Deriving disease subtypes from electronic health records (EHRs) can guide next-generation personalized medicine. However, challenges in summarizing and representing patient data prevent widespread practice of scalable EHR-based stratification analysis. Here we present an unsupervised framework based on deep learning to process heterogeneous EHRs and derive patient representations that can efficiently and effectively enable patient stratification at scale. We considered EHRs of 1,608,741 patients from a diverse hospital cohort comprising a total of 57,464 clinical concepts. We introduce a representation learning model based on word embeddings, convolutional neural networks, and autoencoders (i.e., ConvAE) to transform patient trajectories into lowdimensional latent vectors. We evaluated these representations as broadly enabling patient stratification by applying hierarchical clustering to different multi-disease and disease-specific patient cohorts. ConvAE significantly outperformed several baselines in a clustering task to identify patients with different complex conditions, with 2.61 entropy and 0.31 purity average scores. When applied to stratify patients within a certain condition, ConvAE led to various clinically relevant subtypes for different disorders, including type 2 diabetes, Parkinson's disease, and Alzheimer's disease, largely related to comorbidities, disease progression, and symptom severity. With these results, we demonstrate that ConvAE can generate patient representations that lead to clinically meaningful insights. This scalable framework can help better understand varying etiologies in heterogeneous sub-populations and unlock patterns for EHR-based research in the realm of personalized medicine.

Introduction Informed decision making has been highlighted as an important aspect of lung cancer screening programs. This study seeks to assess the efficacy of a web-based patient decision aid for lung cancer screening, www.shouldiscreen.com. Methods A before-and-after study (August through December 2014) was conducted where participants navigated a web-based decision aid that provided information about low-dose computed tomography lung cancer screening. Using an established prediction model, the decision aid computed baseline lung cancer risk and an individual’s chances of benefiting from, and risk of being harmed by, screening. Outcome measures included knowledge of lung cancer risk factors and lung cancer screening, decisional conflict, concordance, and acceptability of the decision aid. Data were collected from 60 participants who were current or former smokers, had no history of lung cancer, and had not received a chest computed tomographic scan in the previous year. Analysis took place in 2015. Results Knowledge increased after seeing the decision aid compared with before (p<0.001), whereas the score on the Decisional Conflict Scale decreased (p<0.001). Concordance between a participant’s preference to screen and the U.S. Preventive Services Task Force recommendation improved after seeing the decision aid (p<0.001). Risk perceptions among the screen-ineligible group changed (n=49), contrary to those who were eligible (n=11). Ninety-seven percent of the participants reported that the decision aid was likely useful for lung cancer screening decision making. Conclusions The web-based decision aid should be a helpful resource for individuals considering lung cancer screening, as well as for practitioners and health systems with lung cancer screening programs.

PURPOSEThe paradox of primary care is the observation that primary care is associated with apparently low levels of evidence-based care for individual diseases, but systems based on primary care have healthier populations, use fewer resources, and have less health inequality. The purpose of this article is to explore, from a complex systems perspective, mechanisms that might account for the effects of primary care beyond disease-specific care. METHODSIn an 8-session, participatory group model-building process, patient, caregiver, and primary care clinician community stakeholders worked with academic investigators to develop and refine an agent-based computer simulation model to test hypotheses about mechanisms by which features of primary care could affect health and health equity.RESULTS In the resulting model, patients are at risk for acute illness, acute lifechanging illness, chronic illness, and mental illness. Patients have changeable health behaviors and care-seeking tendencies that relate to their living in advantaged or disadvantaged neighborhoods. There are 2 types of care available to patients: primary and specialty. Primary care in the model is less effective than specialty care in treating single diseases, but it has the ability to treat multiple diseases at once. Primary care also can provide disease prevention visits, help patients improve their health behaviors, refer to specialty care, and develop relationships with patients that cause them to lower their threshold for seeking care. In a model run with primary care features turned off, primary care patients have poorer health. In a model run with all primary care features turned on, their conjoint effect leads to better population health for patients who seek primary care, with the primary care effect being particularly pronounced for patients who are disadvantaged and patients with multiple chronic conditions. Primary care leads to more total health care visits that are due to more disease prevention visits, but there are reduced illness visits among people in disadvantaged neighborhoods. Supplemental appendices provide a working version of the model and worksheets that allow readers to run their own experiments that vary model parameters. CONCLUSION This simulation model provides insights into possible mechanisms for the paradox of primary care and shows how participatory group model building can be used to evaluate hypotheses about the behavior of such complex systems as primary health care and population health. INTRODUCTIONM ultiple studies have found that primary care is associated with poorer quality care for individual diseases than is care provided by clinicians focused primarily on those diseases. [1][2][3][4][5][6] Yet, other evidence shows that systems based on primary care have better quality of care, better population health, greater equity, and lower cost. [7][8][9][10][11] This discrepancy between apparently poor disease-specific care and advantageous outcomes at the level of the whole person and system has been called t...