MedReadCtrl: Personalizing medical text generation with readability-controlled instruction learning

2025-07-10

AI Summary

Medical generative AI often fails to simultaneously achieve personalization and readability, particularly for users with varying health literacy levels. Method: This paper introduces the first readability-controllable medical instruction-tuning framework, integrating human feedback–driven readability-level annotation, multi-task instruction tuning, and semantics-preserving text rewriting to enable controllable generation that balances complexity, clinical accuracy, and reliability. Contribution/Results: The framework demonstrates strong cross-task and cross-domain generalization, outperforming GPT-4 across nine medical benchmarks: readability instruction error rate decreases by 12.6% (1.39 vs. 1.59), clinical relevance (ROUGE-L) improves by 14.7 points, expert preference reaches 71.7%, and comprehension among low-literacy users is significantly enhanced.

Technology Category

• Natural Language Processing: Generation
• Humans and AI: AI for Accessibility
• Machine Learning: Large Multimodal Models (LMMs)

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Economics, Online Markets and Human Computation: Trust and reliance of crowd workers and data experts on GenAI
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking

Abstract

Generative AI has demonstrated strong potential in healthcare, from clinical decision support to patient-facing chatbots that improve outcomes. A critical challenge for deployment is effective human-AI communication, where content must be both personalized and understandable. We introduce MedReadCtrl, a readability-controlled instruction tuning framework that enables LLMs to adjust output complexity without compromising meaning. Evaluations of nine datasets and three tasks across medical and general domains show that MedReadCtrl achieves significantly lower readability instruction-following errors than GPT-4 (e.g., 1.39 vs. 1.59 on ReadMe, p<0.001) and delivers substantial gains on unseen clinical tasks (e.g., +14.7 ROUGE-L, +6.18 SARI on MTSamples). Experts consistently preferred MedReadCtrl (71.7% vs. 23.3%), especially at low literacy levels. These gains reflect MedReadCtrl's ability to restructure clinical content into accessible, readability-aligned language while preserving medical intent, offering a scalable solution to support patient education and expand equitable access to AI-enabled care.

Problem

• Personalizing medical text generation for diverse readability levels
• Ensuring AI-generated healthcare content is understandable and accurate
• Improving patient education with scalable readability-controlled language models

Innovation

• Readability-controlled instruction tuning framework
• Adjusts output complexity without compromising meaning
• Restructures clinical content into accessible language