WISTERIA: Learning Clinical Representations from Noisy Supervision via Multi-View Consistency in Electronic Health Records

2026-05-10

AI Summary

This work addresses the challenge that clinical labels in electronic health records (EHRs) are often heterogeneous, noisy, and institution-dependent, yet conventional approaches treat them as deterministic ground truth, hindering the learning of robust and clinically meaningful representations. To overcome this limitation, the authors propose WISTERIA, a novel framework that explicitly models the weakly supervised label generation process in EHR representation learning. WISTERIA treats observed labels as stochastic observations of latent clinical states and leverages consistency regularization across multiple weak supervision operators to implicitly denoise labels. Furthermore, it incorporates medical ontology-aware constraints to preserve and enhance semantic structure. Experiments demonstrate that WISTERIA significantly improves predictive performance and robustness to label noise, outperforming existing sequence-based pretraining methods in cross-institutional generalization.

Abstract

Representation learning in electronic health records (EHR) has largely followed paradigms inherited from natural language processing, relying on sequence modeling and reconstruction based objectives that treat clinical labels as ground truth. However, real world clinical supervision is inherently weak, arising from heterogeneous, noisy, and institution specific labeling processes such as billing codes, heuristic phenotypes, and incomplete annotations. In this work, we propose WISTERIA, a weakly supervised representation learning framework that models labels as stochastic observations of an underlying latent clinical state. Instead of optimizing against a single supervision signal, WISTERIA constructs multiple weak supervision operators and learns representations by enforcing consistency across their induced label distributions. This multi view formulation induces an implicit denoising mechanism, allowing the model to recover clinically meaningful structure by reconciling disagreement between noisy labelers. We further incorporate ontology aware regularization in the label space to impose semantic structure over supervision signals. Empirically, WISTERIA improves predictive performance across standard EHR benchmarks, demonstrates strong robustness to label noise, and exhibits superior cross institutional generalization compared to sequence based pretraining objectives. These results suggest that explicitly modeling the supervision process rather than treating labels as fixed targets provides a more appropriate inductive bias for learning robust and clinically meaningful representations from EHR data.

Problem

• weak supervision
• electronic health records
• noisy labels
• clinical representation learning
• multi-view consistency

Innovation

• weak supervision
• multi-view consistency
• electronic health records
• ontology-aware regularization
• representation learning