Assessment of adiposity distribution and its association with diabetes and insulin resistance: a population-based study

We performed a cross-sectional study in a community in Guangzhou, China from June to November 2011. The study population was from the REACTION study and details of this study have been published previously [22‐24]. At first, 10,104 subjects aged 40 years or older were invited to participate by examination notices or home visits during the recruiting phase. After this, 9916 subjects signed the consent form and agreed to participate in the survey, including those with or without diabetes. Subjects who failed to provide information [WC: n = 186; hip circumference (HC): n = 42; height: n = 29; weight: n = 23; triglycerides (TG): n = 23; high-density lipoprotein cholesterol (HDL-C): n = 2; fasting plasma glucose (FPG): n = 15; oral glucose tolerance test (OGTT) 2 h glucose: n = 60; HbA1c: n = 40] were excluded from the analyses. Accordingly, a total of 9496 eligible individuals were included in the final data analyses. The study protocol was approved by the Institutional Review Board of the Sun Yat-sen Memorial Hospital affiliated to Sun Yat-sen University and was in accordance with the principle of the Helsinki Declaration II. Written informed consent was obtained from each participant before data collection.

Information on lifestyle factors, sociodemographic characteristics and family history were collected by using a standard questionnaire. Smoking or drinking habit was classified as ‘never’, ‘current’ (smoking or drinking regularly in the past 6 months) or ‘ever’ (cessation of smoking or drinking more than 6 months) [25]. A short form of the International Physical Activity Questionnaire (IPAQ) was used to estimate physical activity at leisure time by adding questions on frequency and duration of moderate or vigorous activities and walking [26]. Separate metabolic equivalent hours per week (MET-h/week) were calculated for evaluation of total physical activity.

Venous blood samples were collected for laboratory tests after an overnight fasting of at least 10 h. Measurements of FPG, fasting serum insulin, TG, total cholesterol (TC), HDL-C, low-density lipoprotein cholesterol (LDL-C), creatinine and γ-glutamyltransferase (γ-GGT) were done using an autoanalyser (Beckman CX-7 Biochemical Autoanalyser, Brea, CA, USA). HbA1c was assessed by high-performance liquid chromatography (BioRad, Hercules, CA). The abbreviated Modification of Diet in Renal Disease (MDRD) formula recalibrated for Chinese population was used to calculate estimated glomerular filtration rate (eGFR) expressed in mL/min per 1.73 m² using a formula of eGFR = 175 × [serum creatinine × 0.011]^−1.234 × [age]^−0.179 × [0.79 if female], where serum creatinine was expressed as μmol/L [27].

All participants completed anthropometrical measurements with the assistance of trained staff by using standard protocols. Body height and body weight were recorded to the nearest 0.1 cm and 0.1 kg while participants were wearing light indoor clothing without shoes. BMI was calculated as weight in kilograms divided by height in meters squared (kg/m²). WC was measured at the umbilical level with participant in standing position, at the end of gentle expiration. HC was measured the same way as WC, but the measuring tape was placed around the widest part of the buttocks. Waist-to-hip ratio (WHR) was calculated as WC divided by HC while waist-to-height ratio (WHtR) was calculated as WC divided by height in cm. VAI was determined by gender-specifc equations and calculated using the following formulas. Men: [WC/(39.68 + (1.88 × BMI))] × (TG/1.03) × (1.31/HDL-C); Women: [WC/(36.58 + (1.89 × BMI))] × (TG/0.81) × (1.52/HDL-C) [28]. WC was calculated in cm while HDL-C and TG in mmol/L. BAI was calculated using the equation as [(HC/(height^1.5) − 18] [10]. HC and height were calculated in cm. LAP was calculated as (WC − 65) × TG in men and (WC − 58) × TG in women. WC was calculated in cm and TG in mmol/L [29, 30]. Three times consecutively blood pressure measurements were obtained by an automated electronic device in a 5-min interval by the same observer (OMRON, Omron Company, China). The average of three measurements of blood pressure was used for analysis.

The insulin resistance index (homeostasis model assessment of insulin resistance, HOMA-IR) was calculated as fasting insulin (μIU/mL) × FPG (mmol/L)/22.5 [31]. Insulin resistance was defined by a HOMA-IR index within the top quartile (greater than 2.54) in the present study [32]. Diabetes was diagnosed according to the American Diabetes Association 2010 criteria, which included (1) FPG level of 7.0 mmol/L or greater, or (2) OGTT 2 h glucose level of 11.1 mmol/L or greater, or (3) HbA1c level of 6.5% or greater [33].

Statistical analysis was performed using SAS version 9.2 (SAS Institute Inc, Cary, NC, USA). Continuous variables were presented as mean ± standard deviation (SD) except for skewed variables, which were presented as medians (interquartile ranges). Categorical variables were expressed as numbers (proportions). Differences between groups were performed by one-way ANOVAs. Comparisons between categorical variables were performed by the χ² test. Physical activity levels, TG, FPG, OGTT 2 h glucose, HbA1c, fasting insulin, HOMA-IR, γ-GGT, WHR, WHtR, VAI, BAI and LAP were logarithmically transformed before analysis due to a non-normal distribution.

We primarily analyzed the associations of body fat content estimates (BMI, WC, WHR, WHtR, VAI, BAI and LAP) with FPG, OGTT 2 h glucose, HbA1c, fasting insulin and prevalence of insulin resistance and diabetes. Linear regression analyses were used to test for trends across groups. Pearson’s correlation and multiple regression analysis adjusted for age and sex were performed to test the correlations of body adiposity parameters with FPG, OGTT 2 h glucose, HbA1c and fasting insulin. The unadjusted and multivariate-adjusted logistic regression analysis was used to assess the risk of increased prevalent insulin resistance and diabetes in relation to 1-quartile increase in body adiposity measurements. Model 1 is unadjusted. Model 2 is adjusted for age. Model 3 is adjusted for age, sex, current smoking and drinking status, physical activity level, systolic blood pressure (SBP), LDL-C, γ-GGT, eGFR and antidiabetic treatment. By considering the variation in each quartile group, multivariate-adjusted logistic regression analysis was used to assess the prevalent diabetes and insulin resistance in relation to each quartile increase in body adiposity measurements. Odds ratios (OR) and the corresponding 95% confidence intervals (CI) were calculated. The discriminative ability of adiposity measurements on prevalent insulin resistance and diabetes was examined through area under the receiver operating characteristic curve (AUC) with 95% CI. Differences between the AUC of anthropometric measures were performed with a nonparametric approach [34]. To obtain a better assessment of the estimation power of the adipose distribution parameters, we used the optimal operating point with setting the minimum sensitivity of 70% for which we do not want sensitivity to fall below [35].

All statistical tests were two-sided, and a P value < 0.05 was considered statistically significant.

The mean age was 55.9 ± 8.1 years among the 9496 enrolled individuals. Totally, 2054 subjects diagnosed with diabetes and the prevalence rate was 21.6% in this population. Accordingly, 606 (6.4%) subjects have been diagnosed with diabetes before the survey and among them 479 (5.0%) subjects were receiving antidiabetic treatment. Moreover, there were 1448 (16.3%) newly diagnosed diabetic patients based on the survey. The demographic and biochemical characteristics of the study population were shown in Table 1. Subjects in diabetes group differed from those in non-diabetes group in multiple clinical variables. All of the adiposity measurements in the study were significantly elevated in diabetes group (all P < 0.0001). Moreover, LAP was dramatically elevated in participants with diabetes in comparison with those without the disease [24.4 (14.4–41.2) vs 38.9 (22.5–61.9), P < 0.0001].

Pearson’s correlation analysis revealed that body adiposity measurements including BMI, WC, WHR, WHtR, VAI, BAI and LAP were all significantly associated with increased FPG, OGTT 2 h glucose, HbA1c and fasting insulin (Table 2, all P < 0.0001). The associations were still persisted in multivariate linear regression analysis with age and sex adjusted (all P < 0.0001). Compared with other adiposity measurements, LAP reached the highest correlation coefficient with FPG (r = 0.22, P < 0.0001), OGTT 2 h glucose (r = 0.27, P < 0.0001), HbA1c (r = 0.24, P < 0.0001) and fasting insulin (r = 0.52, P < 0.0001). In multivariate linear regression analysis adjusted for age and sex, LAP was considerably most strongly correlated with HbA1c (β = 0.22, P < 0.0001) and fasting insulin (β = 0.52, P < 0.0001) while VAI was most strongly correlated with FPG (β = 0.21, P < 0.0001) and OGTT 2 h glucose (β = 0.26, P < 0.0001). However, as an independent determinant of body fat percentage, BAI was most weakly associated with FPG, OGTT 2 h glucose and HbA1c than other body adiposity measurements in both Pearson’s correlation and multivariate linear regression analysis. As shown in Fig. 1, with the elevating quartiles of adiposity phenotypes in the study, levels of FPG, OGTT 2 h glucose and HbA1c were all significantly increased (all P for trend < 0.0001).

After this, we further studied the prevalence of diabetes and insulin resistance in different quartiles of body adiposity measurements. As showed in Fig. 2, the change in proportion of prevalent diabetes and insulin resistance were increased with the elevated quartiles of all tested adiposity parameters in the study (all P for trend < 0.001). The positive associations of BMI, WC, WHR, WHtR, VAI, BAI and LAP with diabetes and insulin resistance were consistently detected in both univariable and multivariable logistic regression analysis (Tables 3, 4). As shown in Table 4, compared with participants in quartile 1 of body adiposity measurements, multivariate-adjusted logistic regression analysis showed that participants in quartile 2, quartile 3 and quartile 4, respectively, have a significant correlation with increased odds of diabetes and insulin resistance. Compared with other adiposity phenotypes in the study, LAP have shown the relatively strongest while BAI have shown the relatively weakest associations with increased odds of both diabetes and insulin resistance across all logistic regression models. Further receiver operating characteristics curve represented consistent results and the LAP still performed the best discrimination accuracy for diabetes (AUC: 0.658 95% CI 0.645–0.671, all P < 0.05 except WHR) and insulin resistance (AUC: 0.781 95% CI 0.771–0.792, all P < 0.05) when compared with other adiposity phenotypes. We analysis the optimal operating point for the screening diabetes of the LAP. With a pre-assigned threshold of sensitivity for detecting a true-responder, greater than 70% in the present study, 25.5 of the LAP was thought to have a more appropriate screening effect for prevalent diabetes, of which the sensitivity and specificity were 70.0% and 52.0%, respectively. Accordingly, the sensitivity and specificity were 81.0% and 59.0% by using threshold 25.5 of the LAP for screening prevalent insulin resistance.