The study participants were from the China Cardiometabolic Disease and Cancer Cohort (4 C) Study, which is a nationwide population-based prospective cohort study consisting of community-dwelling adults aged 40 years or older [13]. The study protocol and informed consent were approved by the Committee on Human Research at Ruijin Hospital affiliated to Shanghai Jiao Tong University School of Medicine, Shanghai, China. All participants provided the written informed consent.

The details of the 4 C Study design have been described previously [14, 15]. The baseline survey was conducted between 2011 and 2012. 193,846 participants were recruited from local resident registration systems of 20 community sites representing 16 provincial administrative regions of mainland China. Eligible men and women aged ≥ 40 years were identified from local resident registration systems. Trained community health workers visited eligible individual’s homes and invited them to participate in the study. We intended to enroll both men and women, but more women participated in the study, possibly because women care more about their health and have more flexible schedule due to home working, compared to men. The follow-up visit was conducted between 2014 and 2016, and 170,240 participants (87.8%) attended. As shown in Supplemental Fig. 1, for the current analyses, we excluded 84,167 individuals with missing data on weight at age 20 years or 40 years, and 260 participants whose height, weight or BMI at 20 or 40 years was outside the valid range (height > 200 cm or height < 120 cm or weight > 150 kg or weight < 20 kg or BMI < 13 kg/m2 or BMI > 50 kg/m2), and 12,203 individuals with missing data on any metric of cardiovascular health. A total of 72,610 participants were included in the analyses. For the analyses of incident CVD events, we further excluded 1,326 participants with prior CVD events at baseline and 11,736 participants with missing data on incident CVD events during follow-up, leaving 59,548 participants. And for analyses of incident diabetes, we excluded 17,868 diabetes patients at baseline and 7,934 participants with missing data on incident diabetes during follow-up, leaving 46,808 participants. We compared the key baseline characteristics of the study participants and participants who were excluded from the analysis in Supplemental Table 1.

At baseline and follow-up visits, data collection was performed in local community clinics by trained staff following standardized protocols. Standard questionnaires were administered to collect information on demographic characteristics, education level, family history of CVD events and diabetes, medical history and lifestyle habits. Skilled nurses measured blood pressure according to standard protocols. Blood samples were collected after overnight fasting. Then a standard oral glucose tolerance test (OGTT) was conducted, and 2-hour post-load blood specimens were collected. Fasting and 2-hour post-load plasma glucose were measured at local hospitals. Serum samples were shipped in dry ice to the central laboratory, where lipids profiles and creatinine were measured using an autoanalyzer (Abbott Laboratories, IL). Glycated hemoglobin (HbA1c) was tested using finger capillary whole blood using high-performance liquid chromatography (VARIANT™ II Systems, BIO-RAD, Hercules, CA, USA) at the central laboratory.

Data on weight at age 20 years and 40 years were recalled at the baseline survey. Baseline height and body weight were measured according to a standard protocol. BMI at age 20 and 40 years was calculated as self-reported weight (kg) divided by squared height (m2) at baseline.

Absolute weight change over 20 kg was redefined to 20 kg. We classified absolute weight change from age 20 to 40 years into five groups: weight loss > 2.5 kg, weight loss ≤ 2.5 kg or gain < 2.5 kg (reference group), weight gain between ≥ 2.5 kg and < 5.0 kg, weight gain between ≥ 5.0 kg and < 10.0 kg, and weight gain ≥ 10 kg. On the basis of a study by Chen et al [16], we also defined five BMI change patterns using BMI at age 20 and 40 years: stable normal pattern (< 23.0 at both times), maximum overweight pattern (23.0-24.9 at either time but not ≥ 25.0 at the other time), obese to non-obese pattern (≥ 25.0 at 20 years and < 25.0 at 40 years), non-obese to obese pattern (< 25.0 at 20 years and ≥ 25.0 at 40 years), and stable obesity (≥ 25.0 at both times). We used BMI cut points more suitable for Asian and Chinese in our study [17].

According to the updated definition proposed by AHA in 2022 [12], we redefined and scored the baseline CVH score based on 8 components, including nicotine exposure, physical activity, diet, sleep, BMI, blood lipids, blood glucose, and blood pressure. Each of the 8 CVH metrics was scored from 0 to 100, with 100 points representing ideal cardiovascular health metrics (ICVHM)). The overall CVH score was calculated as the average of 8 component metric scores. Participants with overall CVH scores of 80–100 were considered to have high CVH, while moderate CVH referred to 50–79 and low CVH referred to 0–49. Detailed methods and scoring criteria referred to AHA Presidential Advisory were presented in Supplemental Table 2.

Information on CVD events was obtained from hospital records. Two members of the outcome adjudication committee, masked to the baseline characteristics, independently reviewed all medical material and defined CVD events as a composite of non-fatal myocardial infarction, non-fatal stroke, hospitalized or treated heart failure, and cardiovascular deaths. Discrepancies were resolved by discussion involving other members of the committee.

At the follow-up visit, we repeated OGTT and blood sample collection to assess fasting and 2-hour post-load plasma glucose, and HbA1c. Incident diabetes was diagnosed according to the ADA 2022 criteria [18]: fasting plasma glucose ≥ 126 mg/dL, or OGTT 2-hour post-load plasma glucose ≥ 200 mg/dL, or HbA1c ≥ 6.5%, or a diagnosis by physicians during follow-up.

According to weight change categories from age 20 years to 40 years, baseline characteristics of participants were expressed as means with standard deviations (SD) for continuous variables and numbers with percentages for categorical variables.

The association of weight change categories and BMI change patterns with the risk of incident clinical outcomes was evaluated using Cox proportional hazards models to generate hazard ratio (HR) and 95% confidence interval (CI). In the crude model 1, we adjusted for age, sex and weight at age 20 years (only in weight change analysis). Model 2 further adjusted for education level, CVD events or diabetes family history. Model 3 fully adjusted for individual CVH metric scores, except for BMI score, based on model 2. We also depicted the association between weight change from age 20 years to 40 years and risk for outcomes by restricted cubic splines and calculated the non-linear P value, adjusting for covariates in model 3. Furthermore, we investigated the associations of weight change categories and overall CVH status with incident diabetes and CVD events in which we divided all participants into 13 categories as weight change within 2.5 kg and other weight change patterns with 3 degrees of CVH (high, moderate, and low). Besides, we estimated the associations of weight gain ≥ 10 kg and number of ICVHMs (without ideal BMI) with incident CVD events and diabetes.

Five sensitivity analyses were performed. Considering BMI in the CVH metrics may have significant effect on weight change analysis, a sensitivity analysis was performed to test the combined association of weight change categories and CVH status estimated by 7 metrics without BMI with outcomes (Supplemental Fig. 2). To test the impact of missingness and exclusion of participants on the results, we conducted multiple imputation by fully conditional specification and generated 10 imputed datasets [19], and the results were pooled by Rubin’s Rule. The results were shown in Supplemental Tables 4 and Supplemental Fig. 3. We also performed another complementary analysis to test the combined association of weight change categories and CVH status estimated by 7 metrics without glucose with diabetes (Supplemental Fig. 4). In addition, we evaluated the association between weight gain ≥ 10 kg and outcomes in subgroups of individual CVH metric and further tested their interactions by including the product term (exposure × stratification variable) in the models (Supplemental Fig. 5). We also evaluated the association between weight change and outcomes in females and males and tested their interactions (Supplemental Table 5).

We used SAS software (version 9.4), and R software (version 4.1.1) for statistical analyses. All reported P values are nominal. Statistical significance was a two-tailed P < 0.05.