Articles

Eect of daily aspirin on risk of cancer metastasis: a study of

incident cancers during randomised controlled trials
Peter M Rothwell, Michelle Wilson, Jacqueline F Price, Jill F F Belch, Tom W Meade, Ziyah Mehta

Summary
Background Daily aspirin reduces the long-term incidence of some adenocarcinomas, but eects on mortality due to Published Online
some cancers appear after only a few years, suggesting that it might also reduce growth or metastasis. We established March 21, 2012
DOI:10.1016/S0140-
the frequency of distant metastasis in patients who developed cancer during trials of daily aspirin versus control. 6736(12)60209-8
See Online/Comment
Methods Our analysis included all ve large randomised trials of daily aspirin (75 mg daily) versus control for the DOI:10.1016/S0140-
prevention of vascular events in the UK. Electronic and paper records were reviewed for all patients with incident 6736(11)61654-1
cancer. The eect of aspirin on risk of metastases at presentation or on subsequent follow-up (including post-trial See Online/Articles
follow-up of in-trial cancers) was stratied by tumour histology (adenocarcinoma vs other) and clinical characteristics. DOI:10.1016/S0140-
6736(11)61720-0
See Online/Articles Lancet Oncol
Findings Of 17 285 trial participants, 987 had a new solid cancer diagnosed during mean in-trial follow-up of
DOI:10.1016/S1470-
65 years (SD 20). Allocation to aspirin reduced risk of cancer with distant metastasis (all cancers, hazard ratio [HR] 2045(12)70112-2
064, 95% CI 048084, p=0001; adenocarcinoma, HR 054, 95% CI 038077, p=00007; other solid cancers, Stroke Prevention Research
HR 082, 95% CI 053128, p=039), due mainly to a reduction in proportion of adenocarcinomas that had meta- Unit, Nueld Department of
static versus local disease (odds ratio 052, 95% CI 035075, p=00006). Aspirin reduced risk of adenocarcinoma Clinical Neuroscience,
University of Oxford, Oxford,
with metastasis at initial diagnosis (HR 069, 95% CI 050095, p=002) and risk of metastasis on subsequent
UK (Prof P M Rothwell FMedSci,
follow-up in patients without metastasis initially (HR 045, 95% CI 028072, p=00009), particularly in patients M Wilson MSc, Z Mehta PhD);
with colorectal cancer (HR 026, 95% CI 011057, p=00008) and in patients who remained on trial treatment up Centre for Population Health
to or after diagnosis (HR 031, 95% CI 015062, p=00009). Allocation to aspirin reduced death due to cancer in Sciences, University of
Edinburgh, Edinburgh, UK
patients who developed adenocarcinoma, particularly in those without metastasis at diagnosis (HR 050, 95% CI
(J F Price MD); The Institute of
034074, p=00006). Consequently, aspirin reduced the overall risk of fatal adenocarcinoma in the trial popula- Cardiovascular Research,
tions (HR 065, 95% CI 053082, p=00002), but not the risk of other fatal cancers (HR 106, 95% CI 084132, Vascular and Inammatory
p=064; dierence, p=0003). Eects were independent of age and sex, but absolute benet was greatest in smokers. Diseases Research Unit,
University Division of Medicine
A low-dose, slow-release formulation of aspirin designed to inhibit platelets but to have little systemic bioavailability
and Therapeutics, Ninewells
was as eective as higher doses. Hospital and Medical School,
Dundee, UK
Interpretation That aspirin prevents distant metastasis could account for the early reduction in cancer deaths in trials (Prof J F F Belch MD); and
London School of Hygiene and
of daily aspirin versus control. This nding suggests that aspirin might help in treatment of some cancers and
Tropical Medicine, University
provides proof of principle for pharmacological intervention specically to prevent distant metastasis. of London, London, UK
(Prof T W Meade FRS)
Funding None. Correspondence to:
Prof Peter M Rothwell, Stroke
Prevention Research Unit,
Introduction the reduction in mortality due to some cancers occurred
Nueld Department of Clinical
Cancer is the second most common cause of death within 23 years after randomisation and seemed too Neuroscience, University of
worldwide and 5 million new cases are diagnosed each quick to be accounted for by an eect only on carcino- Oxford, John Radclie Hospital,
year in Europe and the USA alone.13 There has been genesis or early cancer growth.12 Third, there was a Headington, Oxford OX3 9DU, UK
peter.rothwell@clneuro.ox.
evidence from animal studies and from observational reduction in deaths due to metastatic cancer with
ac.uk
studies in humans that daily aspirin might be eective in unknown primary.12 We therefore hypothesised that
preventing several common cancers,48 which has recently these shorter-term eects of aspirin could be due to the
been conrmed by follow-up of patients originally prevention of distant metastasis, consistent with
included in randomised trials of daily aspirin versus experimental evidence in animals that platelets play an
control in the prevention of vascular events.912 Allocation important part in blood-borne metastases,1316 which are
to aspirin in these trials resulted in a 40% reduction in reduced by aspirin,1519 and with observational studies in
cancer deaths from 5 years onwards,12 with sustained humans that showed that aspirin use is associated with
reductions in deaths due to certain cancers at 20-year reduced distant metastasis and with decreased rates of
follow-up.911 recurrence in patients with common adenocarcino-
These recent reports of randomised trials suggest that mas.7,2024 Since the results in animal studies might not
aspirin might also have short-term eects on outcome in apply to humans and because the observational studies
patients who develop cancer while on treatment. First, are subject to several potential biases, we studied
the eect of aspirin on death due to cancer was greater metastasis at initial diagnosis and during subsequent
than the apparent eect on incidence of cancer.12 Second, follow-up in all participants with a new diagnosis of

www.thelancet.com Published online March 21, 2012 DOI:10.1016/S0140-6736(12)60209-8 1

 Articles

cancer in all ve large completed randomised trials of acute vascular events, and trials in the treatment or
aspirin versus control in prevention of vascular events in secondary prevention of cancer or colonic polyps.
the UK.2529
Procedures
Methods The investigators from all ve eligible trials2529 had
Trials collaborated on previous analyses of data for risk of
Recent literature searches identied all randomised cancer.1012 Design of the trials and the methods of follow-
See Online for appendix trials of aspirin (any dose) versus no aspirin in the up are given in appendix p 1. Cancers diagnosed during
absence of another antiplatelet agent, or aspirin (any the trials had been recorded by face-to-face or question-
dose) versus no aspirin in the presence of another naire follow-up, supported by centralised national death
antithrombotic agent (if balanced in the aspirin and no certication and cancer registration (appendix p 1). All
aspirin groups) in prevention of vascular events.11,12 We electronic and archived paper records collected during
restricted inclusion to trials done in the UK because of the trials for patients in whom a new cancer was
the availability of reliable and well validated centralised diagnosed were re-reviewed, blinded to randomised
death certication and cancer registration to ensure treatment allocation, including all records of baseline
identication of incident cancers not recorded on face- and follow-up assessments during the trials and any
to-face follow-up and to determine the post-trial outcome outcome summaries relating to cancers. All death
of cancers diagnosed during the trials. We excluded trials certicates were also reviewed, blinded to treatment
with less than ten incident cancers recorded during allocation, for patients in whom a cancer was diagnosed
follow-up, trials of short-term (90 days) treatment of during the trials, including those for post-trial deaths in
four trials.2527,29
Data were extracted, where available, for date of
Cancers/person-years HR (95% CI) p value diagnosis of solid cancers during the trial (excluding
Aspirin Control non-melanoma skin cancer), site of primary cancer,
All solid cancers* investigations, tumour staging (where available), hist-
Any metastasis 182/60 560 211/47 703 073 (060089) 0002 ology, and treatment. Metastasis at initial diagnosis of
Denite distant metastasis 92/60 656 118/47 830 064 (048084) 0001
cancer or on subsequent follow-up was categorised
Metastasis with site unspecied 90/60 621 93/47 836 085 (063114) 027
(blinded to treatment allocation) in four groups:
Metastasis at initial diagnosis 131/60 666 140/47 891 079 (062101) 006
(1) denite blood-borne distant metastasis, dened as
Metastasis on follow-up 51/60 611 71/47 775 060 (042086) 0006
metastasis to distant tissues clearly stated in records or
death certicates (ie, specic mention of secondary
Local disease only 227/60 079 155/47 477 124 (101153) 0040
tumour in liver, lung, bone, brain, or other distant tissue)
Metastasis status unknown 101/60 604 111/47 855 077 (058101) 0056
and not including spread of cancer likely to have occurred
Adenocarcinoma
by local invasion or via lymphatics (eg, lung cancer to
Any metastasis 99/60 625 135/47 761 060 (046078) 00001
mediastinum) or by peritoneal seeding in abdominal or
Denite distant metastasis 52/60 674 76/47 853 054 (038077) 00007
pelvic cancers; (2) metastasis without specied site (if no
Metastasis with site unspecied 47/60 668 59/47 871 069 (047101) 005
data were recorded for the specic site or sites of
Metastasis at initial diagnosis 70/60 681 82/47 909 069 (050095) 002
metastasis, but the cancer was described as metastatic,
Metastasis on follow-up 29/60 662 53/47 811 046 (029073) 00009
or the patient was said to have carcinomatosis at
Local disease only 122/60 352 86/47 679 124 (094163) 014
presentation, subsequent follow-up, or on death certi-
Metastasis status unknown 44/60 654 43/47 913 088 (057134) 055
cation); (3) no evidence of distant metastasis (ie, local
Non-adenocarcinoma disease), if the cancer was apparently localised at
Any metastasis 83/60 653 76/47 906 096 (070132) 081 diagnosis and was treated without evidence of distant
Denite distant metastasis 40/60 700 42/47 941 082 (053128) 039 spread during subsequent follow-up, or if there was
Metastasis with site unspecied 43/60 670 34/47 928 113 (072178) 059 evidence of local or regional spread only (eg, lung cancer
Metastasis at initial diagnosis 61/60 703 58/47 946 095 (066136) 078 to mediastinum or ribs; spread of prostate or gynae-
Metastasis on follow-up 22/60 667 18/47 927 100 (053188) 100 cological cancer within the pelvis; breast cancer to
Local disease only 105/60 444 69/47 761 124 (091169) 016 axillary lymph nodes; colorectal cancer to draining lymph
Metastasis status unknown 57/60 667 68/47 905 070 (049100) 005 nodes, etc) at presentation or during follow-up; and
Person-years are for time from randomisation to diagnosis of cancer (rather than time to metastasis) because there is
(4) unknown, when there were insucient data, most
no time to metastasis for cancers with local disease only or unknown metastasis status. However, HRs for metastatic commonly in the case of a rapidly fatal cancer for which
cancer based on time from randomisation to metastasis (rather than time to diagnosis) dier only in the third decimal little clinical information was available (often lung or
place. HR=hazard ratio from a Cox regression stratied by trial. Dierence in eect of treatment on risk of any
pancreatic cancer).
metastatic disease versus local disease: *p=00007; p=0002; p=022. Dierence in eect of treatment on risk of any
metastatic disease in adenocarcinomas versus non-adenocarcinomas: p=003.
Statistical analysis
Table: Pooled analysis of ve trials of aspirin versus control showing the eect of allocated treatment on SPSS (version 20) was used for all analyses. All analyses
the risk of in-trial incident cancers stratied by metastatic status and by histological type
were by intention to treat unless stated otherwise.

2 www.thelancet.com Published online March 21, 2012 DOI:10.1016/S0140-6736(12)60209-8

 Articles

Haematological cancers were excluded from analyses of
metastasis, as were primary brain cancers (unless
otherwise specied) because of the rarity of extracranial
metastasis. In each trial, eects of aspirin on risks of
incident cancer, fatal incident cancer, and cancer with
distant metastasis were expressed as odds ratios (ORs)
with 95% CI. A sensitivity analysis was done in which
fatal cancers for which metastasis status was unknown
were assumed to be metastatic. In the absence of
signicant heterogeneity in eects across trials (xed-
eects meta-analysis), individual patient data were
pooled, but tests of statistical signicance of all analyses
of risk and survival (log-rank test) and Cox models in the
pooled data were stratied by trial. Two sets of analyses
of the pooled data were done.
First, in all randomised participants, we established
the eect of aspirin on the risk of cancer with denite
distant metastasis, with stratication by site of metastasis
(lungs, liver, bone, brain, and multiple or other), and on
risk of cancer with potentially distant metastasis
(denite distant or site-unspecied). These analyses
were based on time from randomisation to metastasis
and take account of any eects of aspirin on incidence
of cancer as well as on metastasis. We looked separately
at the risk of cancer in which no metastases were evident
at initial diagnosis or less than 1 year thereafter, but in
Role of the funding source
The study was unfunded and was independent of any
commercial interest.
Results
Trials
Of the ve eligible randomised trials of aspirin versus
control (appendix p 1), two had been done in
primary prevention of vascular disease,25,27 one in
secondary prevention after recent vascular events,26 and
two in groups with asymptomatic peripheral arterial
disease (one in patients with diabetes;28 one in individuals
with low ankle brachial index on screening29). Among the
17 285 participants in the ve trials, there were
1101 incident cancers and 563 deaths due to incident
cancer during trial follow-up. Allocation to aspirin
reduced the incidence of new cancers during the trials
(OR 088, 95% CI 078099, p=004, phet=044; appen-
dix p 2), but tended to have a greater eect on the risk of
A Distant metastasis
30 Aspirin
HR 064 (95% CI 048084), p=0001
Control
25
20
Risk (%)

which metastases were diagnosed on subsequent follow-

15
up. To establish the latency of any eect of aspirin on
risk of metastasis, this analysis was done separately for 10
time from randomisation to diagnosis of cancer and
time to diagnosis of metastasis. 05
Second, in all participants who developed incident
0
cancer during the trials, we established the eect of 0 2 4 6 8 10 12
allocation to aspirin on the proportion of cancers that Number at risk
Aspirin 9919 9663 9312 8820 8071 7283 6032
had metastases (distant or site-unspecied) at presen- Control 7380 7207 6981 6587 5954 5255 4102
tation or follow-up (including post-trial follow-up) versus
local disease only, the risk of metastasis during follow-up B Distant or unspecied metastasis
(time from diagnosis of cancer to diagnosis of metastasis) 30
HR 073 (95% CI 059089), p=0002
in participants in whom no metastases were evident at
initial diagnosis or less than 1 year thereafter, and survival 25

(for death due to cancer; and for any death) after diagnosis
20
of cancer.
Risk (%)

We stratied analyses by histological type of solid 15

cancer (adenocarcinoma vs non-adenocarcinoma) to
establish whether an eect on metastasis might account 10
for the particular eects of aspirin on deaths due to
05
adenocarcinomas (histologically proven or primary
cancers in which adenocarcinomas predominate 0
ie, stomach, small bowel, pancreas, bileduct, colon, 0 2 4 6 8 10 12
rectum, breast, uterus, ovary, and prostate) in our Years to metastasis
Number at risk
previous studies11,12 and in observational studies.2024 Aspirin 9919 9659 9310 8818 8070 7282 6032
Control 7380 7206 6981 6583 5952 5255 4102
Analyses were also done for common individual cancers
and in relation to sex, median age (<60 vs 60 years), Figure 1: The eect of aspirin on risk of metastasis due to any incident cancer
and current smoking. Statistical signicance of these diagnosed during ve trials of aspirin versus control
Analysis is based on time from randomisation to diagnosis of metastasis during
subgroup eects was established with interaction terms or after the trials. Part A shows denite site-specic distant metastasis and
in a Cox model for time from randomisation to part B also includes metastatic cancers in which the site of the metastasis was
diagnosis of metastasis. not specied. HR=hazard ratio from a Cox regression stratied by trial.

www.thelancet.com Published online March 21, 2012 DOI:10.1016/S0140-6736(12)60209-8 3

 Articles

Lung metastasis Liver metastasis

05 Aspirin 10
HR 034 (95% CI 014083), p=0018 Control HR 066 (95% CI 042103), p=007

04 08

03 06
Risk (%)

02 04

01 02

0 0
0 2 4 6 8 10 12 0 2 4 6 8 10 12
Number at risk
Aspirin 9919 9670 9317 8824 8076 7286 6032 9919 9670 9315 8825 8076 7286 6032
Control 7380 7216 6988 6591 5960 5259 4102 7380 7214 6988 6593 5960 5260 4102

Bone metastasis Metastasis to other or multiple sites

05
HR 117 (95% CI 069201), p=056 HR 036 (95% CI 018070), p=0003

04

03
Risk (%)

02

01

0
0 2 4 6 8 10 12 0 2 4 6 8 10 12
Number at risk
Aspirin 9919 9665 9313 8821 8071 7283 6032 9919 9669 9317 8825 8076 7286 6032
Control 7380 7213 6987 6591 5958 5260 4102 7380 7214 6988 6594 5960 5259 4102
Figure 2: The eect of aspirin
Brain metastasis Any malignant brain tumour
on risk of denite distant
metastasis due to any
05 HR 065 (95% CI 034124), p=019 HR 056 (95% CI 033095), p=003
incident cancer diagnosed
during ve trials of aspirin
versus control, stratied by
04
site of metastasis
Analysis is based on time from
randomisation to diagnosis of
03
metastasis during or after the
Risk (%)

trials. The sensitivity analysis

of the eect of aspirin on risk
02
of all malignant brain tumours
is included because of the
clinical inaccuracy in
01
dierentiating between a
primary brain tumour and a
solitary metastasis, particularly
0
if a biopsy was not undertaken. 0 2 4 6 8 10 12 0 2 4 6 8 10 12
29 patients in the analysis of Years to metastasis Years to metastasis
other or multiple metastasis Number at risk
are also included in analyses of Aspirin 9919 9671 9317 8825 8076 7286 6032 9919 9671 9317 8825 8076 7286 6032
Control 7380 7216 6991 6593 5959 5261 4102 7380 7214 6988 6590 5956 5258 4100
site-specic metastasis.
HR=hazard ratio from a Cox
regression stratied by trial.

4 www.thelancet.com Published online March 21, 2012 DOI:10.1016/S0140-6736(12)60209-8

 Articles

death due to incident cancer during the trials (OR 077,
95% CI 065091, p=0002, phet=044; appendix p 2).
Of the 1101 incident cancers, 94 were haematological
and 20 were primary brain tumours, leaving 987 for
analysis of metastasis (table). In 212 (21%, including
84 lung cancers), there was either clinical uncertainty
about the presence of metastasis or insucient infor-
mation in the trial records, often because death occurred
quickly after diagnosis (median time 054 years,
IQR 010145). The proportion of such cases did not
dier between the aspirin and control groups (table) or
between the trials (p=019). Of the 775 eligible incident
solid cancers in which metastasis status was known,
382 (49%) had no evidence of distant metastasis at
presentation or on follow-up, 183 (24%) had metastasis
without specied site (ie, uncertain whether denitely
distant), and 210 (27%) had denite distant metastasis at
presentation or on follow-up.
Overall risk of metastatic cancer
The table shows the eect of aspirin on risk of in-trial
incident cancer, stratied by metastasis status. Aspirin
reduced the risk of cancer with denite distant metastasis
(hazard ratio [HR] 064, 95% CI 048084, p=0001;
gure 1, table) and increased the risk of cancer with local
disease only (HR 124, 95% CI 101153, p=004; table).
The dierence in eect of aspirin on risk of metastatic
cancer versus risk of local cancer was signicant
(p=00007). The reduction in risk of cancer with distant
metastasis remained and was consistent across all ve

A Excluding cancers with metastasis at initial diagnosis

04 Aspirin
Control
HR 046 (95% CI 029073), p=00009 HR 046 (95% CI 029073), p=00009

03
Risk (%)

02

01

0
0 2 4 6 8 0 5 10 15 20
Number at risk
Aspirin 9919 9470 8656 6727 1632 9919 9101 7286 5214 3453
Control 7380 7132 6631 5285 1620 7380 6825 5257 3498 2179

B All cancers
30

HR 060 (95% CI 046078), p=00001 HR 060 (95% CI 046078), p=00001

25

20
Risk (%)

15

10

05

0
0 2 4 6 8 0 5 10 15 20
Years to diagnosis Years to metastasis
Number at risk
Aspirin 9919 9463 8652 6721 1630 9919 9094 7283 5214 3453
Control 7380 7125 6621 5282 1619 7380 6814 5256 3498 2179

Figure 3: Eect of aspirin on risk of any adenocarcinoma with metastasis (denite distant or site unspecied) and risk of adenocarcinoma in which
metastases were only diagnosed on follow-up after the initial diagnosis of cancer (ie, excluding cases with metastasis at presentation)
Analysis is based on time from randomisation to initial diagnosis of cancer during the trials (left-hand side) or time from randomisation to diagnosis of metastasis during
or after the trials (right-hand side). HRs for the analysis of years to diagnosis dier from those for years to metastasis, but only in the third decimal place. HR=hazard ratio
from a Cox regression stratied by trial.

www.thelancet.com Published online March 21, 2012 DOI:10.1016/S0140-6736(12)60209-8 5

 Articles

trials (meta-analysis, OR 075, 95% CI 063089,
p=0001, phet=023; appendix p 3) when fatal cancers for
which metastasis status was unknown were assumed also
to have been metastatic. There was no signicant
heterogeneity in the eect of aspirin on risk of distant
metastases by site (p=010; gure 2), with fewer metastases
to lung (n=24; HR 034, 95% CI 014083), liver (n=79;
HR 066, 95% CI 042103), brain (metastasis, n=38;
HR 065, 95% CI 034125; any malignant brain tumour,
n=58; HR 056, 95% CI 033095), and to other or
multiple sites (n=41; HR 036, 95% CI 018070),
although metastases to bone were not reduced (n=57;
HR 117, 95% CI 069200, p=056).
In the 775 participants with an incident in-trial solid
cancer for which metastasis status was known, cancers
were less likely to be metastatic (at presentation or
follow-up) in the aspirin versus control groups
All randomised patients
100
HR 045 (95% CI 028072), p=00009
80
60
Risk (%)
(182/409 vs 211/366; OR 059, 95% CI 044078,
p=00003; appendix p 4). This eect was signicant for
colorectal cancer (OR 036, 95% CI 018074, p=0005)
and there were similar but non-signicant trends for
other adenocarcinomas (appendix p 4), with a highly
signicant eect on all adenocarcinomas combined
(OR 052, 95% CI 035075, p=00006). However,
there was no signicant eect of aspirin in non-
adenocarcinomas (appendix p 4).
In all randomised participants, the reduction in risk of
metastatic cancer by aspirin was also greater (interaction
p=003) for adenocarcinoma (all metastasis, OR 060,
95% CI 046078, p=00001; table, appendix p 5) than
for non-adenocarcinoma (OR 096, 95% CI 070132,
p=081). Although there were no signicant interactions
between the relative eect of aspirin on the risk of
development of cancer with metastases (distant or
unspecied) at diagnosis or follow-up and age, sex, and
smoking, the absolute benet of aspirin was somewhat
larger in current smokers (dierence, p=008; appendix
Aspirin
Control
pp 67). This interaction with smoking was most
signicant in women (dierence, p=0055), with no
reduction in risk of metastatic cancer in non-smokers
(HR 112, 95% CI 065194), albeit based on small
numbers (aspirin, 28/9784 person-years vs 24/9447).

Time from randomisation to diagnosis of adeno-

40 carcinoma was similar in the aspirin and control groups
(mean dierence 006 years, SD 037, p=074). In
20 patients with colorectal cancer with formal staging at rst
diagnosis, the proportion of advanced cancers (stage III
0 or IV) was lower in the aspirin groups than in control
0 2 4 6 8 10 12
Number at risk
groups (26/55 vs 47/73; p=005).
Aspirin 140 102 73 52 39 33 30 The eect of aspirin on risk of adenocarcinoma
Control 138 96 63 41 29 20 13 with no metastasis at initial diagnosis but in which meta-
Patients on trial treatment at diagnosis stases developed on subsequent follow-up (HR 046,
100 95% CI 029073, p=00009; table) tended to be larger
HR 031 (95% CI 015062), p=00009 than the eect on adenocarcinoma with metastasis
80 identied at presentation (HR 069, 95% CI 050095,
p=002), although no overall eect of aspirin was evident
60
until about 2 years after randomisation (gure 3). How-
Risk (%)

ever, analysis of time from randomisation to diagnosis

40
of adenocarcinomas in which metastasis presented later
suggested that aspirin reduced the risk of subsequent
metastasis even in cancers diagnosed shortly after
20
randomisation (gure 3).
0
0 2 4 6 8 10 12 Outcome after diagnosis of cancer
Number at risk
Years from diagnosis In patients with adenocarcinoma in whom metastases
Aspirin 70 56 41 34 26 21 20 were not found at initial diagnosis (gure 4), the risk of
Control 85 67 49 32 23 17 11
later metastasis was reduced in the aspirin groups
(HR 045, 95% CI 028072, p=00009), particularly in
Figure 4: Risk of metastasis (denite distant or site unspecied) on
follow-up after diagnosis of incident adenocarcinoma during the trials in patients with colorectal cancer (HR 026, 95% CI
patients in whom metastases were not found at initial diagnosis, stratied 011057, p=00008) and in patients still on trial
by whether or not they were still taking allocated trial treatment when the treatment at diagnosis (all adenocarcinomas, HR 031,
cancer was rst diagnosed
95% CI 015062, p=00009; colorectal cancer, HR 013,
Analysis is based on years from initial diagnosis of cancer during the trials to
diagnosis of metastasis during or after the trials. HR=hazard ratio from a Cox 95% CI 003056, p=0007; other adenocarcinomas,
regression stratied by trial. HR 047, 95% CI 021105, p=006). Results were

6 www.thelancet.com Published online March 21, 2012 DOI:10.1016/S0140-6736(12)60209-8

 Articles

similar after exclusion of the BDAT trial in which
treatment allocation was not blinded.
Overall survival rates after diagnosis of adenocar-
cinoma tended to be higher in the trials of low-dose
aspirin than high-dose aspirin (gure 5; appendix p 8),
irrespective of randomised treatment, probably reecting
the dierent eras in which they were done. However, in
both sets of trials, survival rates were higher in the aspirin
groups than in the control groups (death due to cancer,
HR 071, 95% CI 057090, p=0004, gure 5; death
due to any cause, HR 081, 95% CI 065100, p=005,
appendix p 8). This reduction in case-fatality was greatest
in patients without metastasis at initial diagnosis (death
due to cancer HR 050, 95% CI 034074, p=00006,
gure 5; death due to any cause, HR 069, 95% CI
049097, p=003, appendix p 8).
Reliable estimation of eects of aspirin on case-fatality
in individual primary cancers was limited by small
numbers, but there was some evidence of reduced case-
fatality in the three most common adenocarcinomas
(appendix p 9), again driven mainly by eects in cases
without metastasis at rst diagnosis (colorectal, OR 027,
95% CI 011066, p=0004; breast, OR 016, 95% CI
002119, p=007; and prostate, OR 034, 95% CI
012099, p=0049). Aspirin also reduced the risk of
death due to metastatic cancer in which the primary was
unknown (HR 044, 95% CI 021092, p=003). As a
consequence of the reduced case-fatality, the eect of

High-dose aspirin trials (197886) Low-dose aspirin trials (19892008)

All cases (pooled HR 071, 95% CI 057090, p=0004)
10 Aspirin
Control

08

HR 070 (95% CI 046106), p=009

06
Survival

04

02
HR 072 (95% CI 055095), p=002

0
0 5 10 15 0 5 10 15
Number at risk
Aspirin 86 27 15 12 169 51 21 9
Control 42 5 5 3 222 65 20 6

Cases without metastasis at diagnosis (pooled HR 050, 95% CI 034074, p=00006)

10

08
HR 052 (95% CI 026105), p=007

06
Survival

04

02
HR 049 (95% CI 030079), p=0004

0
0 5 10 15 0 5 10 15
Years from diagnosis Years from diagnosis
Number at risk
Aspirin 42 21 14 12 98 48 21 9
Control 18 4 4 3 120 58 20 6

Figure 5: Eect of aspirin on risk of death due to cancer after diagnosis of incident adenocarcinoma in ve randomised trials of aspirin versus control
stratied by dose (<300 mg vs 300 mg daily) of aspirin (and hence the era in which the trials were done) in all cases and in cases without metastases
(denite distant or site unspecied) evident at the time of initial diagnosis
Analysis is based on time from diagnosis of cancer during the trials to death during or after the trials. HR=hazard ratio from a Cox regression stratied by trial.

www.thelancet.com Published online March 21, 2012 DOI:10.1016/S0140-6736(12)60209-8 7

 Articles
aspirin on the overall risk of fatal adenocarcinoma tended
to be greater than that on all incident adenocarcinoma
(appendix p 10). Since aspirin had no eect on incidence
or case-fatality of non-adenocarcinomas, the overall eect
of allocation to aspirin on risk of fatal adenocarcinoma
(HR 065, 95% CI 053082, p=00002; appendix p 10)
was signicantly greater (interaction p=0003) than that
on fatal non-adenocarcinoma (HR 106, 95% CI
084132, p=064).
Discussion
Daily aspirin reduces the incidence of colorectal cancer
after a delay of 810 years9,10 and reduces mortality due
to several other adenocarcinomas after a similar delay,11
consistent with the latency expected before the clinical
manifestation of an eect on early carcinogenesis.
However, the accompanying analysis of data from
51 trials of aspirin versus control showed that aspirin
reduced mortality from all cancer after only 5 years,
with even earlier reductions in deaths due to colorectal
cancer and cancers with unknown primary site,12
suggesting an eect on cancers that were already
present, albeit occult, at randomisation. In this report,
we have shown that these early eects on cancer death
are likely to be due, at least in part, to reductions in
distant metastasis (panel).
Platelets play an important part in cancer growth and
metastasis.1316 Thrombocytosis is common in several
cancers and usually indicates a poor prognosis,15,16 and
the role of platelets in allowing metastasis in mice
was identied more than 40 years ago.13,14 Aspirin was
subsequently shown to prevent these experimentally
induced metastases in mice.17 Our results show that
aspirin has a similar eect in humans. We found no
evidence of earlier diagnosis of cancer in the aspirin
groups, which might have led to an artefactual reduction
Panel: Research in context
Findings
By review of electronic and paper records of all patients with incident cancers in all ve
large UK randomised trials of daily aspirin versus no aspirin for the prevention of vascular
events, we showed that aspirin reduced risk of distant metastasis by 3040% and reduced
risk of metastatic adenocarcinoma by almost half. In patients with adenocarcinoma who
did not have metastasis at initial diagnosis and who remained on trial treatment up to or
after diagnosis, allocation to aspirin reduced risk of metastasis on subsequent follow-up
by about 70%. Eects were seen with a low-dose, slow-release formulation of aspirin that
had little systemic bioavailability.
Interpretation
These ndings provide the rst proof in man that, as well as reducing the long-term risk
of some cancers, aspirin also prevents distant cancer metastasis. Previous animal studies
had shown that platelets play a part in metastasis of cancer via the bloodstream to distant
tissues and that such metastasis might be prevented by aspirin. That aspirin prevents
metastasis at least partly accounts for the reduced cancer mortality recently reported in
trials of aspirin versus control in prevention of vascular events and suggests that aspirin
will also be eective in treatment of some cancers.
8 www.thelancet.com Published online March 21, 2012 DOI:10.1016/S0140-6736(12)60209-8
in later metastasis, and the eect of aspirin on clinically
overt metastasis was still evident 15 years after the
diagnosis of cancer, long after trial treatment was
stopped, suggesting that aspirin prevented formation of
metastases rather than just slowed their growth. Aspirin
might have reduced the propensity of cancers to
metastasise by a direct eect locally, but our ndings in
the TPT trial, in particular, suggest a platelet-mediated
eect. The low-dose, slow-release formulation of aspirin
used in TPT was designed to inhibit platelet function
only in the portal venous system and to have very little
systemic bioavailability owing to almost complete
deacetylation on rst pass through the liver.30 If aspirin
does reduce metastasis purely via platelet inhibition
then some eect on metastasis might also be seen in
trials of alternate-day aspirin and in trials of other
antiplatelet drugs, and combinations of antiplatelet
drugs might be of even greater benet. Previous trials of
dual antiplatelet treatment in prevention of vascular
events have so far been too short to allow any such
eects to be studied.
Our ndings have important implications for both
prevention and treatment of cancer. In relation to
prevention, the eect of aspirin on risk of metastasis
provides an explanation for the earlier than expected
eect on cancer mortality during aspirin trials12 and could
account for the small apparent reduction in cancer
incidence after only 3 years, if prevention of metastasis
resulted in a delay in presentation of some cancers.
However, the large long-term reductions in incidence of
some cancers, such as colorectal and oesophageal
cancers,911 are unlikely to be accounted for by a delayed
eect on metastasis alone and probably reect a reduction
in development of new cancers.
In relation to treatment of cancer, a recent diagnosis of
cancer would usually have excluded patients from the
trials that we studied. However, our ndings suggest
that aspirin can reduce metastasis when started soon
after a diagnosis is made. First, although clinically overt
metastases were not reduced until about 2 years after
randomisation to aspirin, such a delay would be expected
between the initial seeding of a micrometastasis and
the point at which subsequent growth would lead to
clinical presentation or detection on imaging, and aspirin
is likely to have started to prevent distant micrometa-
stasis soon after randomisation. This contention is sup-
ported by the nding that risk of later metastasis was
reduced in patients in whom adenocarcinomas were
diagnosed shortly after randomisation. Second, although
based on behaviour after randomisation, continuation of
aspirin up to or after diagnosis of cancer was associated
with a greater reduction in risk of subsequent metas-
tasis. Third, allocation to aspirin improved overall survival
in patients who developed adenocarcinoma during the
trials, particularly for cancers in which surgery has the
greatest potential to be curative, such as colorectal and
breast cancer, where prevention of later metastasis

consequently has most potential to improve long-term
survival. Observational studies have reported improved
survival rates in both of these cancers in patients starting
aspirin after diagnosis22,24 and the side-eect prole of
aspirin is relatively benign compared with those of most
cancer treatments. Yet, there have been only a few
small randomised trials of aspirin or other antiplatelet
drugs in treatment of cancer,5,18,31 and more trials are
clearly justied.
Our study has some shortcomings. The metastasis
status of about 20% of cancers diagnosed during the
trials was uncertain, often because of limited clinical
investigation in patients who died quickly. However,
this proportion was similar in both treatment groups
and when we assumed that all such patients did have
distant metastasis, our ndings were unchanged. More-
over, no patients were excluded from our analyses of
survival after diagnosis of cancer. The site of metastasis
was uncertain in some cases and information was
sometimes limited to terms such as metastatic cancer or
carcinomatosis in trial records or on post-trial death
certicates. However, eects of aspirin were largest for
denite site-specic blood-borne distant metastasis
alone, and any reduction in regional metastasis would
also be clinically important.
We did not have sucient data to stratify analyses by
exact tumour staging. However, our primary aim was to
study eects of aspirin on risk of distant metastasis, which
is a hard outcome, whereas tumour staging is dependent
on the nature of investigations and treatment, which have
changed over time. Furthermore, insucient data were
available to establish the eect of aspirin in relation to
potentially curative surgery or adjuvant radiotherapy and
chemotherapy. Further work is planned, subject to ethics
approval, to obtain more detail from archived hospital
records. In view of our previous observations on the eect
of aspirin on long-term post-trial incidence and mortality
due to cancer,10,11 further work is also needed to examine
the eect of aspirin on metastasis status of cancers that
were diagnosed after the trials were completed. Also
important to note is that, although the dierence in the
eect of aspirin that we observed on metastasis in
adenocarcinomas versus non-adenocarcinomas was
signicant and the analysis was prompted by data for
dierences in mortality eects from randomised trials,11,12
and by ndings from observational studies,2024,3234 similar
eects in some other cancers are possible.35,36
Our ndings only apply to daily aspirin. A previous
analysis of the eect of previous randomisation to
alternate-day low-dose aspirin on breast cancers during
follow-up in the Womens Health Study (WHS) did not
report data on distant metastasis,37 but some eects
might be expected if the eect of aspirin on metastasis
is platelet-mediated. However, the subgroup eects in
our study (appendix pp 67) suggest that the absolute
risk reduction in WHS is likely to be small given a
mean age at randomisation of 55 years and 87% non-
www.thelancet.com Published online March 21, 2012 DOI:10.1016/S0140-6736(12)60209-8
Articles
smokers. We found no eect of aspirin on risk of
metastatic cancer in non-smoking women, although
this subgroup was small. Indeed, smoking is associated
with an increased risk of distant metastasis in several
cancers,38,39 and more work is required to reliably
determine the interaction between the eect of aspirin
on risk of distant metastasis and smoking status. We
also only studied trials done in the UK. However, these
ve trials were the only trials of daily aspirin in which
detailed paper records or post-trial follow-up of patients
with cancer, or both, were available to us and they
include the majority of patients in all randomised trials
of daily aspirin with a scheduled duration of trial
treatment beyond 5 years.11 Further work is planned to
obtain any data for metastasis status from other trials
with shorter treatment periods that were included in
the all-inclusive analyses of cancer mortality in the
accompanying report,12 and from trials comparing
dierent antiplatelet drugs.
Despite these shortcomings, we think that our main
ndings are valid and generalisable. In fact, they are
likely to underestimate the true eects of aspirin. Most
analyses were done by intention to treat even though only
about 5060% of patients in the aspirin groups were still
on treatment by the end of the trials, and many patients
stopped trial treatment at or before diagnosis of cancer.
The more widespread use nowadays of screening for
cancers such as breast and colon, allowing earlier
detection and surgical removal, might have changed the
benet of aspirin in prevention of cancer. However, any
treatment that reduces the likelihood of metastasis before
a cancer is detected would be expected to improve the
eectiveness of screening. Detection of metastasis on
imaging will also have improved since the earlier trials
that we studied, but the rates of metastasis that we
recorded in the placebo groups did not suggest major
underascertainment.
In summary, that randomisation to daily aspirin
reduces the risk of metastasis in patients who sub-
sequently develop cancer has important implications.
First, it provides an explanation for the short-term
reduction in cancer deaths reported in all trials of daily
aspirin versus control.12 Second, that randomisation only
shortly before diagnosis of cancer still reduced later
metastasis, and that continuation of trial treatment
increased benet, suggest that starting aspirin after
diagnosis will also be benecial. There have been only a
few small randomised trials of aspirin in treatment of
cancer5,18,30,40 and, although two new trials have recently
started (NCT00565708, NCT01058902), there is an urgent
need for more. The accompanying systematic review of
evidence from observational studies suggests that they
can also contribute useful data for the eect of aspirin on
the risk and outcome of cancers.41 Third, stopping of
aspirin after a diagnosis of cancer, which occurs
commonly in routine clinical practice, could be detri-
mental. Finally, our results provide proof of principle for
9
 Articles
pharmacological intervention specically to prevent
distant metastasis. Moreover, our ndings in TPT show
that the eects of aspirin on metastasis are likely to be
platelet-mediated, raising the prospect that more inten-
sive platelet inhibition with combination treatment might
be even more eective.
Contributors
PMR conceived and coordinated the project, extracted and collated data
for metastasis status of cancers from electronic and paper records,
planned and performed analyses, and wrote the paper. MW also checked
and extracted data from some of the trials. ZM collated data and did
analyses. Other authors were investigators on the individual trials and
commented on drafts of the report.
Collaborators
J Boreham, R Peto, A Rudnicka, C P Warlow, R Lee.
Conicts of interest
This study was completely independent of any pharmaceutical company
or other commercial interest. However, PMR has received honoraria for
talks, advisory boards, and clinical trial committees from several
pharmaceutical companies with an interest in antiplatelet agents,
including AstraZeneca, Bayer, Boehringer Ingelheim, Sano-BMS, and
Servier, and is on the executive committee of the ARRIVE Trial. All other
authors declare that they have no conicts of interest.
Acknowledgments
PMR is in receipt of an NIHR Senior Investigator Award, a Wellcome
Trust Senior Investigator Award, and is also supported by the NIHR
Biomedical Research centre, Oxford.
References
1 Jemal A, Center MM, DeSantis C, Ward EM. Global patterns of
cancer incidence and mortality rates and trends.
Cancer Epidemiol Biomarkers Prev 2010; 19: 1893907.
2 Ferlay J, Autier P, Boniol M, Heanue M, Colombet M, Boyle P.
Estimates of the cancer incidence and mortality in Europe in 2006.
Ann Oncol 2007; 18: 58192.
3 The Lancet. Preventable cancer in the USA. Lancet 2010;
375: 1665.
4 Dannenberg AJ, DuBois RD, eds. Progress in experimental tumour
research vol. 37: COX-2a new target for cancer prevention and
treatment. Basel: Karger, 2003.
5 Thun MJ, Henley SJ, Patrono C. Nonsteroidal anti-inammatory
drugs as anticancer agents: mechanistic, pharmacologic, and
clinical issues. J Natl Cancer Inst 2002; 94: 25266.
6 Elwood PC, Gallagher AM, Duthie GG, Mur LA, Morgan G.
Aspirin, salicylates, and cancer. Lancet 2009; 373: 130109.
7 Cuzick J, Otto F, Baron JA, et al. Aspirin and non-steroidal
anti-inammatory drugs for cancer prevention: an international
consensus statement. Lancet Oncol 2009; 10: 50107.
8 Bosetti C, Gallus S, La Vecchia C. Aspirin and cancer risk:
a summary review to 2007. Recent Results Cancer Res 2009;
181: 23151.
9 Flossmann E, Rothwell PM, for the British Doctors Aspirin Trial
and the UK-TIA Aspirin Trial. Eect of aspirin on long-term risk of
colorectal cancer: consistent evidence from randomised and
observational studies. Lancet 2007; 369: 160313.
10 Rothwell PM, Wilson M, Elwin CE, Norrving B, Algra A,
Warlow CP, Meade TW. Long-term eect of aspirin on colorectal
cancer incidence and mortality: 20-year follow-up of ve
randomised trials. Lancet 2010; 376: 174150.
11 Rothwell PM, Fowkes FGR, Belch JFF, Ogawa H, Warlow CP,
Meade TW. Eect of daily aspirin on long-term risk of death due to
cancer: analysis of individual patient data from randomised trials.
Lancet 2011; 377: 3141.
12 Rothwell PM, Price JF, Fowkes FGR, et al. Short-term eects of
daily aspirin on cancer incidence, mortality, and non-vascular death:
analysis of the time course of risks and benets in 51 randomised
controlled trials. Lancet 2012; published online March 21.
DOI:10.1016/S0140-6736(11)61720-0.
13 Gasic GJ, Gasic TB, Stewart CC. Antimetastatic eects associated
with platelet reduction. Proc Natl Acad Sci USA 1968; 61: 4652.
10 www.thelancet.com Published online March 21, 2012 DOI:10.1016/S0140-6736(12)60209-8
14 Gasic GJ, Gasic TB, Galanti N, Johnson T, Murphy S.
Platelet-tumor-cell interactions in mice. The role of platelets in the
spread of malignant disease. Int J Cancer 1973; 11: 70418.
15 Bambace NM, Holmes CE. The platelet contribution to cancer
progression. J Thromb Haemostat 2011; 9: 23749.
16 Gay LJ, Felding-Habermann B. Contribution of platelets to tumour
metastasis. Nature Rev Cancer 2011; 11: 12334.
17 Gasic GJ, Gasic TB, Murphy S. Anti-metastatic eect of aspirin.
Lancet 1972; 300: 93233.
18 Henschke UK, Luande GJ, Choppala JD. Aspirin for reducing
cancer metastasis? J Nat Med Assoc 1977; 69: 58184.
19 Futakuchi M, Ogawa K, Sana M, Tamano S, Takeshitia F, Shirai T.
Suppression of lung metastasis by aspirin but not indomethacin in
an in vivo model of chemically induced hepatocellular carcinoma.
Jpn J Cancer Res 2002; 93: 117581.
20 Sharpe CR, Collet JP, McNutt M, Belzile E, Boivin JF, Hanley JA.
Nested case-control study of the eects of non-steroidal
anti-inammatory drugs on breast cancer risk and stage.
Br J Cancer 2000; 83: 11220.
21 Leitzmann MF, Stampfer MJ, Ma J, et al. Aspirin use in relation to
risk of prostate cancer. Cancer Epidemiol Biomarkers Prev 2002;
11: 110811.
22 Holmes MD, Chen WY, Li L, Hertzmark E, Spiegelman D,
Hankinson SE. Aspirin intake and survival after breast cancer.
J Clin Oncol 2010; 28: 146772.
23 Chan AT, Giovannucci EL, Meyerhardt JA, Schernhammer ES,
Wu K, Fuchs CS. Aspirin dose and duration of use and risk of
colorectal cancer in men. Gastroenterology 2008; 134: 2128.
24 Chan AT, Ogino S, Fuchs CS. Aspirin use and survival after
diagnosis of colorectal cancer. JAMA 2009; 302: 64958.
25 Peto R, Gray R, Collins R, et al. Randomised trial of prophylactic
daily aspirin in British male doctors. BMJ 1988; 296: 31316.
26 Farrell B, Godwin J, Richards S, Warlow C. The United Kingdom
transient ischaemic attack (UK-TIA) aspirin trial: nal results.
J Neurol Neurosurg Psychiatry 1991; 54: 104454.
27 Medical Research Councils General Practice Research Framework.
Thrombosis prevention trial: randomised trial of low-intensity oral
anticoagulation with warfarin and low-dose aspirin in the primary
prevention of ischaemic heart disease in men at increased risk.
Lancet 1998; 351: 23341.
28 Belch J, MacCuish A, Campbell I, et al; Prevention of Progression
of Arterial Disease and Diabetes Study Group; Diabetes Registry
Group; Royal College of Physicians Edinburgh. The prevention of
progression of arterial disease and diabetes (POPADAD) trial:
factorial randomised placebo controlled trial of aspirin and
antioxidants in patients with diabetes and asymptomatic peripheral
arterial disease. BMJ 2008; 337: a1840.
29 Fowkes FG, Price JF, Stewart MC, et al. Aspirin for Asymptomatic
Atherosclerosis Trialists. Aspirin for prevention of cardiovascular
events in a general population screened for a low ankle brachial
index: a randomized controlled trial. JAMA 2010; 303: 84148.
30 Charman WN, Charman SA, Monkhouse DC, et al.
Biopharmaceutical characterisation of a low-dose (75 mg)
controlled-release aspirin formulation. Br J Clin Pharmacol 1993;
36: 47073.
31 Lebeau B, Chastang C, Muir JF, Vincent J, Massin F, Fabre C.
No eect of an antiaggregant treatment with aspirin in small cell
lung cancer treated with CCAVP16 chemotherapy. Results from a
randomized clinical trial of 303 patients. The Petites Cellules
Group. Cancer 1993; 71: 174145.
32 Sansbury LB, Millikan RC, Schroeder JC, Moorman PG,
North KE, Sandler RS. Use of nonsteroidal antiinammatory
drugs and risk of colon cancer in a population-based, case-control
study of African Americans and Whites. Am J Epidemiol 2005;
162: 54858.
33 Chan AT, Giovannucci EL, Meyerhardt JA, Schernhammer ES,
Curhan GC, Fuchs CS. Long-term use of aspirin and nonsteroidal
anti-inammatory drugs and risk of colorectal cancer. JAMA 2005;
294: 91423.
34 Liu JF, Jamieson GG, Wu TC, Zhu GJ, Drew PA. A preliminary
study on the postoperative survival of patients given aspirin after
resection for squamous cell carcinoma of the esophagus or
adenocarcinoma of the cardia. Ann Surg Oncol 2009; 16: 1397402.

35 Slatore CG, Au DH, Littman AJ, Satia JA, White E. Association of
nonsteroidal anti-inammatory drugs with lung cancer: results
from a large cohort study. Cancer Epidemiol Biomarkers Prev 2009;
18: 120307.
36 Fontaine E, McShane J, Page R, et al. Aspirin and non-small cell
lung cancer resections: eect on long-term survival.
Eur J Cardiothorac Surg 2010; 38: 2126.
37 Zhang SM, Cook NR, Manson JE, Lee IM, Buring JE. Low-dose
aspirin and breast cancer risk: results by tumour characteristics
from a randomised trial. Br J Cancer 2008; 98: 98991.
38 Murin S, Inciardi J. Cigarette smoking and risk of pulmonary
metastasis from breast cancer. Chest 2001; 119: 163540.
www.thelancet.com Published online March 21, 2012 DOI:10.1016/S0140-6736(12)60209-8
Articles
39 Pantarotto J, Malone S, Dahrouge S, Gallant V, Eapen L. Smoking is
associated with worse outcomes in patients with prostate cancer
treated by radical radiotherapy. BJU Int 2006; 99: 56469.
40 Lipton A, Scialla S, Harvey H, et al. Adjuvant antiplatelet therapy
with aspirin in colorectal cancer. J Med 1982; 13: 41929.
41 Algra AM, Rothwell PM. Eects of regular aspirin on long-term
cancer incidence and metastasis: a systematic review of evidence
from observational studies versus randomised trials. Lancet Oncol
2012; published online March 21. DOI:10.1016/S1470-2045(12)70112-2.
11