$\mathtt {Deepr}$: A Convolutional Net for Medical Records

Comparing neural language models for medical concept representation and patient trajectory prediction

Effective representation of medical concepts is crucial for secondary analyses of electronic health records. Neural language models have shown promise in automatically deriving medical concept representations from clinical data. However, the comparative performance of different language models for creating these empirical representations, and the extent to which they encode medical semantics, has not been extensively studied. This study aims to address this gap by evaluating the effectiveness of three popular language models - word2vec, fastText, and GloVe - in creating medical concept embeddings that capture their semantic meaning. By using a large dataset of digital health records, we created patient trajectories and used them to train the language models. We then assessed the ability of the learned embeddings to encode semantics through an explicit comparison with biomedical terminologies, and implicitly by predicting patient outcomes and trajectories with different levels of available information. Our qualitative analysis shows that empirical clusters of embeddings learned by fastText exhibit the highest similarity with theoretical clustering patterns obtained from biomedical terminologies, with a similarity score between empirical and theoretical clusters of 0.88, 0.80, and 0.92 for diagnosis, procedure, and medication codes, respectively. Conversely, for outcome prediction, word2vec and GloVe tend to outperform fastText, with the former achieving AUROC as high as 0.78, 0.62, and 0.85 for length-of-stay, readmission, and mortality prediction, respectively. In predicting medical codes in patient trajectories, GloVe achieves the highest performance for diagnosis and medication codes (AUPRC of 0.45 and of 0.81, respectively) at the highest level of the semantic hierarchy, while fastText outperforms the other models for procedure codes (AUPRC of 0.66). Our study demonstrates that subword information is crucial for learning medical concept representations, but global embedding vectors are better suited for more high-level downstream tasks, such as trajectory prediction. Thus, these models can be harnessed to learn representations that convey clinical meaning, and our insights highlight the potential of using machine learning techniques to semantically encode medical data.

Enhancing Patient Outcome Prediction Through Deep Learning With Sequential Diagnosis Codes From Structured Electronic Health Record Data: Systematic Review

BACKGROUND

The use of structured electronic health records in health care systems has grown rapidly. These systems collect huge amounts of patient information, including diagnosis codes representing temporal medical history. Sequential diagnostic information has proven valuable for predicting patient outcomes. However, the extent to which these types of data have been incorporated into deep learning (DL) models has not been examined.

OBJECTIVE

This systematic review aims to describe the use of sequential diagnostic data in DL models, specifically to understand how these data are integrated, whether sample size improves performance, and whether the identified models are generalizable.

METHODS

Relevant studies published up to May 15, 2023, were identified using 4 databases: PubMed, Embase, IEEE Xplore, and Web of Science. We included all studies using DL algorithms trained on sequential diagnosis codes to predict patient outcomes. We excluded review articles and non-peer-reviewed papers. We evaluated the following aspects in the included papers: DL techniques, characteristics of the dataset, prediction tasks, performance evaluation, generalizability, and explainability. We also assessed the risk of bias and applicability of the studies using the Prediction Model Study Risk of Bias Assessment Tool (PROBAST). We used the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) checklist to report our findings.

RESULTS

Of the 740 identified papers, 84 (11.4%) met the eligibility criteria. Publications in this area increased yearly. Recurrent neural networks (and their derivatives; 47/84, 56%) and transformers (22/84, 26%) were the most commonly used architectures in DL-based models. Most studies (45/84, 54%) presented their input features as sequences of visit embeddings. Medications (38/84, 45%) were the most common additional feature. Of the 128 predictive outcome tasks, the most frequent was next-visit diagnosis (n=30, 23%), followed by heart failure (n=18, 14%) and mortality (n=17, 13%). Only 7 (8%) of the 84 studies evaluated their models in terms of generalizability. A positive correlation was observed between training sample size and model performance (area under the receiver operating characteristic curve; P=.02). However, 59 (70%) of the 84 studies had a high risk of bias.

CONCLUSIONS

The application of DL for advanced modeling of sequential medical codes has demonstrated remarkable promise in predicting patient outcomes. The main limitation of this study was the heterogeneity of methods and outcomes. However, our analysis found that using multiple types of features, integrating time intervals, and including larger sample sizes were generally related to an improved predictive performance. This review also highlights that very few studies (7/84, 8%) reported on challenges related to generalizability and less than half (38/84, 45%) of the studies reported on challenges related to explainability. Addressing these shortcomings will be instrumental in unlocking the full potential of DL for enhancing health care outcomes and patient care.

TRIAL REGISTRATION

PROSPERO CRD42018112161; https://tinyurl.com/yc6h9rwu.

Mdpg: a novel multi-disease diagnosis prediction method based on patient knowledge graphs

Diagnosis prediction, a key factor in enhancing healthcare efficiency, remains a focal point in clinical decision support research. However, the time-series, sparse and multi-noise characteristics of electronic health record (EHR) data make it a great challenge. Existing methods commonly address these issues using RNNs and incorporating medical prior knowledge from medical knowledge bases, but they neglect the local spatial characteristics and spatial-temporal correlation of the data. Consequently, we propose MDPG, a diagnosis prediction model based on patient knowledge graphs. Initially, we represent the electronic visit records of patients as a patient-centered temporal knowledge graph, capturing the local spatial structure and temporal characteristics of the visit information. Subsequently, we design the spatial graph convolution block, temporal self-attention block, and spatial-temporal synchronous graph convolution block to capture the spatial, temporal, and spatial-temporal correlations embedded in them, respectively. Ultimately, we accomplish the prediction of patients' future states through multi-label classification. We conduct comprehensive experiments on two real-world datasets independently and evaluate the results using visit-level precision@k and code-level accuracy@k metrics. The experimental results demonstrate that MDPG outperforms all baseline models, yielding the best performance.

Multi-task heterogeneous graph learning on electronic health records

Learning electronic health records (EHRs) has received emerging attention because of its capability to facilitate accurate medical diagnosis. Since the EHRs contain enriched information specifying complex interactions between entities, modeling EHRs with graphs is shown to be effective in practice. The EHRs, however, present a great degree of heterogeneity, sparsity, and complexity, which hamper the performance of most of the models applied to them. Moreover, existing approaches modeling EHRs often focus on learning the representations for a single task, overlooking the multi-task nature of EHR analysis problems and resulting in limited generalizability across different tasks. In view of these limitations, we propose a novel framework for EHR modeling, namely MulT-EHR (Multi-Task EHR), which leverages a heterogeneous graph to mine the complex relations and model the heterogeneity in the EHRs. To mitigate the large degree of noise, we introduce a denoising module based on the causal inference framework to adjust for severe confounding effects and reduce noise in the EHR data. Additionally, since our model adopts a single graph neural network for simultaneous multi-task prediction, we design a multi-task learning module to leverage the inter-task knowledge to regularize the training process. Extensive empirical studies on MIMIC-III and MIMIC-IV datasets validate that the proposed method consistently outperforms the state-of-the-art designs in four popular EHR analysis tasks - drug recommendation, and predictions of the length of stay, mortality, and readmission. Thorough ablation studies demonstrate the robustness of our method upon variations to key components and hyperparameters.

CPLLM: Clinical prediction with large language models

We present Clinical Prediction with Large Language Models (CPLLM), a method that involves fine-tuning a pre-trained Large Language Model (LLM) for predicting clinical disease and readmission. We utilized quantization and fine-tuned the LLM using prompts. For diagnostic predictions, we predicted whether patients would be diagnosed with a target disease during their next visit or in the subsequent diagnosis, leveraging their historical medical records. We compared our results to various baselines, including Retain and Med-BERT, the latter of which is the current state-of-the-art model for disease prediction using temporal structured EHR data. In addition, we also evaluated CPLLM's utility in predicting hospital readmission and compared our method's performance with benchmark baselines. Our experiments ultimately revealed that our proposed method, CPLLM, surpasses all the tested models in terms of PR-AUC and ROC-AUC metrics, providing state-of-the-art performance as a tool for predicting disease diagnosis and patient hospital readmission without requiring pre-training on medical data. Such a method can be easily implemented and integrated into the clinical workflow to help care providers plan next steps for their patients.

Multimodal Attention Network for Dementia Prediction

The early identification of an individual's dementia risk is crucial for disease prevention and the design of insurance products in an aging society. This study aims to accurately predict the future incidence risk of dementia in individuals by leveraging the advantages of neural networks. This is, however, complicated by the high dimensionality and sparsity of the International Classification of Diseases (ICD) codes when utilizing data from Taiwan's National Health Insurance, which includes individual profiles and medical records. Inspired by the click-through rate (CTR) problem in recommendation systems, where future user behavior is predicted based on their past consumption records, we address these challenges with a multimodal attention network for dementia (MAND), which incorporates an ICD code embedding layer and multihead self-attention to encode ICD codes and capture interactions among diseases. Additionally, we investigate the applicability of several CTR methods to the dementia prediction problem. MAND achieves an AUC of 0.9010, surpassing traditional CTR models and demonstrating its effectiveness. The highly flexible pipelined design allows for module replacement to meet specific requirements. Furthermore, the analysis of attention scores reveals diseases highly correlated with dementia, aligning with prior research and emphasizing the interpretability of the model. This research deepens our understanding of the diseases associated with dementia, and the accurate prediction provided can serve as an early warning for dementia occurrence, aiding in its prevention.

Prediction of mortality events of patients with acute heart failure in intensive care unit based on deep neural network

BACKGROUND

Acute heart failure (AHF) in the intensive care unit (ICU) is characterized by its criticality, rapid progression, complex and changeable condition, and its pathophysiological process involves the interaction of multiple organs and systems. This makes it difficult to predict in-hospital mortality events comprehensively and accurately. Traditional analysis methods based on statistics and machine learning suffer from insufficient model performance, poor accuracy caused by prior dependence, and difficulty in adequately considering the complex relationships between multiple risk factors. Therefore, the application of deep neural network (DNN) techniques to the specific scenario, predicting mortality events of patients with AHF under intensive care, has become a research frontier.

METHODS

This research utilized the MIMIC-IV critical care database as the primary data source and employed the synthetic minority over-sampling technique (SMOTE) to balance the dataset. Deep neural network models-backpropagation neural network (BPNN) and recurrent neural network (RNN), which are based on electronic medical record data mining, were employed to investigate the in-hospital death event judgment task of patients with AHF under intensive care. Additionally, multiple single machine learning models and ensemble learning models were constructed for comparative experiments. Moreover, we achieved various optimal performance combinations by modifying the classification threshold of deep neural network models to address the diverse real-world requirements in the ICU. Finally, we conducted an interpretable deep model using SHapley Additive exPlanations (SHAP) to uncover the most influential medical record features for each patient from the aspects of global and local interpretation.

RESULTS

In terms of model performance in this scenario, deep neural network models outperform both single machine learning models and ensemble learning models, achieving the highest Accuracy, Precision, Recall, F1 value, and Area under the ROC curve, which can reach 0.949, 0.925, 0.983, 0.953, and 0.987 respectively. SHAP value analysis revealed that the ICU scores (APSIII, OASIS, SOFA) are significantly correlated with the occurrence of in-hospital fatal events.

CONCLUSIONS

Our study underscores that DNN-based mortality event classifier offers a novel intelligent approach for forecasting and assessing the prognosis of AHF patients in the ICU. Additionally, the ICU scores stand out as the most predictive features, which implies that in the decision-making process of the models, ICU scores can provide the most crucial information, making the greatest positive or negative contribution to influence the incidence of in-hospital mortality among patients with acute heart failure.

Early detection of pediatric health risks using maternal and child health data

Machine learning (ML)-driven diagnosis systems are particularly relevant in pediatrics given the well-documented impact of early-life health conditions on later-life outcomes. Yet, early identification of diseases and their subsequent impact on length of hospital stay for this age group has so far remained uncharacterized, likely because access to relevant health data is severely limited. Thanks to a confidential data use agreement with the California Department of Health Care Access and Information, we introduce Ped-BERT: a state-of-the-art deep learning model that accurately predicts the likelihood of 100+ conditions and the length of stay in a pediatric patient's next medical visit. We link mother-specific pre- and postnatal period health information to pediatric patient hospital discharge and emergency room visits. Our data set comprises 513.9K mother-baby pairs and contains medical diagnosis codes, length of stay, as well as temporal and spatial pediatric patient characteristics, such as age and residency zip code at the time of visit. Following the popular bidirectional encoder representations from the transformers (BERT) approach, we pre-train Ped-BERT via the masked language modeling objective to learn embedding features for the diagnosis codes contained in our data. We then continue to fine-tune our model to accurately predict primary diagnosis outcomes and length of stay for a pediatric patient's next visit, given the history of previous visits and, optionally, the mother's pre- and postnatal health information. We find that Ped-BERT generally outperforms contemporary and state-of-the-art classifiers when trained with minimum features. We also find that incorporating mother health attributes leads to significant improvements in model performance overall and across all patient subgroups in our data. Our most successful Ped-BERT model configuration achieves an area under the receiver operator curve (ROC AUC) of 0.927 and an average precision score (APS) of 0.408 for the diagnosis prediction task, and a ROC AUC of 0.855 and APS of 0.815 for the length of hospital stay task. Further, we examine Ped-BERT's fairness by determining whether prediction errors are evenly distributed across various subgroups of mother-baby demographics and health characteristics, or if certain subgroups exhibit a higher susceptibility to prediction errors.

DNA-T: Deformable Neighborhood Attention Transformer for Irregular Medical Time Series

The real-world Electronic Health Records (EHRs) present irregularities due to changes in the patient's health status, resulting in various time intervals between observations and different physiological variables examined at each observation point. There have been recent applications of Transformer-based models in the field of irregular time series. However, the full attention mechanism in Transformer overly focuses on distant information, ignoring the short-term correlations of the condition. Thereby, the model is not able to capture localized changes or short-term fluctuations in patients' conditions. Therefore, we propose a novel end-to-end Deformable Neighborhood Attention Transformer (DNA-T) for irregular medical time series. The DNA-T captures local features by dynamically adjusting the receptive field of attention and aggregating relevant deformable neighborhoods in irregular time series. Specifically, we design a Deformable Neighborhood Attention (DNA) module that enables the network to attend to relevant neighborhoods by drifting the receiving field of neighborhood attention. The DNA enhances the model's sensitivity to local information and representation of local features, thereby capturing the correlation of localized changes in patients' conditions. We conduct extensive experiments to validate the effectiveness of DNA-T, outperforming existing state-of-the-art methods in predicting the mortality risk of patients. Moreover, we visualize an example to validate the effectiveness of the proposed DNA.

Deep STI: Deep Stochastic Time-series Imputation on Electronic Health Records

Electronic Health Records (EHRs) are a cornerstone of modern healthcare analytics, offering rich datasets for various disease analyses through advanced deep learning algorithms. However, the pervasive issue of missing values in EHRs significantly hampers the development and performance of these models. Addressing this challenge is crucial for enhancing clinical decision-making and patient care. Existing methods for handling missing data, ranging from simple imputation to more sophisticated approaches, often fall short of capturing the temporal dynamics inherent in EHRs. To bridge this gap, we introduce the Deep Stochastic Time-series Imputation (Deep STI) algorithm, an innovative end-to-end deep learning model that seamlessly integrates a sequence-to-sequence generative network with a prediction network. Deep STI is designed to leverage the observed time-series data in EHRs, learning to infer missing values from the temporal context with high accuracy. We evaluated Deep STI on the liver cancer data from the National Taiwan University Hospital (NTUH), Taiwan. Our results showed that Deep STI achieved better 5-year hepatocellular carcinoma predictions (19.21% in the area under the precision-recall curve) than extreme gradient boosting (18.15%) and Transformer (18.09%). The ablation study also illustrates the efficacy of our generative architecture design compared to regular imputations. This approach not only promises to improve the reliability of disease analysis in the presence of incomplete data but also sets a new standard for utilizing EHRs in predictive healthcare. Our work aims to advance the field of healthcare analytics and open new avenues for research in deep learning applications to EHRs.

Targeted-BEHRT: Deep Learning for Observational Causal Inference on Longitudinal Electronic Health Records

Observational causal inference is useful for decision-making in medicine when randomized clinical trials (RCTs) are infeasible or nongeneralizable. However, traditional approaches do not always deliver unconfounded causal conclusions in practice. The rise of "doubly robust" nonparametric tools coupled with the growth of deep learning for capturing rich representations of multimodal data offers a unique opportunity to develop and test such models for causal inference on comprehensive electronic health records (EHRs). In this article, we investigate causal modeling of an RCT-established causal association: the effect of classes of antihypertensive on incident cancer risk. We develop a transformer-based model, targeted bidirectional EHR transformer (T-BEHRT) coupled with doubly robust estimation to estimate average risk ratio (RR). We compare our model to benchmark statistical and deep learning models for causal inference in multiple experiments on semi-synthetic derivations of our dataset with various types and intensities of confounding. In order to further test the reliability of our approach, we test our model on situations of limited data. We find that our model provides more accurate estimates of relative risk [least sum absolute error (SAE) from ground truth] compared with benchmark estimations. Finally, our model provides an estimate of class-wise antihypertensive effect on cancer risk that is consistent with results derived from RCTs.

Attention-Based Deep Learning Model for Prediction of Major Adverse Cardiovascular Events in Peritoneal Dialysis Patients

Major adverse cardiovascular events (MACE) encompass pivotal cardiovascular outcomes such as myocardial infarction, unstable angina, and cardiovascular-related mortality. Patients undergoing peritoneal dialysis (PD) exhibit specific cardiovascular risk factors during the treatment, which can escalate the likelihood of cardiovascular events. Hence, the prediction and key factor analysis of MACE have assumed paramount significance for peritoneal dialysis patients. Current pathological methodologies for prognosis prediction are not only costly but also cumbersome in effectively processing electronic health records (EHRs) data with high dimensionality, heterogeneity, and time series. Therefore in this study, we propose the CVEformer, an attention-based neural network designed to predict MACE and analyze risk factors. CVEformer leverages the self-attention mechanism to capture temporal correlations among time series variables, allowing for weighted integration of variables and estimation of the probability of MACE. CVEformer first captures the correlations among heterogeneous variables through attention scores. Then, it analyzes the correlations within the time series data to identify key risk variables and predict the probability of MACE. When trained and evaluated on data from a large cohort of peritoneal dialysis patients across multiple centers, CVEformer outperforms existing models in terms of predictive performance.

Personalized event prediction for Electronic Health Records

Clinical event sequences consist of hundreds of clinical events that represent records of patient care in time. Developing accurate predictive models of such sequences is of a great importance for supporting a variety of models for interpreting/classifying the current patient condition, or predicting adverse clinical events and outcomes, all aimed to improve patient care. One important challenge of learning predictive models of clinical sequences is their patient-specific variability. Based on underlying clinical conditions, each patient's sequence may consist of different sets of clinical events (observations, lab results, medications, procedures). Hence, simple population-wide models learned from event sequences for many different patients may not accurately predict patient-specific dynamics of event sequences and their differences. To address the problem, we propose and investigate multiple new event sequence prediction models and methods that let us better adjust the prediction for individual patients and their specific conditions. The methods developed in this work pursue refinement of population-wide models to subpopulations, self-adaptation, and a meta-level model switching that is able to adaptively select the model with the best chance to support the immediate prediction. We analyze and test the performance of these models on clinical event sequences of patients in MIMIC-III database.

Patient Graph Deep Learning to Predict Breast Cancer Molecular Subtype

Breast cancer is a heterogeneous disease consisting of a diverse set of genomic mutations and clinical characteristics. The molecular subtypes of breast cancer are closely tied to prognosis and therapeutic treatment options. We investigate using deep graph learning on a collection of patient factors from multiple diagnostic disciplines to better represent breast cancer patient information and predict molecular subtype. Our method models breast cancer patient data into a multi-relational directed graph with extracted feature embeddings to directly represent patient information and diagnostic test results. We develop a radiographic image feature extraction pipeline to produce vector representation of breast cancer tumors in DCE-MRI and an autoencoder-based genomic variant embedding method to map variant assay results to a low-dimensional latent space. We leverage related-domain transfer learning to train and evaluate a Relational Graph Convolutional Network to predict the probabilities of molecular subtypes for individual breast cancer patient graphs. Our work found that utilizing information from multiple multimodal diagnostic disciplines improved the model's prediction results and produced more distinct learned feature representations for breast cancer patients. This research demonstrates the capabilities of graph neural networks and deep learning feature representation to perform multimodal data fusion and representation in the breast cancer domain.

A comparative study of model-centric and data-centric approaches in the development of cardiovascular disease risk prediction models in the UK Biobank

Aims

A diverse set of factors influence cardiovascular diseases (CVDs), but a systematic investigation of the interplay between these determinants and the contribution of each to CVD incidence prediction is largely missing from the literature. In this study, we leverage one of the most comprehensive biobanks worldwide, the UK Biobank, to investigate the contribution of different risk factor categories to more accurate incidence predictions in the overall population, by sex, different age groups, and ethnicity.

Methods and results

The investigated categories include the history of medical events, behavioural factors, socioeconomic factors, environmental factors, and measurements. We included data from a cohort of 405 257 participants aged 37-73 years and trained various machine learning and deep learning models on different subsets of risk factors to predict CVD incidence. Each of the models was trained on the complete set of predictors and subsets where each category was excluded. The results were benchmarked against QRISK3. The findings highlight that (i) leveraging a more comprehensive medical history substantially improves model performance. Relative to QRISK3, the best performing models improved the discrimination by 3.78% and improved precision by 1.80%. (ii) Both model- and data-centric approaches are necessary to improve predictive performance. The benefits of using a comprehensive history of diseases were far more pronounced when a neural sequence model, BEHRT, was used. This highlights the importance of the temporality of medical events that existing clinical risk models fail to capture. (iii) Besides the history of diseases, socioeconomic factors and measurements had small but significant independent contributions to the predictive performance.

Conclusion

These findings emphasize the need for considering broad determinants and novel modelling approaches to enhance CVD incidence prediction.

SCOPE: predicting future diagnoses in office visits using electronic health records

We propose an interpretable and scalable model to predict likely diagnoses at an encounter based on past diagnoses and lab results. This model is intended to aid physicians in their interaction with the electronic health records (EHR). To accomplish this, we retrospectively collected and de-identified EHR data of 2,701,522 patients at Stanford Healthcare over a time period from January 2008 to December 2016. A population-based sample of patients comprising 524,198 individuals (44% M, 56% F) with multiple encounters with at least one frequently occurring diagnosis codes were chosen. A calibrated model was developed to predict ICD-10 diagnosis codes at an encounter based on the past diagnoses and lab results, using a binary relevance based multi-label modeling strategy. Logistic regression and random forests were tested as the base classifier, and several time windows were tested for aggregating the past diagnoses and labs. This modeling approach was compared to a recurrent neural network based deep learning method. The best model used random forest as the base classifier and integrated demographic features, diagnosis codes, and lab results. The best model was calibrated and its performance was comparable or better than existing methods in terms of various metrics, including a median AUROC of 0.904 (IQR [0.838, 0.954]) over 583 diseases. When predicting the first occurrence of a disease label for a patient, the median AUROC with the best model was 0.796 (IQR [0.737, 0.868]). Our modeling approach performed comparably as the tested deep learning method, outperforming it in terms of AUROC (p < 0.001) but underperforming in terms of AUPRC (p < 0.001). Interpreting the model showed that the model uses meaningful features and highlights many interesting associations among diagnoses and lab results. We conclude that the multi-label model performs comparably with RNN based deep learning model while offering simplicity and potentially superior interpretability. While the model was trained and validated on data obtained from a single institution, its simplicity, interpretability and performance makes it a promising candidate for deployment.

A flexible symbolic regression method for constructing interpretable clinical prediction models

Machine learning (ML) models trained for triggering clinical decision support (CDS) are typically either accurate or interpretable but not both. Scaling CDS to the panoply of clinical use cases while mitigating risks to patients will require many ML models be intuitively interpretable for clinicians. To this end, we adapted a symbolic regression method, coined the feature engineering automation tool (FEAT), to train concise and accurate models from high-dimensional electronic health record (EHR) data. We first present an in-depth application of FEAT to classify hypertension, hypertension with unexplained hypokalemia, and apparent treatment-resistant hypertension (aTRH) using EHR data for 1200 subjects receiving longitudinal care in a large healthcare system. FEAT models trained to predict phenotypes adjudicated by chart review had equivalent or higher discriminative performance (p < 0.001) and were at least three times smaller (p < 1 × 10-6) than other potentially interpretable models. For aTRH, FEAT generated a six-feature, highly discriminative (positive predictive value = 0.70, sensitivity = 0.62), and clinically intuitive model. To assess the generalizability of the approach, we tested FEAT on 25 benchmark clinical phenotyping tasks using the MIMIC-III critical care database. Under comparable dimensionality constraints, FEAT's models exhibited higher area under the receiver-operating curve scores than penalized linear models across tasks (p < 6 × 10-6). In summary, FEAT can train EHR prediction models that are both intuitively interpretable and accurate, which should facilitate safe and effective scaling of ML-triggered CDS to the panoply of potential clinical use cases and healthcare practices.

Predicting Visit Cost of Obstructive Sleep Apnea Using Electronic Healthcare Records With Transformer

BACKGROUND

Obstructive sleep apnea (OSA) is growing increasingly prevalent in many countries as obesity rises. Sufficient, effective treatment of OSA entails high social and financial costs for healthcare.

OBJECTIVE

For treatment purposes, predicting OSA patients' visit expenses for the coming year is crucial. Reliable estimates enable healthcare decision-makers to perform careful fiscal management and budget well for effective distribution of resources to hospitals. The challenges created by scarcity of high-quality patient data are exacerbated by the fact that just a third of those data from OSA patients can be used to train analytics models: only OSA patients with more than 365 days of follow-up are relevant for predicting a year's expenditures.

METHODS AND PROCEDURES

The authors propose a translational engineering method applying two Transformer models, one for augmenting the input via data from shorter visit histories and the other predicting the costs by considering both the material thus enriched and cases with more than a year's follow-up. This method effectively adapts state-of-the-art Transformer models to create practical cost prediction solutions that can be implemented in OSA management, potentially enhancing patient care and resource allocation.

RESULTS

The two-model solution permits putting the limited body of OSA patient data to productive use. Relative to a single-Transformer solution using only a third of the high-quality patient data, the solution with two models improved the prediction performance's [Formula: see text] from 88.8% to 97.5%. Even using baseline models with the model-augmented data improved the [Formula: see text] considerably, from 61.6% to 81.9%.

CONCLUSION

The proposed method makes prediction with the most of the available high-quality data by carefully exploiting details, which are not directly relevant for answering the question of the next year's likely expenditure. Clinical and Translational Impact Statement: Public Health- Lack of high-quality source data hinders data-driven analytics-based research in healthcare. The paper presents a method that couples data augmentation and prediction in cases of scant healthcare data.

Predictive analysis for pathogenicity classification of H5Nx avian influenza strains using machine learning techniques

Over the past decades, avian influenza (AI) outbreaks have been reported across different parts of the globe, resulting in large-scale economic and livestock loss and, in some cases raising concerns about their zoonotic potential. The virulence and pathogenicity of H5Nx (e.g., H5N1, H5N2) AI strains for poultry could be inferred through various approaches, and it has been frequently performed by detecting certain pathogenicity markers in their haemagglutinin (HA) gene. The utilization of predictive modeling methods represents a possible approach to exploring this genotypic-phenotypic relationship for assisting experts in determining the pathogenicity of circulating AI viruses. Therefore, the main objective of this study was to evaluate the predictive performance of different machine learning (ML) techniques for in-silico prediction of pathogenicity of H5Nx viruses in poultry, using complete genetic sequences of the HA gene. We annotated 2137 H5Nx HA gene sequences based on the presence of the polybasic HA cleavage site (HACS) with 46.33% and 53.67% of sequences previously identified as highly pathogenic (HP) and low pathogenic (LP), respectively. We compared the performance of different ML classifiers (e.g., logistic regression (LR) with the lasso and ridge regularization, random forest (RF), K-nearest neighbor (KNN), Naïve Bayes (NB), support vector machine (SVM), and convolutional neural network (CNN)) for pathogenicity classification of raw H5Nx nucleotide and protein sequences using a 10-fold cross-validation technique. We found that different ML techniques can be successfully used for the pathogenicity classification of H5 sequences with ∼99% classification accuracy. Our results indicate that for pathogenicity classification of (1) aligned deoxyribonucleic acid (DNA) and protein sequences, with NB classifier had the lowest accuracies of 98.41% (+/-0.89) and 98.31% (+/-1.06), respectively; (2) aligned DNA and protein sequences, with LR (L1/L2), KNN, SVM (radial basis function (RBF)) and CNN classifiers had the highest accuracies of 99.20% (+/-0.54) and 99.20% (+/-0.38), respectively; (3) unaligned DNA and protein sequences, with CNN's achieved accuracies of 98.54% (+/-0.68) and 99.20% (+/-0.50), respectively. ML methods show potential for regular classification of H5Nx virus pathogenicity for poultry species, particularly when sequences containing regular markers were frequently present in the training dataset.

Patient Embeddings From Diagnosis Codes for Health Care Prediction Tasks: Pat2Vec Machine Learning Framework

BACKGROUND

In health care, diagnosis codes in claims data and electronic health records (EHRs) play an important role in data-driven decision making. Any analysis that uses a patient's diagnosis codes to predict future outcomes or describe morbidity requires a numerical representation of this diagnosis profile made up of string-based diagnosis codes. These numerical representations are especially important for machine learning models. Most commonly, binary-encoded representations have been used, usually for a subset of diagnoses. In real-world health care applications, several issues arise: patient profiles show high variability even when the underlying diseases are the same, they may have gaps and not contain all available information, and a large number of appropriate diagnoses must be considered.

OBJECTIVE

We herein present Pat2Vec, a self-supervised machine learning framework inspired by neural network-based natural language processing that embeds complete diagnosis profiles into a small real-valued numerical vector.

METHODS

Based on German outpatient claims data with diagnosis codes according to the International Statistical Classification of Diseases and Related Health Problems, 10th Revision (ICD-10), we discovered an optimal vectorization embedding model for patient diagnosis profiles with Bayesian optimization for the hyperparameters. The calibration process ensured a robust embedding model for health care-relevant tasks by aggregating the metrics of different regression and classification tasks using different machine learning algorithms (linear and logistic regression as well as gradient-boosted trees). The models were tested against a baseline model that binary encodes the most common diagnoses. The study used diagnosis profiles and supplementary data from more than 10 million patients from 2016 to 2019 and was based on the largest German ambulatory claims data set. To describe subpopulations in health care, we identified clusters (via density-based clustering) and visualized patient vectors in 2D (via dimensionality reduction with uniform manifold approximation). Furthermore, we applied our vectorization model to predict prospective drug prescription costs based on patients' diagnoses.

RESULTS

Our final models outperform the baseline model (binary encoding) with equal dimensions. They are more robust to missing data and show large performance gains, particularly in lower dimensions, demonstrating the embedding model's compression of nonlinear information. In the future, other sources of health care data can be integrated into the current diagnosis-based framework. Other researchers can apply our publicly shared embedding model to their own diagnosis data.

CONCLUSIONS

We envision a wide range of applications for Pat2Vec that will improve health care quality, including personalized prevention and signal detection in patient surveillance as well as health care resource planning based on subcohorts identified by our data-driven machine learning framework.

Self-Supervised Graph Learning With Hyperbolic Embedding for Temporal Health Event Prediction

Electronic health records (EHRs) have been heavily used in modern healthcare systems for recording patients' admission information to health facilities. Many data-driven approaches employ temporal features in EHR for predicting specific diseases, readmission times, and diagnoses of patients. However, most existing predictive models cannot fully utilize EHR data, due to an inherent lack of labels in supervised training for some temporal events. Moreover, it is hard for the existing methods to simultaneously provide generic and personalized interpretability. To address these challenges, we propose Sherbet, a self-supervised graph learning framework with hyperbolic embeddings for temporal health event prediction. We first propose a hyperbolic embedding method with information flow to pretrain medical code representations in a hierarchical structure. We incorporate these pretrained representations into a graph neural network (GNN) to detect disease complications and design a multilevel attention method to compute the contributions of particular diseases and admissions, thus enhancing personalized interpretability. We present a new hierarchy-enhanced historical prediction proxy task in our self-supervised learning framework to fully utilize EHR data and exploit medical domain knowledge. We conduct a comprehensive set of experiments on widely used publicly available EHR datasets to verify the effectiveness of our model. Our results demonstrate the proposed model's strengths in both predictive tasks and interpretable abilities.

Real-time prediction of organ failures in patients with acute pancreatitis using longitudinal irregular data

It is extremely important to identify patients with acute pancreatitis who are at high risk for developing persistent organ failures early in the course of the disease. Due to the irregularity of longitudinal data and the poor interpretability of complex models, many models used to identify acute pancreatitis patients with a high risk of organ failure tended to rely on simple statistical models and limited their application to the early stages of patient admission. With the success of recurrent neural networks in modeling longitudinal medical data and the development of interpretable algorithms, these problems can be well addressed. In this study, we developed a novel model named Multi-task and Time-aware Gated Recurrent Unit RNN (MT-GRU) to directly predict organ failure in patients with acute pancreatitis based on irregular medical EMR data. Our proposed end-to-end multi-task model achieved significantly better performance compared to two-stage models. In addition, our model not only provided an accurate early warning of organ failure for patients throughout their hospital stay, but also demonstrated individual and population-level important variables, allowing physicians to understand the scientific basis of the model for decision-making. By providing early warning of the risk of organ failure, our proposed model is expected to assist physicians in improving outcomes for patients with acute pancreatitis.

Hi-BEHRT: Hierarchical Transformer-Based Model for Accurate Prediction of Clinical Events Using Multimodal Longitudinal Electronic Health Records

Electronic health records (EHR) represent a holistic overview of patients' trajectories. Their increasing availability has fueled new hopes to leverage them and develop accurate risk prediction models for a wide range of diseases. Given the complex interrelationships of medical records and patient outcomes, deep learning models have shown clear merits in achieving this goal. However, a key limitation of current study remains their capacity in processing long sequences, and long sequence modelling and its application in the context of healthcare and EHR remains unexplored. Capturing the whole history of medical encounters is expected to lead to more accurate predictions, but the inclusion of records collected for decades and from multiple resources can inevitably exceed the receptive field of the most existing deep learning architectures. This can result in missing crucial, long-term dependencies. To address this gap, we present Hi-BEHRT, a hierarchical Transformer-based model that can significantly expand the receptive field of Transformers and extract associations from much longer sequences. Using a multimodal large-scale linked longitudinal EHR, the Hi-BEHRT exceeds the state-of-the-art deep learning models 1% to 5% for area under the receiver operating characteristic (AUROC) curve and 1% to 8% for area under the precision recall (AUPRC) curve on average, and 2% to 8% (AUROC) and 2% to 11% (AUPRC) for patients with long medical history for 5-year heart failure, diabetes, chronic kidney disease, and stroke risk prediction. Additionally, because pretraining for hierarchical Transformer is not well-established, we provide an effective end-to-end contrastive pre-training strategy for Hi-BEHRT using EHR, improving its transferability on predicting clinical events with relatively small training dataset.

Learning the progression patterns of treatments using a probabilistic generative model

Modeling a disease or the treatment of a patient has drawn much attention in recent years due to the vast amount of information that Electronic Health Records contain. This paper presents a probabilistic generative model of treatments that are described in terms of sequences of medical activities of variable length. The main objective is to identify distinct subtypes of treatments for a given disease, and discover their development and progression. To this end, the model considers that a sequence of actions has an associated hierarchical structure of latent variables that both classifies the sequences based on their evolution over time, and segments the sequences into different progression stages. The learning procedure of the model is performed with the Expectation-Maximization algorithm which considers the exponential number of configurations of the latent variables and is efficiently solved with a method based on dynamic programming. The evaluation of the model is twofold: first, we use synthetic data to demonstrate that the learning procedure allows the generative model underlying the data to be recovered; we then further assess the potential of our model to provide treatment classification and staging information in real-world data. Our model can be seen as a tool for classification, simulation, data augmentation and missing data imputation.

TERTIAN: Clinical Endpoint Prediction in ICU via Time-Aware Transformer-Based Hierarchical Attention Network

Accurately predicting the clinical endpoint in ICU based on the patient's electronic medical records (EMRs) is essential for the timely treatment of critically ill patients and allocation of medical resources. However, the patient's EMRs usually consist of a large amount of heterogeneous multivariate time series data such as laboratory tests and vital signs, which are produced irregularly. Most existing methods fail to effectively model the time irregularity inherent in longitudinal patient medical records and capture the interrelationships among different types of data. To tackle these limitations, we propose a novel time-aware transformer-based hierarchical attention network (TERTIAN) for clinical endpoint prediction. In this model, a time-aware transformer is introduced to learn the personalized irregular temporal patterns of medical events, and a hierarchical attention mechanism is deployed to get the accurate patient fusion representation by comprehensively mining the interactions and correlations among multiple types of medical data. We evaluate our model on the MIMIC-III dataset and MIMIC-IV dataset for the task of mortality prediction, and the results show that TERTIAN achieves higher performance than state-of-the-art approaches.

A multi-granularity convolutional neural network model with temporal information and attention mechanism for efficient diabetes medical cost prediction

As the cost of diabetes treatment continues to grow, it is critical to accurately predict the medical costs of diabetes. Most medical cost studies based on convolutional neural networks (CNNs) ignore the importance of multi-granularity information of medical concepts and time interval characteristics of patients' multiple visit sequences, which reflect the frequency of patient visits and the severity of the disease. Therefore, this paper proposes a new end-to-end deep neural network structure, MST-CNN, for medical cost prediction. The MST-CNN model improves the representation quality of medical concepts by constructing a multi-granularity embedding model of medical concepts and incorporates a time interval vector to accurately measure the frequency of patient visits and form an accurate representation of medical events. Moreover, the MST-CNN model integrates a channel attention mechanism to adaptively adjust the channel weights to focus on significant medical features. The MST-CNN model systematically addresses the problem of deep learning models for temporal data representation. A case study and three comparative experiments are conducted using data collected from Pingjiang County. Through experiments, the methods used in the proposed model are analyzed, and the super contribution of the model performance is demonstrated.

Predicting hypertension onset from longitudinal electronic health records with deep learning

Objective

Hypertension has long been recognized as one of the most important predisposing factors for cardiovascular diseases and mortality. In recent years, machine learning methods have shown potential in diagnostic and predictive approaches in chronic diseases. Electronic health records (EHRs) have emerged as a reliable source of longitudinal data. The aim of this study is to predict the onset of hypertension using modern deep learning (DL) architectures, specifically long short-term memory (LSTM) networks, and longitudinal EHRs.

Materials and Methods

We compare this approach to the best performing models reported from previous works, particularly XGboost, applied to aggregated features. Our work is based on data from 233 895 adult patients from a large health system in the United States. We divided our population into 2 distinct longitudinal datasets based on the diagnosis date. To ensure generalization to unseen data, we trained our models on the first dataset (dataset A "train and validation") using cross-validation, and then applied the models to a second dataset (dataset B "test") to assess their performance. We also experimented with 2 different time-windows before the onset of hypertension and evaluated the impact on model performance.

Results

With the LSTM network, we were able to achieve an area under the receiver operating characteristic curve value of 0.98 in the "train and validation" dataset A and 0.94 in the "test" dataset B for a prediction time window of 1 year. Lipid disorders, type 2 diabetes, and renal disorders are found to be associated with incident hypertension.

Conclusion

These findings show that DL models based on temporal EHR data can improve the identification of patients at high risk of hypertension and corresponding driving factors. In the long term, this work may support identifying individuals who are at high risk for developing hypertension and facilitate earlier intervention to prevent the future development of hypertension.

Effective hospital readmission prediction models using machine-learned features

BACKGROUND

Hospital readmissions are one of the costliest challenges facing healthcare systems, but conventional models fail to predict readmissions well. Many existing models use exclusively manually-engineered features, which are labor intensive and dataset-specific. Our objective was to develop and evaluate models to predict hospital readmissions using derived features that are automatically generated from longitudinal data using machine learning techniques.

METHODS

We studied patients discharged from acute care facilities in 2015 and 2016 in Alberta, Canada, excluding those who were hospitalized to give birth or for a psychiatric condition. We used population-level linked administrative hospital data from 2011 to 2017 to train prediction models using both manually derived features and features generated automatically from observational data. The target value of interest was 30-day all-cause hospital readmissions, with the success of prediction measured using the area under the curve (AUC) statistic.

RESULTS

Data from 428,669 patients (62% female, 38% male, 27% 65 years or older) were used for training and evaluating models: 24,974 (5.83%) were readmitted within 30 days of discharge for any reason. Patients were more likely to be readmitted if they utilized hospital care more, had more physician office visits, had more prescriptions, had a chronic condition, or were 65 years old or older. The LACE readmission prediction model had an AUC of 0.66 ± 0.0064 while the machine learning model's test set AUC was 0.83 ± 0.0045, based on learning a gradient boosting machine on a combination of machine-learned and manually-derived features.

CONCLUSION

Applying a machine learning model to the computer-generated and manual features improved prediction accuracy over the LACE model and a model that used only manually-derived features. Our model can be used to identify high-risk patients, for whom targeted interventions may potentially prevent readmissions.

Temporal deep learning framework for retinopathy prediction in patients with type 1 diabetes

The adoption of electronic health records in hospitals has ensured the availability of large datasets that can be used to predict medical complications. The trajectories of patients in real-world settings are highly variable, making longitudinal data modeling challenging. In recent years, significant progress has been made in the study of deep learning models applied to time series; however, the application of these models to irregular medical time series (IMTS) remains limited. To address this issue, we developed a generic deep-learning-based framework for modeling IMTS that facilitates the comparative studies of sequential neural networks (transformers and long short-term memory) and irregular time representation techniques. A validation study to predict retinopathy complications was conducted on 1207 patients with type 1 diabetes in a French database using their historical glycosylated hemoglobin measurements, without any data aggregation or imputation. The transformer-based model combined with the soft one-hot representation of time gaps achieved the highest score: an area under the receiver operating characteristic curve of 88.65%, specificity of 85.56%, sensitivity of 83.33% and an improvement of 11.7% over the same architecture without time information. This is the first attempt to predict retinopathy complications in patients with type 1 diabetes using deep learning and longitudinal data collected from patient visits. This study highlighted the significance of modeling time gaps between medical records to improve prediction performance and the utility of a generic framework for conducting extensive comparative studies.

DeepMPM: a mortality risk prediction model using longitudinal EHR data

BACKGROUND

Accurate precision approaches have far not been developed for modeling mortality risk in intensive care unit (ICU) patients. Conventional mortality risk prediction methods can hardly extract the information in longitudinal electronic medical records (EHRs) effectively, since they simply aggregate the heterogeneous variables in EHRs, ignoring the complex relationship and interactions between variables and the time dependence in longitudinal records. Recently deep learning approaches have been widely used in modeling longitudinal EHR data. However, most existing deep learning-based risk prediction approaches only use the information of a single disease, neglecting the interactions between multiple diseases and different conditions.

RESULTS

In this paper, we address this unmet need by leveraging disease and treatment information in EHRs to develop a mortality risk prediction model based on deep learning (DeepMPM). DeepMPM utilizes a two-level attention mechanism, i.e. visit-level and variable-level attention, to derive the representation of patient risk status from patient's multiple longitudinal medical records. Benefiting from using EHR of patients with multiple diseases and different conditions, DeepMPM can achieve state-of-the-art performances in mortality risk prediction.

CONCLUSIONS

Experiment results on MIMIC III database demonstrates that with the disease and treatment information DeepMPM can achieve a good performance in terms of Area Under ROC Curve (0.85). Moreover, DeepMPM can successfully model the complex interactions between diseases to achieve better representation learning of disease and treatment than other deep learning approaches, so as to improve the accuracy of mortality prediction. A case study also shows that DeepMPM offers the potential to provide users with insights into feature correlation in data as well as model behavior for each prediction.

AdaDiag: Adversarial Domain Adaptation of Diagnostic Prediction with Clinical Event Sequences

Early detection of heart failure (HF) can provide patients with the opportunity for more timely intervention and better disease management, as well as efficient use of healthcare resources. Recent machine learning (ML) methods have shown promising performance on diagnostic prediction using temporal sequences from electronic health records (EHRs). In practice, however, these models may not generalize to other populations due to dataset shift. Shifts in datasets can be attributed to a range of factors such as variations in demographics, data management methods, and healthcare delivery patterns. In this paper, we use unsupervised adversarial domain adaptation methods to adaptively reduce the impact of dataset shift on cross-institutional transfer performance. The proposed framework is validated on a next-visit HF onset prediction task using a BERT-style Transformer-based language model pre-trained with a masked language modeling (MLM) task. Our model empirically demonstrates superior prediction performance relative to non-adversarial baselines in both transfer directions on two different clinical event sequence data sources.

Measuring the impact of anonymization on real-world consolidated health datasets engineered for secondary research use: Experiments in the context of MODELHealth project

Introduction

Electronic Health Records (EHRs) are essential data structures, enabling the sharing of valuable medical care information for a diverse patient population and being reused as input to predictive models for clinical research. However, issues such as the heterogeneity of EHR data and the potential compromisation of patient privacy inhibit the secondary use of EHR data in clinical research.

Objectives

This study aims to present the main elements of the MODELHealth project implementation and the evaluation method that was followed to assess the efficiency of its mechanism.

Methods

The MODELHealth project was implemented as an Extract-Transform-Load system that collects data from the hospital databases, performs harmonization to the HL7 FHIR standard and anonymization using the k-anonymity method, before loading the transformed data to a central repository. The integrity of the anonymization process was validated by developing a database query tool. The information loss occurring due to the anonymization was estimated with the metrics of generalized information loss, discernibility and average equivalence class size for various values of k.

Results

The average values of generalized information loss, discernibility and average equivalence class size obtained across all tested datasets and k values were 0.008473 ± 0.006216252886, 115,145,464.3 ± 79,724,196.11 and 12.1346 ± 6.76096647, correspondingly. The values of those metrics appear correlated with factors such as the k value and the dataset characteristics, as expected.

Conclusion

The experimental results of the study demonstrate that it is feasible to perform effective harmonization and anonymization on EHR data while preserving essential patient information.

Hospital Intelligent Power Operation and Maintenance Information Evaluation with the Long and Short Memory Neural Network

The invention describes a deep learning-based technique for monitoring power grid information operation and maintenance. Based on the time series data information in the power grid information operation and maintenance monitoring system, this method obtains the cleaned time series data through appropriate data preprocessing technology. This also uses the long-term and short-term memory neural network to realize the prediction function of the time series data to be detected. The reason behind is to construct the normal behavior model of the time series. Additionally, the control chart based on an exponentially weighted moving average is employed to determine whether the time series to be discovered contains any anomalous and abnormal phenomena. In the area of power grid information operation and maintenance monitoring, the method of invention is designed to counter anomaly. This phenomenon is universal in nature and has significant scientific importance for guiding treatment after the anomaly is discovered. Additionally, it guards against serious faults that the abnormality might bring about.

Personalizing Medication Recommendation with a Graph-Based Approach

The broad adoption of electronic health records (EHRs) has led to vast amounts of data being accumulated on a patient’s history, diagnosis, prescriptions, and lab tests. Advances in recommender technologies have the potential to utilize this information to help doctors personalize the prescribed medications. However, existing medication recommendation systems have yet to make use of all these information sources in a seamless manner, and they do not provide a justification on why a particular medication is recommended. In this work, we design a two-stage personalized medication recommender system called PREMIER that incorporates information from the EHR. We utilize the various weights in the system to compute the contributions from the information sources for the recommended medications. Our system models the drug interaction from an external drug database and the drug co-occurrence from the EHR as graphs. Experiment results on MIMIC-III and a proprietary outpatient dataset show that PREMIER outperforms state-of-the-art medication recommendation systems while achieving the best tradeoff between accuracy and drug-drug interaction. Case studies demonstrate that the justifications provided by PREMIER are appropriate and aligned to clinical practices.

An Explainable Transformer-Based Deep Learning Model for the Prediction of Incident Heart Failure

Predicting the incidence of complex chronic conditions such as heart failure is challenging. Deep learning models applied to rich electronic health records may improve prediction but remain unexplainable hampering their wider use in medical practice. We aimed to develop a deep-learning framework for accurate and yet explainable prediction of 6-month incident heart failure (HF). Using 100,071 patients from longitudinal linked electronic health records across the U.K., we applied a novel Transformer-based risk model using all community and hospital diagnoses and medications contextualized within the age and calendar year for each patient's clinical encounter. Feature importance was investigated with an ablation analysis to compare model performance when alternatively removing features and by comparing the variability of temporal representations. A post-hoc perturbation technique was conducted to propagate the changes in the input to the outcome for feature contribution analyses. Our model achieved 0.93 area under the receiver operator curve and 0.69 area under the precision-recall curve on internal 5-fold cross validation and outperformed existing deep learning models. Ablation analysis indicated medication is important for predicting HF risk, calendar year is more important than chronological age, which was further reinforced by temporal variability analysis. Contribution analyses identified risk factors that are closely related to HF. Many of them were consistent with existing knowledge from clinical and epidemiological research but several new associations were revealed which had not been considered in expert-driven risk prediction models. In conclusion, the results highlight that our deep learning model, in addition high predictive performance, can inform data-driven risk factor identification.

Deep Learning Approaches for Predicting Glaucoma Progression Using Electronic Health Records and Natural Language Processing

Purpose

Advances in artificial intelligence have produced a few predictive models in glaucoma, including a logistic regression model predicting glaucoma progression to surgery. However, uncertainty exists regarding how to integrate the wealth of information in free-text clinical notes. The purpose of this study was to predict glaucoma progression requiring surgery using deep learning (DL) approaches on data from electronic health records (EHRs), including features from structured clinical data and from natural language processing of clinical free-text notes.

Design

Development of DL predictive model in an observational cohort.

Participants

Adult patients with glaucoma at a single center treated from 2008 through 2020.

Methods

Ophthalmology clinical notes of patients with glaucoma were identified from EHRs. Available structured data included patient demographic information, diagnosis codes, prior surgeries, and clinical information including intraocular pressure, visual acuity, and central corneal thickness. In addition, words from patients’ first 120 days of notes were mapped to ophthalmology domain-specific neural word embeddings trained on PubMed ophthalmology abstracts. Word embeddings and structured clinical data were used as inputs to DL models to predict subsequent glaucoma surgery.

Main Outcome Measures

Evaluation metrics included area under the receiver operating characteristic curve (AUC) and F1 score, the harmonic mean of positive predictive value, and sensitivity on a held-out test set.

Results

Seven hundred forty-eight of 4512 patients with glaucoma underwent surgery. The model that incorporated both structured clinical features as well as input features from clinical notes achieved an AUC of 73% and F1 of 40%, compared with only structured clinical features, (AUC, 66%; F1, 34%) and only clinical free-text features (AUC, 70%; F1, 42%). All models outperformed predictions from a glaucoma specialist’s review of clinical notes (F1, 29.5%).

Conclusions

We can successfully predict which patients with glaucoma will need surgery using DL models on EHRs unstructured text. Models incorporating free-text data outperformed those using only structured inputs. Future predictive models using EHRs should make use of information from within clinical free-text notes to improve predictive performance. Additional research is needed to investigate optimal methods of incorporating imaging data into future predictive models as well.

Deep learning in multimedia healthcare applications: a review

The increase in chronic diseases has affected the countries' health system and economy. With the recent COVID-19 virus, humanity has experienced a great challenge, which has led to make efforts to detect it and prevent its spread. Hence, it is necessary to develop new solutions that are based on technology and low cost, to satisfy the citizens' needs. Deep learning techniques is a technological solution that has been used in healthcare lately. Nowadays, with the increase in chips processing capabilities, increase size of data, and the progress in deep learning research, healthcare applications have been proposed to provide citizens' health needs. In addition, a big amount of data is generated every day. Development in Internet of Things, gadgets, and phones has allowed the access to multimedia data. Data such as images, video, audio and text are used as input of applications based on deep learning methods to support healthcare system to diagnose, predict, or treat patients. This review pretends to give an overview of proposed healthcare solutions based on deep learning techniques using multimedia data. We show the use of deep learning in healthcare, explain the different types of multimedia data, show some relevant deep learning multimedia applications in healthcare, and highlight some challenges in this research area.

Multi-view Integrative Attention-based Deep Representation Learning for Irregular Clinical Time-series Data

Electronic health record (EHR) data are sparse and irregular as they are recorded at irregular time intervals, and different clinical variables are measured at each observation point. In this work, to handle irregular multivariate time-series data, we consider the human knowledge of the aspects to be measured and time to measure them in different situations, known as multi-view features, which are indirectly represented in the data. We propose a scheme to realize multi-view features integration learning via a self-attention mechanism. Specifically, we devise a novel multi-integration attention module (MIAM) to extract complex information that is inherent in irregular time-series data. We explicitly learn the relationships among the observed values, missing indicators, and time interval between the consecutive observations in a simultaneous manner. In addition, we build an attention-based decoder as a missing value imputer that helps empower the representation learning of the interrelations among multi-view observations for the prediction task this decoder operates only in the training phase so that the final model is implemented in an imputation-free manner. We validated the effectiveness of our method over the public MIMIC-III and PhysioNet challenge 2012 datasets by comparing with and outperforming the state-of-the-art methods in three downstream tasks i.e., prediction of the in-hospital mortality, prediction of the length of stay, and phenotyping. Moreover, we conduct a layer-wise relevance propagation (LRP) analysis based on case studies to highlight the explainability of the trained model.

Autonomous Tool for Monitoring Multi-Morbidity Health Conditions in UAE and India

Multi-morbidity is the presence of two or more long-term health conditions, including defined physical or mental health conditions, such as diabetes or schizophrenia. One of the regular and critical health cases is an elderly person with a multi-morbid health condition and special complications who lives alone. These patients are typically not familiar with advanced Information and Communications Technology (ICT), but they are comfortable using smart devices such as wearable watches and mobile phones. The use of ICT improves medical quality, promotes patient security and data security, lowers operational and administrative costs, and gives the people in charge to make informed decisions. Additionally, the use of ICT in healthcare practices greatly reduces human errors, enhances clinical outcomes, ramps up care coordination, boosts practice efficiencies, and helps in collecting data over time. The proposed research concept provides a natural technique to implement preventive health care innovative solutions since several health sensors are embedded in devices that autonomously monitor the patients' health conditions in real-time. This enhances the elder's limited ability to predict and respond to critical health situations. Autonomous monitoring can alert doctors and patients themselves of unexpected health conditions. Real-time monitoring, modeling, and predicting health conditions can trigger swift responses by doctors and health officials in case of emergencies. This study will use data science to stimulate discoveries and breakthroughs in the United Arab Emirates (UAE) and India, which will then be reproduced in other world areas to create major gains in health for people, communities, and populations.

Time Series Prediction Using Deep Learning Methods in Healthcare

Traditional Machine Learning (ML) methods face unique challenges when applied to healthcare predictive analytics. The high-dimensional nature of healthcare data necessitates labor-intensive and time-consuming processes when selecting an appropriate set of features for each new task. Furthermore, ML methods depend heavily on feature engineering to capture the sequential nature of patient data, oftentimes failing to adequately leverage the temporal patterns of medical events and their dependencies. In contrast, recent Deep Learning (DL) methods have shown promising performance for various healthcare prediction tasks by specifically addressing the high-dimensional and temporal challenges of medical data. DL techniques excel at learning useful representations of medical concepts and patient clinical data as well as their nonlinear interactions from high-dimensional raw or minimally-processed healthcare data.

In this paper we systematically reviewed research works that focused on advancing deep neural networks to leverage patient structured time series data for healthcare prediction tasks. To identify relevant studies, we searched MEDLINE, IEEE, Scopus, and ACM digital library for relevant publications through November 4 th , 2021. Overall, we found that researchers have contributed to deep time series prediction literature in ten identifiable research streams: DL models, missing value handling, addressing temporal irregularity, patient representation, static data inclusion, attention mechanisms, interpretation, incorporation of medical ontologies, learning strategies, and scalability. This study summarizes research insights from these literature streams, identifies several critical research gaps, and suggests future research opportunities for DL applications using patient time series data.

PercolationDF: A percolation-based medical diagnosis framework

Goal: With the continuing shortage and unequal distribution of medical resources, our objective is to develop a general diagnosis framework that utilizes a smaller amount of electronic medical records (EMRs) to alleviate the problem that the data volume requirement of prevailing models is too vast for medical institutions to afford. Methods: The framework proposed contains network construction, network expansion, and disease diagnosis methods. In the first two stages above, the knowledge extracted from EMRs is utilized to build and expense an EMR-based medical knowledge network (EMKN) to model and represent the medical knowledge. Then, percolation theory is modified to diagnose EMKN. Result: Facing the lack of data, our framework outperforms naïve Bayes networks, neural networks and logistic regression, especially in the top-10 recall. Out of 207 test cases, 51.7% achieved 100% in the top-10 recall, 21% better than what was achieved in one of our previous studies. Conclusion: The experimental results show that the proposed framework may be useful for medical knowledge representation and diagnosis. The framework effectively alleviates the lack of data volume by inferring the knowledge modeled in EMKN. Significance: The proposed framework not only has applications for diagnosis but also may be extended to other domains to represent and model the knowledge and inference on the representation.

Using Machine Learning to Predict Complications in Pregnancy: A Systematic Review

Introduction: Artificial intelligence is widely used in medical field, and machine learning has been increasingly used in health care, prediction, and diagnosis and as a method of determining priority. Machine learning methods have been features of several tools in the fields of obstetrics and childcare. This present review aims to summarize the machine learning techniques to predict perinatal complications.

Objective: To identify the applicability and performance of machine learning methods used to identify pregnancy complications.

Methods: A total of 98 articles were obtained with the keywords “machine learning,” “deep learning,” “artificial intelligence,” and accordingly as they related to perinatal complications (“complications in pregnancy,” “pregnancy complications”) from three scientific databases: PubMed, Scopus, and Web of Science. These were managed on the Mendeley platform and classified using the PRISMA method.

Results: A total of 31 articles were selected after elimination according to inclusion and exclusion criteria. The features used to predict perinatal complications were primarily electronic medical records (48%), medical images (29%), and biological markers (19%), while 4% were based on other types of features, such as sensors and fetal heart rate. The main perinatal complications considered in the application of machine learning thus far are pre-eclampsia and prematurity. In the 31 studies, a total of sixteen complications were predicted. The main precision metric used is the AUC. The machine learning methods with the best results were the prediction of prematurity from medical images using the support vector machine technique, with an accuracy of 95.7%, and the prediction of neonatal mortality with the XGBoost technique, with 99.7% accuracy.

Conclusion: It is important to continue promoting this area of research and promote solutions with multicenter clinical applicability through machine learning to reduce perinatal complications. This systematic review contributes significantly to the specialized literature on artificial intelligence and women’s health.

Sequential Data-Based Patient Similarity Framework for Patient Outcome Prediction: Algorithm Development

BACKGROUND

Sequential information in electronic medical records is valuable and helpful for patient outcome prediction but is rarely used for patient similarity measurement because of its unevenness, irregularity, and heterogeneity.

OBJECTIVE

We aimed to develop a patient similarity framework for patient outcome prediction that makes use of sequential and cross-sectional information in electronic medical record systems.

METHODS

Sequence similarity was calculated from timestamped event sequences using edit distance, and trend similarity was calculated from time series using dynamic time warping and Haar decomposition. We also extracted cross-sectional information, namely, demographic, laboratory test, and radiological report data, for additional similarity calculations. We validated the effectiveness of the framework by constructing k-nearest neighbors classifiers to predict mortality and readmission for acute myocardial infarction patients, using data from (1) a public data set and (2) a private data set, at 3 time points-at admission, on Day 7, and at discharge-to provide early warning patient outcomes. We also constructed state-of-the-art Euclidean-distance k-nearest neighbor, logistic regression, random forest, long short-term memory network, and recurrent neural network models, which were used for comparison.

RESULTS

With all available information during a hospitalization episode, predictive models using the similarity model outperformed baseline models based on both public and private data sets. For mortality predictions, all models except for the logistic regression model showed improved performances over time. There were no such increasing trends in predictive performances for readmission predictions. The random forest and logistic regression models performed best for mortality and readmission predictions, respectively, when using information from the first week after admission.

CONCLUSIONS

For patient outcome predictions, the patient similarity framework facilitated sequential similarity calculations for uneven electronic medical record data and helped improve predictive performance.

Deep learning for temporal data representation in electronic health records: A systematic review of challenges and methodologies

OBJECTIVE

Temporal electronic health records (EHRs) contain a wealth of information for secondary uses, such as clinical events prediction and chronic disease management. However, challenges exist for temporal data representation. We therefore sought to identify these challenges and evaluate novel methodologies for addressing them through a systematic examination of deep learning solutions.

METHODS

We searched five databases (PubMed, Embase, the Institute of Electrical and Electronics Engineers [IEEE] Xplore Digital Library, the Association for Computing Machinery [ACM] Digital Library, and Web of Science) complemented with hand-searching in several prestigious computer science conference proceedings. We sought articles that reported deep learning methodologies on temporal data representation in structured EHR data from January 1, 2010, to August 30, 2020. We summarized and analyzed the selected articles from three perspectives: nature of time series, methodology, and model implementation.

RESULTS

We included 98 articles related to temporal data representation using deep learning. Four major challenges were identified, including data irregularity, heterogeneity, sparsity, and model opacity. We then studied how deep learning techniques were applied to address these challenges. Finally, we discuss some open challenges arising from deep learning.

CONCLUSION

Temporal EHR data present several major challenges for clinical prediction modeling and data utilization. To some extent, current deep learning solutions can address these challenges. Future studies may consider designing comprehensive and integrated solutions. Moreover, researchers should incorporate clinical domain knowledge into study designs and enhance model interpretability to facilitate clinical implementation.

Is it alright to use artificial intelligence in digital health? A systematic literature review on ethical considerations

The application of artificial intelligence (AI) not only yields in advantages for healthcare but raises several ethical questions. Extant research on ethical considerations of AI in digital health is quite sparse and a holistic overview is lacking. A systematic literature review searching across 853 peer-reviewed journals and conferences yielded in 50 relevant articles categorized in five major ethical principles: beneficence, non-maleficence, autonomy, justice, and explicability. The ethical landscape of AI in digital health is portrayed including a snapshot guiding future development. The status quo highlights potential areas with little empirical but required research. Less explored areas with remaining ethical questions are validated and guide scholars' efforts by outlining an overview of addressed ethical principles and intensity of studies including correlations. Practitioners understand novel questions AI raises eventually leading to properly regulated implementations and further comprehend that society is on its way from supporting technologies to autonomous decision-making systems.

Synthesizing time-series wound prognosis factors from electronic medical records using generative adversarial networks

Wound prognostic models not only provide an estimate of wound healing time to motivate patients to follow up their treatments but also can help clinicians to decide whether to use a standard care or adjuvant therapies and to assist them with designing clinical trials. However, collecting prognosis factors from Electronic Medical Records (EMR) of patients is challenging due to privacy, sensitivity, and confidentiality. In this study, we developed time series medical generative adversarial networks (GANs) to generate synthetic wound prognosis factors using very limited information collected during routine care in a specialized wound care facility. The generated prognosis variables are used in developing a predictive model for chronic wound healing trajectory. Our novel medical GAN can produce both continuous and categorical features from EMR. Moreover, we applied temporal information to our model by considering data collected from the weekly follow-ups of patients. Conditional training strategies were utilized to enhance training and generate classified data in terms of healing or non-healing. The ability of the proposed model to generate realistic EMR data was evaluated by TSTR (test on the synthetic, train on the real), discriminative accuracy, and visualization. We utilized samples generated by our proposed GAN in training a prognosis model to demonstrate its real-life application. Using the generated samples in training predictive models improved the classification accuracy by 6.66-10.01% compared to the previous EMR-GAN. Additionally, the suggested prognosis classifier has achieved the area under the curve (AUC) of 0.875, 0.810, and 0.647 when training the network using data from the first three visits, first two visits, and first visit, respectively. These results indicate a significant improvement in wound healing prediction compared to the previous prognosis models.

Interpretability analysis of one-year mortality prediction for stroke patients based on deep neural network

Clinically, physicians collect the benchmark medical data to establish archives for a stroke patient and then add the follow up data regularly. It has great significance on prognosis prediction for stroke patients. In this paper, we present an interpretable deep learning model to predict the one-year mortality risk on stroke. We design sub-modules to reconstruct features from original clinical data that highlight the dissimilarity and temporality of different variables. The model consists of Bidirectional Long Short-Term Memory (Bi-LSTM), in which a novel correlation attention module is proposed that takes the correlation of variables into consideration. In experiments, datasets are collected clinically from the department of neurology in a local AAA hospital. It consists of 2,275 stroke patients hospitalized in the department of neurology from 2014 to 2016. Our model achieves a precision of 0.9414, a recall of 0.9502 and an F1-score of 0.9415. In addition, we provide the analysis of the interpretability by visualizations with reference to clinical professional guidelines.

Time-Aware Multi-Type Data Fusion Representation Learning Framework for Risk Prediction of Cardiovascular Diseases

Predicting the future risk of cardiovascular diseases from the historical Electronic Health Records (EHRs) is a significant research task in personalized healthcare fields. In recent years, many deep neural network-based methods have emerged, which model patient disease progression by capturing the temporal patterns in sequential visit data. However, existing methods usually cannot effectively integrate the features of heterogeneous clinical data, and do not fully consider the impact of patients age and irregular time interval between consecutive medical records on the patients disease development. To address these challenges, we propose a Time-Aware Multi-type Data fUsion Representation learning framework (TAMDUR) for CVDs risk prediction. In this framework, we design a time-aware decay function, which is based on the patients age and the elapsed time between visits, to model the disease progression pattern. A parallel combination of Bi LSTM and CNN is constructed to respectively learn the temporal and non-temporal features from various types of clinical data. Finally, a multi-type data fusion representation layer based on self-attention is utilized to integrate various features and their correlations to obtain the final patient representation. We evaluate our model on a real medical dataset, and the experimental results demonstrate that TAMDUR outperforms the state-of-the-art approaches.