Application of Artificial Intelligence in Biochemistry Education: Evaluating GPT Models for Teaching and Learning – A Multicenter Retrospective Study

Background: Biochemistry is a core subject in undergraduate medical training, yet many students find it difficult to understand and retain due to the abstract nature of biochemical pathways and molecular processes. Conventional lecture-based teaching methods may not sufficiently support conceptual learning for all students. In recent years, generative artificial intelligence tools, including Generative Pre-trained Transformer (GPT) models, have been increasingly used by students as supplementary learning aids. Objective: To examine the association between GPT-assisted learning and academic performance, concept retention, and learner satisfaction among undergraduate medical students. Methods: A multicentre retrospective observational study was conducted across medical colleges in Rajasthan, over a one-year period. A total of 150 undergraduate medical students were included and divided into two groups: GPT-assisted learners (n = 75) and conventional learners (n = 75). Academic performance was assessed using internal biochemistry examination scores and concept retention tests, while learner satisfaction was measured using a five-point Likert scale. Statistical analysis included independent t-tests, correlation analysis, multivariate logistic regression, and receiver operating characteristic (ROC) curve analysis, with statistical significance set at p < 0.05. Results: Students who used GPT-assisted learning achieved higher mean examination scores compared with those using conventional learning methods (68.9 ± 9.2 vs 61.4 ± 10.5; p < 0.001). Concept retention scores were also higher among GPT users (72.3 ± 8.6 vs 64.1 ± 9.3; p < 0.001). Learner satisfaction was significantly greater in the GPT-assisted group (4.2 ± 0.6 vs 3.4 ± 0.7; p < 0.001). Multivariate analysis identified GPT usage as an independent predictor of high academic performance (adjusted OR = 2.86; p = 0.003). The combined predictive model demonstrated good accuracy on ROC analysis (AUC = 0.82). Conclusion: The use of GPT-assisted learning was associated with improved academic performance, better concept retention, and higher satisfaction among undergraduate medical students studying biochemistry. When applied responsibly as a supplementary tool, GPT-based systems may offer meaningful support within medical education.