LASSO-based machine learning algorithm to predict the incidence of diabetes in different stages

Area of Science:

• Endocrinology
• Preventive Medicine
• Biostatistics

Background:

• Formal risk assessment is vital for effective diabetes prevention strategies.
• Prediabetes poses a significant risk for progression to type 2 diabetes.
• Developing practical predictive tools is essential for early identification and intervention.

Purpose of the Study:

• To establish a practical nomogram for predicting prediabetes incidence.
• To develop a nomogram for predicting prediabetes conversion to diabetes.
• To create early warning models for high-risk populations.

Main Methods:

• A cohort of 1428 subjects was analyzed.
• Least Absolute Shrinkage and Selection Operator (LASSO) regression was used for risk factor screening and model development.
• Multivariate logistic regression and nomograms were employed for prediction.
• Model performance was assessed using receiver-operating characteristic (ROC) curves and calibration analysis.

Main Results:

• The LASSO algorithm outperformed other methods in diabetes risk prediction.
• Key predictors for prediabetes included Age, Family History (FH), Insulin_F, hypertension, Tgab, HDL-C, Proinsulin_F, and TG.
• Predictors for prediabetes to diabetes conversion were Age, FH, Proinsulin_E, and HDL-C.
• The predictive models demonstrated good discrimination (AUCs of 0.78 and 0.70) and calibration.

Conclusions:

• Novel early warning models for prediabetes and diabetes were successfully established.
• These nomograms provide a practical approach for identifying at-risk individuals.
• The models facilitate timely intervention to prevent diabetes progression.