2025 ISAKOS Biennial Congress ePoster
Enhancing Patient Education On The Role Of Tibial Osteotomy In The Management Of Knee Osteoarthritis Using A Customized Chatgpt: A Readability And Quality Assessment
Stephen Fahy, Dr, BSc, BmBs, MSC,MRCSI, Berlin GERMANY
Stephan Oehme, MD, Berlin, Berlin GERMANY
Danko Dan Milinkovic, MD, Berlin GERMANY
Benjamin Bartek, MD, Berlin GERMANY
Charité – Universitätsmedizin , Berlin, Berlin, GERMANY
FDA Status Not Applicable
Summary
We assessed whether a customised ChatGPT show superiority over general ChatGPT with respect to Patient Education on Knee Osteoarthritis and Tibial Osteotomy.
ePosters will be available shortly before Congress
Abstract
Introduction
Osteoarthritis (OA) of the knee affects 40% of men and 47% of women, significantly
impairing physical and mental health. Initially, non-surgical treatments like
physiotherapy and joint injections are used, but they often become insufficient as the
disease progresses, leading to surgical options such as Total Knee Arthroplasty (TKA),
Unicompartmental Knee Arthroplasty (UKA), and High Tibial Osteotomy (HTO). HTO is
particularly beneficial for younger patients, offering superior postoperative range of
motion. Effective patient education on these options is crucial for informed decisionmaking
and satisfaction.
Materials And Methods
The study generated questions about HTO using Google search's "People Also Asked"
feature, formatted for an 8th-grade reading level. Two ChatGPT-4 models were
assessed: a native model and a fine-tuned version named "The Knee Guide." Finetuning
focused on readability and source citation through Instruction-Based Fine-
Tuning and Reinforcement Learning from Human Feedback (RLHF). Responses to 25
questions were analyzed for quality using DISCERN criteria and readability using
Flesch Reading Ease Score (FRES) and Flesch-Kincaid Grade Level (FKGL).
Statistical analysis was performed using IBM SPSS Statistics.
Results
The mean DISCERN score for the native ChatGPT-4 was 38.41 (poor quality), while
"The Knee Guide" scored 45.95 (moderate quality). "The Knee Guide" outperformed
the native model in clarity, relevance, source citation, and discussion of treatment benefits. The readability of "The Knee Guide" was significantly better, with a mean
FKGL of 8.2 and a mean FRES of 60, compared to FKGL of 13.99 and FRES of 32 for
the native model.
Conclusions
A fine-tuned ChatGPT model provided higher quality, more readable patient education
on HTO compared to the native ChatGPT-4. This improvement was achieved with
minimal expertise and time, suggesting that customized AI tools can enhance patient
education by making complex medical information accessible and understandable to a
broader audience.