Participants were members of the HPFS, a cohort of 51,529 U.S. male dentists, optometrists, osteopaths, podiatrists, pharmacists and veterinarians, who returned a mailed health questionnaire in 1986. Participants were 40–75 years of age at baseline and the questionnaire included a validated assessment of diet22 and medical diagnoses, including cancer. Questionnaires are mailed biennially to update anthropometric, physical activity, smoking, medication, vitamin, diet (collected every four years) and other lifestyle factors and to identify new cases of disease, including prostate cancer. The follow-up rate exceeds 90%. The conduct of this cohort study and these analyses were approved by the Human Subjects Committee of the Harvard School of Public Health.
Ascertainment of cases
We reviewed the medical records and pathology reports of men who reported prostate cancer on each biennial questionnaire. For all cases, medical charts were abstracted to determine clinical data, including Gleason score, tumor-node-metastasis (TNM) stage23 and PSA levels at the time of diagnosis. Development of bony metastases was ascertained through mailed questionnaires to consenting participants and their treating physicians and was based on the date of a positive bone scan or the date confirmed by the patient's physician. Deaths were identified through the National Death Index, postal system and next of kin, with virtually complete follow-up.24 A prostate cancer death was based on evidence of extensive metastatic disease and no other plausible cause of death and was determined by a study physician through medical record review.
Assessment of aspirin use
Information about aspirin use was collected biennially through mailed questionnaires starting with the questionnaire in 1986 in which men were asked if they currently used aspirin (e.g., Anacin, Bufferin, Alka-Seltzer) two or more times per week. Individual medications or brand names were not ascertained. This information was updated every two years. Additional information on frequency and quantity were collected starting in 1992, when men were asked about the number of adult-strength tablets consumed per day or per week (in categories; men were asked to convert 4 baby aspirin into 1 full-strength tablet). In 1993, a supplementary questionnaire was sent to a sample of 211 regular aspirin users to ascertain reasons for aspirin use between 1986 and 1990 (88% response). The major reasons (nonmutually exclusive) for use were cardiovascular disease, 25.4%; to decrease risk for cardiovascular disease, 58.4%; headaches, 25.4%; joint or musculoskeletal pain, 33.0%; and other reasons, 7.0%.25
Men reported use of aspirin and other NSAIDs only if they consumed these medications at least two times a week. As a result, nonusers in any given cycle were defined as those using medications less than two times a week, which is consistent with previous analyses of this cohort.21, 25, 26 We also evaluated short-term associations by assessing risk in current and formers users. Men were classified as current users if they reported use in the previous two years, former users if they were nonusers in that period but reported previous use, and nonusers if they reported no aspirin use or aspirin use less than 2 days a week since the initiation of the study. Although our main results focus on the most recent questionnaire for frequency and quantity, we also assessed the cumulative average of aspirin use for all previous questionnaires up until the time of diagnosis. In any particular cycle, if a man reported aspirin use but did not report a frequency or quantity, his use was captured for analyses of duration but his frequency and quantity for that period were assigned to missing and therefore not incorporated in the respective analyses.
Collection of covariates
The baseline and biennial questionnaires included updated information on demographic, diet, smoking and other lifestyle factors in the cohort. At baseline, demographic information such as weight and height were collected and weight was updated every two years. Information on physical activity was also collected every two years and MET-hours per week was calculated from a list of activities reported in 1986.27 One met-hour is the metabolic expenditure of sitting at rest for one hour. Family history of prostate cancer was based on the 1990 and 1996 questionnaires when participants were asked if they had a father or any biological brothers diagnosed with prostate cancer. Tobacco use, current smoking status, duration and cigarettes per day were collected biennially. Dietary information was collected from a semiquantitative food-frequency questionnaire, which was updated every 4 years; the methodology, validation and details are described elsewhere.22, 28 History of PSA testing was first asked in 1994 and was updated biennially, with participants being asked if a test had been taken for symptoms or for screening in the prior 2 years.
At baseline, we excluded men with a prior history of cancer (except nonmelanoma skin cancer) and those who died or reported implausible dietary data (outside the range of 800–4,200 kcal per day). We evaluated aspirin use two cycles prior to diagnosis (2-year lag) to avoid a potential bias from more frequent use in undiagnosed cases close to the time of diagnosis. A total of 47,271 men accrued follow-up time starting on the month of the 1988 questionnaire return date and ending on the month of date of diagnosis for cases, date of death from other causes for noncases or January 1, 2006, whichever came first. During follow-up, we excluded cases with a missing date of diagnosis and follow-up time for participants who did not provide information about aspirin use. We calculated duration of aspirin use as the time between return dates of the questionnaires. We focused on simple, updated measurements using the most recent cycle (with the two-year lag) for aspirin frequency, in days per week and quantity, in tablets per week.
Cox proportional hazards regression was used to calculate the hazard ratio (HR) and 95% confidence intervals (CI) while adjusting for age (1-month time intervals), time period (2-year intervals), established risk factors (ethnic and family history) and other covariates shown to be associated with incidence or mortality in HPFS:28 height (<66, 66–67.9, 68–69.9, 70–71.9, 72+ inches), body mass index (<21.0, 21–22.9, 23–24.9, 25–27.4, 27.5–29.9, 30+ kg/m2), smoking (never smoker or quit >10 years, current smoker or quit ≦10 years and <15 cigarettes/day, current smoker or quit ≦10 years and ≥15 cigarettes/day), intake of tomato sauce (<0.25, 0.25–1, 1–2, 2+ servings/week), vitamin D (quintiles), total kilocalories (quintiles, kcal per day), fish (<2/month, 2/month-1/week, >1–<3/week, 3+/week), red meat (quintiles, servings per week), vigorous physical activity (quintiles, hours) and the use of statins (yes/no current user). Covariates were updated every 2 or 4 years. Adjusting for a history of myocardial infarction and diabetes, both associated with aspirin use and PSA screening in our cohort, did not change the estimates and were not included in the final models. Tests for linear trend were conducted by assigning the median value in each category of aspirin use (p < 0.05).
We assessed the HR associated with aspirin use and four prostate cancer endpoints, including total, high-grade (Gleason 8–10), regionally advanced (T3b-T4 or N1 and M0) and lethal disease (M1 at diagnosis or development of bony metastases and/or fatal disease during follow-up). The heterogeneity of prostate cancer includes latent forms that remain clinically dormant and more aggressive forms that progress to metastatic and fatal disease. Studies on aspirin7, 8 and other putative risk factors show varying effects for total and advanced disease and suggest different etiological pathways for different prostate cancer endpoints.28 To evaluate the potential influence of aspirin on these different pathways, we analyzed risks associated with overall prostate cancer and three aggressive sub-types of the disease.
We conducted various sensitivity analyses to address potential confounding due to higher PSA screening rates among aspirin users in our cohort. First, we excluded PSA screen-detected disease (T1c) from overall prostate cancer to address an inflation bias that might arise because aspirin users are more likely to undergo PSA screening than nonusers. In another sub-analysis, we restricted the population to men who had ever received a PSA test to reduce detection bias and ensure a similar screening history in cases and noncases. In a third analysis, we adjusted for PSA screening in the prior 2-year cycle (not the most recent cycle, in which cases would have a higher probability of PSA testing that led to diagnosis). In a fourth sensitivity analysis, we restricted our analysis to men who were screened by PSA in the same 2-year cycle and excluded men who were not screened in that interval to ensure equal opportunities for a prostate cancer diagnosis in any given two-year cycle.
We evaluated whether the association of aspirin on prostate cancer risk varied according to age at diagnosis, body mass index (BMI), fish and red meat intake. Men diagnosed at younger ages tend to have a family history of prostate cancer and exhibit more aggressive forms of the disease, and the influence of aspirin's anti-inflammatory actions may vary according to age. The low-grade inflammatory state of overweight and/or obesity may modulate the biological efficacy of aspirin and evidence from other sites (e.g., endometrial cancer and colorectal adenoma) suggests that the impact of aspirin on risk varies by BMI.29, 30 Fish is inversely31, 32 and red meat is positively associated with prostate cancer in many but not all studies,33 and both fatty fish and red meat intake are important dietary sources influencing the AA pathway for COX-1 and COX-2 enzymatic activity. We used the Wald statistic and likelihood ratio test to determine statistical significance at the p < 0.05 level.