ReToP: Learning to Rewrite Electronic Health Records for Clinical Prediction

2026-01-27

AI Summary

This work addresses the challenges posed by the high dimensionality, heterogeneity, and sparsity of electronic health records (EHRs), which hinder clinical prediction performance, and the limitation of existing large language model approaches that are often task-agnostic and fail to effectively incorporate predictive signals. To overcome these issues, the authors propose ReToP, a framework that jointly trains an EHR rewriter and a predictor in an end-to-end manner. ReToP introduces a clinically driven feature selection strategy to generate synthetic pseudo-labels and innovatively designs a Classifier Supervision Contribution (CSC) scoring mechanism to align the rewriting process with the prediction objective—without requiring ground-truth rewritten data. Evaluated on three clinical prediction tasks using the MIMIC-IV dataset, ReToP significantly outperforms strong baselines, demonstrating superior generalization, rewriting fidelity, and focus on task-relevant features.

Technology Category

• Machine Learning: Feature Construction/Reformulation
• Humans and AI: Human-Aware Planning and Behavior Prediction
• Natural Language Processing: Code Generation / Program Synthesis from Natural Language

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Economics, Online Markets and Human Computation: Trust and reliance of crowd workers and data experts on GenAI
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking

Abstract

Electronic Health Records (EHRs) provide crucial information for clinical decision-making. However, their high-dimensionality, heterogeneity, and sparsity make clinical prediction challenging. Large Language Models (LLMs) allowed progress towards addressing this challenge by leveraging parametric medical knowledge to enhance EHR data for clinical prediction tasks. Despite the significant achievements made so far, most of the existing approaches are fundamentally task-agnostic in the sense that they deploy LLMs as EHR encoders or EHR completion modules without fully integrating signals from the prediction tasks. This naturally hinders task performance accuracy. In this work, we propose Rewrite-To-Predict (ReToP), an LLM-based framework that addresses this limitation through an end-to-end training of an EHR rewriter and a clinical predictor. To cope with the lack of EHR rewrite training data, we generate synthetic pseudo-labels using clinical-driven feature selection strategies to create diverse patient rewrites for fine-tuning the EHR rewriter. ReToP aligns the rewriter with prediction objectives using a novel Classifier Supervised Contribution (CSC) score that enables the EHR rewriter to generate clinically relevant rewrites that directly enhance prediction. Our ReToP framework surpasses strong baseline models across three clinical tasks on MIMIC-IV. Moreover, the analysis of ReToP shows its generalizability to unseen datasets and tasks with minimal fine-tuning while preserving faithful rewrites and emphasizing task-relevant predictive features.

Problem

• Electronic Health Records
• Clinical Prediction
• Large Language Models
• Task-Agnostic
• High-Dimensionality

Innovation

• Rewrite-To-Predict
• LLM-based EHR rewriting
• Classifier Supervised Contribution
• synthetic pseudo-labels
• task-aware clinical prediction