Evaluate the Risk Factors of Mild Cognitive Impairment (MCI) in Patients with Type 2 Diabetes Mellitus (T2DM): A Cross-Sectional Study

Abstract

Objective

To explore the potential incremental value of metabolic-inflammatory and insulin-resistance indices for diagnosing mild cognitive impairment (MCI) in type 2 diabetes (T2DM), and to furnish new candidate tools and insights for the risk stratification and early recognition of diabetes-related cognitive dysfunction.

Methods

This cross-sectional, observational study was jointly conducted by two tertiary endocrine centers in northwestern China. A uniform assessment of cognitive status was performed using the Chinese versions of the Montreal Cognitive Assessment (MoCA). MCI was defined as MoCA < 26. Baseline clinical characteristics, metabolic-inflammatory parameters and insulin-resistance indices were systematically compared between groups. Candidate predictors were first compressed via Lasso regression; key retained variables were then used to develop a binary logistic prediction model. The model’s discrimination was assessed via ROC curves, and its calibration was evaluated by the Hosmer‑Lemeshow test combined with calibration plots. Decision-curve analysis (DCA) and clinical impact curves (CIC) were further employed to quantify net benefit and reclassification across a range of risk thresholds, providing a comprehensive appraisal of the model’s clinical utility.

Results

This study included 293 patients with type 2 diabetes mellitus, of whom 187 (63.8%) had mild cognitive impairment (MoCA<26). Compared with the 106 cognitively normal controls, the T2DM-MCI group were older, had fewer years of education and a lower proportion of higher education (all p<0.05). The prevalence of diabetic nephropathy (32.6% vs 18.9%, p = 0.006) and previous cerebral infarction (29.9% vs 10.4%, p < 0.001) was significantly higher, while diabetic peripheral neuropathy and carotid atherosclerosis showed borderline differences (p ≈ 0.06). Concurrently, inflammatory-metabolic burden increased across the board—NMLR, PHR, BRI, ABSI, WWI and LAP were all elevated (p ≤ 0.048)—and adiposity index RFM together with insulin-resistance markers TyG-WtHR and HOMA-IR also rose (p ≤ 0.017). Notably, Lasso regression identified WWI, RFM, and TyG-WtHR as candidate markers. However, none of these variables achieved statistical significance in the subsequent multivariable Logistic regression (all p>0.05). This finding suggests that novel anthropometric and metabolic indices may be linked to cognitive decline primarily through indirect pathways associated with traditional risk factors, offering new mechanistic insights into the obesity-metabolism-cognition axis.

Conclusion

Univariate analysis revealed that WWI, RFM, and TyG-WtHR were significantly correlated with cognitive decline (r=0.232, 0.194, and 0.154, respectively; all p<0.01). Incorporation of these three readily obtainable anthropometric and laboratory parameters into the baseline clinical model increased the area under the curve (AUC) from 0.768 to 0.771, offering potential clues into the mechanistic link between obesity-metabolism and cognitive impairment.

Introduction

Diabetes mellitus encompasses a range of metabolic disorders marked by chronic hyperglycemia, among which type 2 diabetes (T2DM) constitutes over 90% of cases, representing the predominant form.^1,2 Its insidious and slow course is driven by insulin resistance and progressive β-cell failure.^3,4 Once the compensatory hypersecretion of β-cells can no longer counteract insulin resistance, glycaemic homeostasis is broken and sustained hyperglycaemic load—via glucotoxicity, oxidative stress and inflammatory responses—induces functional and structural injury to large- and small-vessel endothelium, resulting in diabetic retinopathy, nephropathy and neuropathy.⁵ When this metabolic disorder extends to the central nervous system, it also damages the blood–brain barrier, triggers neuroinflammation and impairs β-amyloid clearance, leading to progressive decline in memory and executive function and ultimately evolving to mild cognitive impairment or even dementia.⁶

Mild cognitive impairment (MCI) represents a transitional phase from normal cognition to dementia, and its notable heterogeneity along with reversibility has drawn substantial research interest.⁷ According to the latest research data, the rate of MCI regressing to normal cognition (NC) within 5 years is as high as 31.3%. Over the subsequent five years, 48.1% remain NC, 32.7% return to MCI, and only 5.8% progress to dementia, underscoring MCI’s bidirectional trajectory.⁸ Even more concerning, patients with T2DM face a 1.25- to 1.91-fold higher risk of progressing from MCI to dementia compared with non-diabetic controls, with this accelerated deterioration detectable in the prediabetic stage.⁹ Therefore, it is crucial to identify and manage MCI in the T2DM population.

Systemic inflammation is a significant component of diabetes and is considered a primary contributor to cognitive dysfunction.¹⁰ The buildup of inflammatory mediators drives sustained microglial activation, leading to blood-brain barrier disruption and culminating in neuronal injury as well as cognitive deterioration.¹¹ Importantly, these cascading pathological events are quantifiable via blood biomarkers. A growing body of evidence demonstrates that elevated whole-blood-cell-derived inflammatory indices (CBCIIs) are significantly correlated with cognitive impairment in individuals with T2DM.^12–14 Given their low cost and widespread availability, these indices hold promise as supplementary tools for early cognitive screening in this population, potentially providing a readily accessible indicator of cognitive decline.

Obesity has become a global health threat that is closely linked to a variety of adverse outcomes. In addition to raising the risk of dementia, it can also induce early structural and functional brain abnormalities.¹⁵ Abdominal obesity was identified as an independent risk factor for young-onset dementia (YOD). Relative to women with a waist circumference below 75 cm, those with a waist circumference of 95 cm or greater exhibited a 55% higher overall risk of YOD, with this association being especially pronounced for vascular dementia.¹⁶ Building on this evidence, mid-life adiposity assessed by body-mass index reveals a similar pattern: entering the overweight range (BMI ≥ 25 kg/m²) during middle age will increase the risk of developing Alzheimer’s disease (AD), vascular dementia (VaD), and all-cause dementia by approximately 35%, 33%, and 26%, respectively.¹⁷ Meanwhile, evidence from certain researchers indicates that among elderly individuals, elevated BMI is linked to decelerated cognitive decline or a lower dementia incidence, implying that the impact of obesity on cognition follows a non-linear pattern.^18,19 Notably, emerging indices of adiposity and insulin resistance (such as BRI, WHtR, METS-IR, HOMA-IR etc) have demonstrated “threshold effects” or “non-linear platforms” in relation to cognitive impairment.^20–23 Therefore, future intervention strategies are expected to break away from the “one-size-fits-all” weight loss model and customize dynamic weight management plans for different populations.

Although previous evidence has linked visceral adiposity to cognitive decline, most studies focused on single age strata or isolated adiposity metrics, leaving a gap in sensitivity research among individuals with T2DM and MCI who are characterized by the continuous insulin resistance–visceral “fat–inflammation” axis. Therefore, this study integrates the TyG index, HOMA-IR, and multiple visceral adiposity parameters (VAI, LAP, CMI, etc) to construct a composite “glucose–lipid–inflammation” exposure profile, and employs a Lasso-Logistic framework to identify key predictive markers, aiming to improve early MCI detection and quantify the incremental diagnostic value of adiposity-metabolic indices for diabetes-related cognitive dysfunction.

Study Design and the Population

Patients and Selection Criteria

This was a multicenter, cross-sectional study. Between November 2024 and August 2025, 367 patients with T2DM were initially recruited from two specialized endocrine centers in China. Of these, 293 met the eligibility criteria and were included in the final analysis (Figure S1). Patients were included if they met the following criteria: I. aged 40–79 years; II. met the T2DM diagnostic criteria outlined in the Chinese Guidelines for the Prevention and Treatment of T2DM (2020 edition),²⁴ fasting blood glucose ≥7.0 mmol/L; random blood glucose or 2-hour blood glucose in an oral glucose tolerance test ≥11.1 mmol/L; glycated haemoglobin ≥6.5%. The remaining 74 patients were excluded based on the following criteria:¹ type 1 diabetes or other specific types of diabetes;² severe infection, or occurrence of acute diabetic complications (eg, diabetic ketoacidosis, hyperosmolar hyperglycemic state, hypoglycemic coma);³ central nervous system disorders affecting cognitive function, including traumatic brain injury, brain tumor, epilepsy, Parkinson’s disease, dementia, multiple sclerosis, or other inflammatory or structural diseases of the central nervous system;⁴ acute cardio-cerebrovascular events (including stroke or myocardial infarction), severe infection, or major surgery within the past 3 months;⁵ severe psychiatric disorders (eg, major depressive disorder, schizophrenia) or use of antipsychotics, sedative-hypnotics, or other medications with significant cognitive-impairing effects for ≥2 weeks within the past 3 months;⁶ endocrine or systemic disorders affecting cognition, including thyroid dysfunction, Cushing’s syndrome, active malignancy, severe anemia, or end-stage liver or renal disease;⁷ alcohol dependence, drug abuse, or long-term use of systemic glucocorticoids;⁸ severe visual, auditory, speech, or writing impairment that precludes completion of cognitive assessments;⁹ congenital intellectual disability, or any condition associated with lifelong cognitive impairment predating the diagnosis of diabetes.

Study Instruments

Demographic characteristics were recorded, including age, gender, education level, occupation, waist-to-hip ratio (WHR, calculated as waist circumference/hip circumference) and BMI (calculated as weight/height² (kg/m²)). BMI is categorised as normal (18.5–23.9 kg/m²), overweight (24–28 kg/m²) or obese (BMI ≥28 kg/m²). Furthermore, detailed medical histories of diabetic patients were carefully collected, including information on diabetes duration, current treatment, comorbid diseases (including coronary heart disease, fatty liver, hypertension, cerebral infarction, etc), smoking, drinking, ethnicity, and contact information. Blood samples were obtained to determine the following: fasting plasma glucose (FPG), glycosylated haemoglobin (HbA1c), the homeostasis model assessment of insulin resistance (HOMA-IR), serum creatinine, serum uric acid, total cholesterol (TC), triglycerides (TG), low-density lipoprotein (LDL-C) and high-density lipoprotein (HDL-C).

Cognitive performance was measured with the Chinese adaptation of the Montreal Cognitive Assessment (MoCA). Developed by Nasreddine et al²⁵ and later translated and revised by Wen et al,²⁶ the scale yields a total score of 30 points and covers seven domains: visuospatial/executive abilities, naming, attention, language, abstraction, delayed recall, and orientation. A score < 26 indicates MCI. All tests were administered by trained and certified researchers in a quiet room, and results were double-entered into the database on site. Cognitive function was assessed using the MoCA, a screening instrument rather than a diagnostic gold standard; the absence of comprehensive neuropsychological battery testing or clinical dementia evaluation may affect diagnostic precision.

Sample Size Calculation and Statistical Analysis

Sample size was estimated based on the association between the body shape index and cognitive impairment in T2DM. Using the formula n = Zα²/d² × PQ (Q = 1 − P), with α = 0.05 (Zα = 1.96), d = 0.15P and P = 43.2% derived from the literature, the minimum sample size was determined to be 270 participants.

All statistical analyses were carried out using SPSS 20.0 and R software (version 4.4.0). Continuous variables were tested for normality and homogeneity of variance; those meeting the assumptions were summarized as mean ± SD and compared using independent samples t-tests, whereas those not meeting the assumptions were reported as median (IQR) and analyzed with the Mann–Whitney U-test. Ordinal data were assessed using the rank-sum test, and categorical data were expressed as frequencies (percentages) and compared via the χ² test. A p-value < 0.05 (two‑tailed) was considered statistically significant. To identify relevant predictors, Lasso regression (glmnet package) was first applied to screen candidate variables, followed by binary logistic regression to determine independent risk factors, with results presented as odds ratios (ORs) and 95% confidence intervals (CIs). A nomogram was then developed using the nomogramFormula package. The discriminatory ability of the model was assessed via receiver operating characteristic (ROC) curve analysis, and several performance metrics—including the area under the ROC curve (AUC), likelihood ratio, optimal cut‑off point, sensitivity (SEN), specificity (SPE), and their 95% CIs—were computed (Figure 1).

Figure 1.

Study design and analytical workflow.

Notes: Starting from “T2DM with comorbid cognitive dysfunction (MoCA <26)”, candidate indicators were collected along four thematic axes: diabetic complications, inflammation, metabolism, and insulin resistance. A four-step statistical strategy—LASSO regression for initial screening, ROC-curve optimization, logistic-regression modelling, and model verification—was then applied to construct and validate a prediction model suitable for clinical use.

Abbreviations: NMLR, neutrophil-to-monocyte lymphocyte ratio; PHR, platelet-to-lymphocyte ratio; HbA1c, glycated hemoglobin A1c; WWI, weight-adjusted waist circumference index; RFM, relative fat mass index; FINS, fasting insulin; TyG-WtHR, triglyceride-glucose index × waist-to-height ratio; HOMA-IR, homeostasis model assessment of insulin resistance.

Results

Baseline Characteristics of the Study Population

This study enrolled 293 patients with T2DM, predominantly male (200, 68.3%), with a median age of 59 years. The 50–59 age-group accounted for the largest proportion (43.3%, 127/293), about 66.6% (231/293) had a disease duration ≥10 years, and the mean education duration was 12 years. Mild cognitive impairment was present in 63.8% (187/293). Overweight status (BMI 24.0–27.9 kg/m²) was recorded in 139 individuals (47.4%). Regarding occupation, 15 (5.1%) were unemployed, while 186 (63.5%) worked in agriculture or other sectors. Nearly 38.2% (112/293) reported a monthly income below CNY 3000. The characteristics of the subjects are detailed in Table 1 and classified according to the presence of cognitive dysfunction.

Table 1.

Comparison of Demographic Characteristics Between the NCF Group and the MCI Group

Characteristics	Total (n=293)	NCF (n=106)	MCI (n=187)	p value
Age (years)	59 (53, 65)	56.18±8.46	61 (55, 67)	<0.001**
Medical history	10.0 (5.0, 16.5)	9.5 (4.0, 15.0)	11.0 (5.0, 17.0)	0.072
Gender, n (%)				0.033*
Male	200 (68.3%)	80 (75.5%)	120 (64.2%)
female	93 (31.7%)	26 (24.5%)	67 (35.8%)
Smoking, n (%)	115 (39.2%)	45 (42.5%)	70 (37.4%)	0.357
Drinking, n (%)	90 (30.7%)	32 (30.2%)	58 (31.0%)	0.732
Living alone, n (%)	21 (7.2%)	9 (8.5%)	12 (6.2%)	0.518
Length of education (years)	12.0 (9.0, 14.0)	12.5 (11.0, 15.0)	11.0 (8.0, 12.5)	<0.001**
Height (cm)	168 (162, 173)	170 (164, 174)	166.83±7.87	0.064
Weight (kg)	68.21±10.58	68.86±10.27	67.84±10.77	0.171
BMI (kg/m2)
< 18.5	8 (2.7%)	2 (1.9%)	6 (3.2%)
18.5–23.9	120 (41%)	45 (42.5%)	72 (40.1%)	0.907
24.0–27.9	139 (47.4%)	50 (47.2%)	89 (47.6%)
≥ 28.0	26 (8.9%)	9 (8.5%)	17 (9.1%)
Waistline (cm)	90.73±8.45	89.52±8.03	91.41±8.62	0.064
Hipline (cm)	97 (93, 101)	96.44±5.33	98 (93, 102)	0.171
WHR	0.94 (0.89, 0.98)	0.93±0.05	0.94±0.05	0.136
Family history of diabetes, n (%)	152 (51.9%)	61 (57.5%)	91 (48.7%)	0.342
Work, n (%)
Farmer	64 (21.8%)	13 (12.3%)	51 (27.3%)
Employee	122 (41.6%)	61 (57.5%)	61 (32.6%)	<0.001**
Retirement	92 (31.4%)	27 (25.5%)	65 (34.8%)
Unemployed	15 (5.1%)	5 (4.7%)	10 (5.3%)
Monthly income, n(%)				0.002*
CNY<3,000	112 (38.2%)	29 (27.4%)	83 (44.4%)
CNY 3000–6000	101 (34.5%)	35 (33%)	66 (35.3%)
CNY6,000–10,000	55 (18.8%)	29 (27.4%)	26 (13.9%)

Notes: * p<0.05, ** p<0.001. Mean±SD represents the mean and standard deviation; Mean (P25, P75) indicates median and interquartile range; n (%) indicates frequency and percentage.

Abbreviations: NCF, Normal Cognitive Function group; MCI, Mild cognitive impairment group.

Relative to the cognitively normal group, the MCI group exhibited significantly higher rates of diabetic nephropathy (32.6% vs 18.9%, p = 0.006) and prior cerebral infarction (29.9% vs 10.4%, p < 0.001), whereas diabetic peripheral neuropathy and carotid atherosclerosis showed only borderline differences (p ≈ 0.06). In terms of glucose metabolism, individuals with MCI exhibited markedly higher fasting insulin concentrations than controls (10.80 [6.71, 17.75] mIU/L vs 7.35 [5.18, 13.55] mIU/L, p =0.017). Consistently, the HbA1c distribution shifted rightward: the proportion of participants in the intermediate glycaemic band [7.4–9.0] increased from 24.5% to 30.5%, accompanied by a reciprocal decline in the low [4.6–7.4] category (p = 0.027). No significant intergroup differences were found for any of the following parameters: fasting plasma glucose, fasting C-peptide, insulin antibody positivity, total cholesterol, HDL-C, and LDL-C (all p > 0.05) (Table S1).

Univariate analysis revealed that the MCI group carried a heavier inflammatory-metabolic burden, with NMLR, PHR, BRI, ABSI, WWI and LAP all markedly elevated (p ≤ 0.048). In contrast, the remaining inflammatory indices—NLR, dNLR, MLR, NHR, MHR, LHR and SIRI—did not differ significantly between the two groups (all p > 0.05). Regarding insulin resistance and metabolic composite indicators, the MCI group showed significantly higher values of both the HOMA-IR and the composite index of triglyceride-glucose-waist-to-height ratio (TyG-WtHR) (3.52 [2.04,6.17] vs 2.26 [1.55,40.02] and 5.0±0.69 vs 4.81±0.65, respectively, p ≤ 0.017). In contrast, the basic TyG index and other TyG-derived metrics (TyG-WC, TyG-BMI, TyG-WHR) showed no statistically significant differences between the two groups (p > 0.05) (Table S2).

Screening Important Covariates Using Lasso Regression

Multicollinearity among all independent variables was assessed using variance inflation factor analysis, which indicated a severe multicollinearity problem. Accordingly, we employed a Lasso regression approach for variable screening. This strategy serves to simplify the model while effectively addressing collinearity and overfitting concerns. Of the 29 variables initially included (covering clinical characteristics, metabolic-inflammatory parameters, and insulin resistance indices), 16 with non-zero coefficients were retained for further modeling. Figure 2A shows the mean squared error (MSE) vs Log(λ) in LASSO regression. To minimize MSE, λ=0.01671641, log (λ)= −4.091364 Figure 2B presents the regression coefficient vs Log(λ) curve.

Figure 2.

Construction and validation of the ROC-significance prediction model.

Notes: (A) Path diagram of regression coefficients, with λ=0.01671641 and log(λ)=-4.091364 determined to minimize mean squared error; (B) Cross-validation curve for lasso regression; (C) Nomogram of the ROC-significance model; Clinical decision curve (D) and cost–benefit ratios (E and F).

Standardized formula for the model: Ln=(p/1-p) = −0.8999 + Diabetic nephropathy × 0.3432 + Diabetic cerebrovascular disease × 0.9306 + Diabetic peripheral neuropathy × 0.1539 + Carotid atherosclerosis×0.1320-LAP × 0.1246 + TG × 0.0115 + NLR × 0.1760 + TyG - WC × 0.1895 + TyG-WtHR × 0.0101 + RFM × 0.1544+ WWI × 0.1515 + HOMAIR × 0.1448 + Age × 0.0234 - Length of education × 0.1658 + Insulin antibody ×0.0053+Disease duration×0.0069.

ROC Curve Analysis of Candidate Variables

The Lasso regression screened 16 candidate variables with non-zero coefficients (age, years of education, diabetic nephropathy, diabetic cerebrovascular disease, LAP, TG, TyG-WtHR, RFM, WWI, HOMAIR, TyG-WC, NLR, carotid atherosclerosis, diabetic peripheral neuropathy, insulin antibody, medical history). ROC curve analysis was performed for each of these variables. By integrating domain knowledge, the sensitivity, specificity, likelihood ratio, and AUC of the selected features were compared (Table 2).

Table 2.

ROC Curve Analysis

Variables	Cut-off Point	AUC (95%CI)/%	Sensitivity (95%CI)/%	Specificity (95%CI)/%	Likelihood Ratio	P
Education years(year)	12.5	0.690 (0.627–0.752)	76.47 (69.90–81.98)	50.00 (40.65–59.35)	1.529	<0.0001**
Age(year)	64.5	0.656 (0.591–0.721)	35.29 (28.80–42.38)	84.91 (76.88–90.49)	2.338	<0.0001**
WWI	12.17	0.643 (0.578–0.709)	5.882 (3.322–10.22)	99.06 (94.85–99.95)	6.235	<0.0001**
RFM	28.52	0.618 (0.551–0.684)	56.68 (49.52–63.58)	60.38 (50.86–69.17)	1.431	0.001*
Diabetic cerebrovascular disease	0.5	0.598 (0.533–0.663)	29.95 (23.84–36.86)	89.62 (82.37–94.11)	2.886	0.005*
TyG-WtHR	6.399	0.587 (0.520–0.654)	37.97 (31.32–45.10)	75.47 (66.49–82.68)	1.548	0.013*
Diabetic nephropathy	0.5	0.569 (0.502–0.638)	32.62 (26.31–39.63)	81.13 (72.65–87.44)	1.729	0.051*
NLR	5.005	0.566 (0.497–0.634)	2.674 (1.15–6.11)	98.11 (93.38–99.66)	1.417	<0.062
TyG-WC	1031	0.564 (0.495–0.632)	6.417 (3.71–10.88)	97.17 (92.01–99.23)	2.267	0.701
Medical history(year)	25.5	0.563 (0.494–0.632)	3.743 (1.83–7.52)	98.11 (93.38–99.66)	1.984	0.072
HOMA-IR	54.74	0.563 (0.495–0.632)	2.139 (0.84–5.37)	99.06 (94.85–99.95)	2.267	0.072
LAP	35.23	0.563 (0.493–0.632)	60.22 (53.04–66.97)	50.00 (40.56–59.44)	1.204	0.076
Insulin antibodies	22.87	0.557 (0.491–0.624)	10.70 (7.03–15.94)	98.11 (93.38–99.66)	5.668	0.103
Carotid atherosclerosis	0.5	0.538 (0.469–0.608)	85.03 (79.21–89.43)	22.64 (15.71–31.48)	1.099	0.276
Diabetic peripheral neuropathy	0.5	0.537 (0.467–0.607)	86.63 (81.01–90.78)	20.75 (14.12–29.43)	1.093	0.294
TG	1.355	0.534 (0.463–0.605)	53.76 (46.59–60.78)	50 (40.56–59.44)	1.075	0.335

Notes: *p<0.05,**p<0.001.

Abbreviations: WWI, weight-adjusted waist circumference index; RFM, relative fat mass index; TyG-WtHR, triglyceride-glucose-waist-to-height ratio index; NLR, neutrophil‑to‑lymphocyte ratio; TyG‑WC, triglyceride‑glucose‑waist circumference index; HOMA‑IR, homeostasis model assessment of insulin resistance; LAP, lipid accumulation product.

Among the 16 candidate indicators, years of education demonstrated the strongest discriminative performance, with an AUC of 0.69 (95% CI: 0.627–0.752), suggesting that lower educational attainment offers broad early-warning coverage for cognitive impairment. The weight-adjusted waist circumference index (WWI) ranked second, achieving an extremely high specificity of 99.1% and serving as a “red-line” marker; however, its sensitivity was only 5.9%, making it more suitable for confirmation rather than initial screening. The combination of these two indicators is expected to maintain high specificity while improving early detection efficiency.

Age, RFM, and TyG-WtHR formed a “second-tier” group, with AUCs between 0.59 and 0.66, offering moderate overall accuracy and complementary sensitivity profiles that can be used interchangeably across different clinical settings.

The remaining insulin-resistance and metabolic composite indices (HOMA-IR, LAP, TyG-WC, etc) and the neutrophil-to-lymphocyte ratio (NLR) achieved AUCs of approximately 0.563–0.566, values that were not statistically significant; their incremental clinical utility appears limited, and they are better employed as mechanistic exploratory variables rather than standalone screening tools.

Logistic Regression Handles the Task of Binary Classification

For logistic regression analysis, the dependent variable was dichotomized, assigning a value of 1 to cases with cognitive dysfunction and 0 to those without. The significant factors (years of education, age, WWI, insulin antibodies, disease duration, LAP, and RFM) were selected using the “enter” method for logistic regression analysis.

The model revealed that age significantly contributed to the risk of MCI (OR=1.038, 95%CI: 1.002–1.075, p=0.036). Diabetic cerebrovascular disease also markedly increased the risk (OR=3.128, 95% CI: 1.455–6.724, p=0.003), whereas years of education acted as a protective factor (OR=0.825,95%CI: 0.760–0.894, p<0.001). The association between diabetic nephropathy and cognitive impairment approached statistical significance (p=0.058). Although TyG-WtHR, WWI, and RFM did not reach significance in the current model (p>0.05), these variables were retained to safeguard the integrity of forthcoming cohort analyses and to leave room for detecting non-linear or threshold effects in larger, long-term datasets. Consequently, the study incorporated indicators significant on ROC analysis into a binary logistic regression (Table 3), and constructed the following model equation: Ln(p/1-p) = −3.38 -0.193×education years +1.140×Diabetic cerebrovascular disease +0.038×Age+0.623×Diabetic nephropathy +0.180×TyGWtHR +0.211×WWI +0.012×RFM.

Table 3.

Model Comparison: Binary Logistic Regression Results from the ROC-Significance vs Clinical-Only Models

Variable	ROC Significance Model				Clinical Only Model
	B	Wald	OR (95%CI)	p	B	Wald	OR (95%CI)	p
Education years(year)	−0.193	21.807	0.825 (0.760–0.894)	<0.001**	−0.208	26.386	0.812 (0.750–0.879)	<0.001**
Diabetic cerebrovascular disease	1.140	8.527	3.128 (1.455–6.724)	0.003*	1.203	9.563	3.329 (1.553–7.133)	0.002*
Age(year)	0.038	4.398	1.038 (1.002–1.075)	0.036*	0.04	5.763	1.041 (1.007–1.075)	0.016*
Diabetic nephropathy	0.623	3.595	1.865 (0.979–3.551)	0.058*	0.646	4.031	1.909 (1.015–3.588)	0.045*
TyGWtHR	0.180	0.492	1.197 (0.724–1.981)	0.483	–	–	–	–
WWI	0.211	0.480	1.236 (0.679–2.247)	0.488	–	–	–	–
RFM	0.012	0.216	1.012 (0.961–1.066)	0.642	–	–	–	–
log-likelihood ratio	316.441				319.803
Omnibus test	χ2=67.057, p<0.001**				χ2=63.694, p<0.001**
HL test	χ2=2.076, p=0.979				χ2=3.755, p=0.0.879
AUC (95%CI)	0.771 (0.715–0.826)				0.768 (0.712–0.824)
Sensitivity	44.9%				46.5%
Specificity	87.7%				86.8%

Notes: **p < 0.001, *p < 0.05.

Abbreviations: TyGWtHR, triglyceride glucose-waist height ratio index; WWI, weight-adjusted waist index; RFM, relative fat mass index; OR, odds ratio; CI, confidence interval; AUC, area under the curve.

Compared with the clinical-only model, education years, diabetic cerebrovascular disease and age remained independent predictors (p<0.05) in the ROC-significance model, with virtually unchanged effect directions and magnitudes, underscoring the robustness of these clinical variables. After adding TyG-WtHR, WWI and RFM, the ROC-significance model retained the significance of the clinical triad while integrating insulin resistance, central adiposity and lipid accumulation into the same equation, yielding a more complete representation of the “glucose–lipid–inflammation” axis. Multivariable analysis revealed that none of the three added indices achieved independent significance (all p>0.05), suggesting their predictive effects may be explained by conventional risk factors or multicollinearity. AUC rose from 0.768 to 0.771 and specificity increased from 86.8% to 87.7%, while Hosmer-Lemeshow calibration remained excellent (p=0.979). This model-level increment stemmed not from independent contributions of individual metabolic-obesity indices, but likely from synergistic optimization between metabolic-obesity dimensional information and the clinical triad, or from overall improvement in model fit (likelihood-ratio test χ²=316.441, p<0.001). The enhanced model improves false-positive control without sacrificing calibration, offering a candidate risk-stratification tool for subsequent external validation.

Based on the ROC-significance model, we generated a Nomogram using the nomogramFormula package in R, with corresponding scales for variables such as years of education, age, and WWI, representing the acceptable value ranges of each variable. For example, consider a patient with a TyG-WtHR of 4.56, RFM of 25.32, and WWI of 10.24, who has diabetic nephropathy, diabetic cerebrovascular disease, is 54 years old, has 11 years of education, and has a total score of 337. The estimated risk of cognitive impairment for this patient is 83.0% (Figure 2C).

ROC-Significance Model Test

Predicted probabilities from the ROC-significance model closely matched observed outcomes, as illustrated by the calibration curve, with a mean absolute error (MAE) of 0.017 and a mean squared error (MSE) of 0.00044, suggesting excellent calibration and strong predictive performance. Decision curve analysis further revealed that the model delivers maximal net clinical benefit within the 0.2–0.3 risk-threshold window, where the cost–benefit ratio is most favorable (≈1:4 -1:5). These findings suggest that intervention at this threshold is both clinically effective and economically efficient, making it particularly suitable for resource-constrained settings (Figure 2D–F).

Discussion

This study developed and validated a T2DM cognitive-impairment risk model that integrates metabolic, inflammatory, and clinical dimensions through a three-step “LASSO-ROC-Logistic” pipeline. Compared with the baseline model containing only traditional clinical variables, the new model raised the AUC to 0.771 (95% CI 0.715–0.826), a 0.003 increment (DeLong p = 0.02), while retaining excellent calibration and cost-effectiveness, thereby providing a quantitative basis for individualized follow-up schedules and early-intervention intensity.

Using a MoCA cutoff of <26, the present study found a 63.8% prevalence of cognitive impairment in patients with type 2 diabetes—higher than the 48% pooled prevalence (95%CI 40–55%) reported in Chen et al’s 2023 meta-analysis of 17 Chinese studies (n = 4380).²⁷ This difference may be due to the longer disease duration of this cohort (average 10 years) and poor blood sugar control (HbA1c 8.5% ± 2.1%), as long-term high blood sugar aggravates cerebrovascular damage and neurodegeneration, thereby increasing the prevalence of MCI. Notably, MCI was present in 60.4% of the cohort, underscoring that early cognitive decline is already widespread among “relatively young” older adults with diabetes.

Our results revealed 18 variables that were significantly associated with cognitive impairment; four of these are intrinsically fixed—advancing age, female sex, fewer years of schooling, and the structural sequelae of prior stroke—echoing patterns observed in previous reports.^26–28 Even after excluding age and education, about one in twelve patients with diabetes remain in a high-risk state that cannot be modified by any intervention because of their sex and history of stroke. At the “interventionable” level, this study focused on 5 core variables: diabetic cerebrovascular disease and diabetic nephropathy serve as the classic levers for target-organ protection, while TyG-WtHR, WWI, and RFM constitute dynamic endpoints of “glucolipid–adipose–distribution”. A dual vascular-renal protection strategy—combining blood-pressure reduction with RAAS blockade, GLP-1RA, and SGLT2 inhibitors—paired with diet-and-exercise-induced redistribution of adiposity offers a practical and measurable pathway for improving outcomes in patients with T2DM and early cognitive decline.^29,30

Among the inflammatory indices readily available from routine complete blood count, only NMLR and PHR were significantly elevated, indicating that the MCI group exhibited an inflammatory-metabolic shift characterized by myeloid-cell predominance, relative lymphocyte depletion, platelet activation, and a concurrent decline in HDL-C antioxidant capacity. A similar pattern has been observed in older adults with sepsis-associated delirium: both MLR and NMLR rise stepwise with increasing inflammatory burden.³¹ Thus, blood-count-derived inflammatory indices play a pivotal role in flagging individuals at high risk of cognitive decline. Furthermore, this study also revealed that body composition disorders (increased BRI, WWI, LAP, and RFM) and insulin resistance (increased TyG-WtHR index and HOMA-IR) were all significantly associated with cognitive impairment in patients with T2DM. When these two metabolic abnormalities coexist with chronic low-level inflammation, they may synergistically amplify oxidative stress and vascular endothelial damage, further accelerating the deterioration of brain microcirculation and the functional connections of neurons.

There are some limitations to this study. Our strategy of applying Lasso regression to screen for variables would require excluding a large fraction of our different data, which may limit the breadth of our conclusions. As this study is a single-center study with a modest sample size, independent external datasets are lacking to validate the proposed model. Consequently, pending external validation, these indices represent a promising scalable approach rather than a fully validated clinical tool. As the majority of participants were middle-aged or older, the extent to which these findings can be generalized to younger populations remains uncertain. Future studies could be predicted by adopting age-matched cohorts and larger datasets.

In summary, this study systematically mapped out reference ranges for inflammatory, adiposity and insulin-resistance indices in a T2DM population and constructed a hierarchical cognitive-risk assessment system centered on “education years→WWI→age/RFM/TyG-WtHR”. To our knowledge, this is the first study to simultaneously evaluate body composition indices, metabolic indicators, and insulin resistance markers in a T2DM-MCI population using LASSO-logistic regression. By providing standardized and reproducible biological data, the framework offers a promising scalable approach for progressive screening and precision intervention across diverse clinical settings.

Funding Statement

This work was supported by the Major Research Plan of National Natural Science Foundation of China [81670736, Jianfang Fu].

Data Sharing Statement

The datasets generated during and analyzed during the current study are available from the corresponding author on reasonable request.

Ethics Statement

This study complies with the Declaration of Helsinki. This study was respectively approved by the Medical Ethics Committee of the First Affiliated Hospital of Air Force Medical University and the Medical Ethics Committee of the 986^th Hospital of the Air Force (Grant Number: XJLL-KY-20252059, KYLL2025-986-188). All patients who matched the criteria were willing to participate and completed an informed consent form.

Author Contributions

Yanning Cao: Conceptualization, Methodology, Writing – original draft, Writing – review & editing; Cong Lei: Conceptualization, Methodology, Formal analysis, Writing – original draft; Jianhui Liu: Investigation, Data curation,Writing – review & editing; Nan Wu: Investigation, Data curation,Writing – review & editing; Qiaoqiao Zhang: Investigation, Data curation, Writing – review & editing; Xiaohang Zuo: Investigation, Data curation, Writing – review & editing; Wenli Lei: Investigation, Data curation, Writing – review & editing; Na Wang: Investigation, Data curation, Writing – review & editing; Lamei Xu: Investigation, Data curation, Writing – review & editing; Peijun Mao:Conceptualization, Methodology, Writing – original draft, Writing – review & editing; Jie Li: Conceptualization, Methodology, Writing – review & editing; Jianfang Fu: Conceptualization, Methodology, Formal analysis, Writing – original draft, Writing – review & editing,Funding acquisition, Project administration.

All authors gave final approval of the version to be published; have agreed on the journal to which the article has been submitted; and agree to be accountable for all aspects of the work.

Disclosure

Yanning Cao, Nan Wu, Qiaoqiao Zhang, Xiaohang Zuo, Jie Li and Jianfang Fu report grants from National Natural Science Foundation of China. The authors declare they have no other conflicts of interest.