Association of glycolipid metabolism 6 factors index with cardiovascular and metabolic diseases, and its predictive value in disease progression Authors: Jiahui Qi, Lingyu Liu, Yijin Pan, Yihan Wang, Xitong Xu, Haipeng Wang, Yiyan Li, Qian Liu, Songwen Liu, Chen Li, Yuqing He Published: 2026 ABSTRACT BackgroundThe study aims to investigate the potential of glycolipid metabolism 6 factors (GLM6) and glycolipid metabolism 7 factors (GLM7) indexes in identifying high-risk populations for various diseases during routine screening, and to evaluate the potential of GLM6 for long-term dynamic monitoring of disease risk.MethodsResearch data were obtained from the National Health and Nutrition Examination Survey (NHANES) and the China Health and Retirement Longitudinal Study (CHARLS). Logistic regression, Cox regression, and restricted cubic spline (RCS) curve analysis were employed to examine correlations between exposure variables and outcome variables. The eXtreme Gradient Boosting (XGBoost) machine learning algorithm was used to establish models evaluating the predictive efficacy of GLM7 and GLM6 indicators. Subgroup analyses were conducted based on age, gender, smoking status, and drinking status.ResultsThe composite indicators GLM7 and GLM6 showed significant positive correlations with the risk of 9 disease categories, 14 specific diseases, and mortality, with no association observed only for hyperthyroidism. The XGBoost model built on GLM6 performs comparably to the GLM7 model in terms of predictive ability for most diseases and has the potential to identify high-risk populations for various diseases during routine examinations. GLM7 and GLM6 demonstrate sound predictive capabilities for cardiovascular categories and metabolic diseases in the XGBoost model. In the cohort, the baseline GLM6 index, cumulative GLM6 index, and GLM6 dynamic clustering group were significantly positively correlated with cardiovascular disease (CVD) and diabetes (DM). The highest tertile of the cumulative GLM6 index, combined with an unfavorable GLM6 dynamic clustering group, was associated with the highest risk of new-onset CVD and DM, demonstrating a high risk of CVD and DM associated with long-term high GLM6 exposure. Furthermore, female exhibited higher disease risk under prolonged high GLM6 exposure.ConclusionsGLM6 and GLM7 show potential as biomarkers for identifying individuals at high risk of CVD and DM in large-scale population screenings. They can help establish long-term disease risk monitoring networks, and following future multicenter validation studies, they are expected to support the optimization of clinical decision-making and promote the targeted allocation of public health resources.Clinical trial numberNot applicable.Supplementary InformationThe online version contains supplementary material available at https://doi.org/10.1186/s12902-026-02385-w. Background Noncommunicable diseases are the leading cause of the disease burden in most countries [1]. The World Health Organization reports that they account for approximately 75% of all non-pandemic-related deaths worldwide and has set a target to reduce premature deaths from these diseases by one-third by 2030 [2, 3]. Data from the Global Burden of Disease 2021 study indicate that the burden of noncommunicable diseases is on the decline, but they remain a major global health issue. Cardiovascular disease (CVD) accounts for the largest proportion of deaths, followed by cancer, chronic respiratory diseases, and diabetes (DM) [4]. How to identify risks early and implement tiered interventions within the general population has become a central issue in public health and clinical practice. Regular health checkups are an effective way to promote health and identify noncommunicable diseases at an early stage. As the most comprehensive and consistent source of health data, routine physical examinations provide unique value in screening for and predicting disease risks [5–7]. However, given the continuing severe burden of noncommunicable diseases, there is a lack of research integrating routine health examination data into predictive models. Repeated testing of relevant indicators during disease diagnosis and screening not only increases patients’ medical expenses but also wastes healthcare resources [8]. In response to this situation, Wang’s team integrated large-scale epidemiological data (covering 49 conventional indicators) to establish a novel set of indicators—the glycolipid metabolism 7 factors (GLM7)—capable of improving the accuracy of disease management. This demonstrated the value of conventional health examination indicators in disease diagnosis and prediction; however, their application remains limited in terms of disease classification and the analysis of specific subtypes [9]. It only examined the association between the GLM7 index and CVD, DM, liver disease, cancer, and comorbidities (including renal failure, chronic bronchitis, thyroid disorders, and rheumatic diseases), and a comprehensive analysis of the relationship between GLM7 and these diseases is still lacking. Building on this foundation, this study expanded the range of diseases covered to further investigate the association between GLM7 and 9 disease categories and 15 specific diseases. It also included an analysis of the association between GLM7 and mortality risk to comprehensively evaluate the potential of the GLM7 biomarker for screening high-risk populations for multiple diseases. In addition, among the routine indicators included in the GLM7, insulin levels may be missing from health examination data in most general population cohorts and large-scale cohort studies. For example, the health examination data from the China Health and Retirement Longitudinal Study (CHARLS) database used in this study, which comprises cross-sectional population data, did not include insulin measurements. At the same time, the cost of insulin testing is higher than that of the other six markers. Therefore, this study attempted to optimize the selection of markers, retaining only age, body mass index (BMI), fasting blood glucose (FBG), triglycerides (TG), low-density lipoprotein cholesterol (LDL-C), and high-density lipoprotein cholesterol (HDL-C) as the glycolipid metabolism 6 factors (GLM6). The GLM6 index requires fewer routine laboratory tests and uses more commonly performed ones, while also reducing testing costs by nearly half, thereby alleviating the financial burden associated with these costs. Compared to traditional indicators, the GLM7 and GLM6 composite indicators proposed in this study are comprehensive measures that assess disease risk from multiple perspectives and dimensions, and are superior to single predictive indicators. Disease risk prediction models based on routine health examination indicators can identify abnormal values before individuals are even aware of any health issues, thereby enabling “early detection, early diagnosis, and early intervention”. At the same time, the multidimensional indicators used in the study encompass various factors such as age, blood glucose, and blood lipid levels, offering the potential to assess an individual’s overall risk of developing multiple chronic diseases and avoiding the fragmented approach of screening for individual diseases. The multidimensional indicators readily obtained through routine physical examinations help establish a dynamic monitoring network for population health data, providing a basis for assessing disease trends and adjusting strategies. Materials and methods Data sources and study population This study utilized cross-sectional population data obtained from the National Health and Nutrition Examination Survey (NHANES) and the CHARLS database. The NHANES dataset, used as the training set, included population data from 1999 to 2018. Mortality follow-up data for this population were obtained from the National Center for Health Statistics, with follow-up extending through 2019. The CHARLS dataset was used as the validation set, incorporating data from the 2011 and 2015 survey cohorts. The study included a total of 23,829 participants from NHANES and 18,550 participants from CHARLS for the cross-sectional analysis (Fig. 1A). The initial number and final number of participants included in each survey round of NHANES and CHARLS are shown in Supplementary Table 1. All missing data for outcomes included in the dataset are listed in Supplementary Table 2. When analyzing any of these outcomes, the corresponding missing data were excluded before performing the analysis. It should be noted that NHANES and CHARLS differ in terms of population composition, age distribution, laboratory tests, disease diagnosis methods, and follow-up design. NHANES is a nationally representative survey in the United States that includes a multiracial population, whereas CHARLS primarily covers middle-aged and older adults in China. Therefore, this study uses CHARLS as an independent external validation cohort to assess the transferability of the GLM6 model across heterogeneous populations, rather than conducting a strict replication validation under identical disease definitions and population structures. Fig. 1Data filtering flowchart. (A) Cross-sectional study; (B) Cohort study Data filtering flowchart. (A) Cross-sectional study; (B) Cohort study The cohort data included in this study were obtained from the CHARLS database. Data from all five CHARLS surveys were included, with 2011 serving as the baseline year and the follow-up period ending in 2020. The cohort comprised a total of 5,112 participants from CHARLS, of whom 4,633 were followed up for CVD outcomes and 3,974 for DM outcomes (Fig. 1B). The NHANES database received ethical approval from the Institutional Review Board of the National Center for Health Statistics, and the CHARLS database received approval from the Ethics Committee of Peking University (IRB00001052-11015). All participants provided written informed consent before enrollment in the study [10, 11]. The exclusion criteria were defined as: (1) age under 18 years; (2) Missing data for GLM7 (Age, BMI, FBG, insulin, TG, LDL-C, and HDL-C) or GLM6 (Age, BMI, FBG, TG, LDL-C, and HDL-C) indicators; (3) Missing data on relevant diseases. Exposed variables The exposure variables in the cross-sectional study were the GLM7 and GLM6 indices. In the longitudinal cohort, the exposure variables were the GLM6 index calculated from baseline measurements, the GLM6 index representing cumulative exposure assessed by integrating baseline and third-survey data, and the GLM6 dynamic cluster grouping. All exposure variables calculated in the study are dimensionless measures. All exposure variable values in the study were standardized using Z-scores. The GLM7 indicator is calculated using the following formula [9]:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned}&\mathrm{GLM7}\mathrm{=}\mathrm{log}10\cr &\left(\frac{\begin{aligned}&\mathrm{A}\mathrm{g}\mathrm{e}\left(\mathrm{y}\mathrm{e}\mathrm{a}\mathrm{r}\mathrm{s}\right)\times\:\:\mathrm{B}\mathrm{M}\mathrm{I}\left(\mathrm{k}\mathrm{g}{/ \mathrm{m}}^{2}\right)\\ &\times\:\mathrm{F}\mathrm{B}\mathrm{G}\left(\mathrm{m}\mathrm{g}/ \mathrm{d}\mathrm{L}\right)\times\:\:\mathrm{I}\mathrm{n}\mathrm{s}\mathrm{u}\mathrm{l}\mathrm{i}\mathrm{n}\left(\mathrm{p}\mathrm{m}\mathrm{o}\mathrm{l}/ \mathrm{L}\right)\\ & \:\times\:\:\mathrm{T}\mathrm{G}(\mathrm{m}\mathrm{m}\mathrm{o}\mathrm{l}/\mathrm{L})\:\times\:\:\mathrm{L}\mathrm{D}\mathrm{L}-\mathrm{C}(\mathrm{m}\mathrm{m}\mathrm{o}\mathrm{l}/ \mathrm{L})\end{aligned}}{\mathrm{H}\mathrm{D}\mathrm{L}-\mathrm{C}(\mathrm{m}\mathrm{m}\mathrm{o}\mathrm{l}/\mathrm{L})}\right)\end{aligned}$$\end{document} Based on the calculation formula for the GLM7 indicator, we have developed the calculation formula for the GLM6 indicator. The specific formula is as follows:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned}&\mathrm{GLM6}\mathrm{=}\mathrm{log}10\cr &\left(\frac{\begin{aligned}&{}\mathrm{A}\mathrm{g}\mathrm{e}\left(\mathrm{y}\mathrm{e}\mathrm{a}\mathrm{r}\mathrm{s}\right)\times\:\:\mathrm{B}\mathrm{M}\mathrm{I}\left(\mathrm{k}\mathrm{g}{/ \mathrm{m}}^{2}\right)\\ & \times \mathrm{F}\mathrm{B}\mathrm{G}\left(\mathrm{m}\mathrm{g}/ \mathrm{d}\mathrm{L}\right) \:\times\:\:\mathrm{T}\mathrm{G}(\mathrm{m}\mathrm{m}\mathrm{o}\mathrm{l}/\mathrm{L})\\ &\times\:\:\mathrm{L}\mathrm{D}\mathrm{L}-\mathrm{C}(\mathrm{m}\mathrm{m}\mathrm{o}\mathrm{l}/ \mathrm{L})\end{aligned}}{\mathrm{H}\mathrm{D}\mathrm{L}-\mathrm{C}(\mathrm{m}\mathrm{m}\mathrm{o}\mathrm{l}/\mathrm{L})}\right)\end{aligned}$$\end{document} Prior to external validation using NHANES and CHARLS data, this study standardized the components of GLM6. For the external validation, only indicators available in both databases were retained, including age, BMI, FBG, TG, LDL-C, and HDL-C. After standardizing the units of the laboratory indicators, GLM6 was calculated using the same formula. Since insulin levels were not measured in the CHARLS database, the external validation was conducted only for the GLM6 model; the GLM7 model was not subjected to external validation. The study estimated cumulative exposure levels by multiplying the mean values of multi-time-point indicators by the exposure time window. It referenced the calculation paradigm for cumulative exposure indicators used in existing studies and made adjustments to align with the follow-up time points specified in this study. Exposure variables in the longitudinal cohort include GLM6 measures calculated at baseline, GLM6 measures reflecting cumulative exposure levels assessed by integrating baseline and third-survey data, and GLM6 control status trajectories. Calculate the cumulative GLM6 index using the following formula [12]:\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{aligned}&\text{Cumulative GLM6}\cr & \quad=\:\frac{\text{2011 GLM6 + 2015 GLM6}}{\mathrm{2}}\cr & \quad \times\text{Times (2015-2011)}\end{aligned}$$\end{document} The study used the standardized GLM6 values from 2011 to 2015 as clustering variables and employed the k-means clustering method to identify dynamic clustering patterns of the GLM6 index over time. The k-means clustering method determined the optimal K value using the elbow method, selecting K = 3 based on the inflection point of the weighted sum of squares (WSS), and repeated the initialization 50 times to optimize the stability of the results (Supplementary Table 3) [13]. The study employed three methods—Bootstrap resampling, repeated consistency tests, and perturbation tests—to assess clustering stability, and conducted cross-validation using hierarchical clustering, demonstrating that the clustering scheme with K = 3 exhibits good robustness and reproducibility (Supplementary Table 4, Supplementary Figure 1). The study identified three distinct GLM6 dynamic clustering groups (Supplementary Figure 2) [14]: (1) Favorable GLM6 dynamic clustering group: The GLM6 level was low at baseline and subsequently rose slightly, indicating well control; (2) Intermediate GLM6 dynamic clustering group: The GLM6 level was moderate at baseline and subsequently rose slightly, indicating moderate control; (3) Unfavorable GLM6 dynamic clustering group: The GLM6 level was high at baseline and subsequently fell slightly, indicating not well control. Outcome variables The outcome variable in this cross-sectional study is the prevalence of various diseases, with specific diseases and classification definitions provided in Supplementary Table 5. The major disease categories are divided into nine types: cardiovascular category, endocrine and metabolic diseases, respiratory diseases, hepatobiliary diseases, renal and urological diseases, cancer, mental and psychological diseases, musculoskeletal diseases, and other diseases. The individual diseases include 15 types: DM, metabolic syndrome, metabolic dysfunction-associated fatty liver disease (MAFLD), non-alcoholic fatty liver disease (NAFLD), coronary heart disease (CHD), hypertension, heart failure, heart attack, angina, gout, peripheral artery disease (PAD), stroke, liver disease, hypothyroidism, and hyperthyroidism. In addition, the study included all-cause mortality and cardiovascular mortality among participants in 2019 from the NHANES dataset as outcome variables. In the external validation, this study analyzed NHANES and CHARLS by mapping highly comparable outcomes to broader disease categories, such as cardiovascular category and metabolic diseases. It should be noted that the methods used to determine outcomes in the two databases are not entirely consistent. In NHANES, some metabolic disease outcomes are based on laboratory tests, physical examinations, medication information, and self-reported physician diagnoses, whereas in CHARLS, some chronic disease outcomes—particularly CVD and certain other chronic conditions—rely primarily on self-reported physician diagnoses. The outcome variables CVD and DM in the cohort study represent the disease types in the optimal tier for predictive performance in cross-sectional studies. For the first three surveys and subsequent follow-up surveys in the cohort, the outcome variable CVD was defined as meeting any one of the following criteria: (1) Self-reported diagnosis of heart disease by a doctor; (2) Self-reported diagnosis of stroke by a doctor. For the first three surveys in this cohort, diabetes was defined as meeting any one of the following criteria: (1) Self-reported physician diagnosis of DM; (2) Fasting plasma glucose ≥ 126 mg/dL; (3) HbA1c ≥ 6.5%; (4) Use of DM medications. Correlation analysis The study used logistic regression to analyze the linear associations between GLM7, GLM6, the cumulative GLM6 index, and GLM6 control status trajectories and disease risk, presenting the results as odds ratios (OR) and their 95% confidence intervals (CI) [15]. The study used Cox regression and Kaplan-Meier (KM) curves to examine the linear relationship between the GLM6 and GLM7 indices and the risk of all-cause mortality and cardiovascular mortality, presenting the results as hazard ratios (HR) and their 95% CI [16, 17]. Nonlinear associations between all exposure variables and outcome variables were analyzed using restricted cubic spline (RCS) curves with the “rms” R package [18, 19]. The association analysis employed hierarchical modeling based on different covariate adjustment strategies, with detailed covariate adjustments presented in Supplementary Table 6. Model 2 was selected as the base model to avoid over-adjustment for potential mediating or common pathway variables. Model 3, which included additional covariates, was used as a supplementary sensitivity analysis to assess the robustness of the associations after further adjustment for these potential risk factors. Building disease risk prediction models The study used the eXtreme Gradient Boosting (XGBoost) machine learning algorithm to develop a model for predicting disease risk. The GLM6 variable set includes Age, BMI, FBG, TG, LDL-C, and HDL-C; GLM7 adds insulin to the GLM6 set. After screening the complete cases based on the corresponding outcome variable, the sample was randomly divided into a training set and a validation set in a 7:3 ratio. Model hyperparameter tuning is performed on the training set, using cross-validation to evaluate candidate hyperparameter combinations. After optimization, performance metrics are calculated for both the training and validation sets, including the area under the curve (AUC), accuracy, sensitivity, and specificity [20–22]. The study evaluated model calibration using relevant calibration metrics (Brier score, calibration intercept, calibration slope, and P value of the Hosmer-Lemeshow test). Decision curve analysis (DCA) curves were used as a supplementary analysis to assess the performance of some models. Class imbalance was addressed using the SMOTE method, which was applied only to the training set or within cross-validation folds. The test set retained its original class distribution to avoid bias in performance evaluation. To prevent data leakage, the test set was not included in feature selection, resampling, model hyperparameter tuning, or model training. All preprocessing steps were confined to the training data and applied to the test set. Model performance was evaluated on an independent test set. Subgroup analysis In the cross-sectional study, the dataset was divided into four subsets based on different GLM7 (7.0, 7.5, 8.0) and GLM6 (5.0, 5.5, 6.0) threshold ranges; within each subset, subgroups were further stratified by age and gender. In the cohort study, subgroups were defined based on age, gender, smoking status, and drinking status to analyze the heterogeneity in the association between exposure variables and each outcome variable. Sensitivity analysis In the study, there was overlap between the components of the GLM6/GLM7 indices and the outcome definitions. To directly quantify the impact of this variable overlap, we conducted a systematic sensitivity analysis, sequentially excluding overlapping components and re-evaluating the AUC. To evaluate the differences in predictive performance between the GLM6 and GLM7 indices and established predictive markers for cross-sectional outcome variables. The study selected the following four established predictive markers, which have been supported by numerous studies, for receiver operating characteristic (ROC) curve analysis to compare the AUC values of each model: the triglyceride-glucose index (TyG), the triglyceride-glucose-body mass index (TyG-BMI), the metabolic score for insulin resistance (METS-IR), and the atherosclerotic index of plasma (AIP). The calculation formulas for each indicator are shown in Supplementary Table 7. To further explore the threshold limits of GLM6, the study set the number of clusters (K) in the K-means clustering analysis to 2 and evaluated the association between the binary groups of each GLM6 index and outcomes as a supplementary sensitivity analysis (Supplementary Figure 3). This was done to assess the sensitivity of the main findings to the choice of the number of clusters and to infer the clinical significance of the binary cutoff values. In the cohort study, to assess the stability of the primary outcomes, the GLM6 and cumulative GLM6 indices were stratified into three groups across different subpopulations, and the associations between each group and the outcome variables were analyzed. Furthermore, raw data—without imputing missing values for covariates—were used to analyze the associations between GLM6-related indicators and new-onset CVD and DM, thereby further evaluating the robustness of the primary outcomes. Statistical analysis Because NHANES uses a complex sampling design, the regression analyses based on NHANES data were performed using the survey package in R to account for sampling weights, strata, and primary sampling units. To evaluate the influence of survey weighting on association estimates, we also compared weighted and unweighted models. The primary purpose of this study in using CHARLS is to evaluate the model’s predictive performance and transferability at the individual level within an independent cohort, rather than to estimate disease prevalence or population-level effect parameters among the general population of middle-aged and older adults in China. Therefore, model performance metrics such as AUC, accuracy, sensitivity, and specificity in the CHARLS external validation are all calculated based on the unweighted analysis sample. Multicollinearity among the indicators was assessed using a multicollinearity test to calculate the correlation coefficients (r) and variance inflation factor (VIF) between the components of the exposure variables. Since all correlation coefficients were less than 0.7 and all VIFs were less than 10, there was no multicollinearity among the indicators (Supplementary Table 8, Supplementary Figure 4). For all P-values from the main association analyses, the study performed false discovery rate (FDR) adjustment and reported the adjusted P-values. In terms of descriptive statistics, the baseline characteristics of the study population were described by grouping participants according to their disease status. For continuous variables, t-tests were used to compare groups, with results presented as means (standard deviations). For categorical variables, chi-square tests were used to compare groups, with results presented as frequencies (percentages). Statistical analysis was performed using R software version 4.3.2. Differences were considered statistically significant when P < 0.05. Results Baseline characteristics of the study population in a cross-sectional study The study included 23,829 participants from the NHANES database. The levels of GLM6 and GLM7 were significantly higher in all disease groups than in the non-disease group. In terms of demographic characteristics, except the group with mental and psychological disorders, the average age in the other eight disease groups was significantly higher than that of the healthy control group. There were significant differences between the disease group and the non-disease group in terms of racial distribution and educational attainment. The educational level of the groups with cardiovascular category and endocrine metabolic diseases was lower than that of the groups without these conditions. About lifestyle factors, there were significant differences between the disease group and the non-disease group in terms of the proportion of smokers and alcohol consumption. Specifically, the proportion of alcohol consumers was significantly higher in the cardiovascular category and metabolic endocrine disease groups than in the non-disease group. Clinical data indicate that the lipid profiles of the patient group were less favorable, as evidenced by elevated levels of TG and LDL-C and significantly reduced levels of HDL-C (Supplementary Table 9). The study included 18,550 participants from the CHARLS database. GLM6 levels were significantly higher in the groups with CVD and chronic diseases than in the group without these diseases. The disease groups exhibited poorer metabolic health, with higher levels of FBG, TG, and TC, and lower levels of HDL-C. In the group with cancer, the proportion of females was significantly higher than that of males. In the group with liver disease, the proportion of males was significantly higher than in the group without liver disease, and levels of the metabolic markers TG and LDL-C were higher, while HDL-C levels were lower (Supplementary Table 10). Associations between GLM7 and GLM6 and 9 disease types, 15 individual diseases, and mortality rates The multivariable survey-weighted regression analysis showed that GLM7 and GLM6 were significantly positively associated with all nine disease types across all models (all P < 0.05). The largest increases in disease risk were observed for endocrine and metabolic diseases, followed by hepatobiliary diseases. In adjusted Model 2, both GLM7 and GLM6 were significantly positively associated with 14 individual diseases, whereas their associations with hyperthyroidism were not statistically significant (P > 0.05) (Table 1). GLM7 and GLM6 were also significantly positively associated with all-cause mortality and cardiovascular mortality (Table 2). In weighted and non-weighted comparisons, the direction of association was consistent and the statistical significance of the core findings did not materially change, indicating that weighting corrected the sample-representative structure while the core conclusions remained robust (Supplementary Figure 5). Table 1Results of the association analysis between GLM7, GLM6, and 9 disease types, and 15 individual diseases (weighted)OutcomesGLM6GLM7OR (95% CI) P FDR- Adjusted POR (95% CI) P FDR- Adjusted PCardiovascular Category M02.31(2.21, 2.41)< 0.001< 0.0012.18(2.08, 2.28)< 0.001< 0.001 M12.33(2.23, 2.45)< 0.001< 0.0012.19(2.09, 2.3)< 0.001< 0.001 M22.26(2.16, 2.37)< 0.001< 0.0012.12(2.02, 2.22)< 0.001< 0.001Endocrine and metabolic M06.55(6.07, 7.08)< 0.001< 0.0015.69(5.29, 6.12)< 0.001< 0.001 M17.77(7.17, 8.43)< 0.001< 0.0016.28(5.84, 6.76)< 0.001< 0.001 M27.95(7.31, 8.64)< 0.001< 0.0016.25(5.8, 6.74)< 0.001< 0.001Respiratory M01.22(1.17, 1.26)< 0.001< 0.0011.25(1.21, 1.3)< 0.001< 0.001 M11.21(1.17, 1.26)< 0.001< 0.0011.25(1.21, 1.3)< 0.001< 0.001 M21.21(1.16, 1.26)< 0.001< 0.0011.25(1.21, 1.3)< 0.001< 0.001Hepatobiliary M02.95(2.79, 3.12)< 0.001< 0.0015.16(4.78, 5.56)< 0.001< 0.001 M12.93(2.77, 3.1)< 0.001< 0.0015.12(4.75, 5.53)< 0.001< 0.001 M22.94(2.78, 3.12)< 0.001< 0.0015.18(4.8, 5.59)< 0.001< 0.001Renal and urological M01.63(1.54, 1.72)< 0.001< 0.0011.55(1.47, 1.63)< 0.001< 0.001 M11.57(1.48, 1.66)< 0.001< 0.0011.49(1.42, 1.58)< 0.001< 0.001 M21.51(1.43, 1.6)< 0.001< 0.0011.45(1.37, 1.53)< 0.001< 0.001Cancer M01.49(1.4, 1.6)< 0.001< 0.0011.31(1.23, 1.4)< 0.001< 0.001 M11.54(1.44, 1.64)< 0.001< 0.0011.34(1.26, 1.43)< 0.001< 0.001 M21.53(1.43, 1.64)< 0.001< 0.0011.34(1.25, 1.43)< 0.001< 0.001Mental and psychological M01.04(0.99, 1.09)0.1500.1501.08(1.04, 1.13)< 0.001< 0.001 M11.07(1.02, 1.12)0.0080.0081.11(1.06, 1.16)< 0.001< 0.001 M21.08(1.03, 1.14)0.0020.0021.12(1.07, 1.18)< 0.001< 0.001Musculoskeletal M02(1.91, 2.09)< 0.001< 0.0011.8(1.71, 1.89)< 0.001< 0.001 M12.12(2.02, 2.23)< 0.001< 0.0011.88(1.79, 1.98)< 0.001< 0.001 M22.03(1.93, 2.13)< 0.001< 0.0011.8(1.71, 1.9)< 0.001< 0.001Other M01.41(1.35, 1.47)< 0.001< 0.0011.36(1.3, 1.42)< 0.001< 0.001 M11.43(1.37, 1.49)< 0.001< 0.0011.37(1.31, 1.43)< 0.001< 0.001 M21.42(1.35, 1.48)< 0.001< 0.0011.36(1.3, 1.43)< 0.001< 0.001Hypertension M02.24(2.14, 2.34)< 0.001< 0.0012.13(2.03, 2.23)< 0.001< 0.001 M12.27(2.17, 2.38)< 0.001< 0.0012.15(2.05, 2.25)< 0.001< 0.001 M22.21(2.11, 2.32)< 0.001< 0.0012.08(1.99, 2.19)< 0.001< 0.001CHD M01.82(1.61, 2.06)< 0.001< 0.0011.72(1.54, 1.92)< 0.001< 0.001 M11.75(1.55, 1.98)< 0.001< 0.0011.65(1.48, 1.84)< 0.001< 0.001 M21.61(1.41, 1.83)< 0.001< 0.0011.54(1.37, 1.73)< 0.001< 0.001Stroke M01.69(1.51, 1.88)< 0.001< 0.0011.54(1.39, 1.71)< 0.001< 0.001 M11.73(1.55, 1.93)< 0.001< 0.0011.58(1.43, 1.75)< 0.001< 0.001 M21.59(1.43, 1.78)< 0.001< 0.0011.45(1.32, 1.6)< 0.001< 0.001Heart attack M01.86(1.67, 2.08)< 0.001< 0.0011.73(1.58, 1.9)< 0.001< 0.001 M11.8(1.62, 2.01)< 0.001< 0.0011.68(1.53, 1.84)< 0.001< 0.001 M21.67(1.48, 1.88)< 0.001< 0.0011.57(1.42, 1.73)< 0.001< 0.001Angina M01.92(1.7, 2.18)< 0.001< 0.0011.78(1.58, 1.99)< 0.001< 0.001 M11.89(1.67, 2.15)< 0.001< 0.0011.75(1.56, 1.96)< 0.001< 0.001 M21.76(1.54, 2.02)< 0.001< 0.0011.64(1.45, 1.85)< 0.001< 0.001Heart failure M02.07(1.81, 2.37)< 0.001< 0.0011.93(1.7, 2.2)< 0.001< 0.001 M12.07(1.81, 2.37)< 0.001< 0.0011.93(1.69, 2.19)< 0.001< 0.001 M21.9(1.66, 2.18)< 0.001< 0.0011.77(1.56, 2.01)< 0.001< 0.001PAD M01.51(1.23, 1.86)< 0.001< 0.0011.51(1.28, 1.79)< 0.001< 0.001 M11.55(1.26, 1.92)< 0.001< 0.0011.55(1.31, 1.84)< 0.001< 0.001 M21.45(1.15, 1.82)0.0020.0021.46(1.21, 1.76)< 0.001< 0.001DM M03.58(3.34, 3.84)< 0.001< 0.0013.42(3.2, 3.65)< 0.001< 0.001 M13.61(3.37, 3.87)< 0.001< 0.0013.45(3.23, 3.68)< 0.001< 0.001 M23.5(3.26, 3.76)< 0.001< 0.0013.33(3.11, 3.56)< 0.001< 0.001Metabolic syndrome M05.44(5.15, 5.75)< 0.001< 0.0015.43(5.14, 5.75)< 0.001< 0.001 M16.21(5.85, 6.6)< 0.001< 0.0016.08(5.71, 6.46)< 0.001< 0.001 M26.14(5.76, 6.53)< 0.001< 0.0016(5.64, 6.39)< 0.001< 0.001Gout M02.07(1.78, 2.4)< 0.001< 0.0011.95(1.69, 2.25)< 0.001< 0.001 M12.01(1.72, 2.35)< 0.001< 0.0011.89(1.63, 2.19)< 0.001< 0.001 M21.96(1.66, 2.31)< 0.001< 0.0011.85(1.58, 2.16)< 0.001< 0.001Hyperthyroidism M00.79(0.47, 1.33)0.3620.3620.85(0.55, 1.32)0.4580.458 M10.85(0.49, 1.49)0.5700.5700.91(0.57, 1.47)0.6930.693 M20.7(0.41, 1.19)0.1830.1830.75(0.48, 1.17)0.2000.200Hypothyroidism M01.37(0.97, 1.93)0.0750.0801.42(1.01, 1.99)0.0460.049 M11.47(1.05, 2.04)0.0240.0261.51(1.06, 2.13)0.0220.023 M21.66(1.12, 2.48)0.0130.0141.7(1.22, 2.37)0.0020.002Liver disease M01.5(1.35, 1.66)< 0.001< 0.0011.49(1.34, 1.65)< 0.001< 0.001 M11.48(1.34, 1.65)< 0.001< 0.0011.48(1.33, 1.64)< 0.001< 0.001 M21.46(1.31, 1.62)< 0.001< 0.0011.46(1.31, 1.62)< 0.001< 0.001NAFLD M03.59(3.39, 3.82)< 0.001< 0.0019.94(9.08, 10.88)< 0.001< 0.001 M13.59(3.38, 3.81)< 0.001< 0.0019.97(9.12, 10.91)< 0.001< 0.001 M23.64(3.43, 3.86)< 0.001< 0.00110.46(9.55, 11.46)< 0.001< 0.001MAFLD M03.82(3.61, 4.05)< 0.001< 0.00112.51(11.38, 13.74)< 0.001< 0.001 M13.79(3.58, 4.01)< 0.001< 0.00112.43(11.32, 13.66)< 0.001< 0.001 M23.85(3.64, 4.08)< 0.001< 0.00113.21(12, 14.55)< 0.001< 0.001M0, Model 0; M1, Model 1; M3, Model 3Model 0 is the unadjusted variable modelModel 1 adjusts for genderModel 2 adjusts for gender, educational level, poverty-income ratio, marital status, smoking status, and drinking status Table 1Results of the association analysis between GLM7, GLM6, and 9 disease types, and 15 individual diseases (weighted)OutcomesGLM6GLM7OR (95% CI) P FDR- Adjusted POR (95% CI) P FDR- Adjusted PCardiovascular Category M02.31(2.21, 2.41)< 0.001< 0.0012.18(2.08, 2.28)< 0.001< 0.001 M12.33(2.23, 2.45)< 0.001< 0.0012.19(2.09, 2.3)< 0.001< 0.001 M22.26(2.16, 2.37)< 0.001< 0.0012.12(2.02, 2.22)< 0.001< 0.001Endocrine and metabolic M06.55(6.07, 7.08)< 0.001< 0.0015.69(5.29, 6.12)< 0.001< 0.001 M17.77(7.17, 8.43)< 0.001< 0.0016.28(5.84, 6.76)< 0.001< 0.001 M27.95(7.31, 8.64)< 0.001< 0.0016.25(5.8, 6.74)< 0.001< 0.001Respiratory M01.22(1.17, 1.26)< 0.001< 0.0011.25(1.21, 1.3)< 0.001< 0.001 M11.21(1.17, 1.26)< 0.001< 0.0011.25(1.21, 1.3)< 0.001< 0.001 M21.21(1.16, 1.26)< 0.001< 0.0011.25(1.21, 1.3)< 0.001< 0.001Hepatobiliary M02.95(2.79, 3.12)< 0.001< 0.0015.16(4.78, 5.56)< 0.001< 0.001 M12.93(2.77, 3.1)< 0.001< 0.0015.12(4.75, 5.53)< 0.001< 0.001 M22.94(2.78, 3.12)< 0.001< 0.0015.18(4.8, 5.59)< 0.001< 0.001Renal and urological M01.63(1.54, 1.72)< 0.001< 0.0011.55(1.47, 1.63)< 0.001< 0.001 M11.57(1.48, 1.66)< 0.001< 0.0011.49(1.42, 1.58)< 0.001< 0.001 M21.51(1.43, 1.6)< 0.001< 0.0011.45(1.37, 1.53)< 0.001< 0.001Cancer M01.49(1.4, 1.6)< 0.001< 0.0011.31(1.23, 1.4)< 0.001< 0.001 M11.54(1.44, 1.64)< 0.001< 0.0011.34(1.26, 1.43)< 0.001< 0.001 M21.53(1.43, 1.64)< 0.001< 0.0011.34(1.25, 1.43)< 0.001< 0.001Mental and psychological M01.04(0.99, 1.09)0.1500.1501.08(1.04, 1.13)< 0.001< 0.001 M11.07(1.02, 1.12)0.0080.0081.11(1.06, 1.16)< 0.001< 0.001 M21.08(1.03, 1.14)0.0020.0021.12(1.07, 1.18)< 0.001< 0.001Musculoskeletal M02(1.91, 2.09)< 0.001< 0.0011.8(1.71, 1.89)< 0.001< 0.001 M12.12(2.02, 2.23)< 0.001< 0.0011.88(1.79, 1.98)< 0.001< 0.001 M22.03(1.93, 2.13)< 0.001< 0.0011.8(1.71, 1.9)< 0.001< 0.001Other M01.41(1.35, 1.47)< 0.001< 0.0011.36(1.3, 1.42)< 0.001< 0.001 M11.43(1.37, 1.49)< 0.001< 0.0011.37(1.31, 1.43)< 0.001< 0.001 M21.42(1.35, 1.48)< 0.001< 0.0011.36(1.3, 1.43)< 0.001< 0.001Hypertension M02.24(2.14, 2.34)< 0.001< 0.0012.13(2.03, 2.23)< 0.001< 0.001 M12.27(2.17, 2.38)< 0.001< 0.0012.15(2.05, 2.25)< 0.001< 0.001 M22.21(2.11, 2.32)< 0.001< 0.0012.08(1.99, 2.19)< 0.001< 0.001CHD M01.82(1.61, 2.06)< 0.001< 0.0011.72(1.54, 1.92)< 0.001< 0.001 M11.75(1.55, 1.98)< 0.001< 0.0011.65(1.48, 1.84)< 0.001< 0.001 M21.61(1.41, 1.83)< 0.001< 0.0011.54(1.37, 1.73)< 0.001< 0.001Stroke M01.69(1.51, 1.88)< 0.001< 0.0011.54(1.39, 1.71)< 0.001< 0.001 M11.73(1.55, 1.93)< 0.001< 0.0011.58(1.43, 1.75)< 0.001< 0.001 M21.59(1.43, 1.78)< 0.001< 0.0011.45(1.32, 1.6)< 0.001< 0.001Heart attack M01.86(1.67, 2.08)< 0.001< 0.0011.73(1.58, 1.9)< 0.001< 0.001 M11.8(1.62, 2.01)< 0.001< 0.0011.68(1.53, 1.84)< 0.001< 0.001 M21.67(1.48, 1.88)< 0.001< 0.0011.57(1.42, 1.73)< 0.001< 0.001Angina M01.92(1.7, 2.18)< 0.001< 0.0011.78(1.58, 1.99)< 0.001< 0.001 M11.89(1.67, 2.15)< 0.001< 0.0011.75(1.56, 1.96)< 0.001< 0.001 M21.76(1.54, 2.02)< 0.001< 0.0011.64(1.45, 1.85)< 0.001< 0.001Heart failure M02.07(1.81, 2.37)< 0.001< 0.0011.93(1.7, 2.2)< 0.001< 0.001 M12.07(1.81, 2.37)< 0.001< 0.0011.93(1.69, 2.19)< 0.001< 0.001 M21.9(1.66, 2.18)< 0.001< 0.0011.77(1.56, 2.01)< 0.001< 0.001PAD M01.51(1.23, 1.86)< 0.001< 0.0011.51(1.28, 1.79)< 0.001< 0.001 M11.55(1.26, 1.92)< 0.001< 0.0011.55(1.31, 1.84)< 0.001< 0.001 M21.45(1.15, 1.82)0.0020.0021.46(1.21, 1.76)< 0.001< 0.001DM M03.58(3.34, 3.84)< 0.001< 0.0013.42(3.2, 3.65)< 0.001< 0.001 M13.61(3.37, 3.87)< 0.001< 0.0013.45(3.23, 3.68)< 0.001< 0.001 M23.5(3.26, 3.76)< 0.001< 0.0013.33(3.11, 3.56)< 0.001< 0.001Metabolic syndrome M05.44(5.15, 5.75)< 0.001< 0.0015.43(5.14, 5.75)< 0.001< 0.001 M16.21(5.85, 6.6)< 0.001< 0.0016.08(5.71, 6.46)< 0.001< 0.001 M26.14(5.76, 6.53)< 0.001< 0.0016(5.64, 6.39)< 0.001< 0.001Gout M02.07(1.78, 2.4)< 0.001< 0.0011.95(1.69, 2.25)< 0.001< 0.001 M12.01(1.72, 2.35)< 0.001< 0.0011.89(1.63, 2.19)< 0.001< 0.001 M21.96(1.66, 2.31)< 0.001< 0.0011.85(1.58, 2.16)< 0.001< 0.001Hyperthyroidism M00.79(0.47, 1.33)0.3620.3620.85(0.55, 1.32)0.4580.458 M10.85(0.49, 1.49)0.5700.5700.91(0.57, 1.47)0.6930.693 M20.7(0.41, 1.19)0.1830.1830.75(0.48, 1.17)0.2000.200Hypothyroidism M01.37(0.97, 1.93)0.0750.0801.42(1.01, 1.99)0.0460.049 M11.47(1.05, 2.04)0.0240.0261.51(1.06, 2.13)0.0220.023 M21.66(1.12, 2.48)0.0130.0141.7(1.22, 2.37)0.0020.002Liver disease M01.5(1.35, 1.66)< 0.001< 0.0011.49(1.34, 1.65)< 0.001< 0.001 M11.48(1.34, 1.65)< 0.001< 0.0011.48(1.33, 1.64)< 0.001< 0.001 M21.46(1.31, 1.62)< 0.001< 0.0011.46(1.31, 1.62)< 0.001< 0.001NAFLD M03.59(3.39, 3.82)< 0.001< 0.0019.94(9.08, 10.88)< 0.001< 0.001 M13.59(3.38, 3.81)< 0.001< 0.0019.97(9.12, 10.91)< 0.001< 0.001 M23.64(3.43, 3.86)< 0.001< 0.00110.46(9.55, 11.46)< 0.001< 0.001MAFLD M03.82(3.61, 4.05)< 0.001< 0.00112.51(11.38, 13.74)< 0.001< 0.001 M13.79(3.58, 4.01)< 0.001< 0.00112.43(11.32, 13.66)< 0.001< 0.001 M23.85(3.64, 4.08)< 0.001< 0.00113.21(12, 14.55)< 0.001< 0.001M0, Model 0; M1, Model 1; M3, Model 3Model 0 is the unadjusted variable modelModel 1 adjusts for genderModel 2 adjusts for gender, educational level, poverty-income ratio, marital status, smoking status, and drinking status Results of the association analysis between GLM7, GLM6, and 9 disease types, and 15 individual diseases (weighted) M0, Model 0; M1, Model 1; M3, Model 3 Model 0 is the unadjusted variable model Model 1 adjusts for gender Model 2 adjusts for gender, educational level, poverty-income ratio, marital status, smoking status, and drinking status Table 2Results of the association analysis between GLM7, GLM6, and mortality rates (weighted)OutcomesGLM6GLM7HR (95% CI) P FDR- Adjusted PHR (95% CI) P FDR- Adjusted PAll mortality M01.91(1.81, 2.01)< 0.001< 0.0011.62(1.54, 1.7)< 0.001< 0.001 M11.9(1.81, 2)< 0.001< 0.0011.61(1.53, 1.7)< 0.001< 0.001 M21.79(1.69, 1.88)< 0.001< 0.0011.52(1.44, 1.6)< 0.001< 0.001CVD mortality M02.18(1.98, 2.41)< 0.001< 0.0011.82(1.65, 2.01)< 0.001< 0.001 M12.17(1.96, 2.39)< 0.001< 0.0011.8(1.62, 1.98)< 0.001< 0.001 M22.04(1.84, 2.26)< 0.001< 0.0011.7(1.53, 1.89)< 0.001< 0.001M0, Model 0; M1, Model 1; M3, Model 3Model 0 is the unadjusted variable modelModel 1 adjusts for genderModel 2 adjusts for gender, educational level, poverty-income ratio, marital status, smoking status, and drinking status Table 2Results of the association analysis between GLM7, GLM6, and mortality rates (weighted)OutcomesGLM6GLM7HR (95% CI) P FDR- Adjusted PHR (95% CI) P FDR- Adjusted PAll mortality M01.91(1.81, 2.01)< 0.001< 0.0011.62(1.54, 1.7)< 0.001< 0.001 M11.9(1.81, 2)< 0.001< 0.0011.61(1.53, 1.7)< 0.001< 0.001 M21.79(1.69, 1.88)< 0.001< 0.0011.52(1.44, 1.6)< 0.001< 0.001CVD mortality M02.18(1.98, 2.41)< 0.001< 0.0011.82(1.65, 2.01)< 0.001< 0.001 M12.17(1.96, 2.39)< 0.001< 0.0011.8(1.62, 1.98)< 0.001< 0.001 M22.04(1.84, 2.26)< 0.001< 0.0011.7(1.53, 1.89)< 0.001< 0.001M0, Model 0; M1, Model 1; M3, Model 3Model 0 is the unadjusted variable modelModel 1 adjusts for genderModel 2 adjusts for gender, educational level, poverty-income ratio, marital status, smoking status, and drinking status Results of the association analysis between GLM7, GLM6, and mortality rates (weighted) M0, Model 0; M1, Model 1; M3, Model 3 Model 0 is the unadjusted variable model Model 1 adjusts for gender Model 2 adjusts for gender, educational level, poverty-income ratio, marital status, smoking status, and drinking status Further survey-weighted RCS analyses showed nonlinear associations between GLM7 and five disease types: endocrine and metabolic, hepatobiliary, cancer, mental and psychological, and musculoskeletal diseases (Fig. 2). GLM6 showed nonlinear associations with four disease types: cardiovascular category, endocrine and metabolic, mental and psychological, and musculoskeletal diseases (Fig. 3). Fig. 2Nonlinear association analysis results between GLM7 and 9 disease types. (A) GLM7 and cardiovascular category; (B) GLM7 and endocrine and metabolic diseases; (C) GLM7 and respiratory diseases; (D) GLM7 and hepatobiliary diseases; (E) GLM7 and renal and urological diseases; (F) GLM7 and cancer; (G) GLM7 and mental and psychological disorders; (H) GLM7 and musculoskeletal disorders; (I) GLM7 and other disorders Nonlinear association analysis results between GLM7 and 9 disease types. (A) GLM7 and cardiovascular category; (B) GLM7 and endocrine and metabolic diseases; (C) GLM7 and respiratory diseases; (D) GLM7 and hepatobiliary diseases; (E) GLM7 and renal and urological diseases; (F) GLM7 and cancer; (G) GLM7 and mental and psychological disorders; (H) GLM7 and musculoskeletal disorders; (I) GLM7 and other disorders Fig. 3Nonlinear association analysis results between GLM6 and 9 disease types. (A) GLM6 and cardiovascular category; (B) GLM6 and endocrine and metabolic diseases; (C) GLM6 and respiratory diseases; (D) GLM6 and hepatobiliary diseases; (E) GLM6 and renal and urological diseases; (F) GLM6 and cancer; (G) GLM6 and mental and psychological diseases; (H) GLM6 and musculoskeletal disorders; (I) GLM6 and other disorders Nonlinear association analysis results between GLM6 and 9 disease types. (A) GLM6 and cardiovascular category; (B) GLM6 and endocrine and metabolic diseases; (C) GLM6 and respiratory diseases; (D) GLM6 and hepatobiliary diseases; (E) GLM6 and renal and urological diseases; (F) GLM6 and cancer; (G) GLM6 and mental and psychological diseases; (H) GLM6 and musculoskeletal disorders; (I) GLM6 and other disorders GLM7 showed nonlinear associations with six individual diseases: hypertension, DM, metabolic syndrome, gout, NAFLD, and MAFLD (Supplementary Figure 6). GLM6 showed nonlinear associations with seven individual diseases: hypertension, stroke, DM, metabolic syndrome, gout, NAFLD, and MAFLD (Supplementary Figure 7). Nonlinear associations between GLM7 and GLM6 and mortality risk were analyzed. A nonlinear association was observed between GLM6 and all-cause mortality. At high levels, GLM7 and GLM6 were associated with increased all-cause and cardiovascular mortality (Fig. 4). In addition, the KM curves for GLM7 and GLM6 quartiles were clearly separated; higher levels were associated with higher all-cause and cardiovascular mortality and lower survival probability (Fig. 5). Fig. 4Nonlinear association analysis results for GLM7, GLM6, and all-cause mortality risk and cardiovascular mortality risk. (A) RCS curves of GLM6 with all-cause mortality risk; (B) RCS curves of GLM7 with cardiovascular mortality risk; (C) RCS curves of GLM6 with cardiovascular mortality risk; (D) RCS curves of GLM7 with cardiovascular mortality risk Nonlinear association analysis results for GLM7, GLM6, and all-cause mortality risk and cardiovascular mortality risk. (A) RCS curves of GLM6 with all-cause mortality risk; (B) RCS curves of GLM7 with cardiovascular mortality risk; (C) RCS curves of GLM6 with cardiovascular mortality risk; (D) RCS curves of GLM7 with cardiovascular mortality risk Fig. 5Nonlinear association analysis and survival analysis results for GLM7, GLM6, and all-cause mortality risk and CVD mortality risk. (A) KM curves of GLM6 with all-cause mortality risk; (B) KM curves of GLM7 with all-cause mortality risk; (C) KM curves of GLM6 with CVD mortality risk; (D) KM curves of GLM7 with CVD mortality risk Nonlinear association analysis and survival analysis results for GLM7, GLM6, and all-cause mortality risk and CVD mortality risk. (A) KM curves of GLM6 with all-cause mortality risk; (B) KM curves of GLM7 with all-cause mortality risk; (C) KM curves of GLM6 with CVD mortality risk; (D) KM curves of GLM7 with CVD mortality risk Performance of XGBoost machine learning models based on GLM7 and GLM6 in predicting the diagnosis of various diseases GLM7 and GLM6 have the potential to support population risk stratification in disease screening. The study further developed XGBoost machine learning models based on the GLM7 and GLM6 variable sets, which comprise seven and six factors, respectively, for disease screening and diagnosis. The NHANES data were divided into training and test sets in a 7:3 ratio, and the CHARLS data were used for external validation. Among the nine disease categories, the three models with the best predictive performance were those for endocrine and metabolic diseases, hepatobiliary diseases, and cardiovascular category. Among the individual disease models, the five outcomes with the highest AUC values were DM, metabolic syndrome, MAFLD, NAFLD, and CHD (Table 3). Table 3Results of various indicators for the XGBoost model predicting multiple diseases based on the GLM7 and GLM6 indicatorsDiseaseAUCAccuracySensitivitySpecificityGLM6GLM7GLM6GLM7GLM6GLM7GLM6GLM7Endocrine and metabolic0.963 (0.960, 0.967)0.963 (0.959, 0.967)0.8810.8860.8630.8740.9530.938Hepatobiliary0.858 (0.849, 0.867)0.904 (0.896, 0.911)0.7720.8390.7840.7650.7650.878Cardiovascular category0.835 (0.826, 0.845)0.836 (0.827, 0.846)0.7660.7650.7450.7440.7810.781Musculoskeletal0.800 (0.789, 0.811)0.798 (0.788, 0.809)0.7290.7090.7630.7990.7140.669Cancer0.786 (0.769, 0.804)0.785 (0.768, 0.803)0.6770.6840.7880.7790.6660.674Renal and urological0.671 (0.649, 0.693)0.671 (0.649, 0.693)0.5710.5580.6950.7060.5590.544Other0.663 (0.650, 0.677)0.652 (0.639, 0.666)0.6270.6180.6870.6740.5690.566Respiratory0.566 (0.550, 0.583)0.566 (0.550, 0.583)0.5390.5390.5740.5750.530.53Mental and psychological0.562 (0.544, 0.580)0.557 (0.539, 0.576)0.5380.5580.5670.5180.5290.571DM0.947 (0.940, 0.954)0.948 (0.941, 0.955)0.8960.8850.8380.850.9080.892Metabolic syndrome0.935 (0.930–0.941)0.936 (0.931–0.941)0.8540.8580.8750.8590.8390.857MAFLD0.901 (0.894–0.908)0.959 (0.955–0.963)0.7920.8860.8650.8980.7630.881NAFLD0.882 (0.873–0.890)0.943 (0.937–0.948)0.7820.8620.8330.8970.7650.85Hypertension0.862 (0.845–0.878)0.862 (0.845–0.879)0.6920.6730.90.930.6830.662CHD0.816 (0.806–0.826)0.817 (0.807–0.827)0.7440.7520.7690.7370.7270.762Heart attack0.813 (0.787–0.839)0.813 (0.788–0.839)0.6620.6920.8330.8130.6560.689Heart failure0.809 (0.787–0.831)0.806 (0.784–0.828)0.6780.6250.8320.8840.6710.613Angina0.787 (0.760–0.814)0.787 (0.760–0.814)0.5870.6470.8680.7970.5790.642Stroke0.787 (0.759–0.815)0.791 (0.763–0.818)0.7030.5650.7330.9030.7020.548PAD0.772 (0.721–0.823)0.766 (0.715–0.817)0.580.4990.8670.9330.5550.463Gout0.763 (0.734–0.792)0.764 (0.736–0.792)0.6240.6380.7930.7760.6170.633Liver diseases0.673 (0.643–0.704)0.682 (0.650–0.713)0.5320.5020.7490.780.5230.491Hypothyroidism0.478 (0.228–0.729)0.560 (0.336–0.785)0.1660.8590.6670.4440.1630.862 Table 3Results of various indicators for the XGBoost model predicting multiple diseases based on the GLM7 and GLM6 indicatorsDiseaseAUCAccuracySensitivitySpecificityGLM6GLM7GLM6GLM7GLM6GLM7GLM6GLM7Endocrine and metabolic0.963 (0.960, 0.967)0.963 (0.959, 0.967)0.8810.8860.8630.8740.9530.938Hepatobiliary0.858 (0.849, 0.867)0.904 (0.896, 0.911)0.7720.8390.7840.7650.7650.878Cardiovascular category0.835 (0.826, 0.845)0.836 (0.827, 0.846)0.7660.7650.7450.7440.7810.781Musculoskeletal0.800 (0.789, 0.811)0.798 (0.788, 0.809)0.7290.7090.7630.7990.7140.669Cancer0.786 (0.769, 0.804)0.785 (0.768, 0.803)0.6770.6840.7880.7790.6660.674Renal and urological0.671 (0.649, 0.693)0.671 (0.649, 0.693)0.5710.5580.6950.7060.5590.544Other0.663 (0.650, 0.677)0.652 (0.639, 0.666)0.6270.6180.6870.6740.5690.566Respiratory0.566 (0.550, 0.583)0.566 (0.550, 0.583)0.5390.5390.5740.5750.530.53Mental and psychological0.562 (0.544, 0.580)0.557 (0.539, 0.576)0.5380.5580.5670.5180.5290.571DM0.947 (0.940, 0.954)0.948 (0.941, 0.955)0.8960.8850.8380.850.9080.892Metabolic syndrome0.935 (0.930–0.941)0.936 (0.931–0.941)0.8540.8580.8750.8590.8390.857MAFLD0.901 (0.894–0.908)0.959 (0.955–0.963)0.7920.8860.8650.8980.7630.881NAFLD0.882 (0.873–0.890)0.943 (0.937–0.948)0.7820.8620.8330.8970.7650.85Hypertension0.862 (0.845–0.878)0.862 (0.845–0.879)0.6920.6730.90.930.6830.662CHD0.816 (0.806–0.826)0.817 (0.807–0.827)0.7440.7520.7690.7370.7270.762Heart attack0.813 (0.787–0.839)0.813 (0.788–0.839)0.6620.6920.8330.8130.6560.689Heart failure0.809 (0.787–0.831)0.806 (0.784–0.828)0.6780.6250.8320.8840.6710.613Angina0.787 (0.760–0.814)0.787 (0.760–0.814)0.5870.6470.8680.7970.5790.642Stroke0.787 (0.759–0.815)0.791 (0.763–0.818)0.7030.5650.7330.9030.7020.548PAD0.772 (0.721–0.823)0.766 (0.715–0.817)0.580.4990.8670.9330.5550.463Gout0.763 (0.734–0.792)0.764 (0.736–0.792)0.6240.6380.7930.7760.6170.633Liver diseases0.673 (0.643–0.704)0.682 (0.650–0.713)0.5320.5020.7490.780.5230.491Hypothyroidism0.478 (0.228–0.729)0.560 (0.336–0.785)0.1660.8590.6670.4440.1630.862 Results of various indicators for the XGBoost model predicting multiple diseases based on the GLM7 and GLM6 indicators In the unified test set, XGBoost models constructed using the GLM6 and GLM7 variable sets showed broadly similar predictive performance for most disease categories and individual disease outcomes. However, the GLM7 model showed a clearer improvement for hepatobiliary outcomes and liver steatosis-related individual outcomes. Calibration performance varied across disease outcomes; therefore, we additionally reported the Brier score, calibration intercept, calibration slope, and Hosmer-Lemeshow test P values to present the agreement between predicted probabilities and observed risks more transparently (Supplementary Table 11). For the hepatobiliary model, GLM7 showed greater net benefit across a wider range of threshold probabilities. Compared with GLM6, the addition of insulin in GLM7 may provide incremental value mainly for liver steatosis and liver metabolism-related outcomes (Supplementary Figure 8). The study further used GLM6 data from the external CHARLS validation dataset to evaluate model extrapolation performance. In CHARLS, the model showed moderate discriminatory ability for CVD and metabolic diseases, whereas external performance was limited for some other outcomes (Supplementary Table 12). These results suggest that GLM6 may be more appropriate as an initial risk-stratification indicator for cardiovascular category and metabolic diseases in population-level screening. Subgroup analysis in a cross-sectional study The study used subgroup analysis to further investigate the associations of GLM6 and GLM7 across different threshold ranges and disease subgroups. Participants were grouped by age and gender for each disease, and interaction tests were performed (Supplementary Figure 9–12). For CHD, the odds of disease status increased when GLM6 exceeded 5.5. For DM, across the four threshold ranges, higher GLM6 and GLM7 values were associated with higher odds of DM. The increase became more pronounced when GLM6 exceeded 5.5 and GLM7 exceeded 7.5, and DM showed very high odds when GLM6 exceeded 6.0 and GLM7 exceeded 8.0. For MAFLD, the odds first decreased and then increased as the threshold range expanded. These subgroup results indicate that GLM6 and GLM7 threshold ranges differ across diseases and should be discussed in a disease-specific manner. Sensitivity analyses in the cross-sectional study For cardiovascular category and metabolic disease-related outcomes, GLM6 and GLM7 were both significantly positively associated with the relevant outcomes in the main model. After further adjustment for systolic blood pressure, diastolic blood pressure, and HbA1c in Model 3, these associations remained significantly positive, supporting the robustness of the findings (Supplementary Table 13). In the outcome definitions, DM included FBG; metabolic syndrome and endocrine and metabolic diseases included FBG, TG, and HDL-C; and MAFLD and NAFLD included FBG, TG, and BMI. After sequentially excluding the overlapping variables and reassessing AUC values for these outcomes, the AUCs showed only small increases or decreases, without a consistent marked decline. These results suggest that model performance was not fully explained by variable overlap (Supplementary Figure 13). The study compared the discriminative performance of GLM6 and GLM7 with four existing indicators for cross-sectional outcomes (Supplementary Figure 14). The results showed that GLM6 and GLM7 outperformed TyG, TyG-BMI, METS-IR, and AIP for most outcomes. For endocrine and metabolic diseases, GLM6 had the highest AUC (0.881), followed by GLM7 (0.864), both of which were higher than TyG (0.824), TyG-BMI (0.791), METS-IR (0.780), and AIP (0.794). These comparisons further support the good discriminative performance of GLM6 and GLM7 for cardiovascular category and metabolic diseases. Baseline characteristics of the cohort study population For the follow-up outcomes in the cohort study, the CVD and DM identified as the optimal prediction targets in the cross-sectional study were selected. This further explores the association between long-term changes in the GLM6 index and the incidence of CVD and DM. The study included 4,633 participants from the 2011 cohort of the CHARLS database and followed them for CVD until 2020, during which 556 new cases of CVD were identified. Additionally, 3,974 participants from the 2011 cohort of the CHARLS database were included and followed for DM until 2020, during which 131 new cases of DM were identified (Table 4). Table 4Baseline characteristics of the CHARLS cohort at the time of enrollment in 2011CVDDMNoYes P NoYes P N 40775563843131 GLM6 5.58 (0.42)5.69 (0.41)< 0.0015.53 (0.38)5.71 (0.36)< 0.001 Age57.68 (8.61)59.08 (8.08)< 0.00157.77 (8.62)57.89 (8.20)0.875 BMI (kg/m²)225.03 (6460.46)24.35 (4.52)0.464189.30 (5942.45)24.19 (3.89)0.751 FBG (mg/dL)108.12 (31.64)114.35 (43.13)< 0.00199.47 (11.05)103.73 (14.89)< 0.001 TG (mg/dL)128.53 (93.17)142.22 (110.74)0.002118.50 (72.56)146.36 (94.05)< 0.001 LDL-C (mg/dL)115.94 (34.26)119.74 (36.47)0.015116.60 (33.68)121.10 (34.75)0.133 HDL-C (mg/dL)51.66 (15.12)49.64 (15.58)0.00352.47 (14.77)47.55 (14.17)< 0.001 Cumulative GLM6 22.33 (1.43)22.70 (1.45)< 0.00122.11 (1.31)22.74 (1.22)< 0.001 GLM6 dynamic clustering group Favorable1294 (31.7)122 (21.9)< 0.0011153 (30.0)20 (15.3)< 0.001 Intermediate1772 (43.5)263 (47.3)1696 (44.1)55 (42.0) Unfavorable1011 (24.8)171 (30.8)994 (25.9)56 (42.7) Gender Male1825 (44.8)224 (40.3)0.0511645 (42.8)55 (42.0)0.923 Female2252 (55.2)332 (59.7)2198 (57.2)76 (58.0) Education Level Illiterate1136 (27.9)180 (32.4)0.1481081 (28.1)41 (31.3)0.809 Primary School1710 (41.9)224 (40.3)1616 (42.1)51 (38.9) Middle School/High School/Vocational School1191 (29.2)146 (26.3)1104 (28.7)37 (28.2) College degree or higher40 (1.0)6 (1.1)42 (1.1)2 (1.5) Marital status Married3707 (90.9)504 (90.6)0.7523489 (90.8)112 (85.5)0.232 Divorce32 (0.8)4 (0.7)32 (0.8)2 (1.5) Widowhood319 (7.8)47 (8.5)306 (8.0)16 (12.2) Unmarried19 (0.5)1 (0.2)16 (0.4)1 (0.8) Smoking status No2569 (63.0)364 (65.5)0.282459 (64.0)92 (70.2)0.17 Yes1508 (37.0)192 (34.5)1384 (36.0)39 (29.8) Drinking status No2724 (66.8)379 (68.2)0.5572615 (68.0)88 (67.2)0.909 Yes1353 (33.2)177 (31.8)1228 (32.0)43 (32.8) Height (cm) 157.66 (10.06)157.73 (9.17)0.892157.60 (9.93)158.43 (8.16)0.348 Weight (kg) 58.67 (11.10)60.65 (12.26)< 0.00158.20 (11.00)60.94 (11.74)0.005 WC (cm) 83.87 (12.02)86.50 (12.99)< 0.00183.34 (11.98)85.85 (14.32)0.019 TC (mg/dL) 192.74 (37.59)197.61 (37.23)0.004191.69 (37.07)197.59 (37.91)0.074 SBP (mmHg) 128.01 (20.35)133.81 (21.66)< 0.001127.63 (20.38)129.88 (20.58)0.215 DBP (mmHg) 74.99 (11.91)77.48 (12.21)< 0.00174.82 (12.00)76.58 (12.30)0.099 HbA1c (%) 5.23 (0.73)5.36 (0.99)< 0.0015.08 (0.37)5.22 (0.40)< 0.001 Obesity No2455 (60.2)290 (52.2)< 0.0012392 (62.2)66 (50.4)0.008 Yes1622 (39.8)266 (47.8)1451 (37.8)65 (49.6) Hypertension No2663 (65.3)288 (51.8)< 0.0012514 (65.4)70 (53.4)0.006 Yes1414 (34.7)268 (48.2)1329 (34.6)61 (46.6)BMI, Body mass index; FBG, Fasting blood glucose; TG, Triglycerides; LDL-C, Low-density lipoprotein cholesterol; HDL-C, High-density lipoprotein cholesterol; WC, Waist circumference; TC, Total cholesterol; SBP, Systolic blood pressure; DBP, Diastolic blood pressure; HbA1c, Glycated hemoglobin Table 4Baseline characteristics of the CHARLS cohort at the time of enrollment in 2011CVDDMNoYes P NoYes P N 40775563843131 GLM6 5.58 (0.42)5.69 (0.41)< 0.0015.53 (0.38)5.71 (0.36)< 0.001 Age57.68 (8.61)59.08 (8.08)< 0.00157.77 (8.62)57.89 (8.20)0.875 BMI (kg/m²)225.03 (6460.46)24.35 (4.52)0.464189.30 (5942.45)24.19 (3.89)0.751 FBG (mg/dL)108.12 (31.64)114.35 (43.13)< 0.00199.47 (11.05)103.73 (14.89)< 0.001 TG (mg/dL)128.53 (93.17)142.22 (110.74)0.002118.50 (72.56)146.36 (94.05)< 0.001 LDL-C (mg/dL)115.94 (34.26)119.74 (36.47)0.015116.60 (33.68)121.10 (34.75)0.133 HDL-C (mg/dL)51.66 (15.12)49.64 (15.58)0.00352.47 (14.77)47.55 (14.17)< 0.001 Cumulative GLM6 22.33 (1.43)22.70 (1.45)< 0.00122.11 (1.31)22.74 (1.22)< 0.001 GLM6 dynamic clustering group Favorable1294 (31.7)122 (21.9)< 0.0011153 (30.0)20 (15.3)< 0.001 Intermediate1772 (43.5)263 (47.3)1696 (44.1)55 (42.0) Unfavorable1011 (24.8)171 (30.8)994 (25.9)56 (42.7) Gender Male1825 (44.8)224 (40.3)0.0511645 (42.8)55 (42.0)0.923 Female2252 (55.2)332 (59.7)2198 (57.2)76 (58.0) Education Level Illiterate1136 (27.9)180 (32.4)0.1481081 (28.1)41 (31.3)0.809 Primary School1710 (41.9)224 (40.3)1616 (42.1)51 (38.9) Middle School/High School/Vocational School1191 (29.2)146 (26.3)1104 (28.7)37 (28.2) College degree or higher40 (1.0)6 (1.1)42 (1.1)2 (1.5) Marital status Married3707 (90.9)504 (90.6)0.7523489 (90.8)112 (85.5)0.232 Divorce32 (0.8)4 (0.7)32 (0.8)2 (1.5) Widowhood319 (7.8)47 (8.5)306 (8.0)16 (12.2) Unmarried19 (0.5)1 (0.2)16 (0.4)1 (0.8) Smoking status No2569 (63.0)364 (65.5)0.282459 (64.0)92 (70.2)0.17 Yes1508 (37.0)192 (34.5)1384 (36.0)39 (29.8) Drinking status No2724 (66.8)379 (68.2)0.5572615 (68.0)88 (67.2)0.909 Yes1353 (33.2)177 (31.8)1228 (32.0)43 (32.8) Height (cm) 157.66 (10.06)157.73 (9.17)0.892157.60 (9.93)158.43 (8.16)0.348 Weight (kg) 58.67 (11.10)60.65 (12.26)< 0.00158.20 (11.00)60.94 (11.74)0.005 WC (cm) 83.87 (12.02)86.50 (12.99)< 0.00183.34 (11.98)85.85 (14.32)0.019 TC (mg/dL) 192.74 (37.59)197.61 (37.23)0.004191.69 (37.07)197.59 (37.91)0.074 SBP (mmHg) 128.01 (20.35)133.81 (21.66)< 0.001127.63 (20.38)129.88 (20.58)0.215 DBP (mmHg) 74.99 (11.91)77.48 (12.21)< 0.00174.82 (12.00)76.58 (12.30)0.099 HbA1c (%) 5.23 (0.73)5.36 (0.99)< 0.0015.08 (0.37)5.22 (0.40)< 0.001 Obesity No2455 (60.2)290 (52.2)< 0.0012392 (62.2)66 (50.4)0.008 Yes1622 (39.8)266 (47.8)1451 (37.8)65 (49.6) Hypertension No2663 (65.3)288 (51.8)< 0.0012514 (65.4)70 (53.4)0.006 Yes1414 (34.7)268 (48.2)1329 (34.6)61 (46.6)BMI, Body mass index; FBG, Fasting blood glucose; TG, Triglycerides; LDL-C, Low-density lipoprotein cholesterol; HDL-C, High-density lipoprotein cholesterol; WC, Waist circumference; TC, Total cholesterol; SBP, Systolic blood pressure; DBP, Diastolic blood pressure; HbA1c, Glycated hemoglobin Baseline characteristics of the CHARLS cohort at the time of enrollment in 2011 BMI, Body mass index; FBG, Fasting blood glucose; TG, Triglycerides; LDL-C, Low-density lipoprotein cholesterol; HDL-C, High-density lipoprotein cholesterol; WC, Waist circumference; TC, Total cholesterol; SBP, Systolic blood pressure; DBP, Diastolic blood pressure; HbA1c, Glycated hemoglobin In the baseline characteristics of the study population stratified by the incidence of future CVD or DM, participants who were later diagnosed with CVD or DM had higher GLM6 scores, higher FBG and TG levels, lower HDL-C levels, more severe obesity, and more severe hypertension, compared with participants who did not develop CVD or DM. In terms of cumulative exposure, participants who developed CVD or DM during follow-up had significantly higher cumulative GLM6 levels than those who did not develop CVD or DM. With regard to dynamic clustering analysis, compared with participants who did not develop CVD or DM, those who developed CVD or DM during follow-up demonstrated worse control of GLM6 levels and were more likely to belong to the “poor GLM6 dynamic clustering group.” Associations between GLM6 and CVD and DM in the cohort The multivariable logistic regression analysis showed that GLM6 and cumulative GLM6 were significantly positively associated with increased CVD and DM risk across all models (all P < 0.001). GLM6 and cumulative GLM6 were grouped into tertiles, with the lowest tertile used as the reference group. In all models, the highest tertiles of GLM6 and cumulative GLM6 were significantly positively associated with incident CVD and DM. Using the favorable GLM6 dynamic clustering group as the reference, participants in the unfavorable GLM6 dynamic clustering group showed a significant positive association with CVD and DM risk across all models (Table 5). Table 5Associations between baseline GLM6, cumulative GLM6, and GLM6 dynamic clustering group with CVD and DMUnadjusted modelModel 1Model 2OR (95% CI) P OR (95% CI) P OR (95% CI) P CVD GLM61.28 (1.17–1.39)< 0.0011.27 (1.17–1.38)< 0.0011.26 (1.16–1.38)< 0.001Cumulative GLM61.29 (1.18–1.40)< 0.0011.28 (1.17–1.39)< 0.0011.27 (1.17–1.39)< 0.001 DM GLM61.43 (1.24–1.65)< 0.0011.43 (1.24–1.65)< 0.0011.44 (1.25–1.67)< 0.001Cumulative GLM61.57 (1.33–1.85)< 0.0011.57 (1.33–1.85)< 0.0011.59 (1.35–1.87)< 0.001 CVD GLM6T1Ref.Ref.Ref.T21.60 (1.27–2.02)< 0.0011.59 (1.26–2.01)< 0.0011.57 (1.24–1.99)< 0.001T31.96 (1.56–2.46)< 0.0011.93 (1.54–2.43)< 0.0011.91 (1.52–2.41)< 0.001Cumulative GLM6T1Ref.Ref.Ref.T21.46 (1.15–1.84)0.0021.45 (1.15–1.82)0.0021.44 (1.14–1.82)0.002T31.82 (1.45–2.28)< 0.0011.79 (1.43–2.24)< 0.0011.77 (1.41–2.22)< 0.001GLM6 dynamic clustering groupFavorableRef.Ref.Ref.Intermediate1.57 (1.26–1.97)< 0.0011.56 (1.24–1.95)< 0.0011.55 (1.23–1.95)< 0.001Unfavorable1.79 (1.40–2.30)< 0.0011.76 (1.38–2.26)< 0.0011.74 (1.36–2.24)< 0.001 DM GLM6T1Ref.Ref.Ref.T21.65 (0.99–2.76)0.061.65 (0.99–2.77)0.061.65 (0.98–2.77)0.06T32.94 (1.83–4.71)< 0.0012.94 (1.83–4.72)< 0.0013.00 (1.86–4.83)< 0.001Cumulative GLM6T1Ref.Ref.Ref.T21.85 (1.11–3.10)0.021.86 (1.11–3.11)0.021.83 (1.09–3.06)0.02T32.97 (1.84–4.80)< 0.0012.98 (1.84–4.83)< 0.0013.05 (1.87–4.96)< 0.001GLM6 dynamic clustering groupFavorableRef.Ref.Ref.Intermediate1.87 (1.11–3.14)0.021.88 (1.12–3.15)0.021.85 (1.10–3.11)0.02Unfavorable3.25 (1.94–5.45)< 0.0013.26 (1.94–5.48)< 0.0013.34 (1.98–5.64)< 0.001Model 1 adjusts for genderModel 2 adjusts for gender, educational level, marital status, smoking status, and drinking status Table 5Associations between baseline GLM6, cumulative GLM6, and GLM6 dynamic clustering group with CVD and DMUnadjusted modelModel 1Model 2OR (95% CI) P OR (95% CI) P OR (95% CI) P CVD GLM61.28 (1.17–1.39)< 0.0011.27 (1.17–1.38)< 0.0011.26 (1.16–1.38)< 0.001Cumulative GLM61.29 (1.18–1.40)< 0.0011.28 (1.17–1.39)< 0.0011.27 (1.17–1.39)< 0.001 DM GLM61.43 (1.24–1.65)< 0.0011.43 (1.24–1.65)< 0.0011.44 (1.25–1.67)< 0.001Cumulative GLM61.57 (1.33–1.85)< 0.0011.57 (1.33–1.85)< 0.0011.59 (1.35–1.87)< 0.001 CVD GLM6T1Ref.Ref.Ref.T21.60 (1.27–2.02)< 0.0011.59 (1.26–2.01)< 0.0011.57 (1.24–1.99)< 0.001T31.96 (1.56–2.46)< 0.0011.93 (1.54–2.43)< 0.0011.91 (1.52–2.41)< 0.001Cumulative GLM6T1Ref.Ref.Ref.T21.46 (1.15–1.84)0.0021.45 (1.15–1.82)0.0021.44 (1.14–1.82)0.002T31.82 (1.45–2.28)< 0.0011.79 (1.43–2.24)< 0.0011.77 (1.41–2.22)< 0.001GLM6 dynamic clustering groupFavorableRef.Ref.Ref.Intermediate1.57 (1.26–1.97)< 0.0011.56 (1.24–1.95)< 0.0011.55 (1.23–1.95)< 0.001Unfavorable1.79 (1.40–2.30)< 0.0011.76 (1.38–2.26)< 0.0011.74 (1.36–2.24)< 0.001 DM GLM6T1Ref.Ref.Ref.T21.65 (0.99–2.76)0.061.65 (0.99–2.77)0.061.65 (0.98–2.77)0.06T32.94 (1.83–4.71)< 0.0012.94 (1.83–4.72)< 0.0013.00 (1.86–4.83)< 0.001Cumulative GLM6T1Ref.Ref.Ref.T21.85 (1.11–3.10)0.021.86 (1.11–3.11)0.021.83 (1.09–3.06)0.02T32.97 (1.84–4.80)< 0.0012.98 (1.84–4.83)< 0.0013.05 (1.87–4.96)< 0.001GLM6 dynamic clustering groupFavorableRef.Ref.Ref.Intermediate1.87 (1.11–3.14)0.021.88 (1.12–3.15)0.021.85 (1.10–3.11)0.02Unfavorable3.25 (1.94–5.45)< 0.0013.26 (1.94–5.48)< 0.0013.34 (1.98–5.64)< 0.001Model 1 adjusts for genderModel 2 adjusts for gender, educational level, marital status, smoking status, and drinking status Associations between baseline GLM6, cumulative GLM6, and GLM6 dynamic clustering group with CVD and DM Model 1 adjusts for gender Model 2 adjusts for gender, educational level, marital status, smoking status, and drinking status RCS analysis showed no nonlinear association between GLM6 or cumulative GLM6 and CVD or DM in any model (Fig. 6, Supplementary Figure 15). Fig. 6Nonlinear associations between baseline GLM6, cumulative GLM6, with incident CVD and DM. (A) RCS curves for GLM6 and incident CVD; (B) RCS curves for cumulative GLM6 and incident CVD; (C) RCS curves for GLM6 and incident DM; (D) RCS curves for cumulative GLM6 and incident DM. The RCS curve is based on the logistic regression Model 2. Model 2 adjusts for gender, educational level, marital status, smoking status, and drinking status Nonlinear associations between baseline GLM6, cumulative GLM6, with incident CVD and DM. (A) RCS curves for GLM6 and incident CVD; (B) RCS curves for cumulative GLM6 and incident CVD; (C) RCS curves for GLM6 and incident DM; (D) RCS curves for cumulative GLM6 and incident DM. The RCS curve is based on the logistic regression Model 2. Model 2 adjusts for gender, educational level, marital status, smoking status, and drinking status Results of subgroup and sensitivity analyses for cohort GLM6 with CVD and DM The cohort was stratified by age, gender, smoking status, and drinking status. In the age subgroup analysis, compared with participants aged 60 years or older, participants younger than 60 years had higher CVD and DM risk as GLM6 and cumulative GLM6 increased, suggesting greater health risks associated with high GLM6 exposure in the younger population. In the gender subgroup analysis, compared with male, female had lower CVD risk but higher DM risk as GLM6 and cumulative GLM6 increased, suggesting that male may show stronger CVD-related associations with high GLM6, whereas female may show stronger DM-related associations. For smoking status, the association of increasing GLM6 and cumulative GLM6 with CVD was generally similar between smokers and non-smokers. For drinking status, drinkers had higher DM risk as GLM6 and cumulative GLM6 increased (Table 6). Table 6Subgroup analysis results for the baseline GLM6, cumulative GLM6 with CVD and DMOutcome variables/ Exposure variablesSubgroupOR (95% CI) P P Interaction CVD Age GLM6< 601.37(1.2, 1.55)< 0.0010.487≥ 601.18(1.05, 1.33)0.005Cumulative GLM6< 601.32(1.17, 1.49)< 0.0010.986≥ 601.22(1.08, 1.39)0.002 DM Age GLM6< 601.76(1.4,2.21)< 0.0010.073≥ 601.27(1.05,1.54)0.016Cumulative GLM6< 601.83(1.46,2.3)< 0.0010.022≥ 601.32(1.03,1.7)0.030 CVD Gender GLM6Male1.33(1.16, 1.52)< 0.0010.425Female1.24(1.11, 1.38)< 0.001Cumulative GLM6Male1.32(1.15, 1.51)< 0.0010.563Female1.25(1.12, 1.4)< 0.001 DM Gender GLM6Male1.36(1.08,1.71)0.0090.575Female1.48(1.23,1.78)< 0.001Cumulative GLM6Male1.4(1.09,1.81)0.0100.256Female1.7(1.37,2.11)< 0.001 CVD Smoking status GLM6Smoking1.28(1.11, 1.47)0.0010.276No smoking1.27(1.15, 1.41)< 0.001Cumulative GLM6Smoking1.28(1.11, 1.49)0.0010.528No smoking1.28(1.15, 1.43)< 0.001 DM Smoking status GLM6Smoking1.37(1.06,1.76)0.0150.483No smoking1.46(1.23,1.74)< 0.001Cumulative GLM6Smoking1.44(1.07,1.93)0.0170.958No smoking1.62(1.33,1.97)< 0.001 CVD Drinking status GLM6Drinking1.31(1.13, 1.52)< 0.0010.675No drinking1.26(1.14, 1.4)< 0.001Cumulative GLM6Drinking1.3(1.12, 1.52)0.0010.822No drinking1.28(1.15, 1.42)< 0.001 DM Drinking status GLM6Drinking1.66(1.26,2.18)< 0.0010.204No drinking1.34(1.13,1.6)0.001Cumulative GLM6Drinking1.83(1.38,2.42)< 0.0010.205No drinking1.46(1.19,1.79)< 0.001 Table 6Subgroup analysis results for the baseline GLM6, cumulative GLM6 with CVD and DMOutcome variables/ Exposure variablesSubgroupOR (95% CI) P P Interaction CVD Age GLM6< 601.37(1.2, 1.55)< 0.0010.487≥ 601.18(1.05, 1.33)0.005Cumulative GLM6< 601.32(1.17, 1.49)< 0.0010.986≥ 601.22(1.08, 1.39)0.002 DM Age GLM6< 601.76(1.4,2.21)< 0.0010.073≥ 601.27(1.05,1.54)0.016Cumulative GLM6< 601.83(1.46,2.3)< 0.0010.022≥ 601.32(1.03,1.7)0.030 CVD Gender GLM6Male1.33(1.16, 1.52)< 0.0010.425Female1.24(1.11, 1.38)< 0.001Cumulative GLM6Male1.32(1.15, 1.51)< 0.0010.563Female1.25(1.12, 1.4)< 0.001 DM Gender GLM6Male1.36(1.08,1.71)0.0090.575Female1.48(1.23,1.78)< 0.001Cumulative GLM6Male1.4(1.09,1.81)0.0100.256Female1.7(1.37,2.11)< 0.001 CVD Smoking status GLM6Smoking1.28(1.11, 1.47)0.0010.276No smoking1.27(1.15, 1.41)< 0.001Cumulative GLM6Smoking1.28(1.11, 1.49)0.0010.528No smoking1.28(1.15, 1.43)< 0.001 DM Smoking status GLM6Smoking1.37(1.06,1.76)0.0150.483No smoking1.46(1.23,1.74)< 0.001Cumulative GLM6Smoking1.44(1.07,1.93)0.0170.958No smoking1.62(1.33,1.97)< 0.001 CVD Drinking status GLM6Drinking1.31(1.13, 1.52)< 0.0010.675No drinking1.26(1.14, 1.4)< 0.001Cumulative GLM6Drinking1.3(1.12, 1.52)0.0010.822No drinking1.28(1.15, 1.42)< 0.001 DM Drinking status GLM6Drinking1.66(1.26,2.18)< 0.0010.204No drinking1.34(1.13,1.6)0.001Cumulative GLM6Drinking1.83(1.38,2.42)< 0.0010.205No drinking1.46(1.19,1.79)< 0.001 Subgroup analysis results for the baseline GLM6, cumulative GLM6 with CVD and DM The study further examined the associations of GLM6 tertiles, cumulative GLM6 tertiles, and GLM6 dynamic clustering groups with CVD and DM risk in subgroups (Supplementary Table 14). Across age subgroups, the highest CVD and DM risks were observed among participants in the highest GLM6 tertile, the highest cumulative GLM6 tertile, and the unfavorable GLM6 dynamic clustering group. These associations were stronger among participants younger than 60 years, suggesting that GLM6 control levels in this age group require particular attention. Regarding gender, the risk increase among female in the highest cumulative GLM6 group and the unfavorable GLM6 dynamic clustering group was significantly higher than that among male, suggesting greater sensitivity to GLM6-related DM risk in female. Regarding smoking status, non-smokers exposed to high GLM6 levels still showed significant health risks. Regarding drinking status, alcohol consumption amplified GLM6-related DM risk; among participants in the highest GLM6 tertile, the highest cumulative GLM6 tertile, and the unfavorable GLM6 dynamic clustering group, DM risk was significantly higher among drinkers than among non-drinkers, indicating that drinkers require stricter GLM6 control. Overall, GLM6-related indicators showed generally consistent positive associations with CVD and DM across subgroups. The effect sizes for cumulative GLM6 and GLM6 dynamic clustering groups were higher than those for a single GLM6 measurement, supporting the relevance of reducing cumulative GLM6 exposure and maintaining long-term GLM6 control for lowering CVD and DM risk. Sensitivity analyses showed that results based on data without imputation/interpolation and the Model 3 results were consistent with the main model, with associations remaining significantly positive (Supplementary Table 15, Supplementary Table 16). These findings further demonstrate the stability of the associations between long-term dynamic GLM6 indicators and CVD and DM. To further explore GLM6 threshold boundaries, the study evaluated the associations between dichotomized GLM6-related indicators and outcomes. The high group or unfavorable clustering group was significantly positively associated with higher CVD and DM risk, indicating that the associations of GLM6-related indicators with outcomes were robust across grouping strategies (Supplementary Table 17). Compared with the three-level grouping results, the two-level grouping produced larger changes in effect estimates and may provide an exploratory reference for future risk-stratification thresholds. Discussion The global burden of chronic diseases is currently severe, and the shortcomings of traditional health screening systems—such as “isolated indicators, late-stage detection, and fragmented specialties”—are becoming increasingly apparent, making it difficult to detect metabolic disorders in individuals at an early stage. Routine health checkups are a critical component of primary prevention, and the efficient integration of health screening data facilitates early risk stratification in the population [23]. There is a research team that proposes GLM7 as a composite index for early warning and cross-population assessment of overall metabolic risk, to address the challenges of chronic disease prevention and control in this era [9]. However, in real-world clinical settings, there is limited measurement of insulin levels in GLM7, or data is often missing, which severely compromises the reliability of biomarkers that rely on these data [24]. Current research indicates that many insulin resistance biomarkers face challenges regarding their robustness when confronted with missing data and insufficient clinical validation [25]. To further expand the applicability of the indicators, we investigated the use of a composite index based on GLM6 (excluding insulin levels) for disease risk assessment. The results indicate that GLM7 and GLM6 are significantly positively correlated with 9 disease categories, 14 specific diseases, and mortality. Compared with traditional single-disease biomarkers, the association of the composite biomarkers GLM7 and GLM6 with multiple diseases and mortality suggests that they can simultaneously aid in the early screening and risk stratification of various diseases, thereby helping to improve the utilization of healthcare resources and alleviate the financial burden of medical costs. Furthermore, the results show that higher levels of GLM7 and GLM6 are associated with a higher risk of death and shorter survival time. This suggests that changes in GLM7 and GLM6 may reflect organ dysfunction or worsening pathophysiological conditions. These markers can serve as simple indicators for stratifying mortality risk, which facilitates health interventions for high-risk populations and helps improve their life expectancy. The results of the XGBoost model indicate that both GLM7 and GLM6 demonstrate the highest predictive value for endocrine and metabolic diseases, hepatobiliary diseases, and cardiovascular category diseases. The results of the DCA analysis for disease risk prediction models show that the models for cardiovascular category and metabolic diseases have relatively high clinical net benefit values, indicating that GLM6 and GLM7 are suitable for further clinical validation. When validated using the external CHARLS database, the model continued to demonstrate sound predictive performance for metabolic and cardiovascular category diseases. This suggests that GLM6 has potential for transferable application in broad cardiovascular and metabolic risk stratification across heterogeneous populations. However, there are notable differences between NHANES and CHARLS. NHANES covers a broader range of ages and racial/ethnic groups, and some outcomes are based on laboratory tests or physical examination results. CHARLS primarily focuses on middle-aged and older adults in China, and some disease outcomes rely more heavily on self-reported medical diagnoses. These differences may lead to misclassification of outcomes, variations in diagnostic sensitivity, and model calibration bias, thereby affecting the AUC, sensitivity, and specificity in the CHARLS external validation. Therefore, the CHARLS validation in this study is better interpreted as supportive external validation within a heterogeneous cohort rather than as rigorous validation under fully homogeneous conditions. The GLM6 index shows the potential to serve as a marker for identifying individuals at high risk of cardiovascular and metabolic diseases during large-scale population screening. Although the GLM6 model does not include insulin as a parameter, its core components effectively capture the pathophysiological mechanisms underlying peripheral insulin resistance caused by visceral fat accumulation by integrating the interaction between TG and blood glucose in the context of dysglucidemia and dyslipidemia [26]. And studies have shown that single-point insulin levels are subject to significant fluctuations due to pulsatile secretion and hepatic clearance [27]. As a derived composite indicator, GLM6 provides more stable metabolic feedback. Therefore, when assessing mortality risk, although GLM6 includes fewer variables, it is still able to maintain or even improve its ability to distinguish mortality risk. It is undeniable that the GLM7 model offers advantages in certain scenarios. The study findings indicate that when predicting MAFLD and NAFLD, the GLM6 model exhibits lower accuracy and specificity than the GLM7 model. Insulin levels serve as a direct marker of insulin resistance. Since insulin resistance is the core pathophysiological mechanism underlying MAFLD and NAFLD, direct measurement of insulin levels provides a clear indication of the degree of insulin resistance and offers greater accuracy. Therefore, the GLM7 marker demonstrates greater accuracy in identifying individuals at hidden high risk—such as those with prediabetes who exhibit compensatory elevations in insulin levels but whose blood glucose remains within the normal range—and provides a more precise clinical assessment. When screening for insulin resistance-related metabolic disorders, the GLM7 marker should be prioritized to reduce the rate of missed diagnoses. Since the GLM6 index requires only standard health examination parameters, it is particularly suitable for preliminary risk stratification among individuals undergoing routine health checkups, as well as for primary care settings and scenarios where insulin testing is not available. When conducting large-scale population-based screening for common chronic diseases, the GLM6 index can be used to balance cost and effectiveness. To further provide a stronger scientific basis for the potential application of the GLM6 index in population screening, the study conducted a multidimensional validation analysis using data from the CHARLS cohort. In the predictive models constructed using cross-sectional data from NHANES and CHARLS, GLM6 demonstrated the best predictive performance for cardiovascular and metabolic diseases, with the highest performance specifically for DM among metabolic diseases. This indicates that GLM6 places a very high weight on the explanatory power of risk signals for CVD and DM. This study used a cohort follow-up design to investigate the association between long-term exposure to GLM6 and CVD and DM. The results indicate that long-term high exposure to GLM6 increases the risk of new-onset CVD and DM, leading to a loss of healthy life years. Even after excluding and adjusting for insulin markers, GLM6 still demonstrates excellent ability to identify early risks of CVD and DM. The GLM6 indicator optimizes the cost-benefit ratio without compromising accuracy, offering a more efficient approach to the early identification of DM and CVD risks. Since insulin testing is not feasible in resource-limited areas, it provides a superior option for disease early warning in such regions and holds greater potential for widespread adoption [28]. Compared to the GLM7 index, the model based on the composite GLM6 index is more readily applicable to clinical practice. However, the current study results primarily support its use as a candidate marker for future risk stratification and population screening studies. It cannot yet be directly used as a threshold for clinical diagnosis or treatment decisions. If a multicenter prospective validation study is conducted in the future, this indicator could be more easily integrated into hospital electronic medical record systems, facilitating the early identification of risk factors and the implementation of preventive management. In the cohort study, the effect size for the two-group classification using GLM6 showed a significant increase compared to the three-group classification. The crossover point between the two clusters identified by dynamic clustering was around 5.5. This suggests that future research could further explore and validate the GLM6 threshold. However, these cutoff points should currently be interpreted as empirical risk stratification points rather than established clinical diagnostic thresholds. The significant positive correlation between long-term elevated GLM6 levels and disease risk suggests that the value of dynamic monitoring of GLM6 in long-term risk management warrants further evaluation in the future. The cumulative GLM6 score and GLM6 control status can serve as research indicators for assessing the quality of long-term metabolic control, providing a reference for future studies on personalized prevention strategies. Although there is currently a lack of research on the mechanisms linking the GLM6 marker to disease onset, we can reduce the dimensionality and explore the underlying mechanisms by starting with single-marker or lower-dimensional markers. Research has found that body fat influences the association between glucose metabolism disorders and heart failure [29], and that as metabolic risk factors worsen before the onset of DM, FBG levels may undergo abrupt changes [30]. TG and HDL-C levels are associated with insulin resistance, which is characterized by impaired glucose uptake and oxidation, as well as reduced glycogen synthesis [31]. During insulin resistance, an imbalance in glucose metabolism triggers oxidative stress and inflammatory responses, which in turn damage cells and may alter systemic lipid metabolism, leading to dyslipidemia. These various metabolic changes can contribute to the development of CVD and DM [31, 32]. The GLM6 marker may influence the development of CVD and DM through various changes in glucose and lipid metabolism. However, the detailed mechanisms of action require further investigation. This study also has certain limitations. First, the external validation of CHARLS is limited by the population structure, the definition of outcomes, and the method of weighting. This study did not use sampling weights in the CHARLS external validation. Therefore, the results reflect the model’s performance within the sample included in the analysis and do not directly represent the weighted predictive performance among the general population of Chinese middle-aged and older adults. At the same time, NHANES and CHARLS differ in terms of laboratory testing protocols and disease diagnosis methods. For some diseases in CHARLS, diagnosis relies primarily on self-reported physician diagnoses, which may be subject to recall bias, underreporting, and misclassification of outcomes. These factors may all affect cross-cohort comparability and the interpretation of model performance. Second, the study is subject to outcome reporting bias and survivor bias. Self-reported outcomes may be subject to recall bias and reporting bias. Individuals who do not undergo regular health checkups are particularly prone to underreporting. Furthermore, the study cohort consisted of participants who completed the baseline survey and survived throughout the entire study period. Individuals with poorer baseline health or those already suffering from severe complications—who are considered high-risk—may have been excluded from the final analysis due to loss to follow-up or early withdrawal. This may lead the study to underestimate the true strength of the association between the GLM6 marker and adverse outcomes. The impact of this bias is particularly significant in longitudinal cohort studies involving middle-aged and older adults [33]. Finally, the findings of this study lack supporting evidence for clinical translation. Further multicenter validation studies are needed to establish standardized testing methods for the various GLM6 markers and clinical cutoff values for different populations, thereby facilitating clinical translation. Conclusion The results of this study indicate that the composite indices GLM6 and GLM7 show significant positive correlations with nine disease categories, 14 specific diseases, and mortality rates. The XGBoost model developed demonstrates sound predictive capabilities for cardiovascular category and metabolic diseases. The GLM6 index shows the potential to serve as a marker for identifying individuals at high risk of cardiovascular category and metabolic diseases during large-scale population screening. In the cohort study, the group with the highest cumulative GLM6 index tertile and unfavorable dynamic clustering group exhibited the highest risk of new-onset CVD and DM. This highlights the high risk of CVD and DM incidence associated with long-term high GLM6 exposure. The significant positive correlation between the GLM6 dynamic index and the risk of new-onset CVD and DM suggests that the GLM6 index can help establish a long-term disease risk monitoring network. Following future multicenter validation studies, it is expected to aid in optimizing clinical decision-making and provide a reference for subsequent research on personalized prevention strategies. Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1 Supplementary Material 1 Supplementary Material 2 Supplementary Material 2