Learning probabilistic phenotypes from heterogeneous EHR data

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.

Finding Long-COVID: temporal topic modeling of electronic health records from the N3C and RECOVER programs

Post-Acute Sequelae of SARS-CoV-2 infection (PASC), also known as Long-COVID, encompasses a variety of complex and varied outcomes following COVID-19 infection that are still poorly understood. We clustered over 600 million condition diagnoses from 14 million patients available through the National COVID Cohort Collaborative (N3C), generating hundreds of highly detailed clinical phenotypes. Assessing patient clinical trajectories using these clusters allowed us to identify individual conditions and phenotypes strongly increased after acute infection. We found many conditions increased in COVID-19 patients compared to controls, and using a novel method to associate patients with clusters over time, we additionally found phenotypes specific to patient sex, age, wave of infection, and PASC diagnosis status. While many of these results reflect known PASC symptoms, the resolution provided by this unprecedented data scale suggests avenues for improved diagnostics and mechanistic understanding of this multifaceted disease.

Patient Subtyping via Learning Hidden Markov Models from Pairwise Co-occurrences in EHR Data

Patient subtyping is an effective tool to investigate clinically relevant characteristics of medical services, from which useful insights can be drawn for categorizing patients' conditions and informing disease progressions. Here we propose a hidden Markov model (HMM) based treatment of extracting patient subtypes from electronic health records (EHR). Using real-world clinical data, we show that the HMM based model can effectively identify the latent Markovian structure underlying the EHR data, and derive clinically or medically plausible patient subtypes, which can be used to categorize patients of various conditions.

Finding Long-COVID: Temporal Topic Modeling of Electronic Health Records from the N3C and RECOVER Programs

Post-Acute Sequelae of SARS-CoV-2 infection (PASC), also known as Long-COVID, encompasses a variety of complex and varied outcomes following COVID-19 infection that are still poorly understood. We clustered over 600 million condition diagnoses from 14 million patients available through the National COVID Cohort Collaborative (N3C), generating hundreds of highly detailed clinical phenotypes. Assessing patient clinical trajectories using these clusters allowed us to identify individual conditions and phenotypes strongly increased after acute infection. We found many conditions increased in COVID-19 patients compared to controls, and using a novel method to associate patients with clusters over time, we additionally found phenotypes specific to patient sex, age, wave of infection, and PASC diagnosis status. While many of these results reflect known PASC symptoms, the resolution provided by this unprecedented data scale suggests avenues for improved diagnostics and mechanistic understanding of this multifaceted disease.

MixEHR-SurG: A joint proportional hazard and guided topic model for inferring mortality-associated topics from electronic health records

Survival models can help medical practitioners to evaluate the prognostic importance of clinical variables to patient outcomes such as mortality or hospital readmission and subsequently design personalized treatment regimes. Electronic Health Records (EHRs) hold the promise for large-scale survival analysis based on systematically recorded clinical features for each patient. However, existing survival models either do not scale to high dimensional and multi-modal EHR data or are difficult to interpret. In this study, we present a supervised topic model called MixEHR-SurG to simultaneously integrate heterogeneous EHR data and model survival hazard. Our contributions are three-folds: (1) integrating EHR topic inference with Cox proportional hazards likelihood; (2) integrating patient-specific topic hyperparameters using the PheCode concepts such that each topic can be identified with exactly one PheCode-associated phenotype; (3) multi-modal survival topic inference. This leads to a highly interpretable survival topic model that can infer PheCode-specific phenotype topics associated with patient mortality. We evaluated MixEHR-SurG using a simulated dataset and two real-world EHR datasets: the Quebec Congenital Heart Disease (CHD) data consisting of 8211 subjects with 75,187 outpatient claim records of 1767 unique ICD codes; the MIMIC-III consisting of 1458 subjects with multi-modal EHR records. Compared to the baselines, MixEHR-SurG achieved a superior dynamic AUROC for mortality prediction, with a mean AUROC score of 0.89 in the simulation dataset and a mean AUROC of 0.645 on the CHD dataset. Qualitatively, MixEHR-SurG associates severe cardiac conditions with high mortality risk among the CHD patients after the first heart failure hospitalization and critical brain injuries with increased mortality among the MIMIC-III patients after their ICU discharge. Together, the integration of the Cox proportional hazards model and EHR topic inference in MixEHR-SurG not only leads to competitive mortality prediction but also meaningful phenotype topics for in-depth survival analysis. The software is available at GitHub: https://github.com/li-lab-mcgill/MixEHR-SurG.

Integrating Structured and Unstructured EHR Data for Predicting Mortality by Machine Learning and Latent Dirichlet Allocation Method

An ICU is a critical care unit that provides advanced medical support and continuous monitoring for patients with severe illnesses or injuries. Predicting the mortality rate of ICU patients can not only improve patient outcomes, but also optimize resource allocation. Many studies have attempted to create scoring systems and models that predict the mortality of ICU patients using large amounts of structured clinical data. However, unstructured clinical data recorded during patient admission, such as notes made by physicians, is often overlooked. This study used the MIMIC-III database to predict mortality in ICU patients. In the first part of the study, only eight structured variables were used, including the six basic vital signs, the GCS, and the patient's age at admission. In the second part, unstructured predictor variables were extracted from the initial diagnosis made by physicians when the patients were admitted to the hospital and analyzed using Latent Dirichlet Allocation techniques. The structured and unstructured data were combined using machine learning methods to create a mortality risk prediction model for ICU patients. The results showed that combining structured and unstructured data improved the accuracy of the prediction of clinical outcomes in ICU patients over time. The model achieved an AUROC of 0.88, indicating accurate prediction of patient vital status. Additionally, the model was able to predict patient clinical outcomes over time, successfully identifying important variables. This study demonstrated that a small number of easily collectible structured variables, combined with unstructured data and analyzed using LDA topic modeling, can significantly improve the predictive performance of a mortality risk prediction model for ICU patients. These results suggest that initial clinical observations and diagnoses of ICU patients contain valuable information that can aid ICU medical and nursing staff in making important clinical decisions.

Genetic heterogeneity: Challenges, impacts, and methods through an associative lens

Genetic heterogeneity describes the occurrence of the same or similar phenotypes through different genetic mechanisms in different individuals. Robustly characterizing and accounting for genetic heterogeneity is crucial to pursuing the goals of precision medicine, for discovering novel disease biomarkers, and for identifying targets for treatments. Failure to account for genetic heterogeneity may lead to missed associations and incorrect inferences. Thus, it is critical to review the impact of genetic heterogeneity on the design and analysis of population level genetic studies, aspects that are often overlooked in the literature. In this review, we first contextualize our approach to genetic heterogeneity by proposing a high-level categorization of heterogeneity into "feature," "outcome," and "associative" heterogeneity, drawing on perspectives from epidemiology and machine learning to illustrate distinctions between them. We highlight the unique nature of genetic heterogeneity as a heterogeneous pattern of association that warrants specific methodological considerations. We then focus on the challenges that preclude effective detection and characterization of genetic heterogeneity across a variety of epidemiological contexts. Finally, we discuss systems heterogeneity as an integrated approach to using genetic and other high-dimensional multi-omic data in complex disease research.

Multimodal machine learning in precision health: A scoping review

Machine learning is frequently being leveraged to tackle problems in the health sector including utilization for clinical decision-support. Its use has historically been focused on single modal data. Attempts to improve prediction and mimic the multimodal nature of clinical expert decision-making has been met in the biomedical field of machine learning by fusing disparate data. This review was conducted to summarize the current studies in this field and identify topics ripe for future research. We conducted this review in accordance with the PRISMA extension for Scoping Reviews to characterize multi-modal data fusion in health. Search strings were established and used in databases: PubMed, Google Scholar, and IEEEXplore from 2011 to 2021. A final set of 128 articles were included in the analysis. The most common health areas utilizing multi-modal methods were neurology and oncology. Early fusion was the most common data merging strategy. Notably, there was an improvement in predictive performance when using data fusion. Lacking from the papers were clear clinical deployment strategies, FDA-approval, and analysis of how using multimodal approaches from diverse sub-populations may improve biases and healthcare disparities. These findings provide a summary on multimodal data fusion as applied to health diagnosis/prognosis problems. Few papers compared the outputs of a multimodal approach with a unimodal prediction. However, those that did achieved an average increase of 6.4% in predictive accuracy. Multi-modal machine learning, while more robust in its estimations over unimodal methods, has drawbacks in its scalability and the time-consuming nature of information concatenation.

Real-world data: a brief review of the methods, applications, challenges and opportunities

Background

The increased adoption of the internet, social media, wearable devices, e-health services, and other technology-driven services in medicine and healthcare has led to the rapid generation of various types of digital data, providing a valuable data source beyond the confines of traditional clinical trials, epidemiological studies, and lab-based experiments.

Methods

We provide a brief overview on the type and sources of real-world data and the common models and approaches to utilize and analyze real-world data. We discuss the challenges and opportunities of using real-world data for evidence-based decision making This review does not aim to be comprehensive or cover all aspects of the intriguing topic on RWD (from both the research and practical perspectives) but serves as a primer and provides useful sources for readers who interested in this topic.

Results and Conclusions

Real-world hold great potential for generating real-world evidence for designing and conducting confirmatory trials and answering questions that may not be addressed otherwise. The voluminosity and complexity of real-world data also call for development of more appropriate, sophisticated, and innovative data processing and analysis techniques while maintaining scientific rigor in research findings, and attentions to data ethics to harness the power of real-world data.

A semi-supervised adaptive Markov Gaussian embedding process (SAMGEP) for prediction of phenotype event times using the electronic health record

While there exist numerous methods to identify binary phenotypes (i.e. COPD) using electronic health record (EHR) data, few exist to ascertain the timings of phenotype events (i.e. COPD onset or exacerbations). Estimating event times could enable more powerful use of EHR data for longitudinal risk modeling, including survival analysis. Here we introduce Semi-supervised Adaptive Markov Gaussian Embedding Process (SAMGEP), a semi-supervised machine learning algorithm to estimate phenotype event times using EHR data with limited observed labels, which require resource-intensive chart review to obtain. SAMGEP models latent phenotype states as a binary Markov process, and it employs an adaptive weighting strategy to map timestamped EHR features to an embedding function that it models as a state-dependent Gaussian process. SAMGEP's feature weighting achieves meaningful feature selection, and its predictions significantly improve AUCs and F1 scores over existing approaches in diverse simulations and real-world settings. It is particularly adept at predicting cumulative risk and event counting process functions, and is robust to diverse generative model parameters. Moreover, it achieves high accuracy with few (50-100) labels, efficiently leveraging unlabeled EHR data to maximize information gain from costly-to-obtain event time labels. SAMGEP can be used to estimate accurate phenotype state functions for risk modeling research.

Unsupervised probabilistic models for sequential Electronic Health Records

We develop an unsupervised probabilistic model for heterogeneous Electronic Health Record (EHR) data. Utilizing a mixture model formulation, our approach directly models sequences of arbitrary length, such as medications and laboratory results. This allows for subgrouping and incorporation of the dynamics underlying heterogeneous data types. The model consists of a layered set of latent variables that encode underlying structure in the data. These variables represent subject subgroups at the top layer, and unobserved states for sequences in the second layer. We train this model on episodic data from subjects receiving medical care in the Kaiser Permanente Northern California integrated healthcare delivery system. The resulting properties of the trained model generate novel insight from these complex and multifaceted data. In addition, we show how the model can be used to analyze sequences that contribute to assessment of mortality likelihood.

MixEHR-Guided: A guided multi-modal topic modeling approach for large-scale automatic phenotyping using the electronic health record

Electronic Health Records (EHRs) contain rich clinical data collected at the point of the care, and their increasing adoption offers exciting opportunities for clinical informatics, disease risk prediction, and personalized treatment recommendation. However, effective use of EHR data for research and clinical decision support is often hampered by a lack of reliable disease labels. To compile gold-standard labels, researchers often rely on clinical experts to develop rule-based phenotyping algorithms from billing codes and other surrogate features. This process is tedious and error-prone due to recall and observer biases in how codes and measures are selected, and some phenotypes are incompletely captured by a handful of surrogate features. To address this challenge, we present a novel automatic phenotyping model called MixEHR-Guided (MixEHR-G), a multimodal hierarchical Bayesian topic model that efficiently models the EHR generative process by identifying latent phenotype structure in the data. Unlike existing topic modeling algorithms wherein the inferred topics are not identifiable, MixEHR-G uses prior information from informative surrogate features to align topics with known phenotypes. We applied MixEHR-G to an openly-available EHR dataset of 38,597 intensive care patients (MIMIC-III) in Boston, USA and to administrative claims data for a population-based cohort (PopHR) of 1.3 million people in Quebec, Canada. Qualitatively, we demonstrate that MixEHR-G learns interpretable phenotypes and yields meaningful insights about phenotype similarities, comorbidities, and epidemiological associations. Quantitatively, MixEHR-G outperforms existing unsupervised phenotyping methods on a phenotype label annotation task, and it can accurately estimate relative phenotype prevalence functions without gold-standard phenotype information. Altogether, MixEHR-G is an important step towards building an interpretable and automated phenotyping system using EHR data.

PheNominal: an EHR-integrated web application for structured deep phenotyping at the point of care

BACKGROUND

Clinical phenotype information greatly facilitates genetic diagnostic interpretations pipelines in disease. While post-hoc extraction using natural language processing on unstructured clinical notes continues to improve, there is a need to improve point-of-care collection of patient phenotypes. Therefore, we developed "PheNominal", a point-of-care web application, embedded within Epic electronic health record (EHR) workflows, to permit capture of standardized phenotype data.

METHODS

Using bi-directional web services available within commercial EHRs, we developed a lightweight web application that allows users to rapidly browse and identify relevant terms from the Human Phenotype Ontology (HPO). Selected terms are saved discretely within the patient's EHR, permitting reuse both in clinical notes as well as in downstream diagnostic and research pipelines.

RESULTS

In the 16 months since implementation, PheNominal was used to capture discrete phenotype data for over 1500 individuals and 11,000 HPO terms during clinic and inpatient encounters for a genetic diagnostic consultation service within a quaternary-care pediatric academic medical center. An average of 7 HPO terms were captured per patient. Compared to a manual workflow, the average time to enter terms for a patient was reduced from 15 to 5 min per patient, and there were fewer annotation errors.

CONCLUSIONS

Modern EHRs support integration of external applications using application programming interfaces. We describe a practical application of these interfaces to facilitate deep phenotype capture in a discrete, structured format within a busy clinical workflow. Future versions will include a vendor-agnostic implementation using FHIR. We describe pilot efforts to integrate structured phenotyping through controlled dictionaries into diagnostic and research pipelines, reducing manual effort for phenotype documentation and reducing errors in data entry.

Psychosocial-Behavioral Phenotyping: A Novel Precision Health Approach to Modeling Behavioral, Psychological, and Social Determinants of Health Using Machine Learning

BACKGROUND

The context in which a behavioral intervention is delivered is an important source of variability and systematic approaches are needed to identify and quantify contextual factors that may influence intervention efficacy. Machine learning-based phenotyping methods can contribute to a new precision health paradigm by informing personalized behavior interventions. Two primary goals of precision health, identifying population subgroups and highlighting behavioral intervention targets, can be addressed with psychosocial-behavioral phenotypes. We propose a method for psychosocial-behavioral phenotyping that models social determinants of health in addition to individual-level psychological and behavioral factors.

PURPOSE

To demonstrate a novel application of machine learning for psychosocial-behavioral phenotyping, the identification of subgroups with similar combinations of psychosocial characteristics.

METHODS

In this secondary analysis of psychosocial and behavioral data from a community cohort (n = 5,883), we optimized a multichannel mixed membership model (MC3M) using Bayesian inference to identify psychosocial-behavioral phenotypes and used logistic regression to determine which phenotypes were associated with elevated weight status (BMI ≥ 25kg/m2).

RESULTS

We identified 20 psychosocial-behavioral phenotypes. Phenotypes were conceptually consistent as well as discriminative; most participants had only one active phenotype. Two phenotypes were significantly positively associated with elevated weight status; four phenotypes were significantly negatively associated. Each phenotype suggested different contextual considerations for intervention design.

CONCLUSIONS

By depicting the complexity of psychological and social determinants of health while also providing actionable insight about similarities and differences among members of the same community, psychosocial-behavioral phenotypes can identify potential intervention targets in context.

Mixture Model Framework for Traumatic Brain Injury Prognosis Using Heterogeneous Clinical and Outcome Data

Prognoses of Traumatic Brain Injury (TBI) outcomes are neither easily nor accurately determined from clinical indicators. This is due in part to the heterogeneity of damage inflicted to the brain, ultimately resulting in diverse and complex outcomes. Using a data-driven approach on many distinct data elements may be necessary to describe this large set of outcomes and thereby robustly depict the nuanced differences among TBI patients' recovery. In this work, we develop a method for modeling large heterogeneous data types relevant to TBI. Our approach is geared toward the probabilistic representation of mixed continuous and discrete variables with missing values. The model is trained on a dataset encompassing a variety of data types, including demographics, blood-based biomarkers, and imaging findings. In addition, it includes a set of clinical outcome assessments at 3, 6, and 12 months post-injury. The model is used to stratify patients into distinct groups in an unsupervised learning setting. We use the model to infer outcomes using input data, and show that the collection of input data reduces uncertainty of outcomes over a baseline approach. In addition, we quantify the performance of a likelihood scoring technique that can be used to self-evaluate the extrapolation risk of prognosis on unseen patients.

Comparison of the Erectile Dysfunction Drugs Sildenafil and Tadalafil Using Patient Medication Reviews: Topic Modeling Study

BACKGROUND

Topic modeling of patient medication reviews of erectile dysfunction (ED) drugs can help identify patient preferences regarding ED treatment options. The identification of a set of topics important to the patient from social network service drug reviews would inform the design of patient-centered medication counseling.

OBJECTIVE

This study aimed to (1) identify the distinctive topics from patient medication reviews unique to tadalafil versus sildenafil; (2) determine if the primary topics are distributed differently for each drug and for each patient characteristic (age and time on ED drug therapy); and (3) test if the primary topics affect satisfaction with ED drug therapy controlling for patient characteristics.

METHODS

Data were collected from the patient medication reviews of sildenafil and tadalafil posted on WebMD and Ask a Patient. The latent Dirichlet allocation method of natural language processing was used to identify 5 distinctive topics from the patient medication reviews on each drug. Analysis of variance and a 2-sample t test were conducted to compare the topic distribution and assess whether patient satisfaction varies with the primary topics, age, and time on medication for each ED drug. Statistical significance was tested at an alpha of .05.

RESULTS

The patient medication reviews of sildenafil (N=463) had 2 topics on treatment benefit and 1 each on medication safety, marketing claim, and treatment comparison, while the patient medication reviews of tadalafil (N=919) had 2 topics on medication safety and 1 each on the remaining subjects. Sildenafil's reviewers quite frequently (94/463, 20.4%) mentioned erection sustainability as their primary topic, whereas tadalafil's reviewers were more concerned about severe medication safety. Those who mentioned erection sustainability as their primary topic were quite satisfied with their treatment as opposed to those who mentioned severe medication safety as their primary topic (score 3.85 vs 2.44). The discovered topics reflected the marketing claims of blue magic and amber romance for sildenafil and tadalafil, respectively. The topic of blue magic was preferred among younger patients, while the topic of amber romance was preferred among older patients. The topic alternative choices, which appeared for both the ED drugs, reflected patient interest in the comparative effectiveness and price outside the drug labeling information.

CONCLUSIONS

The patient medication reviews of ED drugs reflect patient preferences regarding drug labeling information, marketing claims, and alternative treatment choices. The patient preferences concerning ED treatment attributes inform the design of patient-centered communication for improved ED drug therapy.

Clinical knowledge extraction via sparse embedding regression (KESER) with multi-center large scale electronic health record data

The increasing availability of electronic health record (EHR) systems has created enormous potential for translational research. However, it is difficult to know all the relevant codes related to a phenotype due to the large number of codes available. Traditional data mining approaches often require the use of patient-level data, which hinders the ability to share data across institutions. In this project, we demonstrate that multi-center large-scale code embeddings can be used to efficiently identify relevant features related to a disease of interest. We constructed large-scale code embeddings for a wide range of codified concepts from EHRs from two large medical centers. We developed knowledge extraction via sparse embedding regression (KESER) for feature selection and integrative network analysis. We evaluated the quality of the code embeddings and assessed the performance of KESER in feature selection for eight diseases. Besides, we developed an integrated clinical knowledge map combining embedding data from both institutions. The features selected by KESER were comprehensive compared to lists of codified data generated by domain experts. Features identified via KESER resulted in comparable performance to those built upon features selected manually or with patient-level data. The knowledge map created using an integrative analysis identified disease-disease and disease-drug pairs more accurately compared to those identified using single institution data. Analysis of code embeddings via KESER can effectively reveal clinical knowledge and infer relatedness among codified concepts. KESER bypasses the need for patient-level data in individual analyses providing a significant advance in enabling multi-center studies using EHR data.

Discovering Key Topics From Short, Real-World Medical Inquiries via Natural Language Processing

Millions of unsolicited medical inquiries are received by pharmaceutical companies every year. It has been hypothesized that these inquiries represent a treasure trove of information, potentially giving insight into matters regarding medicinal products and the associated medical treatments. However, due to the large volume and specialized nature of the inquiries, it is difficult to perform timely, recurrent, and comprehensive analyses. Here, we combine biomedical word embeddings, non-linear dimensionality reduction, and hierarchical clustering to automatically discover key topics in real-world medical inquiries from customers. This approach does not require ontologies nor annotations. The discovered topics are meaningful and medically relevant, as judged by medical information specialists, thus demonstrating that unsolicited medical inquiries are a source of valuable customer insights. Our work paves the way for the machine-learning-driven analysis of medical inquiries in the pharmaceutical industry, which ultimately aims at improving patient care.

Phe2vec: Automated disease phenotyping based on unsupervised embeddings from electronic health records

Robust phenotyping of patients from electronic health records (EHRs) at scale is a challenge in clinical informatics. Here, we introduce Phe2vec, an automated framework for disease phenotyping from EHRs based on unsupervised learning and assess its effectiveness against standard rule-based algorithms from Phenotype KnowledgeBase (PheKB). Phe2vec is based on pre-computing embeddings of medical concepts and patients' clinical history. Disease phenotypes are then derived from a seed concept and its neighbors in the embedding space. Patients are linked to a disease if their embedded representation is close to the disease phenotype. Comparing Phe2vec and PheKB cohorts head-to-head using chart review, Phe2vec performed on par or better in nine out of ten diseases. Differently from other approaches, it can scale to any condition and was validated against widely adopted expert-based standards. Phe2vec aims to optimize clinical informatics research by augmenting current frameworks to characterize patients by condition and derive reliable disease cohorts.

Probabilistic Machine Learning for Healthcare

Machine learning can be used to make sense of healthcare data. Probabilistic machine learning models help provide a complete picture of observed data in healthcare. In this review, we examine how probabilistic machine learning can advance healthcare. We consider challenges in the predictive model building pipeline where probabilistic models can be beneficial, including calibration and missing data. Beyond predictive models, we also investigate the utility of probabilistic machine learning models in phenotyping, in generative models for clinical use cases, and in reinforcement learning.

Evaluation of clustering and topic modeling methods over health-related tweets and emails

BACKGROUND

Internet provides different tools for communicating with patients, such as social media (e.g., Twitter) and email platforms. These platforms provided new data sources to shed lights on patient experiences with health care and improve our understanding of patient-provider communication. Several existing topic modeling and document clustering methods have been adapted to analyze these new free-text data automatically. However, both tweets and emails are often composed of short texts; and existing topic modeling and clustering approaches have suboptimal performance on these short texts. Moreover, research over health-related short texts using these methods has become difficult to reproduce and benchmark, partially due to the absence of a detailed comparison of state-of-the-art topic modeling and clustering methods on these short texts.

METHODS

We trained eight state-of- the-art topic modeling and clustering algorithms on short texts from two health-related datasets (tweets and emails): Latent Semantic Indexing (LSI), Latent Dirichlet Allocation (LDA), LDA with Gibbs Sampling (GibbsLDA), Online LDA, Biterm Model (BTM), Online Twitter LDA, and Gibbs Sampling for Dirichlet Multinomial Mixture (GSDMM), as well as the k-means clustering algorithm with two different feature representations: TF-IDF and Doc2Vec. We used cluster validity indices to evaluate the performance of topic modeling and clustering: two internal indices (i.e. assessing the goodness of a clustering structure without external information) and five external indices (i.e. comparing the results of a cluster analysis to an externally known provided class labels).

RESULTS

In overall, for number of clusters (k) from 2 to 50, Online Twitter LDA and GSDMM achieved the best performance in terms of internal indices, while LSI and k-means with TF-IDF had the highest external indices. Also, of all tweets (N = 286, 971; HPV represents 94.6% of tweets and lynch syndrome represents 5.4%), for k = 2, most of the methods could respect this initial clustering distribution. However, we found model performance varies with the source of data and hyper-parameters such as the number of topics and the number of iterations used to train the models. We also conducted an error analysis using the Hamming loss metric, for which the poorest value was obtained by GSDMM on both datasets.

CONCLUSIONS

Researchers hoping to group or classify health related short-text data can expect to select the most suitable topic modeling and clustering methods for their specific research questions. Therefore, we presented a comparison of the most common used topic modeling and clustering algorithms over two health-related, short-text datasets using both internal and external clustering validation indices. Internal indices suggested Online Twitter LDA and GSDMM as the best, while external indices suggested LSI and k-means with TF-IDF as the best. In summary, our work suggested researchers can improve their analysis of model performance by using a variety of metrics, since there is not a single best metric.

Using computable knowledge mined from the literature to elucidate confounders for EHR-based pharmacovigilance

INTRODUCTION

Drug safety research asks causal questions but relies on observational data. Confounding bias threatens the reliability of studies using such data. The successful control of confounding requires knowledge of variables called confounders affecting both the exposure and outcome of interest. However, causal knowledge of dynamic biological systems is complex and challenging. Fortunately, computable knowledge mined from the literature may hold clues about confounders. In this paper, we tested the hypothesis that incorporating literature-derived confounders can improve causal inference from observational data.

METHODS

We introduce two methods (semantic vector-based and string-based confounder search) that query literature-derived information for confounder candidates to control, using SemMedDB, a database of computable knowledge mined from the biomedical literature. These methods search SemMedDB for confounders by applying semantic constraint search for indications treated by the drug (exposure) and that are also known to cause the adverse event (outcome). We then include the literature-derived confounder candidates in statistical and causal models derived from free-text clinical notes. For evaluation, we use a reference dataset widely used in drug safety containing labeled pairwise relationships between drugs and adverse events and attempt to rediscover these relationships from a corpus of 2.2 M NLP-processed free-text clinical notes. We employ standard adjustment and causal inference procedures to predict and estimate causal effects by informing the models with varying numbers of literature-derived confounders and instantiating the exposure, outcome, and confounder variables in the models with dichotomous EHR-derived data. Finally, we compare the results from applying these procedures with naive measures of association (χ2 and reporting odds ratio) and with each other.

RESULTS AND CONCLUSIONS

We found semantic vector-based search to be superior to string-based search at reducing confounding bias. However, the effect of including more rather than fewer literature-derived confounders was inconclusive. We recommend using targeted learning estimation methods that can address treatment-confounder feedback, where confounders also behave as intermediate variables, and engaging subject-matter experts to adjudicate the handling of problematic covariates.

Mining heterogeneous clinical notes by multi-modal latent topic model

Latent knowledge can be extracted from the electronic notes that are recorded during patient encounters with the health system. Using these clinical notes to decipher a patient’s underlying comorbidites, symptom burdens, and treatment courses is an ongoing challenge. Latent topic model as an efficient Bayesian method can be used to model each patient’s clinical notes as “documents” and the words in the notes as “tokens”. However, standard latent topic models assume that all of the notes follow the same topic distribution, regardless of the type of note or the domain expertise of the author (such as doctors or nurses). We propose a novel application of latent topic modeling, using multi-note topic model (MNTM) to jointly infer distinct topic distributions of notes of different types. We applied our model to clinical notes from the MIMIC-III dataset to infer distinct topic distributions over the physician and nursing note types. Based on manual assessments made by clinicians, we observed a significant improvement in topic interpretability using MNTM modeling over the baseline single-note topic models that ignore the note types. Moreover, our MNTM model led to a significantly higher prediction accuracy for prolonged mechanical ventilation and mortality using only the first 48 hours of patient data. By correlating the patients’ topic mixture with hospital mortality and prolonged mechanical ventilation, we identified several diagnostic topics that are associated with poor outcomes. Because of its elegant and intuitive formation, we envision a broad application of our approach in mining multi-modality text-based healthcare information that goes beyond clinical notes. Code available at https://github.com/li-lab-mcgill/heterogeneous_ehr.

sureLDA: A multidisease automated phenotyping method for the electronic health record

OBJECTIVE

A major bottleneck hindering utilization of electronic health record data for translational research is the lack of precise phenotype labels. Chart review as well as rule-based and supervised phenotyping approaches require laborious expert input, hampering applicability to studies that require many phenotypes to be defined and labeled de novo. Though International Classification of Diseases codes are often used as surrogates for true labels in this setting, these sometimes suffer from poor specificity. We propose a fully automated topic modeling algorithm to simultaneously annotate multiple phenotypes.

MATERIALS AND METHODS

Surrogate-guided ensemble latent Dirichlet allocation (sureLDA) is a label-free multidimensional phenotyping method. It first uses the PheNorm algorithm to initialize probabilities based on 2 surrogate features for each target phenotype, and then leverages these probabilities to constrain the LDA topic model to generate phenotype-specific topics. Finally, it combines phenotype-feature counts with surrogates via clustering ensemble to yield final phenotype probabilities.

RESULTS

sureLDA achieves reliably high accuracy and precision across a range of simulated and real-world phenotypes. Its performance is robust to phenotype prevalence and relative informativeness of surogate vs nonsurrogate features. It also exhibits powerful feature selection properties.

DISCUSSION

sureLDA combines attractive properties of PheNorm and LDA to achieve high accuracy and precision robust to diverse phenotype characteristics. It offers particular improvement for phenotypes insufficiently captured by a few surrogate features. Moreover, sureLDA's feature selection ability enables it to handle high feature dimensions and produce interpretable computational phenotypes.

CONCLUSIONS

sureLDA is well suited toward large-scale electronic health record phenotyping for highly multiphenotype applications such as phenome-wide association studies .

Predictive Analytics for Care and Management of Patients With Acute Diseases: Deep Learning-Based Method to Predict Crucial Complication Phenotypes

BACKGROUND

Acute diseases present severe complications that develop rapidly, exhibit distinct phenotypes, and have profound effects on patient outcomes. Predictive analytics can enhance physicians' care and management of patients with acute diseases by predicting crucial complication phenotypes for a timely diagnosis and treatment. However, effective phenotype predictions require several challenges to be overcome. First, patient data collected in the early stages of an acute disease (eg, clinical data and laboratory results) are less informative for predicting phenotypic outcomes. Second, patient data are temporal and heterogeneous; for example, patients receive laboratory tests at different time intervals and frequencies. Third, imbalanced distributions of patient outcomes create additional complexity for predicting complication phenotypes.

OBJECTIVE

To predict crucial complication phenotypes among patients with acute diseases, we propose a novel, deep learning-based method that uses recurrent neural network-based sequence embedding to represent disease progression while considering temporal heterogeneities in patient data. Our method incorporates a latent regulator to alleviate data insufficiency constraints by accounting for the underlying mechanisms that are not observed in patient data. The proposed method also includes cost-sensitive learning to address imbalanced outcome distributions in patient data for improved predictions.

METHODS

From a major health care organization in Taiwan, we obtained a sample of 10,354 electronic health records that pertained to 6545 patients with peritonitis. The proposed method projects these temporal, heterogeneous, and clinical data into a substantially reduced feature space and then incorporates a latent regulator (latent parameter matrix) to obviate data insufficiencies and account for variations in phenotypic expressions. Moreover, our method employs cost-sensitive learning to further increase the predictive performance.

RESULTS

We evaluated the efficacy of the proposed method for predicting two hepatic complication phenotypes in patients with peritonitis: acute hepatic encephalopathy and hepatorenal syndrome. The following three benchmark techniques were evaluated: temporal multiple measurement case-based reasoning (MMCBR), temporal short long-term memory (T-SLTM) networks, and time fusion convolutional neural network (CNN). For acute hepatic encephalopathy predictions, our method attained an area under the curve (AUC) value of 0.82, which outperforms temporal MMCBR by 64%, T-SLTM by 26%, and time fusion CNN by 26%. For hepatorenal syndrome predictions, our method achieved an AUC value of 0.64, which is 29% better than that of temporal MMCBR (0.54). Overall, the evaluation results show that the proposed method significantly outperforms all the benchmarks, as measured by recall, F-measure, and AUC while maintaining comparable precision values.

CONCLUSIONS

The proposed method learns a short-term temporal representation from patient data to predict complication phenotypes and offers greater predictive utilities than prevalent data-driven techniques. This method is generalizable and can be applied to different acute disease (illness) scenarios that are characterized by insufficient patient clinical data availability, temporal heterogeneities, and imbalanced distributions of important patient outcomes.

Incorporating External Information in Tissue Subtyping: A Topic Modeling Approach

Probabilistic topic models, have been widely deployed for various applications such as learning disease or tissue subtypes. Yet, learning the parameters of such models is usually an ill-posed problem and may result in losing valuable information about disease severity. A common approach is to add a discriminative loss term to the generative model's loss in order to learn a representation that is also predictive of disease severity. However, finding a balance between these two losses is not straightforward. We propose an alternative way in this paper. We develop a framework which allows for incorporating external covariates into the generative model's approximate posterior. These covariates can have more discriminative power for disease severity compared to the representation that we extract from the posterior distribution. For instance, they can be features extracted from a neural network which predicts disease severity from CT images. Effectively, we enforce the generative model's approximate posterior to reside in the subspace of these discriminative covariates. We illustrate our method's application on a large-scale lung CT study of Chronic Obstructive Pulmonary Disease (COPD), a highly heterogeneous disease. We aim at identifying tissue subtypes by using a variant of topic model as a generative model. We quantitatively evaluate the predictive performance of the inferred subtypes and demonstrate that our method outperforms or performs on par with some reasonable baselines. We also show that some of the discovered subtypes are correlated with genetic measurements, suggesting that the identified subtypes may characterize the disease's underlying etiology.

A Multiview Model for Detecting the Inappropriate Use of Prescription Medication: Machine Learning Approach

Background

The inappropriate use of prescription medication has recently garnered worldwide attention, but most national policies do not effectively provide for early detection or timely intervention.

Objective

This study aimed to develop and assess the validity of a model that can detect the inappropriate use of prescription medication. This effort combines a multiview and topic matching method. The study also assessed the validity of this approach.

Methods

A multiview extension of the latent Dirichlet allocation algorithm for topic modeling was chosen to generate diagnosis-medication topics, with data obtained from the Chinese Monitoring Network for Rational Use of Drugs (CMNRUD) database. Topic mapping allowed for calculating the degree to which diagnoses and medications were similarly distributed and, by setting a threshold, for identifying prescription misuse. The Beijing Regional Prescription Review Database (BRPRD) database was used as the gold standard to assess the model’s validity. We also conducted a sensitivity analysis using random samples of validated prescriptions and evaluated the model’s performance.

Results

A total of 44 million prescriptions were used to generate topics using the diagnoses and medications from the CMNRUD database. A random sample (15,000 prescriptions) from the BRPRD was used for validation, and it was found that the model had a sensitivity of 81.8%, specificity of 47.4%, positive-predictive value of 14.5%, and negative-predictive value of 96.0%. The model showed superior stability under different sampling proportions.

Conclusions

A method that combines multiview topic modeling and topic matching can detect the inappropriate use of prescription medication. This model, which has mediocre specificity and moderate sensitivity, can be used as a primary screening tool and will likely complement and improve the process of manually reviewing prescriptions.

A new approach to integrating patient-generated data with expert knowledge for personalized goal setting: A pilot study

INTRODUCTION

Self-monitoring technologies produce patient-generated data that could be leveraged to personalize nutritional goal setting to improve population health; however, most computational approaches are limited when applied to individual-level personalization with sparse and irregular self-monitoring data. We applied informatics methods from expert suggestion systems to a challenging clinical problem: generating personalized nutrition goals from patient-generated diet and blood glucose data.

MATERIALS AND METHODS

We applied qualitative process coding and decision tree modeling to understand how registered dietitians translate patient-generated data into recommendations for dietary self-management of diabetes (i.e., knowledge model). We encoded this process in a set of functions that take diet and blood glucose data as an input and output diet recommendations (i.e., inference engine). Dietitians assessed face validity. Using four patient datasets, we compared our inference engine's output to clinical narratives and gold standards developed by expert clinicians.

RESULTS

To dietitians, the knowledge model represented how recommendations from patient data are made. Inference engine recommendations were 63 % consistent with the gold standard (range = 42 %-75 %) and 74 % consistent with narrative clinical observations (range = 63 %-83 %).

DISCUSSION

Qualitative modeling and automating how dietitians reason over patient data resulted in a knowledge model representing clinical knowledge. However, our knowledge model was less consistent with gold standard than narrative clinical recommendations, raising questions about how best to evaluate approaches that integrate patient-generated data with expert knowledge.

CONCLUSION

New informatics approaches that integrate data-driven methods with expert decision making for personalized goal setting, such as the knowledge base and inference engine presented here, demonstrate the potential to extend the reach of patient-generated data by synthesizing it with clinical knowledge. However, important questions remain about the strengths and weaknesses of computer algorithms developed to discern signal from patient-generated data compared to human experts.

Learning endometriosis phenotypes from patient-generated data

Endometriosis is a systemic and chronic condition in women of childbearing age, yet a highly enigmatic disease with unresolved questions: there are no known biomarkers, nor established clinical stages. We here investigate the use of patient-generated health data and data-driven phenotyping to characterize endometriosis patient subtypes, based on their reported signs and symptoms. We aim at unsupervised learning of endometriosis phenotypes using self-tracking data from personal smartphones. We leverage data from an observational research study of over 4000 women with endometriosis that track their condition over more than 2 years. We extend a classical mixed-membership model to accommodate the idiosyncrasies of the data at hand, i.e., the multimodality and uncertainty of the self-tracked variables. The proposed method, by jointly modeling a wide range of observations (i.e., participant symptoms, quality of life, treatments), identifies clinically relevant endometriosis subtypes. Experiments show that our method is robust to different hyperparameter choices and the biases of self-tracking data (e.g., the wide variations in tracking frequency among participants). With this work, we show the promise of unsupervised learning of endometriosis subtypes from self-tracked data, as learned phenotypes align well with what is already known about the disease, but also suggest new clinically actionable findings. More generally, we argue that a continued research effort on unsupervised phenotyping methods with patient-generated health data via new mobile and digital technologies will have significant impact on the study of enigmatic diseases in particular, and health in general.

A Review of Challenges and Opportunities in Machine Learning for Health

Modern electronic health records (EHRs) provide data to answer clinically meaningful questions. The growing data in EHRs makes healthcare ripe for the use of machine learning. However, learning in a clinical setting presents unique challenges that complicate the use of common machine learning methodologies. For example, diseases in EHRs are poorly labeled, conditions can encompass multiple underlying endotypes, and healthy individuals are underrepresented. This article serves as a primer to illuminate these challenges and highlights opportunities for members of the machine learning community to contribute to healthcare.

Inferring multimodal latent topics from electronic health records

Electronic health records (EHR) are rich heterogeneous collections of patient health information, whose broad adoption provides clinicians and researchers unprecedented opportunities for health informatics, disease-risk prediction, actionable clinical recommendations, and precision medicine. However, EHRs present several modeling challenges, including highly sparse data matrices, noisy irregular clinical notes, arbitrary biases in billing code assignment, diagnosis-driven lab tests, and heterogeneous data types. To address these challenges, we present MixEHR, a multi-view Bayesian topic model. We demonstrate MixEHR on MIMIC-III, Mayo Clinic Bipolar Disorder, and Quebec Congenital Heart Disease EHR datasets. Qualitatively, MixEHR disease topics reveal meaningful combinations of clinical features across heterogeneous data types. Quantitatively, we observe superior prediction accuracy of diagnostic codes and lab test imputations compared to the state-of-art methods. We leverage the inferred patient topic mixtures to classify target diseases and predict mortality of patients in critical conditions. In all comparison, MixEHR confers competitive performance and reveals meaningful disease-related topics.

Veterinary informatics: forging the future between veterinary medicine, human medicine, and One Health initiatives-a joint paper by the Association for Veterinary Informatics (AVI) and the CTSA One Health Alliance (COHA)

Objectives

This manuscript reviews the current state of veterinary medical electronic health records and the ability to aggregate and analyze large datasets from multiple organizations and clinics. We also review analytical techniques as well as research efforts into veterinary informatics with a focus on applications relevant to human and animal medicine. Our goal is to provide references and context for these resources so that researchers can identify resources of interest and translational opportunities to advance the field.

Methods and Results

This review covers various methods of veterinary informatics including natural language processing and machine learning techniques in brief and various ongoing and future projects. After detailing techniques and sources of data, we describe some of the challenges and opportunities within veterinary informatics as well as providing reviews of common One Health techniques and specific applications that affect both humans and animals.

Discussion

Current limitations in the field of veterinary informatics include limited sources of training data for developing machine learning and artificial intelligence algorithms, siloed data between academic institutions, corporate institutions, and many small private practices, and inconsistent data formats that make many integration problems difficult. Despite those limitations, there have been significant advancements in the field in the last few years and continued development of a few, key, large data resources that are available for interested clinicians and researchers. These real-world use cases and applications show current and significant future potential as veterinary informatics grows in importance. Veterinary informatics can forge new possibilities within veterinary medicine and between veterinary medicine, human medicine, and One Health initiatives.

Machine learning for syndromic surveillance using veterinary necropsy reports

The use of natural language data for animal population surveillance represents a valuable opportunity to gather information about potential disease outbreaks, emerging zoonotic diseases, or bioterrorism threats. In this study, we evaluate machine learning methods for conducting syndromic surveillance using free-text veterinary necropsy reports. We train a system to detect if a necropsy report from the Wisconsin Veterinary Diagnostic Laboratory contains evidence of gastrointestinal, respiratory, or urinary pathology. We evaluate the performance of several machine learning algorithms including deep learning with a long short-term memory network. Although no single algorithm was superior, random forest using feature vectors of TF-IDF statistics ranked among the top-performing models with F1 scores of 0.923 (gastrointestinal), 0.960 (respiratory), and 0.888 (urinary). This model was applied to over 33,000 necropsy reports and was used to describe temporal and spatial features of diseases within a 14-year period, exposing epidemiological trends and detecting a potential focus of gastrointestinal disease from a single submitting producer in the fall of 2016.

Unsupervised machine learning for the discovery of latent disease clusters and patient subgroups using electronic health records

Machine learning has become ubiquitous and a key technology on mining electronic health records (EHRs) for facilitating clinical research and practice. Unsupervised machine learning, as opposed to supervised learning, has shown promise in identifying novel patterns and relations from EHRs without using human created labels. In this paper, we investigate the application of unsupervised machine learning models in discovering latent disease clusters and patient subgroups based on EHRs. We utilized Latent Dirichlet Allocation (LDA), a generative probabilistic model, and proposed a novel model named Poisson Dirichlet Model (PDM), which extends the LDA approach using a Poisson distribution to model patients' disease diagnoses and to alleviate age and sex factors by considering both observed and expected observations. In the empirical experiments, we evaluated LDA and PDM on three patient cohorts, namely Osteoporosis, Delirium/Dementia, and Chronic Obstructive Pulmonary Disease (COPD)/Bronchiectasis Cohorts, with their EHR data retrieved from the Rochester Epidemiology Project (REP) medical records linkage system, for the discovery of latent disease clusters and patient subgroups. We compared the effectiveness of LDA and PDM in identifying disease clusters through the visualization of disease representations. We tested the performance of LDA and PDM in differentiating patient subgroups through survival analysis, as well as statistical analysis of demographics and Elixhauser Comorbidity Index (ECI) scores in those subgroups. The experimental results show that the proposed PDM could effectively identify distinguished disease clusters based on the latent patterns hidden in the EHR data by alleviating the impact of age and sex, and that LDA could stratify patients into differentiable subgroups with larger p-values than PDM. However, those subgroups identified by LDA are highly associated with patients' age and sex. The subgroups discovered by PDM might imply the underlying patterns of diseases of greater interest in epidemiology research due to the alleviation of age and sex. Both unsupervised machine learning approaches could be leveraged to discover patient subgroups using EHRs but with different foci.

Identifying sub-phenotypes of acute kidney injury using structured and unstructured electronic health record data with memory networks

Acute Kidney Injury (AKI) is a common clinical syndrome characterized by the rapid loss of kidney excretory function, which aggravates the clinical severity of other diseases in a large number of hospitalized patients. Accurate early prediction of AKI can enable in-time interventions and treatments. However, AKI is highly heterogeneous, thus identification of AKI sub-phenotypes can lead to an improved understanding of the disease pathophysiology and development of more targeted clinical interventions. This study used a memory network-based deep learning approach to discover AKI sub-phenotypes using structured and unstructured electronic health record (EHR) data of patients before AKI diagnosis. We leveraged a real world critical care EHR corpus including 37,486 ICU stays. Our approach identified three distinct sub-phenotypes: sub-phenotype I is with an average age of 63.03±17.25 years, and is characterized by mild loss of kidney excretory function (Serum Creatinine (SCr) 1.55±0.34 mg/dL, estimated Glomerular Filtration Rate Test (eGFR) 107.65±54.98 mL/min/1.73 m²). These patients are more likely to develop stage I AKI. Sub-phenotype II is with average age 66.81±10.43 years, and was characterized by severe loss of kidney excretory function (SCr 1.96±0.49 mg/dL, eGFR 82.19±55.92 mL/min/1.73 m²). These patients are more likely to develop stage III AKI. Sub-phenotype III is with average age 65.07±11.32 years, and was characterized moderate loss of kidney excretory function and thus more likely to develop stage II AKI (SCr 1.69±0.32 mg/dL, eGFR 93.97±56.53 mL/min/1.73 m²). Both SCr and eGFR are significantly different across the three sub-phenotypes with statistical testing plus postdoc analysis, and the conclusion still holds after age adjustment.

Automated grouping of medical codes via multiview banded spectral clustering

OBJECTIVE

With its increasingly widespread adoption, electronic health records (EHR) have enabled phenotypic information extraction at an unprecedented granularity and scale. However, often a medical concept (e.g. diagnosis, prescription, symptom) is described in various synonyms across different EHR systems, hindering data integration for signal enhancement and complicating dimensionality reduction for knowledge discovery. Despite existing ontologies and hierarchies, tremendous human effort is needed for curation and maintenance - a process that is both unscalable and susceptible to subjective biases. This paper aims to develop a data-driven approach to automate grouping medical terms into clinically relevant concepts by combining multiple up-to-date data sources in an unbiased manner.

METHODS

We present a novel data-driven grouping approach - multi-view banded spectral clustering (mvBSC) combining summary data from multiple healthcare systems. The proposed method consists of a banding step that leverages the prior knowledge from the existing coding hierarchy, and a combining step that performs spectral clustering on an optimally weighted matrix.

RESULTS

We apply the proposed method to group ICD-9 and ICD-10-CM codes together by integrating data from two healthcare systems. We show grouping results and hierarchies for 13 representative disease categories. Individual grouping qualities were evaluated using normalized mutual information, adjusted Rand index, and F1-measure, and were found to consistently exhibit great similarity to the existing manual grouping counterpart. The resulting ICD groupings also enjoy comparable interpretability and are well aligned with the current ICD hierarchy.

CONCLUSION

The proposed approach, by systematically leveraging multiple data sources, is able to overcome bias while maximizing consensus to achieve generalizability. It has the advantage of being efficient, scalable, and adaptive to the evolving human knowledge reflected in the data, showing a significant step toward automating medical knowledge integration.

Detecting time-evolving phenotypic topics via tensor factorization on electronic health records: Cardiovascular disease case study

OBJECTIVE

Discovering subphenotypes of complex diseases can help characterize disease cohorts for investigative studies aimed at developing better diagnoses and treatments. Recent advances in unsupervised machine learning on electronic health record (EHR) data have enabled researchers to discover phenotypes without input from domain experts. However, most existing studies have ignored time and modeled diseases as discrete events. Uncovering the evolution of phenotypes - how they emerge, evolve and contribute to health outcomes - is essential to define more precise phenotypes and refine the understanding of disease progression. Our objective was to assess the benefits of an unsupervised approach that incorporates time to model diseases as dynamic processes in phenotype discovery.

METHODS

In this study, we applied a constrained non-negative tensor-factorization approach to characterize the complexity of cardiovascular disease (CVD) patient cohort based on longitudinal EHR data. Through tensor-factorization, we identified a set of phenotypic topics (i.e., subphenotypes) that these patients established over the 10 years prior to the diagnosis of CVD, and showed the progress pattern. For each identified subphenotype, we examined its association with the risk for adverse cardiovascular outcomes estimated by the American College of Cardiology/American Heart Association Pooled Cohort Risk Equations, a conventional CVD-risk assessment tool frequently used in clinical practice. Furthermore, we compared the subsequent myocardial infarction (MI) rates among the six most prevalent subphenotypes using survival analysis.

RESULTS

From a cohort of 12,380 adult CVD individuals with 1068 unique PheCodes, we successfully identified 14 subphenotypes. Through the association analysis with estimated CVD risk for each subtype, we found some phenotypic topics such as Vitamin D deficiency and depression, Urinary infections cannot be explained by the conventional risk factors. Through a survival analysis, we found markedly different risks of subsequent MI following the diagnosis of CVD among the six most prevalent topics (p < 0.0001), indicating these topics may capture clinically meaningful subphenotypes of CVD.

CONCLUSION

This study demonstrates the potential benefits of using tensor-decomposition to model diseases as dynamic processes from longitudinal EHR data. Our results suggest that this data-driven approach may potentially help researchers identify complex and chronic disease subphenotypes in precision medicine research.

The parameter Houlihan: A solution to high-throughput identifiability indeterminacy for brutally ill-posed problems

One way to interject knowledge into clinically impactful forecasting is to use data assimilation, a nonlinear regression that projects data onto a mechanistic physiologic model, instead of a set of functions, such as neural networks. Such regressions have an advantage of being useful with particularly sparse, non-stationary clinical data. However, physiological models are often nonlinear and can have many parameters, leading to potential problems with parameter identifiability, or the ability to find a unique set of parameters that minimize forecasting error. The identifiability problems can be minimized or eliminated by reducing the number of parameters estimated, but reducing the number of estimated parameters also reduces the flexibility of the model and hence increases forecasting error. We propose a method, the parameter Houlihan, that combines traditional machine learning techniques with data assimilation, to select the right set of model parameters to minimize forecasting error while reducing identifiability problems. The method worked well: the data assimilation-based glucose forecasts and estimates for our cohort using the Houlihan-selected parameter sets generally also minimize forecasting errors compared to other parameter selection methods such as by-hand parameter selection. Nevertheless, the forecast with the lowest forecast error does not always accurately represent physiology, but further advancements of the algorithm provide a path for improving physiologic fidelity as well. Our hope is that this methodology represents a first step toward combining machine learning with data assimilation and provides a lower-threshold entry point for using data assimilation with clinical data by helping select the right parameters to estimate.

Phenotype Inference with Semi-Supervised Mixed Membership Models

Disease phenotyping algorithms are designed to sift through clinical data stores to identify patients with specific diseases. Supervised phenotyping methods require significant quantities of expert-labeled data, while unsupervised methods may learn spurious or non-disease phenotypes. To address these limitations, we propose the Semi-Supervised Mixed Membership Model (SS3M) - a probabilistic graphical model for learning disease phenotypes from partially labeled clinical data. We show SS3M can generate interpretable, disease-specific phenotypes which capture the clinical features of the disease concepts specified by the labels provided to the model. Furthermore, SS3M phenotypes demonstrate competitive predictive performance relative to commonly used baselines.

VetTag: improving automated veterinary diagnosis coding via large-scale language modeling

Unlike human medical records, most of the veterinary records are free text without standard diagnosis coding. The lack of systematic coding is a major barrier to the growing interest in leveraging veterinary records for public health and translational research. Recent machine learning effort is limited to predicting 42 top-level diagnosis categories from veterinary notes. Here we develop a large-scale algorithm to automatically predict all 4577 standard veterinary diagnosis codes from free text. We train our algorithm on a curated dataset of over 100 K expert labeled veterinary notes and over one million unlabeled notes. Our algorithm is based on the adapted Transformer architecture and we demonstrate that large-scale language modeling on the unlabeled notes via pretraining and as an auxiliary objective during supervised learning greatly improves performance. We systematically evaluate the performance of the model and several baselines in challenging settings where algorithms trained on one hospital are evaluated in a different hospital with substantial domain shift. In addition, we show that hierarchical training can address severe data imbalances for fine-grained diagnosis with a few training cases, and we provide interpretation for what is learned by the deep network. Our algorithm addresses an important challenge in veterinary medicine, and our model and experiments add insights into the power of unsupervised learning for clinical natural language processing.

The Revival of the Notes Field: Leveraging the Unstructured Content in Electronic Health Records

Problem: Clinical practice requires the production of a time- and resource-consuming great amount of notes. They contain relevant information, but their secondary use is almost impossible, due to their unstructured nature. Researchers are trying to address this problems, with traditional and promising novel techniques. Application in real hospital settings seems not to be possible yet, though, both because of relatively small and dirty dataset, and for the lack of language-specific pre-trained models. Aim: Our aim is to demonstrate the potential of the above techniques, but also raise awareness of the still open challenges that the scientific communities of IT and medical practitioners must jointly address to realize the full potential of unstructured content that is daily produced and digitized in hospital settings, both to improve its data quality and leverage the insights from data-driven predictive models. Methods: To this extent, we present a narrative literature review of the most recent and relevant contributions to leverage the application of Natural Language Processing techniques to the free-text content electronic patient records. In particular, we focused on four selected application domains, namely: data quality, information extraction, sentiment analysis and predictive models, and automated patient cohort selection. Then, we will present a few empirical studies that we undertook at a major teaching hospital specializing in musculoskeletal diseases. Results: We provide the reader with some simple and affordable pipelines, which demonstrate the feasibility of reaching literature performance levels with a single institution non-English dataset. In such a way, we bridged literature and real world needs, performing a step further toward the revival of notes fields.

Temporal phenotyping of medically complex children via PARAFAC2 tensor factorization

OBJECTIVE

Our aim is to extract clinically-meaningful phenotypes from longitudinal electronic health records (EHRs) of medically-complex children. This is a fragile set of patients consuming a disproportionate amount of pediatric care resources but who often end up with sub-optimal clinical outcome. The rise in available electronic health records (EHRs) provide a rich data source that can be used to disentangle their complex clinical conditions into concise, clinically-meaningful groups of characteristics. We aim at identifying those phenotypes and their temporal evolution in a scalable, computational manner, which avoids the time-consuming manual chart review.

MATERIALS AND METHODS

We analyze longitudinal EHRs from Children's Healthcare of Atlanta including 1045 medically complex patients with a total of 59,948 encounters over 2 years. We apply a tensor factorization method called PARAFAC2 to extract: (a) clinically-meaningful groups of features (b) concise patient representations indicating the presence of a phenotype for each patient, and (c) temporal signatures indicating the evolution of those phenotypes over time for each patient.

RESULTS

We identified four medically complex phenotypes, namely gastrointestinal disorders, oncological conditions, blood-related disorders, and neurological system disorders, which have distinct clinical characterizations among patients. We demonstrate the utility of patient representations produced by PARAFAC2, towards identifying groups of patients with significant survival variations. Finally, we showcase representative examples of the temporal phenotypic trends extracted for different patients.

DISCUSSION

Unsupervised temporal phenotyping is an important task since it minimizes the burden on behalf of clinical experts, by relegating their involvement in the output phenotypes' validation. PARAFAC2 enjoys several compelling properties towards temporal computational phenotyping: (a) it is able to handle high-dimensional data and variable numbers of encounters across patients, (b) it has an intuitive interpretation and (c) it is free from ad-hoc parameter choices. Computational phenotypes, such as the ones computed by our approach, have multiple applications; we highlight three of them which are particularly useful for medically complex children: (1) integration into clinical decision support systems, (2) interpretable mortality prediction and 3) clinical trial recruitment.

CONCLUSION

PARAFAC2 can be applied to unsupervised temporal phenotyping tasks where precise definitions of different phenotypes are absent, and lengths of patient records are varying.

EHR phenotyping via jointly embedding medical concepts and words into a unified vector space

Background

There has been an increasing interest in learning low-dimensional vector representations of medical concepts from Electronic Health Records (EHRs). Vector representations of medical concepts facilitate exploratory analysis and predictive modeling of EHR data to gain insights about the patterns of care and health outcomes. EHRs contain structured data such as diagnostic codes and laboratory tests, as well as unstructured free text data in form of clinical notes, which provide more detail about condition and treatment of patients.

Methods

In this work, we propose a method that jointly learns vector representations of medical concepts and words. This is achieved by a novel learning scheme based on the word2vec model. Our model learns those relationships by integrating clinical notes and sets of accompanying medical codes and by defining joint contexts for each observed word and medical code.

Results

In our experiments, we learned joint representations using MIMIC-III data. Using the learned representations of words and medical codes, we evaluated phenotypes for 6 diseases discovered by our and baseline method. The experimental results show that for each of the 6 diseases our method finds highly relevant words. We also show that our representations can be very useful when predicting the reason for the next visit.

Conclusions

The jointly learned representations of medical concepts and words capture not only similarity between codes or words themselves, but also similarity between codes and words. They can be used to extract phenotypes of different diseases. The representations learned by the joint model are also useful for construction of patient features.

Innovation in Pharmacovigilance: Use of Artificial Intelligence in Adverse Event Case Processing

Automation of pharmaceutical safety case processing represents a significant opportunity to affect the strongest cost driver for a company's overall pharmacovigilance budget. A pilot was undertaken to test the feasibility of using artificial intelligence and robotic process automation to automate processing of adverse event reports. The pilot paradigm was used to simultaneously test proposed solutions of three commercial vendors. The result confirmed the feasibility of using artificial intelligence–based technology to support extraction from adverse event source documents and evaluation of case validity. In addition, the pilot demonstrated viability of the use of safety database data fields as a surrogate for otherwise time‐consuming and costly direct annotation of source documents. Finally, the evaluation and scoring method used in the pilot was able to differentiate vendor capabilities and identify the best candidate to move into the discovery phase.

DeepTag: inferring diagnoses from veterinary clinical notes

Large scale veterinary clinical records can become a powerful resource for patient care and research. However, clinicians lack the time and resource to annotate patient records with standard medical diagnostic codes and most veterinary visits are captured in free-text notes. The lack of standard coding makes it challenging to use the clinical data to improve patient care. It is also a major impediment to cross-species translational research, which relies on the ability to accurately identify patient cohorts with specific diagnostic criteria in humans and animals. In order to reduce the coding burden for veterinary clinical practice and aid translational research, we have developed a deep learning algorithm, DeepTag, which automatically infers diagnostic codes from veterinary free-text notes. DeepTag is trained on a newly curated dataset of 112,558 veterinary notes manually annotated by experts. DeepTag extends multitask LSTM with an improved hierarchical objective that captures the semantic structures between diseases. To foster human-machine collaboration, DeepTag also learns to abstain in examples when it is uncertain and defers them to human experts, resulting in improved performance. DeepTag accurately infers disease codes from free-text even in challenging cross-hospital settings where the text comes from different clinical settings than the ones used for training. It enables automated disease annotation across a broad range of clinical diagnoses with minimal preprocessing. The technical framework in this work can be applied in other medical domains that currently lack medical coding resources.

When to re-order laboratory tests? Learning laboratory test shelf-life

Most laboratory results are valid for only a certain time period (laboratory tests shelf-life), after which they are outdated and the test needs to be re-administered. Currently, laboratory test shelf-lives are not centrally available anywhere but the implicit knowledge of doctors. In this work we propose an automated method to learn laboratory test-specific shelf-life by identifying prevalent laboratory test order patterns in electronic health records. The resulting shelf-lives performed well in the evaluation of internal validity, clinical interpretability, and external validity.