Associations of aspirin, nonsteroidal anti-inflammatory drug and paracetamol use with PSA-detected prostate cancer: Findings from a large, population-based, case–control study (the ProtecT study)

Abstract

Evidence from laboratory studies suggests that chronic inflammation plays an important role in prostate cancer aetiology. This has resulted in speculation that nonsteroidal anti-inflammatory drugs may protect against prostate cancer development. We analysed data from a cross-sectional case–control study (n_cases= 1,016; n_controls= 5,043), nested within a UK-wide population-based study that used prostate specific antigen (PSA) testing for identification of asymptomatic prostate cancers, to investigate the relationship of aspirin, nonsteroidal anti-inflammatory drug (NSAID) and paracetamol use with prostate cancer. In conditional logistic regression models accounting for stratum matching on age (5-year age bands) and recruitment centre, use of non-aspirin NSAIDs [odds ratio (OR) = 1.32; 95% confidence interval (CI): 1.04–1.67] or all NSAIDs (OR = 1.25; 95% CI = 1.07–1.47) were positively associated with prostate cancer. There were weaker, not conventionally statistically significant, positive associations of aspirin (OR = 1.13; 95% CI = 0.94–1.36) and paracetamol (OR = 1.20; 95% CI = 0.90–1.60) with prostate cancer. Mutual adjustment for aspirin, non-aspirin NSAIDs or paracetamol made little difference to these results. There was no evidence of confounding by age, family history of prostate cancer, body mass index or self-reported diabetes. Aspirin, NSAID and paracetamol use were associated with reduced serum PSA concentrations amongst controls. Our findings do not support the hypothesis that NSAIDs reduce the risk of PSA-detected prostate cancer. Our conclusions are unlikely to be influenced by PSA detection bias because the inverse associations of aspirin, NSAID and paracetamol use with serum PSA would have attenuated (not generated) the observed positive associations.

Prostate cancer is a major cause of morbidity and mortality amongst elderly men.1 Advancing age, black race and a family history of prostate cancer are known to predispose to the disease. However, at the time of writing, little is known about modifiable risk factors for the disease. Chronic inflammation seems to be important in prostate carcinogenesis2 and cyclooxygenase-2 (COX-2) has emerged as a potential target for preventative strategies. COX-2 is an inducible enzyme that facilitates inflammation by catalysing the conversion of arachidonic acid to prostaglandins, powerful inflammatory mediators.3 Evidence to support a role of COX-2 in prostate cancer aetiology has come from studies demonstrating its overexpression in human prostatic tumours4–6 and the finding that genetic variation within prostaglandin endoperoxidase synthase 2, the gene that encodes COX-2, is associated with alterations in prostate cancer risk.7

These findings have led to speculation that nonsteroidal anti-inflammatory drugs (NSAIDs), for example aspirin and ibuprofen, which inhibit the COX pathway could reduce prostate cancer risk. There is evidence that NSAID use is associated with a reduced risk of colorectal,8, 9 oesophageal,10, 11 gastric12 and breast13 cancer. It remains unclear, however, whether NSAID use reduces prostate cancer risk. Some epidemiological studies have reported that aspirin use is inversely associated with prostate cancer incidence.14–18 Others, however, including a recent secondary analysis of a randomised controlled trial (the Aspirin/Folate Polyp Prevention Study),19 found no evidence of an effect.20–25 The results of studies investigating whether all-NSAID use reduces prostate cancer risk have also been mixed, with the majority finding no evidence for an effect15, 16, 23, 25 and a minority reporting positive10, 26, 27 or inverse17, 18, 28 associations. A meta-analysis published in 2010 found no randomised controlled trials, but the body of published observational studies were consistent with an approximately 20% reduction in risk of total and advanced prostate cancer associated with aspirin in both case–control and cohort studies. In contrast, the literature on all-NSAID or non-aspirin (NA)-NSAID use was inconsistent, with cohort studies in particular indicating null effects on total prostate cancer risk and a 12% increased risk of advanced prostate cancer associated with NA-NSAIDs.29

We used data from a large, population-based, case–control study, nested in cross-section within the case-finding phase of the Prostate testing for cancer and Treatment (ProtecT) trial, to test the hypothesis that aspirin, NA-NSAIDs and all NSAIDs are associated with a reduced risk of prostate cancer. To test the specificity of these associations, we also tested the hypothesis that paracetamol, an analgesic that is not an NSAID and is not believed to influence COX-2 activity at therapeutic doses,30 is not associated with prostate cancer risk.

Material and Methods

Study population

ProtecT is an ongoing, multicentre, randomised controlled trial that will compare the efficacy, cost-effectiveness and acceptability of treatments for localised prostate cancer. Between 2001 and 2008, more than 110,000 men aged 50–69 years, from approximately 300 primary care centres (general practices) across the United Kingdom, attended prostate check clinics, where histologically confirmed prostate cancer cases were identified through a combination of prostate specific antigen (PSA) testing, digital rectal examination (DRE) and, for men with abnormal PSA levels or DRE findings, 10-core transrectal ultrasound-guided biopsy (case-finding phase of ProtecT). Repeat biopsies were offered to men with a normal initial biopsy in whom there was a high index of clinical suspicion or in whom PSA concentration was persistently elevated. Tumours were staged using the tumour node metastasis staging system.

Case–control selection

The case–control series used in our study is nested within the ProtecT case-finding phase and involves men who completed a ‘Diet, health and lifestyle questionnaire (Version 3)’ that was first introduced in January 2004. This questionnaire included questions about medication usage that previous versions of the questionnaire did not. Cases were men aged between 50 and 69 years who underwent PSA measurement and had a histological diagnosis of prostate cancer. We defined localised cancer as T1–T2, NX or NO, MX or MO and advanced cancer as T3–T4 or N1 or M1.

All participants with no evidence of prostate cancer after PSA testing, DRE and/or biopsy were eligible to be controls. Controls were stratum matched to cases by age (5-year bands) and the primary care centres from which they were recruited. The index date for controls was the date of the prostate check clinic. Such matching automatically matches for calendar time, because prostate check clinics were completed sequentially. It is important to note that both outcome (case vs. control status) and exposure status (based on completion of questionnaire) were effectively defined in cross-section soon after the prostate check clinic. The design is therefore a cross-sectional, case–control study. The study received ethical approval from Trent Multicentre Research and Ethics Committee. Detailed descriptions of ProtecT and the protocol for case–control selection are published elsewhere.31–33

Exposure assessment

Data were collected using a mailed, self-completed questionnaire. Most participants (83% cases and 86% controls) completed the questionnaire before the results of PSA measurement were known (the timing of questionnaire completion was available on 97% of cases and controls). Data on aspirin, NSAID and paracetamol use were obtained using a table into which participants were asked to write-in ‘details of all the medications and supplements [they] currently take’. The table included columns for ‘medication [name]’, ‘amount taken and how often’ and ‘reason for taking’. No information was collected on past, or duration of, use. The questionnaire guidance instructed participants to include information on both prescribed and over-the-counter medications. Respondents could provide either generic or brand names. A trained research nurse coded responses in accordance with the anatomical therapeutic chemical (ATC) 2006 classification system. Participants who listed ‘aspirin’ as a medication being taken were considered aspirin users. All others were considered non-aspirin users. Participants who listed ‘aspirin’, ‘ibuprofen’, ‘indometacin’, ‘sulindac’, ‘diclofenac’, ‘alclofenac’, ‘etodolac’, ‘acemetacin’, ‘ketorolac’, ‘aceclofenac’, ‘piroxicam’, ‘tenoxicam’, ‘droxicam’, ‘lornoxicam’, ‘meloxicam’, ‘naproxen’, ‘ketoprofen’, ‘fenoprofen’, ‘fenbufen’, ‘flurbiprofen’, ‘tiaprofenic acid’, ‘dexibuprofen’, ‘dexketoprofen’, ‘mefenamic acid’, ‘tolfenamic acid’, ‘celecoxib’, ‘rofecoxib’, ‘valdecoxib’, ‘parecoxib’, ‘etoricoxib’, ‘lumiracoxib’, ‘nabumetone’, ‘azapropazone’ or combinations including these drugs as a medication being taken were considered all-NSAID users. NA-NSAID users were defined as users of any of the above NSAIDs with the exception of aspirin. All others were considered nonusers. Participants who listed ‘paracetamol’ as a medication being taken were considered paracetamol users. All others were considered non-paracetamol users.

Data on exact age measured at the prostate check clinic, family history of prostate cancer, body mass index (BMI) and self-reported diabetes status were also collected at prostate check clinics.

Statistical methods

A direct effect of aspirin, NSAID or paracetamol use on serum PSA concentration could create bias by determining whether or not a participant undergoes prostatic biopsy and therefore receives a histological diagnosis of prostate cancer. To assess the potential for PSA detection bias, associations of aspirin, NSAID and paracetamol use with serum PSA concentration were investigated using linear regression performed on log-transformed PSA concentrations amongst controls. Log-transformed PSA concentrations were used because untransformed PSA concentrations amongst controls demonstrated a left-skewed distribution.

We calculated odds ratios (ORs) and 95% confidence intervals (CIs) for the association of aspirin, all-NSAID, NA-NSAID and paracetamol use with prostate cancer using conditional logistic regression to account for the stratum matching of cases to controls. In multivariable conditional logistic regression, we also investigated the impact of mutual adjustment for aspirin, NA-NSAIDs or paracetamol, as appropriate. In addition to age at the prostate check clinic and family history of prostate cancer (established risk factors), BMI and diabetes status were also considered potential confounders because there is evidence that these variables modify prostate cancer risk,34, 35 whilst predisposing to osteoarthritis36 and cardiovascular disease, respectively,37 illnesses that are frequently treated with NSAIDs. The impact of multivariable adjustment for exact age (continuous variable), family history of prostate cancer (yes/no), BMI (continuous variable) and self-reported diabetes status (yes/no) was investigated, but inclusion of any or all of these variables to the basic conditional logistic regression models altered the ORs by less than 3%, so we present the basic and mutually adjusted models only. In a sensitivity analysis, we restricted our analyses to those 85% of cases and controls who we knew had completed the questionnaire before obtaining their PSA result.

To determine whether associations differed for localised versus advanced cancers, we also compared associations of aspirin, NSAIDs and paracetamol use with localised compared to advanced-stage cancers (reference group) in logistic regression models controlling for age at prostate check clinic. All analyses were performed using Stata 10 software (StataCorp, College Station, TX).

Results

A total of 2,019 cases and 11,809 controls (randomly selected for complete questionnaire data entry from all noncases) were available in the ProtecT case–control series at the time of the current analysis. Of these, 391 cases and 2,301 controls completed an early version of the ProtecT ‘Diet, health and lifestyle questionnaire’ that did not contain detailed questions about medication usage. A further 612 cases and 4,465 controls chose not to complete any version of this questionnaire. Our study population therefore comprised the 1,016 cases and 5,043 controls who completed the ProtecT ‘Diet, health and lifestyle questionnaire (Version 3)’ and thereby provided detailed information about medication usage. The 1,016 cases comprised 901 localised, 112 advanced and 3 indeterminate cancers. Cases were 2.6 months older than controls, were more likely to have a family history of prostate cancer and were slightly less likely to have diabetes mellitus or be obese (Table 1).

Table 1. Characteristics of the ProtecT study population

Participants using NA-NSAIDs were five times more likely to also use paracetamol than participants who did not use NA-NSAIDs (OR = 5.21; 95% CI = 4.01–6.76). Aspirin users were marginally more likely to also use NA-NSAIDs (OR = 1.14; 95% CI = 0.90–1.44) and paracetamol (OR = 1.22; 95% CI = 0.92–1.61) than non-aspirin users. All participants who reported taking aspirin or NA-NSAIDs provided a reason for their use. Aspirin was used as an ‘antithrombotic agent’ by 96.8% of users. NA-NSAIDs were used for their ‘anti-inflammatory and anti-rheumatic’ effects by 100% of users. Aspirin, NSAID and paracetamol use were each associated with a reduction in PSA concentration amongst controls (Table 2).

Table 2. Relationship between aspirin, NSAID and paracetamol use with PSA concentration amongst 5,043 controls

NA-NSAID and all-NSAID use were each associated with an increased risk of prostate cancer (Table 3). There were weaker, not conventionally statistically significant, positive associations of aspirin and paracetamol with prostate cancer. There was no evidence that associations differed by stage when we examined aspirin, NSAIDs and paracetamol in relation to localised compared to advanced cancers: the age-adjusted ORs (outcome: localised versus advanced stage) were as follows: aspirin, OR = 1.19 (95% CI = 0.71–1.99); NA-NSAIDs, OR = 1.08 (95% CI = 0.56–2.09); all-NSAIDs, OR = 1.13 (95% CI = 0.72–1.77) and paracetamol, OR = 1.11 (95% CI = 0.49–2.49). The small numbers of advanced cancers in our study, however, meant that CIs were wide.

Table 3. Association of aspirin, NSAID and paracetamol use with PSA-detected prostate cancer

The results for all-NSAID and paracetamol use were identical when only participants who completed the questionnaire before obtaining their PSA result were analysed. The positive association of NA-NSAIDs with prostate cancer was slightly strengthened when only participants who completed the questionnaire before obtaining their PSA result were analysed (OR basic model = 1.46; 95% CI = 1.14–1.87).

Discussion

In our large, population-based, case–control study, NA-NSAID and all-NSAID use were associated with an increased risk of PSA-detected prostate cancer. These positive findings are unlikely to be due to PSA-based detection bias because aspirin, NSAID and paracetamol use were each associated with a reduction in PSA concentration. This bias could have differentially misclassified some NSAID-exposed cases as controls and so may have attenuated (not generated) the observed positive associations, possibly resulting in an underestimation of the magnitude of the increased risk of prostate cancer amongst users of these drugs.

Our findings are not in accordance with our a priori hypothesis. However, it is of note that they are similar in both nature and magnitude to those of three other large, case–control studies. The first, using data from the UK General Practice Research Database, concluded that NSAID use was associated with a 33% increased risk of prostate cancer (OR = 1.33, 95% CI = 1.07–1.64).10 The second combined data from the North Jutland Prescription Database and the Danish Cancer Registry and found that NSAID use was associated with a 60% increased risk of prostate cancer (standardised incidence ratio = 1.6, 95% CI = 1.3–2.0).26 Finally, a study, as yet unpublished as a complete article, which used comprehensive data from a collection of Finnish registers has reported that NSAID use is associated with an increased risk of prostate cancer (OR = 1.33, 95% CI = 1.28–1.38).27

The authors of these studies could not provide a physiological explanation for the findings but speculated that they might be attributable to protopathic bias, as a result of NSAIDs being prescribed to treat bone pain caused by undiagnosed prostate cancer, or to detection bias, because NSAID users would be expected to attend health services more frequently than nonusers. Whilst we also do not know the mechanism by which NSAID use might increase risk of prostate cancer, our study adds to these results by effectively excluding both protopathic and detection biases as an explanation. We do not believe that protopathic bias will have significantly influenced our results because the increased risk of prostate cancer seen amongst NSAID users persisted when analysis was restricted to local cancers, which do not cause bone pain.38 Whilst we cannot exclude the possibility that the increased risk of prostate cancer observed amongst NSAID users is due to the underlying condition necessitating NSAID therapy, we believe that this is unlikely. In our study, both aspirin use and NA-NSAID use were associated with an increased risk of prostate cancer despite the fact that they have dissimilar therapeutic indications and it seems unlikely that both cardiovascular disease (the main indication for aspirin use within our population) and arthritic diseases (the only indication for NA-NSAID use amongst our population) would predispose to prostate cancer in a similar fashion. Finally, we also do not believe that detection bias is a likely explanation for our positive results because we used a population-based approach, which gave all men, irrespective of medication use, an equal chance of undergoing PSA testing.

NSAID use was associated with reduced PSA concentrations amongst controls in our study. Whilst the precise mechanism by which NSAID use could reduce PSA concentration is uncertain, such an effect is plausible and we speculate that it may be secondary to a reduction in blood flow to inflamed, PSA-producing prostatic tissue thereby limiting this tissue's ability to secrete PSA into the circulation.39 The 7% reduction in PSA concentration that we observed is similar to that reported by other groups.40, 41 However, these groups were unable to determine whether the reduction in PSA concentration that they reported was merely due to an effect of NSAIDs on reducing prostate cancer incidence. The reduction in PSA concentration that we witnessed was present when we tested only amongst controls, although we accept that some controls will have had unrecognised prostate cancer at rates proportional to their PSA level.42

Other studies have found little evidence that paracetamol use alters risk of prostate cancer.25, 27, 43 We found a weak positive association that was not statistically significant. The finding that PSA concentration was also reduced amongst paracetamol users has implications. It is possible that detection bias secondary to the reduction in PSA concentration may have resulted in an underestimation of the risk of prostate cancer associated with paracetamol use, and we cannot exclude the possibility that paracetamol use is associated with an increased prostate cancer risk.

Our study has several strengths in addition to those discussed above. Notable amongst these is the manner in which we assessed exposure to aspirin, NSAID and paracetamol use. Multiple studies have ascertained that compliance rates amongst patients prescribed NSAIDs fall to below 50%44, 45 within 1 year. Furthermore, within the UK, prescription aspirin, NSAID and paracetamol use comprises only a minority of total use, with ‘over-the-counter’ use comprising the majority.46, 47 Studies reliant on prescribing databases10, 15, 16, 26, 27 for assessing exposure to these medications may therefore be using inaccurate data to estimate risk. In contrast, we collected data by asking patients to self-report ‘details of all the medications and supplements [they] currently take’, in the anticipation that participants would include only medications, including those bought ‘over-the-counter’, that they were genuinely using. Further confidence in the accuracy of our exposure assessment comes from the fact that 82% of participants provided information on medication use before receiving results of PSA screening, suggesting that significant recall bias is unlikely.

Similarly, the population-based PSA testing approach used in ProtecT to identify prostate cancer cases should provide a more complete picture of the effects of aspirin, NSAID and paracetamol use on prostate cancer risk than other approaches. Autopsy studies have indicated that the majority of prostate cancers remain asymptomatic for a prolonged period, and a sizeable proportion will consequently never be detected by standard medical services.48 Studies that rely on these medical services to identify cases may therefore miss a substantial proportion of localised, asymptomatic prostate cancers, which would be detected under our population-based PSA testing approach. This might explain why we found an increased risk of prostate cancer amongst NSAID users when many other studies have not,15–17, 23, 25, 28 particularly if NSAIDs predispose predominantly to localised, asymptomatic prostate cancers. A second advantage of using a population-based PSA testing approach is that our results are unlikely to have been affected significantly by survivor bias. Finally, the large size of our study suggests that we are unlikely to have failed to detect a clinically important effect of using these medications on overall prostate cancer risk.

Our study has limitations. As discussed above, all the men who were ultimately diagnosed with prostate cancer had occult prostate cancer at the time of the assessment of medication use. The natural history of occult, asymptomatic prostate cancer is poorly understood49 and the clinical significance of these cancers is uncertain. Therefore the generalisability to more advanced, clinically detected disease may be limited, particularly if NSAIDs have a different association with progression compared to development of prostate cancers. The cross-sectional nature of our study means that conclusions cannot be drawn about the causality of the associations that we have observed. Furthermore, we did not collect information on duration of medication use. Some studies have observed that the effects of NSAIDs on prostate cancer risk vary with duration of use15–17 and we would have liked to have investigated whether the increased risk of prostate cancer observed amongst NSAID users was duration dependent. Data on NSAID use were self-reported so will be subject to a degree of misclassification. As most participants completed the questionnaire before receiving their PSA result, this misclassification is likely to be nondifferential with respect to outcome, attenuating effect estimates to the null. Most people used aspirin for cardioprotection, which would have been largely at low doses; we do not know whether associations would be stronger with analgesic doses.

We hope that these limitations are addressed by future prospective studies in this area. Long-term follow-up of men with PSA-detected localised cancer to test associations of aspirin and NSAID use with clinically defined prostate cancer progression will also be important. Such studies are indicated because the increased risk of prostate cancer that we observed amongst NSAID users, if confirmed, would have important clinical and public health implications, particularly over the safety of using NSAIDs as chemopreventive agents in the primary prevention of prostate cancer. In conclusion, our large, population-based, case–control study found no evidence that aspirin, NA-NSAID and all-NSAID use are associated with a reduced risk of PSA-detected prostate cancer.

Acknowledgements

The authors thank the tremendous contribution of all members of the ProtecT study research group and especially the following who were involved in this research: Mr. Prasad Bollina, Ms. Sue Bonnington, Ms. Debbie Cooper, Mr. Andrew Doble, Mr. Alan Doherty, Ms. Emma Elliott, Mr. David Gillatt, Ms. Pippa Herbert, Mr. Peter Holding, Ms. Joanne Howson, Ms. Mandy Jones, Mr. Roger Kockelbergh, Prof. Howard Kynaston, Ms. Teresa Lennon, Ms. Norma Lyons, Ms. Hilary Moody, Mr. Philip Powell, Mr. Stephen Prescott, Ms. Liz Salter and Ms. Pauline Thompson.