LLM-Guided Rehabilitation in Degenerative Knee Disease

This randomized, four-arm clinical trial will investigate the effects of large language model (LLM)-assisted exercise prescriptions integrated with conventional physiotherapy on pain, function and quality of life in adults with degenerative knee disease. Participants will be randomly assigned to conventional physiotherapy alone or to conventional physiotherapy plus an exercise program planned with the assistance of ChatGPT-5, Gemini 2.5 Pro, or DeepSeek V3.1. All participants will receive supervised outpatient physiotherapy two times per week for 8 weeks (16 sessions in total). Outcomes will be assessed at baseline and at the end of the 8-week intervention period by physiotherapists blinded to group allocation.

Degenerative knee disease, including knee osteoarthritis, is a common cause of pain, disability and loss of independence in middle-aged and older adults. Exercise-based rehabilitation is a core component of evidence-based management, but the content and progression of exercise programs often vary between clinicians. Large language models (LLMs) such as ChatGPT-5, DeepSeek V3.1 and Gemini 2.5 Pro can generate exercise suggestions based on clinical information and may support physiotherapists in designing and progressing individualized programs. However, their impact on clinical outcomes in degenerative knee disease has not been evaluated in randomized controlled trials. This single-center, randomized, parallel-group clinical trial will be conducted in adults with degenerative knee disease referred to the physiotherapy clinic of Kirsehir Ahi Evran University. After written informed consent and baseline assessments, participants will be randomly assigned (1:1:1:1) to one of four groups. All groups will receive a standardized outpatient physiotherapy program including hot pack, conventional TENS, therapeutic ultrasound and strengthening, range of motion, proprioceptive and functional exercises for the knee joint. In the three experimental groups, the content and progression of the exercise program will be planned with the assistance of standardized prompts to ChatGPT-5, DeepSeek V3.1 or Gemini 2.5 Pro, and all LLM-generated suggestions will be reviewed for safety and appropriateness by physiotherapists before being implemented. Sessions will be delivered two times per week for 8 weeks (16 supervised visits) under the supervision of a physical medicine and rehabilitation specialist and physiotherapists. Outcomes will be evaluated at baseline (week 0) and after the 8-week intervention by a physiotherapist blinded to group allocation. Main outcome domains include knee pain, knee-related patient-reported outcomes, objective functional performance tests, isometric muscle strength, knee range of motion, psychosocial factors and health-related quality of life. Data will be analyzed according to CONSORT guidelines for randomized trials.