Improving Medical Communication using Rubric-Guided Counterfactual Recommendations

2026-06-17

AI Summary

This study addresses the limitation of existing lightweight patient feedback, which often emphasizes subjective communication experiences and offers limited actionable guidance for physicians. To bridge this gap, the authors propose a language model–based counterfactual recommendation framework that enhances interpretable communication attributes—such as tone, personalization, actionability, and completeness—through minimal, rule-guided edits while preserving the original medical content. This approach ensures clinicians retain full control over clinical reasoning and phrasing. By integrating counterfactual reasoning, ordinal feature search, and an independent auditing model, the framework achieves generalizable improvements in communication quality. Experimental results demonstrate that the recommendations increase the predicted probability of positive patient feedback by an average of 6.41%, with non-negative effects observed in 93.31% of cases.

Abstract

Text-based telemedicine increasingly relies on lightweight patient feedback, however, such feedback primarily reflects perceived communication quality rather than medical accuracy. We introduce an LM-guided counterfactual recommendation pipeline that discovers and refines interpretable communication features such as tone, personalization, actionability and completeness in addressing patient concerns, without interfering with the medical content. These features are used together with patient-doctor interaction metadata to estimate positive feedback. At inference time, the system searches over low-cost ordinal feature changes and recommends minimal communication changes predicted to increase the probability of positive feedback, while independent auditor models test whether these gains generalize beyond the selection model. Across interactions, recommendations yield a mean +6.41% gain in predicted positive feedback probability under independent auditors, and are non-negative for 93.31% of recommendations. These results suggest that small, interpretable communication changes can capture most predicted gains while preserving the doctor's control over medical reasoning and final wording.

Problem

• telemedicine
• medical communication
• patient feedback
• communication quality
• counterfactual recommendations

Innovation

• counterfactual recommendation
• rubric-guided
• interpretable communication features
• LM-guided pipeline
• medical communication improvement