The China Health and Retirement Longitudinal Study (CHARLS) is an ongoing cohort study that enrolled a nationally representative sample of community-dwellers aged ≥ 45 years in China [13,14]. The baseline survey was conducted in 2011 to 2012 and included 17,708 participants. The study design of CHARLS was approved by the ethical review committee at Peking University (approval No. IRB 00001052-11014), and informed consent was obtained from each participant. The details about the sampling and design of CHARLS have been provided elsewhere [13].

In this study participants who were aged ≥ 60 years at baseline and followed for 4 years were eligible for inclusion. Of them, 2,909 had blood samples collected at baseline and in 2015. After excluding participants with incomplete information on body weight, WC, blood pressure, blood-based biomarkers such as plasma glucose, hemoglobin A1c (HbA1c), total cholesterol (TC), or C-reactive protein (CRP) at baseline or in 2015 (n = 338), being diagnosed with diabetes before the survey in 2015 (n = 513), having data considered abnormal or as outliers on body weight, height, BMI, WC, blood pressure, or their changes (n = 243), 1,815 participants were finally included (Supplementary Fig. 1).

Demographic data including age, sex, education, smoking, drinking, and disease history (e.g., hypertension, cardiovascular disease, kidney disease) were collected using a pre-designed questionnaire, and anthropometric variables, such as body weight, height, and WC, were measured using standard protocols described in the CHARLS handbook [13]. Blood pressure was measured three times, at a 45-second interval, using Omron^TM HEM-7200 Monitor (Omron Co. LTD., Dalian, China), and its average value was used in the present study.

BMI, WHtR, and BRI were calculated using the following equations [6,9].

Blood samples were collected in the morning, with their plasma being centrifuged for measurements of HbA1c, glucose, TC, triglycerides (TG), high-density lipoprotein cholesterol (HDL-C), low-density lipoprotein cholesterol (LDL-C), uric acid (UA), as well as CRP [14]. Blood count analysis for white blood cells (WBCs) was conducted at the local county health centers. The triglycerides-and-glucose (TyG) index, which is recognized as a marker for insulin resistance [15], was calculated as:

Participants in this study were guided to walk twice on a straight 2.5-m flat course, with walking time being recorded by a stopwatch to calculate walking speed [13]. This was used to reflect the physical wellbeing/performance and the level of physical activity of an individual [16,17].

In this study incident diabetes is defined as: fasting plasma glucose (FPG) ≥ 7.0 mmol/L (126 mg/dL), random plasma glucose ≥ 11.1 mmol/L (200 mg/dL), HbA1c ≥ 6.5% (48 mmol/mol), medical history, and/or the use of anti-diabetic drugs [18]. BMI ≥ 24 kg/m² is classified as overweight/obesity, and < 24 kg/m² as normal weight. Hypertension is ascertained by: systolic blood pressure (SBP) ≥ 140 mmHg, diastolic blood pressure ≥ 90 mmHg, medial history, and/or the use of antihypertensive drugs. Dyslipidemia is defined as: TC ≥ 240 mg/dL, LDL-C ≥ 160 mg/dL, HDL-C < 40 mg/dL, TG ≥ 200 mg/dL, medial history, and/or the use of lipid-lowing agents [19].

Changes in central fatness were defined as the change scores in WC, WHtR, or BRI from baseline. Similarly, changes in glycemic control, lipid profiles and inflammatory markers were calculated using the same methods.

Data were expressed as means ± standard deviations (SDs) for continuous variables and numbers (%) for categorical variables, unless otherwise stated. Unpaired t test and chi-square test were used to compare the difference between groups. Changes in central fatness were evaluated in the following two ways: (1) as continuous variables expressed as per 1 SD change; and (2) as categorical variables, which were divided into tertiles: loss (the lowest tertile of the change), stable (the middle tertile), and gain (the highest tertile). Logistic regression analyses were conducted to obtain the odds ratios (ORs) and 95% confidence intervals (CIs) for risk of diabetes in relation to changes in central fatness using an unadjusted model, and then the multivariable models that include age, sex, and BMI (Model 1), as well as history of smoking and drinking, disease status (hypertension, dyslipidemia, and cardiovascular disease), SBP, HbA1c, TC/HDL-C, UA, CRP, and walking speed at baseline (Model 2). To further assess the relationship of changes in central fatness with risk of diabetes, dose-response analyses, which were modeled with the restricted cubic splines, were introduced [20]. Moreover, subgroup analyses on the basis of sex (male vs. female), age category (60–70 years vs. ≥ 70 years), overweight/obesity (yes vs. no), history of smoking and drinking (yes vs. no), walking speed (fast vs. slow) were also performed. The predictive abilities for diabetes with regard to changes in central fatness were assessed by the area under the receiver operating characteristic curves (AUROC), and compared according to the method described previously [21,22]. Multivariable linear regression analysis adjusting for age, sex, and BMI was employed to investigate the relationship of changes in central fatness with changes in several cardiometabolic markers including blood pressure, glycemic control, lipid profiles, and inflammatory markers. We also utilized mediation analysis to explore the mediating role of the changes in these cardiometabolic markers for risk of diabetes. All analyses were conducted using Stata software version 14.0 (StataCorp., College Station, TX, USA), with p < 0.05 considered statistically significant.

During the 4-year follow-up, 177 of the 1,815 participants developed diabetes, and their baseline characteristics are shown in Table 1 and Supplementary Table 1. Compared with participants who did not develop diabetes, those who did were more likely to present with hypertension, had larger BMI, WC, WHtR, and BRI, and bigger values of FPG, HbA1c, TyG index, TC, TG, and UA, and exhibited a higher degree of inflammation assessed by WBC and CRP. Yet the percentages of smokers and drinkers as well as the walking speed did not differ between groups (all p ≥ 0.63).

Of the enrolled participants, their body weight was minorly changed in the follow-up years from baseline (p = 0.20), but their BMI, WC, WHtR, and BRI were increased on average (all p < 0.01). For participants developed diabetes, they had larger increases in WC, WHtR, and BRI than those remained free of diabetes (all p < 0.01). Further analysis showed that the risk of diabetes was increased by 44%, 40%, and 37% per 1 SD higher of WC, WHtR, and BRI at baseline, respectively, in multivariable-adjusted model (all p < 0.05) (Supplementary Table 2); and their relationship was found to be all linearly-shaped (all p_nonlinearity ≥ 0.38) (Supplementary Fig. 2).

The ORs for diabetes associated with changes in central fatness are presented in Table 2, and the baseline characteristics of the participants among the tertiles of these changes are summarized in Supplementary Table 3. In unadjusted model, the ORs for diabetes were 1.31 (95% CI, 1.12 to 1.54), 1.28 (95% CI, 1.10 to 1.50), and 1.29 (95% CI, 1.11 to 1.51) per 1 SD increases in WC, WHtR, and BRI, respectively. The magnitudes of these associations were slightly enlarged after controlling for multiple variables (Model 2). To account for the baseline difference, relative changes in WC, WHtR, and BRI from baseline were also calculated (Supplementary Table 4), which showed similar results as above. Sensitivity analyses upon the exclusion of participants without fasting samples or using drugs for hypertension or dyslipidemia did not significantly affect the primarily outcomes (Supplementary Table 5).

Subgroup analyses showed that the association between changes in central fatness with risk of diabetes was not significantly affected by sex, age category, body weight status, history of smoking or drinking, or walking speed (all p_interaction ≥ 0.11). However, the magnitude was larger in participants with overweight/obesity or in participants with slow walking speed compared with their counterparts (Supplementary Table 6).

Dose-response analyses noted that no evidence of a nonlinear relationship between changes in central fatness and risk of diabetes was observed in the whole population (all p_nonlinearity ≥ 0.11) (Fig. 1), or specifically, in sex-stratified (all p_nonlinearity ≥ 0.06) or overweight/obesity-based subgroups (all p_nonlinearity ≥ 0.24).

Changes in central fatness had moderate predictive abilities for diabetes, with their AUROC ranging from 0.578 to 0.582 (Supplementary Fig. 3). There was no significant difference between the AUROC for changes in BMI and that for changes in WC, WHtR, or BRI (all p ≥ 0.64 for comparisons). Moreover, in the group of participants with stable BMI (defined as the middle tertile of the changes in BMI), similar results were observed but all the AUROC for changes in WC, WHtR, and BRI became non-significant from 0.50 (all p ≥ 0.08).

Linear regression analyses adjusting for age, sex, and BMI showed that changes in WC, WHtR, or BRI were consistently and positively correlated with changes in HbA1c, TyG index, TG, WBC, and CRP in older adults (all p ≤ 0.04) (Table 3).

Mediation analysis suggested that among the candidates listed in Fig. 2, changes in HbA1c or TyG index met the mediation criteria for all the indicators of central fatness associated with risk of diabetes. Results showed further that the mediating effect was approximately 35% for changes in HbA1c and 21% for changes in TyG index for all the indicators of central fatness (Fig. 2).

This prospective cohort study showed several findings in older adults aged ≥ 60 years: (1) increases in central fatness (assessed by WC, WHtR, and BRI) were associated with elevated risk of diabetes in a dose-dependent manner; (2) changes in central fatness did not show superior predictive abilities for diabetes to changes in BMI—the indicator of general fatness; (3) increases in central fatness correlated with increases in HbA1c, TyG index, TG, WBC, and CRP, suggesting that gains in central fatness lead to deteriorated metabolic control; and (4) changes in HbA1c as well as in TyG index partly explained the association of risk of diabetes in relation to changes in central fatness.

A plenty of evidence has suggested that indicators of central fatness including WC and WHtR were positively associated with risk of diabetes [6,9]. However, no studies have evaluated the risk of diabetes in relation to changes in central fatness assessed by WC, WHtR, and BRI, in particular in older adults aged ≥ 60 years. As an attempt to address this issue, our study showed that increases in central fatness were related to elevated risk of diabetes, and that the association was dose-dependently and linearly shaped (as evidenced by our dose-response analysis). Moreover, our subgroup analysis found that in participants with fast walking speed at baseline, increases in central fatness did not result in significantly increased risk of diabetes, with its magnitude being much smaller than that in participants with slow walking speed. This may probably suggest that good physical performance could help to minimize the adverse effect from gains in central fatness, at least partly.

In recent years there has been an ongoing interest in exploring which or the change of which anthropometric index related to obesity shows the largest predictive ability for diabetes [6,9,10,23–25]. Despite no consensus has been reached to date, accumulating evidence has shown that the indicators of central fatness (such as WC) or their changes outperformed the indicator of general fatness (that is, BMI) in predicting diabetes [6,10]. However, in our study we observed similar AUROC for these indicators regarding their changes, even among participants with stable BMI, suggesting that there might be no significant difference on their predictive abilities. It is speculated that this discrepancy between previous studies and ours might be due to the differences in target populations (older versus middle-aged adults).

Notably, our present study showed for the first time that increases in central fatness were related to deteriorated glycemic control (represented by HbA1c), worsened lipid profile (assessed by TG), and greater degree of inflammation (reflected by WBC and CRP). These results may provide some evidence in support of the findings that weight reductions promote cardiometabolic health, such as improving glycemic control as well as lowering inflammation [11]; while importantly, they shed some insights into the mechanism(s) underlined for diabetes in relation to changes in central fatness. Yet we found that only changes in HbA1c and TyG index were identified as potential mediators. Since HbA1c is a biomarker linking with chronic hyperglycemia and TyG index is a reliable indicator for insulin resistance [15], it is likely that deteriorations in glycemic control and insulin resistance may mediate the impact of increases in central fatness on risk of diabetes. However, it should be mentioned that their total mediating effects account for only about 56%. In this regard, future studies that explore the mediation effect from other possible contributors such as cell adhesion molecules [26], adipokines [27], and myokines [28], are worth being conducted.

The strengths of our study include a prospective cohort design, a lab-based clarification of diabetes, the direct measurement of anthropometric parameters that avoid recall bias, and the use of multiple indicators of central fatness. However, this study has several limitations. First, the relatively short follow-up period could not reflect the influences of long-term changes in central fatness on risk of diabetes in older adults. Second, there still exists the possibility of residual confounding from the unmeasured factors such as cytokine changes, although we controlled for multiple variables in our analyses. Third, there is evidence that the use of statin or antihypertensive medication (e.g., diuretics or beta-blockers) might be associated with increased risk of diabetes [29,30]. Yet we were not able to assess the influence of these medications on the association between changes in central fatness and risk of diabetes due to the issue that such information was not collected in CHARLS. Fourth, we could not determine whether the changes in central fatness were gradual or uneven, while this may affect their relationship with risk of diabetes. Finally, our protocol for measuring walking speed, although has been also adopted in the Health and Retirement Study [31], is different from the recommended one—6-m walking test by the Asian Working Group for Sarcopenia [32]. This may potentially affect the assessment of one’s physical performance.

In conclusion, increases in central fatness were related to elevated risk of diabetes in a dose-dependent manner, but this association was not persisted in participants with good physical performance. The association of increases in central fatness with elevated risk of diabetes was underlined by deteriorated glycemic control and insulin resistance, at least partly.