Natural Language Processing Improves Postoperative Discharge Instructions.

Abstract

Purpose: To evaluate the readability, understandability, and actionability of current orthopaedic discharge instructions and assess the effectiveness of generative artificial intelligence (AI) in addressing these areas of improvement.

Methods: Orthopaedic discharge instruction documents were collected from a large academic hospital from March to May 2025. Subcategories frequently associated with postoperative inquiries-activities, complications and follow-up, medications, narcotics, pain management, physical therapy, and wound care-were extracted for analysis. ChatGPT-4 was used to generate improved discharge instructions using structured prompting strategies. Readability and text complexity measures were assessed using 5 validated readability tests whereas understandability and actionability were assessed using a modified Patient Education Materials Assessment Tool (PEMAT). Paired t-tests were used to compare scores between original and ChatGPT-rewritten content whereas comparisons across subcategories were achieved with independent samples t-tests (P < .05).

Results: Fifty-two subcategory excerpts were identified across twelve distinct discharge instructions. ChatGPT significantly reduced the average reading grade level of discharge instructions from 9.8 to 7.5 across all content categories (P < .001). The wound care subcategory showed the lowest reading grade level before and after ChatGPT. The narcotics subcategory showed the highest reading grade level before ChatGPT, whereas the medications and physical therapy subcategories showed the highest reading grade level after ChatGPT. Understandability and actionability scores improved from 63.6% to 87.9% (P < .001) and from 44.4% to 73.6%, respectively (P < .001).

Conclusions: Generative AI significantly improves the readability, understandability, and actionability of orthopaedic discharge instructions; however, not all documents achieved the desired sixth-grade reading level. Subcategory analysis revealed opportunities for targeted interventions to improve patient education in specific topics, such as narcotics, medication, or physical therapy-related instructions.

Clinical relevance: This study describes a cost-effective, scalable AI intervention for improving orthopaedic discharge instructions which may improve clinical outcomes, reduce healthcare burden, and support patients' management of postoperative recovery.