OBJECTIVE: This study aims to explore speech as an alternative modality for human activity recognition (HAR) in medical settings. While current HAR technologies rely on video and sensory modalities, they are often unsuitable for the medical environment due to interference from medical personnel, privacy concerns, and environmental limitations. Therefore, we propose an end-to-end, fully automatic objective checklist validation framework that utilizes medical personnel's uttered speech to recognize and document the executed actions in a checklist format. METHODS: Our framework records, processes, and analyzes medical personnel's speech to extract valuable information about performed actions. This information is then used to fill the corresponding rubrics in the checklist automatically. RESULTS: Our approach to activity recognition outperformed the online expert examiner, achieving an F1 score of 0.869 on verbal tasks and an ICC score of 0.822 with an offline examiner. Furthermore, the framework successfully identified communication failures and medical errors made by physicians and nurses. CONCLUSION: Implementing a speech-based framework in medical settings, such as the emergency room and operation room, holds promise for improving care delivery and enabling the development of automated assistive technologies in various medical domains. By leveraging speech as a modality for HAR, we can overcome the limitations of existing technologies and enhance workflow efficiency and patient safety.

BACKGROUND: The International Classification of Diseases (ICD) codes represent the global standard for reporting disease conditions. The current ICD codes connote direct human-defined relationships among diseases in a hierarchical tree structure. Representing the ICD codes as mathematical vectors helps to capture nonlinear relationships in medical ontologies across diseases. METHODS: We propose a universally applicable framework called "ICD2Vec" designed to provide mathematical representations of diseases by encoding corresponding information. First, we present the arithmetical and semantic relationships between diseases by mapping composite vectors for symptoms or diseases to the most similar ICD codes. Second, we investigated the validity of ICD2Vec by comparing the biological relationships and cosine similarities among the vectorized ICD codes. Third, we propose a new risk score called IRIS, derived from ICD2Vec, and demonstrate its clinical utility with large cohorts from the UK and South Korea. RESULTS: Semantic compositionality was qualitatively confirmed between descriptions of symptoms and ICD2Vec. For example, the diseases most similar to COVID-19 were found to be the common cold (ICD-10: J00), unspecified viral hemorrhagic fever (ICD-10: A99), and smallpox (ICD-10: B03). We show the significant associations between the cosine similarities derived from ICD2Vec and the biological relationships using disease-to-disease pairs. Furthermore, we observed significant adjusted hazard ratios (HR) and area under the receiver operating characteristics (AUROC) between IRIS and risks for eight diseases. For instance, the higher IRIS for coronary artery disease (CAD) can be the higher probability for the incidence of CAD (HR: 2.15 [95% CI 2.02-2.28] and AUROC: 0.587 [95% CI 0.583-0.591]). We identified individuals at substantially increased risk of CAD using IRIS and 10-year atherosclerotic cardiovascular disease risk (adjusted HR: 4.26 [95% CI 3.59-5.05]). CONCLUSIONS: ICD2Vec, a proposed universal framework for converting qualitatively measured ICD codes into quantitative vectors containing semantic relationships between diseases, exhibited a significant correlation with actual biological significance. In addition, the IRIS was a significant predictor of major diseases in a prospective study using two large-scale datasets. Based on this clinical validity and utility evidence, we suggest that publicly available ICD2Vec can be used in diverse research and clinical practices and has important clinical implications.

Pretrained language models (PLMs) have demonstrated strong performance on many natural language processing (NLP) tasks. Despite their great success, these PLMs are typically pretrained only on unstructured free texts without leveraging existing structured knowledge bases that are readily available for many domains, especially scientific domains. As a result, these PLMs may not achieve satisfactory performance on knowledge-intensive tasks such as biomedical NLP. Comprehending a complex biomedical document without domain-specific knowledge is challenging, even for humans. Inspired by this observation, we propose a general framework for incorporating various types of domain knowledge from multiple sources into biomedical PLMs. We encode domain knowledge using lightweight adapter modules, bottleneck feed-forward networks that are inserted into different locations of a backbone PLM. For each knowledge source of interest, we pretrain an adapter module to capture the knowledge in a self-supervised way. We design a wide range of self-supervised objectives to accommodate diverse types of knowledge, ranging from entity relations to description sentences. Once a set of pretrained adapters is available, we employ fusion layers to combine the knowledge encoded within these adapters for downstream tasks. Each fusion layer is a parameterized mixer of the available trained adapters that can identify and activate the most useful adapters for a given input. Our method diverges from prior work by including a knowledge consolidation phase, during which we teach the fusion layers to effectively combine knowledge from both the original PLM and newly-acquired external knowledge using a large collection of unannotated texts. After the consolidation phase, the complete knowledge-enhanced model can be fine-tuned for any downstream task of interest to achieve optimal performance. Extensive experiments on many biomedical NLP datasets show that our proposed framework consistently improves the performance of the underlying PLMs on various downstream tasks such as natural language inference, question answering, and entity linking. These results demonstrate the benefits of using multiple sources of external knowledge to enhance PLMs and the effectiveness of the framework for incorporating knowledge into PLMs. While primarily focused on the biomedical domain in this work, our framework is highly adaptable and can be easily applied to other domains, such as the bioenergy sector.

A clinical sentiment is a judgment, thought or attitude promoted by an observation with respect to the health of an individual. Sentiment analysis has drawn attention in the healthcare domain for secondary use of data from clinical narratives, with a variety of applications including predicting the likelihood of emerging mental illnesses or clinical outcomes. The current state of research has not yet been summarized. This study presents results from a scoping review aiming at providing an overview of sentiment analysis of clinical narratives in order to summarize existing research and identify open research gaps. The scoping review was carried out in line with the PRISMA-ScR (Preferred Reporting Items for Systematic reviews and Meta-Analyses extension for Scoping Reviews) guideline. Studies were identified by searching 4 electronic databases (e.g., PubMed, IEEE Xplore) in addition to conducting backward and forward reference list checking of the included studies. We extracted information on use cases, methods and tools applied, used datasets and performance of the sentiment analysis approach. Of 1,200 citations retrieved, 29 unique studies were included in the review covering a period of 8 years. Most studies apply general domain tools (e.g. TextBlob) and sentiment lexicons (e.g. SentiWordNet) for realizing use cases such as prediction of clinical outcomes; others proposed new domain-specific sentiment analysis approaches based on machine learning. Accuracy values between 71.5-88.2% are reported. Data used for evaluation and test are often retrieved from MIMIC databases or i2b2 challenges. Latest developments related to artificial neural networks are not yet fully considered in this domain. We conclude that future research should focus on developing a gold standard sentiment lexicon, adapted to the specific characteristics of clinical narratives. Efforts have to be made to either augment existing or create new high-quality labeled data sets of clinical narratives. Last, the suitability of state-of-the-art machine learning methods for natural language processing and in particular transformer-based models should be investigated for their application for sentiment analysis of clinical narratives.

Automatic extraction of patient medication histories from free-text clinical notes can increase the amount of relevant information to clinicians for developing treatment plans. In addition to detecting medication events, clinical text mining systems must also be able to predict event context, such as negation, uncertainty, and time of occurrence, in order to construct accurate patient timelines. Towards this goal, we introduce Levitated Context Markers (LCMs), a novel transformer-based model for contextualized event extraction. LCMs are an adaptation of levitated markers -originally developed for relation extraction- that allow pretrained transformer models to utilize global input representations while also focusing on event-related subspans using a sparse attention mechanism. In addition to outperforming a strong baseline model on the Contextualized Medication Event Dataset, we show that LCMs' sparse attention can provide interpretable predictions by detecting relevant context cues in an unsupervised manner.

Data stewardship is a term that is understood in heterogenous ways. In recent organisational developments and efforts to build infrastructures and hire professional staff for research data management in various scientific fields in Europe, data stewardship is understood as mainly aiming at optimising data management in line with the FAIR principles (findability, accessibility, interoperability, reusability) forpurposes of reuse in the interests of the scientific community and the public. In addition, especially in the health and biomedical sciences some understandings of data stewardship mainly focus on the responsibility to respect the informational rights of data subjects. Following on from these different understandings and from recent developments to include ever more stakeholders in data stewardship, we propose a comprehensive understanding of data stewardship. According to this comprehensive understanding, data stewardship includes responsibilities towards all pertinent stakeholders and to equally consider and respect their legitimate rights and interests in order to build and maintain an efficient, trusted and fair data ecosystem. We also point out some of the practical challenges implied in such a comprehensive understanding.

The past decade has witnessed an explosion of textual information in the biomedical field. Biomedical texts provide a basis for healthcare delivery, knowledge discovery, and decision-making. Over the same period, deep learning has achieved remarkable performance in biomedical natural language processing, however, its development has been limited by well-annotated datasets and interpretability. To solve this, researchers have considered combining domain knowledge (such as biomedical knowledge graph) with biomedical data, which has become a promising means of introducing more information into biomedical datasets and following evidence-based medicine. This paper comprehensively reviews more than 150 recent literature studies on incorporating domain knowledge into deep learning models to facilitate typical biomedical text analysis tasks, including information extraction, text classification, and text generation. We eventually discuss various challenges and future directions.

Alerts are a common functionality of clinical decision support systems (CDSSs). Although they have proven to be useful in clinical practice, the alert burden can lead to alert fatigue and significantly reduce their usability and acceptance. Based on a literature review, we propose a unified framework consisting of a set of meaningful timestamps that allows the use of state-of-the-art measures for alert burden, such as alert dwell time, alert think time, and response time. In addition, it can be used to investigate other measures that could be relevant as regards dealing with this problem. Furthermore, we provide a case study concerning three different types of alerts to which the framework was successfully applied. We consider that our framework can easily be adapted to other CDSSs and that it could be useful for dealing with alert burden measurement thus contributing to its appropriate management.

Identifying patient cohorts meeting the criteria of specific phenotypes is essential in biomedicine and particularly timely in precision medicine. Many research groups deliver pipelines that automatically retrieve and analyze data elements from one or more sources to automate this task and deliver high-performing computable phenotypes. We applied a systematic approach based on the Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines to conduct a thorough scoping review on computable clinical phenotyping. Five databases were searched using a query that combined the concepts of automation, clinical context, and phenotyping. Subsequently, four reviewers screened 7960 records (after removing over 4000 duplicates) and selected 139 that satisfied the inclusion criteria. This dataset was analyzed to extract information on target use cases, data-related topics, phenotyping methodologies, evaluation strategies, and portability of developed solutions. Most studies supported patient cohort selection without discussing the application to specific use cases, such as precision medicine. Electronic Health Records were the primary source in 87.1 % (N = 121) of all studies, and International Classification of Diseases codes were heavily used in 55.4 % (N = 77) of all studies, however, only 25.9 % (N = 36) of the records described compliance with a common data model. In terms of the presented methods, traditional Machine Learning (ML) was the dominant method, often combined with natural language processing and other approaches, while external validation and portability of computable phenotypes were pursued in many cases. These findings revealed that defining target use cases precisely, moving away from sole ML strategies, and evaluating the proposed solutions in the real setting are essential opportunities for future work. There is also momentum and an emerging need for computable phenotyping to support clinical and epidemiological research and precision medicine.

The imputation of missing values in multivariate time series (MTS) data is critical in ensuring data quality and producing reliable data-driven predictive models. Apart from many statistical approaches, a few recent studies have proposed state-of-the-art deep learning methods to impute missing values in MTS data. However, the evaluation of these deep methods is limited to one or two data sets, low missing rates, and completely random missing value types. This survey performs six data-centric experiments to benchmark state-of-the-art deep imputation methods on five time series health data sets. Our extensive analysis reveals that no single imputation method outperforms the others on all five data sets. The imputation performance depends on data types, individual variable statistics, missing value rates, and types. Deep learning methods that jointly perform cross-sectional (across variables) and longitudinal (across time) imputations of missing values in time series data yield statistically better data quality than traditional imputation methods. Although computationally expensive, deep learning methods are practical given the current availability of high-performance computing resources, especially when data quality and sample size are of paramount importance in healthcare informatics. Our findings highlight the importance of data-centric selection of imputation methods to optimize data-driven predictive models.

Cancer is the second leading cause of death globally, trailing only heart disease. In the United States alone, 1.9 million new cancer cases and 609,360 deaths were recorded for 2022. Unfortunately, the success rate for new cancer drug development remains less than 10%, making the disease particularly challenging. This low success rate is largely attributed to the complex and poorly understood nature of cancer etiology. Therefore, it is critical to find alternative approaches to understanding cancer biology and developing effective treatments. One such approach is drug repurposing, which offers a shorter drug development timeline and lower costs while increasing the likelihood of success. In this review, we provide a comprehensive analysis of computational approaches for understanding cancer biology, including systems biology, multi-omics, and pathway analysis. Additionally, we examine the use of these methods for drug repurposing in cancer, including the databases and tools that are used for cancer research. Finally, we present case studies of drug repurposing, discussing their limitations and offering recommendations for future research in this area.

BACKGROUND: Electronic health records (EHRs) are generated at an ever-increasing rate. EHR trajectories, the temporal aspect of health records, facilitate predicting patients' future health-related risks. It enables healthcare systems to increase the quality of care through early identification and primary prevention. Deep learning techniques have shown great capacity for analyzing complex data and have been successful for prediction tasks using complex EHR trajectories. This systematic review aims to analyze recent studies to identify challenges, knowledge gaps, and ongoing research directions. METHODS: For this systematic review, we searched Scopus, PubMed, IEEE Xplore, and ACM databases from Jan 2016 to April 2022 using search terms centered around EHR, deep learning, and trajectories. Then the selected papers were analyzed according to publication characteristics, objectives, and their solutions regarding existing challenges, such as the model's capacity to deal with intricate data dependencies, data insufficiency, and explainability. RESULTS: After removing duplicates and out-of-scope papers, 63 papers were selected, which showed rapid growth in the number of research in recent years. Predicting all diseases in the next visit and the onset of cardiovascular diseases were the most common targets. Different contextual and non-contextual representation learning methods are employed to retrieve important information from the sequence of EHR trajectories. Recurrent neural networks and the time-aware attention mechanism for modeling long-term dependencies, self-attentions, convolutional neural networks, graphs for representing inner visit relations, and attention scores for explainability were frequently used among the reviewed publications. CONCLUSIONS: This systematic review demonstrated how recent breakthroughs in deep learning methods have facilitated the modeling of EHR trajectories. Research on improving the ability of graph neural networks, attention mechanisms, and cross-modal learning to analyze intricate dependencies among EHRs has shown good progress. There is a need to increase the number of publicly available EHR trajectory datasets to allow for easier comparison among different models. Also, very few developed models can handle all aspects of EHR trajectory data.

BACKGROUND: Few-shot learning (FSL) is a class of machine learning methods that require small numbers of labeled instances for training. With many medical topics having limited annotated text-based data in practical settings, FSL-based natural language processing (NLP) holds substantial promise. We aimed to conduct a review to explore the current state of FSL methods for medical NLP. METHODS: We searched for articles published between January 2016 and October 2022 using PubMed/Medline, Embase, ACL Anthology, and IEEE Xplore Digital Library. We also searched the preprint servers (e.g., arXiv, medRxiv, and bioRxiv) via Google Scholar to identify the latest relevant methods. We included all articles that involved FSL and any form of medical text. We abstracted articles based on the data source, target task, training set size, primary method(s)/approach(es), and evaluation metric(s). RESULTS: Fifty-one articles met our inclusion criteria-all published after 2018, and most since 2020 (42/51; 82%). Concept extraction/named entity recognition was the most frequently addressed task (21/51; 41%), followed by text classification (16/51; 31%). Thirty-two (61%) articles reconstructed existing datasets to fit few-shot scenarios, and MIMIC-III was the most frequently used dataset (10/51; 20%). 77% of the articles attempted to incorporate prior knowledge to augment the small datasets available for training. Common methods included FSL with attention mechanisms (20/51; 39%), prototypical networks (11/51; 22%), meta-learning (7/51; 14%), and prompt-based learning methods, the latter being particularly popular since 2021. Benchmarking experiments demonstrated relative underperformance of FSL methods on biomedical NLP tasks. CONCLUSION: Despite the potential for FSL in biomedical NLP, progress has been limited. This may be attributed to the rarity of specialized data, lack of standardized evaluation criteria, and the underperformance of FSL methods on biomedical topics. The creation of publicly-available specialized datasets for biomedical FSL may aid method development by facilitating comparative analyses.

Building on previous work to define the scientific discipline of biomedical informatics, we present a framework that categorizes fundamental challenges into groups based on data, information, and knowledge, along with the transitions between these levels. We define each level and argue that the framework provides a basis for separating informatics problems from non-informatics problems, identifying fundamental challenges in biomedical informatics, and provides guidance regarding the search for general, reusable solutions to informatics problems. We distinguish between processing data (symbols) and processing meaning. Computational systems, that are the basis for modern information technology (IT), process data. In contrast, many important challenges in biomedicine, such as providing clinical decision support, require processing meaning, not data. Biomedical informatics is hard because of the fundamental mismatch between many biomedical problems and the capabilities of current technology.

BACKGROUND AND OBJECTIVE: Automatic clinical coding is a crucial task in the process of extracting relevant information from unstructured medical documents contained in Electronic Health Records (EHR). However, most of the existing computer-based methods for clinical coding act as "black boxes", without giving a detailed description of the reasons for the clinical-coding assignments, which greatly limits their applicability to real-world medical scenarios. The objective of this study is to use transformer-based models to effectively tackle explainable clinical-coding. In this way, we require the models to perform the assignments of clinical codes to medical cases, but also to provide the reference in the text that justifies each coding assignment. METHODS: We examine the performance of 3 transformer-based architectures on 3 different explainable clinical-coding tasks. For each transformer, we compare the performance of the original general-domain version with an in-domain version of the model adapted to the specificities of the medical domain. We address the explainable clinical-coding problem as a dual medical named entity recognition (MER) and medical named entity normalization (MEN) task. For this purpose, we have developed two different approaches, namely a multi-task and a hierarchical-task strategy. RESULTS: For each analyzed transformer, the clinical-domain version significantly outperforms the corresponding general domain model across the 3 explainable clinical-coding tasks analyzed in this study. Furthermore, the hierarchical-task approach yields a significantly superior performance than the multi-task strategy. Specifically, the combination of the hierarchical-task strategy with an ensemble approach leveraging the predictive capabilities of the 3 distinct clinical-domain transformers, yields the best obtained results, with f1-score, precision and recall of 0.852, 0.847 and 0.849 on the Cantemist-Norm task and 0.718, 0.566 and 0.633 on the CodiEsp-X task, respectively. CONCLUSIONS: By separately addressing the MER and MEN tasks, as well as by following a context-aware text-classification approach to tackle the MEN task, the hierarchical-task approach effectively reduces the intrinsic complexity of explainable clinical-coding, leading the transformers to establish new SOTA performances for the predictive tasks considered in this study. In addition, the proposed methodology has the potential to be applied to other clinical tasks that require both the recognition and normalization of medical entities.

Multi-view clustering methods are essential for the stratification of patients into sub-groups of similar molecular characteristics. In recent years, a wide range of methods have been developed for this purpose. However, due to the high diversity of cancer-related data, a single method may not perform sufficiently well in all cases. We present Parea, a multi-view hierarchical ensemble clustering approach for disease subtype discovery. We demonstrate its performance on several machine learning benchmark datasets. We apply and validate our methodology on real-world multi-view patient data, comprising seven types of cancer. Parea outperforms the current state-of-the-art on six out of seven analysed cancer types. We have integrated the Parea method into our Python package Pyrea (https://github.com/mdbloice/Pyrea), which enables the effortless and flexible design of ensemble workflows while incorporating a wide range of fusion and clustering algorithms.

The tumoral immune microenvironment (TIME) plays a key role in prognosis, therapeutic approach and pathophysiological understanding over oncological processes. Several computational immune cell-type deconvolution methods (DM), supported by diverse molecular signatures (MS), have been developed to uncover such TIME interplay from RNA-seq tumor biopsies. MS-DM pairs were benchmarked against each other by means of different metrics, such as Pearson's correlation, R2 and RMSE, but these only evaluate the linear association of the estimated proportion related to the expected one, missing the analysis of prediction-dependent bias trends and cell identification accuracy. We present a novel protocol composed of four tests allowing appropriate evaluation of the cell type identification performance and proportion prediction accuracy of molecular signature-deconvolution method pair by means of certainty and confidence cell-type identification scores (F1-score, distance to the optimal point and error rates) as well the Bland-Altman method for error-trend analysis. Our protocol was used to benchmark six state-of-the-art DMs (CIBERSORTx, DCQ, DeconRNASeq, EPIC, MIXTURE and quanTIseq) paired to five murine tissue-specific MSs, revealing a systematic overestimation of the number of different cell types across almost all methods.

OBJECTIVE: In recent years, we have increasingly observed issues concerning quality of online information due to misinformation and disinformation. Aside from social media, there is growing awareness that questionnaire data collected using online recruitment methods may include suspect data provided by bots. Issues with data quality can be particularly problematic in health and/or biomedical contexts; thus, developing robust methods for suspect data identification and removal is of paramount importance in informatics. In this study, we describe an interactive visual analytics approach to suspect data identification and removal and demonstrate the application of this approach on questionnaire data pertaining to COVID-19 derived from different recruitment venues, including listservs and social media. METHODS: We developed a pipeline for data cleaning, pre-processing, analysis, and automated ranking of data to address data quality issues. We then employed the ranking in conjunction with manual review to identify suspect data and remove them from subsequent analyses. Last, we compared differences in the data before and after removal. RESULTS: We performed data cleaning, pre-processing, and exploratory analysis on a survey dataset (N = 4,163) collected using multiple recruitment mechanins using the Qualtrics survey platform. Based on these results, we identified suspect features and used these to generate a suspect feature indicator for each survey response. We excluded survey responses that did not fit the inclusion criteria for the study (n = 29) and then performed manual review of the remaining responses, triangulating with the suspect feature indicator. Based on this review, we excluded 2,921 responses. Additional responses were excluded based on a spam classification by Qualtrics (n=13), and the percentage of survey completion (n=328), resulting in a final sample size of 872. We performed additional analyses to demonstrate the extent to which the suspect feature indicator was congruent with eventual inclusion, as well as compared the characteristics of the included and excluded data. CONCLUSION: Our main contributions are: 1) a proposed framework for data quality assessment, including suspect data identification and removal; 2) the analysis of potential consequences in terms of representation bias in the dataset; and 3) recommendations for implementation of this approach in practice.

With the growth of data and intelligent technologies, the healthcare sector opened numerous technology that enabled services for patients, clinicians, and researchers. One major hurdle in achieving state-of-the-art results in health informatics is domain-specific terminologies and their semantic complexities. A knowledge graph crafted from medical concepts, events, and relationships acts as a medical semantic network to extract new links and hidden patterns from health data sources. Current medical knowledge graph construction studies are limited to generic techniques and opportunities and focus less on exploiting real-world data sources in knowledge graph construction. A knowledge graph constructed from Electronic Health Records (EHR) data obtains real-world data from healthcare records. It ensures better results in subsequent tasks like knowledge extraction and inference, knowledge graph completion, and medical knowledge graph applications such as diagnosis predictions, clinical recommendations, and clinical decision support. This review critically analyses existing works on medical knowledge graphs that used EHR data as the data source at (i) representation level, (ii) extraction level (iii) completion level. In this investigation, we found that EHR-based knowledge graph construction involves challenges such as high complexity and dimensionality of data, lack of knowledge fusion, and dynamic update of the knowledge graph. In addition, the study presents possible ways to tackle the challenges identified. Our findings conclude that future research should focus on knowledge graph integration and knowledge graph completion challenges.

OBJECTIVE: Evidence-based medicine (EBM) is a decision-making process based on the conscious and judicious use of the best available scientific evidence. However, the exponential increase in the amount of information currently available likely exceeds the capacity of human-only analysis. In this context, artificial intelligence (AI) and its branches such as machine learning (ML) can be used to facilitate human efforts in analyzing the literature to foster EBM. The present scoping review aimed to examine the use of AI in the automation of biomedical literature survey and analysis with a view to establishing the state-of-the-art and identifying knowledge gaps. MATERIALS AND METHODS: Comprehensive searches of the main databases were performed for articles published up to June 2022 and studies were selected according to inclusion and exclusion criteria. Data were extracted from the included articles and the findings categorized. RESULTS: The total number of records retrieved from the databases was 12,145, of which 273 were included in the review. Classification of the studies according to the use of AI in evaluating the biomedical literature revealed three main application groups, namely assembly of scientific evidence (n = 127; 47%), mining the biomedical literature (n = 112; 41%) and quality analysis (n = 34; 12%). Most studies addressed the preparation of systematic reviews, while articles focusing on the development of guidelines and evidence synthesis were the least frequent. The biggest knowledge gap was identified within the quality analysis group, particularly regarding methods and tools that assess the strength of recommendation and consistency of evidence. CONCLUSION: Our review shows that, despite significant progress in the automation of biomedical literature surveys and analyses in recent years, intense research is needed to fill knowledge gaps on more difficult aspects of ML, deep learning and natural language processing, and to consolidate the use of automation by end-users (biomedical researchers and healthcare professionals).

Federated learning initiatives in healthcare are being developed to collaboratively train predictive models without the need to centralize sensitive personal data. GenoMed4All is one such project, with the goal of connecting European clinical and -omics data repositories on rare diseases through a federated learning platform. Currently, the consortium faces the challenge of a lack of well-established international datasets and interoperability standards for federated learning applications on rare diseases. This paper presents our practical approach to select and implement a Common Data Model (CDM) suitable for the federated training of predictive models applied to the medical domain, during the initial design phase of our federated learning platform. We describe our selection process, composed of identifying the consortium's needs, reviewing our functional and technical architecture specifications, and extracting a list of business requirements. We review the state of the art and evaluate three widely-used approaches (FHIR, OMOP and Phenopackets) based on a checklist of requirements and specifications. We discuss the pros and cons of each approach considering the use cases specific to our consortium as well as the generic issues of implementing a European federated learning healthcare platform. A list of lessons learned from the experience in our consortium is discussed, from the importance of establishing the proper communication channels for all stakeholders to technical aspects related to -omics data. For federated learning projects focused on secondary use of health data for predictive modeling, encompassing multiple data modalities, a phase of data model convergence is sorely needed to gather different data representations developed in the context of medical research, interoperability of clinical care software, imaging, and -omics analysis into a coherent, unified data model. Our work identifies this need and presents our experience and a list of actionable lessons learned for future work in this direction.

BACKGROUND: An accurate medication history, foundational for providing quality medical care, requires understanding of medication change events documented in clinical notes. However, extracting medication changes without the necessary clinical context is insufficient for real-world applications. METHODS: To address this need, Track 1 of the 2022 National NLP Clinical Challenges focused on extracting the context for medication changes documented in clinical notes using the Contextualized Medication Event Dataset. Track 1 consisted of 3 subtasks: extracting medication mentions from clinical notes (NER), determining whether a medication change is being discussed (Event), and determining the action, negation, temporality, certainty, and actor for any change events (Context). Participants were allowed to participate in any one or more of the subtasks. RESULTS: A total of 32 teams with participants from 19 countries submitted a total of 211 systems across all subtasks. Most teams formulated NER as a token classification task and Event and Context as multi-class classification tasks, using transformer-based large language models. Overall, performance for NER was high across submitted systems. However, performance for Event and Context were much lower, often due to indirectly stated change events with no clear action verb, events requiring farther textual clues for understanding, and medication mentions with multiple change events. CONCLUSIONS: This shared task showed that while NLP research on medication extraction is relatively mature, understanding of contextual information surrounding medication events in clinical notes is still an open problem requiring further research to achieve the end goal of supporting real-world clinical applications.

BACKGROUND: The diagnosis of rare genetic diseases is often challenging due to the complexity of the genetic underpinnings of these conditions and the limited availability of diagnostic tools. Machine learning (ML) algorithms have the potential to improve the accuracy and speed of diagnosis by analyzing large amounts of genomic data and identifying complex multiallelic patterns that may be associated with specific diseases. In this systematic review, we aimed to identify the methodological trends and the ML application areas in rare genetic diseases. METHODS: We performed a systematic review of the literature following the PRISMA guidelines to search studies that used ML approaches to enhance the diagnosis of rare genetic diseases. Studies that used DNA-based sequencing data and a variety of ML algorithms were included, summarized, and analyzed using bibliometric methods, visualization tools, and a feature co-occurrence analysis. FINDINGS: Our search identified 22 studies that met the inclusion criteria. We found that exome sequencing was the most frequently used sequencing technology (59%), and rare neoplastic diseases were the most prevalent disease scenario (59%). In rare neoplasms, the most frequent applications of ML models were the differential diagnosis or stratification of patients (38.5%) and the identification of somatic mutations (30.8%). In other rare diseases, the most frequent goals were the prioritization of rare variants or genes (55.5%) and the identification of biallelic or digenic inheritance (33.3%). The most employed method was the random forest algorithm (54.5%). In addition, the features of the datasets needed for training these algorithms were distinctive depending on the goal pursued, including the mutational load in each gene for the differential diagnosis of patients, or the combination of genotype features and sequence-derived features (such as GC-content) for the identification of somatic mutations. CONCLUSIONS: ML algorithms based on sequencing data are mainly used for the diagnosis of rare neoplastic diseases, with random forest being the most common approach. We identified key features in the datasets used for training these ML models according to the objective pursued. These features can support the development of future ML models in the diagnosis of rare genetic diseases.

In the three years since SARS-CoV-2 was first detected in China, hundreds of millions of people have been infected and millions have died. Along with the immediate need for treatment solutions, the COVID-19 epidemic has reinforced the need for mathematical models that can predict the spread of the pandemic in an ever-changing environment. The susceptible-infectious-removed (SIR) model has been widely used to model COVID-19 transmission, however, with limited success. Here, we present a novel, dynamic Monte-Carlo Agent-based Model (MAM), which is based on the basic principles of statistical physics. Using public aggregative data from Israel on three major outbreaks, we compare predictions made by SIR and MAM, and show that MAM outperforms SIR in all aspects. Furthermore, MAM is a flexible model and allows to accurately examine the effects of vaccinations in different subgroups, and the effects of the introduction of new variants.

As applications of the gene ontology (GO) increase rapidly in the biomedical field, quality auditing of it is becoming more and more important. Existing auditing methods are mostly based on rules, observed patterns or hypotheses. In this study, we propose a machine-learning-based framework for GO to audit itself: we first predict the IS-A relations among concepts in GO, then use differences between predicted results and existing relations to uncover potential errors. Specifically, we transfer the taxonomy of GO 2020 January release into a dataset with concept pairs as items and relations between them as labels(pairs with no direct IS-A relation are labeled as ndrs). To fully obtain the representation of each pair, we integrate the embeddings for the concept name, concept definition, as well as concept node in a substring-based topological graph. We divide the dataset into 10 parts, and rotate over all the parts by choosing one part as the testing set and the remaining as the training set each time. After 10 rotations, the prediction model predicted 4,640 existing IS-A pairs as ndrs. In the GO 2022 March release, 340 of these predictions were validated, demonstrating significance with a p-value of 1.60e-46 when compared to the results of randomly selected pairs. On the other hand, the model predicted 2,840 out of 17,079 selected ndrs in GO to be IS-A's relations. After deleting those that caused redundancies and circles, 924 predicted IS-A's relations remained. Among 200 pairs randomly selected, 30 were validated as missing IS-A's by domain experts. In conclusion, this study investigates a novel way of auditing biomedical ontologies by predicting the relations in it, which was shown to be useful for discovering potential errors.

In the era of digital healthcare, the huge volumes of textual information generated every day in hospitals constitute an essential but underused asset that could be exploited with task-specific, fine-tuned biomedical language representation models, improving patient care and management. For such specialized domains, previous research has shown that fine-tuning models stemming from broad-coverage checkpoints can largely benefit additional training rounds over large-scale in-domain resources. However, these resources are often unreachable for less-resourced languages like Italian, preventing local medical institutions to employ in-domain adaptation. In order to reduce this gap, our work investigates two accessible approaches to derive biomedical language models in languages other than English, taking Italian as a concrete use-case: one based on neural machine translation of English resources, favoring quantity over quality; the other based on a high-grade, narrow-scoped corpus natively written in Italian, thus preferring quality over quantity. Our study shows that data quantity is a harder constraint than data quality for biomedical adaptation, but the concatenation of high-quality data can improve model performance even when dealing with relatively size-limited corpora. The models published from our investigations have the potential to unlock important research opportunities for Italian hospitals and academia. Finally, the set of lessons learned from the study constitutes valuable insights towards a solution to build biomedical language models that are generalizable to other less-resourced languages and different domain settings.

MOTIVATION: Entity linking is the task of linking entity mentions to the database entries corresponding to the entity mentions. Entity linking enables the treatment of superficially different but semantically identical mentions as the same entity. Since millions of concepts are listed in biomedical databases, selecting the correct database entry for each targeted entity is challenging. Simple string matching between the word and each synonym in biomedical databases is insufficient to handle a wide variety of variants of biomedical entities appearing in the biomedical literature. Recent progress in neural approaches is promising for entity linking. Still, existing neural methods require sufficient data, which is difficult to prepare in biomedical entity linking that deals with millions of biomedical concepts. Therefore, we need to develop a new neural method to train entity-linking models over the sparse training data covering a very limited part of the biomedical concepts. RESULTS: We have devised a pure neural model that classifies biomedical entity mentions into millions of biomedical concepts. The classifier employs (1) the layer overwriting that breaks through the performance ceiling during training, (2) training data augmentation using database entries that compensate for the problem of insufficient training data, and (3) the cosine similarity-based loss function that helps distinguish the millions of biomedical concepts. Our system using the proposed classifier was ranked first in the official run of the National NLP Clinical Challenges (n2c2) 2019 Track 3, which targeted linking medical/clinical entity mentions to 434,056 Concept Unique Identifier (CUI) entries. We also applied our system to the MedMentions dataset, which has 3.2M candidate concepts. Experimental results confirmed the same advantages of our proposed method. We further evaluated our system on the NLM-CHEM corpus with 350K candidate concepts, and our system achieved a new state-of-the-art performance on the corpus. AVAILABILITY: https://github.com/tti-coin/bio-linking Contact:makoto.miwa@toyota-ti.ac.jp.

Disease knowledge graphs have emerged as a powerful tool for artificial intelligence to connect, organize, and access diverse information about diseases. Relations between disease concepts are often distributed across multiple datasets, including unstructured plain text datasets and incomplete disease knowledge graphs. Extracting disease relations from multimodal data sources is thus crucial for constructing accurate and comprehensive disease knowledge graphs. We introduce REMAP, a multimodal approach for disease relation extraction. The REMAP machine learning approach jointly embeds a partial, incomplete knowledge graph and a medical language dataset into a compact latent vector space, aligning the multimodal embeddings for optimal disease relation extraction. Additionally, REMAP utilizes a decoupled model structure to enable inference in single-modal data, which can be applied under missing modality scenarios. We apply the REMAP approach to a disease knowledge graph with 96,913 relations and a text dataset of 1.24 million sentences. On a dataset annotated by human experts, REMAP improves language-based disease relation extraction by 10.0% (accuracy) and 17.2% (F1-score) by fusing disease knowledge graphs with language information. Furthermore, REMAP leverages text information to recommend new relationships in the knowledge graph, outperforming graph-based methods by 8.4% (accuracy) and 10.4% (F1-score). REMAP is a flexible multimodal approach for extracting disease relations by fusing structured knowledge and language information. This approach provides a powerful model to easily find, access, and evaluate relations between disease concepts.

OBJECTIVE: To develop a natural language processing (NLP) system to extract medications and contextual information that help understand drug changes. This project is part of the 2022 n2c2 challenge. MATERIALS AND METHODS: We developed NLP systems for medication mention extraction, event classification (indicating medication changes discussed or not), and context classification to classify medication changes context into 5 orthogonal dimensions related to drug changes. We explored 6 state-of-the-art pretrained transformer models for the three subtasks, including GatorTron, a large language model pretrained using > 90 billion words of text (including > 80 billion words from > 290 million clinical notes identified at the University of Florida Health). We evaluated our NLP systems using annotated data and evaluation scripts provided by the 2022 n2c2 organizers. RESULTS: Our GatorTron models achieved the best F1-scores of 0.9828 for medication extraction (ranked 3rd), 0.9379 for event classification (ranked 2nd), and the best micro-average accuracy of 0.9126 for context classification. GatorTron outperformed existing transformer models pretrained using smaller general English text and clinical text corpora, indicating the advantage of large language models. CONCLUSION: This study demonstrated the advantage of using large transformer models for contextual medication information extraction from clinical narratives.

Document-level relation extraction is designed to recognize connections between entities a cross sentences or between sentences. The current mainstream document relation extraction model is mainly based on the graph method or combined with the pre-trained language model, which leads to the relatively complex process of the whole workflow. In this work, we propose biomedical relation extraction based on prompt learning to avoid complex relation extraction processes and obtain decent performance. Particularity, we present a model that combines prompt learning with T5 for document relation extraction, by integrating a mask template mechanism into the model. In addition, this work also proposes a few-shot relation extraction method based on the K-nearest neighbor (KNN) algorithm with prompt learning. We select similar semantic labels through KNN, and subsequently conduct the relation extraction. The results acquired from two biomedical document benchmarks indicate that our model can improve the learning of document semantic information, achieving improvements in the relation F1 score of 3.1% on CDR.

OBJECTIVES: To examine recent literature in order to present a comprehensive overview of the current trends as regards the computational models used to represent the propagation of an infectious outbreak in a population, paying particular attention to those that represent network-based transmission. METHODS: a systematic review was conducted following the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guidelines. Papers published in English between 2010 and September 2021 were sought in the ACM Digital Library, IEEE Xplore, PubMed and Scopus databases. RESULTS: Upon considering their titles and abstracts, 832 papers were obtained, of which 192 were selected for a full content-body check. Of these, 112 studies were eventually deemed suitable for quantitative and qualitative analysis. Emphasis was placed on the spatial and temporal scales studied, the use of networks or graphs, and the granularity of the data used to evaluate the models. The models principally used to represent the spreading of outbreaks have been stochastic (55.36%), while the type of networks most frequently used are relationship networks (32.14%). The most common spatial dimension used is a region (19.64%) and the most used unit of time is a day (28.57%). Synthetic data as opposed to an external source were used in 51.79% of the papers. With regard to the granularity of the data sources, aggregated data such as censuses or transportation surveys are the most common. CONCLUSION: We identified a growing interest in the use of networks to represent disease transmission. We detected that research is focused on only certain combinations of the computational model, type of network (in both the expressive and the structural sense) and spatial scale, while the search for other interesting combinations has been left for the future.

BACKGROUND: As the application of Artificial Intelligence (AI) technologies increases in the healthcare sector, the industry faces a need to combine medical knowledge, often expressed as clinical rules, with advances in machine learning (ML), which offer high prediction accuracy at the expense of transparency of decision making. PURPOSE: This paper seeks to review the present literature, identify hybrid architecture patterns that incorporate rules and machine learning, and evaluate the rationale behind their selection to inform future development and research on the design of transparent and precise clinical decision systems. METHODS: PubMed, IEEE Explore, and Google Scholar were queried in search for papers from 1992 to 2022, with the keywords: "clinical decision system", "hybrid clinical architecture", "machine learning and clinical rules". Excluded articles did not use both ML and rules or did not provide any explanation of employed architecture. A proposed taxonomy was used to organize the results, analyze them, and depict them in graphical and tabular form. Two researchers, one with expertise in rule-based systems and another in ML, reviewed identified papers and discussed the work to minimize bias, and the third one re-reviewed the work to ensure consistency of reporting. RESULTS: The authors screened 957 papers and reviewed 71 that met their criteria. Five distinct architecture archetypes were determined: Rules are Embedded in ML architecture (REML) (most used), ML pre-processes input data for Rule-Based inference (MLRB), Rule-Based method pre-processes input data for ML prediction (RBML), Rules influence ML training (RMLT), Parallel Ensemble of Rules and ML (PERML), which was rarely observed in clinical contexts. CONCLUSIONS: Most architectures in the reviewed literature prioritize prediction accuracy over explainability and trustworthiness, which has led to more complex embedded approaches. Alternatively, parallel (PERML) architectures may be employed, allowing for a more transparent system that is easier to explain to patients and clinicians. The potential of this approach warrants further research. OTHER: A limitation of the study may be that it reviews scientific literature, while algorithms implemented in clinical practice may present different distributions of motivations and implementations of hybrid architectures.

Clinical trials are essential to the process of new drug development. As clinical trials involve significant investments of time and money, it is crucial for trial designers to carefully investigate trial settings prior to designing a trial. Utilizing trial documents from ClinicalTrials.gov, we aim to understand the common characteristics of successful and unsuccessful cancer drug trials to provide insights about what to learn and what to avoid. In this research, we first computationally classified cancer drug trials into successful and unsuccessful cases and then utilized natural language processing to extract eligibility criteria information from the trial documents. To provide explainable and potentially modifiable recommendations for new trial design, contrast mining was applied to discoverhighly contrasted patterns with a significant difference in prevalence between successful (completion with advancement to the next phase) and unsuccessful (suspended, withdrawn, or terminated) groups. Our method identified contrast patterns consisting of combinations of drug categories, eligibility criteria, study organization, and study design for nine major cancers. In addition to a literature review for the qualitative validation of mined contrast patterns, we found that contrast-pattern-based classifiers using the top 200 contrast patterns as feature representations can achieve approximately 80% F1 score for eight out of ten cancer types in our experiments. In summary, aligning with the modernization efforts of ClinicalTrials.gov, our study demonstrates that understanding the contrast characteristics of successful and unsuccessful cancer trials may provide insights into the decision-making process for trial investigators and therefore facilitate improved cancer drug trial design.

Monkeypox is a zoonotic infectious skin disease initially endemic in Africa only. However, some countries are now beginning to report cases of apparent community transmission. In Computer Aided Diagnosis, deep learning has gained substantial improvement over traditional methods. Commonly, training a supervised deep model requires a large number of labeled samples. However, the collection and annotation of new disease images such as human monkeypox are time-consuming and expensive. Thus, we introduce a few-shot learning based approach for the recognition of human monkeypox in images. It requires merely a small number of training samples. In particular, it is a novel framework built with a normal backbone and auxiliary backbones. They are co-trained with Self-supervised Learning and Cross-domain Adaption techniques. The self-supervision penalty is used to help the auxiliary backbones effectively learn priors from source domain. The combined features across different domains are unified through a power transform layer. Extensive experiments are conducted on a task of recognizing chickenpox, measles, and human monkeypox diseases in a three-way few-shot manner. The results demonstrate that our method outperforms mainstream few-shot learning algorithms such as meta-learning based and fine-tuning based methods.

While a large number of knowledge graphs have previously been developed by automatically extracting and structuring knowledge from literature, there is currently no such knowledge graph that encodes relationships between food, biochemicals and mental illnesses, even though a large amount of knowledge about these relationships is available in the form of unstructured text in biomedical literature articles. To address this limitation, this article describes the development of GENA - (Graph of mEntal-health and Nutrition Association), a knowledge graph that represents relations between nutrition and mental health, extracted from biomedical abstracts. GENA is constructed from PubMed abstracts that contain keywords relating to chemicals, food, and health. A hybrid named entity recognition (NER) model is firstly applied to these abstracts to identify various entities of interest. Subsequently, a deep syntax-based relation extraction model is used to detect binary relations between the identified entities. Finally, the resulting relations are used to populate the GENA knowledge graph, whose relationships can be accessed in an intuitive and interpretable manner using the Neo4J Database Management System. To evaluate the reliability of GENA, two annotators manually assessed a subset of the extracted relations. The evaluation results show that our methods obtain high precision for the NER task and acceptable precision and relative recall for the relation extraction task. GENA consists of 43,367 relationships that encode information about nutrition and health, of which 94.04% are new relations that are not present in existing ontologies of food and diseases. GENA is constructed based on scientific principles, and has the potential to be used within further applications to contribute towards scientific research within the domain. It is a pioneering knowledge graph in nutrition and mental health, containing a diverse range of relationship types. All of our source code and results are publicly available at https://github.com/ddlinh/gena-db.

Multimodal data-based classification methods have been widely used in the diagnosis of Alzheimer's disease (AD) and have achieved better performance than single-modal-based methods. However, most classification methods based on multimodal data tend to consider only the correlation between different modal data and ignore the inherent non-linear higher-order relationships between similar data, which can improve the robustness of the model. Therefore, this study proposes a hypergraph p-Laplacian regularized multi-task feature selection (HpMTFS) method for AD classification. Specifically, feature selection for each modal data is considered as a distinct task and the common features of multimodal data are extracted jointly by group-sparsity regularizer. In particular, two regularization terms are introduced in this study, namely (1) a hypergraph p-Laplacian regularization term to retain higher-order structural information for similar data, and (2) a Frobenius norm regularization term to improve the noise immunity of the model. Finally, using a multi-kernel support vector machine to fuse multimodal features and perform the final classification. We used baseline sMRI, FDG-PET, and AV-45 PET imaging data from 528 subjects in the Alzheimer's Disease Neuroimaging Initiative (ADNI) to evaluate our approach. Experimental results show that our HpMTFS method outperforms existing multimodal-based classification methods.

Early disease detection and prevention methods based on effective interventions are gaining attention worldwide. Progress in precision medicine has revealed that substantial heterogeneity exists in health data at the individual level and that complex health factors are involved in chronic disease development. Machine-learning techniques have enabled precise personal-level disease prediction by capturing individual differences in multivariate data. However, it is challenging to identify what aspects should be improved for disease prevention based on future disease-onset prediction because of the complex relationships among multiple biomarkers. Here, we present a health-disease phase diagram (HDPD) that represents an individual's health state by visualizing the future-onset boundary values of multiple biomarkers that fluctuate early in the disease progression process. In HDPDs, future-onset predictions are represented by perturbing multiple biomarker values while accounting for dependencies among variables. We constructed HDPDs for 11 diseases using longitudinal health checkup cohort data of 3,238 individuals, comprising 3,215 measurement items and genetic data. The improvement of biomarker values to the non-onset region in HDPD remarkably prevented future disease onset in 7 out of 11 diseases. HDPDs can represent individual physiological states in the onset process and be used as intervention goals for disease prevention.

Obtaining text datasets with semantic annotations is an effortful process, yet crucial for supervised training in natural language processing (NLP). In general, developing and applying new NLP pipelines in domain-specific contexts for tasks often requires custom-designed datasets to address NLP tasks in a supervised machine learning fashion. When operating in non-English languages for medical data processing, this exposes several minor and major, interconnected problems such as the lack of task-matching datasets as well as task-specific pre-trained models. In our work, we suggest to leverage pre-trained large language models for training data acquisition in order to retrieve sufficiently large datasets for training smaller and more efficient models for use-case-specific tasks. To demonstrate the effectiveness of your approach, we create a custom dataset that we use to train a medical NER model for German texts, GPTNERMED, yet our method remains language-independent in principle. Our obtained dataset as well as our pre-trained models are publicly available at https://github.com/frankkramer-lab/GPTNERMED.

OBJECTIVE: This article summarizes our approach to extracting medication and corresponding attributes from clinical notes, which is the focus of track 1 of the 2022 National Natural Language Processing (NLP) Clinical Challenges(n2c2) shared task. METHODS: The dataset was prepared using Contextualized Medication Event Dataset (CMED), including 500 notes from 296 patients. Our system consisted of three components: medication named entity recognition (NER), event classification (EC), and context classification (CC). These three components were built using transformer models with slightly different architecture and input text engineering. A zero-shot learning solution for CC was also explored. RESULTS: Our best performance systems achieved micro-average F1 scores of 0.973, 0.911, and 0.909 for the NER, EC, and CC, respectively. CONCLUSION: In this study, we implemented a deep learning-based NLP system and demonstrated that our approach of (1) utilizing special tokens helps our model to distinguish multiple medications mentions in the same context; (2) aggregating multiple events of a single medication into multiple labels improves our model's performance.

OBJECTIVE: Feasible, safe, and inclusive eligibility criteria are crucial to successful clinical research recruitment. Existing expert-centered methods for eligibility criteria selection may not be representative of real-world populations. This paper presents a novel model called OPTEC (OPTimal Eligibility Criteria) based on the Multiple Attribute Decision Making method boosted by an efficient greedy algorithm. METHODS: It systematically identifies the optimal criteria combination for a given medical condition with the optimal tradeoff among feasibility, patient safety, and cohort diversity. The model offers flexibility in attribute configurations and generalizability to various clinical domains. The model was evaluated on two clinical domains (i.e., Alzheimer's disease and Neoplasm of pancreas) using two datasets (i.e., MIMIC-III dataset and NewYork-Presbyterian/Columbia University Irving Medical Center (NYP/CUIMC) database). RESULTS: We simulated the process of automatically optimizing eligibility criteria according to user-specified prioritization preferences and generated recommendations based on the top-ranked criteria combination accordingly (top 0.41-2.75%) with OPTEC. Harnessing the power of the model, we designed an interactive criteria recommendation system and conducted a case study with an experienced clinical researcher using the think-aloud protocol. CONCLUSIONS: The results demonstrated that OPTEC could be used to recommend feasible eligibility criteria combinations, and to provide actionable recommendations for clinical study designers to construct a feasible, safe, and diverse cohort definition during early study design.

Despite the creation of thousands of machine learning (ML) models, the promise of improving patient care with ML remains largely unrealized. Adoption into clinical practice is lagging, in large part due to disconnects between how ML practitioners evaluate models and what is required for their successful integration into care delivery. Models are just one component of care delivery workflows whose constraints determine clinicians' abilities to act on models' outputs. However, methods to evaluate the usefulness of models in the context of their corresponding workflows are currently limited. To bridge this gap we developed APLUS, a reusable framework for quantitatively assessing via simulation the utility gained from integrating a model into a clinical workflow. We describe the APLUS simulation engine and workflow specification language, and apply it to evaluate a novel ML-based screening pathway for detecting peripheral artery disease at Stanford Health Care.

OBJECTIVE: The paper presents a novel solution to the 2022 National NLP Clinical Challenges (n2c2) Track 3, which aims to predict the relations between assessment and plan subsections in progress notes. METHODS: Our approach goes beyond standard transformer models and incorporates external information such as medical ontology and order information to comprehend the semantics of progress notes. We fine-tuned transformers to understand the textual data and incorporated medical ontology concepts and their relationships to enhance the model's accuracy. We also captured order information that regular transformers cannot by taking into account the position of the assessment and plan subsections in progress notes. RESULTS: Our submission earned third place in the challenge phase with a macro-F1 score of 0.811. After refining our pipeline further, we achieved a macro-F1 of 0.826, outperforming the top-performing system during the challenge phase. CONCLUSION: Our approach, which combines fine-tuned transformers, medical ontology, and order information, outperformed other systems in predicting the relationships between assessment and plan subsections in progress notes. This highlights the importance of incorporating external information beyond textual data in natural language processing (NLP) tasks related to medical documentation. Our work could potentially improve the efficiency and accuracy of progress note analysis.

Segmentation of the left ventricle is a key approach in Cardiac Magnetic Resonance Imaging for calculating biomarkers in diagnosis. Since there is substantial effort required from the expert, many automatic segmentation methods have been proposed, in which deep learning networks have obtained remarkable performance. However, one of the main limitations of these approaches is the production of segmentations that contain anatomical errors. To avoid this limitation, we propose a new fully-automatic left ventricle segmentation method combining deep learning and deformable models. We propose a new level set energy formulation that includes exam-specific information estimated from the deep learning segmentation and shape constraints. The method is part of a pipeline containing pre-processing steps and a failure correction post-processing step. Experiments were conducted with the Sunnybrook and ACDC public datasets, and a private dataset. Results suggest that the method is competitive, that it can produce anatomically consistent segmentations, has good generalization ability, and is often able to estimate biomarkers close to the expert.

CONTEXT: Identifying clusters (i.e., subgroups) of patients from the analysis of medico-administrative databases is particularly important to better understand disease heterogeneity. However, these databases contain different types of longitudinal variables which are measured over different follow-up periods, generating truncated data. It is therefore fundamental to develop clustering approaches that can handle this type of data. OBJECTIVE: We propose here cluster-tracking approaches to identify clusters of patients from truncated longitudinal data contained in medico-administrative databases. MATERIAL AND METHODS: We first cluster patients at each age. We then track the identified clusters over ages to construct cluster-trajectories. We compared our novel approaches with three classical longitudinal clustering approaches by calculating the silhouette score. As a use-case, we analyzed antithrombotic drugs used from 2008 to 2018 contained in the Échantillon Généraliste des Bénéficiaires (EGB), a French national cohort. RESULTS: Our cluster-tracking approaches allow us to identify several cluster-trajectories with clinical significance without any imputation of data. The comparison of the silhouette scores obtained with the different approaches highlights the better performances of the cluster-tracking approaches. CONCLUSION: The cluster-tracking approaches are a novel and efficient alternative to identify patient clusters from medico-administrative databases by taking into account their specificities.

BACKGROUND: Pharmacokinetic natural product-drug interactions (NPDIs) occur when botanical or other natural products are co-consumed with pharmaceutical drugs. With the growing use of natural products, the risk for potential NPDIs and consequent adverse events has increased. Understanding mechanisms of NPDIs is key to preventing or minimizing adverse events. Although biomedical knowledge graphs (KGs) have been widely used for drug-drug interaction applications, computational investigation of NPDIs is novel. We constructed NP-KG as a first step toward computational discovery of plausible mechanistic explanations for pharmacokinetic NPDIs that can be used to guide scientific research. METHODS: We developed a large-scale, heterogeneous KG with biomedical ontologies, linked data, and full texts of the scientific literature. To construct the KG, biomedical ontologies and drug databases were integrated with the Phenotype Knowledge Translator framework. The semantic relation extraction systems, SemRep and Integrated Network and Dynamic Reasoning Assembler, were used to extract semantic predications (subject-relation-object triples) from full texts of the scientific literature related to the exemplar natural products green tea and kratom. A literature-based graph constructed from the predications was integrated into the ontology-grounded KG to create NP-KG. NP-KG was evaluated with case studies of pharmacokinetic green tea- and kratom-drug interactions through KG path searches and meta-path discovery to determine congruent and contradictory information in NP-KG compared to ground truth data. We also conducted an error analysis to identify knowledge gaps and incorrect predications in the KG. RESULTS: The fully integrated NP-KG consisted of 745,512 nodes and 7,249,576 edges. Evaluation of NP-KG resulted in congruent (38.98% for green tea, 50% for kratom), contradictory (15.25% for green tea, 21.43% for kratom), and both congruent and contradictory (15.25% for green tea, 21.43% for kratom) information compared to ground truth data. Potential pharmacokinetic mechanisms for several purported NPDIs, including the green tea-raloxifene, green tea-nadolol, kratom-midazolam, kratom-quetiapine, and kratom-venlafaxine interactions were congruent with the published literature. CONCLUSION: NP-KG is the first KG to integrate biomedical ontologies with full texts of the scientific literature focused on natural products. We demonstrate the application of NP-KG to identify known pharmacokinetic interactions between natural products and pharmaceutical drugs mediated by drug metabolizing enzymes and transporters. Future work will incorporate context, contradiction analysis, and embedding-based methods to enrich NP-KG. NP-KG is publicly available at https://doi.org/10.5281/zenodo.6814507. The code for relation extraction, KG construction, and hypothesis generation is available at https://github.com/sanyabt/np-kg.

Even if assessing binary classifications is a common task in scientific research, no consensus on a single statistic summarizing the confusion matrix has been reached so far. In recent studies, we demonstrated the advantages of the Matthews correlation coefficient (MCC) over other popular rates such as cross-entropy error, F1 score, accuracy, balanced accuracy, bookmaker informedness, diagnostic odds ratio, Brier score, and Cohen's kappa. In this study, we compared the MCC to other two statistics: prevalence threshold (PT), frequently used in obstetrics and gynecology, and Fowlkes-Mallows index, a metric employed in fuzzy logic and drug discovery. Through the investigation of the mutual relations among three metrics and the study of some relevant use cases, we show that, when positive data elements and negative data elements have the same importance, the Matthews correlation coefficient can be more informative than its two competitors, even this time.

This paper describes contextualized medication event extraction for automatically identifying medication change events with their contexts from clinical notes. The striding named entity recognition (NER) model extracts medication name spans from an input text sequence using a sliding-window approach. Specifically, the striding NER model separates the input sequence into a set of overlapping subsequences of 512 tokens with 128 tokens of stride, processing each subsequence using a large pre-trained language model and aggregating the outputs from the subsequences. The event and context classification has been done with multi-turn question-answering (QA) and span-based models. The span-based model classifies the span of each medication name using the span representation of the language model. In the QA model, event classification is augmented with questions in classifying the change events of each medication name and the context of the change events, while the model architecture is a classification style that is the same as the span-based model. We evaluated our extraction system on the n2c2 2022 Track 1 dataset, which is annotated for medication extraction (ME), event classification (EC), and context classification (CC) from clinical notes. Our system is a pipeline of the striding NER model for ME and the ensemble of the span-based and QA-based models for EC and CC. Our system achieved a combined F-score of 66.47% for the end-to-end contextualized medication event extraction (Release 1), which is the highest score among the participants of the n2c2 2022 Track 1.

Molecular property prediction based on artificial intelligence technology has significant prospects in speeding up drug discovery and reducing drug discovery costs. Among them, molecular property prediction based on graph neural networks (GNNs) has received extensive attention in recent years. However, the existing graph neural networks still face the following challenges in node representation learning. First, the number of nodes increases exponentially with the expansion of the perception field, which limits the exploration ability of the model in the depth direction. Secondly, the large number of nodes in the perception field brings noise, which is not conducive to the model's representation learning of the key structures. Therefore, a graph neural network model based on structure generation is proposed in this paper. The model adopts the depth-first strategy to generate the key structures of the graph, to solve the problem of insufficient exploration ability of the graph neural network in the depth direction. A tendentious node selection method is designed to gradually select nodes and edges to generate the key structures of the graph, to solve the noise problem caused by the excessive number of nodes. In addition, the model skillfully realizes forward propagation and iterative optimization of structure generation by using an attention mechanism and random bias. Experimental results on public data sets show that the proposed model achieves better classification results than the existing best models.

Automated ICD coding is a multi-label prediction task aiming at assigning patient diagnoses with the most relevant subsets of disease codes. In the deep learning regime, recent works have suffered from large label set and heavy imbalance distribution. To mitigate the negative effect in such scenarios, we propose a retrieve and rerank framework that introduces the Contrastive Learning (CL) for label retrieval, allowing the model to make more accurate prediction from a simplified label space. Given the appealing discriminative power of CL, we adopt it as the training strategy to replace the standard cross-entropy objective and retrieve a small subset by taking the distance between clinical notes and ICD codes into account. After properly training, the retriever could implicitly capture the code co-occurrence, which makes up for the deficiency of cross-entropy assigning each label independently of the others. Further, we evolve a powerful model via a Transformer variant for refining and reranking the candidate set, which can extract semantically meaningful features from long clinical sequences. Applying our method on well-known models, experiments show that our framework provides more accurate results guaranteed by preselecting a small subset of candidates before fine-level reranking. Relying on the framework, our proposed model achieves 0.590 and 0.990 in terms of Micro-F1 and Micro-AUC on benchmark MIMIC-III.

OBJECTIVE: Acute kidney injury (AKI), a common condition on the intensive-care unit (ICU), is characterized by an abrupt decrease in kidney function within a few hours or days, leading to kidney failure or damage. Although AKI is associated with poor outcomes, current guidelines overlook the heterogeneity among patients with this condition. Identification of AKI subphenotypes could enable targeted interventions and a deeper understanding of the injury's pathophysiology. While previous approaches based on unsupervised representation learning have been used to identify AKI subphenotypes, these methods cannot assess time series or disease severity. METHODS: In this study, we developed a data- and outcome-driven deep-learning (DL) approach to identify and analyze AKI subphenotypes with prognostic and therapeutic implications. Specifically, we developed a supervised long short-term memory (LSTM) autoencoder (AE) with the aim of extracting representation from time-series EHR data that were intricately correlated with mortality. Then, subphenotypes were identified via application of K-means. RESULTS: In two publicly available datasets, three distinct clusters were identified, characterized by mortality rates of 11.3%, 17.3%, and 96.2% in one dataset and 4.6%, 12.1%, and 54.6% in the other. Further analysis demonstrated that AKI subphenotypes identified by our proposed approach were statistically significant on several clinical characteristics and outcomes. CONCLUSION: In this study, our proposed approach could successfully cluster the AKI population in ICU settings into 3 distinct subphenotypes. Thus, such approach could potentially improve outcomes of AKI patients in the ICU, with better risk assessment and potentially better personalized treatment.

OBJECTIVE: To demonstrate and develop an approach enabling individual researchers or small teams to create their own ad-hoc, lightweight knowledge bases tailored for specialized scientific interests, using text-mining over scientific literature, and demonstrate the effectiveness of these knowledge bases in hypothesis generation and literature-based discovery (LBD). METHODS: We propose a lightweight process using an extractive search framework to create ad-hoc knowledge bases, which require minimal training and no background in bio-curation or computer science. These knowledge bases are particularly effective for LBD and hypothesis generation using Swanson's ABC method. The personalized nature of the knowledge bases allows for a somewhat higher level of noise than "public facing" ones, as researchers are expected to have prior domain experience to separate signal from noise. Fact verification is shifted from exhaustive verification of the knowledge base to post-hoc verification of specific entries of interest, allowing researchers to assess the correctness of relevant knowledge base entries by considering the paragraphs in which the facts were introduced. RESULTS: We demonstrate the methodology by constructing several knowledge bases of different kinds: three knowledge bases that support lab-internal hypothesis generation: Drug Delivery to Ovarian Tumors (DDOT); Tissue Engineering and Regeneration; Challenges in Cancer Research; and an additional comprehensive, accurate knowledge base designated as a public resource for the wider community on the topic of Cell Specific Drug Delivery (CSDD). In each case, we show the design and construction process, along with relevant visualizations for data exploration, and hypothesis generation. For CSDD and DDOT we also show meta-analysis, human evaluation, and in vitro experimental evaluation. CONCLUSION: Our approach enables researchers to create personalized, lightweight knowledge bases for specialized scientific interests, effectively facilitating hypothesis generation and literature-based discovery (LBD). By shifting fact verification efforts to post-hoc verification of specific entries, researchers can focus on exploring and generating hypotheses based on their expertise. The constructed knowledge bases demonstrate the versatility and adaptability of our approach to versatile research interests. The web-based platform, available at https://spike-kbc.apps.allenai.org, provides researchers with a valuable tool for rapid construction of knowledge bases tailored to their needs.

Named entity recognition is a key task in text mining. In the biomedical field, entity recognition focuses on extracting key information from large-scale biomedical texts for the downstream information extraction task. Biomedical literature contains a large amount of long-dependent text, and previous studies use external syntactic parsing tools to capture word dependencies in sentences to achieve nested biomedical entity recognition. However, the addition of external parsing tools often introduces unnecessary noise to the current auxiliary task and cannot improve the performance of entity recognition in an end-to-end way. Therefore, we propose a novel automatic dependency parsing approach, namely the ADPG model, to fuse syntactic structure information in an end-to-end way to recognize biomedical entities. Specifically, the method is based on a multilayer Tree-Transformer structure to automatically extract the semantic representation and syntactic structure in long-dependent sentences, and then combines a multilayer graph attention neural network (GAT) to extract the dependency paths between words in the syntactic structure to improve the performance of biomedical entity recognition. We evaluated our ADPG model on three biomedical domain and one news domain datasets, and the experimental results demonstrate that our model achieves state-of-the-art results on these four datasets with certain generalization performance. Our model is released on GitHub: https://github.com/Yumeng-Y/ADPG.

OBJECTIVE: With the onset of the Coronavirus Disease 2019 (COVID-19) pandemic, there has been a surge in the number of publicly available biomedical information sources, which makes it an increasingly challenging research goal to retrieve a relevant text to a topic of interest. In this paper, we propose a Contextual Query Expansion framework based on the clinical Domain knowledge (CQED) for formalizing an effective search over PubMed to retrieve relevant COVID-19 scholarly articles to a given information need. MATERIALS AND METHODS: For the sake of training and evaluation, we use the widely adopted TREC-COVID benchmark. Given a query, the proposed framework utilizes a contextual and a domain-specific neural language model to generate a set of candidate query expansion terms that enrich the original query. Moreover, the framework includes a multi-head attention mechanism that is trained alongside a learning-to-rank model for re-ranking the list of generated expansion candidate terms. The original query and the top-ranked expansion terms are posed to the PubMed search engine for retrieving relevant scholarly articles to an information need. The framework, CQED, can have four different variations, depending upon the learning path adopted for training and re-ranking the candidate expansion terms. RESULTS: The model drastically improves the search performance, when compared to the original query. The performance improvement in comparison to the original query, in terms of RECALL@1000 is 190.85% and in terms of NDCG@1000 is 343.55%. Additionally, the model outperforms all existing state-of-the-art baselines. In terms of P@10, the model that has been optimized based on Precision outperforms all baselines (0.7987). On the other hand, in terms of NDCG@10 (0.7986), MAP (0.3450) and bpref (0.4900), the CQED model that has been optimized based on an average of all retrieval measures outperforms all the baselines. CONCLUSION: The proposed model successfully expands queries posed to PubMed, and improves search performance, as compared to all existing baselines. A success/failure analysis shows that the model improved the search performance of each of the evaluated queries. Moreover, an ablation study depicted that if ranking of generated candidate terms is not conducted, the overall performance decreases. For future work, we would like to explore the application of the presented query expansion framework in conducting technology-assisted Systematic Literature Reviews (SLR).

Scientific literature presents a wealth of information yet to be explored. As the number of researchers increase with each passing year and publications are released, this contributes to an era where specialized fields of research are becoming more prevalent. As this trend continues, this further propagates the separation of interdisciplinary publications and makes keeping up to date with literature a laborious task. Literature-based discovery (LBD) aims to mitigate these concerns by promoting information sharing among non-interacting literature while extracting potentially meaningful information. Furthermore, recent advances in neural network architectures and data representation techniques have fueled their respective research communities in achieving state-of-the-art performance in many downstream tasks. However, studies of neural network-based methods for LBD remain to be explored. We introduce and explore a deep learning neural network-based approach for LBD. Additionally, we investigate various approaches to represent terms as concepts and analyze the affect of feature scaling representations into our model. We compare the evaluation performance of our method on five hallmarks of cancer datasets utilized for closed discovery. Our results show the chosen representation as input into our model affects evaluation performance. We found feature scaling our input representations increases evaluation performance and decreases the necessary number of epochs needed to achieve model generalization. We also explore two approaches to represent model output. We found reducing the model's output to capturing a subset of concepts improved evaluation performance at the cost of model generalizability. We also compare the efficacy of our method on the five hallmarks of cancer datasets to a set of randomly chosen relations between concepts. We found these experiments confirm our method's suitability for LBD.

Clinical documentation in electronic health records contains crucial narratives and details about patients and their care. Natural language processing (NLP) can unlock the information conveyed in clinical notes and reports, and thus plays a critical role in real-world studies. The NLP Working Group at the Observational Health Data Sciences and Informatics (OHDSI) consortium was established to develop methods and tools to promote the use of textual data and NLP in real-world observational studies. In this paper, we describe a framework for representing and utilizing textual data in real-world evidence generation, including representations of information from clinical text in the Observational Medical Outcomes Partnership (OMOP) Common Data Model (CDM), the workflow and tools that were developed to extract, transform and load (ETL) data from clinical notes into tables in OMOP CDM, as well as current applications and specific use cases of the proposed OHDSI NLP solution at large consortia and individual institutions with English textual data. Challenges faced and lessons learned during the process are also discussed to provide valuable insights for researchers who are planning to implement NLP solutions in real-world studies.

BACKGROUND AND OBJECTIVE: Many classification tasks in translational bioinformatics and genomics are characterized by the high dimensionality of potential features and unbalanced sample distribution among classes. This can affect classifier robustness and increase the risk of overfitting, curse of dimensionality and generalization leaks; furthermore and most importantly, this can prevent obtaining adequate patient stratification required for precision medicine in facing complex diseases, like cancer. Setting up a feature selection strategy able to extract only proper predictive features by removing irrelevant, redundant, and noisy ones is crucial to achieving valuable results on the desired task. METHODS: We propose a new feature selection approach, called ReRa, based on supervised Relevance-Redundancy assessments. ReRa consists of a customized step of relevance-based filtering, to identify a reduced subset of meaningful features, followed by a supervised similarity-based procedure to minimize redundancy. This latter step innovatively uses a combination of global and class-specific similarity assessments to remove redundant features while preserving those differentiated across classes, even when these classes are strongly unbalanced. RESULTS: We compared ReRa with several existing feature selection methods to obtain feature spaces on which performing breast cancer patient subtyping using several classifiers: we considered two use cases based on gene or transcript isoform expression. In the vast majority of the assessed scenarios, when using ReRa-selected feature spaces, the performances were significantly increased compared to simple feature filtering, LASSO regularization, or even MRmr - another Relevance-Redundancy method. The two use cases represent an insightful example of translational application, taking advantage of ReRa capabilities to investigate and enhance a clinically-relevant patient stratification task, which could be easily applied also to other cancer types and diseases. CONCLUSIONS: ReRa approach has the potential to improve the performance of machine learning models used in an unbalanced classification scenario. Compared to another Relevance-Redundancy approach like MRmr, ReRa does not require tuning the number of preserved features, ensures efficiency and scalability over huge initial dimensionalities and allows re-evaluation of all previously selected features at each iteration of the redundancy assessment, to ultimately preserve only the most relevant and class-differentiated features.

BACKGROUND: Scientific discovery progresses by exploring new and uncharted territory. More specifically, it advances by a process of transforming unknown unknowns first into known unknowns, and then into knowns. Over the last few decades, researchers have developed many knowledge bases to capture and connect the knowns, which has enabled topic exploration and contextualization of experimental results. But recognizing the unknowns is also critical for finding the most pertinent questions and their answers. Prior work on known unknowns has sought to understand them, annotate them, and automate their identification. However, no knowledge-bases yet exist to capture these unknowns, and little work has focused on how scientists might use them to trace a given topic or experimental result in search of open questions and new avenues for exploration. We show here that a knowledge base of unknowns can be connected to ontologically grounded biomedical knowledge to accelerate research in the field of prenatal nutrition. RESULTS: We present the first ignorance-base, a knowledge-base created by combining classifiers to recognize ignorance statements (statements of missing or incomplete knowledge that imply a goal for knowledge) and biomedical concepts over the prenatal nutrition literature. This knowledge-base places biomedical concepts mentioned in the literature in context with the ignorance statements authors have made about them. Using our system, researchers interested in the topic of vitamin D and prenatal health were able to uncover three new avenues for exploration (immune system, respiratory system, and brain development) by searching for concepts enriched in ignorance statements. These were buried among the many standard enriched concepts. Additionally, we used the ignorance-base to enrich concepts connected to a gene list associated with vitamin D and spontaneous preterm birth and found an emerging topic of study (brain development) in an implied field (neuroscience). The researchers could look to the field of neuroscience for potential answers to the ignorance statements. CONCLUSION: Our goal is to help students, researchers, funders, and publishers better understand the state of our collective scientific ignorance (known unknowns) in order to help accelerate research through the continued illumination of and focus on the known unknowns and their respective goals for scientific knowledge.

Numerous clinical trials based on a single-cause paradigm have not resulted in efficacious treatments for Alzheimer's disease (AD). Recently, prevention trials that simultaneously intervened on multiple risk factors have shown mixed results, suggesting that careful design is necessary. Moreover, intensive pilot precision medicine (PM) trial results have been promising but may not generalize to a broader population. These observations suggest that a model-based approach to multi-factor precision medicine (PM) is warranted. We systematically developed a system dynamics model (SDM) of AD for PM using data from two longitudinal studies (N=3660). This method involved a model selection procedure in identifying interaction terms between the SDM components and estimating individualized parameters. We used the SDM to explore simulated single- and double-factor interventions on 14 modifiable risk factors. We quantified the potential impact of double-factor interventions over single-factor interventions as 1.5 [95% CI: 1.5-2.6] and of SDM-based PM over a one-size-fits-all approach as 3.5 [3.1, 3.8] ADAS-cog-13 points in 12 years. Although the model remains to be validated, we tentatively conclude that multi-factor PM could come to play an important role in AD prevention.

Phenotype-based prioritization of candidate genes and diseases has become a well-established approach for multi-omics diagnostics of rare diseases. Most current algorithms exploit semantic analysis and probabilistic statistics based on Human Phenotype Ontology and are commonly superior to naive search methods. However, these algorithms are mostly less interpretable and do not perform well in real clinical scenarios due to noise and imprecision of query terms, and the fact that individuals may not display all phenotypes of the disease they belong to. We present a Phenotype-driven Likelihood Ratio analysis approach (PheLR) assisting interpretable clinical diagnosis of rare diseases. With a likelihood ratio paradigm, PheLR estimates the posterior probability of candidate diseases and how much a phenotypic feature contributes to the prioritization result. Benchmarked using simulated and realistic patients, PheLR shows significant advantages over current approaches and is robust to noise and inaccuracy. To facilitate clinical practice and visualized differential diagnosis, PheLR is implemented as an online web tool (https://phelr.nbscn.org).

Artificial Intelligence (AI) based diagnosis systems have emerged as powerful tools to reform traditional medical care. Each clinician now wants to have his own intelligent diagnostic partner to expand the range of services he can provide. However, the implementation of intelligent decision support systems based on clinical note has been hindered by the lack of extensibility of end-to-end AI diagnosis algorithms. When reading a clinical note, expert clinicians make inferences with relevant medical knowledge, which serve as prompts for making accurate diagnoses. Therefore, external medical knowledge is commonly employed as an augmentation for medical text classification tasks. Existing methods, however, cannot integrate knowledge from various knowledge sources as prompts nor can fully utilize explicit and implicit knowledge. To address these issues, we propose a Medical Knowledge-enhanced Prompt Learning (MedKPL) diagnostic framework for transferable clinical note classification. Firstly, to overcome the heterogeneity of knowledge sources, such as knowledge graphs or medical QA databases, MedKPL uniform the knowledge relevant to the disease into text sequences of fixed format. Then, MedKPL integrates medical knowledge into the prompt designed for context representation. Therefore, MedKPL can integrate knowledge into the models to enhance diagnostic performance and effectively transfer to new diseases by using relevant disease knowledge. The results of our experiments on two medical datasets demonstrate that our method yields superior medical text classification results and performs better in cross-departmental transfer tasks under few-shot or even zero-shot settings. These findings demonstrate that our MedKPL framework has the potential to improve the interpretability and transferability of current diagnostic systems.

Recent work has shown that predictive models can be applied to structured electronic health record (EHR) data to stratify autism likelihood from an early age (<1 year). Integrating clinical narratives (or notes) with structured data has been shown to improve prediction performance in other clinical applications, but the added predictive value of this information in early autism prediction has not yet been explored. In this study, we aimed to enhance the performance of early autism prediction by using both structured EHR data and clinical narratives. We built models based on structured data and clinical narratives separately, and then an ensemble model that integrated both sources of data. We assessed the predictive value of these models from Duke University Health System over a 14-year span to evaluate ensemble models predicting later autism diagnosis (by age 4 years) from data collected from ages 30 to 360 days. Our sample included 11,750 children above by age 3 years (385 meeting autism diagnostic criteria). The ensemble model for autism prediction showed superior performance and at age 30 days achieved 46.8% sensitivity (95% confidence interval, CI: 22.0%, 52.9%), 28.0% positive predictive value (PPV) at high (90%) specificity (CI: 2.0%, 33.1%), and AUC4 (with at least 4-year follow-up for controls) reaching 0.769 (CI: 0.715, 0.811). Prediction by 360 days achieved 44.5% sensitivity (CI: 23.6%, 62.9%), and 13.7% PPV at high (90%) specificity (CI: 9.6%, 18.9%), and AUC4 reaching 0.797 (CI: 0.746, 0.840). Results show that incorporating clinical narratives in early autism prediction achieved promising accuracy by age 30 days, outperforming models based on structured data only. Furthermore, findings suggest that additional features learned from clinician narratives might be hypothesis generating for understanding early development in autism.

Understanding the actual work (i.e., "work-as-done") rather than theorized work (i.e., "work-as-imagined") during complex medical processes is critical for developing approaches that improve patient outcomes. Although process mining has been used to discover process models from medical activity logs, it often omits critical steps or produces cluttered and unreadable models. In this paper, we introduce a TraceAlignment-based ProcessDiscovery method called TAD Miner to build interpretable process models for complex medical processes. TAD Miner creates simple linear process models using a threshold metric that optimizes the consensus sequence to represent the backbone process, and then identifies both concurrent activities and uncommon-but-critical activities to represent the side branches. TAD Miner also identifies the locations of repeated activities, an essential feature for representing medical treatment steps. We conducted a study using activity logs of 308 pediatric trauma resuscitations to develop and evaluate TAD Miner. TAD Miner was used to discover process models for five resuscitation goals, including establishing intravenous (IV) access, administering non-invasive oxygenation, performing back assessment, administering blood transfusion, and performing intubation. We quantitively evaluated the process models with several complexity and accuracy metrics, and performed qualitative evaluation with four medical experts to assess the accuracy and interpretability of the discovered models. Through these evaluations, we compared the performance of our method to that of two state-of-the-art process discovery algorithms: Inductive Miner and Split Miner. The process models discovered by TAD Miner had lower complexity and better interpretability than the state-of-the-art methods, and the fitness and precision of the models were comparable. We used the TAD process models to identify (1) the errors and (2)the best locations for the tentative steps in knowledge-driven expert models. The knowledge-driven models were revised based on the modifications suggested by the discovered models. The improved modeling using TAD Miner may enhance understanding of complex medical processes.

OBJECTIVE: Genotype imputation is a commonly used technique that infers un-typed variants into a study's genotype data, allowing better identification of causal variants in disease studies. However, due to overrepresentation of Caucasian studies, there's a lack of understanding of genetic basis of health-outcomes in other ethnic populations. Therefore, facilitating imputation of missing key-predictor-variants that can potentially improve a risk health-outcome prediction model, specifically for Asian ancestry, is of utmost relevance. METHODS: We aimed to construct an imputation and analysis web-platform, that primarily facilitates, but is not limited to genotype imputation on East-Asians. The goal is to provide a collaborative imputation platform for researchers in the public domain towards rapidly and efficiently conducting accurate genotype imputation. RESULTS: We present an online genotype imputation platform, Multi-ethnic Imputation System (MI-System) (https://misystem.cgm.ntu.edu.tw/), that offers users 3 established pipelines, SHAPEIT2-IMPUTE2, SHAPEIT4-IMPUTE5, and Beagle5.1 for conducting imputation analyses. In addition to 1000 Genomes and Hapmap3, a new customized Taiwan Biobank (TWB) reference panel, specifically created for Taiwanese-Chinese ancestry is provided. MI-System further offers functions to create customized reference panels to be used for imputation, conduct quality control, split whole genome data into chromosomes, and convert genome builds. CONCLUSION: Users can upload their genotype data and perform imputation with minimum effort and resources. The utility functions further can be utilized to preprocess user uploaded data with easy clicks. MI-System potentially contributes to Asian-population genetics research, while eliminating the requirement for high performing computational resources and bioinformatics expertise. It will enable an increased pace of research and provide a knowledge-base for genetic carriers of complex diseases, therefore greatly enhancing patient-driven research. STATEMENT OF SIGNIFICANCE: Multi-ethnic Imputation System (MI-System), primarily facilitates, but is not limited to, imputation on East-Asians, through 3 established prephasing-imputation pipelines, SHAPEIT2-IMPUTE2, SHAPEIT4-IMPUTE5, and Beagle5.1, where users can upload their genotype data and perform imputation and other utility functions with minimum effort and resources. A new customized Taiwan Biobank (TWB) reference panel, specifically created for Taiwanese-Chinese ancestry is provided. Utility functions include (a) create customized reference panels, (b) conduct quality control, (c) split whole genome data into chromosomes, and (d) convert genome builds. Users can also combine 2 reference panels using the system and use combined panels as reference to conduct imputation using MI-System.

In the healthcare sector, resorting to big data and advanced analytics is a great advantage when dealing with complex groups of patients in terms of comorbidities, representing a significant step towards personalized targeting. In this work, we focus on understanding key features and clinical pathways of patients with multimorbidity suffering from Dementia. This disease can result from many heterogeneous factors, potentially becoming more prevalent as the population ages. We present a set of methods that allow us to identify medical appointment patterns within a cohort of 1924 patients followed from January 2007 to August 2021 in Hospital da Luz (Lisbon), and to stratify patients into subgroups that exhibit similar patterns of interaction. With Markov Chains, we are able to identify the most prevailing medical appointments attended by Dementia patients, as well as recurring transitions between these. To perform patient stratification, we applied AliClu, a temporal sequence alignment algorithm for clustering longitudinal clinical data, which allowed us to successfully identify patient subgroups with similar medical appointment activity. A feature analysis per cluster obtained allows the identification of distinct patterns and characteristics. This pipeline provides a tool to identify prevailing clinical pathways of medical appointments within the dataset, as well as the most common transitions between medical specialities within Dementia patients. This methodology, alongside demographic and clinical data, has the potential to provide early signalling of the most likely clinical pathways and serve as a support tool for health providers in deciding the best course of treatment, considering a patient as a whole.

Cytokine release syndrome (CRS), also known as cytokine storm, is one of the most consequential adverse effects of chimeric antigen receptor therapies that have shown otherwise promising results in cancer treatment. When emerging, CRS could be identified by the analysis of specific cytokine and chemokine profiles that tend to exhibit similarities across patients. In this paper, we exploit these similarities using machine learning algorithms and set out to pioneer a meta-review informed method for the identification of CRS based on specific cytokine peak concentrations and evidence from previous clinical studies. To this end we also address a widespread challenge of the applicability of machine learning in general: reduced training data availability. We do so by augmenting available (but often insufficient) patient cytokine concentrations with statistical knowledge extracted from domain literature. We argue that such methods could support clinicians in analyzing suspect cytokine profiles by matching them against the said CRS knowledge from past clinical studies, with the ultimate aim of swift CRS diagnosis. We evaluate our proposed methods under several design choices, achieving performance of more than 90% in terms of CRS identification accuracy, and showing that many of our choices outperform a purely data-driven alternative. During evaluation with real-world CRS clinical data, we emphasize the potential of our proposed method of producing interpretable results, in addition to being effective in identifying the onset of cytokine storm.

In recent years, the high-resolution manometry (HRM) technique has been increasingly used to study esophageal and colonic pressurization and has become a standard routine for discovering mobility disorders. In addition to evolving guidelines for the interpretation of HRM like Chicago standard, some complexities, such as the dependency of normative reference values on the recording device and other external variables, still remain for medical professions. In this study, a decision support framework is developed to aid the diagnosis of esophageal mobility disorders based on HRM data. To abstract HRM data, Spearman correlation is employed to model the spatio-temporal dependencies of pressure values of HRM components and convolutional graph neural networks are then utilized to embed relation graphs to the features vector. In the decision-making stage, a novel Expert per Class Fuzzy Classifier (EPC-FC) is presented that employs the ensemble structure and contains expertized sub-classifiers for recognizing a specific disorder. Training sub-classifiers using the negative correlation learning method makes the EPC-FC highly generalizable. Meanwhile, separating the sub-classifiers of each class gives flexibility and interpretability to the structure. The suggested framework is evaluated on a dataset of 67 patients in 5 different classes recorded in Shariati Hospital. The average accuracy of 78.03% for a single swallow and 92.54% for subject-level is achieved for distinguishing mobility disorders. Moreover, compared with the other studies, the presented framework has an outstanding performance considering that it imposes no limits on the type of classes or HRM data. On the other hand, the EPC-FC outperforms other comparative classifiers such as SVM and AdaBoost not only in HRM diagnosis but also on other benchmark classification problems.

The degree of motor impairment and profile of recovery after stroke are difficult to predict for each individual. Measures obtained from clinical assessments, as well as neurophysiological and neuroimaging techniques have been used as potential biomarkers of motor recovery, with limited accuracy up to date. To address this, the present study aimed to develop a deep learning model based on structural brain images obtained from stroke participants and healthy volunteers. The following inputs were used in a multi-channel 3D convolutional neural network (CNN) model: fractional anisotropy, mean diffusivity, radial diffusivity, and axial diffusivity maps obtained from Diffusion Tensor Imaging (DTI) images, white and gray matter intensity values obtained from Magnetic Resonance Imaging, as well as demographic data (e.g., age, gender). Upper limb motor function was classified into "Poor" and "Good" categories. To assess the performance of the DL model, we compared it to more standard machine learning (ML) classifiers including k-nearest neighbor, support vector machines (SVM), Decision Trees, Random Forests, Ada Boosting, and Naïve Bayes, whereby the inputs of these classifiers were the features taken from the fully connected layer of the CNN model. The highest accuracy and area under the curve values were 0.92 and 0.92 for the 3D-CNN and 0.91 and 0.91 for the SVM, respectively. The multi-channel 3D-CNN with residual blocks and SVM supported by DL was more accurate than traditional ML methods to classify upper limb motor impairment in the stroke population. These results suggest that combining volumetric DTI maps and measures of white and gray matter integrity can improve the prediction of the degree of motor impairment after stroke. Identifying the potential of recovery early on after a stroke could promote the allocation of resources to optimize the functional independence of these individuals and their quality of life.

Unpacking and comprehending how black-box machine learning algorithms (such as deep learning models) make decisions has been a persistent challenge for researchers and end-users. Explaining time-series predictive models is useful for clinical applications with high stakes to understand the behavior of prediction models, e.g., to determine how different variables and time points influence the clinical outcome. However, existing approaches to explain such models are frequently unique to architectures and data where the features do not have a time-varying component. In this paper, we introduce WindowSHAP, a model-agnostic framework for explaining time-series classifiers using Shapley values. We intend for WindowSHAP to mitigate the computational complexity of calculating Shapley values for long time-series data as well as improve the quality of explanations. WindowSHAP is based on partitioning a sequence into time windows. Under this framework, we present three distinct algorithms of Stationary, Sliding and Dynamic WindowSHAP, each evaluated against baseline approaches, KernelSHAP and TimeSHAP, using perturbation and sequence analyses metrics. We applied our framework to clinical time-series data from both a specialized clinical domain (Traumatic Brain Injury - TBI) as well as a broad clinical domain (critical care medicine). The experimental results demonstrate that, based on the two quantitative metrics, our framework is superior at explaining clinical time-series classifiers, while also reducing the complexity of computations. We show that for time-series data with 120 time steps (hours), merging 10 adjacent time points can reduce the CPU time of WindowSHAP by 80 % compared to KernelSHAP. We also show that our Dynamic WindowSHAP algorithm focuses more on the most important time steps and provides more understandable explanations. As a result, WindowSHAP not only accelerates the calculation of Shapley values for time-series data, but also delivers more understandable explanations with higher quality.

The emphasis on fairness in predictive healthcare modeling has increased in popularity as an approach for overcoming biases in automated decision-making systems. The aim is to guarantee that sensitive characteristics like gender, race, and ethnicity do not influence prediction outputs. Numerous algorithmic strategies have been proposed to reduce bias in prediction results, mitigate prejudice toward minority groups and promote prediction fairness. The goal of these strategies is to ensure that model prediction performance does not exhibit significant disparity among sensitive groups. In this study, we propose a novel fairness-achieving scheme based on multitask learning, which fundamentally differs from conventional fairness-achieving techniques, including altering data distributions and constraint optimization through regularizing fairness metrics or tampering with prediction outcomes. By dividing predictions on different sub-populations into separate tasks, we view the fairness problem as a task-balancing problem. To ensure fairness during the model-training process, we suggest a novel dynamic re-weighting approach. Fairness is achieved by dynamically modifying the gradients of various prediction tasks during neural network back-propagation, and this novel technique applies to a wide range of fairness criteria. We conduct tests on a real-world use case to predict sepsis patients' mortality risk. Our approach satisfies that it can reduce the disparity between subgroups by 98% while only losing less than 4% of prediction accuracy.

Inferring knowledge from known relationships between drugs, proteins, genes, and diseases has great potential for clinical impact, such as predicting which existing drugs could be repurposed to treat rare diseases. Incorporating key biological context such as cell type or tissue of action into representations of extracted biomedical knowledge is essential for principled pharmacological discovery. Existing global, literature-derived knowledge graphs of interactions between drugs, proteins, genes, and diseases lack this essential information. In this study, we frame the task of associating biological context with protein-protein interactions extracted from text as a classification task using syntactic, semantic, and novel meta-discourse features. We introduce the Insider corpora, which are automatically generated PubMed-scale corpora for training classifiers for the context association task. These corpora are created by searching for precise syntactic cues of cell type and tissue relevancy to extracted regulatory relations. We report F1 scores of 0.955 and 0.862 for identifying relevant cell types and tissues, respectively, for our identified relations. By classifying with this framework, we demonstrate that the problem of context association can be addressed using intuitive, interpretable features. We demonstrate the potential of this approach to enrich text-derived knowledge bases with biological detail by incorporating cell type context into a protein-protein network for dengue fever.

OBJECTIVE: Electronic health records (EHR), containing detailed longitudinal clinical information on a large number of patients and covering broad patient populations, open opportunities for comprehensive predictive modeling of disease progression and treatment response. However, since EHRs were originally constructed for administrative purposes not for research, in the EHR-linked studies, it is often not feasible to capture reliable information for analytical variables, especially in the survival setting, when both accurate event status and event times are needed for model building. For example, progression-free survival (PFS), a commonly used survival outcome for cancer patients, often involves complex information embedded in free-text clinical notes and cannot be extracted reliably. Proxies of PFS time such as time to the first mention of progression in the notes are at best good approximations to the true event time. This leads to difficulty in efficiently estimating event rates for an EHR patient cohort. Estimating survival rates based on error-prone outcome definitions can lead to biased results and hamper the power in the downstream analysis. On the other hand, extracting accurate event time information via manual annotation is time and resource intensive. The objective of this study is to develop a calibrated survival rate estimator using noisy outcomes from EHR data. MATERIALS AND METHODS: In this paper, we propose a two-stage semi-supervised calibration of noisy event rate (SCANER) estimator that can effectively overcome censoring induced dependency and attains more robust performance (i.e., not sensitive to misspecification of the imputation model) by fully utilizing both a small-labeled set of gold-standard survival outcomes annotated via manual chart review and a set of proxy features automatically captured via EHR in the unlabeled set. We validate the SCANER estimator by estimating the PFS rates for a virtual cohort of lung cancer patients from one large tertiary care center and the ICU-free survival rates for COVID patients from two large tertiary care centers. RESULTS: In terms of survival rate estimates, the SCANER had very similar point estimates compared to the complete-case Kaplan Meier estimator. On the other hand, other benchmark methods for comparison, which fail to account for the induced dependency between event time and the censoring time conditioning on surrogate outcomes, produced biased results across all three case studies. In terms of standard errors, the SCANER estimator was more efficient than the KM estimator, with up to 50% efficiency gain. CONCLUSION: The SCANER estimator achieves more efficient, robust, and accurate survival rate estimates compared to existing approaches. This promising new approach can also improve the resolution (i.e., granularity of event time) by using labels conditioning on multiple surrogates, particularly among less common or poorly coded conditions.

OBJECTIVE: To determine whether graph neural network based models of electronic health records can predict specialty consultation care needs for endocrinology and hematology more accurately than the standard of care checklists and other conventional medical recommendation algorithms in the literature. METHODS: Demand for medical expertise far outstrips supply, with tens of millions in the US alone with deficient access to specialty care. Rather than potentially months long delays to initiate diagnostic workup and medical treatment with a specialist, referring primary care supported by an automated recommender algorithm could anticipate and directly initiate patient evaluation that would otherwise be needed at subsequent a specialist appointment. We propose a novel graph representation learning approach with a heterogeneous graph neural network to model structured electronic health records and formulate recommendation/prediction of subsequent specialist orders as a link prediction problem. RESULTS: Models are trained and assessed in two specialty care sites: endocrinology and hematology. Our experimental results show that our model achieves an 8% improvement in ROC-AUC for endocrinology (ROC-AUC = 0.88) and 5% improvement for hematology (ROC-AUC = 0.84) personalized procedure recommendations over prior medical recommender systems. These recommender algorithm approaches provide medical procedure recommendations for endocrinology referrals more effectively than manual clinical checklists (recommender: precision = 0.60, recall = 0.27, F1-score = 0.37) vs. (checklist: precision = 0.16, recall = 0.28, F1-score = 0.20), and similarly for hematology referrals (recommender: precision = 0.44, recall = 0.38, F1-score = 0.41) vs. (checklist: precision = 0.27, recall = 0.71, F1-score = 0.39). CONCLUSION: Embedding graph neural network models into clinical care can improve digital specialty consultation systems and expand the access to medical experience of prior similar cases.

Accurate and efficient extraction of key information related to diseases from medical examination reports, such as X-ray and ultrasound images, CT scans, and others, is crucial for accurate diagnosis and treatment. These reports provide a detailed record of a patient's health condition and are an important part of the clinical examination process. By organizing this information in a structured way, doctors can more easily review and analyze the data, leading to better patient care. In this paper, we introduce a new technique for extracting useful information from unstructured clinical text examination reports, which we refer to as a medical event extraction (EE) task. Our approach is based on Machine Reading Comprehension (MRC) and involves two sub-tasks: Question Answerability Judgment (QAJ) and Span Selection (SS). We use BERT to build a question answerability discriminator (Judger) that determines whether a reading comprehension question can be answered or not, thereby avoiding the extraction of arguments from unanswerable questions. The SS sub-task first obtains the encoding of each word in the medical text from the final layer of BERT's Transformer, then utilizes the attention mechanism to identify important information related to the answer from these word encodings. This information is then input into a bidirectional LSTM (BiLSTM) module to obtain a global representation of the text, which is used, along with the softmax function, to predict the span of the answer (i.e., the start and end positions of the answer in the text report). We use interpretable methods to calculate the Jensen-Shannon Divergence (JSD) score between various layers of the network and confirm that our model has strong word representation capabilities, enabling it to effectively extract contextual information from medical reports. Our experiments demonstrate that our method outperforms existing medical event extraction methods, achieving state-of-the-art results with a notable F1 score.

In recent decades, cardiovascular disease (CVD) has become the leading cause of death in most countries of the world. Since many types of CVD are preventable by modifying lifestyle behaviors, the objective of this paper is to develop an effective personalized lifestyle recommendation algorithm for reducing the risk of common types of CVD. However, in practice, the underlying relationships between the risk factors (e.g., lifestyles, blood pressure, etc.) and disease onset is highly complex. It is also challenging to identify effective modification recommendations for different individuals due to individual's effort-benefits consideration and uncertainties in disease progression. Therefore, to address these challenges, this study developed a novel data-driven approach for personalized lifestyle behaviors recommendation based on machine learning and a personalized exponential utility function model. The contributions of this work can be summarized into three aspects: (1) a classification-based prediction model is implemented to predict the CVD risk based on the condition of risk factors; (2) the generative adversarial network (GAN) is incorporated to learn the underlying relationship between risk factors, as well as quantify the uncertainty of disease progression under lifestyle modifications; and (3) a novel personalized exponential utility function model is proposed to evaluate the modifications' utilities with respect to CVD risk reduction, individual's effort-benefits consideration, and disease progression uncertainty, as well as identify the optimal modification for each individual. The effectiveness of the proposed method is validated through an open-access CVD dataset. The results demonstrate that the personalized lifestyle modification recommended by the proposed methodology has the potential to effectively reduce the CVD risk. Thus, it is promising to be further applied to real-world cases for CVD prevention.

OBJECTIVE: Trust determines the success of Health-Behavior-Change Artificial Intelligence Apps (HBC-AIApp). Developers of such apps need theory-based practical methods that can guide them in achieving such trust. Our study aimed to develop a comprehensive conceptual model and development process that can guide developers how to build HBC-AIApp in order to support trust creation among the app's users. METHODS: We apply a multi-disciplinary approach where medical informatics, human-centered design, and holistic health methods are integrated to address the trust challenge in HBC-AIApps. The integration extends a conceptual model of trust in AI developed by Jermutus et al., whose properties guide the extension of the IDEAS (integrate, design, assess, and share) HBC-App development process. RESULTS: The HBC-AIApp framework consists of three main blocks: (1) system development methods that study the users' complex reality, hence, their perceptions, needs, goals and environment; (2) mediators and other stakeholders who are important for developing and operating the HBC-AIApp, boundary objects that examine users' activities via the HBC-AIApp; and (3) the HBC-AIApp's structural components, AI logic, and physical implementation. These blocks come together to provide the extended conceptual model of trust in HBC-AIApps and the extended IDEAS process. DISCUSSION: The developed HBC-AIApp framework drew from our own experience in developing trust in HBC-AIApp. Further research will focus on studying the application of the proposed comprehensive HBC-AIApp development framework and whether applying it supports trust creation in such apps.

OBJECTIVE: Clinical work involves performing overlapping, time-sensitive tasks that frequently require clinicians to switch their attention between multiple tasks. We developed a methodological approach using EHR-based audit logs to determine switch costs-the cognitive burden associated with task switching-and assessed its magnitude during routine EHR-based clinical tasks. METHOD: Physician trainees (N = 75) participated in a longitudinal study where they provided access to their EHR-based audit logs. Physicians' audit log actions were used to create a taxonomy of EHR tasks. These tasks were transformed into task sequences and the time spent on each task in a sequence was computed. Within these task sequences, instances of task switching (i.e., switching from one task to the next) and non-switching were identified. The primary outcome of interest was the time spent on a post-switch task. Using a mixed-effects regression model, we compared the durations of post-switch and non-switch tasks. RESULTS: 2,781,679 audit log events over 117,822 sessions from 75 physicians were analyzed. Physicians spent most time on chart review (Median (IQR) = 5,439 (2,492-8,336) seconds), note review (1,936 (827-3,321) seconds), and navigating the EHR interface (1,048 (365.5-2,006) seconds) daily. Post task switch activity times were greater for documentation (Median increase = 5 s), order entry (Median increase = 3 s) and results review (Median increase = 3 s). Mixed-effects regression showed that time spent on tasks were longer following a task switch (β = 0.03; 95% CIlower = 0.027, CIupper = 0.034), with greater post-swtich task times for imaging, order entry, note review, handoff, note entry, chart review and best practice advisory tasks. DISCUSSION: Increased task switching time-an indicator of the cognitive burden associated with switching between tasks-is prevalent in routine EHR-based tasks. We discuss the cumulative impact of incremental switch costs have on overall EHR workload, wellness, and error rates. Relying on theoretical cognitive foundations, we suggest pragmatic design considerations for mitigating the effects of cognitive burden associated with task switching.

Linking data across studies offers an opportunity to enrich data sets and provide a stronger basis for data-driven models for biomedical discovery and/or prognostication. Several techniques to link records have been proposed, and some have been implemented across data repositories holding molecular and clinical data. Not all these techniques guarantee appropriate privacy protection; there are trade-offs between (a) simple strategies that can be associated with data that will be linked and shared with any party and (b) more complex strategies that preserve the privacy of individuals across parties. We propose an intermediary, practical strategy to support linkage in studies that share de-identified data with Data Coordinating Centers. This technology can be extended to link data across multiple data hubs to support privacy preserving record linkage, considering data coordination centers and their awardees, which can be extended to a hierarchy of entities (e.g., awardees, data coordination centers, data hubs, etc.) b.

Influenza viruses pose great threats to public health and cause enormous economic losses every year. Previous work has revealed the viral factors associated with the virulence of influenza viruses in mammals. However, taking prior viral knowledge represented by heterogeneous categorical and discrete information into account to explore virus virulence is scarce in the existing work. How to make full use of the preceding domain knowledge in virulence study is challenging but beneficial. This paper proposes a general framework named ViPal for virulence prediction in mice that incorporates discrete prior viral mutation and reassortment information based on all eight influenza segments. The posterior regularization technique is leveraged to transform prior viral knowledge into constraint features and integrated into the machine learning models. Experimental results on influenza genomic datasets validate that our proposed framework can improve virulence prediction performance over baselines. The comparison between ViPal and other existing methods shows the computational efficiency of our framework with comparable or superior performance. Moreover, the interpretable analysis through SHAP (SHapley Additive exPlanations) identifies the scores of constraint features contributing to the prediction. We hope this framework could provide assistance for the accurate detection of influenza virulence and facilitate flu surveillance.

The Biomedical Research field is currently advancing to develop Clinical Trials and translational projects based on Real World Evidence. To make this transition feasible, clinical centers need to work toward Data Accessibility and Interoperability. This task is particularly challenging when applied to Genomics, that entered in routinary screening in the last years via mostly amplicon-based Next-Generation Sequencing panels. Said experiments produce up to hundreds of features per patient, and their summarized results are often stored in static clinical reports, making critical information inaccessible to automated access and Federated Search consortia. In this study, we present a reanalysis of 4620 solid tumor sequencing samples in five different histology settings. Furthermore, we describe all the Bioinformatics and Data Engineering processes that were put in place in order to create a Somatic Variant Registry able to deal with the large biotechnological variability of routinary Genomics Profiling.

Transforming raw EHR data into machine learning model-ready inputs requires considerable effort. One widely used EHR database is Medical Information Mart for Intensive Care (MIMIC). Prior work on MIMIC-III cannot query the updated and improved MIMIC-IV version. Besides, the need to use multicenter datasets further highlights the challenge of EHR data extraction. Therefore, we developed an extraction pipeline that works on both MIMIC-IV and eICU Collaborative Research Database and allows for model cross validation using these 2 databases. Under the default choices, the pipeline extracted 38,766 and 126,448 ICU records for MIMIC-IV and eICU, respectively. Using the extracted time-dependent variables, we compared the Area Under the Curve (AUC) performance with ​​prior works on clinically relevant tasks such as in-hospital mortality prediction. METRE achieved comparable performance with AUC 0.723-0.888 across all tasks with MIMIC-IV. Additionally, when we evaluated the model directly on MIMIC-IV data using a model trained on eICU, we observed that the AUC change can be as small as +0.019 or -0.015. Our open-source pipeline transforms MIMIC-IV and eICU into structured data frames and allows researchers to perform model training and testing using data collected from different institutions, which is of critical importance for model deployment under clinical contexts. The code used to extract the data and perform training is available here: https://github.com/weiliao97/METRE.

A causal effect can be defined as a comparison of outcomes that result from two or more alternative actions, with only one of the action-outcome pairs actually being observed. In healthcare, the gold standard for causal effect measurements is randomized controlled trials (RCTs), in which a target population is explicitly defined and each study sample is randomly assigned to either the treatment or control cohorts. The great potential to derive actionable insights from causal relationships has led to a growing body of machine-learning research applying causal effect estimators to observational data in the fields of healthcare, education, and economics. The primary difference between causal effect studies utilizing observational data and RCTs is that for observational data, the study occurs after the treatment, and therefore we do not have control over the treatment assignment mechanism. This can lead to massive differences in covariate distributions between control and treatment samples, making a comparison of causal effects confounded and unreliable. Classical approaches have sought to solve this problem piecemeal, first by predicting treatment assignment and then treatment effect separately. Recent work extended part of these approaches to a new family of representation-learning algorithms, showing that the upper bound of the expected treatment effect estimation error is determined by two factors: the outcome generalization-error of the representation and the distance between the treated and control distributions induced by the representation. To achieve minimal dissimilarity in learning such distributions, in this work we propose a specific auto-balancing, self-supervised objective. Experiments on real and benchmark datasets revealed that our approach consistently produced less biased estimates than previously published state-of-the-art methods. We demonstrate that the reduction in error can be directly attributed to the ability to learn representations that explicitly reduce such dissimilarity; further, in case of violations of the positivity assumption (frequent in observational data), we show our approach performs significantly better than the previous state of the art. Thus, by learning representations that induce similar distributions of the treated and control cohorts, we present evidence to support the error bound dissimilarity hypothesis as well as providing a new state-of-the-art model for causal effect estimation.

OBJECTIVE: We explore the framing of literature-based discovery (LBD) as link prediction and graph embedding learning, with Alzheimer's Disease (AD) as our focus disease context. The key link prediction setting of prediction window length is specifically examined in the context of a time-sliced evaluation methodology. METHODS: We propose a four-stage approach to explore literature-based discovery for Alzheimer's Disease, creating and analyzing a knowledge graph tailored to the AD context, and predicting and evaluating new knowledge based on time-sliced link prediction. The first stage is to collect an AD-specific corpus. The second stage involves constructing an AD knowledge graph with identified AD-specific concepts and relations from the corpus. In the third stage, 20 pairs of training and testing datasets are constructed with the time-slicing methodology. Finally, we infer new knowledge with graph embedding-based link prediction methods. We compare different link prediction methods in this context. The impact of limiting prediction evaluation of LBD models in the context of short-term and longer-term knowledge evolution for Alzheimer's Disease is assessed. RESULTS: We constructed an AD corpus of over 16 k papers published in 1977-2021, and automatically annotated it with concepts and relations covering 11 AD-specific semantic entity types. The knowledge graph of Alzheimer's Disease derived from this resource consisted of ∼11 k nodes and ∼394 k edges, among which 34% were genotype-phenotype relationships, 57% were genotype-genotype relationships, and 9% were phenotype-phenotype relationships. A Structural Deep Network Embedding (SDNE) model consistently showed the best performance in terms of returning the most confident set of link predictions as time progresses over 20 years. A huge improvement in model performance was observed when changing the link prediction evaluation setting to consider a more distant future, reflecting the time required for knowledge accumulation. CONCLUSION: Neural network graph-embedding link prediction methods show promise for the literature-based discovery context, although the prediction setting is extremely challenging, with graph densities of less than 1%. Varying prediction window length on the time-sliced evaluation methodology leads to hugely different results and interpretations of LBD studies. Our approach can be generalized to enable knowledge discovery for other diseases. AVAILABILITY: Code, AD ontology, and data are available at https://github.com/READ-BioMed/readbiomed-lbd.

A large amount of personal health data that is highly valuable to the scientific community is still not accessible or requires a lengthy request process due to privacy concerns and legal restrictions. As a solution, synthetic data has been studied and proposed to be a promising alternative to this issue. However, generating realistic and privacy-preserving synthetic personal health data retains challenges such as simulating the characteristics of the patients' data that are in the minority classes, capturing the relations among variables in imbalanced data and transferring them to the synthetic data, and preserving individual patients' privacy. In this paper, we propose a differentially private conditional Generative Adversarial Network model (DP-CGANS) consisting of data transformation, sampling, conditioning, and network training to generate realistic and privacy-preserving personal data. Our model distinguishes categorical and continuous variables and transforms them into latent space separately for better training performance. We tackle the unique challenges of generating synthetic patient data due to the special data characteristics of personal health data. For example, patients with a certain disease are typically the minority in the dataset and the relations among variables are crucial to be observed. Our model is structured with a conditional vector as an additional input to present the minority class in the imbalanced data and maximally capture the dependency between variables. Moreover, we inject statistical noise into the gradients in the networking training process of DP-CGANS to provide a differential privacy guarantee. We extensively evaluate our model with state-of-the-art generative models on personal socio-economic datasets and real-world personal health datasets in terms of statistical similarity, machine learning performance, and privacy measurement. We demonstrate that our model outperforms other comparable models, especially in capturing the dependence between variables. Finally, we present the balance between data utility and privacy in synthetic data generation considering the different data structures and characteristics of real-world personal health data such as imbalanced classes, abnormal distributions, and data sparsity.

The widespread adoption of effective hybrid closed loop systems would represent an important milestone of care for people living with type 1 diabetes (T1D). These devices typically utilise simple control algorithms to select the optimal insulin dose for maintaining blood glucose levels within a healthy range. Online reinforcement learning (RL) has been utilised as a method for further enhancing glucose control in these devices. Previous approaches have been shown to reduce patient risk and improve time spent in the target range when compared to classical control algorithms, but are prone to instability in the learning process, often resulting in the selection of unsafe actions. This work presents an evaluation of offline RL for developing effective dosing policies without the need for potentially dangerous patient interaction during training. This paper examines the utility of BCQ, CQL and TD3-BC in managing the blood glucose of the 30 virtual patients available within the FDA-approved UVA/Padova glucose dynamics simulator. When trained on less than a tenth of the total training samples required by online RL to achieve stable performance, this work shows that offline RL can significantly increase time in the healthy blood glucose range from 61.6±0.3% to 65.3±0.5% when compared to the strongest state-of-art baseline (p<0.001). This is achieved without any associated increase in low blood glucose events. Offline RL is also shown to be able to correct for common and challenging control scenarios such as incorrect bolus dosing, irregular meal timings and compression errors. The code for this work is available at: https://github.com/hemerson1/offline-glucose.

Predicting the patient's in-hospital mortality from the historical Electronic Medical Records (EMRs) can assist physicians to make clinical decisions and assign medical resources. In recent years, researchers proposed many deep learning methods to predict in-hospital mortality by learning patient representations. However, most of these methods fail to comprehensively learn the temporal representations and do not sufficiently mine the contextual knowledge of demographic information. We propose a novel end-to-end approach based on Local and Global Temporal Representation Learning with Demographic Embedding (LGTRL-DE) to address the current issues for in-hospital mortality prediction. LGTRL-DE is enabled by (1) a local temporal representation learning module that captures the temporal information and analyzes the health status from a local perspective through a recurrent neural network with the demographic initialization and the local attention mechanism; (2) a Transformer-based global temporal representation learning module that extracts the interaction dependencies among clinical events; (3) a multi-view representation fusion module that fuses temporal and static information and generates the final patient's health representations. We evaluate our proposed LGTRL-DE on two public real-world clinical datasets (MIMIC-III and e-ICU). Experimental results show that LGTRL-DE achieves area under receiver operating characteristic curve of 0.8685 and 0.8733 on the MIMIC-III and e-ICU datasets, respectively, outperforming several state-of-the-art approaches.

In recent years, interest and investment in health and mental health smartphone apps have grown significantly. However, this growth has not been followed by an increase in quality and the incorporation of more advanced features in such applications. This can be explained by an expanding fragmentation of existing mobile platforms along with more restrictive privacy and battery consumption policies, with a consequent higher complexity of developing such smartphone applications. To help overcome these barriers, there is a need for robust, well-designed software development frameworks which are designed to be reliable, power-efficient and ethical with respect to data collection practices, and which support the sense-analyse-act paradigm typically employed in reactive mHealth applications. In this article, we present the AwarNS Framework, a context-aware modular software development framework for Android smartphones, which facilitates transparent, reliable, passive and active data sampling running in the background (sense), on-device and server-side data analysis (analyse), and context-aware just-in-time offline and online intervention capabilities (act). It is based on the principles of versatility, reliability, privacy, reusability, and testability. It offers built-in modules for capturing smartphone and associated wearable sensor data (e.g. IMU sensors, geolocation, Wi-Fi and Bluetooth scans, physical activity, battery level, heart rate), analysis modules for data transformation, selection and filtering, performing geofencing analysis and machine learning regression and classification, and act modules for persistence and various notification deliveries. We describe the framework's design principles and architecture design, explain its capabilities and implementation, and demonstrate its use at the hand of real-life case studies implementing various mobile interventions for different mental disorders used in clinical practice.

OBJECTIVE: International Classification of Disorders version 10 (ICD-10) codes contribute heavily to healthcare data. Medicare claims and other data-sources are used to constitute study populations and appraise healthcare processes. How variability in claims-per-beneficiary impacts diagnostic determinations is inadequately understood. The objective of this study is so assess distributional properties of Medicare claims, and examine claim rates impact on code utilization and rate determinations. METHODS: The study population was Medicare beneficiaries aged 75-79.99 with claim(s) in the 5% standard analytical Carrier and Outpatient files, alive and participating in Medicare part B for all 12 months of 2017. Medicare beneficiary files were processed to create records containing all ICD-10 codes specified, key demographics, Part B and vital status, and the total claims for each 2017 beneficiary. Claim number cohorts were characterized. RESULTS: Beneficiaries meeting inclusion criteria totaled 221,625, these having 7,617,503 claims; 96.4% had between 1 and 120 claims. Median claims were 24 for males (females 25); modal claims were 11 (13). Average distinct codes per beneficiary increased with claims number. The assignment of ICD-10 codes, i.e., 'diagnostic rate estimates' (DRE), increased as claim numbers increased for most codes among those most commonly utilized. For some conditions, mostly benign and age-related, DREs plateaued as claim numbers increased. For other conditions, typically associated with clinical acuity, e.g., chest pain, DREs increased steeply with claims. CONCLUSIONS: Older adult Medicare beneficiaries aged 75-80 exhibited varying claims activity over the course of a year. Although DRE dependence on claim numbers varies across ICD-10 codes, rate estimates are higher for beneficiaries with claim numbers above the median.

OBJECTIVES: To examine the feasibility of promoting engagement with data-driven self-management of health among individuals from minoritized medically underserved communities by tailoring the design of self-management interventions to individuals' type of motivation and regulation in accordance with the Self-Determination Theory. METHODS: Fifty-three individuals with type 2 diabetes from an impoverished minority community were randomly assigned to four different versions of an mHealth app for data-driven self-management with the focus on nutrition, Platano; each version was tailored to a specific type of motivation and regulation within the SDT self-determination continuum. These versions included financial rewards (external regulation), feedback from expert registered dietitians (RDF, introjected regulation), self-assessment of attainment of one's nutritional goals (SA, identified regulation), and personalized meal-time nutrition decision support with post-meal blood glucose forecasts (FORC, integrated regulation). We used qualitative interviews to examine interaction between participants' experiences with the app and their motivation type (internal-external). RESULTS: As hypothesized, we found a clear interaction between the type of motivation and Platano features that users responded to and benefited from. For example, those with more internal motivation reported more positive experience with SA and FORC than those with more external motivation. However, we also found that Platano features that aimed to specifically address the needs of individuals with external regulation did not create the desired experience. We attribute this to a mismatch in emphasis on informational versus emotional support, particularly evident in RDF. In addition, we found that for participants recruited from an economically disadvantaged community, internal factors, such as motivation and regulation, interacted with external factors, most notably with limited health literacy and limited access to resources. CONCLUSIONS: The study suggests feasibility of using SDT to tailor design of mHealth interventions for promoting data-driven self-management to individuals' motivation and regulation. However, further research is needed to better align design solutions with different levels of self-determination continuum, to incorporate stronger emphasis on emotional support for individuals with external regulation, and to address unique needs and challenges of underserved communities, with particular attention to limited health literacy and access to resources.

Biological networks are known to be highly modular, and the dysfunction of network modules may cause diseases. Defining the key modules from the omics data and establishing the classification model is helpful in promoting the research of disease diagnosis and prognosis. However, for applying modules in downstream analysis such as disease states discrimination, most methods only utilize the node information, and ignore the node interactions or topological information, which may lead to false positives and limit the model performance. In this study, we propose an omics data analysis method based on feature linear relationship and graph convolutional network (LCNet). In LCNet, we adopt a way of applying the difference of feature linear relationships during disease development to characterize physiological and pathological changes and construct the differential linear relation network, which is simple and interpretable from the perspective of feature linear relationship. A greedy strategy is developed for searching the highly interactive modules with a strong discrimination ability. To fully utilize the information of the detected modules, the personalized sub-graphs for each sample based on the modules are defined, and the graph convolutional network (GCN) classifiers are trained to predict the sample labels. The experimental results on public datasets show the superiority of LCNet in classification performance. For Breast Cancer metabolic data, the identified metabolites by LCNet involve important pathways. Thus, LCNet can identify the module biomarkers by feature linear relationship and a greedy strategy, and label samples by personalized sub-graphs and GCN. It provides a new manner of utilizing node (molecule) information and topological information in the defined modules for better disease classification.

Physician progress notes are frequently organized into Subjective, Objective, Assessment, and Plan (SOAP) sections. The Assessment section synthesizes information recorded in the Subjective and Objective sections, and the Plan section documents tests and treatments to narrow the differential diagnosis and manage symptoms. Classifying the relationship between the Assessment and Plan sections has been suggested to provide valuable insight into clinical reasoning. In this work, we use a novel human-in-the-loop pipeline to classify the relationships between the Assessment and Plan sections of SOAP notes as a part of the n2c2 2022 Track 3 Challenge. In particular, we use a clinical information model constructed from both the entailment logic expected from the aforementioned Challenge and the problem-oriented medical record. This information model is used to label named entities as primary and secondary problems/symptoms, events and complications in all four SOAP sections. We iteratively train separate Named Entity Recognition models and use them to annotate entities in all notes/sections. We fine-tune a downstream RoBERTa-large model to classify the Assessment-Plan relationship. We evaluate multiple language model architectures, preprocessing parameters, and methods of knowledge integration, achieving a maximum macro-F1 score of 82.31%. Our initial model achieves top-2 performance during the challenge (macro-F1: 81.52%, competitors' macro-F1 range: 74.54%-82.12%). We improved our model by incorporating post-challenge annotations (S&O sections), outperforming the top model from the Challenge. We also used Shapley additive explanations to investigate the extent of language model clinical logic, under the lens of our clinical information model. We find that the model often uses shallow heuristics and nonspecific attention when making predictions, suggesting language model knowledge integration requires further research.

The article presents a workflow to create a question-answering system whose knowledge base combines knowledge graphs and scientific publications on coronaviruses. It is based on the experience gained in modeling evidence from research articles to provide answers to questions in natural language. The work contains best practices for acquiring scientific publications, tuning language models to identify and normalize relevant entities, creating representational models based on probabilistic topics, and formalizing an ontology that describes the associations between domain concepts supported by the scientific literature. All the resources generated in the domain of coronavirus are available openly as part of the Drugs4COVID initiative, and can be (re)-used independently or as a whole. They can be exploited by scientific communities conducting research related to SARS-CoV-2/COVID-19 and also by therapeutic communities, laboratories, etc., wishing to find and understand relationships between symptoms, drugs, active ingredients and their documentary evidence.

Brief Hospital Course (BHC) summaries are succinct summaries of an entire hospital encounter, embedded within discharge summaries, written by senior clinicians responsible for the overall care of a patient. Methods to automatically produce summaries from inpatient documentation would be invaluable in reducing clinician manual burden of summarising documents under high time-pressure to admit and discharge patients. Automatically producing these summaries from the inpatient course, is a complex, multi-document summarisation task, as source notes are written from various perspectives (e.g. nursing, doctor, radiology), during the course of the hospitalisation. We demonstrate a range of methods for BHC summarisation demonstrating the performance of deep learning summarisation models across extractive and abstractive summarisation scenarios. We also test a novel ensemble extractive and abstractive summarisation model that incorporates a medical concept ontology (SNOMED) as a clinical guidance signal and shows superior performance in 2 real-world clinical data sets.

The last three years have been a game changer in the way medicine is practiced. The COVID-19 pandemic changed the obstetrics and gynecology scenery. Pregnancy complications, and even death, are preventable due to maternal-fetal monitoring. A fast and accurate diagnosis can be established by a doctor + Artificial Intelligence combo. The aim of this paper is to propose a framework designed as a merger between Deep learning algorithms and Gaussian Mixture Modelling clustering applied in differentiating between the view planes of a second trimester fetal morphology scan. The deep learning methods chosen for this approach were ResNet50, DenseNet121, InceptionV3, EfficientNetV2S, MobileNetV3Large, and Xception. The framework establishes a hierarchy of the component networks using a statistical fitness function and the Gaussian Mixture Modelling clustering method, followed by a synergetic weighted vote of the algorithms that gives the final decision. We have tested the framework on two second trimester morphology scan datasets. A thorough statistical benchmarking process has been provided to validate our results. The experimental results showed that the synergetic vote of the framework outperforms the vote of each stand-alone deep learning network, hard voting, soft voting, and bagging strategy.

OBJECTIVE: With the increasing amount and growing variety of healthcare data, multimodal machine learning supporting integrated modeling of structured and unstructured data is an increasingly important tool for clinical machine learning tasks. However, it is non-trivial to manage the differences in dimensionality, volume, and temporal characteristics of data modalities in the context of a shared target task. Furthermore, patients can have substantial variations in the availability of data, while existing multimodal modeling methods typically assume data completeness and lack a mechanism to handle missing modalities. METHODS: We propose a Transformer-based fusion model with modality-specific tokens that summarize the corresponding modalities to achieve effective cross-modal interaction accommodating missing modalities in the clinical context. The model is further refined by inter-modal, inter-sample contrastive learning to improve the representations for better predictive performance. We denote the model as Attention-based cRoss-MOdal fUsion with contRast (ARMOUR). We evaluate ARMOUR using two input modalities (structured measurements and unstructured text), six clinical prediction tasks, and two evaluation regimes, either including or excluding samples with missing modalities. RESULTS: Our model shows improved performances over unimodal or multimodal baselines in both evaluation regimes, including or excluding patients with missing modalities in the input. The contrastive learning improves the representation power and is shown to be essential for better results. The simple setup of modality-specific tokens enables ARMOUR to handle patients with missing modalities and allows comparison with existing unimodal benchmark results. CONCLUSION: We propose a multimodal model for robust clinical prediction to achieve improved performance while accommodating patients with missing modalities. This work could inspire future research to study the effective incorporation of multiple, more complex modalities of clinical data into a single model.

OBJECTIVE: We develop a deep learning framework based on the pre-trained Bidirectional Encoder Representations from Transformers (BERT) model using unstructured clinical notes from electronic health records (EHRs) to predict the risk of disease progression from Mild Cognitive Impairment (MCI) to Alzheimer's Disease (AD). METHODS: We identified 3657 patients diagnosed with MCI together with their progress notes from Northwestern Medicine Enterprise Data Warehouse (NMEDW) between 2000 and 2020. The progress notes no later than the first MCI diagnosis were used for the prediction. We first preprocessed the notes by deidentification, cleaning and splitting into sections, and then pre-trained a BERT model for AD (named AD-BERT) based on the publicly available Bio+Clinical BERT on the preprocessed notes. All sections of a patient were embedded into a vector representation by AD-BERT and then combined by global MaxPooling and a fully connected network to compute the probability of MCI-to-AD progression. For validation, we conducted a similar set of experiments on 2563 MCI patients identified at Weill Cornell Medicine (WCM) during the same timeframe. RESULTS: Compared with the 7 baseline models, the AD-BERT model achieved the best performance on both datasets, with Area Under receiver operating characteristic Curve (AUC) of 0.849 and F1 score of 0.440 on NMEDW dataset, and AUC of 0.883 and F1 score of 0.680 on WCM dataset. CONCLUSION: The use of EHRs for AD-related research is promising, and AD-BERT shows superior predictive performance in modeling MCI-to-AD progression prediction. Our study demonstrates the utility of pre-trained language models and clinical notes in predicting MCI-to-AD progression, which could have important implications for improving early detection and intervention for AD.

OBJECTIVE: Prediction of physiological mechanics are important in medical practice because interventions are guided by predicted impacts of interventions. But prediction is difficult in medicine because medicine is complex and difficult to understand from data alone, and the data are sparse relative to the complexity of the generating processes. Computational methods can increase prediction accuracy, but prediction with clinical data is difficult because the data are sparse, noisy and nonstationary. This paper focuses on predicting physiological processes given sparse, non-stationary, electronic health record data in the intensive care unit using data assimilation (DA), a broad collection of methods that pair mechanistic models with inference methods. METHODS: A methodological pipeline embedding a glucose-insulin model into a new DA framework, the constrained ensemble Kalman filter (CEnKF) to forecast blood glucose was developed. The data include tube-fed patients whose nutrition, blood glucose, administered insulins and medications were extracted by hand due to their complexity and to ensure accuracy. The model was estimated using an individual's data as if they arrived in real-time, and the estimated model was run forward producing a forecast. Both constrained and unconstrained ensemble Kalman filters were estimated to compare the impact of constraints. Constraint boundaries, model parameter sets estimated, and data used to estimate the models were varied to investigate their influence on forecasting accuracy. Forecasting accuracy was evaluated according to mean squared error between the model-forecasted glucose and the measurements and by comparing distributions of measured glucose and forecast ensemble means. RESULTS: The novel CEnKF produced substantial gains in robustness and accuracy while minimizing the data requirements compared to the unconstrained ensemble Kalman filters. Administered insulin and tube-nutrition were important for accurate forecasting, but including glucose in IV medication delivery did not increase forecast accuracy. Model flexibility, controlled by constraint boundaries and estimated parameters, did influence forecasting accuracy. CONCLUSION: Accurate and robust physiological forecasting with sparse clinical data is possible with DA. Introducing constrained inference, particularly on unmeasured states and parameters, reduced forecast error and data requirements. The results are not particularly sensitive to model flexibility such as constraint boundaries, but over or under constraining increased forecasting errors.

BACKGROUND: The reuse of data from electronic health records (EHRs) for research purposes promises to improve the data foundation for clinical trials and may even support to enable them. Nevertheless, EHRs are characterized by both, heterogeneous structure and semantics. To standardize this data for research, the Observational Medical Outcomes Partnership (OMOP) Common Data Model (CDM) standard has recently seen an increase in use. However, the conversion of these EHRs into the OMOP CDM requires complex and resource intensive Extract Transform and Load (ETL) processes. This hampers the reuse of clinical data for research. To solve the issues of heterogeneity of EHRs and the lack of semantic precision on the care site, the openEHR standard has recently seen wider adoption. A standardized process to integrate openEHR records into the CDM potentially lowers the barriers of making EHRs accessible for research. Yet, a comprehensive approach about the integration of openEHR records into the OMOP CDM has not yet been made. METHODS: We analyzed both standards and compared their models to identify possible mappings. Based on this, we defined the necessary processes to transform openEHR records into CDM tables. We also discuss the limitation of openEHR with its unspecific demographics model and propose two possible solutions. RESULTS: We developed the OMOP Conversion Language (OMOCL) which enabled us to define a declarative openEHR archetype-to-CDM mapping language. Using OMOCL, it was possible to define a set of mappings. As a proof-of-concept, we implemented the Eos tool, which uses the OMOCL-files to successfully automatize the ETL from real-world and sample EHRs into the OMOP CDM. DISCUSSION: Both Eos and OMOCL provide a way to define generic mappings for an integration of openEHR records into OMOP. Thus, it represents a significant step towards achieving interoperability between the clinical and the research data domains. However, the transformation of openEHR data into the less expressive OMOP CDM leads to a loss of semantics.

Stroke is the second largest cause of mortality in the world. Genome-wide association studies (GWAS) have identified some genetic variants associated with stroke risk, but their putative functional causal genes are unknown. Hence, we aimed to identify putative functional causal gene biomarkers of stroke risk. We used a summary-based Mendelian randomisation (SMR) approach to identify the pleiotropic associations of genetically regulated traits (i.e., gene expression and DNA methylation) with stroke risk. Using SMR approach, we integrated cis-expression quantitative loci (cis-eQTLs) and cis-methylation quantitative loci (cis-mQTLs) data with GWAS summary statistics of stroke. We also utilised heterogeneity in dependent instruments (HEIDI) test to distinguish pleiotropy from linkage from the observed associations identified through SMR analysis. Our integrative SMR analyses and HEIDI test revealed 45 candidate biomarker genes (FDR < 0.05; PHEIDI > 0.01) that were pleiotropically or potentially causally associated with stroke risk. Of those candidate biomarker genes, 10 genes (HTRA1, PMF1, FBN2, C9orf84, COL4A1, BAG4, NEK6, SH2B3, SH3PXD2A, ACAD10) were differentially expressed in genome-wide blood transcriptomics data from stroke and healthy individuals (FDR < 0.05). Functional enrichment analysis of the identified candidate biomarker genes revealed gene ontologies and pathways involved in stroke, including "cell aging", "metal ion binding" and "oxidative damage". Based on the evidence of genetically regulated expression of genes through SMR and directly measured expression of genes in blood, our integrative analysis suggests ten genes as blood biomarkers of stroke risk. Furthermore, our study provides a better understanding of the influence of DNA methylation on the expression of genes linked to stroke risk.

OBJECTIVE: While associations between HLA antigen-level mismatches (Ag-MM) and kidney allograft failure are well established, HLA amino acid-level mismatches (AA-MM) have been less explored. Ag-MM fails to consider the substantial variability in the number of MMs at polymorphic amino acid (AA) sites within any given Ag-MM category, which may conceal variable impact on allorecognition. In this study we aim to develop a novel Feature Inclusion Bin Evolver for Risk Stratification (FIBERS) and apply it to automatically discover bins of HLA amino acid mismatches that stratify donor-recipient pairs into low versus high graft survival risk groups. METHODS: Using data from the Scientific Registry of Transplant Recipients, we applied FIBERS on a multiethnic population of 166,574 kidney transplants between 2000 and 2017. FIBERS was applied (1) across all HLA-A, B, C, DRB1, and DQB1 locus AA-MMs with comparison to 0-ABDR Ag-MM risk stratification, (2) on AA-MMs within each HLA locus individually, and (3) using cross validation to evaluate FIBERS generalizability. The predictive power of graft failure risk stratification was evaluated while adjusting for donor/recipient characteristics and HLA-A, B, C, DRB1, and DQB1 Ag-MMs as covariates. RESULTS: FIBERS's best-performing bin (on AA-MMs across all loci) added significant predictive power (hazard ratio = 1.10, Bonferroni adj. p < 0.001) in stratifying graft failure risk (where low-risk is defined as zero AA-MMs and high-risk is one or more AA-MMs) even after adjusting for Ag-MMs and donor/recipient covariates. The best bin also categorized more than twice as many patients to the low-risk category, compared to traditional 0-ABDR Ag mismatching (∼24.4% vs ∼ 9.1%). When HLA loci were binned individually, the bin for DRB1 exhibited the strongest risk stratification; relative to zero AA-MM, one or more MMs in the bin yielded HR = 1.11, p < 0.005 in a fully adjusted Cox model. AA-MMs at HLA-DRB1 peptide contact sites contributed most to incremental risk of graft failure. Additionally, FIBERS points to possible risk associated with HLA-DQB1 AA-MMs at positions that determine specificity of peptide anchor residues and HLA-DQ heterodimer stability. CONCLUSION: FIBERS's performance suggests potential for discovery of HLA immunogenetics-based risk stratification of kidney graft failure that outperforms traditional assessment.

OBJECTIVE: In the fields of medical care and research as well as hospital management, time series are an important part of the overall data basis. To ensure high quality standards and enable suitable decisions, tools for precise and generic imputations and forecasts that integrate the temporal dynamics are of great importance. Since forecasting and imputation tasks involve an inherent uncertainty, the focus of our work lay on a probabilistic multivariate generative approach that samples infillings or forecasts from an analysable distribution rather than producing deterministic results. MATERIALS AND METHODS: For this task, we developed a system based on generative adversarial networks that consist of recurrent encoders and decoders with attention mechanisms and can learn the distribution of intervals from multivariate time series conditioned on the periods before and, if available, periods after the values that are to be predicted. For training, validation and testing, a data set of jointly measured blood pressure series (ABP) and electrocardiograms (ECG) (length: 1,250=ˆ10s) was generated. For the imputation tasks, one interval of fixed length was masked randomly and independently in both channels of every sample. For the forecasting task, all masks were positioned at the end. RESULTS: The models were trained on around 65,000 bivariate samples and tested against 14,000 series of different persons. For the evaluation, 50 samples were produced for every masked interval to estimate the range of the generated infillings or forecasts. The element-wise arithmetic average of these samples served as an estimator for the mean of the learned conditional distribution. The approach showed better results than a state-of-the-art probabilistic multivariate forecasting mechanism based on Gaussian copula transformation and recurrent neural networks. On the imputation task, the proposed method reached a mean squared error (MSE) of 0.057 on the ECG channel and an MSE of 28.30 on the ABP channel, while the baseline approach reached MSEs of 0.095 (ECG) and 229.1 (ABP). Moreover, on the forecasting task, the presented system achieved MSEs of 0.069 (ECG) and 33.73 (ABP), outperforming the recurrent copula approach, which reached MSEs of 0.082 (ECG) and 196.53 (ABP). CONCLUSION: The presented generative probabilistic system for the imputation and forecasting of (medical) time series features the flexibility to handle masks of different sizes and positions, the ability to quantify uncertainty due to its probabilistic predictions, and an adjustable trade-off between the goals of minimising errors in individual predictions and minimising the distance between the learned and the real conditional distribution of the infillings or forecasts.

In this work, we describe the findings of the 'WisPerMed' team from their participation in Track 1 (Contextualized Medication Event Extraction) of the n2c2 2022 challenge. We tackle two tasks: (i) medication extraction, which involves extracting all mentions of medications from the clinical notes, and (ii) event classification, which involves classifying the medication mentions based on whether a change in the medication has been discussed. To address the long lengths of clinical texts, which often exceed the maximum token length that models based on the transformer-architecture can handle, various approaches, such as the use of ClinicalBERT with a sliding window approach and Longformer-based models, are employed. In addition, domain adaptation through masked language modeling and preprocessing steps such as sentence splitting are utilized to improve model performance. Since both tasks were treated as named entity recognition (NER) problems, a sanity check was performed in the second release to eliminate possible weaknesses in the medication detection itself. This check used the medication spans to remove false positive predictions and replace missed tokens with the highest softmax probability of the disposition types. The effectiveness of these approaches is evaluated through multiple submissions to the tasks, as well as with post-challenge results, with a focus on the DeBERTa v3 model and its disentangled attention mechanism. Results show that the DeBERTa v3 model performs well in both the NER task and the event classification task.

Understanding patients' survival probability as well as the factors affecting it constitute a significant concern for researchers and practitioners, in particular for patients with severe chronic illnesses such as congestive heart failure (CHF). CHF is a clinical syndrome characterized by comorbidities and adverse medical events. Risk stratification to identify patients most likely to die shortly after hospital discharge can improve the quality of care by better allocating organizational resources and personalized interventions. Probability assessment improves clinical decision-making, contributes to personalized care, and saves costs. Although one of the most informative indices is the time to an adverse event for each patient, commonly analyzed using survival analysis methods, these are often challenging to implement due to the complexity of the medical data. Numerous studies have used the Cox proportional hazards (PH) regression method to generate the survival distribution pattern and factors affecting survival. This model, although advantageous for survival analysis, assumes the homogeneity of the hazard ratio across patients and independence of the observations in terms of survival time. These assumptions are often violated in real-world data, especially when the dataset is composed of readmission data for chronically ill patients, since these recurring observations are inherently dependent. This study ran the Cox PH regression on a feature set selected by machine learning algorithms from a rich hospital dataset. The event modeled here was patient mortality within 90 days post-hospital discharge. The sample was composed of medical records of patients hospitalized in the Israeli Sheba Medical Center more than once, with CHF as the primary diagnosis. We modeled the survival of CHF patients using the Cox PH regression with and without the shared frailty correction that addresses the shortcomings of the Cox Model. The results of the two models of the Cox PH regression - with and without the shared frailty correction were compared. The results demonstrate that the shared frailty correction, which was statistically significant in our analysis, improved the performance of the basic Cox PH model. While this is the main contribution, we also show that this model outperforms two commonly used measures (ADHERE and EFFECT) for predicting early mortality of CHF patients. Thus, the results illustrate how applying advanced analytics can outperform traditional methods. An additional contribution is the feature set selected using machine-learning methods that is different from those used in the extant literature.

OBJECTIVE: The study has dual objectives. Our first objective (1) is to develop a community-of-practice-based evaluation methodology for knowledge-intensive computational methods. We target a whitebox analysis of the computational methods to gain insight on their functional features and inner workings. In more detail, we aim to answer evaluation questions on (i) support offered by computational methods for functional features within the application domain; and (ii) in-depth characterizations of the underlying computational processes, models, data and knowledge of the computational methods. Our second objective (2) involves applying the evaluation methodology to answer questions (i) and (ii) for knowledge-intensive clinical decision support (CDS) methods, which operationalize clinical knowledge as computer interpretable guidelines (CIG); we focus on multimorbidity CIG-based clinical decision support (MGCDS) methods that target multimorbidity treatment plans. MATERIALS AND METHODS: Our methodology directly involves the research community of practice in (a) identifying functional features within the application domain; (b) defining exemplar case studies covering these features; and (c) solving the case studies using their developed computational methods-research groups detail their solutions and functional feature support in solution reports. Next, the study authors (d) perform a qualitative analysis of the solution reports, identifying and characterizing common themes (or dimensions) among the computational methods. This methodology is well suited to perform whitebox analysis, as it directly involves the respective developers in studying inner workings and feature support of computational methods. Moreover, the established evaluation parameters (e.g., features, case studies, themes) constitute a re-usable benchmark framework, which can be used to evaluate new computational methods as they are developed. We applied our community-of-practice-based evaluation methodology on MGCDS methods. RESULTS: Six research groups submitted comprehensive solution reports for the exemplar case studies. Solutions for two of these case studies were reported by all groups. We identified four evaluation dimensions: detection of adverse interactions, management strategy representation, implementation paradigms, and human-in-the-loop support. Based on our whitebox analysis, we present answers to the evaluation questions (i) and (ii) for MGCDS methods. DISCUSSION: The proposed evaluation methodology includes features of illuminative and comparison-based approaches; focusing on understanding rather than judging/scoring or identifying gaps in current methods. It involves answering evaluation questions with direct involvement of the research community of practice, who participate in setting up evaluation parameters and solving exemplar case studies. Our methodology was successfully applied to evaluate six MGCDS knowledge-intensive computational methods. We established that, while the evaluated methods provide a multifaceted set of solutions with different benefits and drawbacks, no single MGCDS method currently provides a comprehensive solution for MGCDS. CONCLUSION: We posit that our evaluation methodology, applied here to gain new insights into MGCDS, can be used to assess other types of knowledge-intensive computational methods and answer other types of evaluation questions. Our case studies can be accessed at our GitHub repository (https://github.com/william-vw/MGCDS).

Infections are implicated in the etiology of type 1 diabetes mellitus (T1DM); however, conflicting epidemiologic evidence makes designing effective strategies for presymptomatic screening and disease prevention difficult. Considering the temporality and combination in which infections occur may provide valuable insights into understanding T1DM etiology but is rarely studied due to limited longitudinal datasets and insufficient analytical techniques. The objective of this work was to demonstrate a computational approach to classify the temporality and combination of infections in presymptomatic T1DM. We present a sequential data mining pipeline that leverages routinely collected infectious disease data from a prospective cohort study, the Environmental Determinants of Diabetes in the Young (TEDDY) study, to extract, interpret, and compare infection sequences. We then utilize this pipeline to assess risk for developing presymptomatic biomarkers of islet autoimmunity and clinical onset of T1DM. Overall, we identified 229 significant sequential rules that increased the risk for developing presymptomatic biomarkers of islet autoimmunity or clinical onset of T1DM. Multiple significant sequential rules involving varicella increased the risk for all presymptomatic biomarker-specific outcomes, while a single significant sequential rule involving parasites significantly increased risk for T1DM. Significant sequential rules involving respiratory illnesses were differentially represented among the presymptomatic biomarkers of islet autoimmunity and clinical onset of T1DM. Risk for T1DM was significantly increased by a single episode of sixth disease at 12 months, representing the only single-event sequence that increased disease risk. Together, these findings provide the first insights into the timing and combination of infections in T1DM etiology, which may ultimately lead to personalized disease screening and prevention strategies. The sequential data mining pipeline developed in this work demonstrates how temporal data mining can be used to address clinically meaningful questions. This method can be adapted to other presymptomatic factors and clinical conditions.

Self-supervised learning approaches provide a promising direction for clustering multivariate time-series data. However, real-world time-series data often include missing values, and the existing approaches require imputing missing values before clustering, which may cause extensive computations and noise and result in invalid interpretations. To address these challenges, we present a Self-supervised Learning-based Approach to Clustering multivariate Time-series data with missing values (SLAC-Time). SLAC-Time is a Transformer-based clustering method that uses time-series forecasting as a proxy task for leveraging unlabeled data and learning more robust time-series representations. This method jointly learns the neural network parameters and the cluster assignments of the learned representations. It iteratively clusters the learned representations with the K-means method and then utilizes the subsequent cluster assignments as pseudo-labels to update the model parameters. To evaluate our proposed approach, we applied it to clustering and phenotyping Traumatic Brain Injury (TBI) patients in the Transforming Research and Clinical Knowledge in Traumatic Brain Injury (TRACK-TBI) study. Clinical data associated with TBI patients are often measured over time and represented as time-series variables characterized by missing values and irregular time intervals. Our experiments demonstrate that SLAC-Time outperforms the baseline K-means clustering algorithm in terms of silhouette coefficient, Calinski Harabasz index, Dunn index, and Davies Bouldin index. We identified three TBI phenotypes that are distinct from one another in terms of clinically significant variables as well as clinical outcomes, including the Extended Glasgow Outcome Scale (GOSE) score, Intensive Care Unit (ICU) length of stay, and mortality rate. The experiments show that the TBI phenotypes identified by SLAC-Time can be potentially used for developing targeted clinical trials and therapeutic strategies.

OBJECTIVE: Lung cancer exhibits unpredictable recurrence in low-stage tumors and variable responses to different therapeutic interventions. Predicting relapse in early-stage lung cancer can facilitate precision medicine and improve patient survivability. While existing machine learning models rely on clinical data, incorporating genomic information could enhance their efficiency. This study aims to impute and integrate specific types of genomic data with clinical data to improve the accuracy of machine learning models for predicting relapse in early-stage, non-small cell lung cancer patients. METHODS: The study utilized a publicly available TCGA lung cancer cohort and imputed genetic pathway scores into the Spanish Lung Cancer Group (SLCG) data, specifically in 1348 early-stage patients. Initially, tumor recurrence was predicted without imputed pathway scores. Subsequently, the SLCG data were augmented with pathway scores imputed from TCGA. The integrative approach aimed to enhance relapse risk prediction performance. RESULTS: The integrative approach achieved improved relapse risk prediction with the following evaluation metrics: an area under the precision-recall curve (PR-AUC) score of 0.75, an area under the ROC (ROC-AUC) score of 0.80, an F1 score of 0.61, and a Precision of 0.80. The prediction explanation model SHAP (SHapley Additive exPlanations) was employed to explain the machine learning model's predictions. CONCLUSION: We conclude that our explainable predictive model is a promising tool for oncologists that addresses an unmet clinical need of post-treatment patient stratification based on the relapse risk while also improving the predictive power by incorporating proxy genomic data not available for specific patients.

Daily progress notes are a common note type in the electronic health record (EHR) where healthcare providers document the patient's daily progress and treatment plans. The EHR is designed to document all the care provided to patients, but it also enables note bloat with extraneous information that distracts from the diagnoses and treatment plans. Applications of natural language processing (NLP) in the EHR is a growing field with the majority of methods in information extraction. Few tasks use NLP methods for downstream diagnostic decision support. We introduced the 2022 National NLP Clinical Challenge (N2C2) Track 3: Progress Note Understanding - Assessment and Plan Reasoning as one step towards a new suite of tasks. The Assessment and Plan Reasoning task focuses on the most critical components of progress notes, Assessment and Plan subsections where health problems and diagnoses are contained. The goal of the task was to develop and evaluate NLP systems that automatically predict causal relations between the overall status of the patient contained in the Assessment section and its relation to each component of the Plan section which contains the diagnoses and treatment plans. The goal of the task was to identify and prioritize diagnoses as the first steps in diagnostic decision support to find the most relevant information in long documents like daily progress notes. We present the results of the 2022 N2C2 Track 3 and provide a description of the data, evaluation, participation and system performance.

OBJECTIVE: To represent a patient record with both time-invariant and time-varying features as a single vector using an end-to-end deep learning model, and further to predict the kidney failure (KF) status and mortality of heart failure (HF) patients. MATERIALS AND METHODS: The time-invariant EMR data included demographic information and comorbidities, and the time-varying EMR data were lab tests. We used a Transformer encoder module to represent the time-invariant data, and refined a long short-term memory (LSTM) with a Transformer encoder attached to the top to represent the time-varying data, taking the original measured values and their corresponding embedding vectors, masking vectors, and two types of time intervals as inputs. The proposed representations of patients with time-invariant and time-varying data were used to predict KF status (949 out of 5268 HF patients diagnosed with KF) and mortality (463 in-hospital deaths) for HF patients. Comparative experiments were conducted between the proposed model and some representative machine learning models. Ablation experiments were also performed around the time-varying data representation, including replacing the refined LSTM with the standard LSTM, GRU-D and T-LSTM, respectively, and removing the Transformer encoder and the time-varying data representation module, respectively. The visualization of the attention weights of the time-invariant and time-varying features was used to clinically interpret the predictive performance. We used the area under the receiver operating characteristic curve (AUROC), the area under the precision-recall curve (AUPRC), and the F1-score to evaluate the predictive performance of the models. RESULTS: The proposed model achieved superior performance, with average AUROCs, AUPRCs and F1-scores of 0.960, 0.610 and 0.759 for KF prediction and 0.937, 0.353 and 0.537 for mortality prediction, respectively. Predictive performance improved with the addition of time-varying data from longer time periods. The proposed model outperformed the comparison and ablation references in both prediction tasks. CONCLUSIONS: Both time-invariant and time-varying EMR data of patients could be efficiently represented by the proposed unified deep learning model, which shows higher performance in clinical prediction tasks. The way to use time-varying data in the current study is hopeful to be used in other kinds of time-varying data and other clinical tasks.

Angiogenesis is essential for tumor growth and cancer metastasis. Identifying the molecular pathways involved in this process is the first step in the rational design of new therapeutic strategies to improve cancer treatment. In recent years, RNA-seq data analysis has helped to determine the genetic and molecular factors associated with different types of cancer. In this work we performed integrative analysis using RNA-seq data from human umbilical vein endothelial cells (HUVEC) and patients with angiogenesis-dependent diseases to find genes that serve as potential candidates to improve the prognosis of tumor angiogenesis deregulation and understand how this process is orchestrated at the genetic and molecular level. We downloaded four RNA-seq datasets (including cellular models of tumor angiogenesis and ischaemic heart disease) from the Sequence Read Archive. Our integrative analysis includes a first step to determine differentially and co-expressed genes. For this, we used the ExpHunter Suite, an R package that performs differential expression, co-expression and functional analysis of RNA-seq data. We used both differentially and co-expressed genes to explore the human gene interaction network and determine which genes were found in the different datasets that may be key for the angiogenesis deregulation. Finally, we performed drug repositioning analysis to find potential targets related to angiogenesis inhibition. We found that that among the transcriptional alterations identified, SEMA3D and IL33 genes are deregulated in all datasets. Microenvironment remodeling, cell cycle, lipid metabolism and vesicular transport are the main molecular pathways affected. In addition to this, interacting genes are involved in intracellular signaling pathways, especially in immune system and semaphorins, respiratory electron transport and fatty acid metabolism. The methodology presented here can be used for finding common transcriptional alterations in other genetically-based diseases.

OBJECTIVE: Investigate the preliminary efficacy and feasibility of a personalized mobile health (mHealth) intervention based on social cognitive theory (SCT) to promote physical activity among type 2 diabetes patients via self-monitoring, goal setting, and automatic feedback. METHODS: We conducted a pilot study involving 33 type 2 diabetes patients attending Mitsui Memorial Hospital in Japan using a pre-post evaluation design over 12 weeks. Participants measured daily step count, body weight, and blood pressure at home, with the measurements synchronized with the StepAdd application (app) automatically. Participants used the app to review daily results, update personalized step goals, identify individualized barriers to achieving the step goals, find coping strategies to overcome each barrier, and implement these strategies, thereby building effective coping skills to meet the goals. Pharmacists examined the usage of the app and provided coaching on lifestyle modifications. Ultimately, patients established skills to enhance diabetes self-care by using the app. RESULTS: Daily step count increased dramatically with high statistical significance (p < 0.0001), from a mean of 5436 steps/day to 10,150 steps/day, an 86.7 % increase. HbA1c (p = 0.0001) and BMI (p = 0.0038) also improved. Diabetes self-care in diet, exercise, and foot care as well as self-management behavior, self-regulation, and self-efficacy in achieving daily step goals showed significant improvements. The retention rate of the study was very high, at 97.0 % (n = 32). CONCLUSIONS: A personalized smartphone-based mHealth intervention based on SCT is feasible and effective at promoting physical activity among type 2 diabetes patients. The methodology of the intervention could be readily applied to other patient populations.

OBJECTIVE: To apply the latest guidance for estimating and evaluating heterogeneous treatment effects (HTEs) in an end-to-end case study of the Long-term Anticoagulation Therapy (RE-LY) trial, and summarize the main takeaways from applying state-of-the-art metalearners and novel evaluation metrics in-depth to inform their applications to personalized care in biomedical research. METHODS: Based on the characteristics of the RE-LY data, we selected four metalearners (S-learner with Lasso, X-learner with Lasso, R-learner with random survival forest and Lasso, and causal survival forest) to estimate the HTEs of dabigatran. For the outcomes of (1) stroke or systemic embolism and (2) major bleeding, we compared dabigatran 150 mg, dabigatran 110 mg, and warfarin. We assessed the overestimation of treatment heterogeneity by the metalearners via a global null analysis and their discrimination and calibration ability using two novel metrics: rank-weighted average treatment effects (RATE) and estimated calibration error for treatment heterogeneity. Finally, we visualized the relationships between estimated treatment effects and baseline covariates using partial dependence plots. RESULTS: The RATE metric suggested that either the applied metalearners had poor performance of estimating HTEs or there was no treatment heterogeneity for either the stroke/SE or major bleeding outcome of any treatment comparison. Partial dependence plots revealed that several covariates had consistent relationships with the treatment effects estimated by multiple metalearners. The applied metalearners showed differential performance across outcomes and treatment comparisons, and the X- and R-learners yielded smaller calibration errors than the others. CONCLUSIONS: HTE estimation is difficult, and a principled estimation and evaluation process is necessary to provide reliable evidence and prevent false discoveries. We have demonstrated how to choose appropriate metalearners based on specific data properties, applied them using the off-the-shelf implementation tool survlearners, and evaluated their performance using recently defined formal metrics. We suggest that clinical implications should be drawn based on the common trends across the applied metalearners.

OBJECTIVE: Clinical data's confidential nature often limits the development of machine learning models in healthcare. Generative adversarial networks (GANs) can synthesise realistic datasets, but suffer from mode collapse, resulting in low diversity and bias towards majority demographics and common clinical practices. This work proposes an extension to the classic GAN framework that includes a variational autoencoder (VAE) and an external memory mechanism to overcome these limitations and generate synthetic data accurately describing imbalanced class distributions commonly found in clinical variables. METHODS: The proposed method generated a synthetic dataset related to antiretroviral therapy for human immunodeficiency virus (ART for HIV). We evaluated it based on five metrics: (1) accurately representing imbalanced class distribution; (2) the realism of the individual variables; (3) the realism among variables; (4) patient disclosure risk; and (5) the utility of the generated dataset for developing downstream machine learning models. RESULTS: The proposed method overcomes the issue of mode collapse and generates a synthetic dataset that accurately describes imbalanced class distributions commonly found in clinical variables. The generated data has a patient disclosure risk of 0.095%, lower than the 9% threshold stated by Health Canada and the European Medicines Agency, making it suitable for distribution to the research community with high security. The generated data also has high utility, indicating the potential of the proposed method to enable the development of downstream machine learning algorithms for healthcare applications using synthetic data. CONCLUSION: Our proposed extension to the classic GAN framework, which includes a VAE and an external memory mechanism, represents a promising approach towards generating synthetic data that accurately describe imbalanced class distributions commonly found in clinical variables. This method overcomes the limitations of GANs and creates more realistic datasets with higher patient cohort diversity, facilitating the development of downstream machine learning algorithms for healthcare applications.

BACKGROUND: Identifying practice-ready evidence-based journal articles in medicine is a challenge due to the sheer volume of biomedical research publications. Newer approaches to support evidence discovery apply deep learning techniques to improve the efficiency and accuracy of classifying sound evidence. OBJECTIVE: To determine how well deep learning models using variants of Bidirectional Encoder Representations from Transformers (BERT) identify high-quality evidence with high clinical relevance from the biomedical literature for consideration in clinical practice. METHODS: We fine-tuned variations of BERT models (BERTBASE, BioBERT, BlueBERT, and PubMedBERT) and compared their performance in classifying articles based on methodological quality criteria. The dataset used for fine-tuning models included titles and abstracts of >160,000 PubMed records from 2012 to 2020 that were of interest to human health which had been manually labeled based on meeting established critical appraisal criteria for methodological rigor. The data was randomly divided into 80:10:10 sets for training, validating, and testing. In addition to using the full unbalanced set, the training data was randomly undersampled into four balanced datasets to assess performance and select the best performing model. For each of the four sets, one model that maintained sensitivity (recall) at ≥99% was selected and were ensembled. The best performing model was evaluated in a prospective, blinded test and applied to an established reference standard, the Clinical Hedges dataset. RESULTS: In training, three of the four selected best performing models were trained using BioBERTBASE. The ensembled model did not boost performance compared with the best individual model. Hence a solo BioBERT-based model (named DL-PLUS) was selected for further testing as it was computationally more efficient. The model had high recall (>99%) and 60% to 77% specificity in a prospective evaluation conducted with blinded research associates and saved >60% of the work required to identify high quality articles. CONCLUSIONS: Deep learning using pretrained language models and a large dataset of classified articles produced models with improved specificity while maintaining >99% recall. The resulting DL-PLUS model identifies high-quality, clinically relevant articles from PubMed at the time of publication. The model improves the efficiency of a literature surveillance program, which allows for faster dissemination of appraised research.

OBJECTIVE: We developed and evaluated a novel one-shot distributed algorithm for evidence synthesis in distributed research networks with rare outcomes. MATERIALS AND METHODS: Fed-Padé, motivated by a classic mathematical tool, Padé approximants, reconstructs the multi-site data likelihood via Padé approximant whose key parameters can be computed distributively. Thanks to the simplicity of [2,2] Padé approximant, Fed-Padé requests an extremely simple task and low communication cost for data partners. Specifically, each data partner only needs to compute and share the log-likelihood and its first 4 gradients evaluated at an initial estimator. We evaluated the performance of our algorithm with extensive simulation studies and four observational healthcare databases. RESULTS: Our simulation studies revealed that a [2,2]-Padé approximant can well reconstruct the multi-site likelihood so that Fed-Padé produces nearly identical estimates to the pooled analysis. Across all simulation scenarios considered, the median of relative bias and rate of instability of our Fed-Padé are both <0.1%, whereas meta-analysis estimates have bias up to 50% and instability up to 75%. Furthermore, the confidence intervals derived from the Fed-Padé algorithm showed better coverage of the truth than confidence intervals based on the meta-analysis. In real data analysis, the Fed-Padé has a relative bias of <1% for all three comparisons for risks of acute liver injury and decreased libido, whereas the meta-analysis estimates have a substantially higher bias (around 10%). CONCLUSION: The Fed-Padé algorithm is nearly lossless, stable, communication-efficient, and easy to implement for models with rare outcomes. It provides an extremely suitable and convenient approach for synthesizing evidence in distributed research networks with rare outcomes.

OBJECTIVE: Physical inactivity is a leading modifiable cause of death and disease worldwide. Population-based interventions to increase physical activity are needed. Existing automated expert systems (e.g., computer-tailored interventions) have significant limitations that result in low long-term effectiveness. Therefore, innovative approaches are needed. This special communication aims to describe and discuss a novel mHealth intervention approach that proactively offers participants with hyper-personalised intervention content adjusted in real-time. METHODS: Using machine learning approaches, we propose a novel physical activity intervention approach that can learn and adapt in real-time to achieve high levels of personalisation and user engagement, underpinned by a likeable digital assistant. It will consist of three major components: (1) conversations: to increase user's knowledge on a wide range of activity-related topics underpinned by Natural Language Processing; (2) nudge engine: to provide users with hyper-personalised cues to action underpinned by reinforcement learning (i.e., contextual bandit) and integrating real-time data from activity tracking, GPS, GIS, weather, and user provided data; (3) Q&A: to facilitate users asking any physical activity related questions underpinned by generative AI (e.g., ChatGPT, Bard) for content generation. RESULTS: The detailed concept of the proposed physical activity intervention platform demonstrates the practical application of a just-in-time adaptive intervention applying various machine learning techniques to deliver a hyper-personalised physical activity intervention in an engaging way. Compared to traditional interventions, the novel platform is expected to show potential for increased user engagement and long-term effectiveness due to: (1) using new variables to personalise content (e.g., GPS, weather), (2) providing behavioural support at the right time in real-time, (3) implementing an engaging digital assistant and (4) improving the relevance of content through applying machine learning algorithms. CONCLUSION: The use of machine learning is on the rise in every aspect of today's society, however few attempts have been undertaken to harness its potential to achieve health behaviour change. By sharing our intervention concept, we contribute to the ongoing dialogue on creating effective methods for promoting health and well-being in the informatics research community. Future research should focus on refining these techniques and evaluating their effectiveness in controlled and real-world circumstances.

BACKGROUND: In return for their nutritional properties and broad availability, cereal crops have been associated with different alimentary disorders and symptoms, with the majority of the responsibility being attributed to gluten. Therefore, the research of gluten-related literature data continues to be produced at ever-growing rates, driven in part by the recent exploratory studies that link gluten to non-traditional diseases and the popularity of gluten-free diets, making it increasingly difficult to access and analyse practical and structured information. In this sense, the accelerated discovery of novel advances in diagnosis and treatment, as well as exploratory studies, produce a favourable scenario for disinformation and misinformation. OBJECTIVES: Aligned with, the European Union strategy "Delivering on EU Food Safety and Nutrition in 2050″ which emphasizes the inextricable links between imbalanced diets, the increased exposure to unreliable sources of information and misleading information, and the increased dependency on reliable sources of information; this paper presents GlutKNOIS, a public and interactive literature-based database that reconstructs and represents the experimental biomedical knowledge extracted from the gluten-related literature. The developed platform includes different external database knowledge, bibliometrics statistics and social media discussion to propose a novel and enhanced way to search, visualise and analyse potential biomedical and health-related interactions in relation to the gluten domain. METHODS: For this purpose, the presented study applies a semi-supervised curation workflow that combines natural language processing techniques, machine learning algorithms, ontology-based normalization and integration approaches, named entity recognition methods, and graph knowledge reconstruction methodologies to process, classify, represent and analyse the experimental findings contained in the literature, which is also complemented by data from the social discussion. RESULTS AND CONCLUSIONS: In this sense, 5814 documents were manually annotated and 7424 were fully automatically processed to reconstruct the first online gluten-related knowledge database of evidenced health-related interactions that produce health or metabolic changes based on the literature. In addition, the automatic processing of the literature combined with the knowledge representation methodologies proposed has the potential to assist in the revision and analysis of years of gluten research. The reconstructed knowledge base is public and accessible at https://sing-group.org/glutknois/.

OBJECTIVE: Despite the high prevalence of alcohol use disorder (AUD) in the United States, limited research is focused on the associations among AUD, pain, and opioids/benzodiazepine use. In addition, little is known regarding individuals with a history of AUD and their potential risk for pain diagnoses, pain prescriptions, and subsequent misuse. Moreover, the potential risk of pain diagnoses, prescriptions, and subsequent misuse among individuals with a history of AUD is not well known. The objective was to develop a tailored dataset by linking data from 2 New York State (NYS) administrative databases to investigate a series of hypotheses related to AUD and painful medical disorders. METHODS: Data from the NYS Office of Addiction Services and Supports (OASAS) Client Data System (CDS) and Medicaid claims data from the NYS Department of Health Medicaid Data Warehouse (MDW) were merged using a stepwise deterministic method. Multiple patient-level identifier combinations were applied to create linkage rules. We included patients aged 18 and older from the OASAS CDS who initially entered treatment with a primary substance use of alcohol and no use of opioids between January 1, 2003, and September 23, 2019. This cohort was then linked to corresponding Medicaid claims. RESULTS: A total of 177,685 individuals with a primary AUD problem and no opioid use history were included in the dataset. Of these, 37,346 (21.0%) patients had an OUD diagnosis, and 3,365 (1.9%) patients experienced an opioid overdose. There were 121,865 (68.6%) patients found to have a pain condition. CONCLUSION: The integrated database allows researchers to examine the associations among AUD, pain, and opioids/benzodiazepine use, and propose hypotheses to improve outcomes for at-risk patients. The findings of this study can contribute to the development of a prognostic prediction model and the analysis of longitudinal outcomes to improve the care of patients with AUD.

BACKGROUND: Causal feature selection is essential for estimating effects from observational data. Identifying confounders is a crucial step in this process. Traditionally, researchers employ content-matter expertise and literature review to identify confounders. Uncontrolled confounding from unidentified confounders threatens validity, conditioning on intermediate variables (mediators) weakens estimates, and conditioning on common effects (colliders) induces bias. Additionally, without special treatment, erroneous conditioning on variables combining roles introduces bias. However, the vast literature is growing exponentially, making it infeasible to assimilate this knowledge. To address these challenges, we introduce a novel knowledge graph (KG) application enabling causal feature selection by combining computable literature-derived knowledge with biomedical ontologies. We present a use case of our approach specifying a causal model for estimating the total causal effect of depression on the risk of developing Alzheimer's disease (AD) from observational data. METHODS: We extracted computable knowledge from a literature corpus using three machine reading systems and inferred missing knowledge using logical closure operations. Using a KG framework, we mapped the output to target terminologies and combined it with ontology-grounded resources. We translated epidemiological definitions of confounder, collider, and mediator into queries for searching the KG and summarized the roles played by the identified variables. We compared the results with output from a complementary method and published observational studies and examined a selection of confounding and combined role variables in-depth. RESULTS: Our search identified 128 confounders, including 58 phenotypes, 47 drugs, 35 genes, 23 collider, and 16 mediator phenotypes. However, only 31 of the 58 confounder phenotypes were found to behave exclusively as confounders, while the remaining 27 phenotypes played other roles. Obstructive sleep apnea emerged as a potential novel confounder for depression and AD. Anemia exemplified a variable playing combined roles. CONCLUSION: Our findings suggest combining machine reading and KG could augment human expertise for causal feature selection. However, the complexity of causal feature selection for depression with AD highlights the need for standardized field-specific databases of causal variables. Further work is needed to optimize KG search and transform the output for human consumption.

BACKGROUND: Online health communities (OHCs) have emerged as prominent platforms for behavior modification, and the digitization of online peer interactions has afforded researchers with unique opportunities to model multilevel mechanisms that drive behavior change. Existing studies, however, have been limited by a lack of methods that allow the capture of conversational context and socio-behavioral dynamics at scale, as manifested in these digital platforms. OBJECTIVE: We develop, evaluate, and apply a novel methodological framework, Pragmatics to Reveal Intent in Social Media (PRISM), to facilitate granular characterization of peer interactions by combining multidimensional facets of human communication. METHODS: We developed and applied PRISM to analyze peer interactions (N = 2.23 million) in QuitNet, an OHC for tobacco cessation. First, we generated a labeled set of peer interactions (n = 2,005) through manual annotation along three dimensions: communication themes (CTs), behavior change techniques (BCTs), and speech acts (SAs). Second, we used deep learning models to apply our qualitative codes at scale. Third, we applied our validated model to perform a retrospective analysis. Finally, using social network analysis (SNA), we portrayed large-scale patterns and relationships among the aforementioned communication dimensions embedded in peer interactions in QuitNet. RESULTS: Qualitative analysis showed that the themes of social support and behavioral progress were common. The most used BCTs were feedback and monitoring and comparison of behavior, and users most commonly expressed their intentions using SAs-expressive and emotion. With additional in-domain pre-training, bidirectional encoder representations from Transformers (BERT) outperformed other deep learning models on the classification tasks. Content-specific SNA revealed that users' engagement or abstinence status is associated with the prevalence of various categories of BCTs and SAs, which also was evident from the visualization of network structures. CONCLUSIONS: Our study describes the interplay of multilevel characteristics of online communication and their association with individual health behaviors.

BACKGROUND: Adverse outcome pathway (AOP) networks are versatile tools in toxicology and risk assessment that capture and visualize mechanisms driving toxicity originating from various data sources. They share a common structure consisting of a set of molecular initiating events and key events, connected by key event relationships, leading to the actual adverse outcome. AOP networks are to be considered living documents that should be frequently updated by feeding in new data. Such iterative optimization exercises are typically done manually, which not only is a time-consuming effort, but also bears the risk of overlooking critical data. The present study introduces a novel approach for AOP network optimization of a previously published AOP network on chemical-induced cholestasis using artificial intelligence to facilitate automated data collection followed by subsequent quantitative confidence assessment of molecular initiating events, key events, and key event relationships. METHODS: Artificial intelligence-assisted data collection was performed by means of the free web platform Sysrev. Confidence levels of the tailored Bradford-Hill criteria were quantified for the purpose of weight-of-evidence assessment of the optimized AOP network. Scores were calculated for biological plausibility, empirical evidence, and essentiality, and were integrated into a total key event relationship confidence value. The optimized AOP network was visualized using Cytoscape with the node size representing the incidence of the key event and the edge size indicating the total confidence in the key event relationship. RESULTS: This resulted in the identification of 38 and 135 unique key events and key event relationships, respectively. Transporter changes was the key event with the highest incidence, and formed the most confident key event relationship with the adverse outcome, cholestasis. Other important key events present in the AOP network include: nuclear receptor changes, intracellular bile acid accumulation, bile acid synthesis changes, oxidative stress, inflammation and apoptosis. CONCLUSIONS: This process led to the creation of an extensively informative AOP network focused on chemical-induced cholestasis. This optimized AOP network may serve as a mechanistic compass for the development of a battery of in vitro assays to reliably predict chemical-induced cholestatic injury.

Monoclonal antibodies (MAs) are increasingly used in the therapeutic arsenal. Clinical Data Warehouses (CDWs) offer unprecedented opportunities for research on real-word data. The objective of this work is to develop a knowledge organization system on MAs for therapeutic use (MATUs) applicable in Europe to query CDWs from a multi-terminology server (HeTOP). After expert consensus, three main health thesauri were selected: the MeSH thesaurus, the National Cancer Institute thesaurus (NCIt) and the SNOMED CT. These thesauri contain 1,723 MAs concepts, but only 99 (5.7 %) are identified as MATUs. The knowledge organisation system proposed in this article is a six-level hierarchical system according to their main therapeutic target. It includes 193 different concepts organised in a cross lingual terminology server, which will allow the inclusion of semantic extensions. Ninety nine (51.3 %) MATUs concepts and 94 (48.7 %) hierarchical concepts composed the knowledge organisation system. Two separates groups (an expert group and a validation group) carried out the selection, creation and validation processes. Queries identify, for unstructured data, 83 out of 99 (83.8 %) MATUs corresponding to 45,262 patients, 347,035 hospital stays and 427,544 health documents, and for structured data, 61 out of 99 (61.6 %) MATUs corresponding to 9,218 patients, 59,643 hospital stays and 104,737 hospital prescriptions. The volume of data in the CDW demonstrated the potential for using these data in clinical research, although not all MATUs are present in the CDW (16 missing for unstructured data and 38 for structured data). The knowledge organisation system proposed here improves the understanding of MATUs, the quality of queries and helps clinical researchers retrieve relevant medical information. The use of this model in CDW allows for the rapid identification of a large number of patients and health documents, either directly by a MATU of interest (e.g. Rituximab) but also by searching for parent concepts (e.g. Anti-CD20 Monoclonal Antibody).

Triplet extraction is one of the fundamental tasks in biomedical text mining. Compared with traditional pipeline approaches, joint methods can alleviate the error propagation problem from entity recognition to relation classification. However, existing methods face challenges in detecting overlapping entities and overlapping relations, which are ubiquitous in biomedical texts. In this work, we propose a novel pipeline method of end-to-end biomedical triplet extraction. In particular, a span-based detection strategy is used to detect the overlapping triplets by enumerating possible candidate spans and entity pairs. The strategy is further used to capture different contextualized representations via an entity model and a relation model, respectively. Furthermore, to enhance interrelation between spans, entity information from the output of the entity model is used to construct the input for the relation model without utilizing any external knowledge. Our approach is evaluated on the drug-drug interaction (DDI) and chemical-protein interaction (CHEMPROT) datasets, exhibiting improvement of the absolute F1-score in relation extraction by 3.5%-3.7% compared prior work. The experimental results highlight the importance of overlapping triplet detection using the span-based approach, acquisition of various contextualized representations via different in-domain pre-trained language models, and early fusion of entity information in the relation model.

OBJECTIVE: Deep neural network (DNN) techniques have demonstrated significant advantages over regression and some other techniques. In recent studies, DNN-based analysis has been conducted on data with high-dimensional input such as omics measurements. In such analysis, regularization, in particular penalization, has been applied to regularize estimation and distinguish relevant input variables from irrelevant ones. A unique challenge arises from the "lack of information" attributable to high dimensionality of input and limited size of training data. For many data/studies, there exist other data/studies that may be relevant and can potentially provide additional information to boost performance. METHODS: In this study, we conduct integrative analysis of multiple independent datasets/studies, with the goal of borrowing information across each other and improving overall performance. Significantly different from regression-based integrative analysis (where alignment can be easily achieved based on covariates), alignment across multiple DNNs can be nontrivial. We develop ANNI, an Aligned DNN technique for Integrative analysis with high-dimensional input. Penalization is applied for regularized estimation, selection of important input variables, and, equally importantly, information borrowing across multiple DNNs. An effective computational algorithm is developed. RESULTS: Extensive simulations demonstrate competitive performance of the proposed technique. The analysis of cancer omics data further establishes its practical utility.

OBJECTIVE: Ovarian cancer is a significant health issue with lasting impacts on the community. Despite recent advances in surgical, chemotherapeutic and radiotherapeutic interventions, they have had only marginal impacts due to an inability to identify biomarkers at an early stage. Biomarker discovery is challenging, yet essential for improving drug discovery and clinical care. Machine learning (ML) techniques are invaluable for recognising complex patterns in biomarkers compared to conventional methods, yet they can lack physical insights into diagnosis. eXplainable Artificial Intelligence (XAI) is capable of providing deeper insights into the decision-making of complex ML algorithms increasing their applicability. We aim to introduce best practice for combining ML and XAI techniques for biomarker validation tasks. METHODS: We focused on classification tasks and a game theoretic approach based on Shapley values to build and evaluate models and visualise results. We described the workflow and apply the pipeline in a case study using the CDAS PLCO Ovarian Biomarkers dataset to demonstrate the potential for accuracy and utility. RESULTS: The case study results demonstrate the efficacy of the ML pipeline, its consistency, and advantages compared to conventional statistical approaches. CONCLUSION: The resulting guidelines provide a general framework for practical application of XAI in medical research that can inform clinicians and validate and explain cancer biomarkers.

In biomedical literature, cross-sentence texts can usually express rich knowledge, and extracting the interaction relation between entities from cross-sentence texts is of great significance to biomedical research. However, compared with single sentence, cross-sentence text has a longer sequence length, so the research on cross-sentence text information extraction should focus more on learning the context dependency structural information. Nowadays, it is still a challenge to handle global dependencies and structural information of long sequences effectively, and graph-oriented modeling methods have received more and more attention recently. In this paper, we propose a new graph attention network guided by syntactic dependency relationship (SR-GAT) for extracting biomedical relation from the cross-sentence text. It allows each node to pay attention to other nodes in its neighborhood, regardless of the sequence length. The attention weight between nodes is given by a syntactic relation graph probability network (SR-GPR), which encodes the syntactic dependency between nodes and guides the graph attention mechanism to learn information about the dependency structure. The learned feature representation retains information about the node-to-node syntactic dependency, and can further discover global dependencies effectively. The experimental results demonstrate on a publicly available biomedical dataset that, our method achieves state-of-the-art performance while requiring significantly less computational resources. Specifically, in the "drug-mutation" relation extraction task, our method achieves an advanced accuracy of 93.78% for binary classification and 92.14% for multi-classification. In the "drug-gene-mutation" relation extraction task, our method achieves an advanced accuracy of 93.22% for binary classification and 92.28% for multi-classification. Across all relation extraction tasks, our method improves accuracy by an average of 0.49% compared to the existing best model. Furthermore, our method achieved an accuracy of 69.5% in text classification, surpassing most existing models, demonstrating its robustness in generalization across different domains without additional fine-tuning.

INTRODUCTION: Clinical trials (CTs) often fail due to inadequate patient recruitment. Finding eligible patients involves comparing the patient's information with the CT eligibility criteria. Automated patient matching offers the promise of improving the process, yet the main difficulties of CT retrieval lie in the semantic complexity of matching unstructured patient descriptions with semi-structured, multi-field CT documents and in capturing the meaning of negation coming from the eligibility criteria. OBJECTIVES: This paper tackles the challenges of CT retrieval by presenting an approach that addresses the patient-to-trials paradigm. Our approach involves two key components in a pipeline-based model: (i) a data enrichment technique for enhancing both queries and documents during the first retrieval stage, and (ii) a novel re-ranking schema that uses a Transformer network in a setup adapted to this task by leveraging the structure of the CT documents. METHODS: We use named entity recognition and negation detection in both patient description and the eligibility section of CTs. We further classify patient descriptions and CT eligibility criteria into current, past, and family medical conditions. This extracted information is used to boost the importance of disease and drug mentions in both query and index for lexical retrieval. Furthermore, we propose a two-step training schema for the Transformer network used to re-rank the results from the lexical retrieval. The first step focuses on matching patient information with the descriptive sections of trials, while the second step aims to determine eligibility by matching patient information with the criteria section. RESULTS: Our findings indicate that the inclusion criteria section of the CT has a great influence on the relevance score in lexical models, and that the enrichment techniques for queries and documents improve the retrieval of relevant trials. The re-ranking strategy, based on our training schema, consistently enhances CT retrieval and shows improved performance by 15% in terms of precision at retrieving eligible trials. CONCLUSION: The results of our experiments suggest the benefit of making use of extracted entities. Moreover, our proposed re-ranking schema shows promising effectiveness compared to larger neural models, even with limited training data. These findings offer valuable insights for improving methods for retrieval of clinical documents.

BACKGROUND: Electronic Clinical Narratives (ECNs) store valuable individual's health information. However, there are few available open-source data. Besides, ECNs can be structurally heterogeneous, ranging from documents with explicit section headings or titles to unstructured notes. This lack of structure complicates building automatic systems and their evaluation. OBJECTIVE: The aim of the present work is to provide the scientific community with a Spanish open-source dataset to build and evaluate automatic section identification systems. Together with this dataset, the purpose is to design and implement a suitable evaluation measure and a fine-tuned language model adapted to the task. MATERIALS AND METHODS: A corpus of unstructured clinical records, in this case progress notes written in Spanish, was annotated with seven major section types. Existing metrics for the presented task were thoroughly assessed and, based on the most suitable one, we defined a new B2 metric better tailored given the task. RESULTS: The annotated corpus, as well as the designed new evaluation script and a baseline model are freely available for the community. This model reaches an average B2 score of 71.3 on our open source dataset and an average B2 of 67.0 in data scarcity scenarios where the target corpus and its structure differs from the dataset used for training the LM. CONCLUSION: Although section identification in unstructured clinical narratives is challenging, this work shows that it is possible to build competitive automatic systems when both data and the right evaluation metrics are available. The annotated data, the implemented evaluation scripts, and the section identification Language Model are open-sourced hoping that this contribution will foster the building of more and better systems.

BACKGROUND AND OBJECTIVE: Valvular heart disease (VHD) is associated with elevated mortality rates. Although transthoracic echocardiography (TTE) is the gold standard detection tool, phonocardiography (PCG) could be an alternative as it is a cost-effective and noninvasive method for cardiac auscultation. Many researchers have dedicated their efforts to improving the decision-making process and developing robust and precise approaches to assist physicians in providing reliable diagnoses of VHD. METHODS: This research proposes a novel approach for the detection of anomalous valvular heart sounds from PCG signals. The proposed approach combines orthogonal non-negative matrix factorization (ONMF) and convolutional neural network (CNN) architectures in a three-stage cascade. The aim of the proposal is to improve the learning process by identifying the optimal ONMF temporal or spectral patterns for accurate detection. In the first stage, the time-frequency representation of the input PCG signal is computed. Next, band-pass filtering is performed to locate the spectral range that is most relevant for the presence of such cardiac abnormalities. In the second stage, the temporal and spectral cardiac structures are extracted using the ONMF approach. These structures are utilized in the third stage and fed into the CNN architecture to detect abnormal heart sounds. RESULTS: Several state-of-the-art CNN architectures, such as LeNet5, AlexNet, ResNet50, VGG16 and GoogLeNet, have been evaluated to determine the effectiveness of using ONMF temporal features for VHD detection. The results reveal that the integration of ONMF temporal features with a CNN classifier significantly improve VHD detection. Specifically, the proposed approach achieves an accuracy improvement of approximately 45% when ONMF spectral features are used and 35% when time-frequency features from the short-time Fourier transform (STFT) spectrogram are used. Additionally, feeding ONMF temporal features into low-complexity CNN architectures yields competitive results comparable to those obtained with complex architectures. CONCLUSIONS: The temporal structure factorized by ONMF plays a critical role in distinguishing between normal heart sounds and abnormal heart sounds since the repeatability of normal heart cycles is disrupted by the presence of cardiac abnormalities. Consequently, the results highlight the importance of appropriate input data representation in the learning process of CNN models in the biomedical field of valvular heart sound detection.