AI integration into undergraduate health education streams- a multicenter study in Sri Lanka

BACKGROUND: Despite rapid integration of AI in healthcare, formal AI education remains limited in health professional curricula globally, particularly in low-and middle-income countries (LMICs). This study provides the first large-scale, multi-disciplinary quantitative assessment of AI-related knowledge, attitudes, and practices (KAP) among undergraduate healthcare students at Sri Lankan state universities and identifies independent predictors of AI engagement. METHODS: A multi-centre descriptive cross-sectional survey was conducted between March and June 2025 across five randomly selected Sri Lankan state universities, adhering to STROBE reporting guidelines. The minimum required sample of 385 was calculated using Cochran's formula (Z = 1.96; p = 0.50; d = 0.05); 1,100 students were invited to achieve this target. A self-administered online questionnaire assessed KAP across five domains. Face and content validity were established through expert review and pre-testing with 15 participants outside the final sample. Two multivariable logistic regression models identified independent predictors of academic AI tool use and clinical AI exposure. RESULTS: Of 430 respondents (67.2% female; response rate 39.1%), AI awareness was near-universal (99.3%) and academic AI tool use was high (88.4%). Only 8.1% had received any formal AI training and only 21.6% reported clinical AI exposure. Curriculum integration was the strongest independent predictor of both academic AI use (AOR 2.58, 95% CI: 1.01-6.56, p = 0.047) and clinical AI exposure (AOR 5.16, 95% CI: 2.82-9.47, p < 0.001). Workshop participation independently predicted clinical AI exposure (AOR 2.71, 95% CI: 1.25-5.85, p = 0.011). The most frequently reported barriers were insufficient practical experience (55.8%), lack of guidance (55.8%), and limited resources (50.5%). CONCLUSION: Sri Lankan healthcare undergraduates demonstrate near-universal AI awareness but a pronounced competency gap characterised by limited formal training and minimal clinical AI engagement. Curriculum integration and structured experiential training were the strongest independent predictors of meaningful AI engagement. These findings support the urgent need for national curriculum reform and context-sensitive AI education frameworks for resource-constrained health professional training environments.