Associations between aspirin use and the risk of cancers: a meta-analysis of observational studies

1Aspirin has been used as an analgesic and in the prevention of cardiovascular diseases events in the past decades and is one of the most commonly used drugs worldwide. Clinical and epidemiological studies reported that the rates of aspirin usage in different populations across different countries ranging from 11% to 54%. Since the 1970s, many researchers started to focus on the effects of aspirin on cancers. However, these original studies were not comprehensive, and the effects on some cancers were controversial.

2Although several meta-analyses have been conducted to assess the associations between aspirin use and the risk of cancers(e.g., gastric, esophageal, pancreatic, lung, squamous cell carcinoma, breast, ovarian, and prostate cancers), most of these studies were restricted to certain types of cancers, and some types such as hepatobiliary and cervical cancer could not be investigated. In addition, 70 new studies have been published since 2012. Therefore, this comprehensive systematic review and updated meta-analysis was conducted to explore the reliability of risk estimates between aspirin usage and most types of cancers and provide a landscape of aspirin use and cancer incidence.

3This systematic review was conducted in accordance with the checklist proposed by the Meta-analysis of Observational Studies in Epidemiology group. We searched multiple electronic bibliographic databases to identify studies published from database inception till March 2017, including PubMed, Embase, and Web of Science databases, with the following search terms: (“cancer” OR “neoplasm” OR “carcinoma”) AND (“aspirin” OR “acetylsalicylic acid” OR “non-steroidal anti-inflammatory drugs” OR “NSAIDs”). We restricted our search to human studies and published in English. In addition, reference lists from relevant reviews and retrieved articles were searched for qualifying studies.

4The inclusion criteria were: 1) case-control or cohort studies; 2) studies that evaluated the relationships between the use of aspirin and the risk of cancers; 3) studies that reported risk estimates with 95% confidence interval (CI) or provided information that enabled us to calculate them. The exclusion criteria were: 1) studies that used other combinations of NSAIDs, which prevented the determination of the specific effect of aspirin, and 2) studies involving patients with specific diseases (e.g., Barrett’s esophagus, Crohn’s disease, or ulcerative colitis). Only the latest or the most informative study was included when multiple studies were published on the same study population.

5The following information was obtained from each study: first author name, year of publication, study period, study location, study design, number of cases, number of participants, gender, definition of aspirin exposure, as curtained methods of exposure, odds ratios (ORs), hazard ratios (HRs) or relative risks (RRs) with their corresponding 95% CIs, and confounding factors adjusted in the analysis. The most fully-adjusted risk estimates with its corresponding 95% CIs (when available) were preferentially extracted. Data extraction was conducted independently by two authors (Y.Q. and T.T.Y.), and discrepancies were resolved by discussion with a third investigator (Z.X.L.).

6Quality assessment of eligible studies was performed independently by two reviewers (Y.Q. and T.T.Y.) according to the Newcastle-Ottawa Quality Assessment Scale. This scale allocates a maximum of nine points based on the selection (0–4 points), comparability (0–2 points), and exposure/outcome of the study participants (0–3 points). Scores of 0–3, 4–6, and 7–9 were classified as low, moderate, and high-quality studies respectively.

7RRs were used as the common measurement of the associations between aspirin use and the risk of cancer. Because cancer is a rare event in general, we could generally ignore the distinctions among the various measures of relative risk (e.g., odds ratios, rate ratios, and risk ratios), and considered that ORs and HRs were similar to RRs. When risk estimates for different durations of aspirin use or different levels of aspirin utilization were available, the study-specific RRs were subsequently recalculated in the primary analysis by pooling the risk estimates compared with the reference group. A random effects model was selected to estimate the pooled RRs (95% CI) for the associations between aspirin use and the risk of cancer if the risk estimates for different subtypes of cancer were available. Summary estimates were derived from meta-analyses using random effects models. Studies involving different populations or different types of cancers were treated as independent studies.

8To assess the heterogeneity in results of individual studies, I2 statistic (values of 25%, 50%, and 75% represented cutoff points for low, moderate, and high degrees of heterogeneity, respectively) were used. Publication bias was assessed with Funnel plots, the Begg’s rank correlations and Egger’s regression model. Subgroup analyses for study design, study location, gender, exposure assessment, quality assessment, duration of aspirin use (years), and frequency of aspirin use (tablets/week) were conducted to explore the potential heterogeneity among studies. Subgroup analysis was not conducted for strata with less than five studies. Because time-related biases are common in observational studies of medications and are often responsible for apparent protective effects of drugs, we conducted analyses both including and excluding studies with immortal time bias (bias because of the inclusion of follow-up time during which events cannot occur). Statistical analyses were performed with Stata version 12.0. (College Station, TX, USA). All reported probabilities (P values) were two-tailed with a significance level of 0.05.

9Figure 1 shows the process for the identification of eligible studies. A total of 28,683 studies were identified and 298 studies remained in the analysis after assessing the titles and abstracts according to the criteria mentioned above. In total, 307 potentially relevant articles were reviewed in their entirety. Among them, 89 articles were further excluded due to the following reasons: 26 articles were not observational design, 11 articles defined exposure combined with other NSAIDs, 8 articles evaluated cancer mortality, 39 articles were duplicate publications on the same subject population, and 5 articles (1 for Crohn’s disease, 1 for ulcerative colitis, 3 for Barrett’s esophagus) included patients with specific diseases. Ultimately, 218 studies with 309 independent reports were included in the present meta-analysis.

10The main characteristics of the 218 eligible articles published between 1985 and 2016 are summarized in Tables 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20 and 21. Results were presented according to study design. This study altogether included 161 cohort studies and 148 case-control studies. Among them, 135 studies were conducted in North America, 12 in Asia, 61 in Europe, 8 in Oceania, and 2 were multi-country studies. Overall, the summarized RR was 0.89(95%CI: 0.87–0.91), indicating a decreased risk of cancer associated with the use of aspirin. The combined RRs were 0.82 (95% CI: 0.79–0.85) for the case-control studies and 0.94 (95% CI: 0.92–0.97) for the cohort studies. We also observed a apparent beneficial effect of aspirin use when excluding 41 studies deemed to be prone to immortal time bias (RR = 0.87, 95%CI:0.85–0.89) in the meta-analysis.

11Figures 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17 and 18 and Additional file 1: Table S1 shows the RRs for the 21 separate cancer sites that we assessed and that of the total cancers. The use of aspirin was associated with a reduced cancer risk for ten specific sites: gastric cancer (RR =0.75, 95%CI:0.65–0.86), esophagus cancer (RR = 0.75, 95%CI:0.62–0.89), colorectal cancer(RR = 0.79, 95%CI:0.74–0.85), pancreatic cancer (RR = 0.80, 95%CI:0.68–0.93), breast cancer (RR = 0.92, 95%CI:0.88–0.96), ovarian cancer (RR = 0.89, 95%CI:0.83–0.95), endometrial cancer (RR = 0.92, 95%CI:0.85–0.99), prostate cancer (RR = 0.94, 95%CI:0.90–0.99), and small intestine neuroendocrine tumors (RR = 0.17, 95%CI:0.05–0.58). However, there was no significant association between aspirin use and the risk of some cancers, including hepato-biliary, lung, cervical uterus, renal, renal pelvis and ureter, bladder, brain, head and neck, thyroid, and skin cancers, as well as lymphoma and leukemia.

12Additional file 1: Tables S1–S18 shows the RRs for cancers at 17 sites, in subgroups of studies defined by their design, study location, gender, exposure assessment, quality assessment, duration of aspirin use, and frequency of aspirin use.

13We conducted a subgroup analysis stratified by questionnaires and medical records, and found a lower risk in medical records with most cancers (gastric, esophageal, colorectal, hepato-biliary, and pancreatic cancers), however, significant heterogeneity of effects was noted for those subgroups (Additional file 1: Tables S2–S18). As we expected, the decreased risk of colorectal cancer (RRs = 0.76, 95%CI: 0.66–0.87 for ≥5 years), pancreatic cancer (RRs = 0.75, 95%CI: 0.57–0.99 for ≥5 years), ovarian cancer (RRs = 0.77, 95%CI: 0.63–0.93 for ≥5 years), and brain cancer (RRs = 0.65, 95%CI: 0.43–0.97 for ≥5 years) were more pronounced with longer duration of aspirin use. However, the aspirin-associated RR for 21 specific cancers did not vary significantly by other characteristics (gender, quality assessment and frequency of aspirin use).

14The funnel plot showed asymmetry (Fig. 19). In addition, the Begg’s test and Egger’s test provided evidence of publication bias among the included studies (Begg’s test Z = 4.34, P < 0.001; Egger’s test Z = − 5.27, P < 0.001).

15The results of our meta-analysis supported the presence of inverse associations between aspirin use and the risk of overall cancer, gastric, esophageal, colorectal, pancreatic, breast, ovarian, endometrial, and prostate cancers, as well as small intestine neuroendocrine tumors. However, no significant associations were observed between the use of aspirin and the risk of other cancers, including hepato-biliary, lung, cervical uterus, renal, renal pelvis and ureter, bladder, brain, head and neck, thyroid, and skin cancers, as well as lymphoma, and leukemia.

16There are several potential biological mechanisms through which aspirin could reduce the risk of cancer. First, aspirin and other NSAIDs have been proven to inhibit the activity of the enzyme cyclooxygenase 2 (COX-2), which is responsible for the synthesis of prostaglandins. COX-2 has been reported to be overexpressed in many cancers and participates in key cellular activities, including cell proliferation, apoptosis, angiogenesis, and metastasis. Second, aspirin could activate the NF-kappa B (NF-κB) signaling pathway, which triggers apoptosis in neoplasia. In addition, some studies showed that aspirin might induce gene selection and modulate mitochondrial voltage dependent anion channels (VDACs) to reduce the risk of cancer progression and metastasis.

17The results of this meta-analysis indicated that utilization of aspirin had different protective effects on the development of cancer. This difference may be attributed to the different expression levels of COX in various cancers. Furthermore, Zumwalt et al. reported that the effectiveness of aspirin was primarily determined by specific genetic variants. Aspirin inhibited cell growth in all cancer cell lines regardless of mutational background, however, the effects were exacerbated in cells with PIK3CA mutations, which might explain the different effects of aspirin on cancers.

18The decreased risk of gastric, esophageal, pancreatic, lung, breast, and ovarian cancers was observed in the case-control studies but not in the cohort studies. One possible explanation for the difference might be that cases in the case-control studies might have a recall bias and tended to overestimate the risk of cancer by aspirin use. Another possible explanation is that misclassification or measurement errors for aspirin use in the cohort studies might have distorted the association because most of our analyses were based on baseline data, and there might be a discrepancy between initial recruitment and subsequent aspirin consumption.

19The longer those who had used aspirin, the lower their risk of cancer was, with longer duration of use associated with an RR of 0.90 (95% CI 0.89–0.74), based on 118 studies that reported associations with longer (≥5 years) duration of aspirin use and 105 studies that reported associations with shorter (< 5 years) duration of aspirin use. For most cancers (colorectal, pancreatic, ovarian, and brain cancers), risk reductions were more pronounced with longer duration of use, and these results agree with those of previous studies. In addition, the United States Preventive Services Task Force (USPSTF) indicated that cancer prevention was a significant aspect in the overall health benefit of aspirin, but this benefit was not apparent until several years after the initiation of aspirin therapy. It is of note that a significant inverse association with prostate cancer was observed in the patients who took aspirin for less than 5 years. Indeed, after the study that relied on the General Practice Research Database was excluded, the discrepancy disappeared. Considering that aspirin use was off-prescription in the United Kingdom, misclassification was likely to occur in this study because many commonly used aspirins do not require a prescription. Therefore, it can be deduced that the patients who used aspirin for at least 5 years were more likely to realize the potential cancer prevention benefit.

20There was no statistically significant difference between the pooled RRs for the frequency of aspirin in most studies. Given that a few studies were included in the subgroup analysis on the basis of the frequency of aspirin use and most studies lacked information on this variable, the results on the risks associated with the frequency of aspirin use should be interpreted with caution. Further studies that explore the associations between the frequency of aspirin use and cancer risk are necessary to elucidate the effects of aspirin.

21In addition, our results indicated that the strongest reduction in the risk of most cancers associated with aspirin was found in North American countries. However, two-thirds of the included studies were performed in North America and a few studies were performed in Asian and European countries, which might distort the accuracy of the results. Therefore, more studies are necessary to examine the discrepancies among the different countries and regions.

22Bosetti et al. (2011) conducted a meta-analysis on aspirin and 12 selected cancer sites based on 139 observational studies and 187,167 cases. Our study included 218 studies involving 737,409 cases and examined the correlation between aspirin use and the risk of skin, head and neck, hepatobiliary, thyroid, cervical uterus, renal pelvis, ureter, and brain cancers, lymphoma, small intestine neuroendocrine tumors, and leukemia, thereby providing more comprehensive and reliable evidence for this correlation. More importantly, this study was the first meta-analysis to evaluate the association between aspirin use and the risk of hepatobiliary cancer and we found a non-significant effect of aspirin on the risk of hepatobiliary cancer (OR = 0.64, 95% CI: 0.40–1.02).

23Algra and Rothwell (2012) conducted a meta-analysis on the association between aspirin use and the risk of cancer based on 195 studies and 215,211 cases. Compared with their review, our meta-analysis have added approximately 70 new articles published since 2012, with a total of 737,409 cases, which significantly enhanced the statistical power to determine this potential association. In addition, the exposure in the previous review was inconsistent, which may mislead the estimation. Many studies defined aspirin as the exposure but only a few studies defined NSAIDs as the exposure, and thus the specific effect of aspirin on cancers was not defined. The exposure to aspirin in our meta-analysis was consistent and ensured the reliability of the findings.

24This study is the most up-to-date comprehensive review of the effect of aspirin use on the risk of all types of cancers, and the large sample size provides reliable results with greater precision and power. The potential limitations of this study should be noted. First, there was substantial heterogeneity across the included studies, which was likely due to differences in the definitions of exposure, units, assessment methods, and the adjusted variables across different studies. Second, misclassification or measurement errors for aspirin use might distort the association because our analyses were based on baseline data, and changes in the exposure to aspirin were not updated during the follow-up period. Third, the visual inspection of a funnel plot showed asymmetry, and the Begg’s test and Egger’s test also identified evidence of publication bias among the studies included in our meta-analysis.

25Our meta-analysis indicated a beneficial role for aspirin for overall cancers; however, the results should be interpreted with caution. Considering that most evaluated studies were based on secondary prevention rather than on primary prevention, the totality of evidence for the high-risk population was incomplete, and it is appropriate to let the beneficial role remain uncertain. At present, we should accept the uncertainties, and future chemoprevention trials should clarify the extent to which aspirin decreases cancers incidence.

26Evidence from observational studies indicates that utilization of aspirin is associated with reduced risk of gastric, colorectal, esophageal, pancreatic, ovarian, endometrial, breast, and prostate cancers, in addition to small intestine neuroendocrine tumors. A stronger protective effect was observed in the North American populations and patients who used aspirin for at least 5 years. It is important to address immortal time bias not only to ensure the integrity of the meta-analysis, but also to ensure the integrity of pharmacoepidemiological studies. Moreover, given the confidence limits of the evaluated studies, adequately powered mechanistic studies should help elucidate the mechanisms underlying this correlation.