EURURO Vol. 72 No. 5

822 860

1. Introduction
Regular aspirin probably protects against some malignan-
cies . Meta-analyses and a recent study suggest
modest decreases in overall and advanced prostate cancers
(PCs), although some studies reported no benefit . An
international consensus concluded that the potential effects
of aspirin on PC warrant further evaluation Evidence for
lethal PC is particularly limited.
The chemopreventive effects of aspirin may result from
antiplatelet or anti-inflammatory properties . Aspi-
rin inhibits cyclo-oxygenase enzymes, which are over-
expressed in several malignancies and are implicated
in cell proliferation, angiogenesis, and cancer progression
Platelet-tumor signaling may play a key role in
metastatic initiation, and platelet depletion decreases
metastatic burden in animal models .
Most low-grade and early-stage PCs are indolent ,
so the significance of overall incidence as an outcome is
questionable. Therefore, we investigated regular aspirin and
lethal PC.
2. Participants and methods
The Physicians’ Health Study began in 1981/82 as a randomized,
placebo-controlled trial of aspirin and b-carotene for prevention of
cardiovascular disease and cancer (n = 22 071) The participants
were male physicians aged 40–84 yr without a history of cancer (except
for non-melanoma skin cancer), myocardial infarction (MI), stroke, or
transient ischemic attack. Participants were randomized using a
2 2 factorial design to aspirin 325 mg, b-carotene 50 mg, both, or
double placebo (all taken every other day). The aspirin trial ended in
1988 because of a 44% reduction in first MI in the aspirin group.
Thereafter, most participants elected to receive complimentary un-
blinded aspirin. The b-carotene component continued up to 1995; no
associations with cancer were observed .
We conducted two related analyses. Our risk analysis investigated
prediagnostic aspirin and the risk of lethal PC among all participants who
provided sufficient aspirin information (n = 22 037). Our survival
analysis investigated postdiagnostic aspirin and survival among
participants initially diagnosed with nonmetastatic PC between enrol-
ment and 2009 ( n = 3462).
The primary exposure was regular aspirin ( >3 d/wk for 1 yr),
defined a priori according to the study design. Aspirin use was
ascertained from baseline until 2009. Annual questionnaires asked
how many days/year participants missed study pills, and days/year of
personal aspirin use. After the aspirin trial ended in 1988, participants
reported days/year of complimentary study and personal aspirin use.
After diagnosis, PC patients reported regular aspirin use (yes/no) on PC
follow-up questionnaires up to 2015.
The primary outcome was lethal PC (metastatic PC or death from PC),
chosen a priori on the basis of clinical significance and our hypothesis
that aspirin would be associated with lethal PC. Secondary risk analysis
outcomes included overall mortality, overall PC, high-grade PC (Gleason
8–10), and advanced PC (TNM stage T3b, N1, or M1 at diagnosis).
Secondary survival analysis outcomes included PC mortality and overall
mortality.
We recorded stage (81% complete), Gleason score (based on biopsy,
82% complete), prostate-specific antigen (PSA) at diagnosis, and
treatment(s) from self-reports and medical records. We conducted
national death index searches to confirm the date and cause of death for
the whole cohort up to 2009 and for PC patients up to 2015. Cause of
death was assigned by a three-physician endpoint committee after
review of death certificates, medical records, and information from
family. Follow-up is >96% complete for PC incidence and >99% for
mortality.
2.1. Statistical analysis
2.1.1. Main analyses
Descriptive statistics characterized the study population. We used Cox
proportional-hazards regression models to estimate age and multivari-
ate-adjusted hazard ratios (HRs) and 95% confidence intervals (CIs).
We categorized aspirin use as current (within 1 yr), past, and never
from baseline to PC diagnosis, death, or 2009 in the risk analysis, and
from diagnosis to death or 2015 in the survival analysis. We examined
duration among current users (1–4 vs 5 yr) and time since stopping
among past users ( 10, 8–9, 6–7, 4–5, 2–3 yr). If missing, the most recent
aspirin report was carried forward for one questionnaire cycle, and then
set to missing.
Risk analysis follow-up was from 1981/82 baseline to event, death, or
end of follow-up (2009, the last aspirin assessment for all participants).
Aspirin was updated throughout follow-up for non-cases and until PC
diagnosis for cases. We adjusted risk analysis models for baseline age (in
years), race (white vs other), body mass index (BMI; in kg/m 2), height (in
inches), smoking (current, former [quit 10 yr ago], never/remote [quit
>10 yr ago]), hypertension (yes vs no), and type 2 diabetes (yes vs no).
Survival analysis follow-up time was from PC diagnosis until an
event, death, or the end of follow-up (2015). Time-varying aspirin was
updated approximately annually after diagnosis. We adjusted for at-
diagnosis age, race, Charlson comorbidity index (0, 1–2, >2 comorbid-
ities) , BMI, smoking (current, former, never/remote), hypertension
(yes vs no), and type 2 diabetes (yes vs no), stage (T1–2, T3, T4/N1), PSA
(none, <10, 10–20, > 20 ng/ml), Gleason ( 6, 7, 8–10), and treatment
(radical prostatectomy, radiation, other/none).
2.1.2. Secondary analyses
Aspirin may confer different effects at different stages of PC progression.
Because cases were generally diagnosed later in disease progression
before PSA screening, we stratified both analyses by year of diagnosis:
pre-PSA era (<1992) and PSA-era ( 1992).
Cancer, even when undiagnosed, might influence aspirin use (reverse
causation). Thus, in the risk analysis we lagged aspirin exposure by both
2 yr and 4 yr; for example, we applied 1986 aspirin use to the 1988 time
period (2-yr lag) and the 1990 time period (4-yr lag), thus using a prior
exposure uninfluenced by possible underlying disease. In the survival
analysis, we stopped updating aspirin 3 yr after diagnosis; when PC
progresses, 3.5 yr is the average time from treatment to biochemical
recurrence, a precursor to lethality Stopping the updating
prevents this progression from influencing exposure status. Finally, we
performed an intention-to-treat (ITT) analysis based on original randomi-
zation.
The proportional hazards assumption held throughout. We used SAS
version 9 (SAS Institute, Cary, NC, USA). Two-sided p values <0.05
defined statistical significance.
3. Results
3.1. Risk analysis
From 1981/82 to 2009, 502 participants developed lethal
disease. Baseline characteristics were similar among regular
aspirin users and nonusers ).
Compared to never use, past regular aspirin use was
associated with a lower risk of lethal PC (HR 0.54, 95% CI
E U R O P E A N U R O L O G Y 7 2 ( 2 0 1 7 ) 8 2 1 – 8 2 7 822