Smoking and the risk of type 2 diabetes in Japan: A systematic review and meta-analysis

Cigarette smoking is the leading avoidable cause of disease burden. Observational studies have suggested an association between smoking and risk of type 2 diabetes mellitus (T2DM). We conducted a meta-analysis of prospective observational studies to investigate the association of smoking status, smoking intensity, and smoking cessation with the risk of T2DM in Japan, where the prevalence of smoking has been decreasing but remains high. We systematically searched MEDLINE and the Ichushi database to December 2015 and identified 22 eligible articles, representing 343,573 subjects and 16,383 patients with T2DM. We estimated pooled relative risks (RRs) using a random-effects model and conducted subgroup analyses by participant and study characteristics. Compared with nonsmoking, the pooled RR of T2DM was 1.38 (95% confidence interval [CI], 1.28–1.49) for current smoking (19 studies) and 1.19 (95% CI, 1.09–1.31) for former smoking (15 studies). These associations persisted in all subgroup and sensitivity analyses. We found a linear dose-response relationship between cigarette consumption and T2DM risk; the risk of T2DM increased by 16% for each increment of 10 cigarettes smoked per day. The risk of T2DM remained high among those who quit during the preceding 5 years but decreased steadily with increasing duration of cessation, reaching a risk level comparable to that of never smokers after 10 years of smoking cessation. We estimated that 18.8% of T2DM cases in men and 5.4% of T2DM cases in women were attributable to smoking. The present findings suggest that cigarette smoking is associated with an increased risk of T2DM, so tobacco control programs to reduce smoking could have a substantial effect to decrease the burden of T2DM in Japan.


Introduction
The United States Surgeon General's report recently documented a 40% increase in the risk of type 2 diabetes mellitus (T2DM) among cigarette smokers compared with nonsmokers, based on a systematic review and meta-analysis of 46 prospective studies, and concluded that cigarette smoking is a cause of T2DM. 1 This conclusion has been supported by a more recent and vigorous systematic review and meta-analysis of 88 prospective studies. 2 Although it remains debatable whether a causal relationship between smoking and T2DM has been established, 2e4 eliminating smoking may considerably reduce the burden of T2DM. For example, Pan et al estimated that 11.7% of diabetes cases among men and 2.4% of diabetes cases among women would be attributable to active smoking if smoking is causally related to diabetes. 2 Because there are substantial differences in the prevalence of smoking among countries, 5 the burden of diabetes that is attributable to smoking likely varies across countries. Quantification of the country-specific burden of diabetes associated with smoking would help guide country-specific evidence-based policies.
In Japan, the prevalence of diabetes has been steadily increasing and is expected to increase 10% by 2030. 6 Obesity is not common in Japan, 7 so preventative strategies that target weight loss may not be as effective in Japan as in Western populations. 8 Given the high prevalence of smoking, especially among young men (approximately 32%), 7 tobacco control may have a substantial importance in managing diabetes in Japan. However, there has been no systematic evaluation of the association between smoking and the risk or burden of diabetes in Japan. Recent systematic reviews 1,2 of worldwide studies did not include two Japanese studies. 9,10 Furthermore, increasing evidence from epidemiological studies also suggests that passive smoking is associated with an increased risk of diabetes. 2,11e13 Therefore, the present study was performed to provide 1) a quantitative summary of the association between smoking status (current smoking, former smoking, smoking cessation years, and passive smoking) and the risk of T2DM in Japan and 2) the population attributable fraction (PAF) of diabetes due to smoking in Japan.

Search strategy
We conducted a systematic search of MEDLINE for the literature published through December 2015 of studies addressing the association between tobacco smoking and T2DM. The Ichushi (Japana Centra Revuo Medicina) database was also searched to identify studies written in Japanese. We used the following texts and keywords in combination with both MeSH terms and text words: diabetes mellitus, type 2 or diabetes mellitus, prediabetic state, smoking, smoking cessation, passive smoking, tobacco, smokeless tobacco use, cigarette, incidence, cohort studies, follow-up studies, survival analysis, Japan, and Japanese. We also searched the reference lists of publications included in the meta-analysis and relevant reviews.

Selection criteria and data extraction
We identified articles eligible for further review by performing an initial screen of identified abstracts or titles. The second screening was based on the full-text review. Two investigators (SA and AG) independently assessed the full text for eligibility; discrepancies were resolved via consensus or determined by a third investigator (TM). Only prospective cohort studies of Japanese populations living in Japan were included. We also considered studies for inclusion if the investigators reported data from an original study and the study was conducted among adults without T2DM at baseline. Exclusion criteria were studies that included participants with a specific disease. In case of multiple publications related to the same study, we included the reports with the longest follow-up or the largest number of incident cases of T2DM.
From full-text articles, we extracted data on the year of publication, study design, number of participants, exposures, the time of the exposures assessment, outcomes, confounders, and the measures of association. The main exposure variable of interest was the presence or absence of tobacco smoking at baseline. The preferred reference group was never smokers. The majority of studies defined a group of former smokers, but a few studies defined smokers and nonsmokers without mentioning whether former smokers were included in the nonsmoking group.
The outcome variable of interest was T2DM. The definitions and diagnostic criteria to define T2DM varied somewhat across studies. The criteria used to define T2DM have changed over time, as is evident by comparing the World Health Organization 1985 criteria 14 17 (FPG !126 mg/dL or HbA1c ! 6.5%). The diagnosis of diabetes was based on objective measurement (blood tests) except for in one study, which solely based diagnosis on self-reporting by patients. We included information available from publications, but when we did not obtain sufficient information about the outcome, exposure, and study design from the article, we communicated with the authors of the original reports for further details.

Quality assessment of the included studies
Using the Newcastle-Ottawa Scale, 18 we assessed the overall quality of each study by totaling scores of the 9 criteria (0e9 stars): the representativeness of the exposed cohort, the selection of the nonexposed cohort, ascertainment of exposure, and outcome of interest not present at the start of the study (maximum of 4 stars); comparability of the cohorts on the basis of study design and analysis (maximum of 2 stars); and finally, the assessment of the outcome (maximum of 3 stars). Studies with scores of !6, 4e5, and 0e3 were defined as a high, moderate, and low quality studies, respectively.

Statistical analysis
Relative risks (RRs) were used as the common measure of association across studies. Hazard ratios and incidence density ratios were directly considered as RRs, and odds ratios were regarded as approximate to RRs in view of the low incidence rates. Pooled risk estimates were performed according to the type of smoking. We used DerSimonian and Laird random-effects models for calculating the summary estimates. 19 We used funnel plots and Egger's regression asymmetry test to assess publication bias. 20 Additionally, we performed trim-and-fill procedures to further evaluate possible effects of publication bias. 21 We also conducted subgroup analyses according to follow-up years ( 10 vs. >10 years), sample size ( 20,000 vs. >20,000), number of confounding factors ( 8 vs. >8 factors), mean age ( 50 vs. >50 years), and diagnostic criteria of diabetes (FPG!126 mg/dL only vs. FPG!126 mg/dL or HbA1c ! 6.5). We assessed the difference in association between groups using meta-regression analysis. We undertook sensitivity analyses by excluding studies in which former smokers were included in the nonsmoker group.
In assessing dose-response relationships, we treated the number of cigarettes smoked per day as the explanatory variable. Because most studies reported cigarette consumption as categorical 22    data, we assigned the mid-value of each category and 0 to nonsmokers (reference).
For the highest open-ended category, the assigned number of cigarettes smoked per day was calculated as the lower boundary multiplied by 1.2. 22 We used a random-effects generalized least-squares regression model to assess the pooled dose-response relation between smoking and risk of T2DM. To examine the potential nonlinear relationship of the number of cigarettes consumed with the risk of T2DM, we used restricted cubic splines with three knots placed at the 10th, 50th, and 90th percentiles of the distribution. We assessed the effects of smoking cessation on the risk of T2DM from 3 studies 9,23,24 that reported the risk of T2DM in relation to duration of smoking cessation. We considered three categories of smoking cessation: less than 5 years, 5e9 years, and 10 years or more. Never smoking was used as reference category in all the studies. One study 23 presented results for two separate categories of less than 5 years since quitting smoking (<3 and 3e5 years) and another study 24 presented results for three separate categories of 10 years or more since quitting smoking (10 to <15, 15 to <20, and !20 years). We combined those additional categories using our meta-analysis approach. We estimated the PAF using the formula [prevalence of smoking Â (RRÀ1)/{prevalence of smoking Â (RRÀ1) þ 1}], where RR indicates pooled RRs. The national prevalence of past and current smokers among adults (!20 years of age) 25 was used to estimate the PAF. We used Stata version 13.1 (StataCorp, College Station, TX, USA) for all analyses.

Study selection
Our initial search identified 128 potential articles, of which 23 articles were considered potentially eligible based on the title and abstract screening (Fig.1). Another four articles were identified from NOTE: Weights are from random effects analysis   26 assessed the association with smoking intensity. Of two articles from another study, 23,27 one assessed the association with current and former smoking, 27 whereas the other 23 assessed the association with smoking cessation years. We found only one article investigating the association between passive smoking and risk of diabetes. 28

Study characteristics
A total of 19 independent prospective cohort studies including 343,573 individuals and 16,383 incident cases were identified (Table 1). 9,10,24,27e42 The selected studies were published between 1997 and 2015. The number of subjects per study ranged from 1266 to 128,141. The average follow-up duration ranged from 4 to 16 years. Diabetes was ascertained using biological screening in all studies except one (patient report). 24 Smoking status was selfreported in all studies. Six studies involved men only, 35e37, 39,41,42 and the remaining studies involved both men and women. Regarding cutoff points for diabetes, 11 studies used a FPG threshold of !126 mg/dL, 10,28,30,32e34,36e39,42 three studies used FPG !126 mg/dL or HbA1c ! 6.5%, 9,27,31 one study used FPG !140 mg/dL, 35 one study used FPG !110 mg/dL, 40 one study used FPG !126 mg/dL or 2-h post-load glucose !200 mg/dL, 29 and one study used HbA1c ! 6.1%. 41 Most studies adjusted for age (19 studies), body mass index (18 studies), alcohol consumption (13 studies), physical activity (11 studies), and heredity (9 studies), whereas fewer were adjusted for education (1 study), diet (1 study), and waist circumference (1 study) (eTable 1). The characteristics of studies on the number of years of smoking cessation 9,24,27 are shown in eTable 2. All the studies included in the present metaanalysis were generally of high quality (eTable 3).

Stratified analysis
To examine sources of heterogeneity on the association between active smoking and T2DM, we conducted stratified analysis across a number of key study characteristics ( Table 2). An increased risk of diabetes in current smokers was found in most of the subgroups. A slightly stronger association between smoking and T2DM was found in studies that adjusted for more than eight confounding factors, with a mean follow-up period of 10 years, and when diabetes was diagnosed using both FPG and HbA1c. However, we found no significant differences across strata (the P values for metaregression were >0.05 for all).

Publication bias
Visual inspection of a funnel plot indicated asymmetry in studies related to T2DM, which raises the possibility of publication bias (eFigure 1A). We then performed sensitivity analysis using a trim-and-fill method (eFigure 1B), which hypothetically imputes six negative and unpublished studies that were missing from the initial analysis. The imputed studies produced a symmetrical funnel plot. The pooled analysis including the six hypothetical studies also showed a statistically significant association between active smoking and T2DM (RR 1.27; 95% CI, 1.16e1.38).

Passive smoking and risk of diabetes
Exposure to passive smoking, excluding active smoking, was associated with an increased risk of diabetes compared with no current exposure to passive smoking, 28 although the association was not statistically significant, with a RR of 1.20 (95% CI, 0.54e2.68) (eTable 4).

Discussion
In this systematic review and meta-analysis involving 343,573 subjects and 16,383 patients with T2DM from 19 prospective cohort studies in Japan, we found an increased risk of T2DM in smokers compared with nonsmokers. An increased risk associated with current smoking was observed in most of the subgroups. There was a dose-response relationship among smokers, with the risk of T2DM being increased by 16% for each increment of 10 cigarettes smoked per day. The risk of T2DM remains high among those who quit during the preceding 5 years but decreased steadily with increasing duration of cessation, reaching a risk level comparable to that of never smokers after 10 years of smoking cessation.
We found a significant 38% higher risk of T2DM for current smoking compared with nonsmoking among the Japanese. The magnitude of this association is consistent with three metaanalyses conducted so far on active smoking and T2DM worldwide. 1,2,43 In previous meta-analyses, Pan et al found a 35% higher risk, 2 the United States Surgeon General's report found a 37% higher risk, 1 and Willi et al found a 44% higher risk 43 of T2DM in current smokers compared with current nonsmokers. The present metaanalysis of Japanese studies additionally included two large studies (1 recent study 9 and another study written in Japanese 10 ) that were not included in previous meta-analyses. We found a dose-response relationship between smoking and T2DM, a finding that is also consistent with previous studies. 1,2,43 The association between smoking and T2DM is biologically plausible. Smoking leads to insulin resistance or inadequate compensatory insulin secretion 44,45 through various underlying effects, including oxidative stress, inflammation, and endothelial dysfunction. 46,47 Nicotine in cigarettes may also exert a direct toxic effect on beta-cell function. 48 In addition, although smoking tends to decrease weight, it leads to central adiposity, 49,50 which has been linked to inflammation 51 and insulin resistance. 52 Compared with nonsmoking, former smoking was associated with a 19% higher risk of T2DM in the present study. This estimate is similar to those of two other recent meta-analyses, 1,2 where former smoking was associated with a 14% higher risk of diabetes. In a meta-analysis, Pan et al reported that former smoking was associated with a 16% higher risk of diabetes in Asians and a 15% higher risk of diabetes in East Asians. 2 That study included data of 13 Japanese studies, but our study included two additional studies. 9,10 When we examined the association between the duration of quitting smoking and T2DM, we found a significantly increased risk of T2DM for the first 5 years of smoking cessation compared with nonsmoking, although this increase did not exceed the risk of T2DM among current smokers. The risk of T2DM decreased steadily with increasing duration of cessation, reaching a risk level comparable to that of never smokers after 10 years of smoking cessation. Consistent with our finding, two recent meta-analyses reported that smoking cessation was associated with a substantial decrease in diabetes risk in the long term. 2,53 Taken together, Cigarette smoking per day Fig. 4. Linear dose-response relationship between cigarette smokes per day and relative risk of diabetes among total subjects (P for non-linearity ¼ 0.08). Data were modeled with random-effects restricted cubic spline models with three knots placed at 10th, 50th, and 90th percentiles of cigarette smokes per day. Lines with long dashes represent the pointwise 95% confidence intervals for the fitted linear trend (solid line). Line with short dashes represents the linear trend. Fig. 5. Relationship between duration of smoking cessation and relative risk of diabetes. Data were pooled using random-effects meta-analysis from three studies that presented data for duration of smoking cessation. Error bars show 95% confidence intervals.
although the risk of T2DM remains high after short-term smoking cessation, it decreases eventually in the long run. As smoking cessation usually leads to weight gain, 54 a concern has been raised about the possibility of increased risk of T2DM after quitting smoking. In fact, mechanistic studies showed deterioration of insulin sensitivity and lipid profiles after smoking cessation. 55,56 Some studies in Japan, 24 Korea, 57 and the United States 58,59 have reported a sizable increase (>15%) in the risk of diabetes among new quitters compared with current smokers. However, we did not find any further increase in risk after smoking cessation. Given the limited number of studies with conflicting data, further studies are required to elucidate whether short-term smoking cessation could lead to an increased risk of T2DM after quitting smoking.
So far, four meta-analyses based on worldwide data reported that passive smoking is associated with a 21e28% higher risk of T2DM. 2,11e13 We found only one prospective study 28 on the association between passive smoking and T2DM in Japan, which reported a nonsignificant 20% higher risk of T2DM associated with passive smoking (eTable 4). In that study, exposure to passive smoke in the workplace was associated with an increased risk of diabetes (hazard ratio 1.81; 95% CI, 1.06e3.08). In Japan, the Health Promotion Act was enacted in 2003 to restrict exposure to secondhand smoke in the workplace. 60 According to this act, smoking is banned in public spaces, relegating smokers to designated areas. The law was strengthened in 2010. 60 Accordingly, passive smoking is expected to decrease in Japan, so the impact of passive smoking on diabetes might have been decreasing.
The prevalence of smoking has been declining in Japan; however, smoking remains a public health threat, as 20.7% of adults smoke. 25 We estimated that 12.0% of T2DM cases in men and 3.6% in women were attributable to current smoking in Japan. For former smoking, these figures were 6.8% and 1.8% in men and women, respectively. A recent meta-analysis reported that worldwide, 11.7% of T2DM cases in men and 2.4% in women were attributable to current smoking, 2 but the PAF for former smoking was not reported. The present findings suggest that active and former smoking jointly accounted for nearly one-fifth of T2DM cases in Japanese men (18.8%). Smoking prevention should be encouraged and effective smoking control programs should be implemented to reduce smoking-related diabetes in Japan.
The strengths of our study include being based on high-quality cohort studies with large sample sizes and the inclusion of all relevant studies among Japanese populations, covering recent welldesigned cohort studies. This enabled us to draw strong conclusions. Despite these strengths, some limitations of this metaanalysis must be considered. First, there was heterogeneity in the RRs across studies that might result from differences in participant characteristics and definitions of outcome measures. However, we conducted stratified analyses and found summary RRs consistently greater than 1 across a number of study-and participant-level characteristics. Second, funnel plot analysis showed some asymmetry among male smokers, suggesting the possibility of publication bias and missing of some gray literature. We used a trim-andfill method, which can capture all unpublished and gray literature, and obtained a similar result, suggesting that the association was not affected by unpublished negative studies. Third, the possibility of residual confounding and unmeasured factors cannot be ruled out in observational studies. Specifically, smoking is related to other unhealthy lifestyle factors, such as unhealthy diet, excessive alcohol use, physical inactivity, and comorbidities. 61 We confirmed, however, that there were no substantial differences in association between studies with adjustment of !8 confounding variables and those with adjustment of fewer variables.
In conclusion, the results of this meta-analysis suggest that current smoking is associated with an increased risk of T2DM in a dose-dependent manner among the Japanese. Although the risk of diabetes remains high after short-term smoking cessation, the risk decreases substantially in the long run. Tobacco smoking accounts for 18.8% of T2DM cases among men and 5.4% of T2DM cases among women in Japan. These findings greatly strengthen the evidence on the association between smoking and T2DM and provide an additional rationale for the intensified implementation of tobacco control programs, especially in countries, like Japan, where tobacco smoking is still prevalent.

Conflicts of interest
None declared.