Fax: +418-691-5562 or +418-691-5439
Dosage, duration and timing of nonsteroidal antiinflammatory drug use and risk of prostate cancer
Article first published online: 16 MAY 2003
Copyright © 2003 Wiley-Liss, Inc.
International Journal of Cancer
Volume 106, Issue 3, pages 409–415, 1 September 2003
How to Cite
Perron, L., Bairati, I., Moore, L. and Meyer, F. (2003), Dosage, duration and timing of nonsteroidal antiinflammatory drug use and risk of prostate cancer. Int. J. Cancer, 106: 409–415. doi: 10.1002/ijc.11250
- Issue published online: 1 JUL 2003
- Article first published online: 16 MAY 2003
- Manuscript Accepted: 18 MAR 2003
- Manuscript Revised: 17 MAR 2003
- Manuscript Received: 5 NOV 2002
- Canadian Institute for Health Research. Grant Number: MOP-36447
- prostatic neoplasm;
- nonsteroidal anti-inflammatory drug;
- case-control study;
- dose–response relationship
Experimental studies suggest that NSAIDs could reduce prostate cancer risk. Results of observational studies on the relation between NSAIDs and prostate cancer risk have, however, been inconsistent. Moreover, none has addressed the issues of dosage, duration and timing of exposure. In a population-based, age-matched case-control study, we measured the association between prostate cancer risk and NSAIDs defined in terms of mean daily dose, cumulative duration and timing of exposure. Eight-year drug exposure history was obtained from the Quebec health insurance system database. Parallel analyses were performed for aspirin and NSAIDs other than aspirin. We controlled for detection bias and assessed the potential impact of protopathic bias. Analyses were performed with conditional logistic regression. Among the 2,221 cases and 11,105 controls, there was a negative trend between cumulative duration of aspirin use and prostate cancer risk (p = 0.0009). Also, exposure to a mean daily dose of aspirin of at least 80 mg, maintained throughout the entire 8 years of follow-up, was associated with an 18% reduction in prostate cancer risk (OR = 0.82, 95% CI 0.71–0.95). In more recent users of such a dose, the risk reduction was 7%. However, 1 year after the end of a 7-year regular aspirin exposure, no residual protective effect persisted. No association was observed between prostate cancer risk and exposure to NSAIDs other than aspirin. The results suggest that long-term and regular use of aspirin, at a dosage beneath that usually recommended for an anti-inflammatory effect, may prevent prostate cancer. © 2003 Wiley-Liss, Inc.
In North America and Western Europe, prostate cancer has become one of the most commonly diagnosed cancers and one of the leading causes of cancer death. Since the incidence of this disease increases strikingly with age,1 the number of new cases will continue to rise in the Western world as the population ages. So far, the control of prostate cancer has focused on early detection and treatment; but to develop preventive strategies, researchers are investigating the protective effect of a wide range of agents.2 Among them are NSAIDs.2
Many epidemiologic studies have found that long-term use of NSAIDs is associated with a lower risk of colorectal cancer and adenomatous polyps.3 NSAIDs may also help to lower the incidence of, or death from, cancers at several other sites.3 Despite continuing uncertainty, there is mounting evidence that the antineoplastic effect of NSAIDs is principally mediated through restoration of apoptosis and inhibition of angiogenesis.3 The pathways through which NSAIDs exert their antineoplastic effects have not been completely elucidated but probably involve their shared ability to inhibit the COX-2 isoenzyme.3 COX-2 catalyzes the transformation of arachidonic acid into prostaglandins, which are in turn involved in inflammation and carcinogenesis.3 Moreover, some studies indicate that NSAIDs could also have antitumorigenic activity through COX-2-independent processes.3, 4, 5, 6
Several in vitro and in vivo laboratory studies have demonstrated that NSAIDs can decrease growth and increase apoptosis of prostate cancer cells.7, 8, 9, 10, 11, 12, 13, 14, 15, 16 The results of these studies suggest that NSAIDs may exert their prostatic antitumorigenic effect through both COX-2-dependent7–9, 11, 12 and COX-2-independent13, 14, 15, 16 pathways. In epidemiologic studies, however, the protective effect of NSAIDs on prostate cancer development has not been conclusively demonstrated. In 3 studies, exposure to NSAIDs or to a proxy of NSAID use was associated with an increased risk of prostate cancer,17, 18, 19 which was statistically significant in 1 case.17 In 5 other observational studies, NSAID exposure was associated with nonstatistically significant reductions of prostate cancer risk ranging 5–18%.20, 21, 22, 23, 24 Conversely, in 1 case-control study25 and in 1 cohort study,26 investigators reported statistically significant risk reductions of prostate cancer of at least 50% in NSAID users. Discrepancies between these results could be partly explained by misclassification of exposure, detection bias, different ages of study populations and various lengths of follow-up.
Using the computerized database of the Quebec health insurance system, the RAMQ, we conducted a population-based, age-matched case-control study. This database allows a more precise classification of exposure than has been possible in most previous studies. Our aim was to determine whether the use of NSAIDs, and aspirin in particular, over an 8-year observation period was associated with the risk of prostate cancer. We also assessed the effect of dosage, duration of exposure and timing of exposure, issues that have not been directly addressed.
MATERIAL AND METHODS
We defined the source population as male RAMQ enrollees aged 73–79, between 1 January 1993 and 31 December 1995. From 1975 to 1997, the RAMQ reimbursed all drugs prescribed on an outpatient basis to persons aged 65 and older, who became automatically eligible on the day of their sixty-fifth birthday. The insurance plan covered all Quebec's residents with the exception of first nations' members, military employees and inmates of federal penitentiaries, who benefit from a federal program. From 1975 until May 1992, all eligible people presenting at a pharmacy with a medical prescription could receive their medication free of charge. Afterward, a copayment of $2 per prescription, up to a maximum fee of $100 per year, was imposed. The RAMQ reimbursed pharmacists for all drugs dispensed under this insurance plan. Moreover, in Quebec, all residents benefit from a universal insurance plan for medical services.
We excluded all men who had a prostate cancer reported to the Quebec Cancer Registry between 1984 and the date of their inclusion in the source population. We also excluded all immigrants, refugees and persons who had resided outside Quebec for 6 months or more after their sixty-fifth birthday, to guarantee that the exposure period was free of any meaningful interruption.
Case and control definition
Cases were subjects from the source population who were diagnosed with prostate cancer (ICD-9 code 185) between 1 January 1993 and 31 December 1995 and reported to the Quebec Cancer Registry. We considered the diagnosis date indicated in the first report to the cancer registry to be the date of diagnosis. Potential controls for each case were subjects from the source population who were born on the exact same date as the case and alive on the date of diagnosis of the case. From every set of potential controls, 5 men were randomly selected. The date of diagnosis was assigned to each of the matched controls as their index date.
Information on the use of NSAID, aspirin and other drugs
Drug-exposure data were obtained from the RAMQ pharmacists' computerized claims database, which keeps a record of all drugs reimbursed since 1981. The database has been described in detail elsewhere, and a high level of reliability and validity of the prescription data has been demonstrated.27 Dispensation of prescribed drugs could be reconstituted from the date of the sixty-fifth birthday or 1 January 1981 until the index date. Therefore, the information on the dispensation of prescribed drugs was available for a period of 8–14.9 years before the index date, depending on the subject's age. However, to include all subjects in the analyses and to have an acceptable statistical precision, we only considered the exposure during the 8 years immediately preceding the index date.
Available information included the date of delivery; the class and identity of the drug (American Hospital Formulary System classification); the number of tablets, capsules or other vehicle dispensed; the drug dosage; and the duration of the prescription period. All prescriptions of oral NSAIDs and aspirin filled during the observation period were used in the creation of exposure variables, as described below. Since in most cases aspirin was prescribed at levels below that necessary for an anti-inflammatory effect, we constructed separate exposure variables for aspirin and all other NSAIDs. For comparison of dosages of NSAIDs other than aspirin, we used the DDD, which is the average dosage of a drug taken by adults for the most frequent indication, according to the WHO.28
For every subject, we measured the cumulative exposure to prescribed NSAIDs and aspirin from 3 months to 8 years before the index date. We calculated the ORs of prostate cancer (and 95% CIs) in relation to the use of NSAIDs and aspirin with conditional logistic regression. All categorical variables were introduced in the model using dummy variables with “no use” as the reference category.
We first compared the effect of no use to any use of drug. Second, we assessed the effect of duration of exposure by categorizing the cumulative duration of drug use into mutually exclusive classes up to 6 years or more. Third, we assessed the association between mean daily dose and the risk of prostate cancer separately for the subjects who had cumulated less than 4 years and at least 4 years of drug use. Mean daily dose was computed over the period between the date of the first prescription of aspirin or NSAIDs, found during the 8-year observation period, and 3 months before the index date. Mean daily dose was thereafter divided into 3 mutually exclusive categories. Finally, to measure the effect of timing of exposure, we compared nonusers to the subjects exposed exclusively 3 months to 4 years before the index date (recent users) and the subjects exposed throughout the period of 3 months to 8 years before the index date (chronic users). Variables were constructed separately for aspirin and NSAIDs and parallel analyses performed. To sum dosages of different classes of NSAIDs, we computed, for each prescription, a standardized total dose by dividing the total number of milligrams of drug dispensed by the drug's DDD. Thereafter, to obtain a mean daily dose of NSAIDs for the follow-up period, we summed all standardized total doses dispensed during the follow-up period and divided by the number of days in that period. The standardized mean daily dose of NSAIDs can therefore be interpreted as a percentage of a usual daily dose to obtain an anti-inflammatory effect.
The potentially confounding variable considered was the dispensation of antihypertensive agents, which have been negatively associated with prostate cancer.29 In the analysis of the relation between NSAIDs other than aspirin and prostate cancer, we also considered the use of aspirin as a potential confounder. Detection bias might occur if those taking NSAIDs or aspirin were under increased medical surveillance compared to nonusers. To control such bias, we included in all analyses a dichotomized variable indicative of medical contacts that occurred 1–3 months before the index date. We measured the existence of medical contacts by counting the number of different days where one or more prescriptions of any medication were dispensed. Since they are used in the treatment of benign prostatic hyperplasia, we also considered the dispensation of finasteride and terazosin 1 month to 2 years before the index date as a potential source of detection bias. The criterion for confounding was a change of 10% or more in the parameter estimate when the covariable was introduced into the model.
To assess the presence of a protopathic bias, which might be caused by changes in drug use during the period preceding the diagnosis of prostate cancer,30 we examined the influence of 3 different lag times on the ORs. To do so, all of the analyses described above were performed with exposure measures excluding either the last 3 months, the last year or the last 2 years before the index date. We hypothesized that if men took NSAIDs or aspirin to control symptoms of a yet undiagnosed prostate cancer, the ORs should tend toward positive values with shorter lag-time periods.
The study population comprised 2,221 cases and 11,105 controls. At the index date, subjects were on average aged 75.7 years (range 73–79). Characteristics of the study population with regard to potential confounding and detection bias variables are given in Table I. In the following analyses, only the variable indicating a medical contact during the last months before the index date (recent medical contacts) had an influence on the parameter estimates and was kept in the models to control for detection bias.
|Drug dispensed||Cases (n = 2,221)||Controls (n = 11,105)||Age-adjusted OR (95% CI)|
|Recent medical contacts1||1,887||7,710||2.48 (2.19–2.81)|
|Finasteride or terazosin2||77||204||1.91 (1.47–2.49)|
|Antihypertensive agents3 (3 months to 8 years before index date)||1,319||5,975||1.26 (1.15–1.38)|
There were 155,988 prescriptions of aspirin and 108,332 prescriptions of NSAIDs other than aspirin dispensed to the study population (Table II). Diclofenac, naproxyn, ibuprofen and indomethacin accounted for 65% of the 108,332 NSAID prescriptions (Table II). Cumulative duration of aspirin use reached 4,722,901 days, whereas cumulative duration of NSAID use reached 2,599,234 days (Table II). Fifty-eight percent of subjects were never exposed to prescribed aspirin, while 38% were never exposed to prescribed NSAIDs. The prescribed duration of drug use for aspirin was on average 30.28 days, while it was 23.99 days for NSAIDs. The prescribed daily dose of aspirin, as measured by dividing the total number of milligrams prescribed by the total number of days the drug was recommended, was on average of 706 mg. It was below 325 mg in 51% of prescriptions and above 2,600 mg in 1% of them. The WHO does not provide a DDD for the anti-inflammatory effect of aspirin, but as an indication, the Canadian Pharmacists Association suggests a daily dose of 2,600–3,900 mg for aspirin to provide an anti-inflammatory effect.31 For NSAID prescriptions, the average prescribed daily dose was 116% of the drug's respective DDD. About 5% of NSAID prescriptions recommended a daily dose below 50% of the drug's DDD, while 38% recommended a daily dose above 100% of the drug's DDD. Throughout the observation period, the prescribed duration and dosage of aspirin and NSAIDs to the study population were fairly constant from one year to the other.
|Class of drugs||Prescriptions (% of total)1||Cumulative duration, days (% of total)1||DDD (mg)|
|Diclofenac||22,122 (20.42)||529,410 (20.37)||100|
|Naproxyn||22,075 (20.38)||517,381 (19.91)||500|
|Ibuprofen||13,309 (12.29)||298,651 (11.49)||1200|
|Indomethacin||13,007 (12.01)||312,976 (12.04)||100|
|Piroxicam||10,133 (9.35)||283,238 (10.90)||20|
|Ketoprofen||9,270 (8.56)||229,138 (8.82)||150|
|Sulindac||5,572 (5.14)||136,022 (5.23)||400|
|Flurbiprofen||4,761 (4.39)||106,440 (4.10)||200|
|Tiaprofenic acid||4,237 (3.91)||104,907 (4.04)||600|
|Others3||3,846 (3.55)||81,071 (3.12)||—|
Use of aspirin 3 months to 8 years before the index date, defined as a binary variable and compared to no use during the same period, was associated with an adjusted OR of prostate cancer of 0.94 (95% CI 0.85–1.03). Use of NSAIDs during the same period was associated with an adjusted OR of prostate cancer of 1.07 (95% CI 0.97–1.18).
To assess the duration–response gradient, we categorized aspirin use into mutually exclusive classes up to 6 years or more of cumulative duration of use (Table III). Until 4 years of cumulative exposure, no clear trend in the risk of prostate cancer relative to nonusers was observed. However, relative to nonusers, the risk of prostate cancer was 0.92 (95% CI 0.71–1.19) in those who cumulated between 4 and 5 years of use, 0.70 (95% CI 0.52–0.95) in those who cumulated 5–6 years of use and 0.66 (95% CI 0.51–0.84) in those who cumulated at least 6 years of use. The χ2 for trend was statistically significant (p = 0.0009). No trend was found in the association between the cumulative duration of exposure to NSAIDs other than aspirin and prostate cancer (p for trend = 0.152) (Table III).
|Cumulative duration of use (years)||Cases (n = 2,221)||Controls (n = 11,105)||Age-adjusted OR (95% CI)||Multivariate model OR (95% CI)1|
|>0–1||444||1,972||1.19 (1.06–1.35)||1.02 (0.91–1.16)|
|≥1–2||144||701||1.09 (0.90–1.32)||0.88 (0.73–1.07)|
|>2–3||111||457||1.29 (1.04–1.60)||1.04 (0.83–1.29)|
|≥3–4||97||393||1.31 (1.04–1.65)||1.04 (0.82–1.31)|
|≥4–5||76||355||1.13 (0.88–1.46)||0.92 (0.71–1.19)|
|≥5–6||53||306||0.92 (0.68–1.24)||0.70 (0.52–0.95)|
|≥6||78||477||0.86 (0.68–1.11)||0.66 (0.51–0.84)|
|>0–1||1,098||5,215||1.22 (1.10–1.35)||1.06 (0.95–1.17)|
|>1–2||163||644||1.46 (1.21–1.76)||1.21 (1.00–1.46)|
|>2–3||61||348||1.02 (0.77–1.35)||0.80 (0.60–1.07)|
|>3–4||50||198||1.46 (1.06–2.01)||1.14 (0.83–1.57)|
|>4–5||38||122||1.80 (1.24–2.62)||1.40 (0.96–2.04)|
|≥5||60||231||1.50 (1.12–2.01)||1.14 (0.85–1.54)|
In the group with less than 4 years of aspirin use, we found no association between mean daily dose and prostate cancer risk (Table IV). Conversely, in the group with at least 4 years of cumulative aspirin use, the risk of prostate cancer was 0.81 (95% CI 0.62–1.05) in those with a mean daily dose below 325 mg and 0.72 (95% CI 0.59–0.88) in those with a mean daily dose of at least 325 mg (Table IV). For NSAIDs other than aspirin, the analysis did not reveal any trend between the mean daily dose and the risk of prostate cancer for either the subjects who cumulated less than 4 years of use or those with 4 or more years of cumulative use (Table IV).
|Mean daily dose||Cases (n = 2,221)||Controls (n = 11,105)||Age-adjusted OR (95% CI)||Multivariate model OR (95% CI)1|
|Less than 4 years of use|
|<325 mg||645||2,896||1.18 (1.06–1.31)||0.99 (0.89–1.10)|
|≥325 mg||151||629||1.27 (1.05–1.54)||1.03 (0.85–1.25)|
|At least 4 years of use|
|<325 mg||73||375||1.03 (0.79–1.33)||0.81 (0.62–1.05)|
|≥325 mg||134||761||0.93 (0.77–1.13)||0.72 (0.59–0.88)|
|Less than 4 years of use|
|<50% of DDD2||1,304||6,179||1.22 (1.11–1.35)||1.05 (0.95–1.16)|
|≥50% of DDD||68||227||1.74 (1.31–2.31)||1.37 (1.03–1.82)|
|At least 4 years of use|
|<50% of DDD||9||31||1.70 (0.81–3.57)||1.30 (0.62–2.76)|
|≥50% of DDD||89||321||1.60 (1.25–2.04)||1.23 (0.96–1.58)|
Moreover, to understand the effect of the timing of exposure, we compared nonusers to recent users and to chronic users. Recent users were defined as those with no exposure to aspirin 4–8 years prior to the index date but a mean exposure of at least 80 mg per day from 3 months to 4 years prior to the index date, whereas chronic users were those with a mean daily dose of aspirin of at least 80 mg throughout the entire period of 3 months to 8 years before the index date. Compared to nonusers, the risk of prostate cancer was 0.94 (95% CI 0.79–1.12) in recent users and 0.82 (95% CI 0.71–0.95) in chronic users (Table V). Among chronic users of aspirin, the mean daily dose received over the 8-year observation period was below 325 mg, between 325 and 650 mg, between 650 and 1,300 mg and above 1,300 mg in 44%, 34%, 19% and 3% of subjects, respectively. Definition of recent and chronic use of NSAIDs was similar to that used for aspirin use except that minimal exposure was considered to be 10% of a usual daily dose to obtain an anti-inflammatory effect instead of 80 mg. No trend was noticed when comparing nonusers to recent and chronic NSAID users (Table V). Among chronic users of NSAIDs, the mean daily dose received over the 8-year observation period was above 50% of the usual daily dose to obtain an anti-inflammatory effect in 38% of subjects.
|Timing of exposure1||Cases (n = 2,221)||Controls (n = 11,105)||Age-adjusted OR (95% CI)||Multivariate model OR (95% CI)2|
|Recent exposure3||179||785||1.21 (1.02–1.44)||0.94 (0.79–1.12)|
|Chronic exposure4||292||1,496||1.03 (0.90–1.19)||0.82 (0.71–0.95)|
|Other exposure||532||2,380||1.18 (1.06–1.33)||1.01 (0.90–1.13)|
|Recent exposure||52||281||1.07 (0.79–1.46)||0.88 (0.64–1.20)|
|Chronic exposure||274||1,050||1.51 (1.30–1.76)||1.20 (1.02–1.40)|
|Other exposure||1,144||5,427||1.22 (1.10–1.35)||1.06 (0.95–1.17)|
Interestingly, as shown in Table VI, the risk of prostate cancer in regular aspirin users (on average at least 80 mg per day) throughout the period of 1–8 years before the index date but who had no exposure in the last year before the index date was not different from the risk in nonusers (OR = 1.03, 95% CI 0.84–1.25).
|Timing of exposure1||Cases (n = 2,221)||Controls (n = 11,105)||Age-adjusted OR (95% CI)||Multivariate model OR(95% CI)2|
|Former regular exposure3||134||626||1.13 (0.93–1.38)||1.03 (0.84–1.25)|
|Other exposure||869||4,035||1.14 (1.03–1.25)||0.92 (0.83–1.02)|
|Former regular exposure3||162||785||1.20 (0.99–1.44)||1.02 (0.84–1.23)|
|Other exposure||1,308||5,973||1.27 (1.15–1.40)||1.08 (0.97–1.19)|
All of the above analyses were performed with a drug-exposure measurement that excluded the last 3 months before the index date (3 months lag time). To estimate the impact of protopathic bias, we compared these ORs to those obtained with a drug-exposure measurement that excluded either the last year or the last 2 years before the index date. For aspirin, as in the 3-month lag-time analyses, the ORs obtained with a 1- or 2-year lag time revealed a negative association between prostate cancer risk and drug exposure (results not shown). Also, as we shortened the lag time, the negative association became stronger, suggesting that the 3-month lag-time model was not affected by any significant protopathic bias. For NSAIDs, we did not observe any clear trend in the association between drug exposure and prostate cancer risk for any of the 3 lag-time periods, and shortening the lag-time period did not appear to affect the ORs in any clear direction (results not shown).
In these data, we observed, with regard to the cumulative duration of aspirin use, an absence of trend until 4 years of use but a statistically significant negative trend afterward, with a 34% reduction in prostate cancer risk in men who cumulated 6 or more years of drug use. We also found that the effect of dosage was only apparent in subjects with at least 4 years of use. In this group, compared to nonusers, we observed a 19% and 28% prostate cancer risk reduction when exposed to <325 mg daily and ≥325 mg daily, respectively. It is likely that the group with fewer than 4 years of cumulative use included a high proportion of irregular aspirin users, and this might explain why no association was observed. With regard to the timing of exposure, we noted that, in comparison to nonusers, subjects exposed to a mean daily dose of aspirin of at least 80 mg throughout the 8 years preceding the index date had an 18% reduction in prostate cancer risk. The more recent users (last 4 years before index date) of such a dose had a 7% reduction in prostate cancer risk. However, subjects with regular aspirin use who stopped taking aspirin 1 year before the index date had a risk of prostate cancer comparable to nonusers. We did not observe any association between prostate cancer risk and exposure to NSAIDs other than aspirin.
The population-based case-control design used in our study reduces the possibility of selection bias. Also, ascertaining exposure through a preregistered database precludes recall bias. Nonetheless, other systematic errors could have affected our results. Even though we adjusted for medical contacts in the months preceding the index date, a difference could persist between comparison groups with regard to the probability of having a prostate cancer detected. If this is so, it would most likely underestimate the association measures. Alternatively, earlier mortality may have precluded a diagnosis of prostate cancer in NSAID or aspirin users compared to nonusers. In the prospective study conducted by Roberts et al.,26 however, mortality rates during follow-up were similar among NSAID users, of whom 87% were using aspirin, and nonusers. Likewise, in their study population, there was no difference in preclinical signs of prostate cancer at the beginning of follow-up between NSAID users and nonusers. Our association measures could also be blurred by misclassification of exposure. First, we did not account for drugs obtained without prescriptions. However, since the RAMQ reimbursed all prescribed drugs to senior Quebec citizens, it is unlikely that any significant amounts of drugs were systematically bought over the counter. Second, our exposure measure relied on drug dispensation rather than actual drug consumption, and there could be a significant gap between these 2 realities, particularly for NSAIDs other than aspirin, which are mainly prescribed for the management of temporary pain or inflammation. In the case of aspirin, which is often prescribed for cardiovascular disease prophylaxis, it appears less plausible that drug consumption deviated from that recommended on prescription. This could be the reason why no protective effect was demonstrated for NSAIDs other than aspirin. Third, misclassification of exposure could originate from errors in the RAMQ pharmacist claims database, but it should be of limited importance since the database has been shown to be very reliable.27 These 3 possible sources of misclassification of exposure should equally affect cases and controls. Prostate cancer cases were identified through the Quebec Cancer Registry, which relies solely on hospitalization data. Since prostate cancer investigation and treatment, particularly in those over 70 years of age, do not necessarily involve hospitalization, this registry does not capture all incident prostate cancer cases. This underdeclaration should not bias our results unless it differently affects exposed and nonexposed cases. It is difficult to speculate on factors that could have induced more or less declaration to the Quebec Cancer Registry in the exposed or nonexposed and in those exposed to aspirin compared to those exposed to NSAIDs other than aspirin. Nonetheless, this bias, if present, could not explain the dose–response relationship observed in those chronically exposed to aspirin.
Apart from age and use of antihypertensive drugs, we did not control for any of the established or suspected prostate cancer risk factors. As suggested by Barry,32 men taking low-dose prophylactic aspirin might have other health-related behaviors that attenuate their risk of prostate cancer. They could, e.g., reduce their intake of animal fat or supplement their diet with vitamins or other micronutrients, some of which are under investigation as potential prostate cancer chemopreventive agents. It has also been hypothesized that genetic polymorphisms might be a potential confounder of the association between NSAID exposure and prostate cancer risk.33 However, confounding by race is unlikely in our study because of the characteristics of the Quebec senior population in the middle of the 1990s and of the eligibility criteria that limit the inclusion of immigrants into the study population. Confounding by family predisposition to prostate cancer also appears unlikely because most hereditary prostate cancer is diagnosed before the age of 70.34 Finally, it could be that men took NSAID or aspirin to control symptoms caused by a yet undiagnosed prostate cancer. However, in the presence of such a protopathic bias, the lag-time analysis should have revealed weaker rate ratios with the shorter lag-time periods.
In the interpretation of our findings, certain limits must be recognized. We measured dosage and duration of exposure within a time span of 8 years before index date. Therefore, we could not assess the effect of lifetime cumulative dosage or duration of use. Neither could we evaluate the effect of age at onset of exposure. Also, since a fairly important proportion of the subjects in the higher categories of exposure were already taking aspirin at the beginning of the 8-year observation period (approx. 50%), it is plausible that the protective effect appears after a longer duration of exposure than what we observed. Moreover, we only assessed the effect of NSAID or aspirin exposure on the risk of prostate cancer as a whole. Norrish et al.23 observed a stronger protective effect of NSAIDs on advanced prostate cancer compared to total prostate cancer, suggesting that NSAIDs could act differently on aggressive tumors than on small tumors with slow doubling time.
Other investigators have reported statistically significant inverse associations between NSAID use and prostate cancer. In a community-based cohort study, Roberts et al.26 observed that men aged 50–79 who were exposed daily to NSAIDs had a 55% reduction in the risk of prostate cancer. In men aged 70–79, they found an 83% prostate cancer risk reduction. In this study, 87% of NSAID exposure was attributable to aspirin use. Nelson and Harris25 found that men exposed to at least 1 pill per day of aspirin or ibuprofen had a 66% reduction of prostate cancer risk compared to nonusers. The risk reduction was about the same for men exposed to at least 1 pill per day of prescribed NSAIDs. Prior to these 2 studies, others had observed less impressive negative associations that had not reached statistical significance. Paganini-Hill et al.,22 Schreinemachers and Everson21 and Norrish et al.23 reported a protective effect of NSAID exposure on prostate cancer incidence, with rate ratios ranging from 0.84 to 0.95. Thun et al.20 found an 18% reduction of prostate cancer mortality when comparing a group of men exposed to 16 pills or more per month to the nonexposed, and Bucher et al.24 found that at autopsy analgesic abusers had a reduced risk of prostate cancer of 16% in comparison to nonabusers.
A case-control study observed a statistically significant positive association between NSAID use and prostate cancer.17 This study reported an OR of 1.33 in the group exposed to at least 7 prescriptions during the observation period compared to nonusers. This is the only published study on the association of NSAIDs and prostate cancer that ascertained exposure through a preregistered computerized database, as we have done. Three major methodologic differences could explain the discrepancy between the results of this study and ours. First, they measured exposure during the 12–36 months before the index date, while we measured exposure over the 3–96 months before the index date. Second, they analyzed jointly aspirin and other NSAIDs. In our data, no protective effect was demonstrated when aspirin and NSAIDs were considered together. Third, Langman et al.17 did not apply any correction for detection bias. As suggested by Roberts et al.,26 men taking NSAIDs on a regular basis are more likely to have frequent medical contacts, digital rectal examinations, and prostate-specific antigen measurements than men who do not take NSAIDs. Without any control of this enhanced probability of detection in the exposed, any protective association might be blurred.35 In our data, an inverse association is present only when adjustment is made on a proxy of medical contacts in the months preceding the index date. Also, the 2 recent studies that demonstrated strong protective effects of NSAIDs on prostate cancer used designs that precluded detection bias. In one of them, controls were chosen within a population of men who consulted for prostate cancer screening;25 and in the other, both the exposed and nonexposed were questioned and examined at regular intervals for prostate cancer detection.26 Finally, there are 2 other published epidemiologic studies with results that do not support the hypothesis of a protective effect of NSAIDs on prostate cancer risk. Looking at the association between rheumatoid arthritis and prostate cancer incidence, Gridley et al.18 found a nonsignificant positive association, and Neuget et al.19 observed a rate ratio of 1.60 (95% CI 0.82–3.11) in aspirin users compared to nonusers.
With respect to duration and timing of exposure, our results are in accordance with the published literature on the association between NSAIDs and colorectal cancer. First, drug use needs to be regular and maintained over many years before cancer incidence decreases; and second, tumor growth resumes soon after termination of NSAID treatment.3 Moreover, regular and long-term use of aspirin at doses similar to those recommended for the prevention of cardiovascular disease has been demonstrated to reduce the risk of colorectal cancer.36
In this population-based case-control study, we observed that regular exposure to aspirin maintained over many years, at dosages beneath those usually recommended for an anti-inflammatory effect, was associated with a 20–30% reduction in prostate cancer risk. We also found that 1 year after the end of regular and long-term use of aspirin, no residual protective effect on prostate cancer risk persisted. This might suggest that the drug delays, rather than prevents, cancer development. In our findings, the absence of a protective effect of NSAIDs other than aspirin on prostate cancer could result from misclassification bias. It could also suggest that aspirin exerts a particular pharmacologic effect not shared by the other NSAIDs. Prostate cancer detection can be strongly influenced by care-seeking behaviors, and this might have blurred the results of many studies on the association of NSAIDs or aspirin and prostate cancer. In future studies, emphasis should be placed on the control of such detection bias.
The authors acknowledge the assistance provided by Ms. D. Labrie-Pelletier and Mr. M. Saindon of the RAMQ; Mr. M. Beaupré of the Fichier des tumeurs du Québec; as well as Ms. J. Moisan of l'Unité de recherche en santé des populations and the Faculty of Pharmacy of Laval University. We also thank Dr. S. Suissa, of the Department of Epidemiology and Biostatistics of McGill University, who provided helpful comments and suggestions in the planning of data analyses.
- 28World Health Organization. Anatomical therapeutic classification (ATC) index including defined daily doses (DDD) for plain substances. Geneva: World Health Organization, 2001.
- 31Canadian Pharmacists Association. Compendium of pharmaceutical products, 37th ed. Ottawa: Canadian Pharmacists Association, 2002.