Improving Medical Communication using Rubric-Guided Counterfactual Recommendations

The article introduces a language model-guided counterfactual recommendation pipeline designed to enhance medical communication in telemedicine by refining features such as tone, personalization, and completeness without altering medical content. The system achieves a mean increase of 6.41% in predicted positive feedback probability through low-cost ordinal feature modifications, demonstrating that small, interpretable changes can significantly improve patient feedback while maintaining the integrity of medical reasoning. This approach is relevant for practitioners as it provides a method to optimize communication strategies in healthcare settings, potentially improving patient engagement and satisfaction.