Machine Learning Prediction of Prediabetes in a Young Male Chinese Cohort with 5.8-Year Follow-Up

Area of Science:

• Endocrinology
• Metabolic Health
• Preventive Medicine

Background:

• Prediabetes risk factors in young men are understudied.
• Early identification is crucial for intervention.
• Normal fasting plasma glucose (FPG) at baseline does not preclude future prediabetes risk.

Purpose of the Study:

• To evaluate machine learning (Mach-L) for prediabetes prediction in young men.
• To compare Mach-L performance against traditional multiple linear regression (MLR).
• To identify significant risk factors for incident prediabetes.

Main Methods:

• Enrolled 6247 young ethnic Chinese men with normal baseline FPG.
• Utilized various Mach-L algorithms (Random Forest, Gradient Boosting, Elastic Net).
• Developed two predictive models: one including baseline FPG, one excluding it.

Main Results:

• Mach-L models demonstrated superior accuracy over MLR across multiple error metrics.
• Key predictors in Model 1 (all variables) were FPG, body fat (BF), creatinine, thyroid stimulating hormone (TSH), white blood cell count (WBC), and age.
• In Model 2 (excluding FPG), important factors included BF, WBC, age, TSH, triglycerides (TG), and LDL-C.

Conclusions:

• Machine learning offers a more accurate approach to predicting prediabetes in young men.
• Baseline FPG is a primary predictor, but BF, WBC, age, and TSH are critical when FPG is excluded.
• These findings highlight modifiable and non-modifiable factors for early prediabetes intervention strategies.