Trajectory guided steering reduces bias in clinical ai predictions
Trajectory-Guided Steering: Mitigating Spurious Feature Bias in Clinical Classification
Computation and Language
Summary
Clinical AI models sometimes make mistakes because they rely on unrelated details like a patient's gender when diagnosing. The authors found that by watching how a model’s guesses change inside it, they can tell if correcting these mistakes will help. Their method, called Trajectory-Guided Steering, decides whether and how to fix each prediction without needing special data about the bias. This method improves accuracy while reducing bias related to gender in medical diagnoses.
What this means in practice
- •For hospital data teams: Reduce gender-related bias in AI-assisted clinical diagnosis without needing specially crafted correction data during prediction.
- •For biomedical software developers: Improve clinical AI tools by automatically deciding when and how to adjust predictions to avoid errors caused by spurious clinical note details.
Authors
Fardin Ahsan Sakib, Ziwei Zhu, Ozlem Uzuner
Abstract
Large language models rely on spurious features in clinical classification: swapping a single gender token in clinical notes alters gender-irrelevant diagnoses in up to 53.5% of cases. Existing inference-time corrections either need counterfactual data built around the specific spurious feature or behave unpredictably. We observe that how a model's preference between answers evolves across layers, its prediction trajectory, indicates for each document whether a correction is likely to help. We propose Trajectory-Guided Steering (TGS), which decides from the trajectory alone whether to correct each prediction and in which direction, with no counterfactual data and no knowledge of the spurious feature. On 16 clinical classification tasks from i2b2, TGS fixes 17.7% of errors while breaking only 1.3% of correct predictions and stays net-positive across all tested hyperparameter settings; all baselines are net-negative or near zero. The same corrections reduce reliance on the spurious feature: on the most affected diseases, gender-swap inconsistency falls sharply (hypertension 53.5% -> 38.0%, diabetes 35.1% -> 21.5%). Applying the correction to every document instead raises breaks on correct positive predictions from 0.1% to 6.4%, so per-document selection is essential. TGS also transfers to three model families and three other clinical and biomedical datasets.