Based on the China Multi-Ethnicity Cohort (CMEC) study, a multistage, stratified cluster sampling method was adopted in Guizhou Province as the study site to conduct a baseline survey among permanent residents aged 30–79 years from July 2018 to August 2019. Initially, a total of 18,790 people from the Miao, Dong, and Bouyei ethnicity groups were selected from the database. Furthermore, 9,183 non-hypertensive participants, 28 participants without high-density lipoprotein or low-density lipoprotein, 29 participants with incomplete diabetes mellitus data, and 26 participants with body mass index or 78 other incomplete covariate variables data were excluded. Finally, 9,446 participants were included in the analyses. The final study population consisted of 9,446 individuals ( Figure 1 ). All participants signed informed consent forms before the investigation. This study was approved by the Sichuan University Medical Ethical Review Board (K2016038) and the Medical Ethics Committee of the Affiliated Hospital of Guizhou Medical University [2018(094)].

A questionnaire covering sociodemographic characteristics, medical history, family history, and lifestyle factors was used during an interview by the same group of trained experienced personnel. The CMEC collected participants’ blood samples after at least 8 h of fasting and measured levels of cholesterol (CHOL), triglyceride (TG), low-density lipoprotein (LDL-C), and high-density lipoprotein (HDL-C) by an AU5800 Automated Chemistry Analyzer (Beckman Coulter Commercial Enterprise, Shanghai, China).

In order to include prehypertensive patients in this study, variables were defined as follows: hypertension was defined as systolic blood pressure ≥130 mmHg, diastolic blood pressure ≥80 mmHg (17, 18), or a self-reported previous diagnosis of hypertension. DM was defined using a previous diagnosis or glycosylated hemoglobin (HBA1C1) ≥6.5% or FPG ≥7.0 mmol/L. Hyperlipidemia was defined as a self-reported previous diagnosis of hyperlipidemia by a physician. The cardiovascular disease (CVD) outcome was defined as a previous diagnosis of coronary heart disease, stroke, or peripheral arterial disease and was recorded in the registration platform.

Anthropometric measurements were conducted by well-trained nurses and physicians. Weight was assessed using an electronic scale with the patients in light clothes and without shoes. Participants were asked to stand with arms hanging freely and height was acquired by a stadiometer with the patient barefoot. BMI was calculated as weight (kg) divided by height squared (m²). Waist circumference (WC) was measured using an inelastic tape measure. Participants were asked to wear light clothes, and WC was assessed midway between the last rib and iliac crest with the participants breathing out gently. Duplicate measures were taken for all measurements with a tolerance error of 1 cm for height and circumference measurements and 1 kg for the weight measurement. A third measurement was needed if the difference of the first two measures was greater than the tolerance limit. The average of the two closest measurements was recorded at last. BAI was calculated as proposed by Bergman et al. (19): hip circumference (cm) divided by [height (m)]^1.5 minus 18. WHR was calculated according to the following formula: WC (cm)/hip circumference (cm). WHtR was calculated according to the following formula: WC (cm)/height (cm). The AIP was calculated as the logarithm of the ratio of TG/HDL-C.

In this study, the covariates included demographic characteristics (age, sex, and ethnicity), socioeconomic indicators (education level, household income, residence, and job type), physical activity, alcohol consumption status, smoking status, habit of drinking tea, systolic blood pressure, diastolic blood pressure, high-density lipoprotein, low-density lipoprotein, triglyceride, cholesterol, history of CVD, and hyperlipemia. Ethnicity was classified as Dong, Miao, and Bouyei. The participants’ education level was divided into three levels: lower than high school, high school or equivalent, and above high school. The job type was grouped into four categories: primary industry practitioner, secondary industry practitioner, tertiary industry practitioner, and unemployed or other. Household income per year was divided into three levels: <100,000 CNY, 100,000–199,999 CNY, and ≥200,000 CNY. Smoking condition was categorized into three groups: never, current, and ever. Alcohol consumption condition was categorized into three groups: never, occasionally, and frequently. Habit of drinking tea was categorized into two groups: yes and no. Physical activity was categorized into three groups: low, moderate, and high.

Continuous variables collected from the participants were tested for normality according to the characteristics of the data, and data obeying a normal distribution were expressed as mean ± standard deviation (SD), while data not obeying a normal distribution were expressed as median and interquartile range (IQR). Qualitative variables were expressed as relative numbers or percentages. Differences in characteristics between the DM and non-DM groups were evaluated using the Student’s t-test (for continuous variables) or chi-square test (for categorical variables). Differences between subjects grouped by quartiles of obesity index were compared in multivariable logistic regression, using quartile 1 (Q1) as the reference group, and the values of odds ratios (ORs) and 95% confidence intervals (CIs) were calculated. Three logistic models were developed for the analyses: model 1 was not adjusted for any confounding variables and was a univariate analysis; model 2 was adjusted for the main demographic variables (sex, age, and ethnicity); and model 3 added place of household registration, education level, job type, family income per year, smoking status, alcohol intake status, habit of drinking tea, physical activity, systolic blood pressure (SBP), diastolic blood pressure (DBP), HDL-CH, LDL-CH, TG, CHOL, CVD, and hyperlipemia to model 2. Linear regression was used for continuous variables. To visualize more closely the linear or non-linear correlation between obesity indices and DM, curves were fitted using restricted cubic spline (RCS). Finally, mediation analyses were performed using the mediation package, and confidence intervals for the mediating effect were assessed using the bootstrap method to determine the proportion of the mediating effect accounted for by the AIP. Using these statistical methods, the possible causal relationship between four obesity indices and DM can be examined more broadly. Statistical analyses were performed using the R software (version 4.2.0). Statistical significance was defined as P <0.05.

Table 1 shows the baseline characteristics of the study participants. In the whole group of 9,446 hypertensive patients, there were 1,111 patients with diabetes, and the prevalence of diabetes was 11.76%. The medium age of the study population was 53.7 (46.8, 62.5) years for non-DM people and 56.4 (51.3, 65.3) years for DM people. In addition, the prevalence of DM was significantly different stratified by age, sex, ethnicity, place of household registration, education level, job type, family income per year, smoking status, alcohol intake status, habit of drinking tea, physical activity, SBP, DBP, HDL-CH, LDL-CH, TG, CHOL, CVD, and hyperlipemia. Furthermore, there were significant differences in ethnicity between the two groups ( Table 1 ).

Regarding subjects in the lowest quartile of each indicator, ORs (95% CIs) for those in the highest quartile were 4.961 (3.351, 7.345), 5.965 (3.825, 9.304), 5.903 (3.821, 9.121), 3.932 (2.725, 5.672), and 3.280 (2.331, 4.615) for BAI, WHR, WHtR, BMI, and AIP, respectively, after adjusting for all covariates ( Table 2 ). The exposure–response trend of DM with five indicators was observed (all p-trend < 0.05). Every 0.1-unit increment of BAI, WHR, WHtR, and AIP was associated with a 30.5%, 95.8%, 118%, and 21.2% increased risk of DM, respectively; every 1-unit increment of BMI was associated with a 14.4% increased risk of DM.

Figure 2 depicts the relationship between four obesity indicators and diabetes mellitus using the RCS curve, with A–D depicting the dose–response relationship between BAI, WHR, WHtR, BMI, and diabetes mellitus after adjusting for all covariates, respectively. As the index increased, the incidence of diabetes mellitus gradually increased. Overall, there was a linear negative correlation between BAI, WHtR, BMI, and diabetes mellitus, which was particularly evident in BMI (non-linear p = 0.672). Furthermore, we could find a non-linear relationship between WHR and the prevalence of diabetes mellitus (non-linear p ≤ 0.001).

Figure 3 presents the direct and indirect effects of BAI, WHR, WHtR, and BMI on diabetes mellitus with AIP as mediators after adjusting all covariates. Overall, the AIP mediated the relationship between all four obesity indicators and diabetes mellitus. The findings suggest that the association of each obesity indicator with diabetes mellitus was mediated by the AIP, and the proportion of the total effect of obesity indicator mediated by the AIP was 19.6%, 15.3%, 15.8%, and 17.2% for BAI, WHR, WHtR, and BMI, respectively.