Although the definition of obesity (BMI ≥30 kg/m²), a key risk factor of diabetes, has been validated in white populations (5), its appropriateness in Asian populations has been questioned (6). Recognizing this, a World Health Organization (WHO) expert panel was convened in 2002 to discuss the potential for developing Asian-specific BMI cutoff points for obesity (7). The consultation concluded that with the data available at the time, there was no clear BMI cutoff point that would be universally applicable to Asians and that the prespecified BMI ranges would be retained (i.e., underweight <18.5 kg/m², normal 18.5 to <25 kg/m², overweight 25 to <30 kg/m², and obese ≥30 kg/m²) for assessing the risk of obesity-related chronic diseases. Nevertheless, the WHO expert panel recommended potential BMI categories for public health action in people of Asian descent (i.e., underweight <18.5 kg/m², increasing but acceptable risk 18.5 to <23 kg/m², increased risk 23 to <27.5 kg/m², and high risk ≥27.5 kg/m²) (7). The WHO panel also emphasized the need for longitudinal studies using disease outcomes data to better understand the relationship between BMI and obesity-related diseases and to derive and validate ethnic-specific BMI cutoff points. Because of these recommendations, several studies have attempted to redefine obesity for Asians. A systematic review (8) of these studies found that, although most studies were in favor of lowering the BMI cutoff point for obesity in Asians, the majority of these studies used cross-sectional data and some did not report health outcomes.

The two main objectives of this multiethnic cohort study were 1) to compare the incidence and age at diagnosis of diabetes across white, South Asian, Chinese, and black subjects living in Ontario, Canada, one of the world’s most ethnically diverse regions; and 2) to derive ethnically appropriate population-based BMI cutoff values for obesity in assessing diabetes risk using clinically ascertained diabetes. A cutoff point of a BMI of 30 kg/m² was chosen because it represents the current standard for obesity, a key risk factor of diabetes.

Our study included survey participants living in Ontario who were aged ≥30 years at the time of survey and who identified themselves as white, South Asian (i.e., of Indian, Pakistani, Bangladeshi, or Sri Lankan origin), Chinese, or black. The surveys collected information on various sociodemographic characteristics (e.g., income, education) and risk factors (e.g., BMI from self-reported height and weight, current smoking, and inadequate physical activity). Additional details on these variables are published elsewhere (4).

The ethnicity-sex patterns of diabetes incidence observed in this study are consistent with results from earlier prevalence studies in Canada (4), the U.K. (2), the U.S. (3), and elsewhere (16). Although most of the earlier studies compared two or three ethnic groups and relied mainly on cross-sectional data, our study is the first to conduct a cohort study to compare the risk of incident diabetes in the world’s four major ethnic groups. An incidence study among individuals from different groups initially free of disease and living in a similar environment is stronger than a prevalence study because it allows for more certainty that ethnicity is a true independent risk factor for the development of diabetes while simultaneously accounting for other confounding factors that might influence the development of diabetes.

The relatively low crude rate of diabetes in the Chinese population living in Ontario might be partly attributed to the relatively low average BMI observed in this population. Our data suggest that a population shift from the normal BMI range to the obese BMI range would result in an alarming 11.7-fold–increased rate of diabetes in Chinese subjects compared with a 4.5–6.3-fold increased rate of diabetes in the other ethnic groups. Likewise, a recent trend toward urbanization and the associated rise in obesogenic behaviors already has resulted in a rapid rise in prediabetes and diabetes in the Chinese population in China (17).

Several hypotheses have been proposed to explain why non-Europeans have a higher risk of diabetes than people of European descent. Researchers suggest that non-European ethnic groups are more likely to have inherited the thrifty gene because their ancestors were more likely than Europeans to have been exposed to extended periods of starvation. The thrifty gene was advantageous during feast or famine cycles because it enabled individuals to store calories more efficiently during times of food shortages. In the present day, however, where there is an abundance of high-fat and high-calorie foods, the thrifty gene makes it difficult for individuals to control their weight. Other hypotheses for the higher risk of diabetes in nonwhite versus white subjects include a genetic susceptibility to insulin resistance, particularly in South Asian subjects (18,19); a higher likelihood of intrauterine deprivation coupled with weight gain and physical inactivity later in life; and higher central adiposity at similar BMI levels (20).

Our findings suggest that the current definition of obesity may provide a false sense of security for South Asian, Chinese, and black populations and that, even at a BMI range thought to be acceptable, the risk of diabetes may be markedly underestimated in nonwhite ethnic groups who appear to be particularly sensitive to weight gain in terms of diabetes risk. To better reflect the risk of incident diabetes, our data suggest that the current BMI cutoff value for obesity should be lowered in South Asian, Chinese, and black groups. Most of the earlier studies that attempted to redefine BMI cutoff points in Asian populations relied on prevalence data, in which BMI and obesity-related conditions were ascertained at the same point in time. This is potentially problematic because having diabetes could influence metabolic and lifestyle changes, which in turn could influence body weight. A number of previous studies also were impeded by a lack of data on specific clinical outcomes (20,21). Nevertheless, our findings are consistent with earlier reports (6,22,23) on Asian populations that recommended BMI cutoff values between 22 and 27 kg/m² for predicting the risk of diabetes, as well as of hypertension, cardiovascular disease, and mortality.

A major strength of this study was our ability to compare diabetes risk across diverse ethnic groups living in the same geographic region. This was especially important in identifying ethnic-specific BMI cutoff points for assessing diabetes risk because body composition and other determinants of obesity and diabetes, such as the built environment and availability of healthy foods, can vary widely by geographic location (24,25). Another strength of this study was our ability to achieve complete follow-up of all study participants using a highly sensitive and specific validated source of incident diabetes and several linked administrative databases of Canada’s single-payer universal health care system.

We recognize the following limitations of our study. 1) Our baseline variables were based on self-reported data; however, wherever possible, survey data were augmented using information from administrative sources. 2) BMI was calculated from self-reported height and weight and may be influenced by ethnic differences in reporting. However, an independent analysis of self-reported and measured BMI collected on a representative sample of participants of the CCHS cycle 3.1 found very high concordance between self-reported and measured BMI, irrespective of ethnicity (Supplementary Table A1) (M. Shields, Statistics Canada, unpublished data). 3) The number of nonwhite subjects was relatively smaller than the number of white subjects; however, with 2,614 nonwhite participants (who represented >1.8 million nonwhite people in Ontario), our study represents the largest multiethnic cohort study of its kind. 4) We did not have data on waist-to-hip ratio, data on family history of diabetes, or detailed information about diet. 5) We were unable to account for undiagnosed cases of diabetes or the possibility of the proportion of undiagnosed diabetes being different in the different ethnic groups. 6) We were unable to identify whether incident diabetes cases were type 1 or type 2; however, we greatly reduced the likelihood of identifying type 1 diabetes cases by limiting our cohort to individuals aged ≥30 years.