Assessing the Capabilities of Artificial Intelligence (AI) Tools in Community Medicine: A Comparative Study of ChatGPT, Gemini, and Bing in Community-Based Clinico-Social Case Interpretation

Abstract

Background and objective Artificial intelligence (AI) is being increasingly integrated into healthcare, offering opportunities to enhance decision-making, education, and patient engagement. Community medicine often involves interpreting clinico-social case studies that combine clinical, social, and environmental dimensions. However, there is limited research evaluating how AI tools perform in analyzing such community-based clinico-social cases. This study aimed to address that gap in the literature. Methods A comparative cross-sectional study was conducted using 30 standardized clinico-social case studies covering communicable and non-communicable diseases, maternal and child health, adolescent health, and social pathology. Three conversational AI models -ChatGPT (OpenAI, San Francisco, CA), Microsoft Bing AI (Microsoft Corp., Redmond, WA), Gemini (Google, Mountain View, CA) - were provided with identical prompts to interpret cases. Their responses were assessed by a panel of five community medicine experts using a 100-point rubric across five domains: diagnosis, intervention, recognition of social determinants, ethical reasoning, and public health appropriateness. Descriptive statistics and Spearman's rank correlation were used for analysis. Results All three AI tools demonstrated near-ceiling performance. Gemini achieved the highest total score (97.00 ± 1.74), excelling in diagnosis and public health appropriateness. ChatGPT (96.00 ± 2.98) performed best in intervention suggestions, while Bing AI (95.70 ± 3.64) showed slightly lower but comparable scores. Correlation analysis revealed weak-to-moderate alignment, with limited statistically significant associations across domains. Conclusions ChatGPT, Gemini, and Bing exhibit broadly similar capabilities in interpreting clinico-social cases, with domain-specific strengths. Their complementary nature suggests that they can aid medical education and public health practice, but should supplement rather than replace human expertise to minimize bias and errors.

Ähnliche Arbeiten