Type II diabetes has emerged as a leading cause of death and disability worldwide^(1, Reference King and Rewers²⁾. In 2010, an estimated 285 million people were living with diabetes⁽Reference Shaw, Sicree and Zimmet³⁾. Although often seen as a disease of affluence, almost three out of every four of these people live in low- and middle-income countries, with increasing evidence suggesting that diabetes transitions over time from a disease of affluence to a disease of the poor⁽Reference Sicree, Shaw and Zimme⁴⁾. The number of people with diabetes in developing nations is expected to rise 69 % by the year 2030⁽Reference Shaw, Sicree and Zimmet³⁾ – outpacing even the ‘pessimistic scenario’ forecasts of earlier global disease models⁽Reference Murray and Lopez⁵⁾.

Why has diabetes risen so rapidly in the developing world, at a much higher rate than that predicted based on our knowledge from high-income countries? According to recent reviews, most efforts to mitigate the problem in developing countries have focused on increasingly sedentary lifestyles contributing to obesity⁽Reference Prentice^6–Reference Seidell⁸⁾. However, this begs the question of how to intervene, as at present only very aggressive programmes, such as those implemented in Singapore, have been able to reverse a rise in obesity rates at a population level through exercise-based approaches⁽Reference Simmons, Unwin and Griffin⁹⁾. Despite clinical trials showing that such interventions to reduce obesity could help prevent diabetes and its complications⁽Reference Simmons, Unwin and Griffin^9, Reference Ramachandran, Snehalatha and Mary¹⁰⁾, the real-world implementation of activity-based efforts has had disappointingly low effectiveness⁽Reference Simmons, Unwin and Griffin⁹⁾.

One recent but contested proposal in high-income countries is to additionally focus on the food environment, for example by reducing the intake of high-sugar foods through taxes and related regulations⁽Reference Brownell, Farley and Willett^11, Reference Brownell and Frieden¹²⁾. Large prospective epidemiological studies demonstrate not only that overall joules contribute to obesity and diabetes, but also that the sugar content of foods is significantly associated with the onset of type II diabetes⁽Reference Salmeron, Manson and Stampfer^13, Reference Salmeron, Ascherio and Rimm¹⁴⁾. Products with refined sugars confer a high glycaemic load while having poor satiating properties, which is thought to contribute to excessive weight gain, the metabolic syndrome and insulin resistance⁽Reference Ludwig^15–Reference Schulze, Manson and Ludwig¹⁷⁾. However, emerging data also indicate that refined sugars such as fructose, which are common in processed foods, may contribute to diabetes independently of their impact on obesity, by altering the physiology of insulin responses to induce insulin resistance and impaired glucose tolerance⁽Reference Elliott, Keim and Stern^18–Reference Yoshida, McKeown and Rogers²⁰⁾. While concerns have been raised about sugar-containing food products in Western nations, do such products significantly contribute to diabetes worldwide, such that they should become a point of interest for public health policy makers?

To date, other major hypotheses have been put forward as being the principal determinants of diabetes in developing countries. One commonly held view is that demographic shift, particularly population ageing, is the major driver of diabetes emergence. As the WHO stated, ‘age is the single most important determinant’ of diabetes⁽Reference Roglic and Unwin²¹⁾, based on the increasing prevalence of diabetes with age⁽Reference Yang, Lu and Jia²²⁾. Urbanization has also been invoked as a second determinant of diabetes⁽Reference Diamond²³⁾: ‘rising prevalence of diabetes in India and other developing countries is chiefly attributed to urbanization’, reported some researchers, under the premise that city living is conducive to a more sedentary lifestyle than life in rural agricultural zones⁽Reference Ramachandran, Mary and Yamuna²⁴⁾.

Curiously, while ageing and urbanization have been used to estimate the future trajectory of diabetes, the models based on these two factors have notably underestimated actual rates of diabetes now being observed in low- and middle-income countries⁽Reference Murray and Lopez^5, Reference King, Aubert and Herman²⁵⁾. Other social and economic changes besides ageing and urbanization may therefore be contributing increases in diabetes rates⁽Reference Yusuf, Reddy and Ounpuu²⁶⁾. A third hypothesis is that food markets integrate into the global economy and as per capita income increases, people shift their dietary habits – a ‘nutrition transition’ in which people in developing countries spend a greater percentage of their money on processed foods, including new imports⁽Reference Drewnowski and Popkin^27, Reference Reddy and Yusuf²⁸⁾. Unhealthy foods with high sugar, fat and oil content are becoming cheaper even for people who have not experienced a rise in income⁽Reference Monteiro, Moura and Conde²⁹⁾, leading to a phenomenon of ‘dietary dependency’ in which integration into global markets increases access to, and associated consumption of, foods high in sugar and related carbohydrates⁽Reference Brownell and Warner^30, Reference Stuckler³¹⁾. It is thus plausible that all three factors – ageing, urbanization and the nutrition transition – are contributing to the global diabetes pandemic. Indeed, urbanization (due to increased access to unhealthy foods) and economic development (due to rising personal incomes) may be underlying factors contributing to the nutrition transition.

While urbanization, ageing and development are difficult to modify, are particular components of the diet appropriate policy targets for those who wish to stem the rising tide of diabetes? In the current paper we use data available from 173 countries to test the potential role of various nutritional components on the global food market. Our rationale was to use comparable data on diabetes prevalence worldwide and determine what statistical correlates might explain the variation in diabetes rates among countries, when controlling for other intervening factors such as differing levels of economic development. Our study starts with measures of proximal causes of varying diabetes rates – analysing how diabetes prevalence relates to physical inactivity, obesity as well as kilojoule exposure to different types of foods (if they have any effect independent of obesity). We then assess which underlying factors, from urbanization to ageing to economic growth to market integration, can explain significant proximal risks and overall diabetes rates.

Experimental methods

Data on diabetes prevalence by country (the percentage of adults aged 20 to 79 years with diabetes) was obtained from the International Diabetes Federation's (IDF) Diabetes Atlas Database for the year 2007⁽³²⁾. We use the 2007 data because this is the year for which comparative estimates of other key variables of interest (described below) are also available and because the 2007 IDF data have been subject to extensive checks on quality; furthermore, longitudinal comparisons of the IDF data are not recommended due to changes in data collection strategy between years⁽³²⁾.

To assess the population determinants of diabetes, the analysis proceeded in two steps. First, we examined the relationship between diabetes prevalence and proximate risk factors for diabetes. Given the potential importance of nutrition as a risk factor for diabetes, we performed the regression of the prevalence of diabetes v. country-specific food marketplace data taken from the FAO describing market size (in kJ/person per d) of: (i) sugars and related sweeteners; (ii) cereals; (iii) fruits and vegetables; (iv) vegetable oils and related fats, (v) meats and other animal products; and (vi) overall total joules⁽³³⁾. The food market size data are considered a standard measure of population exposure to different food items. Of note, ‘related sweeteners’ refers to carbohydrates like high- fructose corn syrup and honey, which are broken down into blood glucose, not sugar substitute products used as alternatives to sugar (e.g. saccharine or sucralose). Because diabetes develops after cumulative exposure to dietary risks (i.e. high intake of joules today does not lead to immediate obesity, but a prolonged exposure to high joules is required), we calculated the mean kilojoules per person per day over a 10-year period in each of these food categories to represent typical long-term exposure to each of these dietary components. The rationale for this decision is that the pathophysiology of diabetes is as a delayed-onset disorder, in which the metabolic syndrome leading from underlying risk factors to frank diabetes mellitus symptomatology is thought to take roughly one decade⁽Reference Fauci³⁴⁾. We also incorporated measures of sedentary lifestyle and obesity as two other proximal determinants of diabetes rates; the number of passenger cars per 1000 people was used as a standard international indicator of sedentary lifestyle (along with urbanization), and the prevalence of both overweight and obese adults (percentage of adults with BMI ≥ 25 kg/m²) was incorporated as a measure of excess weight in the population.

We also analysed how well key hypothesized population-level risk factors for diabetes – ageing, urbanization, rising incomes and market integration – relate to diabetes prevalence as well as any significant proximal risk correlates to diabetes prevalence. As described further below, we found that sugar exposure was the major proximal dietary risk correlate to diabetes prevalence, independent of obesity or sedentary lifestyle, so we investigated specifically how the underlying population-level factors may relate to sugar exposure as well as to diabetes prevalence. Data on sugar exposure are available longitudinally (not just for the year 2007), so we conducted a full time-series (panel) data analysis for this segment of the study.

Thus, our cross-national statistical models were as follows:

$${{\rm{Diabete}}{{{\rm{s}}}_i}\, = \,\alpha \, + \,\beta {\rm{DIE}}{{{\rm{T}}}_i}\, + \,\beta {\rm{FA}}{{{\rm{T}}}_i}\, + \,\beta {\rm{SE}}{{{\rm{D}}}_i}\, + \,{{{\epsilon}}_i}\eqno\rm$$

$${\rm{Diabete}}{{{\rm{s}}}_i}\, = \,\alpha \, + \,\beta {\rm{SO}}{{{\rm{C}}}_i}\, + \,{{{\epsilon}}_i}\eqno\rm$$

$$\ \,{\rm{Suga}}{{{\rm{r}}}_{i,t}}\, = \,\alpha \, + \,\beta {\rm{SO}}{{{\rm{C}}}_{i,t}}\, + \,{{{\rm{\mu }}}_i}\, + \,{{{\epsilon}}_{i,t}}\eqno\rm$$

where i is the country, α is the constant in the model, β is the individual regression coefficient for each variable, μ is a country-specific dummy for the fixed-effects time-series regression (correcting for differences inherent to each country between years in the sugar analysis, to avoid confounding by surveillance biases, as judged appropriate by a Hausman test), t is time, and ε is the error term. DIET is the average market size over 10 years in kilojoules of exposure per person per day for each food category prior to the year 2007 (sugars, labelled SUGAR; cereals; vegetable oils; meats; and total joules); FAT is the percentage of adults who are overweight; SED is the number of passenger cars per 1000 people. In the second stage of models testing population factors leading to dietary change and/or overall diabetes rates, SOC is one of the following social, demographic and economic determinants, testing each underlying hypothesis in turn: (i) AGE, the percentage of the population aged 65 years and above; (ii) URBAN, the percentage of the population living in urban areas as determined by the UN World Urbanization Prospects group (both expressed as percentages to allow comparison with international health impact assessments conducted by the World Bank); (iii) GDP, the measure of average income (per capita, in constant 2005 $US adjusted for purchasing power parity for cross-country comparability); and (iv) MKT, the number of dollars spent on food imports per capita (in constant 2005 $US for comparability). We tested each of these SOC variables separately in order to tease apart their individual contributions to diabetes prevalence. All food data were from the FAO’s FAOSTAT database from the years 1960 to 2007 (the full duration of data available); all other data were taken from the World Bank World Development Indicators Database 2011 edition and were available for the year 2007 to compare against diabetes prevalence, with the exception for overweight prevalence and passenger cars, for which 2005 data were available^{(33, 35)}.

Results

Table 1 shows the results of our models studying the proximal correlates to diabetes prevalence. Model 1 shows that higher total joules were significantly associated with greater diabetes prevalence (P < 0·001). However, in investigating the dietary components corresponding to diabetes prevalence, specific dietary factors, rather than the overall level of joules, were critically correlated with diabetes. In particular, once sugar and sweeteners were included in the model the effect of total joules disappeared, suggesting that high sugar was the principal determinant of the association between high joules and diabetes. As shown in Table 1, each additional exposure to sugars and related sweeteners of 100 kJ/person per d was associated with a 2·8 % rise in diabetes prevalence in a country, even after accounting for other components of the diet such as oils and meats (P < 0·001). The only other dietary factor significantly correlated with diabetes prevalence was cereals (P < 0·01), which are also high in carbohydrates.

Table 1

Proximal determinants of diabetes prevalence, 2007

The table describes cross-national models of diabetes prevalence. Standard errors are robust. Food exposure is in average kJ/person per d over last 10 years. Diabetes prevalence data are from the International Diabetes Federation Diabetes Atlas, 4th edition (2007). Dietary data are from the FAO's FAOSTAT database, 2011 update.

* P < 0·05, ** P < 0·01, *** P < 0·001.

Sedentary lifestyles and overweight were also significant correlates to diabetes prevalence. As shown in Table 1, sugar remained a significant predictor of diabetes independent of these factors (P < 0·001). Notably, sugar correlated with prevalence of overweight more strongly than did any other component of the diet (r = 0·69; see Supplementary Materials, Part 1).

Figure 1 shows the unadjusted correlation between exposure to sugar and related sweeteners and diabetes prevalence. It shows uneven patterns of sugar exposure worldwide, ranging from <10 kJ/person per d in poorer regions to >150 kJ/person per d in the USA. Notably, high sugar producers, such as Brazil, Jamaica, Dominica, Costa Rica, Cuba, Mexico, Jamaica, Trinidad and Tobago, exhibit high levels of diabetes prevalence. These sugar producers have very high sugar-to-total joule ratios and, as shown in Fig. 1, experience some of the highest rates of diabetes among low- and middle-income countries.

Fig. 1

Sugar exposure and diabetes prevalence, 2007: (a) low- and middle-income countries; (b) high-income countries

Additionally, we evaluated commonly proposed distal population-level risk factors for high diabetes rates: ageing, urbanization and economic development. As shown in Table 2, GDP, sedentary lifestyle and urbanization were significantly associated with increased risks of diabetes, but the significant correlations disappeared once sugar exposure was incorporated into the models. This supports the hypothesis that the observed correlation between income, lifestyle and urbanization with diabetes is driven substantially by increasing sugar exposure. Only ageing remained significantly correlated with diabetes prevalence after sugar was taken into account.

Table 2

Underlying determinants of diabetes prevalence, 2007

The table describes cross-national models of diabetes prevalence. Results are from twenty separate regression models for each factor's impact on diabetes, estimated in turn with and without the impact of sugar included. Standard errors are robust. n 173.

*P < 0·05, **P < 0·01, ***P < 0·001.

†This is the independent effect of increasing sugar exposure by 100 kJ/person per d when each individual factor is included in the model, e.g. Diabetes_i = α+βSUGAR_i+βSOC_i+ε_i (see Experimental methods).

‡Gross domestic product in constant 2005 $US, adjusted for purchasing power parity.

Determinants of sugar exposure

In view of the importance of sugar as a significant correlate of diabetes, we investigated what factors may be driving the magnitude of sugar exposure. Figure 2 depicts the relationship between economic growth (measured by GDP per capita) and exposure to sugar and related sweeteners. As shown, sugar and sweetener exposure tends to rise rapidly in low- and middle-income countries, reaching about 15 % of dietary joules at income of around $US 10 000 per capita; exposure subsequently plateaus. Supplementary Materials, Part 2 shows that, at around $US 10 000 per capita, consumption is close to 750 kJ/person per d – about 2·5-fold the recommended maximum sugar consumption level.

Fig. 2

Sugar exposure and economic development, 2007: $$$$, low- and middle-income countries; $$$$, high-income countries (GDP, gross domestic product in constant 2005 $US, adjusted for purchasing power parity)

Because data on sugar are comparable longitudinally (unlike data on diabetes prevalence), we used longitudinal models to study what factors in addition to GDP may be critically increasing exposure to sugar and related sweeteners over the past several years. Our models control for country ‘fixed effects’ – that is, they control for differences between countries that are inherent to each country, such as differences in surveillance infrastructure that could otherwise lead to biases in comparison. We did this using a standard econometric technique in which a dummy variable is created for each country and included in the regressions, so that unobserved variables that are specific to each country – e.g. the strength of a country's public health surveillance system – are controlled for implicitly. Table 3 shows the critical determinants of rising sugar exposure, both in terms of overall joules from sugar and in terms of the proportion of sugar in the diet relative to other food (joules of sugar as a percentage of overall joules). Higher GDP was a significant correlate to overall sugar exposure per capita per day (P < 0·01). Food importation was a significant correlate of both higher sugar joules overall and an increasing dietary composition of sugar (higher proportion of diet made up by sugar; P < 0·05).

Table 3

Drivers of increasing sugar exposure over time, 1960 to 2007

The table describes cross-national time-series (panel) models of sugar exposure clustered by country to reflect non-independence of sampling. Models include country-specific fixed effects. Standard errors are robust. Sugar data from the FAO FAOSTAT database, 2011 update.

*P < 0·05, **P < 0·01, ***P < 0·001.

†Gross domestic product in constant 2005 $US, adjusted for purchasing power parity.

Robustness checks

We replicated our models using independent non-IDF estimates of diabetes prevalence for the year 2008, based on a recent systematic review of health examination surveys and other epidemiological studies of fasting plasma glucose levels and diabetes prevalence⁽Reference Danaei, Finucane and Lu³⁶⁾. Supplementary Materials, Parts 3 and 4 shows consistency in our results when replicated using the alternative data set.