Authors: Baiyang Gu, Yifan Bu, Zhihua Hao, Mingrong Song, Jing Chen
Categories: Research, METS-IR, Glucose metabolism, Lipid metabolism, Metabolic disorders, Cognitive function
Source: European Journal of Medical Research
Authors: Baiyang Gu, Yifan Bu, Zhihua Hao, Mingrong Song, Jing Chen
Metabolic disorders are closely associated with cognitive decline, and the metabolic score for insulin resistance (METS-IR), as a comprehensive indicator reflecting metabolic status, may offer a novel perspective for cognitive function assessment. This study aims to explore the association between METS-IR and cognitive function in older adults and evaluate its potential value in assessing the risk of cognitive impairment.
A cross-sectional analysis was carried out utilizing data from the National Health and Nutrition Examination Survey conducted between 2011 and 2014. The relationship between METS-IR and cognitive function was analyzed using weighted multivariable linear regression. Nonlinear relationships were examined using restricted cubic spline techniques and threshold effect analyses. In addition, subgroup interaction analyses were performed based on demographic characteristics, lifestyle factors, and metabolic diseases. Finally, the discriminative ability of METS-IR for cognitive decline was evaluated using receiver operating characteristic curve analysis.
METS-IR was significantly negatively correlated with all cognitive function test scores. After adjusting for demographic characteristics, lifestyle factors, and metabolic-related diseases, the negative association remained significant. Threshold effect analysis suggested that there is a tipping point, where the association between METS-IR and cognitive impairment becomes more evident. Subgroup analyses indicated that most subgroups did not exert a significant modifying effect on the association between METS-IR and cognitive function; however, smoking may affect the strength of the relationship between METS-IR and cognitive decline. The ROC curves demonstrated that METS-IR demonstrated relatively stable discriminative ability for cognitive decline, highlighting its potential as a supplementary tool for risk stratification.
Elevated METS-IR is independently associated with cognitive decline in older adults and shows a threshold effect. Its accessibility and stable discriminative ability support its potential as a supplementary tool for risk stratification. Future longitudinal studies are needed to verify the causal relationship and explore the feasibility of improving cognitive function through metabolic interventions. This study provides new evidence for the association between metabolic health and brain health, offering scientific basis for early screening and intervention strategies for high-risk populations.
The online version contains supplementary material available at 10.1186/s40001-025-03522-2.
Dementia is a neurodegenerative condition marked by a gradual deterioration in cognitive function. As a result of this decrease in cognitive function, many patients experience psychological and behavioral issues and might even become unable to live on their own. This situation creates considerable physical, mental, and financial challenges for both the individuals affected and their caregivers [1]. Currently, more than 50 million people worldwide are affected, resulting in economic losses exceeding $1.3 trillion. With the intensification of population aging, it is projected that by 2050, the number of affected individuals may surpass 150 million [2], leading to an even heavier burden of disease. Undoubtedly, dementia has emerged as one of the most significant challenges to global public health. As late-stage dementia is nearly irreversible [3], the academic community is increasingly focusing on the early stages of dementia. Mild cognitive impairment is mainly identified by a reduction in cognitive function that exceeds what is typical for a person's age and education [4]. It is considered a precursor and risk state for dementia [5, 6], and is often regarded as a critical window for preventing the onset of dementia. Early identification and intervention for mild cognitive impairment are beneficial in reducing the incidence of dementia.
The diagnostic methods for cognitive impairment primarily include neuropsychological assessment, imaging studies, and biomarker testing [7]. The early symptoms of cognitive impairment are often subtle, overlapping with normal aging or mild psychophysiological disorders, making slight cognitive changes easily overlooked. Although neuropsychological tests have limitations in sensitivity and specificity [8], they remain the current gold standard for evaluating cognitive function in large-scale population studies. Early imaging manifestations of cognitive impairment are often not apparent, and imaging diagnostics are costly. Biomarker detection using cerebrospinal fluid is not only expensive but also invasive, which often leads to low acceptance among patients [9]. Against this backdrop, identifying a reliable and easily obtainable indicator to serve as a complementary measure is valuable for early risk stratification and hypothesis generation regarding the mechanisms of cognitive decline. In this study, cognitive tests are treated as population-level outcomes, rather than a reference standard to be surpassed. The objective is to examine whether an easily obtainable composite indicator of metabolic status is associated with cognitive performance and to explore its potential as a pragmatic, supplementary tool for early risk stratification while providing preliminary, hypothesis-generating evidence to inform subsequent experimental hypotheses about the metabolic pathways implicated in cognitive decline.
Evidence suggests that metabolic disorders, such as insulin resistance and dyslipidemia, may significantly increase the risk of cognitive decline [10–12]. Insulin resistance may affect the brain through two inter-related pathways. On one hand, brain insulin resistance has been associated with neuroinflammation, abnormal protein accumulation, and oxidative stress, ultimately contributing to cognitive decline [13–15]. On the other hand, peripheral insulin resistance is also closely linked to an increased risk of cognitive impairment [16–18] and may exacerbate central insulin signaling deficits and cognitive deterioration [19–21]. This study focuses on peripheral insulin resistance, because it is more readily measurable in large-scale epidemiological data sets and provides a practical tool for risk stratification at the population level. Lipid-related abnormalities are among the neuropathological features first noted in Alzheimer’s disease [22]. Hyperlipidemia significantly increases the risk of dementia [23], and lipid metabolic disorders may contribute to cognitive decline through multiple pathways, such as accelerating abnormal protein aggregation, damaging blood vessels and brain structures, inducing mitochondrial dysfunction, and mediating neuroinflammation [12, 24]. Disruptions in glucose metabolism and lipid metabolism often occur together and mutually reinforce each other in terms of pathophysiological mechanisms. These disruptions not only collectively contribute to metabolic syndrome but also jointly participate in the onset and progression of cognitive decline.
Identifying risk factors for cognitive decline and implementing reasonable preventive measures for susceptible populations can effectively reduce the prevalence of mild cognitive impairment and dementia. Such approaches also help control the disease burden more effectively and provide significant cost benefits. The Metabolic Score for Insulin Resistance (METS-IR) integrates multidimensional information, including fasting blood glucose (FBG), triglycerides (TG), high-density lipoprotein cholesterol (HDL-C), and body mass index (BMI), to comprehensively reflect an individual's metabolic status [25]. METS-IR is an efficient and cost-effective assessment tool, demonstrating particular strength in assessing all-cause mortality, cardiovascular events, and metabolic syndrome [25, 26]. In the detection of metabolic syndrome, its area under the curve (AUC) reaches 0.856, outperforming traditional indicators with similar functions, such as Triglyceride Glucose Index (TyG) and TG/BMI, and exhibiting favorable sensitivity and specificity [27]. Prior studies have linked single metabolic indices to cognition, yet the value of a composite burden index remains uncertain. It is well-established that commonly used metabolic indices (e.g., BMI, HOMA-IR, and TG) are associated with cognition; however, each typically captures only one facet of metabolic dysregulation, such as adiposity for BMI, insulin resistance for HOMA-IR, and lipid abnormalities for triglycerides. By contrast, METS-IR integrates glucose- and lipid-related components and may better reflect overall metabolic status and its relationship to cognition. This study does not position METS-IR as a mechanistic biomarker; rather, it reflects joint dysglycemia, dyslipidemia, and adiposity burden and can reveal dose–response and threshold patterns that are useful for hypothesis generation and risk stratification in cross-sectional settings. Moreover, METS-IR is readily obtainable, and these characteristics collectively provide a potential comprehensive framework for assessing cognitive decline.
Based on this foundation, the present study used data from the 2011–2014 cycles of the National Health and Nutrition Examination Survey (NHANES) to examine the association between METS-IR and cognitive function among older adults. NHANES is a nationally representative survey with a rigorous sampling design and standardized data collection procedures, ensuring comparability across subpopulations and minimizing measurement bias. These features provide unique advantages for investigating the relationship between metabolic and cognitive function. The 2011–2014 cycles were selected, because they include multidimensional cognitive assessments (CERAD W-L, AFT, and DSST) as well as the metabolic and biochemical parameters required to calculate METS-IR, allowing for a comprehensive evaluation of metabolic–cognitive associations. In contrast, subsequent NHANES waves did not include comparable cognitive assessments. Therefore, this period provides the most suitable and reliable data set for exploring the relationship between METS-IR and cognitive performance in older adults. The aim of this study was to assess the feasibility of using METS-IR as a supplementary tool for monitoring cognitive status and stratifying risk, thereby providing a convenient and reliable reference for clinical practice and preliminary evidence to inform subsequent experimental hypotheses.
The data for this study were obtained from the publicly available NHANES database. The NHANES program is approved by the National Center for Health Statistics Research Ethics Review Board, and informed consent was obtained from all participants. A total of 19,931 participants were drawn from the 2011–2014 NHANES data set, and the following inclusion and exclusion criteria were 1. a total of 16,299 individuals under the age of 60 were excluded, leaving 3,632 participants. 2. A total of 698 participants missing cognitive score data were excluded, leaving 2,934 participants. 3. Excluding 1,559 participants who lacked data on BMI, TG, FBG, and HDL-C, the final cohort consisted of 1,375 individuals who had complete data sets included in the study. The process of selection is depicted in Fig. 1.Fig. 1Diagram of eligible participant’s selection process. NHANES: the National Health and Nutrition Examination Survey. A total of 1375 participants aged ≥ 60 years with complete data on cognitive tests and METS-IR were included
The METS-IR was calculated using FBG, TG, HDL-C, and BMI data from the laboratory tests and physical measurements in the NHANES data set. The specific calculation formula is as \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ {\text{METS}} - {\text{IR}}, = ,{\text{Ln }}\left[ {{2}, \times ,{\text{FBG }}\left( {{\text{mg}}/{\text{dL}}} \right), + ,{\text{TG }}\left( {{\text{mg}}/{\text{dL}}} \right)} \right], \times ,{\text{BMI}}/{\text{Ln HDL}} - {\text{C }}\left( {{\text{mg}}/{\text{dL}}} \right).
### Outcome variables The NHANES study utilized various standardized cognitive tests to assess participants' cognitive function. These tests included the CERAD word learning subtest (CERAD W-L), the animal fluency test (AFT), and the digit symbol substitution test (DSST), which are widely used in large-scale cognitive impairment screening. CERAD W-L includes the immediate recall test and the delayed recall test, which are commonly used to assess the ability to encode and retrieve verbal material [28], The AFT is primarily used to evaluate verbal fluency and executive function [29], while the DSST assesses sustained attention, processing speed, and executive function [30]. These tests provide a comprehensive evaluation of participants' cognitive function, with lower scores indicating poorer cognitive performance. In this study, cognitive function was categorized into two participants scoring below the 25th percentile on any cognitive test were defined as the cognitive impairment group, while the remaining participants were classified as the normal cognitive function group. This method has been validated in previous studies for assessing cognitive function [31]. To account for the fact that a single cognitive test may only reflect a specific aspect of cognitive function, combining the scores from these tests offers a more comprehensive reflection of overall cognitive performance. Thus, in this study, cognitive function scores were standardized and merged using Z-score conversion to generate a unified standardized comprehensive cognitive function score. The specific method is as each cognitive test score was standardized using Z-score transformation, following the Z = (X − μ)/σ. The standardized scores were then averaged to produce the standardized comprehensive cognitive function score according to the standardized comprehensive cognitive function score = (CERAD W-L~Z~ + AFT~Z~ + DSST~Z~)/3. Higher scores indicate better cognitive function [32]. ### Covariates Based on previous studies [33], this study evaluated several potential covariates, including age, gender, race, smoking, drinking, stroke, overweight, coronary heart disease (CHD), and sleep disorders. In this study, the primary estimand was the overall population-level association between metabolic status (METS-IR) and cognitive performance. Socio-economic factors were not included in the primary models, as they represent broad socio-demographic constructs and proxies for cognitive reserve. Adjusting for these factors could shift the estimand toward a direct effect net of social determinants and risk over-adjustment; therefore, only core demographic and lifestyle covariates were included to reduce confounding while preserving the total association relevant for risk stratification. In accordance with NHANES documentation and established epidemiological practice, key covariates were categorized rather than modeled solely as continuous variables. This approach reflects cumulative exposure and long-term risk, and better captures the nonlinear relationships between exposures and health outcomes. It provides greater clinical interpretability, is consistent with recommended standards, and ensures comparability with existing literature. Race was categorized into five Mexican–American, other Hispanic, non-Hispanic White, non-Hispanic Black, and others. Smoking was classified as No (fewer than 100 cigarettes smoked in a lifetime) and Yes (at least 100 cigarettes smoked in a lifetime). Drinking was categorized as No (fewer than 12 alcoholic drinks in the past 12 months) and Yes (12 or more alcoholic drinks in the past 12 months). Stroke was defined as Yes (The doctor or other health professional ever told you that you had a stroke) and No (The doctor or other health professional never told you that you had a stroke). Overweight status was categorized as Yes (The doctor or other health professional ever said you were overweight) and No (The doctor or other health professional never said you were overweight). CHD was classified as Yes (The doctor or other health professional ever told you that you had coronary heart disease) and No (The doctor or other health professional never told you that you had coronary heart disease). Sleep disorder was grouped as Yes (The doctor or other health professional ever told you that you had a sleep disorder) and No (The doctor or other health professional never told you that you had a sleep disorder). ### Statistical analysis R (version 4.2.1) and Empower Stats (version 2.0) were utilized for statistical analyses in this research. All statistical procedures incorporated sampling weights from NHANES. Continuous variables were expressed as mean ± standard deviation, whereas categorical variables were represented in percentages to comprehensively describe the characteristics of different groups. To compare differences between the normal cognitive function group and the cognitive decline group, weighted linear regression and weighted chi-square tests were utilized to assess baseline variations in both continuous and categorical variables. Multivariable linear regression models were utilized to investigate the relationship between METS-IR and cognitive function. Collinearity was assessed using the variance inflation factor (VIF), and the primary model assumptions were examined. The results of these evaluations are provided in the supplementary materials. Three models were constructed to assess this Model 1 (no covariate adjustments). Model 2 (adjusted for age, race, and gender). Model 3 (further adjusted for potential confounding factors, including gender, age, race, smoking, drinking, stroke, overweight, CHD, and sleep disorders). Restricted cubic spline (RCS) curves were applied to Model 3 to explore the potential nonlinear relationship between METS-IR and cognitive function. Threshold effect analysis was performed using a two-piecewise linear regression model to detect potential inflection points in the relationship between METS-IR and cognitive function. The log-likelihood ratio test was applied to compare the one-line linear regression model with the two-piecewise linear regression model and to determine whether a threshold effect was present. Subgroup analyses were performed by categorizing participants into various levels based on factors, such as gender, age, race, smoking, drinking, stroke, overweight, CHD, and sleep disorders. To assess heterogeneity among the subgroups, interaction terms were incorporated. Multiple testing correction was conducted using the Benjamini–Hochberg False Discovery Rate (BH-FDR) method. Finally, receiver operating characteristic (ROC) curve analysis was used to evaluate the discriminative ability of METS-IR for cognitive decline. Under the same modeling framework, this study assessed the discrimination of BMI, FBG, HDL-C, TG, METS-IR, Homeostatic Model Assessment for Insulin Resistance (HOMA-IR), and TyG for classifying cognitive decline, and assessed correlations between METS-IR and HOMA-IR/TyG using Pearson and Spearman methods. The formula for HOMA-IR HOMA-IR = [Insulin (µU/mL)] × [FBG (mmol/L)]/22.5; the formula for TyG TyG = Ln ([TG (mg/dL)] × [FBG (mg/dL)]/2). ## Results ### Baseline characteristics of the population The study population consisted of 1,375 participants drawn from the NHANES from 2011 to 2014. Cognitive decline was used as a stratification variable, and participants were categorized and described based on gender, age, race, smoking, drinking, stroke, overweight, CHD, and sleep disorders. The average age of participants was 69.62 ± 6.78 years, with 49.02% (674 participants) being male and 50.98% (701 participants) being female. Among the participants, 28.15% (CERAD W-L), 29.31% (AFT), 26.55% (DSST), and 25.02% (comprehensive cognitive function score) exhibited cognitive decline, as measured by their respective cognitive assessments. It should be noted that cognitive decline was operationally defined as scoring in the lowest quartile of the cognitive test distribution. This categorization reflects relative performance within the study sample, rather than the actual prevalence of cognitive impairment in the population. Compared to individuals with normal cognitive function, those with cognitive decline were generally older and more likely to be male, non-Hispanic Black, alcohol consumers, or CHD patients. In addition, patients with a history of stroke exhibited a significantly higher prevalence of cognitive impairment in both the CERAD W-L and DSST tests, indicating that stroke is strongly associated with poorer cognitive performance. Overweight showed a potential association with decreased cognitive function, though the statistical significance of this relationship was weak. Overall, cognitive decline was associated with multiple risk factors, including advanced age, male gender, non-Hispanic Black race, drinking, stroke, and CHD. There was also a possible trend toward an association between overweight and cognitive function. However, smoking and sleep disorders did not appear to have significant effects in this study. Detailed data are presented in Table 1. Table 1Characteristics of participantsCharacteristicsCERAD W-L*P* valueAFT*P* valueDSST*P* valueComprehensive Cognitive Score*P* valueNormal cognitive *N = *988Low cognitive *N = *387Normal cognitive *N = *972Low cognitive *N = *403Normal cognitive *N = *1010Low cognitive *N = *365Normal cognitive *N = *1031Low cognitive *N = *344Age, years68.59 ± 6.5272.27 ± 6.73 < 0.00169.07 ± 6.7370.95 ± 6.73 < 0.00168.88 ± 6.6371.69 ± 6.79 < 0.00168.71 ± 6.5472.35 ± 6.77 < 0.001Gender, *n* (%) < 0.0010.2580.0850.041 Male440 (44.53)234 (60.47)486 (50.00)188 (46.65)481 (47.62)193 (52.88)489 (47.43)185 (53.78) Female548 (55.47)153 (39.53)486 (50.00)215 (53.35)529 (52.38)172 (47.12)542 (52.57)159 (46.22)Race, *n* (%)0.004 < 0.001 < 0.001 < 0.001 Mexican American78 (7.89)45 (11.63)78 (8.02)45 (11.17)71 (7.03)52 (14.25)79 (7.66)44 (12.79) Other Hispanic90 (9.11)53 (13.70)99 (10.19)44 (10.92)76 (7.52)67 (18.36)89 (8.63)54 (15.70) Non-Hispanic White520 (52.63)175 (45.22)549 (56.48)146 (36.23)579 (57.33)116 (31.78)568 (55.09)127 (36.92) Non-Hispanic Black196 (19.84)83 (21.45)165 (16.98)114 (28.29)169 (16.73)110 (30.14)187 (18.14)92 (26.74) Other Race104 (10.53)31 (8.01)81 (8.33)54 (13.40)115 (11.39)20 (5.48)108 (10.48)27 (7.85)Smoking, *n* (%)0.7890.6510.9700.987Yes495 (50.10)197 (50.90)493 (50.72)199 (49.38)508 (50.30)184 (50.41)519 (50.34)173 (50.29)No493 (49.90)190 (49.10)479 (49.28)204 (50.62)502 (49.70)181 (49.59)512 (49.66)171 (49.71)Drinking, *n* (%)0.4730.0020.001 < 0.001 Yes681 (68.93)259 (66.93)689 (70.88)251 (62.28)715 (70.79)225 (61.64)730 (70.81)210 (61.05) No307 (31.07)128 (33.07)283 (29.12)152 (37.72)295 (29.21)140 (38.36)301 (29.19)134 (38.95)Stroke, *n* (%)0.0190.0680.0060.051 Yes59 (5.97)37 (9.56)60 (6.17)36 (8.93)59 (5.84)37 (10.14)64 (6.21)32 (9.30) No929 (94.03)350 (90.44)912 (93.83)367 (91.07)951 (94.16)328 (89.86)967 (93.79)312 (90.70)Overweight, *n* (%)0.0280.0870.1240.122 Yes403 (40.79)133 (34.37)393 (40.43)143 (35.48)406 (40.20)130 (35.62)414 (40.16)122 (35.47) No585 (59.21)254 (65.63)579 (59.57)260 (64.52)604 (59.80)235 (64.38)617 (59.84)222 (64.53)CHD, *n* (%)0.0030.1060.9830.022 Yes83 (8.40)53 (13.70)88 (9.05)48 (11.91)100 (9.90)36 (9.86)91 (8.83)45 (13.08) No905 (91.60)334 (86.30)884 (90.95)355 (88.09)910 (90.10)329 (90.14)940 (91.17)299 (86.92)Sleep disorders, *n* (%)0.8770.2940.4440.772 Yes123 (12.45)47 (12.14)126 (12.96)44 (10.92)129 (12.77)41 (11.23)129 (12.51)41 (11.92) No865 (87.55)340 (87.86)846 (87.04)359 (89.08)881 (87.23)324 (88.77)902 (87.49)303 (88.08)FBG, mg/dL112.70 ± 32.26115.25 ± 33.950.237111.44 ± 30.59120.05 ± 38.69 < 0.0001111.73 ± 30.24121.58 ± 42.57 < 0.0001112.22 ± 31.26118.80 ± 38.800.007HDL-C, mg/dL57.33 ± 16.7055.01 ± 17.530.03757.25 ± 16.7655.30 ± 17.360.08357.39 ± 16.5553.82 ± 18.460.00557.30 ± 16.6454.32 ± 18.080.018TG, mg/dL124.61 ± 74.72128.20 ± 71.260.462126.08 ± 72.18122.71 ± 80.510.493125.14 ± 72.95126.70 ± 79.580.777125.46 ± 74.15124.96 ± 73.240.929BMI29.20 ± 6.3428.48 ± 6.640.08829.09 ± 6.1828.86 ± 7.190.58229.00 ± 6.3429.25 ± 6.800.60429.17 ± 6.3728.38 ± 6.570.097METS-IR43.09 ± 12.1742.61 ± 12.160.55242.92 ± 11.8143.22 ± 13.450.71242.76 ± 12.1044.22 ± 12.470.10743.04 ± 12.1542.69 ± 12.270.697Participants with cognitive decline (lowest quartile) were generally older, more likely to be male, non-Hispanic Black, alcohol consumers, or with CHD/strokeCERAD W-L: Consortium to Establish a Registry for Alzheimer’s Disease Word Learning test; AFT: Animal Fluency Test; DSST: Digit Symbol Substitution Test; CHD: Coronary Heart Disease; FBG: Fasting Blood Glucose; HDL-C: High-Density Lipoprotein Cholesterol; TG: Triglycerides; BMI: Body Mass Index; METS-IR: Metabolic Score for Insulin Resistance Given that METS-IR shares certain constituent components with traditional metabolic indices, this study examined pairwise correlations to quantify overlap and inform model specification. METS-IR showed moderate correlations with HOMA-IR (r≈0.35) and TyG (r≈0.51), indicating shared but nonidentical information. Full correlation matrices are presented in Supplementary Table 2. Because BMI, FBG, HDL-C, and TG are embedded in the METS-IR formula, this study did not include METS-IR together with BMI, FBG, HDL-C, or TG in the same regression model; instead, each indicator was evaluated individually, with receiver operating characteristic curves derived accordingly. ### The relationship between METS-IR and cognitive function METS-IR was negatively correlated with CERAD W-L, AFT, DSST, and comprehensive cognitive function scores, and this association became more pronounced after controlling for confounding factors. This suggests that high METS-IR may be an independent risk factor for cognitive decline. Specifically, METS-IR exhibited a negative association with overall cognitive function. In Model 3, each 1-unit increase in METS-IR was associated with a 0.01-point reduction in comprehensive cognitive function score (*β* = − 0.01, 95% CI − 0.01, − 0.01, *P* < 0.0001). METS-IR showed a significant negative correlation with CERAD W-L scores; for each 1-unit increase in METS-IR, CERAD W-L scores decreased by 0.05 points (*β* = − 0.05, 95% CI − 0.08, − 0.01, *P = * 0.0050). Similarly, METS-IR was significantly negatively correlated with AFT scores, with each 1-unit increase in METS-IR corresponding to a 0.08-point reduction in AFT scores (*β* = − 0.08, 95% CI − 0.11, − 0.05, *P* < 0.0001). For DSST scores, each 1-unit increase in METS-IR led to a 0.10-point reduction (β = − 0.10, 95% CI − 0.18, − 0.03, *P = * 0.0085). These findings indicate that the association of METS-IR with cognitive test scores is indicator-specific, suggesting that high METS-IR may independently relate to lower performance in certain cognitive domains. When METS-IR was further divided into quartiles, the associations between each quartile and cognitive test scores were analyzed using Q1 (lowest METS-IR group) as the reference. This approach provides an intuitive assessment of risk stratification, allows for the identification of threshold effects, and enhances clinical interpretability by categorizing individuals into relative risk groups. In Model 3, a significant negative association between METS-IR and CERAD W-L scores was observed only in the highest quartile (*β* = − 1.16, 95% CI − 2.24, − 0.08, *P = * 0.0353), suggesting a threshold effect—memory function (CERAD W-L) is significantly impaired when METS-IR reaches a high level. For AFT, the strongest negative association was observed in Q4 (highest quartile) (*β* = − 2.38, 95% CI − 3.31, − 1.44, P < 0.0001), with the strength of the association increasing across quartiles, supporting a dose–response relationship. Similar results were found for DSST, where Q4 demonstrated a significant negative association (*β* = − 2.69, 95% CI − 5.25, − 0.13, *P = * 0.0395), indicating poorer performance in processing speed or executive function in the highest METS-IR group. Regarding comprehensive cognitive function score, Q4 showed a significantly lower score compared to Q1 (*β* = − 0.26, 95% CI − 0.38, − 0.13, *P* < 0.0001), suggesting that the higher the METS-IR, the worse the overall cognitive function. In summary, METS-IR was significantly negatively associated with multiple dimensions of cognitive function, including memory ability, verbal fluency, processing speed, and overall cognitive function, independent of demographic characteristics, lifestyle, and metabolic diseases. Notably, the strength of associations in Q4 was stronger in Model 3 compared to unadjusted models, indicating that metabolic-related factors such as overweight status and CHD may enhance the negative effects of METS-IR. Detailed data are presented in Table 2. Table 2Associations between METS-IR with cognitive functionModel 1*P* valueModel 2*P* valueModel 3*P* value*β* (95%CI)*β* (95%CI)*β* (95%CI)CERAD W-LMETS-IR– 0.02 (– 0.05, 0.01)0.1245– 0.03 (– 0.06, – 0.01)0.0071– 0.05 (– 0.08, – 0.01)0.0050METS-IR (quartile) Q1000 Q2– 0.74 (– 1.67, 0.19)0.1185– 0.46 (– 1.31, 0.39)0.2915– 0.57 (– 1.44, 0.30)0.1979 Q3– 0.65 (– 1.59, 0.30)0.1790– 0.14 (– 1.00, 0.72)0.7483– 0.35 (– 1.32, 0.62)0.4829 Q4– 0.72 (– 1.65, 0.20)0.1255– 1.00 (– 1.85, – 0.15)0.0210– 1.16 (– 2.24, – 0.08)0.0353AFT METS-IR– 0.02 (– 0.04, 0.01)0.1678– 0.03 (– 0.05, – 0.01)0.0037– 0.08 (– 0.11, – 0.05) < 0.0001METS-IR (quartile) Q1000 Q2– 0.51 (– 1.31, 0.30)0.2208– 0.52 (– 1.27, 0.22)0.1687– 0.99 (– 1.75, – 0.24)0.0098 Q3– 0.67 (– 1.48, 0.15)0.1107– 0.49 (– 1.25, 0.26)0.2025– 1.52 (– 2.36, – 0.68)0.0004 Q4– 0.52 (– 1.33, 0.28)0.2043– 0.93 (– 1.67, – 0.18)0.0148– 2.38 (– 3.31, – 1.44) < 0.0001DSST METS-IR– 0.07 (– 0.14, 0.01)0.0779– 0.10 (– 0.16, – 0.04)0.0013– 0.10 (– 0.18, – 0.03)0.0085METS-IR (quartile) Q1000 Q21.02 (– 1.48, 3.53)0.42492.19 (0.15, 4.24)0.03581.97 (– 0.10, 4.04)0.0625 Q3– 0.98 (– 3.51, 1.55)0.44771.23 (– 0.85, 3.30)0.24720.78 (– 1.52, 3.09)0.5063 Q4– 2.29 (– 4.78, 0.21)0.0723– 2.80 (– 4.85, – 0.75)0.0075– 2.69 (– 5.25, – 0.13)0.0395Comprehensive Cognitive Score METS-IR– 0.00 (– 0.01, 0.00)0.0567– 0.01 (– 0.01, – 0.00)0.0001– 0.01 (– 0.01, – 0.01) < 0.0001METS-IR (quartile) Q1000 Q2– 0.05 (– 0.17, 0.07)0.4151– 0.01 (– 0.11, 0.09)0.7912– 0.05 (– 0.15, 0.05)0.3054 Q3– 0.09 (– 0.21, 0.03)0.1306– 0.01 (– 0.11, 0.09)0.7880– 0.10 (– 0.21, 0.02)0.0935 Q4– 0.11 (– 0.23, 0.01)0.0613– 0.16 (– 0.26, – 0.06)0.0014– 0.26 (– 0.38, – 0.13) < 0.0001METS-IR was significantly negatively associated with multiple dimensions of cognitive functionCI: confidence interval; CERAD W-L: Consortium to Establish a Registry for Alzheimer’s Disease Word Learning test; AFT: Animal Fluency Test; DSST: Digit Symbol Substitution Test. METS-IR: Metabolic Score for Insulin Resistance ### RCS curve plotting and threshold effect analysis To investigate the connection between METS-IR and cognitive abilities, RCS curves were generated and analyses of the threshold effect were performed. The RCS curves were generated from fully adjusted models that included all covariates described in the Statistical Analysis section. The results showed a significant negative correlation between METS-IR and CERAD W-L scores (*P = * 0.006), with an approximately linear decreasing trend. In addition, higher METS-IR levels were significantly associated with lower AFT scores, demonstrating a linear trend. A significant linear relationship was also observed between increased METS-IR levels and reduced DSST cognitive scores, presenting a strong negative correlation. Furthermore, METS-IR was significantly negatively correlated with comprehensive cognitive function score, and the trend appeared to be linear. Detailed data are presented in Fig. 2.Fig. 2Association between METS-IR and cognitive function. **A** The association between METS-IR and CERAD W-L; (**B**) The association between METS-IR and AFT; (**C**) The association between METS-IR and DSST; (**D**) The association between METS-IR and Comprehensive Cognitive Score. CI: confidence interval; CERAD W-L: Consortium to Establish a Registry for Alzheimer’s Disease Word Learning test; AFT: Animal Fluency Test; DSST: Digit Symbol Substitution Test. METS-IR: Metabolic Score for Insulin Resistance. The RCS curves were generated from fully adjusted models. Higher METS-IR showed significant linear negative associations with all cognitive measures A threshold effect analysis was conducted. The findings indicated a significant negative correlation between METS-IR, treated as a continuous variable, and all scores related to cognitive function (overall *P* values < 0.05). This suggests a possible link between increased METS-IR levels and a decline in cognitive function. The overall linear effect between METS-IR and CERAD W-L scores was significant, though no significant association was observed when METS-IR was < 33.25, indicating that the threshold effect occurred at the inflection point of METS-IR = 33.25. The overall trend showed that higher METS-IR levels were associated with lower CERAD W-L scores. For AFT scores, the overall linear effect was significant, with negative associations observed both above and below the threshold. However, when METS-IR was < 39.92, the effect was stronger, potentially reflecting a critical sensitivity of AFT scores to METS-IR. Similarly, a significant linear effect was observed between METS-IR and DSST scores, with a pronounced negative correlation appearing above the threshold (METS-IR > 28.92). The overall linear effect between METS-IR and comprehensive cognitive function score was also significant, and a marked negative correlation was observed above the threshold (METS-IR > 27.78). In conclusion, elevated METS-IR was related to lower cognitive function, with the association becoming more pronounced beyond specific thresholds. Detailed data are provided in Table 3. Table 3Threshold effect analysis for correlations between METS-IR and cognitive functionOutcomesCERAD W-LAFTDSSTComprehensive cognitive scoreLinear model β (95%CI) *p* value– 0.05 (– 0.08, 0.01) 0.0050– 0.08 (– 0.11, – 0.05) < 0.0001– 0.10 (– 0.18, – 0.03) 0.0085– 0.01 (– 0.01, – 0.01) < 0.0001Non-linear model Inflection point (K)33.2539.9228.9227.78 < K-segment effectβ (95%CI) *p* value– 0.11 (– 0.28, 0.05) 0.1641– 0.11 (– 0.19, – 0.04) 0.00241.14 (0.36, 1.91) 0.00410.04 (– 0.01, 0.09) 0.1112 > K-segment effectβ (95%CI) *p* value– 0.04 (– 0.07, – 0.00) 0.0267– 0.07 (– 0.11, – 0.04) < 0.0001– 0.14 (– 0.22, – 0.06) 0.0007– 0.01 (– 0.01, – 0.01) < 0.0001 LLR0.3910.3500.0020.047(A) The association between METS-IR and CERAD W-L; (B) The association between METS-IR and AFT; (C) The association between METS-IR and DSST; (D) The association between METS-IR and Comprehensive Cognitive ScoreCI: confidence interval; LLR: Log-Likelihood Ratio test; CERAD W-L: Consortium to Establish a Registry for Alzheimer’s Disease Word Learning test; AFT: Animal Fluency Test; DSST: Digit Symbol Substitution Test. METS-IR: Metabolic Score for Insulin Resistance. Elevated METS-IR was related to poorer cognitive function, especially above certain thresholds ### Subgroup analysis Subgroup analyses were conducted to evaluate the differences in the association between METS-IR and cognitive decline across various demographic groups. Interaction effects were tested for age, gender, race, smoking, drinking, stroke, overweight, CHD, and sleep disorders, and multiple comparisons were adjusted using the BH-FDR method. The results indicated that the association between METS-IR and CERAD W-L scores did not exhibit significant interactions across any of the subgroups (*P > *0.05), suggesting that these factors did not significantly modify the relationship. Similarly, no significant interactions were observed between METS-IR and AFT scores in most subgroups (*P > *0.05), indicating a lack of notable effect modification; however, the interaction with smoking status reached statistical significance (*P* < 0.05), implicating smoking as a potential modifier in the association between METS-IR and AFT scores. The interaction with sleep disorders approached statistical significance (*P = * 0.0648), suggesting that sleep disorders may also represent a potential modifying factor. In analyses of the association between METS-IR and DSST scores, no significant interactions were found across any subgroups (*P > *0.05), further indicating minimal effect modification by these variables. Collectively, these results suggest a robust and broadly consistent adverse effect of METS-IR on cognitive function across various populations. Comprehensive data can be found in Fig. 3.Fig. 3Subgroup analysis of cognitive function related to METS-IR. **A** Subgroup analysis of METS-IR and CERAD W-L; (**B**) Subgroup analysis of METS-IR and AFT; (**C**) Subgroup analysis of METS-IR and DSST; (**D**) Subgroup analysis of METS-IR and Comprehensive Cognitive Score. CI: confidence interval; CERAD W-L: Consortium to Establish a Registry for Alzheimer’s Disease Word Learning test; AFT: Animal Fluency Test; DSST: Digit Symbol Substitution Test; CHD: Coronary Heart Disease; FBG: Fasting Blood Glucose; HDL-C: High-Density Lipoprotein Cholesterol; TG: Triglycerides; BMI: Body Mass Index; METS-IR: Metabolic Score for Insulin Resistance. METS-IR showed a robust and consistent negative association with cognitive function across populations ### Stability of METS-IR in relation to risk of cognitive decline To assess the model's stability, this research examined the discriminative ability of METS-IR in determining the risk of cognitive decline and made adjustments for various covariates. Model 1 included no covariate adjustment, whereas Model 2 was fully adjusted for all covariates. The results are shown in Fig. 4. ROC analysis indicated that, regardless of whether covariates were adjusted, METS-IR demonstrated relatively stable classification performance in identifying the risk of cognitive decline for CERAD W-L, AFT, DSST, and comprehensive cognitive function scores (AUC > 0.5). Although the AUC for AFT-defined cognitive decline was 0.68—indicating modest discrimination above chance—it still reflects a meaningful ability to distinguish higher from lower risk individuals in this population context. The inclusion of METS-IR in the baseline model led to an improvement in the accuracy of risk stratification, particularly marked in overall cognitive function (ΔAUC 0.135). These findings highlight the potential application value of METS-IR in the risk assessment of cognitive decline. To evaluate the incremental discrimination of METS-IR relative to single or traditional metabolic indices, this study assessed, under an identical modeling framework, the ability of BMI, FBG, HDL-C, TG, HOMA-IR, and TyG to classify cognitive decline. Results are shown in Supplementary Fig. 2. The AUC for METS-IR (AUC 0.762) was modestly higher than those for BMI (AUC 0.661), FBG (AUC 0.667), HDL-C (AUC 0.666), TG (AUC 0.661), HOMA-IR (AUC 0.733), and TyG (AUC 0.733), suggesting slightly better discrimination than single or traditional metabolic indices.Fig. 4ROC curve analysis of METS-IR in assessing cognitive decline. **A** ROC analysis of METS-IR and cognitive decline on CERAD W-L; (**B**) ROC analysis of METS-IR and cognitive decline on AFT; (**C**) ROC analysis of METS-IR and cognitive decline on DSST; (**D**) ROC analysis of METS-IR and cognitive decline on Comprehensive Cognitive Score. ROC: Receiver Operating Characteristic; AUC: Area Under the Curve; CERAD W-L: Consortium to Establish a Registry for Alzheimer’s Disease Word Learning test; AFT: Animal Fluency Test; DSST: Digit Symbol Substitution Test. METS-IR: Metabolic Score for Insulin Resistance. Model 1 unadjusted; Model 2 fully adjusted. METS-IR demonstrated stable discriminative ability across tests ## Discussion This study represents the initial investigation into the relationship between METS-IR and cognitive decline among older adults, utilizing data from the nationally representative NHANES database. A multidimensional cognitive assessment was conducted on 1375 individuals aged 60 and above, which included evaluations based on CERAD W-L, AFT, DSST, and comprehensive cognitive function scores. The results revealed significant negative associations between METS-IR and all cognitive function indicators. Importantly, these relationships persisted strongly even after accounting for demographic characteristics, lifestyle variables, and diseases related to metabolism. It should be noted that mean METS-IR values did not differ significantly between cognitive decline and non-cognitive decline groups in the descriptive comparison (Table 1). This likely reflects information loss from lowest quartile outcome dichotomization and lack of covariate adjustment. By contrast, the multivariable regressions treated both METS-IR and cognitive scores as continuous variables with covariate adjustment, thereby revealing associations not apparent in crude comparisons. Separately, the quartile analyses of the exposure (METS-IR) in Table 2 are presented as a secondary, clinically interpretable they display ordered differences across risk strata and illustrate potential thresholds for risk stratification, consistent with the spline results. Subgroup analyses further indicated that the adverse association between METS-IR and cognition remained consistent across diverse population strata, suggesting a universal nature of this association. Furthermore, threshold effect analysis highlighted the existence of a critical value effect of METS-IR on cognitive function, suggesting that there is a tipping point, where the association between METS-IR and cognitive impairment becomes more evident. Subgroup analysis further indicated that specific factors, such as sleep disorders and ethnicity, may enhance the strength of these associations. These findings provide novel evidence linking metabolic disorders to cognitive decline and propose actionable composite metabolic indices for the early identification of at-risk populations in clinical practice. The connection between metabolic disorders and cognitive decline has attracted considerable attention in recent years. This study identified an association between METS-IR and cognitive decline. Although the analysis focused on peripheral metabolic dysregulation as measured by METS-IR, it is noteworthy that peripheral disturbances may also contribute to central metabolic dysfunction, thereby mediating cognitive impairment; these processes are not mutually exclusive. Such connections are supported by previous evidence. First, persistent hyperglycemia leads to the accumulation of advanced glycation end-products (AGEs) and harmful lipids, which can directly compromise the functionality of cerebrovascular endothelial cells and interfere with the stability of the blood–brain barrier. These processes simultaneously activate microglia, triggering neuroinflammatory responses [34–36]. Second, sustained abnormalities in glucose and lipid metabolism may accelerate Alzheimer’s disease (AD) pathology by promoting β-amyloid (Aβ) deposition and tau protein hyperphosphorylation [24, 37]. This finding corresponds with the observation in this study that cognitive impairment, particularly memory function as reflected by CERAD W-L scores, was most pronounced in the highest METS-IR quartile group. Notably, the inclusion of BMI in the METS-IR formula suggests potential independent effects of obesity-induced metabolic disorders. Inflammatory factors secreted by adipocytes could interfere with neurogenesis through the “brain–adipose axis” [38, 39]. The synergistic interactions between peripheral and central mechanisms, as well as between glucose and lipid metabolism, align with the composite nature of METS-IR and help explain why it shows significant associations with cognition and may offer greater potential for discovery compared to traditional insulin resistance indices, such as HOMA-IR. This simultaneously underscores the importance of integrating systemic and central perspectives in future research. METS-IR exhibited moderate correlations with HOMA-IR and TyG, indicating overlapping but nonidentical information—consistent with the view that METS-IR integrates glucose–lipid–adiposity burden rather than reflecting a single pathway. This finding suggests that, as an integrative index, METS-IR is not a simple substitute for traditional indices but captures additional metabolic information not fully covered by them. Moreover, in classification analyses, the AUC for METS-IR was modestly higher than that of single and traditional metabolic indices (including HOMA-IR, TyG, BMI, and so forth), indicating better discrimination of cognitive impairment. From a pathophysiological perspective, this may imply that interactions and potential synergy between disturbances in glucose and lipid metabolism are more closely linked to cognitive impairment than either disturbance alone. Accordingly, the broader metabolic dimensions encompassed by METS-IR may better integrate the cumulative impact of multiple metabolic insults on brain health and support a more comprehensive clinical classification of cognitive risk. In this context, this study does not position METS-IR as a mechanistic biomarker; rather, it reflects joint dysglycemia–dyslipidemia–adiposity burden and can reveal dose–response and threshold patterns that are useful for hypothesis generation and risk stratification in cross-sectional settings. This framing is supported by the observed inverse associations across cognitive domains and the robustness of results under a common modeling framework, and it underscores the pragmatic use of METS-IR as a composite burden signal rather than a pathway-specific measure. METS-IR exhibits differential associations across cognitive domains. A dose–response relationship observed in AFT (verbal fluency) and DSST (processing speed) shows stronger associations (Q4 group *β* values of − 2.38 and − 2.69, respectively) compared to the association with CERAD W-L (memory function; Q4 group β = − 1.16). This phenomenon may reflect the priority damage caused by metabolic disorders to the frontostriatal pathway—a neural network responsible for executive function and information processing speed—which is more sensitive to energy metabolism abnormalities [40, 41]. Insulin resistance reduces glucose uptake in the prefrontal cortex [42], resulting in impaired synaptic plasticity, which is directly associated with the decline in DSST performance. The hippocampus, as a critical memory center, might maintain functionality during the early stages of metabolic disorders through compensatory mechanisms, such as enhanced fatty acid oxidation [43, 44]. However, once METS-IR exceeds a critical threshold, hippocampal function deteriorates. This is supported by threshold analysis in the present study, which indicates that the association with CERAD W-L becomes markedly stronger when METS-IR exceeds 33.25. These findings provide an important window of opportunity for clinical intervention in cognitive dysfunction. Analysis of the threshold effect demonstrated a nonlinear association between METS-IR and cognitive function. Cognitive decline accelerates once METS-IR exceeds the critical threshold of 27.78 (total cognitive score). This suggests that the synergistic effects of metabolic abnormalities—including dysregulated blood glucose, abnormal lipid profiles, and obesity—may exceed physiological compensation thresholds, initiating a vicious cycle of oxidative stress and mitochondrial dysfunction in the brain [45, 46]. Notably, the DSST turning point (28.92) is lower than other indicators, indicating that processing speed may serve as the most sensitive early warning signal for metabolic disorders. This aligns with neuroimaging findings, where changes in white matter integrity precede gray matter atrophy [47], suggesting that microvascular damage and axonal transport disruption may represent the initial events in cognition decline mediated by metabolic disorders. Subgroup analyses demonstrated that the detrimental association of METS-IR with cognitive function remained consistent across different population groups, indicating a generalizable relationship. Smoking and sleep disorders may modulate the strength of this association, a finding of significant clinical relevance. Among individuals who smoke, the association between increased METS-IR and decreased AFT scores was attenuated. This phenomenon may be partially attributable to the transient neurostimulatory effects of nicotine, which can temporarily enhance verbal fluency [48]. However, in the long term, smoking itself constitutes a neurotoxic factor that may promote cognitive impairment through the facilitation of vascular injury [49]. Therefore, the observed attenuation of this association among smokers should not be interpreted as evidence of a protective effect, but rather as the result of short-term effects or confounding factors. The association between METS-IR and AFT scores approached significance in individuals with sleep disorders (*P = * 0.0648), suggesting that sleep disorders may also serve as a potential effect modifier. Sleep disorders may amplify the neurotoxicity of metabolic disturbances through dual on one hand, sleep deprivation exacerbates insulin resistance and inflammatory responses [50, 51], creating a metabolic-sleep vicious cycle; on the other hand, reduced clearance function of the cerebrospinal Glymphatic system during sleep [52] may synergize with METS-IR-associated amyloid pathology. This underscores the importance of personalized interventions targeting high-risk populations, such as patients with metabolic syndrome complicated by sleep disorders, as they may yield superior cognitive protection benefits. Compared to previous studies that predominantly relied on single metabolic indices, this study is the first to introduce METS-IR into the field of cognitive research and explicitly demonstrate its association with cognitive decline while verifying the stability of this association. These findings hold significant clinical value for the early prevention and stratified management of cognitive decline. Although the observed effect sizes were modest, they remain clinically meaningful at the population level. Cognitive decline is a multifactorial process, and even small shifts in risk associated with metabolic dysregulation may accumulate over time and translate into substantial public health impact when extrapolated across aging populations. Importantly, this study does not propose METS-IR as a replacement for existing cognitive screening methods, but rather as a complementary indicator that may aid in risk stratification and serve as preliminary evidence to support subsequent mechanistic hypotheses. METS-IR reflects both glucose–lipid metabolism and abnormalities in body composition, providing a more comprehensive measure of the overall effects of metabolic disorders. In addition, it can be calculated from routine physical examination data and demonstrates stable discriminative ability, making it a practical supplementary tool for cognitive screening. For example, when METS-IR exceeds the threshold of 28 in elderly health examinations, this may warrant further cognitive assessment and consideration of metabolic management strategies to help mitigate cognitive decline. At the same time, it should be noted that the comprehensive nature of METS-IR inevitably reduces its specificity for identifying which single metabolic factor—such as TG, FBG, HDL-C, or BMI—is most strongly associated with the observed outcomes. However, the primary aim of this study is not to determine which individual factor is most important, but rather to assess whether overall metabolic imbalance is associated with cognitive decline. In metabolic–cognition research, these factors are often highly correlated, with strong collinearity and potential interactions. Analyzing them individually may lead to unstable results and biased interpretation, whereas a composite index provides a more robust way to capture the joint metabolic burden at the population level. This supports the use of METS-IR as a complementary first-line tool for risk stratification and threshold identification. Nevertheless, the biological mechanisms linking metabolic disorders to cognitive decline remain to be clarified. Future work that integrates experimental and clinical studies will be essential to unravel the interactions and pathways connecting these two processes, thereby offering deeper mechanistic insights. Importantly, this epidemiological study represents one component of a broader research program, establishing population-level associations and providing preliminary evidence to support the generation of subsequent experimental hypotheses. Future experimental studies are planned to directly examine whether targeted modulation of metabolic status is associated with improvements in cognitive outcomes. This layered strategy forms an integrated research narrative, spanning from epidemiological association to mechanistic investigation. Nevertheless, this study has several limitations. First, the cross-sectional design does not allow causal inferences, and longitudinal cohort studies will be required to confirm the stability and directionality of the associations observed here. In addition, to provide more clinical proof of concept, randomized controlled trials (RCTs) should ideally be designed around the key thresholds identified in this study to test whether targeted metabolic management can influence cognitive outcomes. Second, some covariates were self-reported, which may introduce bias. Furthermore, NHANES does not collect validated continuous measures of sleep quality, lifetime alcohol consumption beyond the past 12 months, or distinctions between current and former overweight. These limitations may have constrained the ability to fully capture long-term lifestyle exposures. In addition, NHANES does not provide clinical diagnoses of MCI or dementia; therefore, cognitive decline in this study was defined operationally based on cognitive test performance distributions. While this approach does not directly correspond to clinical diagnoses, it is commonly used in population-based studies and offers a practical way to capture relative cognitive variation. Moreover, although NHANES collects information on socio-economic status and some socio-economic factors may be related to cognitive outcomes, this study did not adjust for these variables in the primary models. The objective was to describe the overall population-level association between metabolic dysregulation and cognition, including socially mediated pathways that are relevant for screening and public-health burden. Adjusting for socio-economic factors would re-define the estimand to a direct effect that excludes such pathways and may attenuate the association through over-adjustment. This choice represents a trade-off, and future longitudinal studies and causal-modeling approaches that integrate socio-demographic factors may help further disentangle direct effects from socially mediated components. Third, the METS-IR formula does not include novel indices, such as visceral adiposity or inflammatory indices, which may also be relevant to cognition. Fourth, although the CERAD-WL, AFT, and DSST assessments, respectively, address memory, language fluency and executive function, as well as sustained attention and processing speed, and thus capture major dimensions of cognition to some extent, these measures remain limited in scope and do not encompass all key cognitive domains. The use of more comprehensive cognitive assessment tools will be important for achieving a finer characterization of cognitive structures and trajectories. Future research should consider expanding the dimensionality of cognitive testing to more fully capture the multifaceted nature of cognitive function. It is noteworthy that while diabetes, dyslipidemia, and overweight/obesity are typically regarded as late-stage consequences of chronic metabolic dysregulation, their underlying pathophysiological alterations often begin to manifest as early as middle age, when the prevalence of metabolic disorders is already substantial [53, 54]. This highlights the potential value of METS-IR as an early screening tool beyond older adults, offering opportunities to identify individuals at risk of cognitive impairment during midlife. Employing METS-IR in this age group for risk assessment could provide a critical window for primary prevention and early intervention against cognitive decline. However, the NHANES database includes cognitive assessment data only for individuals aged 60 years and above, precluding direct evaluation of the application of METS-IR in middle-aged cohorts in the present study. Future research is urgently needed to investigate the discriminatory performance and risk thresholds of METS-IR across different age groups, particularly in middle-aged individuals, to inform more comprehensive strategies for the early prevention and intervention of cognitive disorders. ## Conclusion This study highlights a significant connection between increased levels of METS-IR and cognitive decline in older adults. Furthermore, the study revealed that once METS-IR exceeds a specific threshold, its detrimental association with cognitive function becomes more pronounced. This suggests that the early identification and management of high-risk populations with metabolic disorders may play a crucial role in preventing mild cognitive impairment or dementia. Because METS-IR integrates dysglycemia, dyslipidemia, and adiposity, the composite metabolic dimension may relate more closely to low cognitive performance than single-dimension measures. These findings describe an understanding of the mechanisms underlying cognitive decline from an integrated metabolic perspective. METS-IR, as a composite index reflecting insulin resistance and metabolic disorders, offers advantages, such as accessibility, convenience, and low cost. In future clinical applications, it could function as a means of screening to identify individuals at heightened risk for cognitive decline. This would facilitate the development of stratified, multidimensional intervention plans tailored to at-risk populations, thereby maximizing cognitive protection. However, this research highlights the need for additional longitudinal studies to assess the stability and reliability of METS-IR as an associative indicator of cognitive decline, and to investigate its potential use alongside new biomarkers. Furthermore, subsequent studies should assess whether targeted improvement of metabolic status is associated with more favorable cognitive trajectories. Overall, the results underscore the importance of metabolic health in preserving brain health and provide a basis for developing screening and intervention strategies informed by METS-IR. ## Supplementary Information Supplementary Material 1.