Methods In a population-based survey among South Asian and African Mauritians (n=7978), 1241 deaths occurred during 15 years of follow-up. In a model that included variables used in previous CVD risk calculations (a Framingham-type model), the association between waist circumference and mortality was examined before and after adjustment for hip circumference. The percentage with an increase in estimated 10-year cumulative mortality of >25% and a decrease of >20% after waist and hip circumference were added to the model was calculated.

Results Waist circumference was strongly related to mortality only after adjustment for hip circumference and vice versa. Adding waist and hip circumference to a Framingham-type model increased estimated 10-year cumulative CVD mortality by >25% for 23.7% of those who died and 15.7% of those censored. Cumulative mortality decreased by >20% for 4.5% of those who died and 14.8% of those censored.

Hip circumference, although directly correlated with waist circumference (and BMI), has been shown to be inversely associated with blood glucose, blood pressure and lipids.18 It is also associated with reduced risk of CVDs,19–22 all-cause23^,24 and cardiovascular mortality,19^,25 particularly when adjusted for waist circumference.21^,23^,26

This difference between the adverse effects of elevated waist circumference on the one hand and the protective effects of hip circumference on the other may be due to the observation that most of the pathological influence of adiposity is due to visceral, rather than superficial, fat.27 Waist circumference is increased by increasing superficial fat, and failing to account for this may underestimate the real effects of visceral obesity. The waist-to-hip ratio attempts to do this, but assumes that this can be achieved by a simple ratio. Furthermore, a specific protective role of gluteofemoral body fat via the regulation of fatty acid release and uptake and a beneficial adipokine profile is also now being recognised.28 By improving the assessment of adiposity, we may be able to improve CVD risk prediction. We therefore hypothesized that adding waist and hip circumference to traditional risk factors would significantly improve CVD risk prediction.

A population-based survey was conducted in Mauritius, with measurements taken in 1987, 1992 and 1998, details of which were published previously.29–32 In 1987, 10 randomly selected clusters and a purposely selected area of Chinatown in the capital city, Port Louis, were surveyed. In 1992 and 1998, previous participants were re-surveyed, along with any new eligible residents of the 10 clusters. Three additional clusters were added in 1992 and re-surveyed in 1998. All individuals aged ≥25 years (1987 and 1992) or ≥20 years (1998) were tested. A total of 8932 individuals (71% Asian Indian, 29% African ancestry), participated in the three surveys [excluding Chinese individuals (n=626) who were removed from this analysis]. Sixty percent participated in more than one survey and 28% in all three surveys. Response for each survey was >85%.30 Informed written consent was obtained from all participants. The mortality follow-up protocol was reviewed and approved by the local ethics committee (Ministry of Health and Quality of life, Mauritius). The 1987, 1992 and 1998 surveys were additionally approved by the ethics committee of the International Diabetes Institute (Melbourne, Australia).

Mortality follow-up was conducted between 1 April and 30 October 2007. Contact with participants (or their next of kin, other household members or neighbours when participants were not contactable) was used to determine vital and migration status. National death registries were also used to verify vital status and cause of death. Identity was validated using name, age, address, ethnicity, number of children and occupation from previous surveys. Where possible, death certificates or certified extracts of the certificate were inspected and cause of death noted.

Individuals were followed from the date of the first interview. Participants whose vital status was unknown, but who contributed some follow-up time were censored at the date of last contact (n=333). Participants not contributing any follow-up time (attended one survey and then were lost) were excluded from all mortality analyses (n=409), as were those with a history of prior CVD or lower limb amputation (n=545). Vital status was established for a total of 7978 participants.

Measurements at each survey were conducted according to identical protocols, except where stated. A 75-g oral glucose-tolerance test was performed on all non-pregnant participants, except those taking insulin or oral hypoglycaemic drugs. Biochemical measurements, height, weight and blood pressure were measured as previously described.29^,30^,33 Hip circumference was measured as the maximum circumference at the level of the buttocks. Waist circumference was measured as the minimum circumference, in the horizontal plane, between the xiphisternum and umbilicus in 1987 and as the circumference at the mid-point between the iliac crest and the lower margin of the ribs in 1992 and 1998.34 Mean waist circumference measured at the mid-point between the ribs and the iliac crest has been shown to be higher (1.5cm in men, 2.7cm in women) than at the narrowest waist.35 Therefore, 1987 waist circumference values were increased accordingly. Obesity (as a dichotomous variable) was classified using recommended Asian waist circumference cut-points (≥80cm in women and ≥90cm in men) for the South Asian population.36 Information about smoking status, ethnicity, treatment for hypertension and previous CVD or amputation was collected by interviewer-administered questionnaire. Treatment for hypercholesterolaemia was not assessed in this study.

Those who died from CVD or other causes were older and had higher levels of waist circumference and waist-to-hip ratio than survivors at study entry (Table 1). Hip circumference was positively correlated with waist circumference (r=0.76). Diabetes prevalence was 13.0% in 1987, 17.8% in 1992 and 23.3% in 1998 in this cohort.

Before adjustment for hip circumference, no association was seen between waist circumference and CVD (P=0.28) or all-cause mortality (P=0.85) in a model also containing those variables used in risk prediction models based on the Framingham Heart Study (age, sex, total cholesterol, HDL cholesterol, systolic blood pressure, diabetes mellitus, blood pressure treatment and cigarette smoking). After additional adjustment for hip circumference, a strong and graded relationship with both all-cause and CVD mortality was seen for waist circumference (Figure 1, with sex- and ethnicity-specific results presented in Figures 2 and ​and33 and Supplementary Figures S1–S6 available as Supplementary Data at IJE online). Similarly, a strong negative relationship between hip circumference and both CVD and all-cause mortality was only observed once adjusted for waist circumference (Figure 1). The risk of death from both CVD and all causes was influenced by both waist circumference and hip circumference. The hazard ratio for CVD death in a fully adjusted model was 1.31 per standard deviation (SD) of waist circumference and 0.72 per SD of hip circumference (both P<0.0001). Findings were unchanged when those with diabetes were excluded, and in subgroup analysis of those with either diabetes, impaired fasting glucose or impaired glucose tolerance or those with none of these conditions (results not shown). For men only, we found a non-linear relationship between waist circumference and all-cause and CVD mortality (in both ethnic groups, P<0.05). A J-shaped relationship was apparent with increasing risk of death for greater waist circumference only >75cm.

We tested whether the relationship of each of waist and hip circumference with mortality could be equally well described by linear effects, which was not the case (P-value for the addition of non-linear terms to a model with linear terms <0.0001 for all-cause and P=0.026 for CVD mortality). In addition, we tested whether either waist or hip could be omitted from the model, which was strongly rejected (P-value for the addition of the second term <0.002 for both variables and both outcomes). Neither the inclusion of waist-to-hip ratio (P=0.242 for all-cause; P=0.171 for CVD) nor the product of waist and hip circumference [P=0.08 (all-cause) and P=0.083 (CVD)] improved the model with non-linear terms of waist and hip circumference. We tested whether waist-to-hip ratio could replace the non-linear waist and hip circumference terms and this was also strongly rejected (for the addition of waist and hip terms to a model containing waist-to-hip ratio, P<0.0001 for all-cause; P=0.010 for CVD). Finally, the effect on the relationship of waist circumference and CVD mortality upon the addition to the model of height rather than hip circumference was estimated, with waist circumference remaining non-significant following the addition of height (P=0.26).

To assess change in predictive power of a Framingham-type model upon the addition of waist and hip circumference, estimated 10-year cumulative CVD mortality was calculated for the base model, and for an extended model including non-linear terms of waist and hip circumference. These were plotted against each other separately for those with and without an event (Supplementary Figure S7 available as Supplementary Data at IJE online) with the figure also including the quartiles, 10th and 90th centiles of the distributions. This figure shows that adding waist and hip circumference decreased the estimated cumulative CVD mortality in those without events, and increased it for those who died, thereby increasing predictive power. A calibration plot of observed and predicted 10-year cumulative CVD mortality risk for these two models (Supplementary Figure S8 available as Supplementary Data at IJE online) shows improvement in model fit upon the addition of waist and hip circumference.

Of those who died from cardiovascular causes, 23.7% had an increase in estimated risk >25% upon the addition of waist and hip circumference to the model, whereas 4.5% had a decrease in estimated risk of >20% (Table 2). Estimated mortality risk increased for two-thirds of those who died from CVD upon addition of waist and hip to the model. For those who were censored (i.e. remained alive or died of a non-CVD cause), the percentage whose estimated risk increased by >25% and the percentage whose estimated risk decreased by >20% was similar, at ~15%. A similar pattern was observed for all-cause mortality.

Using the same Framingham-type model in the South Asian population only, those above the recommended Asian waist circumference cut-points for obesity (80cm in women, 90cm in men) had a hip-adjusted risk of CVD death 1.8 [95% confidence interval (CI) 1.2–2.7] times greater in women and 1.5 (95% CI 1.1–2.2) times greater in men, compared with those below the obesity cut-point. Before the inclusion of hip circumference, central obesity according to the relevant cut-points was not associated with CVD mortality [women: 1.3 (95% CI 1.0–1.8), P=0.078; men: 1.0 (95% CI 0.8–1.3), P=0.947]. In the Creole population, a similar effect of the Asian obesity cut-point was seen, although the increase in the relative risk upon the inclusion of hip circumference in the model was less marked among women (see Supplementary Table S1 available as Supplementary Data at IJE online).

Narrow hip circumference has been shown to be a strong risk factor for mortality in several previous studies among Europid populations,19^,23–26 although in one study this finding was among men only.19 A small number of studies have noted that, as we found here, the strength of the relationship between hip circumference and either heart disease or mortality only becomes apparent following adjustment for waist circumference.20^,21^,23 Almost no attention has been given, however, to the impact of adjusting waist circumference for hip circumference. Only two studies that we are aware of have replicated our finding that the effect of waist circumference was underestimated (or absent) before adjustment for hip circumference. Canoy et al.,21 in their report on the effects of body fat on risk of coronary heart disease in the European Prospective Investigation of Cancer-Norfolk study, noted that ‘adjustment for hip circumference increased the risk prediction afforded by using waist circumference alone by ≈10% to ≈18% in men and women’.21 Likewise, in the large and multi-ethnic Epidemiological arm of the Diabetes Reduction Assessment with Ramipril and Rosiglitazone Medication (DREAM) trial (EpiDREAM) cohort, De Koning et al.40 observed that the true effect of waist circumference on glucose levels only became apparent after adjustment for both waist and hip circumferences. We are not aware of any studies that have previously reported the effect on mortality risk of waist circumference adjusted for hip circumference.

Several mechanisms may explain our findings. The first is hinted at by the positive correlation between hip and waist circumference, despite increases in one measure being protective and increases in the other detrimental. Specifically, waist circumference effectively measures both visceral and subcutaneous adipose tissue, whereas hip circumference provides a more specific measure of subcutaneous gluteofemoral adipose tissue. Adjusting a measure of visceral and subcutaneous adiposity for a measure of subcutaneous adiposity may therefore provide a more precise measure of visceral adipose tissue than is possible by measurement of waist circumference alone. Physiological explanations for our findings may also be important, and include previously observed differences in the effect of gluteofemoral subcutaneous adipose and visceral subcutaneous adipose tissue on metabolic health,28^,41^,42 and the possibility that larger hip circumference reflects greater muscle mass. Manolopolous et al.28 noted in their review on the topic that based on emerging evidence from in vitro and in vivo studies of adipose tissue physiology, ‘gluteofemoral fat may protect our bodies, irrespective of gender, by trapping excess fatty acids and preventing chronic exposure to elevated lipid levels’. In addition, a beneficial adipokine profile (leptin and adiponectin in particular) was also suggested as a potential protective mechanism. Whether the effects of lower body adipose tissue differ between Europid and non-Europid populations is largely unknown. On the one hand, Chowdhury et al.,43 in their study of Indian and Swedish men of the same age and BMI, found higher glucose levels in the South Asian men were due to differences in lower leg muscle mass rather than differences in visceral fat, potentially suggesting that physiological differences between ethnic groups may lead to different effects of hip circumference on mortality risk. On the other hand, a powerful confounding effect of hip circumference on the relationship of waist circumference with glucose was observed in the multi-ethnic EpiDREAM study [including Europid (53%), African (6.5%), Latin American (8.1%), South Asian (20.3%) and indigenous (12.1%) participants], which is suggestive of a similar effect in populations other than the South Asian and African populations included here.40 Further studies including both waist and hip circumference in different population groups and with different outcomes are clearly required.26

Our finding of a marked increase in the predictive capacity of a Framingham Heart Study-based model for CVD or all-cause mortality following the addition of waist and hip circumference suggests that these anthropometric measures may be a useful addition to such models in these populations. Taking account of both waist and hip separately, rather than as a ratio measure, has been shown in this study to be a more powerful risk marker of premature death. This supports the previous conclusion of Seidell et al.42 that the waist-to-hip ratio is a poor reflection of the specific effects of each measurement,23 as individuals with an identical waist-to-hip ratio can have different waist and hip circumferences,44 and using a ratio measure prohibits examination of non-linear associations between the numerator (waist) and the denominator (hip).20

The relationship between obesity and premature mortality has not been widely studied in South Asian populations. A large study from India, and two studies from Bangladesh, found leanness, rather than being overweight or obese, to be a major mortality risk factor.45–47 Those studies, however, were conducted in lean, rural populations and did not account for the protective effect of hip circumference. In rural populations from less developed countries (a large percentage of the native South Asian populations), low BMI reflects under-nutrition, low socio-economic status, poor hygiene and higher resultant mortality.45^,48 Urbanization, ageing of populations and increasingly Western lifestyles mean that overweight and obesity are becoming increasingly important causes of death and disease in the developing world.49 A very recent pooled analysis of obesity and risk of death that included two of the cohorts mentioned above, as well as a cohort study from Mumbai with 13001 deaths, also failed to show an association between BMI and risk of death.50 Again, this study did not account for the protective effect of hip circumference, with their finding mirroring our observation of no association between abdominal obesity and premature mortality before hip circumference was taken into account.