The data were from NHANES, which offers a comprehensive health and nutrition evaluation of noninstitutionalized civilians in the United States through a stratified, multi-stage, probabilistic cluster design that ensures national representation. The National Center for Health Statistics (NCHS) in the U.S. manages NHANES, involving in-person interviews, physical examinations, and laboratory tests. The NCHS Disclosure Review Board approved the survey methodologies. The NCHS Ethics Review Board granted ethical clearance, and participants submitted written informed consent to the NCHS Ethics Review Board. Protocol details are available at https://www.cdc.gov/nchs/ahcd/ahcd_confidentiality.htm. The Ethics Committee of Shanghai Tongji Hospital provided ethical approval for this study. Data from seven NHANES cycles (1999–2018) included a total of 101,316 individuals initially. Due to missing clinical data, unknown medical history, and incomplete LE8 data, the number of participants aged 20 years and older eligible for statistical analysis in the NHANES cycles (2007-2018) was 7,739 ( Supplementary Figure 1 ).

The LE8 system assesses CVH by considering four health behavior scores: diet (as measured by the 2015 Healthy Diet Index from 24-hour recalls), physical activity, nicotine exposure, and sleep. It also incorporates four health factor scores: BMI, blood lipids, blood glucose, and blood pressure (13). Blood glucose was not included in the analysis as a LE8 metric, as prediabetes and diabetes were the outcomes of interest (19). Each indicator in the LE8 system is scored on a scale from 0 to 100, and the overall score is calculated as the mean of the scores. Detailed information is available in Supplementary Table 1 . In accordance with the guidelines of the American Heart Association (AHA), LE8, including health behavior and health factors, was analyzed continuously and categorically by tertiles: high (80–100), moderate (50–79), and low (0–49) (13).

Prediabetes and diabetes were defined according to the 2003 American Diabetes Association (ADA) criteria (20). Diabetes was defined by one of the following criteria: (1) fasting plasma glucose (FPG) ≥126 mg/dL (7mmol/L); (2) self-reported use of hypoglycemic medications; (3) 2 h plasma glucose (2hPG) ≥200 mg/dL (11.1mmol/L); (4) self-reported diabetes. After exclusion of diabetic subjects based on the above criteria, isolated impaired glucose tolerance (iIGT) was defined as FPG <100 mg/dL (5.6mmol/L) and 2hPG 140–199 mg/dL (7.8–11mmol/L); isolated impaired fasting glucose (iIFG) was defined as FPG 100–125 mg/dL (5.6–6.9mmol/L) and 2hPG <140 mg/dL(7.8mmol/L); both IFG and IGT (IFG+IGT) were FPG 100–125 mg/dL (5.6–11mmol/L)and 2hPG 140–199 mg/dL (7.8–11mmol/L); normal glucose metabolism (NGM) referred to FBG <100 mg/dL (5.6mmol/L) and 2hPG <140 mg/dL (7.8mmol/L).

Covariates included age, sex (male and female), race and ethnicity (non-Hispanic black, non-Hispanic white, Mexican American, or other races), poverty-to-income ratio (PIR) [low (<1.3), middle (1.3–3.5), and high (≥3.5)], marital status (divorced/separated/widowed, married/living with partner and never married), and education level (college graduate or above, high school or below, and college level). Insulin resistance (IR) was assessed using the homeostasis model of IR (HOMA-IR)=(fasting plasma insulin [μU/mL])×(fasting plasma glucose [mmol/L])÷22.5 (21).

NHANES employed design weighting to ensure the representativeness of the data. We performed weighted data analyses using relevant survey weights (MEC2yr) to generate nationally representative estimates. After applying the NHANES survey weights, our analyses provided estimates that represented approximately 59,055,054 individuals in the U.S. population. Categorical variables in the baseline information were expressed as weighted percentages, while continuous variables were presented as weighted means along with their respective confidence intervals (CIs). Group disparities were assessed by weighted variance tests and weighted chi-square tests. Additionally, we computed age-standardized prevalence estimates and 95% CIs for various score levels.

We utilized weighted multivariate logistic regression to examine the relationship between CVH with both diabetes and prediabetes. This analysis was stepwise adjusted for age, sex, marital status, education level, and race/ethnicity. We also employed restricted cubic spline regression to explore potential non-linear associations between the LE8 score and its sub-scale scores with diabetes and prediabetes. Furthermore, we conducted stratified analyses by age, sex, marital status, education level, and race/ethnicity to assess whether the association between LE8 and diabetes and prediabetes was influenced by these factors. The significance of the interaction was assessed by the interaction P-value between the LE8 score and the stratified factors.

All statistical analyses were performed using R software (Version 4.2.1, The R Foundation; http://www.R-project.org) and EmpowerStats software (Version 5.0, X&Y Solutions, Inc., Boston, MA; http://www.empowerstats.com). A significance level of p < 0.05 was considered statistically significant.

Tables 1 and 2 present the demographic characteristics and metabolic risk factors of the study population, categorized according to ADA criteria. The weighted average age of participants was 47.23 years, with 48.76% male, and a majority being non-Hispanic white (73.45%). Among the 7,739 participants, 1,949 had iIFG, 1,165 had iIGT, 799 had IFG+IGT, and 537 were diagnosed with diabetes. The mean CVH score for the entire study population was 66.94 ± 0.33. Stratified by glycemic metabolism status, the mean CVH scores for NGM, iIFG, iIGT, IFG+IGT, and diabetic subjects were 70.21 ± 0.39, 64.44 ± 0.53, 63.94 ± 0.85, 60.69 ± 0.67 and 59.28 ± 0.66, respectively.

In multiple comparisons, significant differences were observed in CVH scores among NGM, prediabetes, and diabetes groups. The diabetes group exhibited significantly lower CVH scores compared to the NGM and prediabetes groups. Upon further comparison of the prediabetes subgroups, no significant difference was found in CVH scores between iIFG and iIGT, whereas the IFG+IGT group’s CVH score was significantly lower than those of the iIFG and iIGT groups (see Supplementary Figure 2 ). On average, patients with prediabetes or diabetes were older, less educated, and had poorer lipid profiles compared to those with NGM. Patients with IFG+IGT exhibited more pronounced metabolic deficits and more unfavorable CVD risk profiles, such as higher HOMA-IR, larger waist circumference, and altered lipid metabolism indices, compared to those with iIFG and iIGT (see Table 2 ).

When CVH was further subdivided into health behaviors (diet, physical activity, nicotine exposure, and sleep) and health factors (body mass index, blood lipids, and blood pressure), significant differences in scores were observed between the NGM, iIFG, iIGT, IFG+IGT, and diabetic groups ( Table 2 ). The mean health behavior scores for the NGM, iIFG, iIGT, IFG+IGT, and diabetic subjects were 67.89 ± 0.56, 65.88 ± 0.72, 65.46 ± 1.26, 65.33 ± 0.90, and 63.69 ± 0.87, respectively; mean health factor scores were 73.31 ± 0.45, 62.53 ± 0.64, 61.90 ± 1.29, 54.49 ± 0.91, and 53.41 ± 1.15, respectively. Compared to patients with prediabetes (including iIFG, iIGT, IFG+IGT) or diabetes, those with NGM had higher health behaviors and health factors scores, except for nicotine exposure and sleep health scores, which were not statistically different.

The results of the logistic regression analysis of glucose metabolism status are summarized in Tables 3 and 4 . This comprehensive evaluation revealed a robust inverse correlation between CVH scores and both prediabetes and diabetes. After adjusting for the variables of age, sex, ethnicity, marital status, PIR, and education, for every 10-point increase in CVH scores, the odds ratios (ORs) for each category were reduced as follows: iIFG (OR 0.77, 95% CI 0.72–0.82), iIGT (OR 0.76, 95% CI 0.69–0.84), IFG+IGT (OR 0.65, 95% CI 0.60–0.70), prediabetes (OR 0.74, 95% CI 0.70–0.78), and diabetes (OR 0.63, 95% CI 0.57–0.69). Notably, the inverse correlation between IFG+IGT and CVH was more pronounced than the negative association observed with iIFG or iIGT.

To further explore the relationship between CVH scores and prediabetes as well as diabetes, participants were stratified into low, medium, and high CVH groups based on CVH scores (refer to Table 3 ). Following multivariable adjustments, participants with moderate and high CVH had significantly lower odds of developing diabetes compared to those with low CVH, with ORs of 0.29 (95% CI 0.19–0.45) and 0.04 (95% CI 0.02–0.08), respectively. Participants in the moderate and high CVH groups also exhibited a significantly lower risk of iIFG, iIGT, IFG+IGT, and prediabetes compared to those in the low CVH group. Notably, the inverse association between IFG+IGT and CVH was more pronounced in the high CVH group compared to other prediabetes subgroups, with an OR of 0.13 (95% CI 0.09–0.21) per 10-point increase in CVH, second only to diabetes (OR 0.06, 95% CI 0.03, 0.12) (see Table 4 ). Logistics regression analysis of the association between CVH scores and prediabetes/diabetes showed that as the CVH scores went from low to high, the ORs for all types of glucose metabolism disorders, including iIFG, iIGT, IFG+IGT, and diabetes, then progressively decreased (p for trend <0.0001) (see Tables 3 and 4 ).

Furthermore, the age-adjusted prevalence of iIFG, iIGT, IFG+IGT, prediabetes, and diabetes showed varying decreases with increasing CVH levels, health behavior scores, and health factor scores, as illustrated in Figure 1 . When FPG and 2hPG were used as continuous glycemic measures, both FPG and 2hPG exhibited an inverse association with CVH scores as presented in Figure 2 (β=-0.21, p < 0.0001 for both).

We also focused on subgroups with abnormal glucose metabolism, including iIFG, iIGT, IFG+IGT, prediabetic, and diabetic subgroups ( Figure 3 ). CVH scores showed a negative correlation with all of these subgroups. The negative associations between CVH scores and IFG+IGT, prediabetes, and diabetes did not significantly differ across age, gender, race, and PIR subgroups. Of note, our analysis revealed that, except for the iIGT subgroup, education level influenced the negative association between CVH scores and the risk of iIFG, IFG+IGT, prediabetes, or diabetes (p < 0.05 for the interaction). The inverse association between CVH scores and the risk of these four subgroups was more pronounced among participants with education levels of college graduation or above, with ORs of 0.69 (95% CI 0.62–0.77), 0.54 (95% CI 0.47–0.62), 0.64 (95% CI 0.58–0.71), and 0.52 (95% CI 0.44–0.61) for each 10-point increase in CVH scores, respectively.

Logistic regression analysis revealed significant negative associations between both health factor scores and health behavior scores and the risk of prediabetes and diabetes (refer to Tables 3 and 4 ). The risk of iIFG, iIGT, IFG+IGT, prediabetes, and diabetes was significantly lower in the subgroups with moderate and high health factor scores compared with the subgroups with low health factor scores. However, the risk of iIFG, iIGT, IFG+IGT, prediabetes, and diabetes was not statistically different between the two subgroups with low health behavior scores versus moderate health behavior scores (see Tables 3 and 4 ). After adjusting for multiple confounding variables, individuals with high health behaviors scores had significantly lower odds of iIFG, iIGT, IFG+IGT, prediabetes, and diabetes, with ORs of 0.69 (95% CI 0.53–0.90), 0.62 (95% CI 0.42–0.92), 0.60 (95% CI 0.44–0.82), 0.66 (95% CI 0.53–0.82), and 0.49 (95% CI 0.33–0.72), respectively.

Additionally, the associations of each CVH component with prediabetes and diabetes were shown in Figure 4 . Notably, among the health factors, BMI score exhibited the strongest association with prediabetes and diabetes, with ORs of 0.86 (95% CI 0.84–0.88) and 0.80 (95% CI 0.77–0.84) per 10-point increase, respectively. Following BMI score, BP score, and blood lipids score were secondary factors influencing the association of CVH with prediabetes and diabetes. In contrast, health behaviors demonstrated weaker associations with prediabetes and diabetes compared to health factors. Unexpectedly, nicotine exposure score displayed no significant association with either prediabetes (p=0.24) or diabetes (p=0.95). When categorizing prediabetes into the three subtypes, the associations between CVH components and iIFG, iIGT, and IFG+IGT followed a similar pattern of correlation as in prediabetes. In conclusion, both health behaviors and health factors were negatively associated with prediabetes and diabetes, with the negative associations being more pronounced for health factors, especially body mass index.

As shown in Figure 5 , restricted cubic spline (RCS) analysis with multivariate adjustment revealed that iIFG, iIGT, IFG+IGT, prediabetes, and diabetes were correlated with CVH scores (all p < 0.05). Among them, IFG+IGT, prediabetes, and diabetes had a significant non-linear dose-response relationship with CVH scores (p for nonlinear <0.05). In contrast, iIFG and iIGT had a linear dose-response association with CVH scores, with p-values of 0.09 and 0.84, respectively, for the non-linear test.

The main strength of our study was that we explored, for the first time, the association between CVH, as defined by LE8, and prediabetes, including its three distinct subtypes. Additionally, we conducted extensive stratified analyses based on sociodemographic characteristics and risk factors for diabetes/prediabetes. However, this study had certain limitations. Firstly, it was challenging to establish a causal relationship between LE8 and prediabetes/diabetes owing to the cross-sectional design of the study. Secondly, individuals diagnosed with prediabetes or diabetes may have implemented necessary lifestyle changes, however, due to the cross-sectional design, we were unable to explore the causal relationship between lifestyle changes and risk of prediabetes/diabetes, which can be better explored in future large follow-up cohort studies. Thirdly, the diagnosis of diabetes partially relied on self-reporting, potentially leading to misclassification and estimation bias. Fourthly, the exclusion of participants with an unknown medical history or incomplete LE8 data might have led to a reduced study population and potential selection bias.