Integrating conventional and penalized regression models to identify lipid-related predictors of glycemic transitions in prediabetes

Xu, Ziyun, Zhu, Liwei, Wu, Jingyi, Zhang, Lei, Zhou, Yicen, Huang, Bo, Jiang, Yan, Du, Zhenfang, Chen, Shen, Tan, Huifeng, Qiang, Sheng
Lipids in Health and Disease · Springer Nature · 2026-11-21 · DOI: 10.1186/s12944-026-03121-y

Abstract

Background Prediabetes is a heterogeneous metabolic state that may progress to diabetes or revert to normal fasting glucose (NFG). This study evaluated the associations of seven non-traditional lipid-related indices with glucose state transitions in adults with prediabetes. Methods This retrospective secondary analysis used data from a multicenter health examination cohort in China. Prediabetes was defined as a baseline fasting plasma glucose (FPG) level of 5.6–6.9 mmol/L. Diabetes was defined as a final FPG level of ≥7.0 mmol/L or a recorded diagnosis of diabetes, whereas NFG was defined as a final FPG level of <5.6 mmol/L among participants without diabetes. Seven indices were evaluated: LCI, AIP, non-HDL-C, CRI-I, CRI-II, RC, and the RC/HDL-C ratio. Multivariable Logistic regression, restricted cubic splines, breakpoint analysis, subgroup analysis, Elastic-net Logistic regression, SHAP analysis, and receiver operating (ROC) analysis were performed. Results Among 15,387 participants followed for a median of 2.88 years, 1,797 (11.68%) progressed to diabetes, 6,984 (45.39%) remained prediabetic, and 6,606 (42.93%) reverted to NFG. After full adjustment, higher CRI-I, RC/HDL-C ratio, RC, AIP, and LCI were associated with higher risk of progression to diabetes and lower odds of reversion to NFG, whereas non-HDL-C and CRI-II were not significantly associated with either outcome. AIP showed a comparatively pronounced association with diabetes progression (OR per 1-SD increase = 1.294, 95% CI 1.152–1.453, p < 0.001) and was inversely associated with NFG reversion (OR = 0.867, 95% CI 0.806–0.932, p < 0.001). Nonlinear associations were identified for several indices. Elastic-net Logistic regression retained AIP and the RC/HDL-C ratio for diabetes progression and AIP for NFG reversion. The SHAP-selected model for diabetes progression achieved ROC AUC of 0.811 (95% CI 0.801–0.821). Conclusion Higher CRI-I, RC/HDL-C ratio, RC, AIP, and LCI were associated with a greater likelihood of progression to diabetes and a lower likelihood of reversion to NFG. AIP showed comparatively pronounced associations across both outcomes and may complement established clinical factors in characterizing glycemic trajectories, although prospective studies and external validation are required before clinical application.

Read at publisher →

Similar articles