The RAISE machine-learning models strongly correlated with clamp-measured insulin sensitivity in black (r=0.81) and white (r=0.85) youth (p<0.001), outperforming surrogate indices.
Observational (n=1,057)
Yes
Does the RAISE machine-learning model improve the estimation of insulin sensitivity and prediction of metabolic syndrome in black and white youth with obesity compared to surrogate indices?
The RAISE machine-learning model offers a practical, race-specific tool for estimating insulin sensitivity and predicting metabolic syndrome in youth with obesity without requiring fasting glucose and insulin measurements.
Effect estimate: r=0.81 (Black) and r=0.85 (White)
p-value: p=<0.001
Introduction and Objective: In vivo measures of insulin sensitivity (IS) are impractical for clinical/epidemiological studies. Thus, various mathematical IS models are suggested. Yet, such models ignore racial/ethnic disparities in IS in youth. Here, we used machine learning (LASSO age 14.2±1.9 yrs; 131 B/109 W; BMI 33.9±6.1 kg/m2). Predictive parameters included physical and fasting metabolic parameters (Image). The best-performing models were externally tested by comparinig predictive powers of IS indices for metabolic syndrome (NCEP-ATP III) from the NHANES dataset (n=817; age 15.1±0.2 yrs; 441 B/376 W; BMI 32.3±0.4 kg/m2). Results: LASSO was the best-performing model in B youth, while XGBoost was in W youth with different parameters (Image). Internal validation against hyperinsulinemic-euglycemic clamp IS showed a strong correlation for RAISE in B (r=0.81) and W youth (r=0.85, all p0.001). RAISE models achieved the highest predictive power (ROC-AUC) for metabolic syndrome in both races outperforming surrogate indices of IS (Image). Conclusion: Our data suggest that RAISE, by bypassing fasting glucose and insulin measurements, offers a cost-effective and practical tool for assessing IS in diverse pediatric populations across clinical and epidemiological settings. Disclosure W. Cho: None. J. Kim: None. S. Arslanian: None.
CHO et al. (Fri,) conducted a observational in Obesity in youth (n=1,057). RAISE machine-learning models vs. Surrogate indices of insulin sensitivity was evaluated on Correlation with hyperinsulinemic-euglycemic clamp insulin sensitivity (r=0.81 (Black) and r=0.85 (White), p=<0.001). The RAISE machine-learning models strongly correlated with clamp-measured insulin sensitivity in black (r=0.81) and white (r=0.85) youth (p<0.001), outperforming surrogate indices.