# Scalable and accurate deep learning for electronic health records

**Authors:** Alvin Rajkomar, Eyal Oren, Kai Chen, Andrew M. Dai, Nissan Hajaj, Peter J. Liu, Xiaobing Liu, Mimi Sun, Patrik Sundberg, Hector Yee, Kun Zhang, Gavin E. Duggan, Gerardo Flores, Michaela Hardt, Jamie Irvine, Quoc Le, Kurt Litsch, Jake Marcus, Alexander Mossin, Justin Tansuwan  
**arXiv:** [1801.07860](https://arxiv.org/abs/1801.07860)  
**Date:** 2018-01-24  
**Summary page:** https://www.summarizepaper.com/en/paper/1801.07860v1/

## Key points

- Electronic health record (EHR) data can revolutionize personalized medicine and improve healthcare quality
- Constructing predictive statistical models from normalized EHR data is labor-intensive and discards most of the information in each patient's record
- Researchers proposed a representation based on the Fast Healthcare Interoperability Resources (FHIR) format to address this issue
- This approach uses deep learning methods to accurately predict multiple medical events from multiple centers without site-specific data harmonization
- The researchers validated their approach using de-identified EHR data from two U.S. academic medical centers with 216,221 adult patients hospitalized for at least 24 hours
- The deep learning models achieved high accuracy for tasks such as predicting in-hospital mortality, 30-day unplanned readmission, prolonged length of stay, and all of a patient's final discharge diagnoses
- These models outperformed state-of-the-art traditional predictive models in all cases and offered transparency into the predictions through a case-study neural network attribution system that highlights evidence in the patient's chart and allows clinicians to gain insight into how predictions are made
- This approach has significant potential to create accurate and scalable predictions for various clinical scenarios while providing explanations directly linked to evidence in patients' charts - an essential step towards achieving personalized medicine and improving healthcare quality.

## Summary

The use of electronic health record (EHR) data for predictive modeling is expected to revolutionize personalized medicine and improve healthcare quality. To address the labor-intensive process of constructing predictive statistical models from normalized EHR data, which discards most of the information in each patient's record, a team of researchers proposed a representation based on the Fast Healthcare Interoperability Resources (FHIR) format. This approach uses deep learning methods to accurately predict multiple medical events from multiple centers without site-specific data harmonization. The researchers validated their approach using de-identified EHR data from two U.S. academic medical centers with 216,221 adult patients hospitalized for at least 24 hours. In the sequential format they proposed, this volume of EHR data unrolled into a total of 46,864,534,945 data points including clinical notes. The deep learning models achieved high accuracy for tasks such as predicting in-hospital mortality (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 state-of-the-art traditional predictive models in all cases and offered transparency into the predictions through a case-study neural network attribution system that highlights evidence in the patient's chart and allows clinicians to gain insight into how predictions are made. This approach has significant potential to create accurate and scalable predictions for various clinical scenarios while providing explanations directly linked to evidence in patients' charts - an essential step towards achieving personalized medicine and improving healthcare quality.