Predicting Insulin Resistance in a Pediatric Population With Obesity

Objectives: Insulin resistance (IR) affects children and adolescents with obesity and early diagnosis is crucial to prevent long-term consequences. Our aim was to identify predictors of IR and develop a multivariate model to accurately predict IR. Methods: We conducted a cross-sectional analysis of demographical, clinical, and biochemical data from a cohort of patients attending a specialized Paediatric Nutrition Unit in Portugal over a 20-year period. We developed multivariate regression models to predict IR. The participants were randomly divided into 2 groups: a model group for developing the predictive models and a validation group for cross-validation of the study. Results: Our study included 1423 participants, aged 3–17 years old, randomly divided in the model (n = 879) and validation groups (n = 544). The predictive models, including uniquely demographic and clinical variables, demonstrated good discriminative ability [area under the curve (AUC): 0.834–0.868; sensitivity: 77.0%–83.7%; specificity: 77.0%–78.7%] and high negative predictive values (88.9%–91.6%). While the diagnostic ability of adding fasting glucose or triglycerides/high density lipoprotein cholesterol index to the models based on clinical parameters did not show significant improvement, fasting insulin appeared to enhance the discriminative power of the model (AUC: 0.996). During the validation, the model considering demographic and clinical variables along with insulin showed excellent IR discrimination (AUC: 0.978) and maintained high negative predictive values (90%–96.3%) for all models. Conclusion: Models based on demographic and clinical variables can be advantageously used to identify children and adolescents at moderate/high risk of IR, who would benefit from fasting insulin evaluation.