Assessing the performance of large language models (GPT-3.5 and GPT-4) and accurate clinical information for pediatric nephrology

Background Artificial intelligence (AI) has emerged as a transformative tool in healthcare, offering significant advancements in providing accurate clinical information. However, the performance and applicability of AI models in specialized fields such as pediatric nephrology remain underexplored. This study is aimed at evaluating the ability of two AI-based language models, GPT-3.5 and GPT-4, to provide accurate and reliable clinical information in pediatric nephrology. The models were evaluated on four criteria: accuracy, scope, patient friendliness, and clinical applicability. Methods Forty pediatric nephrology specialists with >= 5 years of experience rated GPT-3.5 and GPT-4 responses to 10 clinical questions using a 1-5 scale via Google Forms. Ethical approval was obtained, and informed consent was secured from all participants. Results Both GPT-3.5 and GPT-4 demonstrated comparable performance across all criteria, with no statistically significant differences observed (p > 0.05). GPT-4 exhibited slightly higher mean scores in all parameters, but the differences were negligible (Cohen's d < 0.1 for all criteria). Reliability analysis revealed low internal consistency for both models (Cronbach's alpha ranged between 0.019 and 0.162). Correlation analysis indicated no significant relationship between participants' years of professional experience and their evaluations of GPT-3.5 (correlation coefficients ranged from - 0.026 to 0.074). Conclusions While GPT-3.5 and GPT-4 provided a foundational level of clinical information support, neither model exhibited superior performance in addressing the unique challenges of pediatric nephrology. The findings highlight the need for domain-specific training and integration of updated clinical guidelines to enhance the applicability and reliability of AI models in specialized fields. This study underscores the potential of AI in pediatric nephrology while emphasizing the importance of human oversight and the need for further refinements in AI applications.