The Relationship of Sugar to Population-Level Diabetes Prevalence: An Econometric Analysis of Repeated Cross-Sectional Data

Bridget Wagner,Sanjay Basu,Nancy Hills,Robert H Lustig,Paula Yoffe

doi:10.1371/journal.pone.0057873

Abstract

While experimental and observational studies suggest that sugar intake is associated with the development of type 2 diabetes, independent of its role in obesity, it is unclear whether alterations in sugar intake can account for differences in diabetes prevalence among overall populations. Using econometric models of repeated cross-sectional data on diabetes and nutritional components of food from 175 countries, we found that every 150 kcal/person/day increase in sugar availability (about one can of soda/day) was associated with increased diabetes prevalence by 1.1% (p <0.001) after testing for potential selection biases and controlling for other food types (including fibers, meats, fruits, oils, cereals), total calories, overweight and obesity, period-effects, and several socioeconomic variables such as aging, urbanization and income. No other food types yielded significant individual associations with diabetes prevalence after controlling for obesity and other confounders. The impact of sugar on diabetes was independent of sedentary behavior and alcohol use, and the effect was modified but not confounded by obesity or overweight. Duration and degree of sugar exposure correlated significantly with diabetes prevalence in a dose-dependent manner, while declines in sugar exposure correlated with significant subsequent declines in diabetes rates independently of other socioeconomic, dietary and obesity prevalence changes. Differences in sugar availability statistically explain variations in diabetes prevalence rates at a population level that are not explained by physical activity, overweight or obesity.

Highlights

Global diabetes prevalence has more than doubled over the last three decades, with prevalence rates far exceeding modeled projections, even after allowing for improved surveillance
Each 150 kilocalorie/person/day increase in total calorie availability related to a 0.1% rise in diabetes prevalence, whereas a 150 kilocalories/person/day rise in sugar availability was associated with a 1.1% rise in diabetes prevalence after all control variables were incorporated into the model
In countries like the Philippines, Romania, Sri Lanka, Georgia and Bangladesh, where high and rising diabetes rates were observed in the context of low obesity rates, sugar availability rose by over 20% during the study period. (It is possible that weight gain, rather than overt obesity, might account for some of the changes in diabetes, our models were repeated with overweight prevalence rather than obesity in Table S3, and with measures of physical inactivity rather than BMI in Table S4, but the results did not change)

Summary

Introduction

Global diabetes prevalence has more than doubled over the last three decades, with prevalence rates far exceeding modeled projections, even after allowing for improved surveillance. 1 in 10 adults worldwide are affected by diabetes [1]. Most of the worldwide rise is thought to be type 2 diabetes linked to the ‘‘metabolic syndrome’’ – the cluster of metabolic perturbations that includes dyslipidemia, hypertension, and insulin resistance. Obesity does not fully explain variations and trends in diabetes prevalence rates observed in many countries. High diabetes yet low obesity prevalence are observed in countries with different ethnic compositions, such as the Philippines, Romania, France, Bangladesh and Georgia, there are likely surveillance quality differences between nations [6,7]. There are not obvious ethnic or socio-demographic commonalities between these countries to explain these observations This population-level puzzle is accompanied by individual-level data. About 20% of obese individuals appear to have normal insulin regulation and normal metabolic indices (no indication of diabetes) and normal longevity [8], while up to 40% of normal weight people in some populations manifest aspects of the ‘‘metabolic syndrome’’ [9,10,11,12]

Methods

Results

Conclusion