Aspirin and cancer survival: a systematic review and meta-analyses of 118 observational studies of aspirin and 18 cancers

Background Despite the accumulation of research papers on aspirin and cancer, there is doubt as to whether or not aspirin is an acceptable and effective adjunct treatment of cancer. The results of several randomised trials are awaited, and these should give clear evidence on three common cancers: colon, breast and prostate. The biological effects of aspirin appear likely however to be of relevance to cancer generally, and to metastatic spread, rather than just to one or a few cancers, and there is already a lot of evidence, mainly from observational studies, on the association between aspirin and survival in a wide range of cancers. Aims In order to test the hypothesis that aspirin taking is associated with an increase in the survival of patients with cancer, we conducted a series of systematic literature searches to identify clinical studies of patients with cancer, some of whom took aspirin after having received a diagnosis of cancer. Results Three literature searches identified 118 published observational studies in patients with 18 different cancers. Eighty-one studies report on aspirin and cancer mortality and 63 studies report on all-cause mortality. Within a total of about a quarter of a million patients with cancer who reported taking aspirin, representing 20%–25% of the total cohort, we found aspirin to be associated with a reduction of about 20% in cancer deaths (pooled hazard ratio (HR): 0.79; 95% confidence intervals: 0.73, 0.84 in 70 reports and a pooled odds ratio (OR): 0.67; 0.45, 1.00 in 11 reports) with similar reductions in all-cause mortality (HR: 0.80; 0.74, 0.86 in 56 studies and OR: 0.57; 0.36, 0.89 in seven studies). The relative safety of aspirin taking was examined in the studies and the corresponding author of every paper was written to asking for additional information on bleeding. As expected, the frequency of bleeding increased in the patients taking aspirin, but fatal bleeding was rare and no author reported a significant excess in fatal bleeds associated with aspirin. No author mentioned cerebral bleeding in the patients they had followed. Conclusions There is a considerable body of evidence suggestive of about a 20% reduction in mortality in patients with cancer who take aspirin, and the benefit appears not to be restricted to one or a few cancers. Aspirin, therefore, appears to deserve serious consideration as an adjuvant treatment of cancer, and patients with cancer, and their carers, have a right to be informed of the available evidence.


Introduction
The first suggestive evidence of benefit to patients with cancer from aspirin was reported over 50 years ago. Studies of animals with cancer showed that aspirin is associated with a reduction in the development of metastases [1,2]. Since then, despite the reporting of much further evidence on biological effects of aspirin, and the reporting of many studies on aspirin and survival, there is still uncertainty about the role of aspirin as a possible adjuvant treatment of patients with cancer.
A number of small and inadequate randomised trials have been reported [3][4][5][6] and the pooling of results from these gives a suggestive reduction of 9% in cancer deaths in the 722 patients with cancer who had been randomised to aspirin (hazard ratio (HR): 0.91; 95% confidence interval (CI): 0.79, 1.04) [7]. While this result is only suggestive, a trial which developed within the cohort of the US Physicians Health Study of cancer prevention by aspirin is more strongly supportive. Just over 500 subjects in the cohort developed cancer, and those who had been randomised to aspirin showed a reduction in cancer deaths (HR: 0.68, 95% CI: 0.52, 0.90) [8].
Another source of evidence on the range of cancers to which aspirin may be relevant comes from opportunistic long-term follow-up studies of patients who had been involved in early randomised trials of aspirin and vascular disease. In addition to reporting a subsequent reduction in cancer incidence, Rothwell et al [9] and Mills and Wu [10] showed that deaths from a wide range of cancers were reduced in subjects who had been randomised to aspirin, and furthermore, the occurrence of metastatic spread was reduced in a range of cancers, including colon, brain, liver, lung and 'other or multiple sites' [11].
A number of new ad hoc randomised trial have been set up to test aspirin treatment in a few cancers and results from these are awaited [12][13][14][15]. These, however, are testing aspirin in only a very few cancers -principally colon, but also breast and prostate -while the effects of aspirin on biological mechanisms relevant to cancer lead to the possibility of benefit in most, if not all cancers [16][17][18]. Indeed, because of its manifold effects on biological processes, Zhang et al [19] suggest that aspirin is 'a master regulator of the hallmarks of cancer'.
The bulk of published evidence on aspirin and the treatment of cancer comes, however, from observational studies and in this report, we present the results of 118 published observational studies to test the hypothesis that aspirin is of benefit to a wide range of cancers and not just one or a few common cancers. We also present evidence that aspirin, relative to cancer and in comparisons with other cancer treatments, is a very safe drug.
Given that most of the available studies have been on the three common cancers: colon, breast and prostate, and in view of the fact that aspirin is being tested in randomised trials, we first present pooled evidence on aspirin and these three cancers. We then present evidence from 36 published reports of 15 other cancers, each of which has been examined in only one or a very few studies.
The procedures adopted followed the PRISMA guidelines throughout [21] and a full description of the search strategy is given in Supplementary File 1. In brief: each of the three systematic searches using keywords was conducted by AW and DM in MEDLINE and EMBASE. The searches were limited to human studies in peer-reviewed journals. Relevant studies were selected by two authors (PE and GM) if (a) the studied population comprised patients diagnosed with cancer; (b) aspirin appears to have been taken regularly after cancer diagnosis; (c) the studies were randomised trials, case-control studies or cohort studies. Reference lists of the relevant studies identified were searched for other relevant reports. At least one author on each selected paper in all three searches was written to and asked specifically about gastrointestinal (GI) bleeding in the patients included in their study, together with appropriate further questions.
Data on cancer deaths and deaths from all-causes in the most recent search to March 2020 are listed in Supplementary File 3, first for studies that had expressed association as HRs, followed by studies which had used odds ratios (ORs), risk ratios (RRs) or percent survival. The methodological quality of the studies was assessed and graded independently by two authors (AW and PE) using the Newcastle-Ottawa Scale [22]. We have also added to each paper listed in Supplementary File 3 a comment as to the level of certainty that aspirin had been taken -or had not been taken -throughout follow-up.
Most of the risk estimates reported by the authors were expressed as HRs, and these and their 95% CI were taken from the original articles and log-transformed to obtain the estimate of the treatment effect (TE). The standard errors (seTE) were determined by subtracting the lower log-transformed CI boundary from the upper log-transformed CI boundary and dividing this by 3.92 (2*1.96). Where HRs were not reported, ORs, RRs and their 95% CI, or number of events among patients taking aspirin and those not on aspirin, were taken from the original articles. ORs and exact 95% CIs were calculated where needed, and all were then log-transformed for meta-analysis.
Summary risk estimates of random effects models are shown as forest plots in Supplementary File 4. HR meta-analyses were conducted using the meta package, version 4.13.0 in R 4.0.2, open source. Analysis with the metagen function used sm = HR for the underlying summary method and the DerSimonian-Laird method [23] was used to estimate the between-study variance (τ and τ 2 ). Meta-analyses of the reports as ORs were conducted using Stata/SE 16.1, and used the restricted maximum likelihood method to estimate the between-study variance and these are shown as forest plots in Supplementary File 5.
Finally, funnel plots were constructed and estimates of the probability of publication bias were derived. The forest plot added trim and fill which mirrored the studies followed by a cumulative forest plot based on decreasing standard error. This was only undertaken on a minimum of 10 papers hence there is only one examination for OR. These are all shown in Supplementary File 6.

Results
Three systematic literature searches on the topic of this report were conducted by the authors: in 2016 [7], in 2018 [20] and in 2020 up to March 2020 (Supplementary File 1). In each report, there are two outcomes, death from cancer and death from any cause, almost all of which have been presented as HRs. The new studies are described in Supplementary File 3 and their results are listed and pooled in Supplementary File 4. Some of the deaths have however been reported as OR, relative risk, etc., and all these have been converted to ORs. These ORs are presented separately from the HRs in Supplementary File 3 and are listed and pooled in Supplementary File 5. Some results however have been presented as additional survival in months or years, or during defined periods of time, such as 5 years. These are mentioned in the text, but do not appear in any table or Supplementary file.
In addition, we were concerned about undesirable side effects of the aspirin and in addition to abstracting relevant data from the published reports, following each of the three searches we wrote to an author of every report, asking for details of any unwanted side effect and in particular bleeding attributable to aspirin. A few authors supplied evidence on bleeding further to that in their published report, and these details are quoted in the text.

Mortality
For colon cancer mortality, our three literature searches identified a total of 24 studies in which the association with aspirin was reported as HRs. Together, these give a pooled HR of 0.72 (95% CI: 0.63, 0.82), and a single report showed an OR (OR of 0.78 (0.66, 0.93) ( Table 1  Not all the estimates of association in these reports of 'other' cancers are significant at p < 0.05. However, only three are consistent with an oropharynx possible harmful effect of aspirin, having a confidence limit which includes 1, but none of the three is significant at p < 0.05 with both confidence limits above 1. Together, these reports of 'other' cancers give a pooled HR for cancer mortality of 0.79 (0.70, 0.88) in 18 studies and a pooled OR of 0.49 (0.26, 0.95) in five studies. All-cause mortality in 21 of these other cancers gives a pooled HR of 0.67 (0.60, 0.75) in 21 studies and the five studies that did not report HRs give a pooled OR of 0.47 (0.26, 0.83).
The forest plots of all these data are shown in Supplementary Files 4 and 5, and Table 2 brings together all the available data on cancer deaths and on all-cause mortality. Finally, as a test of the hypothesis posed in this report, we compared the association of aspirin and cancer mortality in the 15 less common cancers with cancer mortality in colon cancer. In this comparison, we use colon cancer mortality as the 'gold estimate' of the effect of aspirin on the grounds that the effect of aspirin has been more thoroughly investigated in colon cancer, than in any other cancer; colon cancer is the only cancer for which the UK National Institute for Clinical Excellence has given a limited recommendation for aspirin, [

Bleeding
A search for evidence on bleeding, and fatal bleeding attributable to aspirin was made, and this included writing to the corresponding author on all the 118 papers included in the three searches. Many of the studies however had been based on recorded data, with no direct contact with the patients involved, and authors of such reports had little or no knowledge about bleeding in the patients they described.
Many of the authors reported the expected excess in GI bleeding in the patients on aspirin. However, only a very few reported fatal bleeds. In one study, 3% of the patients taking aspirin and 3.2% of those not taking aspirin had had a fatal bleed [40]. Tsoi et al [49], who studied a cohort of over 18,000 patients with colon cancer reported that deaths of aspirin users who developed GI bleeding were 0.40%, compared with 0.36% of the patients not taking aspirin. A study of patients with liver cancer treated by transarterial chemoembolisation reported that six patients in the aspirin group and seven patients in the non-aspirin group died because of upper GI bleeding [93]. One paper makes mention of the reduction in bleeding in patients who took a PPI along with the aspirin (OR: 0.85; 0.80, 0.91) [49]. All the references to bleeds relate to GI bleeds and no author made mention of cerebral bleeding.

Discussion
This report provides both confirmatory and new evidence on the benefit of aspirin in reducing mortality in patients being treated for cancer. Replication is an important procedure in science and the present study confirms the findings of our first report with 50 studies [7], and our second report with 29 studies [20]. The present study is a further replicate with 39 new observational studies.
The meta-analyses we now present are all based on pooling of the data provided by 118 observational studies comprising about a quarter of a million patients with cancer who were recorded as taking aspirin. This reveals that aspirin taking is associated with a reduction of cancer deaths of about one fifth in a range of 18 cancers (HR: 0.79 (0.73, 0.84) in 70 observational studies and OR: 0.67 (0.45, 1.00) in 11 studies (Table 2 and Supplementary Files 4 and 5). The effect of aspirin on all-cause mortality is closely similar (HR: 0.80 (0.74, 0.86) in 56 observational studies and OR: 0.57 (0.36, 0.89) in 7 studies). A reasonable interpretation of these results is -that at any time after a diagnosis of a wide range of different cancers, about 20% more of the patients who take aspirin are likely to be alive, compared with patients not taking aspirin.
The evidence of publication bias throughout this work is a most important issue. Bias due to the selective publication of positive findings for aspirin was expected, and for some of the pooled results the magnitude of this bias is greater than could be reasonably expected in chance grounds alone (Supplementary File 6). While conclusions drawn from these 118 papers have, therefore, to be cautious, the evidence is strengthened by the absence of significant bias at p < 0.05 for the data for colon cancer. It is also encouraging that the trim and fill analysis on the less common cancers maintained the beneficial TE for both cancer specific mortality and all-cause mortality.

Bleeding
A bleed, either GI or intra-cerebral, is a crisis for a patient, but the seriousness of a bleed attributable to aspirin should be evaluated against the likely benefits attributable to its use and furthermore the severity of the additional bleeds attributable to aspirin should be considered and not just their frequency [122,123]. In relation to the treatment of cancer, our examination of the 118 reports gives a considerable degree of reassurance on aspirin, and particularly on the most serious bleeds. It is of relevance that most of the patients appear to have been taking low-dose aspirin primarily for cardiovascular protection.
Low-dose aspirin is however associated with additional GI bleeds in between 0.8 and 5.0 patients per 1,000 person years aged 50-84 years in the general population [124]. This represents an increase above spontaneous GI bleeding of between about 50% [125] and 90% [121]. It is important to note that these increases imply that only one in every two or every three bleeds that occur in patients taking low-dose aspirin is likely to be truly attributable to the aspirin, the other bleeds being spontaneous and nothing to do with aspirin.
The most serious bleeds are those that lead to death and our concern on this led us, early in our investigations of aspirin and cancer, to conduct a careful evaluation of fatal bleeding [126] A meta-analysis based on 11 randomised trials showed that the additional bleeds attributable to aspirin are less serious than spontaneous bleeds and are seldom, if ever fatal (relative risk of death: 0.45; (0.25, 0.80), and the risk of a fatal bleed in the totality of subjects randomised to aspirin, relative to subjects randomised to placebo was RR: 0.77 (0.41, 1.42). As we reported in our overview [126], others have reported similar findings of a reduction in the proportion of fatal bleeds in patients taking aspirin [9,[127][128][129][130].
Findings on bleeding in the recent ASPREE trial of prophylactic aspirin are of interest as more than 19,000 subjects with a median age of 74 years were followed for 5 years. Eighty-nine subjects randomised to aspirin, or 1.9 in every 1,000 experienced a bleed each year, compared with 48 bleeds, or just over 1.1 per thousand per year in those not taking aspirin. Granted this was not a trial of aspirin treatment, but it is of relevance to the safety of the drug that only two fatal bleeds occurred, and neither was in a subject taking aspirin [131].
The most serious side effect of aspirin, intra-cerebral bleeding, is fortunately rare [132], and no author in our literature review mentioned cerebral bleeding within the patients they followed. The risk associated with aspirin is estimated to be around 1.39 (95% CI: 1.08, 1.78) [125] equivalent to one or two additional haemorrhagic strokes per year in every 10,000 subjects [133].
Hypertension is the major factor in haemorrhagic stroke and in one major overview of randomised trials there was a doubling of cerebral haemorrhages for a rise of 20 mmHg in blood pressure (RR: 2.18; 95% CI: 1.65, 2.87) [127]. The relevance of hypertension was further highlighted in a trial of aspirin based on 20,000 patients with hypertensive disease, all of whom were adequately treated with anti-hypertensive drugs. There were no additional cerebral bleeds attributable to aspirin: the same number of patients on aspirin experienced cerebral bleeds (19 patients) as those on placebo (20 patients) [133].

Strengths and limitations of this study
In addition to the risks of publication bias as detailed above, a most important limitation is that almost all the evidence we present are from observational studies. A number of randomised trials of therapeutic aspirin are in progress but these focus entirely on either one, or a few of the common cancers: colon [12][13][14][15], breast [12,14] and prostate [12,15]. Our concern, however, is for all cancers and not one or a few cancers, and as others have pointed out many of the actions of aspirin on cancer development, growth and metastatic spread, appear likely to be relevant to a wide range of cancers [6][7][8][9][10][11][12][13][14][15][16][17][18][19].
It is important to note that amongst the uncertainties in these observational studies, two uncertainties appear to stand out in their probable relevance to every observational study, and to the possible size of their effects. These are: first: uncertainties about the classification of patients with regard to continuous aspirin taking, and uncertainties about the non-taking of aspirin by the 'controls', and secondly, comorbidity in the patients taking aspirin.
Few authors give reassurance about continued aspirin taking during follow-up, and no authors comment on the possibility of 'contamination' of control subjects starting to take 'over the counter' aspirin during the follow-up. An additional column in Supplementary File 3 lists quotations from the papers reviewed and these show that most authors assumed that if there is evidence of aspirin taking at the time of diagnosis, it can reasonably be assumed that aspirin taking was continuous during follow-up. Thus, 'Low-dose aspirin use was defined as a minimum of one filled prescription after cancer diagnosis' [89] and another: 'the patients were receiving aspirin from diagnosis to at least 1 year after treatment initiation' [90]. One author pointed out however that 'the inverse association with aspirin appeared to be only among men who reported using aspirin regularly', [76] and another noted that a reduction in mortality was 'notably among patients filling prescriptions for a large quantity of low dose aspirin tablets during the (follow-up) period [77]. Another author found that prescribed aspirin alone was not associated with decreased mortality, but when OTC aspirin was added, a large reduction was detected [39].
A recent study by a group in Dublin examined the influence of approaching death on end-of-life aspirin use in patients with breast or colorectal cancer. They found that the use of aspirin declined 'considerably' during the 2 years before death, and at the time of death rates of aspirin use had dropped from around 60% to around 20% for colorectal cancer and from around 80% to around 45% for breast cancer [134].
The only comment about aspirin taking by control subjects comes from an overview of 12 studies in which the authors state that the pooled survival in patients on aspirin was only HR: 0.96 (0.88, 1.04) but if non-aspirin taking was more tightly defined as less than once per week, the HR was 0.89 (0.82, 0.98) [135].
The other important limitation is confounding by co-morbidity. Many authors mention that the aspirin takers in their study were older than the control patients not on aspirin. While this can be adjusted for statistically, the fact that a number of studies state that most of the patients who were taking aspirin were doing so because of a prior vascular event or prevalent vascular disease. Clearly, the morbidity that had led some of the patients to take aspirin can have eroded any benefit achievable by aspirin and while many of the papers mention this, few give details.
Yet a further limitation arises from possible miscoding of the causes of death in these studies. In the SEER programme on mortality in patients with cancer in the USA, it was found that 11% of cancer deaths had been attributed to vascular disease [136]. Any such miscoding will lead to an underestimate of the reduction in cancer deaths associated with aspirin.
The very broad range in the estimates of effect of aspirin leading to high heterogeneity estimates in our meta-analyses is worrying, and some of the differences between studies seem to defy any reasonable explanation. And yet, this was predicted from the beginning of the work on aspirin treatment [7]. There are many biases and sources of possible differences between the series of patients in the various studies, including differences in age and social factors, differences in other treatments and in general clinical management [41,48]. Then there are possible differences in consistency of aspirin taking and the differences in co-morbidity already mentioned. Both poor aspirin taking and co-morbidity in patients taking aspirin will increase heterogeneity, and are probably inevitable in a series of studies such as we present. On the other hand, it seems unlikely that such differences could account for the overall benefits we find to be associated with aspirin taking.

Conclusions
We judge that the body of evidence now available on the efficacy and the safety of aspirin justifies its use as an adjunct treatment in a wide range of cancers. Clinical care includes an ethical imperative for shared decision making [137] and we, therefore, believe that doctors should present, and patients with cancer should be encouraged to raise the topic of aspirin taking with their doctors. At the same time, we stress that aspirin is not a possible alternative to any other treatment, although in poorer countries aspirin could be one of very few, or perhaps the only acceptable treatment on the grounds of cost and availability [138].
Further research into aspirin and cancer would clearly be of great value, and studies including observational and randomised trial should be encouraged, especially if focused upon one of the less common cancers.

Conflicts of interest
The author(s) declare that they have no conflict of interest. All the authors have read the paper and agree with its content.

Funding
No special funding was obtained for any of the work described in this paper.