Diabetes mellitus (DM), a chronic metabolic disorder affecting approximately 285 million adults worldwide in 2010, is estimated to affect more than 439 million people by 2030 [1]. Cigarette smoking is a problematic social and health issue affecting more than 1.1 billion people worldwide [2]. Cigarette smoking is known to promote DM development by chronic systemic and pancreatic β-cell inflammation, aggravation of abdominal fat accumulation, increased sympathetic activity, and direct toxic damage of pancreatic β-cells [3,4,5]. Smoking also increases risks of microvascular and macrovascular complications by inflammation and endothelial dysfunction in already diagnosed diabetic patients [3,6,7].

As proven in many longitudinal studies, smoking is a definite major risk factor of many cardiovascular diseases, including hypertension, myocardial infarction, and stroke [8,9,10,11,12]. Several studies have shown valid association between smoking and DM [13,14,15,16]. However, direct causal association between the two in Asian population has not been reported, although smoking rates are higher in Asians compared to those in Caucasians.

In previous studies, classification of smoking status was usually based on self-reported questionnaire as it is easy and simple. However, this may result in underestimation of actual smokers and misclassification of smoking groups due to inaccurate self-report of current smoking status [17,18]. Cotinine, a main metabolite of nicotine, is a biomarker with high sensitivity and specificity used for cigarette smoking assessment [19,20,21]. A cross-sectional study has shown positive association between environmental smoking and DM in self-reported never smokers using serum cotinine, demonstrating the effectiveness of cotinine in verifying smoking exposure [22]. Therefore, the accuracy of smoking status classification can be enhanced by co-use of traditional self-reported questionnaires and objective biomarkers such as urinary cotinine.

The objective of the present large longitudinal study was to further elucidate the association between cigarette smoking and DM by assessing incidence of new-onset diabetes mellitus (NODM) in study cohort without baseline DM using both self-reported questionnaire and urine cotinine.

According to this longitudinal study, there were higher proportions of baseline self-reported former and current smokers and cotinine-verified current smokers in those who developed NODM than those who did not. Risks for NODM were significantly higher in both self-reported and cotinine-verified current smoking groups compared to the self-reported never smoking and cotinine-verified current non-smoking group. Also, higher amount and longer duration of smoking were dose-dependently associated with increased risk of NODM.

Similar to results of our study, many previous cross-sectional and longitudinal studies have also stated that cigarette smoking is a risk factor for DM, with risk being as high as 30% to 40% in current smokers compared to never smokers [3,4,6,7]. Major difference between these above studies and ours was that the majority of previous studies classified smoking status using only self-reported questionnaire or information from medical records, which was the simplest and most conventional tool for smoking status classification. However, this method might hinder accurate classification of smoking status as not all study participants might have answered questionnaires truthfully due to social or medical disapproval. This misclassification error rate may be between 0.9% to 9.8% depending on different studies which is large enough to produce bias and wrong interpretation of study results. Our study minimized this classification error by employing both questionnaire (‘self-reported’) and urinary cotinine (‘cotinine-verified’) smoking classification systems. Our study showed that the predictability of cotinine-based categorization was slightly higher than questionnaire-based one according to Harrell's C index (C index=0.882 [95% CI, 0.867 to 0.896] vs. C index=0.879 [95% CI, 0.865 to 0.894]). The practical advantages of urine cotinine are that it is very easy to use, accurate (high sensitivity 84.8%, specificity 98.2%) [24], quick (1 to 2 minutes for results), non-invasive, and cheap (1 to 2 dollars). It is currently widely used in many public health centers and hospital smoke cessation clinics for smoking status evaluation. It is also an approved biomarker for smoking status assessment in ‘Korean National Environmental Health Study.’ Therefore, employment of dual smoking classification system is the key strength of this study.

Urine cotinine level may also increase in secondhand smokers (SHSs) and e-cigarette users. However, in this study the percentage of self-reported never-smokers with urine cotinine >50 ng/mL were 1.7%, which includes possible misclassified actual current smokers as well as SHS exposed non-smokers. Therefore, the actual percentage of SHS is assumed to be lower. There is currently no cut-off urine cotinine value to distinguish SHS from current smokers. However, many studies comment that less than 20 ng/mL is usually detected in SHS [25,26]. Also, the percentage of nicotine containing e-cigarette use in among Korean population in 2011 to 2012 were less than 2%. Therefore, the possibility that secondary smoking or e-cigarette acted as bias in our study seems low.

Consistent with results of this study, our previous cross-sectional study has evaluated the relationship of DM prevalence with self-reported and cotinine-verified smoking and found a positive association between the two [13]. However, our previous study had a cross-sectional design which could not validate the causal relationship between the two. Since the present study has a longitudinal design to evaluate the risk of NODM in a study population without baseline DM, it can be speculated that current smoking may be a possible causal factor of NODM development.

Another significant finding of this study was that current smokers who smoked more than 20 cigarettes per day and those who smoked longer than 10 years had 64% and 34% higher risks, respectively, of NODM development compared to never smokers. In particular, current smokers who smoked more than 20 cigarettes had approximately 50% increased risk of NODM compared to those who smoked less than 10 cigarettes per day. Dose-dependent DM risk elevation is a consistent finding in many previous studies [27,28,29]. Processes of pancreatic β-cell dysfunction and insulin resistance are aggravated and accelerated in individuals who are exposed to increased amount and longer duration of nicotine. Therefore, assessment of amount and longevity of smoking or nicotine exposure should be conducted thoroughly when determining the risk of NODM development.

An interesting finding of this study was that self-reported former smoking did not increase the risk of NODM development. There have been conflicting opinions on whether former smoking is associated with increased DM risk [28,30,31]. In early stages of smoking cessation, weight gain, especially that in the central abdominal region, can lead to impaired glucose tolerance, insulin resistance, and DM. However, in a long-term aspect, smoking cessation is associated with overall reduction in DM risk [30]. Results of the present study suggest that effects of smoking on insulin resistance, glucose control, and consequent DM development may be reversible. However, because lifestyle patterns and metabolic syndrome factors including blood pressure, TG, HDL-C and glucose levels in former smokers were slightly better than current smokers, this could have had a counterbalancing effect on association between smoking and NODM incidence.

Results of our study did not show significant interaction between genders, although it was more evident in males (P for interaction=0.108). The reason for this may be due to relatively smaller proportions of females who were self-reported and cotinine-verified current smokers in this study (self-reported: 2.6%; cotinine-verified: 5.8%). For more accurate analysis, future study should be conducted with similar gender proportions of current smokers. However, this is hard to realize as there is a big percentage difference between Korean male and female current smokers (male vs. female current smokers in Korea in 2016: 40.7% vs. 6.4%) [32].

This study has some limitations. First, diagnosis of NODM in this study did not include 2-hour plasma glucose level, one of criteria in DM diagnosis guideline by American Diabetes Association. This may have underestimated NODM incidence. In addition, median follow-up period of 27 months might be insufficient for accurate evaluation of NODM as progression from normal to impaired glucose tolerance and subsequent DM might occur throughout a period of 5 to 15 years [33]. The average age of study population was approximately 38 years. However, DM prevalence in Korea is approximately twice in those over 65 years compared to that in those over 30 years (29.8% vs. 14.4%) [34]. This indicates that NODM incidence might have been underestimated. Also, information on duration and amount of former smoking in self-reported former smokers is not available. The above limitations might be the reason for weak statistical power in analyzing the association between self-reported former smoking and DM. Therefore, future studies with longer follow-up period and additional information on former smokers are needed. Second, genetic components, which were not evaluated in this study, may have affected NODM development. Current studies have identified more than 50 loci associated with pancreatic β-cell dysfunction and DM development, including peroxisome proliferator-activated receptor gamma (PPARγ) gene [35,36]. As NODM in mid-thirties is a considerably early phenomenon compared to average DM development process, genetic factors might have confounding influence in this study. However, gene analysis would be difficult to conduct in large-scaled population-based study due to ethical and cost issues. Third, combined effects of abdominal obesity evaluated by waist circumference, lack of physical exercise, higher percentage of frequent alcohol consumption, and co-morbid cardiometabolic diseases might have affected results of our study as they are known risk factors for DM. However, most of the above factors were adjusted in our multivariate analyses and results were consistent with statistical significance.

The major strength of our study was that it was the first and the largest observational study of Asian population to evaluate the association between cigarette smoking and DM risk by employing two different smoking classification methods to minimize classification error.

In conclusion, this longitudinal study shows that baseline self-reported and cotinine-verified current smoking are associated with increased risks of NODM in Korean adults compared to baseline never smoking while former smoking does not show statistically significant association. NODM risk is also increased dose-dependently in current smokers with higher daily amount and longer duration of smoking. Therefore, cigarette smoking cessation should be actively recommended and educated to patients in clinical settings as it may be a reversible and modifiable risk factor of DM development.