Investigation of ChatGPT-Generated Responses Regarding to Exercise-Based Cardiac Rehabilitation in Individuals with Heart Failure

Objective: The aim of this study was to investigate the potential of ChatGPT in designing an exercise-based cardiac rehabilitation (CR) program for heart failure (HF). Materials and Methods: Two physiotherapists specializing in cardiopulmonary physiotherapy developed 25 questions regarding exercise-based cardiac rehabilitation in HF. ChatGPT-generated responses were evaluated by two cardiopulmonary physiotherapists by comparing them with current clinical guidelines and professional expertise. The relevance, accuracy, clarity, completeness, and consistency criteria for each query were evaluated using a 5-point Likert-type scale. Each criterion's inter-rater agreement was tested using Cohen's Kappa and intraclass correlation coefficient (ICC) analysis. Results: The two highest means among the evaluation criteria belonged to clarity (mean=4.74, κ =0.78, ICC= 0.74)) and relevance (mean=4.68, κ =0.63, ICC=0.77). The lowest mean and highest consensus belonged to the consistency criteria (mean=3.58, κ =0.93, ICC=0.98). The means for accuracy and completeness were 4.14 and 4.28, respectively. Both criteria had moderate inter-rater agreement (κ =0.40-0.60) and good level ICC (0.75-0.90). Conclusion: ChatGPT directly addressed questions regarding exercise-based CR for HF patients, providing clear replies. However, the responses need to be improved in terms of accuracy, completeness, and, most importantly, consistency. ChatGPT has the potential to support the physiotherapist in the clinical practice. However, as it currently stands, directly using an exercise training program developed by ChatGPT for heart failure patients does not appear to be a reliable approach