Medicine (Baltimore). 2026 Jun 12;105(24):e49278. doi: 10.1097/MD.0000000000049278.
ABSTRACT
Insulin resistance is closely associated with mild cognitive impairment (MCI) and sarcopenia. The predictive utility of the metabolic score for insulin resistance (METS-IR) for MCI across different sarcopenia strata and its potential as a substitute for life’s essential 8 (LE-8) remains unclear. This prospective cohort study used data from the China Health and Retirement Longitudinal Study. Participants aged ≥45 years without baseline MCI or memory-related disease and with biomarker and cognitive data were included. METS-IR was calculated from fasting blood glucose, triglycerides, high-density lipoprotein cholesterol, and body mass index. LE-8 was scored using an adapted American Heart Association framework. Associations between METS-IR and incident MCI stratified by sarcopenia status were assessed using Cox proportional hazards models and restricted cubic splines. Machine-learning analyses using 11 classifiers and stratified 10-fold cross-validation evaluated the incremental and substitutional value of METS-IR within LE-8-based prediction models. A total of 4980 participants were included. Higher METS-IR was associated with a less favorable cardiometabolic profile. In multivariable-adjusted Cox models, compared with the lowest quartile of METS-IR, hazard ratios for incident MCI were 0.94 (95% confidence interval [CI]: 0.79-1.11) for quartile 2, 0.65 (95% CI: 0.53-0.78) for quartile 3, and 0.72 (95% CI: 0.59-0.87) for quartile 4. Restricted cubic spline analyses showed a significant nonlinear association between METS-IR and incident MCI in the overall sample (P for overall < .001, P for nonlinear = .017). A significant overall association was also observed in the possible sarcopenia subgroup, although nonlinearity was not statistically significant. No significant nonlinear or overall association was observed in the nonsarcopenia or confirmed sarcopenia subgroups. In prediction analyses, LE-8-based models achieved moderate discrimination (area under curve, AUC: 0.525-0.696). Adding METS-IR yielded a modest but statistically significant improvement only in AdaBoost (ΔAUC = +0.00487, P < .05). When selected biochemical LE-8 components were substituted by METS-IR, discrimination improved in several algorithms in multiple models, particularly in multilayer perceptron (P < .01). METS-IR showed a significant nonlinear association with incident MCI, with the lowest risk observed at intermediate-to-higher METS-IR levels. The incremental and substitutional value beyond LE-8 was limited, model-dependent, and requires external validation.
PMID:42299594 | DOI:10.1097/MD.0000000000049278
