Key result
Routine blood biomarker model predicts ~306% higher 5-year incident prediabetes risk in high-risk individuals.
Why the study?
Type 2 diabetes often develops silently over years, making early risk identification critical for prevention.
Does a Random Survival Forest model using routine blood biomarkers predict incident prediabetes over 5 years in adults without diabetes?
Cohort (n=80,000)
Does a Random Survival Forest model using routine blood biomarkers predict incident prediabetes over 5 years in adults without diabetes?
Hazard Ratio: 4.06 (95% CI 3.47–4.75)
A machine learning model utilizing routine blood biomarkers can accurately predict the 5-year risk of developing prediabetes in adults without diabetes, enabling early intervention.
Introduction and Objective: Type 2 diabetes (T2D) often develops silently over years, making early risk identification critical for prevention. This study aimed to determine whether routine blood biomarkers can predict 5-year T2D risk in adults without diabetes. Methods: We analyzed de-identified blood diagnostic data from 80,000 adults without diabetes, defined as HbA1c <6.0% or fasting glucose < 6.1 mmol/L in Ontario, Canada (2014-2025). Only blood biomarker data were used, therefore this population could also include individuals with well-controlled diabetes. Participants were aged ≥18 years and had all required biomarkers measured in the same visit: HbA1c, fasting glucose, lipid profile, creatinine, and CBC. Age and sex were included as covariates and individuals also had to have sufficient data to calculate at least one of these insulin resistance indexes: HOMA-IR, TG/HDL-C ratio, Triglyceride/Glucose (TyG) index. A Random Survival Forest (RSF) model was developed to predict time to incident prediabetes (HbA1c ≥6.0% or 2-hour OGTT ≥7.8 mmol/L) over 5 years. Results: In the validation cohort, the RSF model demonstrated strong discrimination (C-index: 0.84 [95% CI 0.83-0.84]; 5-year time-dependent AUC: 0.87 [95% CI 0.87-0.88]) and calibration (5-year Integrated Brier Score: 0.050 [95% CI 0.049-0.051]). After adjusting for age, sex and HbA1C, stratification by predicted risk quantiles showed effective separation, with significantly elevated progression risk in high (HR 4.06 [95% CI 3.47-4.75]) and medium (HR 2.31 [95% CI 2.01-2.67]) versus low quantile. Conclusion: Routine blood biomarkers can predict future T2D risk in adults without diabetes. The RSF approach demonstrated strong predictive performance highlighting its potential for integration into routine screening to enable early intervention. Disclosure S. Tang: Employee; Current; LifeLabs. F. Cai: Employee; Current; LifeLabs. D. Nyholt: Employee; Current; LifeLabs.
No takes yet. Share an insight, caveat, or question.
Tang et al. (2026) conducted a cohort in Type 2 diabetes risk (n=80,000). High predicted risk quantile (Random Survival Forest model) vs. Low predicted risk quantile was evaluated on Time to incident prediabetes (HbA1c ≥6.0% or 2-hour OGTT ≥7.8 mmol/L) over 5 years (HR 4.06, 95% CI 3.47-4.75). A Random Survival Forest model using routine blood biomarkers strongly predicted 5-year incident prediabetes, with high-risk individuals showing significantly elevated risk versus low-risk (HR 4.06; 95% CI 3.47-4.75).
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: