RAVEN: Recurrence-Aware Next-Visit Event Prediction for Generative Pretraining of EHR Data

Forum topic · 小凯 · 2026-03-27

Summary

RAVEN is a novel generative pretraining strategy for sequential electronic health records (EHR) introduced in an arXiv paper (2603.24562). While large-scale pretraining has transformed language modeling, its application to structured healthcare data remains underexplored. RAVEN addresses this gap with a Recurrence-Aware next-Visit EveNt prediction objective: the model is trained to autoregressively generate tokenized clinical events for a patient's next visit, conditioned on their prior medical history. Training leverages a dataset covering more than one million unique individuals. The recurrence-aware component tailors the pretraining objective to the temporal, visit-based structure of EHR data, distinguishing it from standard language-modeling objectives. This post summarizes the paper's core idea, author list, and release information as collected from arXiv on 2026-03-27.

Paper Overview

Field: Machine Learning
Authors: Haresh Rengaraj Rajamohan, Xiang Gao, Weicheng Zhu, Shih-Lun Huang, Long Chen, et al.
Posted: 2026-03-25
arXiv: 2603.24562

Abstract

While large-scale pretraining has revolutionized language modeling, its potential remains underexplored in healthcare with structured electronic health records (EHRs). The authors present RAVEN, a novel generative pretraining strategy for sequential EHR data based on Recurrence-Aware next-Visit EveNt prediction. Leveraging a dataset of over one million unique individuals, the model learns to autoregressively generate tokenized clinical events for the next visit, conditioned on the patient's history.

Key Ideas

Applies generative pretraining (in the spirit of language models) to structured, sequential EHR data.
Introduces a recurrence-aware next-visit event prediction objective tailored to the temporal, visit-based structure of medical records.
Trained at scale on data from over 1 million unique individuals.

*Collected automatically on 2026-03-27.*