Artificial intelligence in cancer survivorship: evaluating ChatGPT's performance in rehabilitation-related queries

This study aimed to evaluate the performance of ChatGPT in responding to patient-centered queries related to cancer rehabilitation. It examined the accuracy, completeness, and overall quality of ChatGPT's responses across major domains of cancer rehabilitation from a physical therapy perspective.

Fifteen representative patient queries covering eight domains of cancer rehabilitation-including general exercise, fatigue, bone metastasis, neuropathy, balance and falls, breast cancer, women's health, and pediatric cancer-were developed by two licensed physical therapists. Each query was input into ChatGPT-4, and responses were independently evaluated for accuracy, completeness, and global quality based on established clinical practice guidelines and expert consensus.

ChatGPT demonstrated a mean accuracy score of 3.93 (SD = 1.10) on a 5-point scale, a mean completeness score of 2.13 (SD = 0.83) on a 3-point scale, and a mean global quality score of 3.60 (SD = 1.12) on a 5-point scale. Responses were generally accurate and clinically relevant but often lacked details regarding exercise dosage, safety precautions, and guideline-concordant recommendations.

ChatGPT shows potential as an accessible tool for general cancer rehabilitation education but requires refinement to ensure accuracy, contextual relevance, and safety.

AI-driven conversational models may improve patient access to rehabilitation information and self-management support when used under professional oversight and guided by evidence-based frameworks.