Title: Total Cholesterol Modeling Based on Blood Glucose: A Local Linear Estimator Approach in Diabetes Mellitus Patients
Authors: Marisa Rifada, Roro Indrianingtyas, Audi Amaris, Akhbar Ramadhan Putra, Bitasya Nanda Dwita
Volume: 10
Issue: 7
Pages: 87-93
Publication Date: 2026/07/28
Abstract:
Diabetes mellitus is a major global health threat often accompanied by dyslipidemia, which elevates cardiovascular risks. While rising cases demand precise predictive tools for cholesterol-related complications, existing studies largely rely on simple linear models that fail to capture the nonlinear, heterogeneous nature of clinical data. This study modeled total cholesterol levels based on blood glucose in diabetes patients using a flexible nonparametric local linear estimator. Secondary data from 42 outpatients at RSUD Kabupaten Karangasem. The proposed approach was compared against parametric linear, quadratic, and cubic regression models. Bandwidth selection employed Cross-Validation, Generalized Cross-Validation, and Rule of Thumb across seven kernel functions. Results revealed a significant positive correlation between blood glucose and cholesterol levels (r = 0.710, p < 0.05). Furthermore, the local linear estimator utilizing a Gaussian kernel and Cross-Validation bandwidth (h = 1.49) demonstrated superior predictive performance. It achieved an MSE of 104.56, a MAPE of 2.27%, and an R-squared of 90.51%, vastly outperforming parametric models, which yielded R-squared values of 40.70-48.77%. In conclusion, the nonparametric local linear approach offers a highly accurate, flexible method for predicting cholesterol levels from blood glucose data. It presents a robust decision-support tool for clinicians managing diabetes-related cardiovascular complications.