Introduction
The location of fat accumulation in the body, rather than total fat volume, is increasingly shown to be more important for the risk of type 2 diabetes [
1]. Both visceral adipose tissue (VAT) and truncal fat depot have been associated with type 2 diabetes [
2‐
4] and the metabolic syndrome [
5,
6].
VAT is a hormonally active component of body fat. The risk of developing diabetes has been shown to be higher in individuals with excess visceral adiposity, with [
3] or without [
7] manifestations of obesity. Therefore, VAT plays a key role in the association between adiposity and glucose metabolism [
4,
8‐
10]. However, traditional anthropometric measures such as BMI and waist circumference (WC) are not able to distinguish VAT from subcutaneous adipose tissue [
11]. Furthermore, VAT accounts for an increased cardiometabolic risk regardless of BMI levels [
12]. Truncal fat depot can be partitioned into upper body (android or central) and lower body (gynoid or peripheral) areas. High android to gynoid per cent fat ratio has shown a greater correlation with cardiometabolic dysregulation compared with BMI [
13]. Among the elderly, the android fat depot seems to be more closely associated with the metabolic syndrome compared with abdominal visceral fat [
5].
Computed tomography (CT) [
2,
12] and MRI [
3] are the gold standard measures for quantification of VAT. Dual-energy x-ray absorptiometry (DXA) is a well-validated imaging method for precise measurement of body fat mass in various body compartments (i.e. android and gynoid fat) [
14]. However, these imaging modalities for assessing adipose tissue distribution are inconvenient and expensive. Recently, different metabolic indices combining both anthropometric and lipid measures have been used as estimators of visceral adiposity dysfunction [
15] and lipid overaccumulation [
16,
17]. These novel indices, including visceral adiposity index (VAI), lipid accumulation product (LAP) and the product of triacylglycerol (TG) and glucose (TyG), have been suggested as early markers of insulin resistance, mainly in cross-sectional studies [
18‐
20]. However, the associations of these novel metabolic indices with incident type 2 diabetes remain unclear. We therefore studied the associations of different novel metabolic indices and their formula components with incident type 2 diabetes among women and men from the large prospective population-based Rotterdam Study. We further assessed the associations of truncal fat depot measured by DXA with incident type 2 diabetes.
Methods
Statistical analysis
Considering sex differences in fat distribution and that the formulae of metabolic indices differ by sex, all analyses were performed among women and men separately. Descriptive characteristics were presented as means ± SD for continuous variables, and numbers (percentages) for dichotomous variables. One-way ANOVA for continuous variables and the χ2 test for categorical variables were used to compare general characteristics between women and men as well as between participants with and without DXA measurements. Markers with a right-skewed distribution (insulin, glucose, HDL-cholesterol, TG, VAI, LAP, per cent android fat, per cent gynoid fat, android to gynoid fat ratio and per cent total fat mass) were transformed to the natural logarithmic scale.
Cox proportional hazards models were used to investigate associations of different combined metabolic indices (VAI, LAP, TyG), the anthropometric (BMI, WC) or laboratory components (inverse HDL-cholesterol, TG) included in their formulae, as well as DXA measurements of body fat (android, gynoid, total fat mass, the ratio of per cent android to gynoid fat mass) with incident type 2 diabetes. Inverse HDL-cholesterol was used to facilitate easier comparison between the estimates. The proportional hazards assumption of the Cox model was checked by visual inspection of log minus log plots and by performing a test for heterogeneity of the exposure over time. There was no evidence of violation of the proportionality assumption in any of the models (p value for time-dependent interaction terms >0.05). The first model was adjusted for age and cohort. The second model was additionally adjusted for BMI. The third model was additionally adjusted for systolic BP, hypertension medication, smoking and prevalent CVD. The fourth model was additionally adjusted for HDL-cholesterol, TG and serum lipid-reducing agents. In the fifth model, FPG was added. As glucose measurement is a means to diagnose type 2 diabetes, model 5 should be considered a conservative model. For each novel lipid index, the covariates that were already in the index formula were excluded from the multivariable-adjusted model.
To check whether the association of different markers with incident diabetes differed by obesity status, the analyses were further stratified based on a BMI cut-off of 30 kg/m
2 and performed among non-obese (BMI <30 kg/m
2) and obese (BMI ≥30 kg/m
2) individuals. The
p value is derived from the
z score calculated from the ratio between the difference of the two estimates and the SE of the difference [
27]. The
p value indicates whether the difference between the estimates is significant. To compare the estimates between women and men, an interaction test was applied to model 4 (analyses of the total population).
Multiple imputation procedure was performed (five imputations) to impute missing data for covariates. All analyses were conducted using IBM SPSS software, version 21 (IBM, Armonk, NY, USA). A p value below 0.05 was considered statistically significant.
Discussion
In the large population-based Rotterdam Study, the novel metabolic indices VAI, LAP and TyG were stronger risk markers for incident diabetes compared with traditional anthropometric and lipid measures among women. The magnitude of association of these novel metabolic indices with diabetes was comparable with that of DXA-measured body fat compositions in women. Among men, neither combined metabolic indices nor DXA measures of body fat were superior to traditional anthropometric and lipid measures, in particular BMI, in association with diabetes.
VAT is a hormonally active component of total body fat, which may play a key role in the association between adiposity and glucose metabolism [
4,
8‐
10]. Excess visceral adiposity has been linked to a higher risk of type 2 diabetes, regardless of obesity [
2,
3,
7,
12]. The three combined metabolic indices VAI, LAP and TyG have been introduced as indicators of ‘visceral adipose function’ [
15] and insulin resistance [
18‐
20] and have been linked to cardiometabolic risk [
15], prediabetes [
28] and diabetes [
28] in cross-sectional studies. Our study is the first to simultaneously investigate the longitudinal associations of all these new indices, as well as their components, with incident type 2 diabetes among women and men. The three novel combined metabolic indices were all independently associated with increased risk of diabetes in our study. VAI and LAP combine both anthropometric and metabolic variables to evaluate, respectively, adiposity dysfunction and lipid overaccumulation, whereas TyG includes only metabolic variables. TyG is among the most mentioned insulin resistance indices in the literature [
29‐
36]. TyG has also been suggested as a promising biomarker for glycaemic control in people with type 2 diabetes [
30], even better than HOMA [
29]. In comparison with FPG, TyG improved diabetes risk prediction in individuals with normal FPG [
37]. LAP includes WC and TG, similarly to hypertriglyceridaemic waist [
17], and is an index of excessive lipid accumulation. Since precise measurement of visceral fat content requires the use of expensive imaging techniques such as CT or MRI [
2,
12], simple and economical quantification of these visceral adiposity indices could lead to improvements in identification of individuals at high risk of developing type 2 diabetes.
The counterbalance between insulin secretion and insulin resistance is critical for type 2 diabetes pathogenesis. VAI, LAP and TyG have been introduced as early indicators of insulin resistance [
18‐
20]. In our study, these three indices were all moderately correlated with an index of insulin resistance (HOMA-IR) and showed a smaller correlation with insulin secretion (HOMA-B). As VAI and LAP combine both lipid variables and adiposity status, they could serve as better surrogates for insulin resistance compared with either lipid or adiposity measures alone. The largest correlation of TyG with insulin resistance in our study is in line with other study findings, supporting a central role of both lipotoxicity and glucotoxicity in modulating insulin resistance [
38]. Since obesity has a strong impact on dyslipidaemia, insulin resistance and the development of type 2 diabetes, we further stratified the analyses based on obesity status. Correlation of different combined adiposity indices with HOMA measures did not materially differ between non-obese and obese individuals. The overall tendency towards stronger associations of these metabolic indices with incident diabetes among non-obese individuals might be due to their lower discriminatory power among higher risk obese individuals.
While the exact mechanisms responsible for the relationship between excess abdominal/visceral fat and cardiometabolic risk are still unclear, several hypotheses have been proposed [
39‐
41]. Subcutaneous fat faces obesogenic stress with a limited capacity for regional adipocyte hypertrophy or hyperplasia. Once this capacity is surpassed, adipose tissue storage is forced into other regions, such as organs or compartments of the body, termed ectopic. Visceral fat is considered the classic ectopic fat depot and is associated with dysfunctional adiposity or adiposopathy [
42,
43].
In our study, WC, TG, VAI, LAP and TyG showed a stronger association with incident type 2 diabetes among women compared with men. Similarly, the correlations between VAI, LAP and TyG with HOMA-IR in our study were overall stronger among women. The greater association of VAT with diabetes and adverse cardiovascular risk profiles among women has been suggested in several studies [
44,
45]. Sex differences in adverse metabolic outcomes associated with visceral fat have been related to a significantly lower visceral fat area in non-diabetic women compared with non-diabetic men, and a similar visceral fat area for both diabetic women and men [
44]. Among individuals with more visceral fat, a greater portion of hepatic NEFA delivery originates from VAT lipolysis [
46]. Contribution of visceral lipolysis to hepatic NEFA delivery in relation to visceral fat has been found to be greater in women than in men [
46]. Moreover, correlation between VAT area and serum TG has been found to be stronger in women than in men [
47].
No previous study has investigated the associations of DXA measures of body fat with incident type 2 diabetes. Our study suggests that per cent gynoid fat and per cent android to gynoid fat ratio among women and total fat mass among men are independent risk markers for diabetes. Previous studies have shown important relations between android to gynoid fat ratio and metabolic risk in healthy adults. Android or truncal obesity has been associated with the risk of metabolic disorders and CVD [
48], yet there is evidence that gynoid fat distribution may be protective [
49]. Android fat depot is the adipose tissue mainly around the trunk including, but not exclusively, visceral fat. Compared with abdominal visceral fat, android fat depot has shown a larger association with the metabolic syndrome in elderly people [
5]. In line with our findings, high per cent android to gynoid fat ratio has shown a larger correlation with cardiometabolic dysregulation compared with per cent android fat, per cent gynoid fat or BMI [
13]. Compared with women with a predominantly gynoid fat distribution, android obesity in women has been correlated with a higher incidence of glucose intolerance [
50]. Excess android fat mass has recently been associated with high TG and low HDL-cholesterol levels in men and high LDL- and low HDL-cholesterol levels in women. Excess gynoid fat mass has been positively correlated with total cholesterol in men and has shown a favourable association with TG and HDL-cholesterol in women [
51]. Increased gynoid fat mass has also been shown to be protective against the progression of non-alcoholic fatty liver disease in Japanese women with type 2 diabetes [
52]. It therefore seems that regional fat distribution in the android and gynoid regions have varying effects on lipid profiles among women and men. In line with this, we found an inverse association in women between gynoid fat and android to gynoid fat ratio and type 2 diabetes and a positive association in men between total fat mass and type 2 diabetes.
In our study, the magnitude of association between DXA measures of body fat and diabetes was comparable with that of combined metabolic indices and traditional anthropometric and lipid measures. Considering the costs and radiation exposure associated with DXA measurement, its use in the general population as a screening tool for diabetes may therefore not be justified, and using well-established and simple anthropometric variables such as BMI might suffice.
To our knowledge, this is the first prospective population-based cohort study to simultaneously investigate the associations between novel metabolic indices as well as DXA measures with incident diabetes among women and men over a long follow-up period. We used data from a well-characterised prospective cohort study, which allowed for direct comparison of several metabolic indices as well as correction for a wide range of covariates.
The limitations of our study also warrant attention. Our population comprised individuals aged 45 years and older of European ancestry. One might speculate that the impact of VAT on diabetes incidence would have been even stronger in a younger population. Thus, generalisation of our results to younger age groups and other ethnicities should be made with caution. Moreover, as with other cohort studies, the possibility of selection bias could not be entirely ruled out. Due to the unavailability of CT or MRI in our population, visceral adiposity was not directly measured but estimated. Also, we did not have DXA measures specifically for visceral fat in the Rotterdam Study. Instead, android fat measured by DXA was used as a proxy for visceral fat. Thus, comparison of our results against the gold standard measures for visceral fat is not possible. We did not include variables such as socioeconomic status, family history of diabetes, dietary intake and physical activity in our multivariable models, as they were not available.
In conclusion, the novel combined metabolic indices VAI, LAP and TyG were stronger risk markers for incident type 2 diabetes compared with traditional anthropometric and lipid measures among women. The predictive value of these novel metabolic indices for type 2 diabetes was also comparable with that of DXA-measured body fat compositions in women. Neither combined metabolic indices nor DXA measures of body fat were superior to traditional anthropometric and lipid measures in association with type 2 diabetes among men. In particular, BMI remained the best marker for type 2 diabetes risk in men and among the best markers in women. BMI could therefore be used as a simple and useful tool for diabetes risk screening in the general population.
Publisher’s note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.