Leveraging Deep Representations of Radiology Reports in Survival Analysis for Predicting Heart Failure Patient Mortality

TL;DR

This paper introduces a novel approach that uses BERT-derived features from radiology reports to improve survival analysis predictions for heart failure patient mortality, outperforming previous models.

Contribution

The study demonstrates that BERT-based hidden layer representations significantly enhance survival prediction accuracy over traditional feature-based methods.

Findings

BERT-based features outperform predefined features by 5.7% on average.

The method achieves higher C-index and time-dependent AUC scores.

The approach is publicly available for further research.

Abstract

Utilizing clinical texts in survival analysis is difficult because they are largely unstructured. Current automatic extraction models fail to capture textual information comprehensively since their labels are limited in scope. Furthermore, they typically require a large amount of data and high-quality expert annotations for training. In this work, we present a novel method of using BERT-based hidden layer representations of clinical texts as covariates for proportional hazards models to predict patient survival outcomes. We show that hidden layers yield notably more accurate predictions than predefined features, outperforming the previous baseline model by 5.7% on average across C-index and time-dependent AUC. We make our work publicly available at https://github.com/bionlplab/heart_failure_mortality.