Artificial intelligence predicts healthcare workers’ antibiotic use intentions from psychological and behavioral measures across multiple theories

This cross-sectional study of 1,135 healthcare workers in four Chinese public hospitals combined behavioural theory with explainable machine learning to identify psychological drivers of appropriate antibiotic use and address non-guideline-concordant prescribing. Integrating constructs from multiple behavioural frameworks with LASSO regression and interpretable ML models (including XGBoost, LightGBM, and CatBoost), the analysis showed that social support, cognitive processing, knowledge and skills, and health beliefs were the strongest predictors of clinicians’ intention to use antimicrobials appropriately, with notable nonlinear interactions between social support and cognitive processing. Ensemble models achieved very high predictive performance for medium and high intention groups (F1 > 0.94), though identifying low-intention prescribers remained more difficult. Overall, the study demonstrates that combining behavioural science with explainable AI can enable scalable identification of prescribing risk profiles and support the design of psychologically tailored, real-time antibiotic stewardship interventions to combat AMR.