The evidence strength of a meta-analysis of aspirin for primary prevention of cancer

Abstract

Dr. Tarek Haykal et al. (145:1795–1809, 2019) reported a meta-analysis of aspirin for the primary prevention of cancer in individuals without known cancer. The authors found that aspirin use was not associated with significant reduction in cancer mortality or incidence, but with higher rates of bleeding. The findings of this study added some evidence to the clinical practice. However, several issues might have compromised the strength of the evidence of this systematic review. If the investigators could have further clarified the inclusion and exclusion criteria, included all eligible studies, extracted data more meticulously, and performed more necessary sensitivity analyses to confirm the robustness of their findings, the strength of evidence of this meta-analysis would have been stronger.

Dr. Tarek Haykal et al. (2019) reported a meta-analysis evaluating the efficacy and safety of aspirin as a primary prevention method for cancer in individuals without known cancer. The authors found that long-term aspirin use was not associated with significant reduction in cancer-related mortality or incidence, but with higher rates of bleeding. Considering that aspirin’s role in cancer primary prevention remains controversial, this evidence-based study is important and might help to add evidence to this clinical practice. However, several issues might have compromised the trustworthiness and strength of the evidence of this systematic review.

First, based on the inclusion criteria and search strategy described in the study, more than ten eligible randomized controlled trials (RCTs) with over 70,000 participants were excluded even if they met the inclusion criteria (Table 1), including several important RCTs with large size of participants, such as ARRIVE (Gaziano et al. 2018) BDS (Peto et al. 1988) HOT (Hansson et al. 1998) and JPPP (Yokoyama et al. 2018) Furthermore, two trials included in the meta-analysis were duplicated (Ogawa et al. 2008; Okada et al. 2018) which were two different reports of the same study, the Japanese Primary Prevention of Atherosclerosis With Aspirin for Diabetes (JPAD) trial (Morimoto et al. 2007). In addition, the inclusion criteria were not explicit enough, for example, the study duration needed for cancer primary prevention trials could not be short, so the inclusion criterion for follow-up duration should be described and the studies on the short-term effect of aspirin should be excluded, but the authors did not discuss this in their analysis.

Table 1.

Some studies that were improperly excluded in the meta-analysis

Reference (trial name, year)	Trial design	Total randomized
AAA, 2010 (Fowkes et al. 2010)	RCT, double-blind	3350
AFPPS, 2003 (Baron et al. 2003)	RCT, double-blind	1121
AMIS, 1980 (Schoenberger 1980)	RCT, double-blind	4524
ARRIVE, 2018 (Gaziano et al. 2018)	RCT, double-blind	12,546
BDS, 1988 (Peto et al. 1988)	RCT, open-label, blind endpoint	5139
CDPA, 1980 (The Coronary Drug Project Research Group 1980)	RCT, double-blind	1529
CLIPS, 2007 (Critical Leg Ischaemia Prevention Study G et al. 2007)	RCT, double-blind	366
DAMAD, 1989 (Effect of aspirin alone and aspirin plus dipyridamole in early diabetic retinopathy. A multicenter randomized controlled clinical trial. The DAMAD Study Group 1989)	RCT, double-blind	475
EAFT, 1993 (Secondary prevention in non-rheumatic atrial fibrillation after transient ischaemic attack or minor stroke. EAFT (European Atrial Fibrillation Trial) Study Group 1993)	RCT, double-blind	782
HOT, 1998 (Hansson et al. 1998)	RCT, double-blind	18,790
JPPP, 2018 (Yokoyama et al. 2018)	RCT, open-label, blind endpoint	14,658
REDUCE, 2015 (Vidal et al. 2015)	Multicenter, RCT, double-blind	6390
ukCAP, 2008 (Algra and Rothwell 2012)	RCT, double-blind	945

AAA Aspirin for Asymptomatic Atherosclerosis Trial, AFPPS The Aspirin/Folate Polyp Prevention Study, AMIS Aspirin Myocardial Infarction Study, ARRIVE Aspirin to Reduce Risk of Initial Vascular Events, BDS British Doctors Study, CDPA Coronary Drug Project Research, CLIPS Critical Leg Ischaemia Prevention Study, DAMAD the Dipyridamole Aspirin Microangiopathy of Diabetes study, EAFT European atrial fibrillation trial, HOT Hypertension Optimal Treatment, JPPP Japanese Primary Prevention Project, RCT Randomized controlled trial, REDUCE the Reduction by Dutasteride of Prostate Cancer Events study, ukCAP The United Kingdom Colorectal Adenoma Prevention

Second, only the inclusion criteria were described, but without exclusion criteria, thus leading to uncertainty of the study selection, for example, how did the authors deal with the studies on the effect of aspirin for those high-incidence familial cancer syndromes or premalignant conditions (e.g., Lynch syndrome, etc.) (Barton 2012; Burn et al. 2008) or those studies where the participants concurrently took other medications which might mask the actual association between the aspirin treatment and cancer outcomes (Hull et al. 2018).

Third, the authors reported that there was no publication bias based on funnel plots for the primary outcome (cancer-related mortality), so their interpretation of the result might not be accurate. In fact, the asymmetry of funnel plots suggested possible publication bias. In this situation, some other tests (e.g., Egger’s test) should be applied to further examine the potential bias (Egger et al. 1997).

Fourth, the sensitivity analysis was only done by removing each trial which was far from enough, while sensitivity analyses based on more important factors, such as aspirin dose or follow-up duration, were not performed to examine the consistency of the results. Their selective inclusion of some trials while excluding other relevant data could have introduced bias and potentially undermined the validity of the results and conclusions.

Last, there might be errors in the data used for the cancer-related mortality outcome. The authors used total events for the aspirin group (729 events) and the control group (748 events) from Cook et al. (2013) which was an observational follow-up of Women’s Health Study (WHS) (Cook et al. 2005). The active intervention for the participants in WHS had ended in 2004 and the results were published in 2005 (Cook et al. 2005). Those data from the posttrial observational study might not be eligible for a meta-analysis of RCTs. The data of cancer deaths reported by Cook et al. (2005) were 284 for the aspirin group and 299 for the control group, which would be more appropriate for a meta-analysis of this outcome. The table 2 in the article by Dr. Tarek Haykal et al. (2019) showed the follow-up period of the study by Cook et al. (2013) was 10 years, which was wrong for the study by Cook et al. (2013), but it was right for the study by Cook et al. (2005). Similarly, they used cancer deaths’ data from the UK-TIA (UTKIA in the article might be a typo) (20 for the aspirin group) for the pooled estimate, but the number reported in that RCT was 21 for the aspirin group (Farrell et al. 1991).

The above questions do not imply that the findings of this meta-analysis are overturned, which might remain similar even when all eligible RCTs are included and the data are updated. However, the strength of evidence would become stronger if this meta-analysis could have further clarified the inclusion and exclusion criteria, included all eligible studies, extracted data more meticulously, and performed more necessary sensitivity analyses to confirm the robustness of their findings. Besides, the Cochrane’s Risk of Bias Tool and the GRADE system could be used to assess the methodologic quality of the RCTs and the quality of evidence (Guyatt et al. 2013; Higgins 2011).

Compliance with ethical standards

Conflict of interest

The authors declare that they have no conflict of interest.