Colorectal polyp outcomes after participation in the seAFOod polyp prevention trial: Evidence of rebound elevated colorectal polyp risk after short‐term aspirin use

The seAFOod polyp prevention trial was a randomised, placebo‐controlled, 2 × 2 factorial trial of aspirin 300 mg and eicosapentaenoic acid (EPA) 2000 mg daily in individuals who had a screening colonoscopy in the English Bowel Cancer Screening Programme (BCSP). Aspirin treatment was associated with a 20% reduction in colorectal polyp number at BCSP surveillance colonoscopy 12 months later. It is unclear what happens to colorectal polyp risk after short‐term aspirin use.


| INTRODUC TI ON
The seAFOod polyp prevention trial was a randomised, double-blind, placebo-controlled, 2 × 2 factorial trial of the colorectal cancer (CRC) chemoprevention efficacy of aspirin 300 mg daily and eicosapentaenoic acid (EPA) 2000 mg free fatty acid equivalents daily in 'high risk' patients undergoing colonoscopy surveillance in the English Bowel Cancer Screening Programme (BCSP). 1,2 Trial participants were aged 55-73 years and had been invited for screening colonoscopy on the basis of a positive faecal occult blood test or 'high risk' screening flexible sigmoidoscopy. 1,2 The trial population was predominantly (80%) male and White European, in keeping with the demographic characteristics of individuals undergoing BCSP colonoscopy. 1,2 Patients with a known genetic CRC predisposition were excluded. 1,2 The intervention period between the screening colonoscopy and the first surveillance colonoscopy was 12 months in individuals deemed 'high risk' (defined as ≥5 polyps or ≥3 polyps, if one or more polyps were ≥10 mm in size), a duration which has been associated with a similar degree of colorectal polyp risk reduction compared with a 3year intervention period in previous aspirin polyp prevention trials. 3 The primary finding from the seAFOod trial was that aspirin and EPA did not reduce colorectal polyp incidence, measured as the 'adenoma detection rate' (the % of individuals with one or more colorectal polyps 12 months after clearance colonoscopy) by an 'at the margins' analysis of the individual interventions. 1,2 However, aspirin use was associated with a significant reduction in overall colorectal polyp risk (measured as mean polyp number per participant). 1,2 There was also colorectal site-and polyp type (conventional adenoma or serrated polyp)-specific chemoprevention activity of aspirin and EPA; notably, randomisation to aspirin was associated with reduced risk of serrated lesions, unlike EPA treatment, which was associated with a statistically significant reduction in risk of left-sided (distal to the splenic flexure) conventional adenomas. 1,2 The seAFOod trial was not powered for a pre-specified 'inside the table' analysis of the four treatment groups, including combined aspirin and EPA therapy. 1,2 However, laboratory studies have since suggested that aspirin and EPA may have a positive interaction for CRC risk prevention via synthesis of novel oxylipins such as E-type resolvins, 4 and/or substrate diversion of EPA secondary to cyclooxygenase inhibition by aspirin. 5 It is not known whether CRC chemoprevention agents inhibit either, or both, initiation and/or growth of tumours at the earliest stages of intestinal tumorigenesis. Inhibition of tumour initiation is hypothesised to provide prolonged benefit after cessation of a chemoprevention agent. Alternatively, tumour growth suppression alone (without inhibition of tumour initiation) is hypothesised to lead to a 'rebound' increase in colorectal polyp incidence, whereby undetectable tumours that initiated but did not grow to become macroscopically visible during chemopreventative agent use, are de-repressed and become detectable upon cessation of chemoprevention therapy. Knowledge of whether a 'rebound' increase in colorectal polyps (indicative of increased CRC risk) occurs will be critical in order to define guidelines for the duration of chemoprevention use, optimal timing of cessation of therapy and best use of accompanying colonoscopic surveillance. This is particularly important in the elderly, in whom the common practice is to stop aspirin therapy driven by increasing concern about elevated bleeding risk. 6 seAFOod trial participants were invited by the English BCSP to undergo surveillance colonoscopy at 3 years (or earlier dictated by prior colorectal polyp findings) after the initial 'high risk' 1-year surveillance procedure (which was the seAFOod trial exit colonoscopy), with subsequent surveillance dependent on the most recent findings ( Figure S1). All trial participants provided informed consent for collection and use of post-trial BCSP data up to 6 years after trial participation (covering a maximum of two 3-year surveillance cycles; Figure S1). Therefore, we obtained post-trial colonoscopy data in order to investigate colorectal polyp risk following cessation of short-term chemoprevention with aspirin and EPA. All participants in the seAFOod polyp prevention trial provided specific, written informed consent to allow access to post-trial BCSS data. 1,2 Endoscopy surveillance data on procedures that occurred in the BCSP for a maximum of 6 years after the seAFOod trial exit colonoscopy (which was the first 'high risk' BCSP surveillance colonoscopy at 12 months after screening) were obtained, where linkage was possible ( Figure S1). The first trial exit colonoscopy was in November 2012 and the last trial exit colonoscopy was in June 2017.

| Bowel cancer screening programme data
The BSCP data extraction date was 7 October 2021. BCSS data were linked to the seAFOod trial database using date of birth, sex and hospital site, as well as dates of the screening and first surveillance colonoscopy. No other data on seAFOod trial participants were available after the trial finished, including subsequent use of aspirin and/or omega-3 polyunsaturated fatty acid supplements.
Participants were not told their treatment allocation after the trial had completed and the Plain English Summary of the trial described the null primary outcome of the trial without any recommendation for future CRC chemoprevention.

| Data collection
Data were obtained at individual-, test (each endoscopic procedure)-, BCSP episode (which can include one or more endoscopic procedures as a single administrative clinical episode)-and polyp-level. For each post-trial surveillance colonoscopy, the time elapsed since the first surveillance colonoscopy and extent of examination (complete or incomplete colonoscopy) were noted, as well as the number of separate procedures per individual. Endoscopic procedures occurring within 6 months of the date of the trial exit colonoscopy were assumed to be part of the same episode of BCSP care as the surveillance colonoscopy at 12 months after screening and were discounted. subcentimetre polyps}]), which was in use during the seAFOod trial, was used to report findings at procedural and BCSP episode level. 8 Since the seAFOod trial finished, updated British Society of Gastroenterology (BSG) Guidelines on colonoscopic surveillance after polypectomy have been adopted by the BSCP since mid-2020. 7 Current risk stratification now consists of a single 'high-risk' category (≥5 polyps or ≥2 polyps if one or more are ≥10 mm in size) for 3-year surveillance, as opposed to other findings that do not warrant further colonoscopic surveillance. Therefore, data were also analysed at test and episode level according to the new BCSP classification, to ensure that the results reflected current BCSP practice and current US Multi-Society Task Force on CRC guidelines on colonoscopic surveillance. 9 The baseline characteristics (age, sex, body mass index [BMI], smoking status) of seAFOod trial participants that had post-trial BCSP surveillance data (included in this analysis) versus trial participants that did not have any linked endoscopic BCSP surveillance data after the seAFOod trial were compared.

| Post-trial surveillance colonoscopy outcomes
The primary outcome for the post-trial analysis was stipulated as total colorectal polyps by an 'at the margins' analysis, consistent with the primary analysis of the seAFOod trial, 1,2 by comparing data from those randomised to aspirin versus no aspirin, and individuals randomised to EPA versus no EPA.
Colorectal polyp data are reported as the % value for each PDR outcome, as well as the mean and 95% confidence interval (CI) values for colorectal polyp number (MPP). Secondary colorectal polyp size and burden data are also reported as mean and 95% CI values. The Student's t-test or one-way analysis of variance was used for univariate comparison of continuous participant characteristics and follow-up duration across two or more groups, as appropriate. A chi-squared test was used to compare sex ratios across treatment groups. Treatment group differences in PDR were tested by logistic regression. Data are reported as the odds ratio (OR) and 95% CI. A negative-binomial regression model was used to compare colorectal polyp number (and polyp burden) outcomes based on the skewed distribution of individual colorectal polyp frequencies that was apparent at the trial exit colonoscopy ( Figure S2). 1,2 The incidence rate ratio (IRR) and 95% CI values for groups based on prior seAFOod trial treatment allocation are reported for comparison with the respective placebo group. The mean colorectal polyp size per person was compared across previous trial treatment allocation by linear regression and reported as the mean size difference and 95% CI. All models were adjusted for age and sex and also included BCSP research site as a random effect. Statistical significance for all analyses was assumed at p ≤ 0.05.
In addition to the main analysis of cumulative post-trial colorectal polyp outcomes, which included BCSP surveillance data from colonoscopies for up to 6 years after the trial exit colonoscopy ( Figure S1), we also performed a sensitivity analysis of total colorectal polyp outcomes at a single point in time after seAFOod trial participation.
This was restricted to individuals who had 3-year surveillance colonoscopy outcomes only (using a 2.5-to 3.5-year post-trial window), excluding individuals who had undergone any post-trial procedure prior to this.

| seAFOod trial colonoscopy outcomes
We repeated the 'at the margins' analysis of total colorectal polyp number from the seAFOod trial in the subgroup of trial participants that had post-trial colonoscopy data, in order to confirm the original trial treatment effect of aspirin on total colorectal polyp risk in the current study cohort. The same Poisson regression model included in the original final trial analysis, adjusted for the BCSP research site as a random effect and for repeat colonoscopy at trial baseline, was used. 1,2 A separate negative-binomial model was also used based on the distribution of individual colorectal polyp frequencies that was apparent at the trial exit colonoscopy ( Figure S2). 1,2 In addition, we performed post hoc 'inside the on the basis that the total number of colorectal polyps detected at the seAFOod trial exit colonoscopy in the group receiving both aspirin and EPA together was noticeably less than in the groups who received either agent alone, or placebo only. 1,2 Using the original seAFOod trial outcomes data and the same regression model used in the primary trial analysis, we generated the IRR and 95% CI for the group that received combined treatment compared with the other three groups. 1,2 An interaction between aspirin and EPA was analysed by comparing individuals allocated to active EPA with those who received placebo EPA, using a Poisson regression model, 1,2 which also included age, sex, BMI, smoking status, alcohol intake, baseline red blood cell % EPA level and randomisation to aspirin as co-variables, with statistical significance reported as the Wald test for interaction. Evidence of an interaction between aspirin and EPA prompted a secondary comparison of the posttrial surveillance colonoscopy outcomes across the four treatment groups, in addition to the primary analysis according to factorial margins.

| Patient and public involvement
A patient and public representative was a member of the STOP-ADENOMA study group and supported the application to the Public Health England Office for Data Release for access to the BCSP data, by writing a General Data Protection Regulation Privacy Notice in conjunction with the co-investigators.

| The post-trial study population and BCSP procedures
Five hundred and seven individuals, who had been randomised to the seAFOod trial, had undergone one or more colonoscopies in the English BCSP, which were more than 6 months after, and less than 6 years after, trial participation ( Figure 1A). No BCSS data were available for 200 participants for several reasons ( Figure 1A). Only 71 individuals were invited for BCSP colonoscopy but did not attend, a number which is likely to have been increased by the ongoing COVID-19 pandemic ( Figure 1A).  (Table S1). 1,2 We confirmed that the post-trial surveillance study population of 507 individuals displayed a similar on-trial treatment effect size for aspirin and EPA as the published primary analysis of the full seAFOod trial population. 1 Figure S4.
We also re-classified surveillance outcomes according to the current risk stratification used by the English BCSP surveillance pathway. 8 Overall, 96 (16.0%) of 602 procedures were classified as 'high-risk' (≥5 polyps, or ≥2 polyps if one or more are ≥10 mm in size) and 359 (59.6%) procedures found at least one colorectal polyp but did not fulfil current 'high-risk' criteria. Findings were similar at episode level with 90 (15.7%) of 574 BCSP episodes that occurred during the posttrial follow-up period being classified as 'high risk'.
In total, 1298 'polyps' were recorded, of which 936 (72.1%) were adenomas and 273 (21.0%) were serrated-hyperplastic polyps (Table 1). Current BCSP guidance is clear that diminutive (≤5 mm) rectal, hyperplastic-looking polyps are not recorded, 7 but practice was more variable during the seAFOod trial and the follow-up period encompassed by this study. Therefore, a small number (n = 15) of diminutive rectal hyperplastic polyps were recorded in the BCSS and are included in the analysis (Table 1)

| The effect of seAFOod trial interventions on post-trial colorectal polyp outcomes
Respective active versus placebo intervention groups were wellmatched for the duration of post-trial follow-up and the number of surveillance procedures that had been performed per person (  Table S2).
Although the groups that were defined by the prior trial interven-  Figure S2 and

| Evidence for an interaction between aspirin and EPA for reduction in total colorectal polyp risk
In addition to the pre-specified analysis of colorectal polyp outcomes in the seAFOod trial, we noted that the number of colorectal polyps detected in individuals who had been randomised to both aspirin and EPA was lower than the other three treatment groups.  Table 4).
In view of the unexpected interaction between aspirin and EPA for the reduction in colorectal polyp risk, analysis of the post-trial colonoscopy data was also performed 'inside the table' across all four treatment groups (Table S4). 1,2 Results mirrored the 'at the margins' analysis of the individual interventions shown in Table 2 with an increase in PDR for total colorectal polyps and distal colorectal polyps in the treatment combinations that included active aspirin compared with the group that received placebo aspirin and placebo EPA only (Table S4). A similar relationship with previous single-agent aspirin use was also observed for increased distal colorectal polyp number (  These data support the hypothesis that short-term aspirin use inhibits colorectal tumour growth, but not initiation, leading to derepression of polyp growth upon treatment cessation, reflected as 'rebound' increased colorectal polyp risk during subsequent colonoscopic surveillance. This relationship was restricted to aspirin, which had chemopreventative efficacy against adenomatous and serrated TA B L E 2 Comparison of colorectal polyp outcomes during post-trial colonoscopic surveillance between individuals who received either active intervention or its respective placebo in the 2 × 2 factorial seAFOod trial.  The effect size (a 10%-15% increase in PDR and MPP in those allocated aspirin, as opposed to placebo during the seAFOod trial)

No aspirin (n = 256) Aspirin (n = 251) p
is clinically meaningful based on a similar effect size for colorectal polyp risk reduction in aspirin polyp prevention trials that is coupled with observational data on CRC risk reduction, 3,10,11 as well as PDR differences reported for endoscopic interventions that improve colorectal polyp detection, which have been adopted into clinical practice. 12 Secondary colorectal polyp size and polyp burden analyses concurred with the PDR outcomes in that trial allocation to aspirin was associated with larger colorectal polyps, including serrated lesions.
In general, serrated-hyperplastic polyps detected during surveillance were larger than adenomatous polyps. This observation might be explained by a faster growth rate and/or higher colonoscopy miss rate for serrated lesions, 13  39.9%; relative risk 1.00 [95% CI 0.80-1.24]). 18 No colorectal polyp number outcomes were reported in the AFPPS follow-up study. 18 The seAFOod trial recruited a 'high risk' cohort with greater colorectal polyp multiplicity at trial entry and post-trial colonoscopic surveillance continued in a quality-assured national programme (that is reflected in the much higher PDR [70%-80%] during follow-up in the BCSP compared with the AFPPS follow-up study),

TA B L E 3
Comparison of total colorectal polyp number in individuals who received combined aspirin and EPA treatment in the seAFOod trial with participants who received either agent alone or placebos only. which could explain the discrepancy in results between the two studies. 1,2 Overall, the yield of advanced neoplasia was low, with no CRCs detected during the entire surveillance period. Advanced colorectal polyp detection during post-trial surveillance was similar in individuals who had received aspirin, as opposed to placebo aspirin, during the seAFOod trial. This is consistent with the observation from posttreatment follow-up of the APPROVe trial that prior use of the selective COX-2 inhibitor rofecoxib was associated with increased risk of any colorectal adenoma, but not advanced adenomatous polyps. 15   A consistent methodological weakness of post-trial observational follow-up studies, which is shared by our study, is the absence of data on post-trial aspirin usage. At the end of the seAFOod trial, participants were not informed of their treatment allocation and the Plain English Summary of the trial results, which was available online and on request, did not recommend aspirin and/or EPA use.
Therefore, one would not expect widespread aspirin use in previous seAFOod trial participants (clinically indicated aspirin use was an exclusion criterion) or that post-trial aspirin use would be unbalanced across the randomised seAFOod trial treatment groups. Moreover, any bias related to post-trial aspirin use would be expected to reduce the 'rebound' effect size compared with the placebo arm based on the reduction in colorectal polyp number associated with current aspirin use. 1,2 Another important observation is that combination treatment with aspirin and EPA was associated with significantly lower colorectal polyp recurrence in the seAFOod trial compared with either agent alone. This was apparent in the primary report of the seAFOod trial, 1,2 but an 'inside the table' analysis of the seAFOod trial data was not pre-specified based on the understanding of the mechanism(s) of action of aspirin and EPA, at the time, which focused on inhibition of COX by both agents, as well as observation that there was no interaction between the two agents when anti-platelet (COX-1-dependent) activity is measured. 22 The pharmacological basis of an interaction between aspirin and with the exclusion of diminutive rectal hyperplastic polyps from the BCSP risk stratification algorithm. 7 However, we continued to describe colorectal polyp outcomes according to the terminology and reporting tools employed by the BCSP at the time, in order to allow direct comparison between on-trial and post-trial colorectal polyp outcomes.
Study limitations include the lack of data on post-trial aspirin and omega-3 polyunsaturated fatty acid use, absence of data on incident co-morbidities and other drug use, as well as a relatively short post-trial follow-up period, which encompassed only one 'intermediate risk' 3-year colonoscopy for the majority of trial participants.
Unfortunately, seAFOod trial participants were not asked to provide consent for future contact at trial entry so we have been unable to contact individuals to enquire about concurrent drug use and permission to access BCSP records at the end of the pre-determined 6-year follow-up period.
Despite the fact that the post-trial surveillance groups, which were defined by randomised trial treatment allocation, were well- We suggest that future biomarker-driven CRC chemoprevention trials include prospective post-intervention follow-up to corroborate our findings that a short-term intervention (aspirin ± EPA), which is associated with reduced on-treatment colorectal polyp risk, is subsequently linked to 'rebound' increased colorectal polyp incidence when the intervention is stopped. A 'rebound' increase in colorectal polyp risk (and possible increased CRC risk, albeit after a much longer latent period) after short-term aspirin use should be considered as part of the ongoing debate about how best to harness the undoubted cancer-preventative properties of aspirin whilst minimising risk in a precision manner, 27 one approach to which is to stop aspirin use in line with age-dependent elevated risk of bleeding. Morris: Conceptualization (equal); data curation (equal); funding acquisition (equal); investigation (equal); methodology (equal); resources (equal); writing -review and editing (equal). Mark A. Hull:

ACK N O WLE D G E M ENTS
The authors wish to thank all the participants and research staff who