The higher proportion of excess deaths in females compared to males is explained by their lower background mortality levels, and the larger increase in the absolute risks of dying in women compared with men if diabetic, in almost all age groups. Although diabetes is often perceived as a disease of affluent countries, the proportion of all deaths that are attributable to diabetes in developing countries is not negligible. This issue of considerable preventable mortality due to diabetes has been recognized for type 1 diabetes   2 , but the vast majority of persons have type 2 diabetes.

A potential source of error in this study is that relative risks of dying in persons with diabetes, compared to those without diabetes, were obtained from studies conducted in a relatively small number of countries, most of them developed. Because little data from developing countries has been published in a format suitable for this study, unpublished data from the DECODE and DECODA studies have been used (Jaakko Tuomilehto, personal communication). This is unlikely to have overestimated the burden of diabetes mortality in developing countries, as available studies from Mauritius and Brazil show that the risk of death is about three times higher in persons with known diabetes, compared to individuals with normal blood glucose, or the general population   3   4 . This is consistent with the sparse information available from low-income countries indicating a poor prognosis for persons with diabetes   5 .

The latest validated available country-specific mortality rates were for the year 2001 and it is possible that the overall number of deaths in each country has not been accurately estimated because these mortality rates, rather than those for the year 2007, were applied to the estimated population size in the year 2007. It is unlikely, however, that the country-specific adult mortality rates have changed substantially since 2001.

The age and sex-specific prevalence of diabetes by country used in this study was estimated from population-based surveys. However, population-based studies of diabetes prevalence have not been conducted for more than one-third of the countries of the world, so the diabetes prevalence for many countries was obtained by extrapolation, and error is possible. The prevalence estimates include both diagnosed and undiagnosed diabetes.

The relative risks of dying were derived from cohort studies of patients with diabetes. The proportion of undiagnosed diabetes varies between populations, but is rarely lower than 30%, and is often higher than 50%, even in developed countries. The relative risk of dying may not be the same for diagnosed and undiagnosed diabetes, but published data from the DECODE study show that mortality in people with undiagnosed diabetes is as high as in people with previously diagnosed diabetes   6 . Moreover, people with a lesser degree of hyperglycaemia, impaired glucose tolerance (IGT), have a 40% increased mortality, regardless whether they progress to diabetes or not   7 .

One of the reasons for non-diagnosis of diabetes could be lack of symptoms, which could reflect a milder metabolic disturbance and possibly a better prognosis. If so, the calculated 3.8 million deaths could be an overestimate of the true number. The results of the DECODE study indicate that risk of death is not significantly different between previously and newly diagnosed persons with diabetes, when unrecognized diabetes is defined by the two-hour post-load glucose value rather than by fasting glucose   6 .

Although these mortality estimates are unlikely to be accurate, given the assumptions on which the calculations are based, they do provide a more realistic estimate of diabetes-attributable mortality than currently exist. The number of deaths related to hyperglycaemia would likely be even higher if mortality attributable to IGT were taken into account   8 .