An XGBoost-based model for detecting undiagnosed type 2 diabetes using routine physical and lifestyle data from a multi-center Chinese population.

Introduction: This study aimed to develop and validate an interpretable machine learning model to identify individuals with undiagnosed type 2 diabetes (T2D) using data readily available from routine health checkups. Methods: In this retrospective study, we analyzed data from 12 tertiary hospitals in China. Following the application of inclusion and exclusion criteria, data from 11,382 individuals formed the training set for developing an XGBoost model, which was optimized using 5-fold cross-validation. An independent test set of 1,026 individuals from the same multi-center data source was used for internal validation. Model performance was primarily assessed using the area under the receiver operating characteristic curve (AUC). Results: The final model incorporated 12 predictors. Fasting blood glucose was the most influential predictor (50.6%), followed by creatinine (6.6%), triglyceride (5.6%), age (5.1%), and low-density lipoprotein (5.0%). On the independent test set, the model achieved an AUC of 77.2% (95%CI: 70.3%-84.1%). Conclusion: The XGBoost model demonstrated moderate predictive performance for T2D risk using routine checkup data. This approach shows potential for integration into clinical practice as an assistive screening tool, enabling automated risk profiling during standard health examinations. By flagging high-risk individuals, it can support clinicians in decision-making regarding further diagnostic testing. Future work should focus on external validation and prospective implementation studies.