Prediabetes and stroke risk in U.S. adults: a machine learning-based analysis with causal mediation insights.
Level 3 - non-randomized controlled study
Cross-sectional analysis of national survey data with multivariable and machine learning modeling
PubMed 41731628 · doi:10.1186/s13098-026-02128-w
What was done
This cross-sectional study analyzed data from 14,500 adults participating in the National Health and Nutrition Examination Survey (NHANES) from 2005 to 2018. Prediabetes was defined as fasting plasma glucose of 100–125 mg/dL and/or HbA1c of 5.7–6.4% in participants without diabetes. The association with self-reported stroke was evaluated using multivariable logistic regression and propensity score matching. Mediation analyses assessed the roles of hypertension and body mass index (BMI), and six machine learning algorithms were trained and validated for stroke risk prediction.
What was found
Among 14,500 participants, 3,318 had prediabetes. Prediabetes was independently associated with prevalent stroke (OR 1.30, 95% CI: 1.06–1.60). In feature importance analyses across 40 predictors, prediabetes ranked sixth (composite score 1.74), ahead of smoking and physical activity. Hypertension and BMI jointly mediated 35–45% of the prediabetes-stroke association. An ensemble machine learning model achieved an AUC of 0.848 (95% CI: 0.833–0.863) versus 0.776 (95% CI: 0.758–0.794) for logistic regression, identifying a top 1% risk subgroup with a 27% stroke prevalence.
Why it matters
These findings suggest that prediabetes is independently linked to stroke prevalence before progression to diabetes, with a substantial portion of the risk operating through hypertension and elevated BMI. Early metabolic management addressing weight and blood pressure in prediabetic individuals may aid stroke prevention.
Limits
The study is cross-sectional, precluding temporal or causal inferences regarding stroke onset. Stroke occurrence was based entirely on participant self-report without medical record or neuroimaging verification. Stroke subtypes (ischemic versus hemorrhagic) were not reported or distinguished in the abstract.
Cited by
- contradicts Approximately 115 million adults in the United States have pre-diabetes.