Diabetes mellitus is a major public health challenge in both developed and developing countries. An estimated 135 million people worldwide had diagnosed diabetes in 1995 (of which more than 95% is Type 2), and this number is expected to rise to at least 300 million by 2025. The major part of this increase will occur in developing countries and it is estimated that by the year 2025, more than 75% of people with diabetes will reside in developing countries.1 This pandemic is largely driven by the globalization of western culture and lifestyles, specifically the inter-related problems of increasing obesity2,3 and decreasing physical activity levels3 worldwide. Diets high in saturated fat with limited intake of fruit and vegetables are also incriminated in the development of glucose intolerance and the occurrence of Type 2 diabetes.4,5
Insulin resistance is an early and potentially modifiable metabolic defect in the pathogenesis of Type 2 diabetes. For almost a decade smoking has been linked with insulin resistance in clinical studies and with markers of insulin resistance including central obesity and dyslipidaemia.6,7 Given the potential public health consequences of even a small increase in risk of a common condition such as diabetes associated with smoking, a common exposure, the possibility that smoking may play a causal role in the development of Type 2 diabetes has received surprisingly little attention. Two papers in this issue of the Journal addressed this hypothesis.8,9 Will et al. present findings from a prospective study involving over 275 000 men and 434 000 women aged ⩾30 years recruited into the US Cancer Prevention study between 1959 and 1960 and followed until 1972 for incident cases of diagnosed diabetes or listing of diabetes as an underlying or contributing cause of death.8 The findings from this study are consistent with a positive association between the number of cigarettes smoked per day and the incidence of diabetes mellitus in both men and women. However, in the age-adjusted data, the evidence of a dose-response relation is limited and the effect is largely confined to those smoking more than two packs of cigarettes per day. The increased risk of diabetes observed in smokers remained significant on adjustment for potential confounders including body mass index (BMI) at baseline, alcohol use, race, amount of exercise, educational level and dietary intakes of fats and carbohydrate. On quitting smoking, rates of diabetes fell gradually to that of non-smokers, providing some evidence of reversibility of the effect.
Considered in isolation, prospective studies of diabetes incidence such as that reported by Will et al. face potentially intractable problems of confounding and bias, particularly ascertainment bias. In these data the measures of exercise and dietary exposures were rudimentary and the adequacy of adjustment for these factors is problematical. Given the insidious onset of Type 2 diabetes and the high prevalence of undiagnosed cases in the general population, there must be concern that smokers are at increased risk of testing for diabetes given the range of common conditions associated with smoking. The argument, advanced by Will et al. that smokers may in fact, be less likely than non-smokers to use health services is unconvincing and it is noteworthy that the major paper cited in support of this argument is based on a prospective study of US physicians.
This paper must, however, be set in the context of other prospective studies on the relationship between smoking and diabetes incidence. Although the majority of such studies to date have not detected a significant positive effect, most have not focused on smoking and diabetes as the major hypothesis and the majority have lacked power to detect the relatively small but important effects reported in the current study. As highlighted by Will et al.8 it is also noteworthy that of six prospective studies with data on the number of cigarettes smoked per day, four have reported positive associations with risk of diabetes. Interestingly, the prospective data from the British Regional Heart Study is regarded as one of the negative studies in this context.3 It should be noted however, that in this study of middle-aged men current smoking was associated with a 50% increased risk of diabetes relative to never smokers in analyses adjusted for age and BMI, (relative risk 1.5, 95% C.I : 1.0–2.2). This association was attenuated on adjustment for physical activity and other potential confounders and no association with the number of cigarettes smoked was detected. However, in further analyses from this cohort, based on 17 years of follow-up and a substantially increased number of cases, the findings are broadly consistent with those from Will et al. (Wannamethee SG, Perry IJ, Shaper AG. Personal communication).
Given the problems of interpretation associated with prospective studies of incident cases of diagnosed diabetes, the data from Sargent et al. are illuminating.9 This work is based on cross-sectional analysis of the association between cigarette smoking and haemoglobin A1C in 2704 men and 3358 women aged 45 to 74 who were recruited into the East Anglian component of the European Prospective Investigation into Cancer (EPIC-Norfolk). Participants with known diabetes were excluded from the analyses. Mean haemoglobin A1C concentrations (a marker of long-term glucose homeostasis) were lowest in never smokers, intermediate in former smokers and highest in current smokers. There was a dose-response relationship between haemoglobin A1C levels and both the number of cigarettes smoked per day and with total smoking as measured by pack-years. This association persisted in analysis adjusted for a range of potential confounders including BMI, waist-hip ratio, physical activity (based on an instrument with acceptable and well documented reliability and validity) and dietary variables, assessed using a standard food frequency questionnaire and plasma vitamin C concentration. In men mean haemoglobin A1C fell with increasing time since quitting smoking. The association between smoking and haemoglobin A1C levels persisted in analysis from which individuals reporting major illnesses were excluded. Given the focus on haemoglobin A1C as the outcome measure and the exclusion of known cases of diabetes, the findings cannot be attributed to ascertainment bias.
Given the previous data on smoking and risk of diabetes (reviewed by Will et al.8) and the evidence linking smoking with insulin resistance, these two papers considered together provide substantial evidence incriminating cigarette smoking as a cause of Type 2 diabetes. Concerns about residual confounding will remain. However, given that the effects of smoking on factors such as central obesity and on taste perception and diet may mediate (in part) the effect of smoking on risk of diabetes, adjustment for these variables may not be appropriate. There is a need in observational epidemiology to focus on primary environmental, lifestyle and genetic determinants of health and disease as opposed to the minutiae of multivariate models with adjustments for confounding of uncertain appropriateness or effectiveness. Moreover, just as in nutritional epidemiology we are moving from considering the health effects of nutrients in isolation to considering the effects of foods and food groups, we should also consider the effects of inter-related lifestyle variables such as smoking, lack of exercise and dietary fat in combination.
Thus the case for smoking as a causal factor in the development of diabetes is now gathering momentum. We need further evidence on the consistency of the association in different populations, ideally from cohort studies with fasting glucose measurements at baseline and follow-up. We also need data on the reversibility of the effect given the tendency towards weight gain on quitting smoking and well-designed clinical studies of the effects of acute and chronic smoking on insulin resistance. A simple causal model is unlikely. It is probable that smoking will ultimately emerge as a causal factor in diabetes in its own right via effects on glucose homeostasis and as a marker for additional causal factors such as physical inactivity and an atherogenic diet. Although cardiovascular disease has been long investigated as a consequence of insulin resistance, it may also cause insulin resistance via effects of atheroma on the rate of glucose uptake into muscle and liver. Thus the atherogenic effects of smoking may contribute to the effect of smoking on risk of diabetes. Type 2 diabetes and cardiovascular disease are increasingly regarded as overlapping syndromes with common causal factors.3,10 Smoking is now in the frame and should be presumed guilty until proven innocent.