MedEinst: Benchmarking the Einstellung Effect in Medical LLMs through Counterfactual Differential Diagnosis

TL;DR

MedEinst introduces a counterfactual benchmark to evaluate how medical LLMs rely on shortcuts instead of evidence, revealing their susceptibility to the Einstellung Effect in atypical diagnoses.

Contribution

This work presents MedEinst, a novel benchmark with paired cases to detect bias traps, and proposes ECR-Agent, a reasoning system aligned with Evidence-Based Medicine standards.

Findings

Frontier models have high accuracy but high bias trap rates.

Existing benchmarks fail to detect the Einstellung Effect.

ECR-Agent improves reasoning fidelity in medical diagnosis.

Abstract

Despite achieving high accuracy on medical benchmarks, LLMs exhibit the Einstellung Effect in clinical diagnosis--relying on statistical shortcuts rather than patient-specific evidence, causing misdiagnosis in atypical cases. Existing benchmarks fail to detect this critical failure mode. We introduce MedEinst, a counterfactual benchmark with 5,383 paired clinical cases across 49 diseases. Each pair contains a control case and a "trap" case with altered discriminative evidence that flips the diagnosis. We measure susceptibility via Bias Trap Rate--probability of misdiagnosing traps despite correctly diagnosing controls. Extensive Evaluation of 17 LLMs shows frontier models achieve high baseline accuracy but severe bias trap rates. Thus, we propose ECR-Agent, aligning LLM reasoning with Evidence-Based Medicine standard via two components: (1) Dynamic Causal Inference (DCI) performs structured reasoning through dual-pathway perception, dynamic causal graph reasoning across three levels (association, intervention, counterfactual), and evidence audit for final diagnosis; (2) Critic-Driven Graph and Memory Evolution (CGME) iteratively refines the system by storing validated reasoning paths in an exemplar base and consolidating disease-specific knowledge into evolving illness graphs. Source code is to be released.