Aspirin Use and Lung Cancer Risk: A Possible Relationship? Evidence from an Updated Meta-Analysis

Background and Purpose Growing evidence has emerged and controversial results reported on possible relationship between aspirin use and lung cancer risk. We, therefore, conducted this updated and comprehensive meta-analysis to evaluate this issue, with focus on dose-risk and duration-risk relationships. Methods We searched electronic databases including PUBMED, EMBASE and Cochrane library to identify eligible studies. Relative risk (RR) and its 95% confidence interval (CI) were used for cohort studies, while odds ratio (OR) were employed for case-control studies. The random effects and fixed effects models were used for analyses. Results 18 studies were identified including 19835 lung cancer cases, which were eligible for inclusion in the present meta-analysis. Pooled data from case-control studies showed a significant inverse association between regular aspirin use and lung cancer risk. But for cohort studies, insignificant association was detected with little evidence of heterogeneity (RR: 1.05, 95%CI: 0.95 – 1.16; I2: 10.3%, p value: 0.351). In case-control studies, standard aspirin use (>325mg) was related to lower lung cancer incidence, compared with low-dose aspirin use (75–100mg). A similar trend was observed in cohort studies. Besides, when analysis was restricted to long time regular aspirin use (>5 years), insignificant results were reported in both cohort and case-control studies. Finally, regular aspirin use might result in higher reduction of non-small cell lung cancer incidence among men. Conclusions Our findings do not support the protective effect of regular aspirin use on lung cancer risk. Long time aspirin use, sex, dose and type of lung cancer might alter the effect of aspirin use on lung cancer risk. More well-designed studies are needed to further clarify these associations.


Results
18 studies were identified including 19835 lung cancer cases, which were eligible for inclusion in the present meta-analysis. Pooled data from case-control studies showed a significant inverse association between regular aspirin use and lung cancer risk. But for cohort studies, insignificant association was detected with little evidence of heterogeneity (RR: 1.05, 95%CI: 0.95 -1.16; I 2 : 10.3%, p value: 0.351). In case-control studies, standard aspirin use (>325mg) was related to lower lung cancer incidence, compared with low-dose aspirin use (75-100mg). A similar trend was observed in cohort studies. Besides, when analysis was restricted to long time regular aspirin use (>5 years), insignificant results were reported in both cohort and case-control studies. Finally, regular aspirin use might result in higher reduction of non-small cell lung cancer incidence among men.

Data Source and Search Strategy
PUBMED, EMBASE and Cochrane library were searched between January 1966 and December 2013 to identify eligible studies, using the following key words: 'aspirin or acetylsalicylic acid or non-steroidal anti-inflammatory agent or NSAID', 'lung cancer' and 'risk'. Besides, the reference list of the articles and reviews retrieved were manually searched to identify additional eligible studies.

Criteria for Inclusion and Exclusion
Studies were eligible for inclusion if they: 1) were case-control or cohort studies, or RCT; 2) evaluated the influence of aspirin use on lung cancer risk separately from other NSAIDs; 3) had explicit description of aspirin exposure and 4) provided relative ratios (RRs) or odds ratios (ORs) with 95%confidence intervals (CIs) or sufficient information to calculate them. Review articles, case reports, letters to the editor and editor comments were excluded. If data were duplicated in more than one study, the most recent or comprehensive report was selected for analysis.

Data Extraction
Data extraction was carried out independently by two investigators (Hai-yan Jiang and Tianbao Huang) according to the meta-analysis of observation studies in epidemiology guidelines [19], and discrepancies were adjudicated by Xu-dong Yao. The data extracted from each study included first author name, year of publication, gender, country, study design, type of controls, sample size, definition of aspirin exposure, RRs or ORs and their 95%CIs and adjusted factors. Data on standard and low-dose regular aspirin use and relevant incidence of lung cancer were also extracted. Moreover, if data were available, lung cancer cases were subdivided into small cell lung cancer (SCLC) and non-small cell lung cancer (NSCLC). When multiple estimates were reported, the one adjusted for the most variables was selected.

Statistical Analysis
RRs and their 95% CIs were used to assess the data extracted from cohort studies, while odds ratios (ORs) were employed for case-control studies. Considering the recall bias and selective bias of case-control studies, we did not pool data from RCT, cohort studies and case-control trials. When data of different durations of use or different intake levels were available, we subsequently restricted the analyses to regular standard aspirin use or low dose aspirin use with longest duration. Regular standard aspirin use was defined as "regular aspirin use with dose !325mg", while low-dose aspirin use was considered for "regular aspirin use with dose between 75 to 100mg". Long time regular aspirin use was defined as "regular aspirin use for more than 5 years, regardless of the dose". The statistical heterogeneity among studies was evaluated by using the Cochrane's Q and I 2 statistics. As for Q statistics, heterogeneity was considered for P<0.05. When P>0.05 and I 2 <50%, the studies were considered with acceptable heterogeneity, and the fixed effects model was used. Otherwise, the random effects model was used.
Subgroup analyses based on gender (female only vs. male only vs. both sex) and lung cancer histology (NSCLC vs. SCLC, adenocarcinoma vs. squamous cell carcinoma) were carried out to detect the source of heterogeneity. Smoking is well known as one of the most important risk factors for lung cancer. Therefore, sensitivity analyses, excluding studies which did not adjust for smoking, were conducted to minimize the influence of smoking on lung cancer risk.
Finally, potential publication bias was evaluated graphically with funnel plots of log risk ratio against the standard error of the study. If the funnel plot was asymmetrical, the rank correlation method proposed by Begg et al. and linear regression approach suggested by Egger et al. were used to evaluate the potential publication bias. At P < 0.05, sensitivity analyses were conducted to explore whether the final effects were strongly influenced by individual studies. All statistical analyses were performed by using the STATA Statistical Software version 11.1 (Stata Corp., College Station, Texas, USA), and two-tailed P values were determined.
Any aspirin use was associated with a decreased risk of lung cancer in case-control studies with a large evidence of heterogeneity (pooled OR: 0.71, 95%CI: 0.56-0.91; I 2 : 82.3%). However, no association was detected in cohort studies (pooled RR: 0.97, 95%CI: 0.81-1.15). There was also some evidence of heterogeneity, which could be subsided by excluding the two studies [29,30] that did not adjust for smoking (I 2 = 55.2%, I 2 = 10.3%, respectively) (detailed estimated data are shown in Table 2 and Table 3). As for regular standard aspirin use, similar insignificant results were obtained in cohort studies (pooled RR: 0.97, 95%CI: 0.82-1.15) (Fig 2). However, in case-control studies, a stronger inverse association was detected (pooled OR: 0.61, 95%CI: 0.42-0.89). For low-dose regular aspirin use, both groups did not yield any significant associations, which confirmed the data abstracted from a post-trial of RCT [12]. In this largescale study, 33682 women were included; after over ten years of follow-up, no significant association was found between the 100mg alternate-day aspirin use group and placebo patients (HR: 1.04, 95%CI: 0.86-1.26). Some of the included studies [7,8,10,18,20,[22][23][24][25]28] reported Table 1. Characteristics of studies included in the meta-analysis of aspirin use and risk of lung cancer.

Author
Year the estimated data about long time aspirin use and lung cancer risk. The pooled estimates revealed similar results as regular aspirin use (Fig 3). Furthermore, we conducted a subgroup analysis based on gender and found that males may benefit more from aspirin use than females (pooled RR for men: 0.83, 95%CI: 0.48-1.41; pooled RR for females: 1.02, 95%CI: 0.70-1.47). Forest plot and meta-analysis of regular aspirin use and incidence of lung cancer, stratified by dose of usage. Standard aspirin use refers to "regular aspirin use with dose !325mg", while low-dose aspirin use is considered for "regular aspirin use with dose between 75 to 100mg". The solid diamonds and horizontal lines correspond to the study-specific estimated risks and 95% CIs. Besides, the hollow diamonds represent the pooled relative risk and 95% CIs. Abbreviation: RR: relative risk, CI: confidence interval.
doi:10.1371/journal.pone.0122962.g002 Forest plot and meta-analysis of long time regular aspirin use and lung cancer risk among cohort studies, stratified by study design. ES refers to relative risk among cohort studies and odds ratio among case-control studies. Long time regular aspirin use was defined as "regular aspirin use for more than 5 years, regardless of the dose". The solid diamonds and horizontal lines correspond to the study-specific estimated risks and 95% CIs. Besides, the hollow diamonds represent the pooled relative risk and 95% CIs. Abbreviation: CI: confidence interval. doi:10.1371/journal.pone.0122962.g003 However, statistical significance level was not reached. After stratification by pathological type, we found that aspirin might protect more people suffering from NSCLC (pooled RR: 0.67, 95% CI: 0.25-1.86). Finally, obvious publication bias was found by Egger's (P = 0.015) and Begg's (P = 0.028) tests (Fig 4). Subsequently, sensitivity analyses indicated that no single study influenced the pooled RR qualitatively.

Discussion
The findings of the present meta-analysis showed that there is no explicit association between aspirin use and lung cancer risk among cohort studies, which was consistent with previous meta-analyses [15][16][17]. Among the previous meta-analyses, Bosetti C et.al [17] aimed at extensively explaining the association between aspirin use and risk of 12 selected cancer, without further exploring the influence of aspirin use on lung cancer risk. Besides, another two studies [15,16] also tried to investigate the possible relations, taking the dose-risk and duration-risk relationships into consideration. But neither of them presented meta-analyses stratified by study design throughout the manuscript. In this meta-analysis, we found that statistical significance was mostly contributed by case-control studies. A spurious protective effect may be presented due to its nature of selection and recall bias. Four of seven case-control studies in our meta-analysis were hospital-based, which might more easily introduce bias. It is known that cohort studies are less susceptible to bias due to their prospective design. So the results summarized in cohort studies are more credible and stable. To minimize the influence of study design, we performed this meta-analysis separately assessing cohort-and case-control studies. Besides, all these studies were based on electronic database before September 2011. After that, several studies [12,18] were published which, in some extend, might provide more evidence on these associations. Thus we performed this meta-analysis to further understand whether the aspirin use could influence the risk of lung cancer. When we limited our analysis to long-term regular aspirin use, a strong inverse association was reported in case-control studies (pooled OR: 0.69, 95%CI: 0.49-0.98), but not for cohort studies (pooled RR: 0.98, 95%CI: 0.86-1.11). Thorat et al. [31] claimed that growing evidence supports aspirin's effect in reducing cancer incidence and mortality, but the duration of use needs to be at least 5 years. In our analysis, after we subdivided regular aspirin use into standard and low-doses, an insignificant difference was obtained between these two subgroups among cohort studies (pooled RR = 0.97, pooled RR = 0.95, respectively). But in case-control studies, slightly stronger insignificant inverse association was detected among individuals with standard aspirin use compared with those taking low-dose aspirin (pooled RR = 0.62, pooled RR = 0.70, respectively). A recently published population-based cohort study [14]indicated that low-dose aspirin in the year prior to diagnosis was associated with lower tumor extent and fewer metastatic diseases in lung cancer, suggesting a protective effect of low-dose aspirin use on lung cancer. Taken together, these data suggest that long-term aspirin use might reduce the incidence of lung cancer; however, the appropriate dose needed is still unclear.
Smoking is considered the most well-known environmental risk factor for lung cancer [32]. Almost all the included studies were adjusted for smoking status or pack-years except two reports [29]. After exclusion of both studies, the estimated data only slightly changed, maybe because actual pack-years and smoking status are hard to adjust. However, it was impossible to conduct an analysis on aspirin use and risk of lung cancer stratified by smoking pack-years/status due to the different measurements used in various studies. Three of the eight studies [18,23,24] demonstrated that aspirin use was associated with a reduced risk of lung cancer among smokers. Another study [26]reported that estimated RR in male smokers was higher than in non-smokers (RR = 1.08, RR = 0.96, respectively), but the difference was not statistically significant.
In cohort studies, a stronger insignificantly inverse association between aspirin use and lung cancer was reported in male populations than in females (pooled RRs of 0.83 and 1.02 for men and women, respectively). A research conducted by Chen et al. [33]found that the levels of serum E2, prolactin and follicle-stimulating hormone were remarkably increased in male patients with lung cancer, suggesting that sexual hormones may play a role in lung cancer genesis. Several other studies suggested that estrogen receptors may influence the effect of aspirin use on lung cancer [16,34]. Further studies are needed to clarify the influence of sex on the association between aspirin use and lung cancer risk.
Several limitations of the present meta-analysis should be mentioned. Firstly, approximately half of the included studies were case-control trials. Although they were not combined in this meta-analysis, there are still several potential confounding factors, which may influence the conclusion obtained from these studies. Secondly, the included studies were different in terms of populations, doses, durations of aspirin use, selection of controls and confounders adjusted, all of which may affect the possible association between aspirin use and lung cancer risk. Subsequently, subgroup-and sensitivity analyses were performed. However, these analyses were far from removing all heterogeneity. Thirdly, our literature search was restricted to published studies. It is known that negative studies are less likely to be published in indexed journals, which may bias our results. Finally, due to lack of relevant data, it was impossible to clarify the dose-response association. Therefore, it was hard to quantitatively assess the aspirin use on lung cancer risk. Taken together, prospective studies with different dose of aspirin use, especially RCTs, were needed to solve this question.

Conclusion
In summary, our meta-analysis indicated that regular/any aspirin use might have no association with lung cancer risk based on findings of cohort studies. Long time aspirin use, sex, dose and type of lung cancer might alter the effect of aspirin use on lung cancer risk. Further welldesigned large-scale studies are needed to confirm our findings.