Enhancing the prediction of disease outcomes using electronic health records and pretrained deep learning models

TL;DR

This study demonstrates that a pretrained encoder-decoder deep learning model significantly enhances the accuracy of predicting various patient outcomes from electronic health records, outperforming existing models.

Contribution

Introduces a denoising sequence-to-sequence model pretrained on large EHR datasets that improves outcome prediction accuracy over current state-of-the-art models.

Findings

Outperformed pretrained BERT on multiple outcomes

Effective in predicting self-harm and pancreatic cancer

Deep bidirectional autoregressive representations enhance predictions

Abstract

Question: Can an encoder-decoder architecture pretrained on a large dataset of longitudinal electronic health records improves patient outcome predictions? Findings: In this prognostic study of 6.8 million patients, our denoising sequence-to-sequence prediction model of multiple outcomes outperformed state-of-the-art models scuh pretrained BERT on a broad range of patient outcomes, including intentional self-harm and pancreatic cancer. Meaning: Deep bidirectional and autoregressive representation improves patient outcome prediction.