Letters, reviews, meta-analyses, comments, animal studies, and ecological studies were excluded in the current systematic review and meta-analysis. Following our initial search, 19,278 articles were identified. After removing 1,293 duplicates, 17,985 reports remained for further assessment. After title and abstract careful checking and review, 17,968 articles were excluded and 17 publications remained for full-text assessment. Four studies were excluded due to the following reasons: two studies had reported lung cancer mortality (21, 22). In addition, one thesis (23) and one Mendelian study (24) were also excluded. Finally, a total of 12 prospective cohort studies (15–20, 25–30) and one pooled analysis (31) were included in this systematic review and meta-analysis. Figure 1 illustrates the study selection process.
A recent pooled analysis by Zhu et al. (31) included 17 cohort studies; however, 12 of them were unpublished data with no available full-texts. Therefore, we decided to analyze data once by including the study by Zhu et al. (31) and excluding the 5 studies (15, 18, 19, 25, 26) that overlapped with the Zhu et al., and once again by adding the 5 studies and excluding the study of Zhu et al. for better understanding of the association.
Narita et al. (15) had reported effect sizes separately for men and women, however, we combined these two effect sizes and then, included in our analysis. Three studies had not reported the 95% CIs for the association between coffee consumption and risk of lung cancer (28–30). Therefore, we derived relevant data for these studies from the previous meta-analysis (14).
Findings from the systematic review
Study characteristics
Overall, 12 cohort studies (15–20, 25–30) and one pooled analysis (31) were included in the present systematic review (Table 1). These studies were reported from 1986 to 2020; four were from the United States (16, 18, 25, 26), three from Norway (20, 28, 30), two from Japan (15, 29), one each from Thailand (17), Singapore (19), and Korea (27). The median follow-up duration ranged from 10 to 17.7 years. For the exposure assessment, 8 studies had used food frequency questionnaire (15, 16, 18–20, 26–28), 1 had collected data based on diet history questionnaire (25), and 1 had used a structured questionnaire (17). Others had reported using a questionnaire (30) and 24-h dietary recall history (29). For the outcome assessment, all, but two, of the included studies had used cancer registries. Outcome assessment in the PLCO study (25) was self-reported and Nomura et al. had used histologic examination (29). Based on the NOS, all included studies were of high quality.
Findings from the meta-analysis
First, we examined the association by including data from 7 cohort studies not included in the pooled analysis paper of Zhu et al. along with the findings from the pooled analysis. The overall effect size based on these 8 studies (16, 17, 20, 27–31) revealed a statistically significant association between coffee consumption and risk of lung cancer (RR: 1.39; 95% CI: 1.12, 1.73; Fig. 2).
We also found an evidence of statistically significant between-study heterogeneity (I2 = 81.9%, P < 0.001). No evidence of publication bias was seen (P = 0.84).
In a further analysis, we excluded the study of Zhu et al. and included 12 cohort studies in the analysis. Combining 12 effect sizes from 12 studies (15–20, 25–30), we observed a statistically significant positive association between coffee consumption and risk of lung cancer (RR: 1.29; 95% CI: 1.12, 1.50; Fig. 3).
However, a significant between-study heterogeneity was found (I2 = 74.2%, P < 0.001). A sensitivity analysis showed that no particular study had a significant influence on the summary effects. In addition, we observed no proof of significant publication bias using Egger’s test (P = 0.49).
To find sources of heterogeneity, we performed subgroup analyses based on fixed-effects model. In the subgroup analyses, we found that sex, follow-up duration, and country might explain between-study heterogeneity (Table 2).
Table 2
Subgroup analyses for the association between coffee consumption and risk of lung cancer
Variables | Effect sizes, n | I2, % | Q test | RR (95% CI) | P-between |
Sex | | | | | 0.004 |
Men | 2 | 0 | 0.68 | 1.67 (1.05, 2.68) | |
Women | 1 | - | - | 2.01 (1.47, 2.75) | |
Both | 9 | 74.4 | < 0.001 | 1.21 (1.13, 1.29) | |
Follow-up duration (year) | | | | | 0.01 |
< 15 | 6 | 75.6 | 0.001 | 1.33 (1.23, 1.45) | |
≥ 15 | 5 | 74.2 | 0.004 | 1.13 (1.01, 1.26) | |
Country | | | | | 0.006 |
USA | 4 | 36.2 | 0.19 | 1.17 (1.09, 1.27) | |
Non-USA | 8 | 77 | < 0.001 | 1.43 (1.27, 1.61) | |
Exposure assessment | | | | | 0.67 |
FFQ | 9 | 66.4 | 0.002 | 1.25 (1.17, 1.34) | |
Non-FFQ | 3 | 89.3 | < 0.001 | 1.18 (0.92, 1.52) | |
Outcome assessment | | | | | 0.13 |
Registries | 10 | 77 | < 0.001 | 1.27 (1.19, 1.37) | |
Non-registries | 2 | 15.6 | 0.27 | 1.12 (0.97, 1.31) | |
CI: confidence interval; RR: rate or risk ratio; FFQ: food frequency questionnaire.
A significant positive association between coffee consumption and risk of lung cancer was seen in men (RR: 1.67; 95% CI: 1.05, 2.68), and both sexes (RR: 1.21; 95% CI: 1.13, 1.29). In addition, we observed a significant positive association between coffee consumption and risk of lung cancer in studies with < 15-year duration of follow-up (RR: 1.33; 95% CI: 1.23, 1.45), as well as those with ≥ 15-year of follow-up (RR: 1.13; 95% CI: 1.01, 1.26), those conducted in USA (RR: 1.17; 95% CI: 1.09, 1.27), and those conducted in non-USA countries (RR: 1.43; 95% CI: 1.27, 1.61).