Confidence interval estimation in a multipredictor spline quadratic regression model for modeling hba1c levels in diabetes mellitus cases

Abstract

Modeling HbA1c levels in patients with diabetes mellitus is commonly conducted using parametric regression; however, this approach is often inadequate in capturing nonlinear relationships with metabolic predictors. This study aims to develop a multipredictor quadratic spline regression model with confidence interval estimation to model HbA1c levels flexibly. The model incorporates five predictor variables: body weight, fasting blood glucose, HDL cholesterol, LDL cholesterol, and triglycerides, and is implemented using RStudio version 2024.12.0. The results clearly demonstrate nonlinear relationships between HbA1c levels and all predictors. LDL cholesterol shows the strongest influence. Threshold effects are observed for body weight and HDL cholesterol, while glucose and triglycerides exhibit moderate nonlinear patterns. The visualization of the fitted curves and confidence bands supports a more interpretable representation of the model. Overall, the quadratic spline regression with confidence intervals provides a flexible and informative framework for modeling HbA1c levels, particularly when nonlinear associations are present.