3.1 Literature selection
A total of 592 studies were included in our study, including the results of keywords search (n=586) and additional records identified from reference lists of other studies (n=6). After the detailed reading of full texts, five articles[29-33] reported the relationship between smoking and incidence of insomnia were potentially eligible for meta-analysis (Figure 1) and one other article was eligible for further discussion as it divided insomnia into different symptoms. The main information of eligible studies was given in Table 1. We assessed the quality of included studies from three aspects and their scores were no less than seven points (total score=seven points) (Table 1 and Supplementary Table 1).
3.2 Risk of bias assessment for included studies
3.2.1 Bias from character of population
Cohorts of four included studies [29-31, 33] were representative of general community population, whereas other two studies [32, 34] restricted their cohorts to specific population, such as adult employers. One study  didn’t provide gender information. The diversity of included population means potential different educational level, socioeconomic status and mental health level, which might contribute to the bias of synthesized results.
3.2.2 Bias from covariates adjustment
Different studies provided different information of covariates adjustment. All five studies [29-34] adjusted the results for age and two studies [30, 33] adjusted the result for gender. One study  further adjusted the result for baseline insomnia symptoms, marital status, occupational class, sleep duration and common mental disorders, one study  further adjusted for parental education, family income, and parent-reported poor health condition at baseline and one study  further adjusted for medical disorder, BMI, physical inactivity, alcohol. One study  adjusted for sampling weight.
3.2.3 Bias from definition of endpoint
Although included studies [29-34] all provided clear definition of insomnia, the definition varied from studies which might contribute to the bias of synthesized results.
3.3 Smoking and Insomnia Incidence
Figure 2 showed the ORs of five studies about smoking and incidence of insomnia. The pooled OR for smoking was 1.07 (95% CI:1.02, 1.13), which indicated a significant association between smoking and incidence of insomnia. The heterogeneity was not significant (I2=14.9%, p=0.310). Funnel plot and Egger's regression revealed that there was no significant evidence for publication bias (Egger's regression: p>0.1, Supplementary figure 1). Sensitivity analysis indicated heterogeneity was mainly from one study  (Figure 1 and Supplementary figure 2). After reviewing the whole article, we noticed that in this study, participants which had a poor quality of sleep while didn’t meet the criterion of insomnia at the baseline were also included in the analysis while in other included studies, this part of participants was excluded. Besides, PSG (polysomnography) was used to diagnose insomnia in this study, which might find out potential insomnia that was neglected in other included studies. All these factors might cause the heterogeneity.
Subgroup-analysis by smoking status showed significant association between regular smoking and incidence of insomnia (OR=1.07, 95% CI:1.01, 1.13). As for occasional and ex-smokers, the pooled analysis didn’t indicate a significant association (occasional smoker: OR=2.09, 95% CI:0.44, 9.95; ex-smoker: OR=1.02, 95% CI:0.67, 1.54) (Figure 2).
Subgroup-analysis by age suggested significant association both in adults (>18) (OR=1.06, 95%CI: 1.02,1.13) and juveniles (≤18) (OR=3.97, 95%CI: 1.04,15.18) (Table 2 and Figure 1). The subgroup-analysis by gender ratio suggested significant association in population with gender ratio (women/men) <1 (OR=1.06, 95%CI: 1.01,1.13), but not in population with gender ratio (women/men) ≥1 (OR=1.08, 95%CI: 0.93,1.25) (Table 2 and Figure 2). However, the pooled result in population with gender ratio (women/men) ≥1 showed a high heterogeneity (I2=72.8%, p=0.055). Because there were only two studies in this subgroup and they reported different results, we failed to exclude any one of them to reduce heterogeneity. The region-specific analysis suggested a positive association in North America (HR: 1.06, 95% CI: 1.01, 1.13) and Asia (only one study) (OR=3.97, 95%CI: 1.04,15.18). The association between smoking and incidence of insomnia was not significant in Europe (OR=1.09, 95% CI: 0.96, 1.25) (Table 2).
Wetler et al (1994)  suggested that current smoking was related to very frequent (OR=2.32, 95%CI:1.04,5.16), frequent (OR=1.56, 95%CI:1.02,2.38), and infrequent difficulty getting to sleep (OR=1.44, 95%CI:1.05,1.98) for males. For females, current smoking was associated with very frequent (OR=1.88, 95%CI:1.09,3.23), frequent (OR=1.78, 95%CI:1.26,2.52) difficulty getting to sleep. Besides, current smoking was associated with very frequent nonrestorative sleep (OR=2.20, 95%CI:1.15,4.19) for males and very frequent (OR=2.61, 95%CI:1.62,4.19), frequent (OR=1.73, 95%CI:1.24,2.40), and infrequent (OR=1.81, 95%CI:1.37,2.41) nonrestorative sleep for females. Current smoking was associated with excessive daytime sleepiness for females (very frequent: OR=2.61, 95%CI:1.43,4.78, frequent: OR=1.48, 95%CI:1.04,2.09, and infrequent: OR=1.47, 95%CI:1.12,1.91) but not for males. (Supplementary Table 2)