Obesity is recognised as a major cause of type 2 diabetes mellitus (T2DM) [1–4]. Changing lifestyles over the last decades have given rise to a global epidemic of overweight and obesity which has spread from developed to developing countries and from adults to children and adolescents [5–7]. In 2000, 15% of the Spain population were obese [8], and 9% suffered from T2DM [9]. If trends remain unchanged, conservative estimates are that 12% of the country population will be diabetic by 2030 [9]. The associated future burden of chronic diseases and health costs of the obesity epidemic are of serious concern.

Obesity is defined by the accumulation of excess body fat with potential harmful health effects [10]. A strong link between excess body weight and T2DM risk has long been established in the epidemiological literature. Further evidence showed that ectopic visceral fat accumulation, but not subcutaneous adipose tissue [11], largely accounted for the metabolic complications of obesity [12], such as abnormalities in glucose and lipid metabolism [13] and hepatic insulin resistance [13, 14], preceding the development of T2DM.

Anthropometry provides the universal basis for the clinical identification of obese people because the anthropometric methodology is easy to implement, inexpensive, and valid. Simple measures such as height, weight or waist and hip circumferences characterise overall and regional adiposity near as accurately as sophisticated reference methods [11]. The body mass index (BMI), which reflects body general adiposity, is valid for defining obesity at the population level but does not properly account for the wide variation in body fat distribution within individuals [10, 15]. BMI does not differentiate between fat and muscle mass [15], follows a non-linear association with body fat percent [16] and is poorly prognostic of obesity-related co-morbidities in subjects of short stature [17] or older age [18]. A proper characterisation of obesity-associated risk requieres that at least one indicator of visceral fat depots is measured. Waist measures such as waist circumference (WC) or waist-to-hip ratio (WHR) are stronger proxies for abdominal obesity than body mass index [11, 19]. Even for narrow ranges of BMI large differences in WC exist that account for sizeable variations in risk of chronic conditions such as diabetes or cardiovascular disease [12, 20], suggesting that indices of general and central adiposity provide complementary information. On an individual basis, there is discrepancy about which measure would better predict risk of T2DM as two previous meta-analyses found no clear differences for BMI, WC or WHR as predictors of diabetes [21, 22]. A further recent meta-analysis suggested that waist-to-height ratio (WHtR) would show a superior predictive ability than BMI or WC, although of limited clinical utility [23]. Using a combination of anthropometric measures to account for both the amount and distribution of body fat seems the best way of characterising obesity-associated T2DM risk [1, 2]. Body proportions related to height, such as the WHtR, relative height or leg length, cause further variations in BMI which are not associated with body fat percent [24, 25], and may provide additional insight into the association of obesity and T2DM [1, 26–28].

Given the aetiologic effect of excess body fat on T2DM and in order for preventive actions to be effective, the most accurate tools for the early identification of at-risk subjects must be adopted. No specific cut-offs exist yet for defining central obesity in Spanish population on the basis of WC and there is little prospective evidence in support of the appropriateness of applying American [29] or European [30] standards to the country population. The main objective of the present study was to obtain specific estimates of diabetes risk in Spain according to different anthropometric variables in a large cohort of participants from the European Prospective Investigation into Cancer and Nutrition (EPIC)-Spain study, and to define those anthropometric values that would better predict future risk of T2DM in this population.

In supplementary analyses, height, but not sitting-height ratio, was shown to be significantly associated with a decreased risk of T2DM both in men and women, regardless of total body weight or waist circumference (Additional file 1: Table S1). Predicted T2DM risk was consistently higher at incresing BMI WHO categories, and also independently in those with central obesity according to the NCEP-ATPIII criterion (Additional file 1: Table S2). Results did not change after further adjustment for dietary variables or exclusion of participants with chronic conditions at baseline (Additional file 1: Table S3). However, evidence of heterogeneity existed by follow-up strata, and by groups of age and menopausal status. The effect of a larger WC on T2DM risk was evaluated separately in normal weight, overweight and obese participants (Additional file 1: Table S4). Women presented higher risk estimates at high WC values, with obese participants of large WC (≥ 88 cm) showing up to 5.6-fold times higher risk of diabetes than their normal weight, low WC (< 80 cm) counterparts (versus 2.8 times higher risk in men). Of note, less than 5% of women with normal WC became diabetic after 10 years of follow-up, even if obese and in the 60–69 years old group (Additional file 1: Table S5). Absolute risks were higher in men than women for any age, waist and BMI category. The population risk of diabetes attributable to excess body weight (BMI ≥ 25 kg/m²) in the EPIC-Spain cohort was estimated in 46% (95% CI, 33 - 59%) for men and 61% (95% CI, 52 - 71%) for women (Additional file 1: Table S6). If only obesity were targeted, the proportion of avoidable cases would reach an estimated 17% (95% CI, 13 - 21%) in men and 31% (95% CI, 27 - 36%) in women.

The modelling of T2DM risk related to the anthropometric indices evaluated using restricted cubic splines revealed a similar pattern for the studied indices within each sex, but appreciably different between sexes, showing curvilinear relationships with steeper slopes in women, while associations tended to reach a plateau at highest anthropometric values in men (Figure 1).

Figure 2 shows the cumulative hazard of diabetes by age for combined BMI and WC strata. Central obesity added independently to the risk of diabetes in each BMI category, but much more significantly in women, for whom the estimated cumulative hazard of diabetes in normal-waist participants remained below 15% even among the oldest.

General and central obesity were both independent predictors of T2DM risk in this large prospective cohort of volunteers from different Spanish regions. Risk estimates at elevated values of BMI and WC were higher for women than for men. Also central obesity, although a moderate independent predictor of T2DM in men, showed the strongest association with diabetes in women. The measure of waist circumference in the clinical practice would thus be a valuable and inexpensive aid in the evaluation of diabetes risk, especially for women.

The practical need to distinguish between general and central obesity from an anthropometric perspective, raises the question about which indicator would be the best proxy for central obesity either in clinical and epidemiological settings. Literature showed that relative measures of WC, such as the waist-to-hip ratio, predicted risk of disease no better than WC alone [22], which guided the definition of central obesity to rely exclusively on WC [29, 30]. But recent findings have given support to the use of WHtR as a better central obesity index in the prediction of T2DM among different ethnic groups [26, 28, 37–39]. Height influences the shape and frame size of individuals, and the evidence suggests that taller people tend to be leaner (and meager) than their shorter counterparts [24]. In the present study, taller participants had around 25% reduced risk of becoming diabetic independently of their weight and abdominal perimeter. Since height is also conceived as a proxy indicator of childhood nutritional status, its strong association with diabetes may thus integrate a double dimension of actual physical characteristics and infancy socio-economic circumstances [40]. Our results supports waist-to-height ratio as the best single measure to predict risk of diabetes, presenting the largest area under the ROC curve in both sexes (AUC_men = 0.69; AUC_women = 0.78), in line with previous evidence, with an optimal cut-off value of 0.6, above the proposed threshold of 0.5 [37]. The consistency of the predictive ability of WHtR with regard to T2DM across different ethnic backgrounds, and sex groups, together with its simplicity, supports including this promising index systematically in future epidemiological studies in the field of diabetes and other chronic diseases [37, 41]. However, the definition of universal clinical thresholds (if at all possible) still warrants further investigation in different age groups and disease contexts. From a clinical point of view, although waist measures provide no dramatic gain in discriminatory ability as compared to BMI, given their independent and graded association with T2DM, the authors firmly believe that it deserves consideration the inclusion of central obesity measures (WC the simplest, WHtR the most discriminative) in clinical practice guidelines for the management of T2DM patients to assist diagnosis and decision-making by the physician, at least in the case of female patients. The combination of BMI and WC would allow a much more accurate ranking of individuals according to their disease risk. Our results show large differences in 10-year absolute risk of T2DM between normal-waist and high-waist female participants, further supporting the relevance of WC as a complementary measure for evaluating diabetes risk in women.

As a point for discussion, no specific cut-off points have yet been adopted to define central obesity in Spain. With regard to diabetes, our results suggest optimal cut-offs at 99 cm in men and 90 cm in women. Two other criteria for defining central obesity are available which are mostly used in the diagnosis of metabolic syndrome, those proposed by the IDF for European population (94/80 cm) [30] and those of the NCEP-ATPIII (102/88 cm) [29]. Although less sensitive, the NCEP-ATPIII criterion was much more specific when applied to our cohort; besides, the IDF criterion failed to correctly classify a large proportion of the sample. Until adapted, specific values are defined, our data strongly supports the adoption of NCEP-ATPIII values in the Spanish population to enable international comparability of country data.

Obesity, both general and central, had a greater influence on the risk of T2DM in women than men, consistent with previous evidence [2]. Genetic effects determine sex differences in body composition [42] and hormonal factors [43] have been invoked to account for the weaker association between obesity and T2DM in men. However, reasons for this sex-specific effect are not clear. A gynoid fat pattern, characterised by fat tissue depots in thighs and hip, has been shown to be inversely associated to fasting and postload glucose concentrations, and to diabetes risk, independently of BMI and waist circumference [44]. Larger thigh and hip circumferences in women could reflect increased femoral and gluteal subuctaneous fat mass, which have been reported to show high lipoprotein lipase activity and low lipolytic activity [44, 45], thus contributing to fatty acid uptake and storage [45]. In turn, abdominal fat depots are more strongly related to insulin resistance than periferal or gluteal adiposity by releasing larger amounts of free fatty acids into the blood that become lipotoxic for hepatic and muscle cells [11, 46]. Thus, women with an android fat pattern (a pattern of central adiposity) might be at higher relative risk of T2DM because of both a diabetogenic effect of intra-abdominal fat and a reduced ‘protection’ by hip and thigh fat depots, as compared to gynoid females. In support of the latter, Cameron et al. have recently shown the important confounding effect of hip circumference in the association between central obesity and all-cause and cardiovascular mortality, hip circumference being inversely associated to mortality after adjustment by waist circumference [47]. In our study, hip circumference was no longer associated to diabetes risk once weight and waist circumference were accounted for in categorical models, but it showed a significant independent inverse relationship with diabetes in a continuous model (HR_{men, per standard deviation} = 0.90, 95% CI, 0.82 - 0.99; HR_{women, per standard deviation} = 0.73, 95% CI, 0.65 - 0.83; data not shown), which points out to a protective independent effect of hip circumference similar to that suggested by previous authors [44, 47]. These results highlight the aetiological role of central obesity in promoting T2DM, but also revive the discussion on the importance of hip circumference as an independent predictor of chronic disease and mortality, especially among women.

The large and increasing prevalence of overweight and obesity, and the ageing of the population in Spain, as in many other countries [6, 7, 48], raises serious concerns about the future burden of diabetes morbidity and mortality. In our study, we have estimated that around 46% of cases among men, and 61% of cases among women, could be avoided by maintaining a normal weight (BMI < 25 kg/m²). If our relative risk estimates were extrapolated to the general Spanish population, using the most accurate and up-to-date nation-wide figures available for Spain (an overall prevalence of overweight or obesity of 70.8% and 53.9% for men and women respectively [49]), the proportion of diabetes cases that could be avoided if the population kept a normal weight could be calculated in 40.1% (95% CI, 27.3 - 52.9%) in men and 49.7% in women (95% CI, 38.9 - 60.4%). Even further gains are conceivable if lower reductions in BMI or abdominal girth were achieved. However, further extrapolation of these results must be applied with caution, since prevalences and relative risk estimates may vary in other settings.

Some limitations should be considered when interpreting these data. The EPIC-Spain sample was not representative of the general population, consisted of a large proportion of blood donors and included predominantly women, which would limit the generalizability of the results. Furthermore, the prevalence of elevated BMI in this cohort was very high as compared to other populations, particularly those from Asian origin [50], this meaning that the estimated risks (relative and aboslute) and cut-off values obtained should not be directly extrapolated to populations of different geographical, ethnic, or cultural contexts. Unfortunately, no data was available on family history of diabetes, and thus the genetic background of participants could not be accounted for in the analyses. The limited availability of primary care data in one of the study centres might not have resulted in a significant misclassification bias, however, since the associations remained unchanged after excluding the participants from this centre in a sensitivity analysis. Finally, since no additional anthropometric measures were performed during the follow-up, the possibility that participants could have lost or put on weight after recruitment could not be evaluated. Furthermore, it was not possible to assess to what extent potential reductions in weight or WC during the study period would be able to decrease T2DM risk in this population. Important strengths are the large sample size, the prospective design with a long follow-up time, and the use of anthropometric measures, not self-reports. Also, the large number of cases available allowed for robust estimates of T2DM risk across the full range of relevant anthropometric indicators. Finally, a large set of confounders was available to control for, including dietary and lifestyle variables, and mis-reporting of energy intake.

Diabetes risk was consistently associated with higher overall and central obesity indices in adult population, even at BMI and WC values regarded as normal. The study provides specific estimates of diabetes risk (absolute and relative) by categories of anthropometric variables in Spanish men and women, makes the importance of central adiposity indices manifest, especially for women, and promotes the measurement of waist circumference in the clinical setting to assist the evaluation of metabolic risk.