The molecular pathogenesis of prostate cancer has been characterized by alterations of genes and proteins involved in proinflammatory pathways.31,32
Epidemiological evidence suggests there is a close association between inflammation and prostate cancer.3–6
Several studies have demonstrated higher expression of Cyclo-oxygenase 2 (COX-2) in prostate cancer.33–36
Studies have revealed the COX-2 dependent and independent mode of action of selective COX-2 inhibitors against prostate cancer.37–39
Inflammatory pathway not only may be involved in carcinogenesis, but may also facilitate progression, local invasion, recurrence and metastasis.40,41
In a phase II study, Pruthi and colleagues42
showed a slowing effect of COX inhibitor celecoxib on the rate of prostate-specific antigen (PSA) rise after biochemical failure of local treatment of prostatic carcinoma. These findings have brought up the possibility of using anti-inflammatory drugs as a means of preventing this disease.
This meta-analysis attempts to evaluate the effectiveness of NSAIDs in reducing the risk of prostate cancer. Based on available studies, use of NSAIDs may have a 5% to 8% protective effect against prostate cancer. When ASA and other NSAIDs are analyzed separately, a statistically significant protective effect is still seen. The risk reduction is 5% for ASA and 8% for other NSAIDs. Our results are comparable to those of the meta-analysis by Mahmud and coauthors.7
Four studies reported OR for ASA and non-ASA NSAID separately.12,13,20,26
These studies posed a challenge to calculate the pooled OR, since they used the same sample for 2 different exposures. To remedy this problem we performed 4 different analyses: excluding these studies, including only ASA or only NSAID data, and including both exposures as if they were from 2 independent samples. The results were similar. The pooled OR varied between 0.93 and 0.95 with very close 95% CI. We have only presented the latter analysis in .
There are several potential pitfalls to this meta-analysis. Like any other systematic review, there may be a publication bias which cannot be entirely ruled out based on funnel plot. There is moderate heterogeneity in the results of the included studies, with the lowest for ASA only studies. This may be due to several factors such as study quality, methodological design, tools used to confirm the exposure or the outcome, variations in defining the exposure and background differences in patient populations. Patients exposed to any of the NSAIDs or ASA may differ in several other aspects that are relevant to the risk of the development of prostate cancer or its detection. For instance, results of a recent study showed that the potential protective effect of NSAIDs on prostate cancer may only exist among certain subgroups of men with particular variants of inflammatory response genes LTA
We performed subgroup analysis to further explore the possible factors affecting our analysis based on RRs shown in . We evaluated the studies according to their design: case–control versus cohort versus nested case–control. The type of study did not affect the findings significantly. We divided the studies into 2 groups according to their quality score: high quality with a score 7 or more out of 10 and lower quality with a score less than 7. Subgroup analysis revealed no difference between the findings in these 2 groups (). We acknowledge that using a scoring system in this situation is controversial. The result of our subgroup analysis may have been affected by the inability of the scoring system to probe into areas specific to studies of prostate cancer.
Results of subgroup analysis
In most studies use of NSAIDs was categorized as frequent, ever used or never used and the exact nature and duration of drug use is largely unknown. Although NSAIDs inhibit COX enzymes, the degree of this effect is variable among different NSAIDs. Furthermore, concurrent use of NSAIDs and ASA is another possibility, as many of these medications are available over the counter. Also, based on available studies, it was not possible to investigate the potential effect of different dosing regimens. Therefore, a clinically meaningful recommendation about the optimal duration and dose of NSAIDs or ASA use to prevent prostate cancer is evasive.
To confirm the exposure, questionnaires were used in some studies and prescription database in others. Recall bias is a well-known weakness of questionnaires. The extent of recall bias is related to characteristics of the exposure of interest and of the respondents.43,44
The level of experience of interviewers, reliability of the tool used to obtain data, and the characteristics of interviewees are among the most important factors that can affect the recall bias. On the other hand, medication database may fail to document over-the-counter use. We performed a subgroup analysis to further explore this issue. Studies using interview and/or questionnaires showed a small protective effect for NSAIDs, which was not seen when a database was used.
Meta-analysis of observational studies may suffer from bias in the original studies. Sources of data for studies included in our analysis were from various countries such as the United States, Canada, New Zealand and France. Background incidence depends on ethnicity and variability of PSA screening programs, which may be different in various countries (screening bias). Confounding by indication may affect the findings. This occurs when the medication used is associated with a condition that may affect the risk of prostate cancer detection. It is possible that users of ASA have a shorter life span than nonusers because of cardiovascular diseases. This may result in a lower chance of being screened and diagnosed with prostate cancer, overestimating the protective effect.45
Other types of bias may exist in studies dealing with prostate cancer, such as protopathic bias (when an exposure is influenced by early stages of a disease) and survivor bias (large number of exposed participants who have died from old age). Dissimilarity of study population could be a source of bias in case–control studies. For example, participants selected from referral centres cannot be a true representative of the general population (referral bias). The best way to eliminate bias is a prospective randomized experimental design. But until this type of study is available we are limited to observational studies that should be interpreted with all these limitations in mind.