Large language models struggle to update clinical judgments reliably
Large language models exhibit unreliable updating of clinical judgment as patient evidence evolves
Artificial IntelligenceComputation and Language
Summary
Large language models (LLMs) can help with medical decisions, but this study found they often fail to adjust their opinions correctly when new patient information appears. Sometimes they overreact to worsening symptoms and are influenced heavily by previous beliefs, even when current evidence is clear. The researchers identified two specific ways these models can get confused and found it hard to fix with simple prompting. They propose a method to reduce unreliable updates, although it may limit how often the model changes its opinion.
What this means in practice
- •For clinical ai developers: Improve medical AI tools by identifying and mitigating bias in updating patient condition assessments over time.
- •For healthcare data engineers: Design workflows and systems that incorporate model reliability checks for evolving patient data in intensive care units.
Authors
Min Zeng, Rui Zhang
Abstract
Large language models (LLMs) are increasingly explored for clinical reasoning, but whether they appropriately revise judgments as patient evidence evolves remains unclear. We evaluated longitudinal belief updating using matched intensive-care trajectories from electronic health records. Across diverse LLMs, conditioning on a preceding judgment more often increased than reduced prediction error when estimates changed, replicated for a second endpoint. Controlled interventions revealed two failure modes. First, with preceding assessment fixed, models responded more strongly to worsening than matched improving respiratory evidence; this asymmetry persisted after headroom normalization at moderate and strong evidence levels. Second, with current evidence fixed, increasing prior risk from 10% to 90% shifted estimates by 26.2 percentage points, demonstrating causal influence of prior model beliefs. Prompting did not restore reliable updating. Evidence-Validated Longitudinal Update (EVLU) identified fewer, more reliable revisions, revealing a reliability-coverage trade-off. These findings establish longitudinal belief updating as a distinct dimension of LLM reliability.