Risk warning model for predicting sleep disorders in healthcare workers on long-term shifts.

Long-term shift work significantly impacts the health of healthcare workers, with sleep disorders (SD) being a common and urgent concern. Traditional predictive models often perform poorly in identifying minority class samples-specifically healthcare workers experiencing SD-due to dataset imbalances. This study aimed to construct a risk warning model by introducing the synthetic minority over-sampling technique (SMOTE) to improve the predictive accuracy for SD among healthcare workers engaged in long-term shift work, providing a scientific basis for early intervention. A retrospective analysis was conducted on the sleep conditions of 181 healthcare workers at CR&WISCO General Hospital, Wuhan University of Science and Technology. Participants were divided into two groups based on their sleep status: 70 individuals in group A (AG) with SD, and 111 individuals in group B (BG) without SD. The application of the SMOTE-based risk warning model was analyzed for predicting SD in healthcare workers under long-term shift work, and the model's performance was validated against two other models and three verification datasets. Multivariate logistic regression analysis of SD among healthcare workers under long-term shift work revealed that gender, age, occupation, education level, professional title, authorization strength, shift duration, work hours, anxiety, and depression were identified as independent influencing factors. The SMOTE warning model achieved a sensitivity of 83.22%, specificity of 78.67%, accuracy of 85.35%, positive predictive value (PPV) of 74.60%, and negative predictive value (NPV) of 87.67%, significantly outperforming the original dataset, backpropagation (BP) model, and the random forest (RF) model ( Supplementary Information: The online version contains supplementary material available at 10.1007/s41105-025-00583-y.