Scalable and accurate deep learning with electronic health records

TLDR

Predictive modeling with EHR data is expected to advance personalized medicine, yet traditional approaches require labor‑intensive extraction of curated predictors, discarding most of the information in a patient’s record. The authors propose a FHIR‑based representation of raw EHR records to enable scalable predictive modeling. They validated this approach on de‑identified data from two US academic centers comprising 216,221 hospitalized adults, unrolling the records into 46.9 billion data points, including clinical notes. Deep learning models built on this representation achieved high accuracy across multiple tasks—AUROC 0.93–0.94 for in‑hospital mortality, 0.75–0.76 for 30‑day readmission, 0.85–0.86 for prolonged length of stay, and 0.90 for discharge diagnoses—outperforming traditional models and identifying relevant chart information.

Abstract

Abstract Predictive modeling with electronic health record (EHR) data is anticipated to drive personalized medicine and improve healthcare quality. Constructing predictive statistical models typically requires extraction of curated predictor variables from normalized EHR data, a labor-intensive process that discards the vast majority of information in each patient’s record. We propose a representation of patients’ entire raw EHR records based on the Fast Healthcare Interoperability Resources (FHIR) format. We demonstrate that deep learning methods using this representation are capable of accurately predicting multiple medical events from multiple centers without site-specific data harmonization. We validated our approach using de-identified EHR data from two US academic medical centers with 216,221 adult patients hospitalized for at least 24 h. In the sequential format we propose, this volume of EHR data unrolled into a total of 46,864,534,945 data points, including clinical notes. Deep learning models achieved high accuracy for tasks such as predicting: in-hospital mortality (area under the receiver operator curve [AUROC] across sites 0.93–0.94), 30-day unplanned readmission (AUROC 0.75–0.76), prolonged length of stay (AUROC 0.85–0.86), and all of a patient’s final discharge diagnoses (frequency-weighted AUROC 0.90). These models outperformed traditional, clinically-used predictive models in all cases. We believe that this approach can be used to create accurate and scalable predictions for a variety of clinical scenarios. In a case study of a particular prediction, we demonstrate that neural networks can be used to identify relevant information from the patient’s chart.