This study draws upon data from the CHARLS, a nationally representative cohort study initiated in 2011, with subsequent follow-up waves in 2013, 2015, 2018, and 2020 (16). A total of 12,527 participants were excluded based on the following criteria: missing data on eGDR (n = 7,767); a diagnosis of DM in 2011 (n = 1,486); a history of brain injury, intellectual disability, stroke, or memory impairment, or incomplete information (n = 524); a diagnosis of cognitive impairment or missing cognitive impairment data in 2011 (n = 2,490); age under 45 years (n = 124); or loss to follow-up (n = 136). Following these exclusions, the final sample comprised 5,178 eligible participants (Figure 1). All study participants provided written informed consent before enrollment. This research project received ethical approval from Peking University’s Biomedical Ethics Review Committee (IRB00001052-11015), with data collection strictly limited to consenting individuals for final analysis.

The estimated glucose disposal rate was derived from the equation: eGDR (mg/kg/min) = 21.158-(0.09 × waist circumference [cm])-(3.407 × hypertension [1 = yes, 0 = no])-(0.551 × hemoglobin A1c) [%]). Participants were then stratified by eGDR quartiles for insulin resistance level comparisons.

The CHARLS employed the Mini-Mental State Examination (MMSE) to measure cognitive performance, utilizing this standardized tool’s capacity to evaluate both global functioning and specific domains including memory retention and cognitive processing. For memory assessment, researchers administered a ten-item verbal recall test, with participants required to repeat words both immediately following presentation and after a 5-min delay, where one point was allocated for each accurate response (potential score: 0–20). The evaluation of fundamental cognitive capacities incorporated three components: arithmetic tasks involving successive subtraction from 100, geometric figure replication to assess spatial reasoning, and temporal awareness questions regarding date identification. Performance on these measures contributed equally to a maximum of 11 points. By aggregating results from both domains (total possible: 31 points), investigators identified cognitive impairment using a validated cutoff of <11 points (17, 18).

This investigation expanded upon existing literature by incorporating a multidimensional array of covariates spanning sociodemographic attributes, health behaviors, and clinical biomarkers. Participant profiles captured age, sex, residential classification (urban/rural), geographical location (northern/southern China), educational background (categorized as ≤9 years, 10–12 years, or ≥13 years of schooling), and partnership status (married/cohabiting versus single/divorced/widowed). Health behavior indicators documented tobacco use, alcohol consumption, and sensory impairments, alongside psychosocial factors (social engagement levels and depressive symptoms) and cardiometabolic risk markers (elevated blood pressure and adiposity). Biochemical analyses quantified glycemic control (HbA1c, fasting blood glucose [FBG]), hematologic parameters (hemoglobin), and lipid profiles (total cholesterol [TC], triglycerides [TG], high-density lipoprotein cholesterol [HDL-C], low-density lipoprotein cholesterol[LDL-C]). Adiposity was determined via body mass index (BMI) (weight[kg]/height[m]²), classifying obesity at ≥28 kg/m². Hypertension criteria included: (1) clinical diagnosis, (2) antihypertensive medication use, or (3) systolic/diastolic pressures exceeding 140/90 mmHg. Diabetes mellitus was operationalized through: (1) self-reported diagnosis, (2) glucose-lowering drug use, (3) FPG ≥ 126 mg/dL (7.0 mmol/L), or 4) HbA1c ≥ 6.5%. Depressive symptomatology was evaluated using the CESD-10 instrument (score range: 0–30 points).

Comparisons across eGDR quartiles were conducted to examine variations in demographic, health, and metabolic characteristics, including age, sex, education, marital status, rural residence, geographic region, BMI, WC, systolic blood pressure (SBP), diastolic blood pressure (DBP), obesity, smoking, alcohol use, hemoglobin, FBG, vision impairment, HbA1c, TC, TG, HDL, LDL, diabetes, hearing loss, depressive symptoms, and social isolation. Continuous data following normal distributions were summarized as means with standard deviations (mean ± SD) and compared using parametric analysis of variance, while non-normally distributed measures were reported as medians with interquartile ranges [median (IQR)] and analyzed through non-parametric Kruskal-Wallis tests. Categorical data were expressed as frequency counts with percentages [n (%)], with group differences examined via χ² tests. The dose-response association between eGDR and cognitive impairment was investigated using restricted cubic splines (RCS), with Cox proportional hazards models applied to evaluate this relationship through both continuous and categorical parameterizations of eGDR. Three progressively adjusted models were constructed: a crude model (unadjusted), a partially adjusted model (controlling for demographic factors including age, sex, residence location, marital status, and education level, along with behavioral covariates of smoking and alcohol consumption), and a fully adjusted model (incorporating all potential confounders). Additional stratified analyses were performed using multivariable Cox regression to identify potential effect modifications across population subgroups. Kaplan-Meier survival analysis with log-rank tests compared cognitive impairment risk across eGDR quartiles. Sensitivity analyses were conducted under four conditions: (1) excluding participants with cognitive impairment onset by 2013 and (2) redefining diabetes based solely on FBG and HbA1c levels.

Table 1 displays the baseline characteristics of participants stratified by quartiles of eGDR. Significant differences were observed across most demographic and health variables among the eGDR quartiles (P < 0.05). Participants in the lowest quartile of eGDR (Quartile 1, indicating higher insulin resistance) were generally older, had higher waist circumference, HbA1c, FBG, BMI, and blood pressure levels compared to those in higher eGDR quartiles. Conversely, HDL levels were lowest and triglyceride levels highest in Quartile 1, indicative of poorer metabolic health in this group. Notably, gender, vision and hearing impairment, educational level, alcohol consumption, and social isolation did not vary significantly across eGDR quartiles (P > 0.05).

During the follow-up period, 1,913 participants (36.94%) developed cognitive impairment (Supplementary Table 1). The RCS analysis (Figure 2) revealed a significant non-linear association between eGDR and cognitive impairment incidence, with a higher risk of cognitive impairment observed as eGDR decreased (indicating increased insulin resistance) (P < 0.05). This association persisted across all adjusted models, suggesting a potential threshold effect in the link between eGDR and cognitive impairment risk.

The Cox proportional hazards models demonstrated an inverse relationship between eGDR and cognitive impairment risk. Progressive multivariable adjustment revealed consistent associations: each unit reduction in eGDR corresponded to a 21.8% lower risk (HR = 0.792, 95%CI: 0.745–0.801, P = 0.014) in the unadjusted model, 19.5% (hazard ratio [HR] = 0.805, 95% confidence interval [CI]: 0.795–0.818, P = 0.014) after demographic adjustment, and 15.8% (HR = 0.842, 95%CI: 0.793–0.881, P = 0.039) in the fully-adjusted model (Table 2). When analyzed categorically, the highest three eGDR quartiles showed non-significant protective trends (all HR < 1, P > 0.05) in Model III (Table 2). Supporting these findings, Kaplan-Meier curves displayed significant divergence in cognitive impairment incidence by eGDR quartile (log-rank P = 0.003), with progressively shorter median survival times observed in lower quartiles (Figure 3).

In the fully adjusted models, three metabolic indices demonstrated distinct associations with cognitive impairment. The metabolic score for insulin resistance (METS-IR) exhibited a significant inverse relationship, with each standard deviation increase corresponding to a reduced risk of cognitive impairment (HR = 0.99, 95%CI: 0.98–1.00, P = 0.002) (Supplementary Table 2). This protective effect was more pronounced in the quartile analyses, where participants in the highest METS-IR quartile had an 18% lower risk of cognitive impairment compared to those in the lowest quartile (HR = 0.82, 95%CI: 0.72–0.94, P = 0.005) (Supplementary Table 2). For the atherogenic index of plasma (AIP), linear regression analysis revealed a non-significant trend (HR = 0.90, 95%CI: 0.78–1.05, P = 0.170), although participants in the highest AIP quartile approached marginal significance (HR = 0.89, 95%CI = 0.78–1.02, P = 0.100) (Supplementary Table 3). In contrast, the triglyceride glucose index (TyG) demonstrated a near-significant linear association with cognitive impairment risk (HR = 0.85, 95% CI: 0.72–1.00, P = 0.050), with participants in the highest TyG quartile showing robust protection against cognitive impairment (HR = 0.83, 95%CI: 0.73–0.95, P = 0.010) (Supplementary Table 4).

The association between eGDR and cognitive impairment risk demonstrated significant heterogeneity by smoking status. Among never-smokers, each SD increment in eGDR corresponded to a 12.2% lower risk (HR = 0.822, 95%CI: 0.784–0.861, P = 0.038). Smokers showed a similar but non-significant inverse relationship (P = 0.216), with significant between-group heterogeneity (pinteraction = 0.023). No significant effect modification was observed for age, sex, or alcohol consumption (all pinteraction > 0.05, Table 3).

Sensitivity analyses using alternative modeling approaches consistently showed modest associations between continuous eGDR measurements and cognitive outcomes (Table 4). Both models produced comparable effect estimates, reinforcing the primary findings while demonstrating robustness to different analytical specifications.