The present study used a subset of participants from the Chinese Multi-Provincial Cohort Study [17, 18]. A stratified random sampling for each sex and 10-year age group was performed. Overall, in 1992, a group of 1450 individuals aged 35–64 years received health survey in an urban community of Chengdu, Sichuan province, China. In 2007, we conducted another health survey on the same group of participants. The two surveys were supported by a project from the National Eighth Five-Year Research Plan and megaprojects of science research for China's 11th 5-year plan, respectively. Among the 1450 individuals, 711 individuals received an interview health survey in 2007, and telephone follow-ups were conducted for the remaining individuals (n = 518). After excluding the individuals who were lost to follow-up and the obese individuals (body mass index, BMI ≥ 28 kg/m2) , a total of 1157 nonobese participants with complete data were analyzed (Fig. 1). Other detailed information of these participants has been reported elsewhere [17, 18, 20].
The surveys were approved by the Ministry of Health of China, as well as by the Ethics Committee of West China Hospital of Sichuan University. The study protocol conforms to the ethical guidelines of the Declaration of Helsinki. All participants provided written informed consent.
At baseline in 1992, the survey content included standardized questionnaire, anthropometric measurements, and laboratory tests. Standardized questionnaire collected the information on demographic characteristics, such as age, sex, etc. Based on the standard methods , we performed anthropometric measurements, which included blood pressure, height, weight, waist circumference, hip circumference. Laboratory tests consisted of fasting plasma glucose (FPG) and fasting lipid profiles, including triglycerides, total cholesterol, high density lipoprotein-cholesterol (HDL-C), and low-density lipoprotein cholesterol (LDL-C).
According to the original study, the criteria for the new MH definition are as follows: 1) systolic blood pressure (SBP) less than 130 mmHg and no use of blood pressure-lowering medication, 2) waist to hip ratio (WHR) less than 0.95 for women and less than 1.03 for men, 3) no prevalent diabetes . Individuals who met all the criteria were categorized as MH, otherwise, were categorized as MUH.
Other definitions used in the study were as follows. Cardiovascular diseases were defined as self-reported coronary heart disease and/or cerebral stroke. Diabetes was defined by self-reported history or FPG ≥ 7.0 mmol/L. WHR was calculated as follows: WHR = waist circumference/hip circumference. BMI was calculated as follows: BMI = Weight (Kg)/Height2 (m2). Smoking was categorized as never, current, and past. Alcohol intake was defined as average intake of alcohol ≥ 50 g/day. Physical activity was defined as exercise one or more times per week, at least 20 min for each time [17, 18, 20].
The primary end point was all-cause mortality from study baseline in 1992 to follow-up in 2007. The occurrence of all-cause mortality and the cause of mortality was confirmed via telephone contact with referring relatives.
For summarizing baseline characteristics of subjects, continuous variables were presented as mean ± standard deviation (SD) and median with interquartile range (IQR) where appropriate, and categorical variables as number (percentage) for each group. Comparisons of baseline characteristics between subjects who finished follow-up and those who lost to follow-up were performed using the analysis of variance or Kruskal–Wallis tests for continuous variables, and the chi-square or Fisher exact tests for categorical variables.
Given the observational nature of the present study, propensity scores (PS) were developed to account for potential confounding by observed baseline characteristics. PS methods replace an entire set of baseline characteristics with a single composite score, and this can be accomplished with a number of potential confounders in excess of what is possible with conventional regression methods [22, 23]. The individual propensities for diagnosis of MH were estimated with the use of a multivariable logistic-regression model that included the following covariates, including age, sex, smoking, drinking, exercise, cardiovascular diseases, diastolic blood pressure (DBP), total cholesterol, HDL-C, LDL-C, triglycerides, and BMI. Then, associations between MUH and all-cause mortality were estimated by Cox regression models with the use of three PS methods, including overlap weighting, propensity-score matching (PSM), and the PS as an additional covariate. Direct acyclic graph was built to select variables for adjustment in multivariable Cox proportional regression models.
Overlap weighting was chosen as the primary method for confounder adjustment in this study, because it could minimize the influence of extreme PS on model output . Overlap weighting could assign weights to each patient that are proportional to the probability of that patient belonging to the opposite exposed group. Specifically, exposed participants are weighted by the unexposed probability (1 – PS), and unexposed participants are weighted by the exposed probability (PS). Overlap weighting assigns greater weight to participants in which treatment cannot be predicted and lesser weight to patients with extreme PS (approaching 0.0 or 1.0) preventing outliers from dominating the analysis and decreasing precision, which is a concern with inverse probability weighting . Furthermore, overlap weighting has the favorable property of resulting in the exact balance (standardized mean differences [SMD] = 0) of all variables included in the multivariable logistic regression model used to derive the PS. PSM was also used to adjust for clinically relevant baseline characteristics that were potentially confounding variables, and participants were matched 1:1 using the nearest neighbor method, with a fixed caliper width of 0.08. After overlap weighting and PSM, SMD were estimated for the baseline covariates before and after the processes to assess pre-match imbalance and post-match balance, and absolute SMD of less than 0.1 for a given covariate indicate a relatively small imbalance . In addition, cumulative hazard plots were also produced to display the cumulative incidence of all-cause mortality in different methods.
To estimate the plausibility of bias from unmeasured and residual confounding, we calculated E-values, which could assess the potential for unmeasured confounding between MUH and all-cause mortality, and it quantifies the required magnitude of an unmeasured confounder that could negate the observed association between MUH and all-cause mortality . In addition, mediation analysis, a single mediator model, was also conducted to assess whether the relationship between MUH and all-cause mortality was mediated or suppressed by other variables. In these analyses, mortality status was used as the outcome variable. MUH was used as the predictor, and other variables were used as mediators, separately.
The statistical analyses were performed with R software, version 4.1.0 (R Project for Statistical Computing) mainly including the “MatchIt” , “survival” , “survey” , “cobalt” , “mediation” , and “Evalue”  packages. For all statistical analyses, a two-sided p value of 0.050 was considered statistically significant.