Since early identification of potential critical patients in the Emergency Department (ED) can lower mortality and morbidity, this study seeks to develop a machine learning model capable of predicting possible critical outcomes based on the history and vital signs routinely collected at triage. We compare emergency physicians and the predictive performance of the machine learning model. Predictors including patients' chief complaints, present illness, past medical history, vital signs, and demographic data of adult patients (aged ≥ 18 years) visiting the ED at Shuang-Ho Hospital in New Taipei City, Taiwan, are extracted from the hospital's electronic health records. Critical outcomes are defined as in-hospital cardiac arrest (IHCA) or intensive care unit (ICU) admission. A clinical narrative-aware deep neural network was developed to handle the text-intensive data and standardized numerical data, which is compared against other machine learning models. After this, emergency physicians were asked to predict possible clinical outcomes of thirty visits that were extracted randomly from our dataset, and their results were further compared to our machine learning model. A total of 4,308 (2.5 %) out of the 171,275 adult visits to the ED included in this study resulted in critical outcomes. The area under the receiver operating characteristic curve (AUROC) and the area under the precision-recall curve (AUPRC) of our proposed prediction model is 0.874 and 0.207, respectively, which not only outperforms the other machine learning models, but even has better sensitivity (0.95 vs 0.41) and accuracy (0.90 vs 0.67) as compared to the emergency physicians. This model is sensitive and accurate in predicting critical outcomes and highlights the potential to use predictive analytics to support post-triage decision-making.

Product-specific guidances (PSGs) recommended by the United States Food and Drug Administration (FDA) are instrumental to promote and guide generic drug product development. To assess a PSG, the FDA assessor needs to take extensive time and effort to manually retrieve supportive drug information of absorption, distribution, metabolism, and excretion (ADME) from the reference listed drug labeling. In this work, we leveraged the state-of-the-art pre-trained language models to automatically label the ADME paragraphs in the pharmacokinetics section from the FDA-approved drug labeling to facilitate PSG assessment. We applied a transfer learning approach by fine-tuning the pre-trained Bidirectional Encoder Representations from Transformers (BERT) model to develop a novel application of ADME semantic labeling, which can automatically retrieve ADME paragraphs from drug labeling instead of manual work. We demonstrate that fine-tuning the pre-trained BERT model can outperform conventional machine learning techniques, achieving up to 12.5% absolute F1 improvement. To our knowledge, we were the first to successfully apply BERT to solve the ADME semantic labeling task. We further assessed the relative contribution of pre-training and fine-tuning to the overall performance of the BERT model in the ADME semantic labeling task using a series of analysis methods, such as attention similarity and layer-based ablations. Our analysis revealed that the information learned via fine-tuning is focused on task-specific knowledge in the top layers of the BERT, whereas the benefit from the pre-trained BERT model is from the bottom layers.

PURPOSE: Recent developments in the field of artificial intelligence and acoustics have made it possible to objectively monitor cough in clinical and ambulatory settings. We hypothesized that time patterns of objectively measured cough in COVID-19 patients could predict clinical prognosis and help rapidly identify patients at high risk of intubation or death. METHODS: One hundred and twenty-three patients hospitalized with COVID-19 were enrolled at University of Florida Health Shands and the Centre Hospitalier de l'Université de Montréal. Patients' cough was continuously monitored digitally along with clinical severity of disease until hospital discharge, intubation, or death. The natural history of cough in hospitalized COVID-19 disease was described and logistic models fitted on cough time patterns were used to predict clinical outcomes. RESULTS: In both cohorts, higher early coughing rates were associated with more favorable clinical outcomes. The transitional cough rate, or maximum cough per hour rate predicting unfavorable outcomes, was 3·40 and the AUC for cough frequency as a predictor of unfavorable outcomes was 0·761. The initial 6 h (0·792) and 24 h (0·719) post-enrolment observation periods confirmed this association and showed similar predictive value. INTERPRETATION: Digital cough monitoring could be used as a prognosis biomarker to predict unfavorable clinical outcomes in COVID-19 disease. With early sampling periods showing good predictive value, this digital biomarker could be combined with clinical and paraclinical evaluation and is well adapted for triaging patients in overwhelmed or resources-limited health programs.

OBJECTIVE: To identify and synthesise research on applications of natural language processing (NLP) for information extraction and retrieval from clinical notes in dentistry. MATERIALS AND METHODS: A predefined search strategy was applied in EMBASE, CINAHL and Medline. Studies eligible for inclusion were those that that described, evaluated, or applied NLP to clinical notes containing either human or simulated patient information. Quality of the study design and reporting was independently assessed based on a set of questions derived from relevant tools including CHecklist for critical Appraisal and data extraction for systematic Reviews of prediction Modelling Studies (CHARMS). A narrative synthesis was conducted to present the results. RESULTS: Of the 17 included studies, 10 developed and evaluated NLP methods and 7 described applications of NLP-based information retrieval methods in dental records. Studies were published between 2015 and 2021, most were missing key details needed for reproducibility, and there was no consistency in design or reporting. The 10 studies developing or evaluating NLP methods used document classification or entity extraction, and 4 compared NLP methods to non-NLP methods. The quality of reporting on NLP studies in dentistry has modestly improved over time. CONCLUSIONS: Study design heterogeneity and incomplete reporting of studies currently limits our ability to synthesise NLP applications in dental records. Standardisation of reporting and improved connections between NLP methods and applied NLP in dentistry may improve how we can make use of clinical notes from dentistry in population health or decision support systems. PROTOCOL REGISTRATION: PROSPERO CRD42021227823.

In clinical research as well as patient care, structured documentation of findings is an important task. In many cases, this is achieved by means of electronic case report forms (eCRF) using corresponding information technology systems. To avoid double data entry, eCRF systems can be integrated with electronic health records (EHR). However, when researchers from different institutions collaborate in collecting data, they often use a single joint eCRF system on the Internet. In this case, integration with EHR systems is not possible in most cases due to information security and data protection restrictions. To overcome this shortcoming, we propose a novel architecture for a federated electronic data capture system (fEDC). Four key requirements were identified for fEDC: Definitions of forms have to be available in a reliable and controlled fashion, integration with electronic health record systems must be possible, patient data should be under full local control until they are explicitly transferred for joint analysis, and the system must support data sharing principles accepted by the scientific community for both data model and data captured. With our approach, sites participating in a joint study can run their own instance of an fEDC system that complies with local standards (such as being behind a network firewall) while also being able to benefit from using identical form definitions by sharing metadata in the Operational Data Model (ODM) format published by the Clinical Data Interchange Standards Consortium (CDISC) throughout the collaboration. The fEDC architecture was validated with a working open-source prototype at five German university hospitals. The fEDC architecture provides a novel approach with the potential to significantly improve collaborative data capture: Efforts for data entry are reduced and at the same time, data quality is increased since barriers for integrating with local electronic health record systems are lowered. Further, metadata are shared and patient privacy is ensured at a high level.

Electronic health records (EHR) are collected as a routine part of healthcare delivery, and have great potential to be utilized to improve patient health outcomes. They contain multiple years of health information to be leveraged for risk prediction, disease detection, and treatment evaluation. However, they do not have a consistent, standardized format across institutions, particularly in the United States, and can present significant analytical challenges- they contain multi-scale data from heterogeneous domains and include both structured and unstructured data. Data for individual patients are collected at irregular time intervals and with varying frequencies. In addition to the analytical challenges, EHR can reflect inequity- patients belonging to different groups will have differing amounts of data in their health records. Many of these issues can contribute to biased data collection. The consequence is that the data for under-served groups may be less informative partly due to more fragmented care, which can be viewed as a type of missing data problem. For EHR data in this complex form, there is currently no framework for introducing realistic missing values. There has also been little to no work in assessing the impact of missing data in EHR. In this work, we first introduce a terminology to define three levels of EHR data and then propose a novel framework for simulating realistic missing data scenarios in EHR to adequately assess their impact on predictive modeling. We incorporate the use of a medical knowledge graph to capture dependencies between medical events to create a more realistic missing data framework. In an intensive care unit setting, we found that missing data have greater negative impact on the performance of disease prediction models in groups that tend to have less access to healthcare, or seek less healthcare. We also found that the impact of missing data on disease prediction models is stronger when using the knowledge graph framework to introduce realistic missing values as opposed to random event removal.

BACKGROUND AND OBJECTIVES: Named Entity Recognition (NER) and Relation Extraction (RE) are two of the most studied tasks in biomedical Natural Language Processing (NLP). The detection of specific terms and entities and the relationships between them are key aspects for the development of more complex automatic systems in the biomedical field. In this work, we explore transfer learning techniques for incorporating information about negation into systems performing NER and RE. The main purpose of this research is to analyse to what extent the successful detection of negated entities in separate tasks helps in the detection of biomedical entities and their relationships. METHODS: Three neural architectures are proposed in this work, all of them mainly based on Bidirectional Long Short-Term Memory (Bi-LSTM) networks and Conditional Random Fields (CRFs). While the first architecture is devoted to detecting triggers and scopes of negated entities in any domain, two specific models are developed for performing isolated NER tasks and joint NER and RE tasks in the biomedical domain. Then, weights related to negation detection learned by the first architecture are incorporated into those last models. Two different languages, Spanish and English, are taken into account in the experiments. RESULTS: Performance of the biomedical models is analysed both when the weights of the neural networks are randomly initialized, and when weights from the negation detection model are incorporated into them. Improvements of around 3.5% of F-Measure in the English language and more than 7% in the Spanish language are achieved in the NER task, while the NER+RE task increases F-Measure scores by more than 13% for the NER submodel and around 2% for the RE submodel. CONCLUSIONS: The obtained results allow us to conclude that negation-based transfer learning techniques are appropriate for performing biomedical NER and RE tasks. These results highlight the importance of detecting negation for improving the identification of biomedical entities and their relationships. The explored techniques show robustness by maintaining consistent results and improvements across different tasks and languages.

Segmentation of rectal cancerous regions from Magnetic Resonance (MR) images can help doctor define the extent of the rectal cancer and judge the severity of rectal cancer, so rectal tumor segmentation is crucial to improve the accuracy of rectal cancer diagnosis. However, accurate segmentation of rectal cancerous regions remains a challenging task due to the shape of rectal tumor has significant variations and the tumor and surrounding tissue are indistinguishable. In addition, in the early research on rectal tumor segmentation, most deep learning methods were based on convolutional neural networks (CNNs), and traditional CNN have small receptive field, which can only capture local information while ignoring the global information of the image. Nevertheless, the global information plays a crucial role in rectal tumor segmentation, so traditional CNN-based methods usually cannot achieve satisfactory segmentation results. In this paper, we propose an encoder-decoder network named Dual Parallel Net (DuPNet), which fuses transformer and classical CNN for capturing both global and local information. Meanwhile, as for capture features at different scales as well as to avoid accuracy loss and parameters reduction, we design a feature adaptive block (FAB) in skip connection between encoder and decoder. Furthermore, in order to utilize the apriori information of rectal tumor shape effectively, we design a Gaussian Mixture prior and embed it in self-attention mechanism of the transformer, leading to robust feature representation and accurate segmentation results. We have performed extensive ablation experiments to verify the effectiveness of our proposed dual parallel encoder, FAB and Gaussian Mixture prior on the dataset from the Shanxi Cancer Hospital. In the experimental comparison with the state-of-the-art methods, our method achieved a Mean Intersection over Union (MIoU) of 89.34% on the test set. In addition to that, we evaluated the generalizability of our method on the dataset from Xinhua Hospital, the promising results verify the superiority of our method.

BACKGROUND: Many important clinical decisions require causal knowledge (CK) to take action. Although many causal knowledge bases for medicine have been constructed, a comprehensive evaluation based on real-world data and methods for handling potential knowledge noise are still lacking. OBJECTIVE: The objectives of our study are threefold: (1) propose a framework for the construction of a large-scale and high-quality causal knowledge graph (CKG); (2) design the methods for knowledge noise reduction to improve the quality of the CKG; (3) evaluate the knowledge completeness and accuracy of the CKG using real-world data. MATERIAL AND METHODS: We extracted causal triples from three knowledge sources (SemMedDB, UpToDate and Churchill's Pocketbook of Differential Diagnosis) based on rule methods and language models, performed ontological encoding, and then designed semantic modeling between electronic health record (EHR) data and the CKG to complete knowledge instantiation. We proposed two graph pruning strategies (co-occurrence ratio and causality ratio) to reduce the potential noise introduced by SemMedDB. Finally, the evaluation was carried out by taking the diagnostic decision support (DDS) of diabetic nephropathy (DN) as a real-world case. The data originated from a Chinese hospital EHR system from October 2010 to October 2020. The knowledge completeness and accuracy of the CKG were evaluated based on three state-of-the-art embedding methods (R-GCN, MHGRN and MedPath), the annotated clinical text and the expert review, respectively. RESULTS: This graph included 153,289 concepts and 1,719,968 causal triples. A total of 1427 inpatient data were used for evaluation. Better results were achieved by combining three knowledge sources than using only SemMedDB (three models: area under the receiver operating characteristic curve (AUC): p < 0.01, F1: p < 0.01), and the graph covered 93.9 % of the causal relations between diseases and diagnostic evidence recorded in clinical text. Causal relations played a vital role in all relations related to disease progression for DDS of DN (three models: AUC: p > 0.05, F1: p > 0.05), and after pruning, the knowledge accuracy of the CKG was significantly improved (three models: AUC: p < 0.01, F1: p < 0.01; expert review: average accuracy: + 5.5 %). CONCLUSIONS: The results demonstrated that our proposed CKG could completely and accurately capture the abstract CK under the concrete EHR data, and the pruning strategies could improve the knowledge accuracy of our CKG. The CKG has the potential to be applied to the DDS of diseases.

BACKGROUND AND OBJECTIVE: Work-related stress affects a large part of today's workforce and is known to have detrimental effects on physical and mental health. Continuous and unobtrusive stress detection may help prevent and reduce stress by providing personalised feedback and allowing for the development of just-in-time adaptive health interventions for stress management. Previous studies on stress detection in work environments have often struggled to adequately reflect real-world conditions in controlled laboratory experiments. To close this gap, in this paper, we present a machine learning methodology for stress detection based on multimodal data collected from unobtrusive sources in an experiment simulating a realistic group office environment (N=90). METHODS: We derive mouse, keyboard and heart rate variability features to detect three levels of perceived stress, valence and arousal with support vector machines, random forests and gradient boosting models using 10-fold cross-validation. We interpret the contributions of features to the model predictions with SHapley Additive exPlanations (SHAP) value plots. RESULTS: The gradient boosting models based on mouse and keyboard features obtained the highest average F1 scores of 0.625, 0.631 and 0.775 for the multiclass prediction of perceived stress, arousal and valence, respectively. Our results indicate that the combination of mouse and keyboard features may be better suited to detect stress in office environments than heart rate variability, despite physiological signal-based stress detection being more established in theory and research. The analysis of SHAP value plots shows that specific mouse movement and typing behaviours may characterise different levels of stress. CONCLUSIONS: Our study fills different methodological gaps in the research on the automated detection of stress in office environments, such as approximating real-life conditions in a laboratory and combining physiological and behavioural data sources. Implications for field studies on personalised, interpretable ML-based systems for the real-time detection of stress in real office environments are also discussed.

Healthcare datasets obtained from Electronic Health Records have proven to be extremely useful for assessing associations between patients' predictors and outcomes of interest. However, these datasets often suffer from missing values in a high proportion of cases, whose removal may introduce severe bias. Several multiple imputation algorithms have been proposed to attempt to recover the missing information under an assumed missingness mechanism. Each algorithm presents strengths and weaknesses, and there is currently no consensus on which multiple imputation algorithm works best in a given scenario. Furthermore, the selection of each algorithm's parameters and data-related modeling choices are also both crucial and challenging. In this paper we propose a novel framework to numerically evaluate strategies for handling missing data in the context of statistical analysis, with a particular focus on multiple imputation techniques. We demonstrate the feasibility of our approach on a large cohort of type-2 diabetes patients provided by the National COVID Cohort Collaborative (N3C) Enclave, where we explored the influence of various patient characteristics on outcomes related to COVID-19. Our analysis included classic multiple imputation techniques as well as simple complete-case Inverse Probability Weighted models. Extensive experiments show that our approach can effectively highlight the most promising and performant missing-data handling strategy for our case study. Moreover, our methodology allowed a better understanding of the behavior of the different models and of how it changed as we modified their parameters. Our method is general and can be applied to different research fields and on datasets containing heterogeneous types.

The meaningful use of electronic health records (EHR) continues to progress in the digital era with clinical decision support systems augmented by artificial intelligence. A priority in improving provider experience is to overcome information overload and reduce the cognitive burden so fewer medical errors and cognitive biases are introduced during patient care. One major type of medical error is diagnostic error due to systematic or predictable errors in judgement that rely on heuristics. The potential for clinical natural language processing (cNLP) to model diagnostic reasoning in humans with forward reasoning from data to diagnosis and potentially reduce cognitive burden and medical error has not been investigated. Existing tasks to advance the science in cNLP have largely focused on information extraction and named entity recognition through classification tasks. We introduce a novel suite of tasks coined as Diagnostic Reasoning Benchmarks, Dr.Bench, as a new benchmark for developing and evaluating cNLP models with clinical diagnostic reasoning ability. The suite includes six tasks from ten publicly available datasets addressing clinical text understanding, medical knowledge reasoning, and diagnosis generation. DR.BENCH is the first clinical suite of tasks designed to be a natural language generation framework to evaluate pre-trained language models for diagnostic reasoning. The goal of DR. BENCH is to advance the science in cNLP to support downstream applications in computerized diagnostic decision support and improve the efficiency and accuracy of healthcare providers during patient care. We fine-tune and evaluate the state-of-the-art generative models on DR.BENCH. Experiments show that with domain adaptation pre-training on medical knowledge, the model demonstrated opportunities for improvement when evaluated in DR. BENCH. We share DR. BENCH as a publicly available GitLab repository with a systematic approach to load and evaluate models for the cNLP community. We also discuss the carbon footprint produced during the experiments and encourage future work on DR.BENCH to report the carbon footprint.

OBJECTIVE: The study aims to investigate whether machine learning-based predictive models for cardiovascular disease (CVD) risk assessment show equivalent performance across demographic groups (such as race and gender) and if bias mitigation methods can reduce any bias present in the models. This is important as systematic bias may be introduced when collecting and preprocessing health data, which could affect the performance of the models on certain demographic sub-cohorts. The study is to investigate this using electronic health records data and various machine learning models. METHODS: The study used large de-identified Electronic Health Records data from Vanderbilt University Medical Center. Machine learning (ML) algorithms including logistic regression, random forest, gradient-boosting trees, and long short-term memory were applied to build multiple predictive models. Model bias and fairness were evaluated using equal opportunity difference (EOD, 0 indicates fairness) and disparate impact (DI, 1 indicates fairness). In our study, we also evaluated the fairness of a non-ML baseline model, the American Heart Association (AHA) Pooled Cohort Risk Equations (PCEs). Moreover, we compared the performance of three different de-biasing methods: removing protected attributes (e.g., race and gender), resampling the imbalanced training dataset by sample size, and resampling by the proportion of people with CVD outcomes. RESULTS: The study cohort included 109,490 individuals (mean [SD] age 47.4 [14.7] years; 64.5% female; 86.3% White; 13.7% Black). The experimental results suggested that most ML models had smaller EOD and DI than PCEs. For ML models, the mean EOD ranged from -0.001 to 0.018 and the mean DI ranged from 1.037 to 1.094 across race groups. There was a larger EOD and DI across gender groups, with EOD ranging from 0.131 to 0.136 and DI ranging from 1.535 to 1.587. For debiasing methods, removing protected attributes didn't significantly reduced the bias for most ML models. Resampling by sample size also didn't consistently decrease bias. Resampling by case proportion reduced the EOD and DI for gender groups but slightly reduced accuracy in many cases. CONCLUSIONS: Among the VUMC cohort, both PCEs and ML models were biased against women, suggesting the need to investigate and correct gender disparities in CVD risk prediction. Resampling by proportion reduced the bias for gender groups but not for race groups.

Invasive fungal infections (IFI) are particularly dangerous to high-risk patients with haematological malignancies and are responsible for excessive mortality and delays in cancer therapy. Surveillance of IFI in clinical settings offers an opportunity to identify potential risk factors and evaluate new therapeutic strategies. However, manual surveillance is both time- and resource-intensive. As part of a broader project aimed to develop a system for automated IFI surveillance by leveraging electronic medical records, we present our approach to detecting evidence of IFI in the key diagnostic domain of histopathology. Using natural language processing (NLP), we analysed cytology and histopathology reports to identify IFI-positive reports. We compared a conventional bag-of-words classification model to a method that relies on concept-level annotations. Although the investment to prepare data supporting concept annotations is substantial, extracting targeted information specific to IFI as a pre-processing step increased the performance of the classifier from the PR AUC of 0.84 to 0.92 and enabled model interpretability. We have made publicly available the annotated dataset of 283 reports, the Cytology and Histopathology IFI Reports corpus (CHIFIR), to allow the clinical NLP research community to further build on our results.

Learning latent representations of patients with a target disease is a core problem in a broad range of downstream applications, such as clinical endpoint prediction. The suffering of patients may have multiple subtypes with certain similarities and differences, which need to be addressed for learning effective patient representation to facilitate the downstream tasks. However, existing studies either ignore the distinction of disease subtypes to learn disease-level representations, or neglect the correlations between subtypes and only learn disease subtype-level representations, which affects the performance of patient representation learning. To alleviate this problem, we studied how to effectively integrate data from all disease subtypes to improve the representation of each subtype. Specifically, we proposed a knowledge-aware shared-private neural network model to explicitly use disease-oriented knowledge and learn shared and specific representations from the disease and its subtype perspectives. To evaluate the feasibility of the proposed model, we conducted a particular downstream task, i.e., clinical endpoint prediction, on the basis of the learned patient presentations. The results on the real-world clinical datasets demonstrated that our model could yield a significant improvement over state-of-the-art models.

Interpreting medical images such as chest X-ray images and retina images is an essential step for diagnosing and treating relevant diseases. Proposing automatic and reliable medical report generation systems can reduce the time-consuming workload, improve efficiencies of clinical workflows, and decrease practical variations between different clinical professionals. Many recent approaches based on image-encoder and language-decoder structure have been proposed to tackle this task. However, some technical challenges remain to be solved, including the fusion efficacy between the language and visual cues and the difficulty of obtaining an effective pre-trained image feature extractor for medical-specific tasks. In this work, we proposed the weighted query-key interacting attention module, including both the second-order and first-order interactions. Compared with the conventional scaled dot-product attention, this design generates a strong fusion mechanism between language and visual signals. In addition, we also proposed the contrastive pre-training step to reduce the domain gap between the image encoder and the target dataset. To test the generalizability of our learning scheme, we collected and verified our model on the world-first multi-modality retina report generation dataset referred to as Retina ImBank and another large-scale retina Chinese-based report dataset referred to as Retina Chinese. These two datasets will be made publicly available and serve as benchmarks to encourage further research exploration in this field. From our experimental results, we demonstrate that our proposed method has outperformed multiple state-of-the-art image captioning and medical report generation methods on IU X-RAY, MIMIC-CXR, Retina ImBank, and Retina Chinese datasets.