Quality and safety of AI-generated exercise advice for patients with hypertension and comorbid chronic diseases: An expert evaluation study
Binfeng Yuan, Hao Ren, Manman ShiBackground
Generative artificial intelligence may support chronic disease management by providing accessible and individualized health information, but its quality and safety in exercise guidance for patients with hypertension and comorbid chronic diseases remain unclear.
Objective
This study aimed to evaluate AI-generated exercise advice for patients with hypertension and comorbid chronic diseases from both patient and physician perspectives.
Methods
Twenty patients with hypertension and at least one comorbid chronic disease were included. De-identified physical examination reports and self-reported exercise-related information were entered into ChatGPT to generate individualized exercise advice. Patients evaluated understandability, actionability, personal relevance, and acceptability. Five physicians independently evaluated clinical accuracy, safety, individualization, prescription completeness, and clinical usefulness using a five-point Likert scale.
Results
Patient evaluations were generally favorable, with overall agreement of 85.0% for understandability, 80.0% for acceptability, 75.0% for actionability, and 70.0% for personal relevance. Physician evaluations were also positive, with the highest ratings for clinical accuracy (4.04 ± 0.80; 82.0% agreement) and safety (3.95 ± 0.82; 78.0% agreement). Individualization (3.75 ± 0.89; 68.0%) and prescription completeness (3.67 ± 0.91; 64.0%) received relatively lower ratings.
Conclusion
Patients reported favorable perceived usability, and physicians generally rated the exercise advice favorably. These ratings do not establish clinical safety or effectiveness. AI-generated advice may support exercise-related health education when reviewed by a clinician, but patient evaluation bias, limited model reproducibility, and the absence of case-level explanations for safety concerns restrict interpretation.