Association between atherogenic index of plasma and hypertension in children and adolescents based on LightGBM prediction model

The prevalence of hypertension in children and adolescents is on the rise, highlighting the need to identify effective biomarkers for risk assessment. The Atherogenic Index of Plasma (AIP), which reflects dyslipidemia, has demonstrated predictive value in adult cardiovascular diseases. However, its association with hypertension in children and adolescents remains unclear. A total of 28,844 children and adolescents from 18 prefecture-level cities in Henan Province, China, were enrolled between 2023 and 2024. After screening, 27991 participants were included in the final analysis. Blood pressure was measured on three non-consecutive days. AIP was calculated as log₁₀ (triglycerides / high-density lipoprotein cholesterol). Multivariate logistic regression, restricted cubic splines, and mediation effect analysis were employed to explore the association between AIP and hypertension. 11 types of machine learning models were constructed to evaluate the predictive value of AIP: first, the dataset was split into a training set and a testing set; variable selection was performed using the Boruta algorithm only within the training set to avoid information leakage from the testing set, and SHAP (SHapley Additive exPlanations) analysis was further conducted to interpret the role of each variable. The AIP level in the hypertensive group was significantly higher than that in the non-hypertensive group (P<0.001). After adjustment by multiple models, participants in the 4th quartile (Q4) of AIP had a 20% higher risk of hypertension compared with those in the 1st quartile (Q1) (OR=1.20, P=0.027). The association was stronger in males (Q4 HR=2.38) than in females (Q4 HR=1.76), and there was a non-linear association between AIP and hypertension (P for non-linearity=0.007). Waist-to-height ratio (WHtR) (mediation proportion: 51.7%) and uric acid (UA) (mediation proportion: 20.7%) were identified as key mediators. The LightGBM model exhibited the relatively optimal predictive performance (AUC=0.7376), and SHAP analysis confirmed that AIP had an independent predictive value for hypertension. Elevated AIP is significantly associated with an increased risk of hypertension in children and adolescents, with gender differences and non-linear characteristics observed. AIP may serve as a potential biomarker for hypertension risk assessment in this population.