In this study, current smoking increased the risk for incident diabetes mellitus in a large cohort of elderly women in a dose-dependent fashion. No statistically significant interaction existed between smoking and BMI on diabetes risk. All results were similar after adjusting for multiple demographic and lifestyle variables.
Our results are consistent with previous studies examining the risk of diabetes based on smoking status in women. Analysis of data on 434,637 women collected from 1959–1972 in the Cancer Prevention Study I (CPS-I) observed an increased risk for diabetes among women smoking >1 pack per day.(Will et al., 2001
) Body weight did not modify the association between smoking and diabetes among men and women. However, this study was not specifically designed to measure such an association, and the authors did not report the test of for an interaction between BMI and smoking status.(Will et al., 2001
The Nurses Health Study observed a multivariate-adjusted relative risk of diabetes of 1.42 (95% CI 1.18–1.72) among women smoking ≥25 cigarettes daily compared to non-smokers over 12 years of follow-up. The relative risk for diabetes among women with a BMI ≤29 smoking ≥25 cigarettes daily was not significantly elevated when compared to non-smokers (1.23, 95% CI 0.87–1.74). However, among those with a BMI >29, the relative risk increased to 1.40 (95% CI 1.11–1.75).(Rimm et al., 1993
) At 16-years of follow-up, the Nurses Health Study reported a multivariate-adjusted relative risk of diabetes mellitus of 1.39 (95% CI 1.02–1.88) for normal weight and 1.40 (95% CI 1.14–1.71) for overweight current smokers smoking ≥15 cigarettes daily. Among obese individuals, they observed a significantly elevated multivariate-adjusted relative risk among former (1.24, 95% CI 1.12–1.39), light (1–14 cigarettes daily) current smokers (1.47, 95% CI 1.17–1.85), and heavy (≥15 cigarettes daily) current smokers (1.31, 95% CI 1.10–1.56).(Hu et al., 2001
) These results compare similarly to our multivariate-adjusted hazard ratios across respective BMI strata. While the results are similar, the Nurses Health Study included only working women in health professions.(Rimm et al., 1995
, Hu et al., 2001
) The similarity of our data from the IWHS cohort makes the association of smoking with diabetes risk applicable to a more general population of older females.
A recent meta-analysis of 25 studies representing 1.2 million participants and 45,844 incident cases of diabetes found a pooled relative risk for diabetes of 1.44 (95% CI 1.31–1.58) for current smokers compared to non-smokers. Of the 7 studies reporting gender-specific data for women, the pooled relative risk for diabetes among current smokers (not corrected for BMI) was 1.25 (95% CI 1.03–1.46) compared to non-smokers. Regardless of BMI and gender, the pooled relative risk for diabetes among former smokers was 1.23 (95% CI 1.14–1.33).(Willi et al., 2007
) In contrast to our findings, this meta-analysis reported a modest effect modification of BMI on the risk for diabetes among smokers. The relative risk for smokers with a BMI ≥25 was 1.57 (95% CI 1.35–1.82), while those with a BMI <25 had a relative risk of 1.34 (95% CI 1.13–1.58).(Willi et al., 2007
) However, this stratification included both men and women across only two levels of BMI and two levels of current smoking intensity.
Our results align with previous studies to support the well-documented association between cigarette smoking and diabetes risk in women.(Willi et al., 2007
, Foy et al., 2005
, Will et al., 2001
, Hu et al., 2001
, Rimm et al., 1993
) While smoking impacts BMI, our findings suggest a BMI-independent mechanism through which smoking elevates diabetes risk. Our study also adjusts for more potential confounding variables than most previous studies. For example, among other large studies specifically examining the risk for diabetes mellitus in women based on smoking status, the Nurses Health Study did not adjust for diet or physical activity.(Hu et al., 2001
) The CPS-I study adjusted for education, diet components, physical activity, and alcohol, but did not consider calorie intake or marital status in their model.(Will et al., 2001
) The Insulin Resistance Atherosclerosis Study accounted for abdominal obesity and alcohol consumption but not physical activity, diet components, education, or marital status.(Foy et al., 2005
Proposed mechanisms to explain the observed relationship between smoking and diabetes include decreased insulin sensitivity, abdominal obesity, endothelial dysfunction, and antiestrogenic effects of smoking in women.(Eliasson, 2003
, Celermajer et al., 1993
, Khaw et al., 1988
, Simon et al., 1997
) Smoking has been shown to acutely provoke hyperglycemia, elevated insulin levels, and hypertension.(Frati et al., 1996
, Facchini et al., 1992
) However, others imply an insulin-independent mechanism through which smoking may mediate diabetes risk.(Godsland et al., 1992
, Wareham et al., 1996
) Our results imply that effect modification of BMI on the smoking-mediated risk for diabetes mellitus may be less important than previous data suggest.(Rimm et al., 1993
The major strength of this study is the large, prospective cohort design with long-term follow-up. This allowed us to conduct multivariate analyses adjusted for 17 anthropometric, demographic, medical, and dietary variables potentially associated with incidence of diabetes.
Our study possesses several limitations. First, the cohort consists of a homogenous white, elderly, female population, impacting the ability to generalize our findings. Second, we based our results on self-reported data, including height, weight, smoking status, and diabetes status. Incident diabetes was not based on biological confirmation. However, previous studies demonstrated self-report as a valid method for detecting diabetes diagnoses and smoking status.(Midthjell et al., 1992
, Kenkel et al., 2003
) Another large study showed a <1 kg difference in reported and measured weights in women ≥60 years old.(Kuczmarski et al., 2001
) Other large studies examining smoking and/or BMI as risk factors for diabetes mellitus have also been self-reported.(Rimm et al., 1993
, Hu et al., 2001
, Simon et al., 1997
, Manson et al., 2000
) Previous studies have supported the reliability and validity of our food frequency questionnaire(Munger et al., 1992
) and measurement of waist and hip circumferences. (Kushi et al., 1988
) Third, our study only analyzed smoking status and BMI at baseline. It could not account for changes in status during follow-up that could lead to misclassification of exposure. Fourth, the baseline prevalence of smoking in our study (14%) at the time of the first IWHS survey (1986) was lower than the national smoking prevalence (30.1% in 1985).(Centers for Disease Control and, 2007
) However, this difference would not impact the internal validity of our results. Fifth, although this analysis adjusted for a large number of factors, the possibility of residual or unmeasured confounding exists. Sixth, although a relatively small portion of the cohort (N=1,189) did not return any follow-up questionnaires, their exclusion introduces the possibility of selection bias. Finally, our study’s large sample size allowed us to examine the association of diabetes and smoking status within strata defined by BMI. Even with this large sample size, statistical power to detect effect modification remains modest. We were, however, unlikely to have missed detecting large effects.
Our results allow clinicians to counsel their female patients that smoking acts independently of body weight to increase diabetes mellitus risk. The independent smoking-associated risk for diabetes may add to the diabetes risk associated with weight gain commonly following smoking cessation. This supports the need for clinician-directed weight control following smoking cessation.
While these data show no interaction between smoking and BMI on the risk for diabetes in women, the role of other potential interactions on diabetes risk with smoking remains unclear. Future studies must address other potential mediators such as physical activity, dietary components, alcohol consumption, or socioeconomic status. Evaluating diabetes incidence through physician-report or biological measures would also provide more credence to previous associations.