Machine learning screening of risk factors for diabetic microvascular complications and construction of a gradient boosting decision tree predictive model.
To develop a machine learning-based classification model to aid in the early diagnosis of diabetic microvascular complications.
This study analyzed clinical and laboratory data from 1,498 patients, categorized into two groups: diabetes alone and diabetes with microvascular complications. Independent risk factors for complications were identified through intergroup comparison, collinearity analysis, and logistic regression. Nine machine learning models were subsequently developed and compared. A comprehensive evaluation of the binary classification performance of the Gradient Boosting Decision Tree (GBDT) model was performed.
Urea, fibrinogen (FIB), prothrombin time (PT), D-dimer (DD), creatine kinase MB isoenzyme (CKMB), lipoprotein(a) (Lpa), activated partial thromboplastin time (APTT), triglycerides (TG), and cholinesterase (CHE) were identified as independent risk factors for diabetic microvascular complications. Among the nine predictive models constructed, the GBDT model demonstrated superior performance across multiple metrics, including the area under the receiver operating characteristic curve (AUC) and sensitivity, indicating strong generalization ability on the validation set. Further evaluation confirmed its consistent and robust predictive performance across training, validation, and test datasets. Calibration curve analysis showed good agreement between predicted probabilities and actual outcomes. Decision curve analysis demonstrated the model's clinical utility, and the Kolmogorov-Smirnov (KS) curve indicated excellent discriminatory power.
The GBDT model, constructed based on the identified risk factors, exhibits outstanding predictive performance and promising application potential. It provides important theoretical support and a practical tool for the early identification and targeted intervention of diabetic microvascular complications.
This study analyzed clinical and laboratory data from 1,498 patients, categorized into two groups: diabetes alone and diabetes with microvascular complications. Independent risk factors for complications were identified through intergroup comparison, collinearity analysis, and logistic regression. Nine machine learning models were subsequently developed and compared. A comprehensive evaluation of the binary classification performance of the Gradient Boosting Decision Tree (GBDT) model was performed.
Urea, fibrinogen (FIB), prothrombin time (PT), D-dimer (DD), creatine kinase MB isoenzyme (CKMB), lipoprotein(a) (Lpa), activated partial thromboplastin time (APTT), triglycerides (TG), and cholinesterase (CHE) were identified as independent risk factors for diabetic microvascular complications. Among the nine predictive models constructed, the GBDT model demonstrated superior performance across multiple metrics, including the area under the receiver operating characteristic curve (AUC) and sensitivity, indicating strong generalization ability on the validation set. Further evaluation confirmed its consistent and robust predictive performance across training, validation, and test datasets. Calibration curve analysis showed good agreement between predicted probabilities and actual outcomes. Decision curve analysis demonstrated the model's clinical utility, and the Kolmogorov-Smirnov (KS) curve indicated excellent discriminatory power.
The GBDT model, constructed based on the identified risk factors, exhibits outstanding predictive performance and promising application potential. It provides important theoretical support and a practical tool for the early identification and targeted intervention of diabetic microvascular complications.