Multilevel Self-Attention Model and its Use on Medical Risk Prediction.

Various deep learning models have been developed for different healthcare predictive tasks using Electronic Health Records and have shown promising performance. In these models, medical codes are often aggregated into visit representation without considering their heterogeneity, e.g., the same diagnosis might imply different healthcare concerns with different procedures or medications. Then the visits are often fed into deep learning models, such as recurrent neural networks, sequentially without considering the irregular temporal information and dependencies among visits. To address these limitations, we developed a Multilevel Self-Attention Model (MSAM) that can capture the underlying relationships between medical codes and between medical visits. We compared MSAM with various baseline models on two predictive tasks, i.e., future disease prediction and future medical cost prediction, with two large datasets, i.e., MIMIC-3 and PFK. In the experiments, MSAM consistently outperformed baseline models. Additionally, for future medical cost prediction, we used disease prediction as an auxiliary task, which not only guides the model to achieve a stronger and more stable financial prediction, but also allows managed care organizations to provide a better care coordination.

Graph embedding on biomedical networks: methods, applications and evaluations.

MOTIVATION: Graph embedding learning that aims to automatically learn low-dimensional node representations, has drawn increasing attention in recent years. To date, most recent graph embedding methods are evaluated on social and information networks and are not comprehensively studied on biomedical networks under systematic experiments and analyses. On the other hand, for a variety of biomedical network analysis tasks, traditional techniques such as matrix factorization (which can be seen as a type of graph embedding methods) have shown promising results, and hence there is a need to systematically evaluate the more recent graph embedding methods (e.g. random walk-based and neural network-based) in terms of their usability and potential to further the state-of-the-art. RESULTS: We select 11 representative graph embedding methods and conduct a systematic comparison on 3 important biomedical link prediction tasks: drug-disease association (DDA) prediction, drug-drug interaction (DDI) prediction, protein-protein interaction (PPI) prediction; and 2 node classification tasks: medical term semantic type classification, protein function prediction. Our experimental results demonstrate that the recent graph embedding methods achieve promising results and deserve more attention in the future biomedical graph analysis. Compared with three state-of-the-art methods for DDAs, DDIs and protein function predictions, the recent graph embedding methods achieve competitive performance without using any biological features and the learned embeddings can be treated as complementary representations for the biological features. By summarizing the experimental results, we provide general guidelines for properly selecting graph embedding methods and setting their hyper-parameters for different biomedical tasks. AVAILABILITY AND IMPLEMENTATION: As part of our contributions in the paper, we develop an easy-to-use Python package with detailed instructions, BioNEV, available at: https://github.com/xiangyue9607/BioNEV, including all source code and datasets, to facilitate studying various graph embedding methods on biomedical tasks. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

A deep learning model for pediatric patient risk stratification.

OBJECTIVES: Current models for patient risk prediction rely on practitioner expertise and domain knowledge. This study presents a deep learning model-a type of machine learning that does not require human inputs-to analyze complex clinical and financial data for population risk stratification. STUDY DESIGN: A comparative predictive analysis of deep learning versus other popular risk prediction modeling strategies using medical claims data from a cohort of 112,641 pediatric accountable care organization members. METHODS: "Skip-Gram," an unsupervised deep learning approach that uses neural networks for prediction modeling, used data from 2014 and 2015 to predict the risk of hospitalization in 2016. The area under the curve (AUC) of the deep learning model was compared with that of both the Clinical Classifications Software and the commercial DxCG Intelligence predictive risk models, each with and without demographic and utilization features. We then calculated costs for patients in the top 1% and 5% of hospitalization risk identified by each model. RESULTS: The deep learning model performed the best across 6 predictive models, with an AUC of 75.1%. The top 1% of members selected by the deep learning model had a combined healthcare cost $5 million higher than that of the group identified by the DxCG Intelligence model. CONCLUSIONS: The deep learning model outperforms the traditional risk models in prospective hospitalization prediction. Thus, deep learning may improve the ability of managed care organizations to perform predictive modeling of financial risk, in addition to improving the accuracy of risk stratification for population health management activities.

Comorbidities in Childhood Celiac Disease: A Phenome Wide Association Study Using the Electronic Health Record.

OBJECTIVES: Celiac disease (CD) is associated with a variety of extraintestinal autoimmune and inflammatory findings that manifest clinically as symptoms and comorbidities. Understanding these comorbidities may improve identification of the disease and prevent sequelae. In this study, we use an unbiased electronic health record (EHR)-based Phenome Wide Association Study (PheWAS) method to confirm known comorbidities, discover novel associations and enhance characterization of the clinical presentation of CD in children. METHODS: Data were extracted from the Nationwide Children's Hospital EHR. Confirmed CD cases (n = 433) were matched with 4330 randomly selected controls. Utilizing an EHR-based PheWAS method to analyze associations of phenotypes with CD, we conducted an unbiased screening of all International Classification of Diseases, 10th revision diagnostic codes and examined significance by performing Fisher's Exact tests. We further tested for the association between CD and 14 previously identified comorbidities in an a priori fashion. RESULTS: We found 45 International Classification of Diseases, 10th revision codes significantly associated with CD. Thirteen are known comorbidities and nine are expected symptoms of CD, thus validating our study methods. Further investigation found symptoms that characterized CD clinically and discovered a significant association between eosinophilic disorders of the esophagus and CD. Of 14 previously identified comorbidities, 8 were significantly associated with CD. CONCLUSIONS: An EHR-based PheWAS method is a powerful, efficient, and cost-effective method to screen for possible CD comorbidities and validate associations at the population level. Ours is the first PheWAS of CD to confirm a significant association of eosinophilic disorders of the esophagus with CD in a controlled study.

Preterm Birth and Gestational Length in Four Race-Nativity Groups, Including Somali Americans.

OBJECTIVE: To compare preterm birth rates and gestational length in four race-nativity groups including Somali Americans. METHODS: Using a retrospective cohort study design of Ohio birth certificates, we analyzed all singleton births between 2000 and 2015 from four groups of women categorized as U.S.-born, non-Hispanic white (USBW), U.S.-born, non-Hispanic black (USBB), African-born black (ABB, primarily of West African birth country), and Somalia-born (SB). An algorithm trained on maternal names was used to confirm Somali ethnicity. Gestational length was analyzed as completed weeks or aggregated by clinically relevant periods. Risk of spontaneous and health care provider-initiated preterm birth was calculated in a competing risk model. RESULTS: Births to women in the designated groups accounted for 1,960,693 births (USBW n=1,638,219; USBB n=303,028; ABB n=10,966, and SB n=8,480). Women in the SB group had a lower preterm birth rate (5.9%) compared with women in the USBB (13.0%), ABB (8.4%), and USBW (7.9%) groups (P<.001). Women in the SB group had a higher frequency of postterm pregnancy (5.8% vs less than 1%, P<.001 for all groups). The lower rate of preterm birth in the SB group was unrelated to differences in parity or smoking or whether preterm birth was spontaneous or health care provider-initiated. The lower rate of preterm birth and tendency for prolonged gestation was attenuated in ethnic Somali women born outside Somalia. CONCLUSION: We report a positive disparity in preterm birth and a tendency for prolonged gestation for ethnic Somali women in Ohio. Etiologic studies in multiethnic cohorts aimed to uncover the sociobiological determinants of gestational length may lead to practical approaches to reduce prematurity in the general population.

'RE:fine drugs': an interactive dashboard to access drug repurposing opportunities.

The process of discovering new drugs has been extremely costly and slow in the last decades despite enormous investment in pharmaceutical research. Drug repurposing enables researchers to speed up the process of discovering other conditions that existing drugs can effectively treat, with low cost and fast FDA approval. Here, we introduce 'RE:fine Drugs', a freely available interactive website for integrated search and discovery of drug repurposing candidates from GWAS and PheWAS repurposing datasets constructed using previously reported methods in Nature Biotechnology. 'RE:fine Drugs' demonstrates the possibilities to identify and prioritize novelty of candidates for drug repurposing based on the theory of transitive Drug-Gene-Disease triads. This public website provides a starting point for research, industry, clinical and regulatory communities to accelerate the investigation and validation of new therapeutic use of old drugs.Database URL: http://drug-repurposing.nationwidechildrens.org.

Drug Repurposing Hypothesis Generation Using the "RE:fine Drugs" System.

The promise of drug repurposing is that existing drugs may be used for new disease indications in order to curb the high costs and time for approval. The goal of computational methods for drug repurposing is to enable solutions for safer, cheaper and faster drug discovery. Towards this end, we developed a novel method that integrates genetic and clinical phenotype data from large-scale GWAS and PheWAS studies with detailed drug information on the concept of transitive Drug-Gene-Disease triads. We created "RE:fine Drugs," a freely available, interactive dashboard that automates gene, disease and drug-based searches to identify drug repurposing candidates. This web-based tool supports a user-friendly interface that includes an array of advanced search and export options. Results can be prioritized in a variety of ways, including but not limited to, biomedical literature support, strength and statistical significance of GWAS and/or PheWAS associations, disease indications and molecular drug targets. Here we provide a protocol that illustrates the functionalities available in the "RE:fine Drugs" system and explores the different advanced options through a case study.

A Review on Genomics APIs.

The constant improvement and falling prices of whole human genome Next Generation Sequencing (NGS) has resulted in rapid adoption of genomic information at both clinics and research institutions. Considered together, the complexity of genomics data, due to its large volume and diversity along with the need for genomic data sharing, has resulted in the creation of Application Programming Interface (API) for secure, modular, interoperable access to genomic data from different applications, platforms, and even organizations. The Genomics APIs are a set of special protocols that assist software developers in dealing with multiple genomic data sources for building seamless, interoperable applications leading to the advancement of both genomic and clinical research. These APIs help define a standard for retrieval of genomic data from multiple sources as well as to better package genomic information for integration with Electronic Health Records. This review covers three currently available Genomics APIs: a) Google Genomics, b) SMART Genomics, and c) 23andMe. The functionalities, reference implementations (if available) and authentication protocols of each API are reviewed. A comparative analysis of the different features across the three APIs is provided in the Discussion section. Though Genomics APIs are still under active development and have yet to reach widespread adoption, they hold the promise to make building of complicated genomics applications easier with downstream constructive effects on healthcare.

An automated approach for ranking journals to help in clinician decision support.

Point of care access to knowledge from full text journal articles supports decision-making and decreases medical errors. However, it is an overwhelming task to search through full text journal articles and find quality information needed by clinicians. We developed a method to rate journals for a given clinical topic, Congestive Heart Failure (CHF). Our method enables filtering of journals and ranking of journal articles based on source journal in relation to CHF. We also obtained a journal priority score, which automatically rates any journal based on its importance to CHF. Comparing our ranking with data gathered by surveying 169 cardiologists, who publish on CHF, our best Multiple Linear Regression model showed a correlation of 0.880, based on five-fold cross validation. Our ranking system can be extended to other clinical topics.

Towards Transforming Expert-based Content to Evidence-based Content.

The goal of this paper is to find relevant citations for clinicians' written content and make it more reliable by adding scientific articles as references and enabling the clinicians to easily update it using new information. The proposed approach uses information retrieval and ranking techniques to extract and rank relevant citations from MEDLINE for any given sentence. Additionally, this system extracts snippets of relevant content from ranked citations. We assessed our approach on 4,697 MEDLINE papers and their corresponding full-text on the subject of Heart Failure. We implemented multi-level and weight ranking algorithms to rank the citations. We demonstrate that using journal relevance and study design type improves results obtained from only using content similarity by approximately 40%. We also show that using full-text, rather than abstract text, leads to extracting higher quality snippets.

Automatically identifying health- and clinical-related content in wikipedia.

Physicians are increasingly using the Internet for finding medical information related to patient care. Wikipedia is a valuable online medical resource to be integrated into existing clinical question answering (QA) systems. On the other hand, Wikipedia contains a full spectrum of world's knowledge and therefore comprises a large partition of non-health-related content, which makes disambiguation more challenging and consequently leads to large overhead for existing systems to effectively filter irrelevant information. To overcome this, we have developed both unsupervised and supervised approaches to identify health-related articles as well as clinically relevant articles. Furthermore, we explored novel features by extracting health related hierarchy from the Wikipedia category network, from which a variety of features were derived and evaluated. Our experiments show promising results and also demonstrate that employing the category hierarchy can effectively improve the system performance.

MedTxting: learning based and knowledge rich SMS-style medical text contraction.

In mobile health (M-health), Short Message Service (SMS) has shown to improve disease related self-management and health service outcomes, leading to enhanced patient care. However, the hard limit on character size for each message limits the full value of exploring SMS communication in health care practices. To overcome this problem and improve the efficiency of clinical workflow, we developed an innovative system, MedTxting (available at http://medtxting.askhermes.org), which is a learning-based but knowledge-rich system that compresses medical texts in a SMS style. Evaluations on clinical questions and discharge summary narratives show that MedTxting can effectively compress medical texts with reasonable readability and noticeable size reduction. Findings in this work reveal potentials of MedTxting to the clinical settings, allowing for real-time and cost-effective communication, such as patient condition reporting, medication consulting, physicians connecting to share expertise to improve point of care.