Results

Parts of this chapter have been reproduced from Hull et al.1 © 2018 The Author(s). Published by Elsevier Ltd. This is an Open Access article under the CC BY 4.0 license (https://creativecommons.org/licenses/by/4.0/).

Recruitment and follow-up

Between November 2011 and June 2016, 3911 ‘high-risk’ individuals were screened for eligibility, of whom 3202 (82%) were not randomised (Table 6 and Figures 2–4). The mean duration from receipt of local research and development (R&D) approval to FPFV for the 53 sites that randomised patients was 6.3 months. Conversion from screening to randomisation of patients did not appear to be influenced by the size of the site or the number of patients undergoing screening colonoscopy. Sites did report that the time taken to undertake trial screening and recruitment (as well as limited availability of, and changes to, research staff) were reasons for low recruitment.

FIGURE 2 Flow of participants through the trial. a, AE was an option to discontinue from the trial in an earlier version of the CRF; however, this option was removed in later versions. HNPCC, hereditary non-polyposis colon cancer. (Figure not reproduced here; see the original article.)

FIGURE 3 Cumulative trial recruitment by month. (Figure not reproduced here; see the original article.)

FIGURE 4 Actual monthly trial recruitment. (Figure not reproduced here; see the original article.)

Overall, 2179 (56%) patients met one or more of the exclusion criteria (see Figure 4). The other 1023 individuals either did not wish to take part in the trial or were not randomised for unknown reasons. The most frequent reasons for no randomisation were not being willing to give written consent [n = 981 (25%)], taking more than three doses per week of a prescribed/OTC aspirin/NSAID [n = 594 (15%)], the need for more than one repeat colonoscopy or FS [n = 328 (8%)] and known bleeding diathesis or anticoagulant use or antiplatelet therapy [n = 313 (8%)]. Overall, only 18% (n = 709) of screened ‘high-risk’ individuals were randomised (see Figure 2).

By 12 June 2016 (the date stipulated by the MHRA as the end date for randomisation related to the approved shelf life of the second capsule IMP (90% EPA-TG), 709 patients had been randomised, which was 83% of the target (n = 853). Reduced recruitment between February and October 2014 (see Table 6 and Figures 3 and 4) was related to diminishing stocks of active and placebo EPA-FFA at sites, prior to the introduction of 90% EPA-TG IMP in November 2014. The mean number of randomisations per month during the whole recruitment phase of the trial was 13. Excluding the run-in period of 12 months, during which sites opened at a variable rate, and the period when capsule IMP stocks were diminishing (February to October 2014), the mean number of randomisations per month was 16 (range 7–26 per month).

Of the 709 participants randomised, 177 participants were randomised to receive EPA + aspirin, 179 were randomised to receive EPA + placebo aspirin, 177 were randomised to receive placebo EPA + aspirin and 176 were randomised to receive placebo EPA + placebo aspirin (see Figure 2). Two participants withdrew immediately after randomisation; one was randomised to EPA + placebo aspirin and one was randomised to placebo EPA + aspirin (see Figure 2).

A total of 422 (60%) participants were randomised to active or placebo EPA-FFA and 287 (40%) participants were randomised to active or placebo 90% EPA-TG. Participants took only one form of capsule IMP (EPA-FFA or EPA-TG) in all cases.

A total of 641 participants (90% of those randomised) underwent surveillance colonoscopy, with endoscopic data being available for 640 of these participants (see Figure 2). The main reasons for not performing surveillance colonoscopy were withdrawal of consent and loss to follow-up (Table 7 and see Figure 2). The ‘other’ reasons for discontinuation in the trial were the participant was withdrawn because of being randomised in error (n = 3), the histology report confirmed a malignant polyp and the participant was referred to MDT for review, failure to comply with trial treatment as a result of ongoing medical problems, the participant being prescribed aspirin and development of cancer (all n = 1).

Baseline characteristics

Baseline characteristics were well balanced across all four treatment groups with respect to demographic data, medical history and baseline colorectal adenoma details (Table 8).

The mean age of participants was 65 years. A male-to-female ratio of approximately 4 : 1 was observed, which reflects the sex distribution of ‘high-risk’ patients in the English BCSP.95,99 A total of 581 (82%) participants were overweight; of these, 270 (38%) were obese. Despite the high prevalence of excess body weight, the prevalence of known type 2 diabetes was low, and only approximately half of the participants were on regular prescribed drugs at trial entry; the drug was usually a statin, proton pump inhibitor or metformin. Baseline drug use was balanced across the treatment groups.

There was a slight imbalance across groups with respect to smoking status, with fewer current smokers in the EPA + placebo aspirin group.

Overall, 158 (22%) participants had a medical history of abdominal pain or dyspepsia, which is consistent with the high prevalence of these symptoms in the general population.113 Ten per cent of participants recorded a prior episode of bleeding and 20 (3%) reported halitosis at baseline. A medical history of symptoms and clinical events at baseline relevant to EPA and/or aspirin therapy was balanced across the treatment groups. Fewer participants reported diarrhoea at trial entry in the EPA + aspirin group than in the other three groups.

‘High-risk’ participants had a mean of five colorectal adenomas and one advanced colorectal adenoma at entry screening colonoscopy. Of 3531 evaluable colorectal adenomas at baseline, 3360 (97%) were conventional (tubular/tubulo-villous and villous) and 91 (3%) were serrated adenomas. Overall, 584 (83%) ‘high-risk’ participants had at least one colorectal adenoma proximal to the splenic flexure at screening colonoscopy. Screening colonoscopy findings were balanced across the treatment groups (see Table 8).

Compliance with the allocated intervention

Compliance with both capsule and tablet IMPs was excellent, with mean percentage compliance levels, calculated by capsule/tablet counting, of between 94% and 97%. A total of 10 participants never took any dose of either capsules or tablets and a further seven participants took only one of the treatments (EPA/placebo EPA or aspirin/placebo aspirin) (Table 9).

Concomitant medication during the trial

During the trial intervention phase, approximately one-quarter of participants started regular prescribed medication, in addition to existing drugs, which included a statin, calcium + vitamin D supplement, metformin, proton pump inhibitors, aspirin, fish oil and a non-aspirin NSAID. Slightly more participants started regular, concomitant medication in the placebo + placebo group (Table 10).

Dietary fish intake before trial participation

Most participants completed a FFQ at baseline; there were similar percentages across dietary fish intake categories in each of the treatment groups. The total fish and oily fish intake at baseline was similar across the four treatment groups (Table 11).

Polyunsaturated fatty acid levels in trial participants

Given the importance of the comparison of EPA levels in individuals randomised to either FFA or TG formulations of EPA, to determine if it was appropriate to pool the primary and secondary outcome data for these groups, PUFA data are presented below before the analyses for the primary and other secondary outcomes.

The RBC EPA level was measured as an accepted biomarker of tissue EPA exposure, as well as the rectal mucosal EPA level measured at the end of the intervention period.

Red blood cell EPA and rectal mucosal EPA levels were compared between each treatment group, at each time point and between users of EPA-FFA and EPA-TG formulations. As expected, active EPA groups had higher RBC levels of EPA than placebo EPA users after the intervention began (Figures 5–7). RBC EPA levels were similar between participants, who received either active FFA or TG formulations of EPA, at baseline, mid-treatment and at the end of the intervention period (see Figures 5–7). Moreover, the increase in RBC EPA level from baseline to the 6-month and 12-month time points for all participants was similar across the two EPA formulations and for rectal mucosal EPA levels at 12 months (Table 12) (see Figures 5–7). There was also no clear difference in RBC or rectal mucosal DHA content between EPA-FFA and EPA-TG users (Table 13; see also Table 12).

FIGURE 5 The RBC EPA levels by formulation and by trial group at baseline. (Figure not reproduced here; see the original article.)

FIGURE 6 The RBC EPA levels by formulation and by trial group at 6 months. (Figure not reproduced here; see the original article.)

FIGURE 7 The RBC EPA levels by formulation and by trial group at 12 months. (Figure not reproduced here; see the original article.)

Rectal mucosal EPA levels at the end of the intervention period were higher in those who received EPA-TG than in those who received EPA-FFA, but with substantial overlap between the two groups (Figure 8). This was not reflected in the rectal mucosal EPA-to-AA ratio at the end of the intervention period, which was similar for both EPA-FFA and EPA-TG users (Table 14).

FIGURE 8 Rectal mucosal EPA levels by formulation and by trial group at 12 months. (Figure not reproduced here; see the original article.)

As there was no clear difference in RBC or rectal mucosal EPA incorporation, or EPA-to-AA ratio, between those allocated EPA-FFA and those allocated EPA-TG, it was felt appropriate to combine primary and secondary outcome data from those who received either type of capsule IMP.

Analysis populations

The ITT population was defined as all randomised participants with post-randomisation data (Table 15). Participants in the ITT population were analysed regardless of adherence to their allocated group and without imputation for missing data. Although 709 participants were randomised, two participants withdrew immediately and provided no data. Therefore, 707 participants were included in the ITT population (see Table 15). All baseline summaries and efficacy analyses were based on this population. Data from surveillance colonoscopy at 12 months was available for 640 participants.

The per-protocol population included all participants who had taken sufficient medication (participants who had taken > 75% of expected capsules and/or > 50% of expected tablets), were not found to be ineligible post randomisation and had not used any OTC medication containing aspirin, NSAIDs or fish oil during the treatment period. The population comprised 659 participants and was used for one of the sensitivity analyses to determine the robustness of the primary analysis.

The safety population was defined as all participants who took at least one dose of allocated treatment. It comprised 697 participants. The 10 participants who did not receive active or placebo EPA or aspirin were excluded from this population.

Primary outcome

Colorectal adenoma data from the 12-month surveillance colonoscopy were available for 640 participants: 161 (91%), 153 (86%), 163 (93%) and 163 (93%) (percentage of the total number of participants in each group) in the EPA + aspirin, EPA + placebo aspirin, placebo EPA + aspirin and placebo EPA + placebo aspirin groups, respectively. The median time between randomisation and the 12-month surveillance colonoscopy was between 344 and 348 days in the four treatment groups (Table 16).

Of 161 participants in the EPA + aspirin group, 98 (61%) had at least one colorectal adenoma (the ADRa) at the surveillance colonoscopy, whereas 97 out of 153 participants (63%) in the EPA + placebo aspirin group had at least one colorectal adenoma (the ADRa) at the surveillance colonoscopy. The ADRa was 61% (100/163) in the placebo EPA + aspirin group and 61% (100/163) in the placebo EPA + placebo aspirin group.

When summarised according to factorial margins, the ADRa was similar across interventions, with an ADRa of 195 (62%) for those who received active EPA versus an ADRa of 200 (61%) for those who did not receive EPA, and an ADRa of 198 (61%) for individuals who received active aspirin versus an ADRa of 197 (62%) for those who did not receive aspirin (62%) (see Table 16).

Primary outcome analysis

The test of interaction showed that there was no evidence of any interaction between EPA and aspirin for the ADRa (p = 0.85). Therefore, primary and secondary outcomes were analysed according to factorial margins, that is the treatment effects for EPA and aspirin were reported separately.

Table 17 shows adjusted risk differences of the ADRa for the treatment effect of EPA and aspirin. The point estimates and 95% CIs for both EPA and aspirin showed that there was no evidence of a statistically significant difference. These analyses were adjusted for whether or not the participant had a repeat endoscopic procedure (i.e. full colonoscopy, partial colonoscopy or FS) and included BCSP site as a random effect. Supportive analyses that were not adjusted by repeat colonoscopy showed similar results (see Appendix 4, Table 40).

The point estimates for EPA and aspirin were –0.9% and –0.6%, respectively, with 95% CIs that included zero, indicating no statistically significant difference from no treatment for both interventions. The trial was designed to detect an absolute ADRa difference of 10%. Figure 9 and Table 17 show that the lower limit of the 95% CIs did not reach –10% for either EPA or aspirin.

FIGURE 9 Forest plot of the treatment effect of EPA and aspirin on the ADRa. The blue vertical line represents the prespecified target reduction of ADRa (10%). Adjusted by site as a random effect and by repeat colonoscopy at baseline. (Figure not reproduced here; see the original article.)

Sensitivity analyses for the primary outcome

Several sensitivity analyses were conducted to investigate the robustness of the primary analysis. Although the point estimates and 95% CIs varied between different analyses, they were supportive of the primary analysis (Table 18 and Figure 10).

FIGURE 10 Forest plots for the sensitivity analyses of the ADRa. (a) EPA vs. placebo; (b) aspirin vs. placebo; and (c) multilevel model. Results using the multilevel model are presented as odds ratios, whereas all other sensitivity analyses are presented using the risk difference. Adjusted by site as a random effect and by repeat colonoscopy at baseline. Some analyses were adjusted by further variables. (Figure not reproduced here; see the original article.)

The multilevel model included both BCSP centre and site as random effects to account for sites embedded within the same BCSP (see Figure 10).

Secondary outcomes

Secondary outcomes were analysed in a similar way to the primary outcome. Results were summarised according to the outcome type, that is the risk difference for the binary outcome (ADRa) and the IRR for count outcomes (Tables 19–25 and Figure 17). Point estimates and CIs reported are according to factorial margins, that is EPA compared with no EPA and aspirin compared with no aspirin.

FIGURE 11 Distribution of the total number of colorectal adenomas per participant at 12 months by trial group. (a) EPA + aspirin; (b) EPA + placebo aspirin; (c) placebo EPA + aspirin; and (d) placebo + placebo. (Figure not reproduced here; see the original article.)

Summary data on colorectal adenomas reported at surveillance colonoscopy are listed in Table 19.

The median number of colorectal adenomas per participant at the 12-month surveillance colonoscopy was 1 for all four trial groups (see Table 20). The median incidence rate was 1 colorectal adenoma per person per year for all four groups. However, fewer colorectal adenomas were detected in the combination EPA + aspirin group (see Table 19), with a reduced total MAP value (see Table 20).

The factorial margin analysis revealed that the IRR for EPA versus no EPA was 0.91 (95% CI 0.79 to 1.05) in favour of EPA (see Table 20). The IRR for aspirin versus no aspirin was 0.78 (95% CI 0.68 to 0.90). The distribution of individual total colorectal adenoma counts was similar across all four groups (see Figure 11).

The advanced ADRa was 8 (5%) in participants who received both EPA and aspirin, as well as EPA + placebo aspirin, but was 10 (6%) and 11 (7%) in participants who received placebo EPA + aspirin and placebo EPA + placebo aspirin, respectively (see Table 21). Risk differences for the advanced ADRa for EPA and aspirin were –0.6% and –0.3%, respectively. The mean incidence rate was 0.1 advanced colorectal adenomas per person per year. Analysis of the advanced MAP ‘at the margins’ revealed IRRs of 0.82 for EPA and 0.99 for aspirin, but with 95% CIs crossing unity. The distribution of individual advanced colorectal adenoma counts was similar across all four groups (Figure 12).

FIGURE 12 Distribution of the number of advanced colorectal adenomas per participant at 12 months by trial group. (a) EPA + aspirin; (b) EPA + placebo aspirin; (c) placebo EPA + aspirin; and (d) placebo + placebo. (Figure not reproduced here; see the original article.)

The conventional ADRa was 55% for EPA + aspirin (88/161), 54% for EPA + placebo aspirin (83/153) and 56% for both placebo EPA + aspirin (91/163) and placebo EPA + placebo aspirin (92/163) (see Table 22). Risk differences for EPA and aspirin were –3.3% and 1.7%, respectively. The median incidence rate was 1 conventional adenoma per person per year for all four groups, with a lower mean value for the combined EPA + aspirin group (see Table 22). IRRs for the number of conventional colorectal adenomas were 0.86 (95% CI 0.74 to 0.99) for EPA versus no EPA and 0.82 (95% CI 0.71 to 0.94) for aspirin versus no aspirin (see Table 22). The distribution of individual conventional colorectal adenoma counts was similar across all four groups (Figure 13).

FIGURE 13 Distribution of the number of conventional colorectal adenomas per participant at 12 months by trial group. (a) EPA + aspirin; (b) EPA + placebo aspirin; (c) placebo EPA + aspirin; and (d) placebo + placebo. (Figure not reproduced here; see the original article.)

The serrated ADRa values were small for all treatment groups: 4 (2%) for EPA + aspirin, 11 (7%) for EPA + placebo aspirin, 6 (4%) for placebo EPA + aspirin and 7 (4%) for placebo EPA + placebo aspirin (see Table 23). The risk differences for the serrated ADRa were 0% (95% CI –3.2% to 3.2%) for EPA and –2.7% (95% CI –6.1% to 0.7%) for aspirin. The median incidence rate was zero for all four groups and the mean incidence rate was 0.1 serrated adenomas per person per year for all trial groups except the EPA + aspirin group, which was zero. IRRs for the number of serrated adenomas were 1.44 (95% CI 0.79 to 2.60) for EPA versus no EPA and 0.46 (95% CI 0.25 to 0.87) for aspirin versus no aspirin (see Table 23). The distribution of individual serrated colorectal adenoma counts was similar across all four groups (Figure 14).

FIGURE 14 Distribution of the number of serrated colorectal adenomas per participant at 12 months by trial group. (a) EPA + aspirin; (b) EPA + placebo aspirin; (c) placebo EPA + aspirin; and (d) placebo + placebo. (Figure not reproduced here; see the original article.)

The distribution of individual left-sided colorectal adenoma counts in each treatment group is shown in Figure 15. Risk differences for the number of participants with at least one left colorectal adenoma (left ADRa) were –7.8% for EPA versus no EPA (95% CI –15.5% to –0.2%) and –1.8% for aspirin versus no aspirin (95% CI –9.4% to 5.8%). The mean incidence rate ranged from 0.4 to 0.7 left colorectal adenomas per person per year. IRRs for left MAP were 0.75 (95% CI 0.60 to 0.94) for EPA versus no EPA and 0.85 (95% CI 0.69 to 1.06) for aspirin versus no aspirin (see Table 24).

FIGURE 15 Distribution of the number of left colorectal adenomas per participant at 12 months by trial group. (a) EPA + aspirin; (b) EPA + placebo aspirin; (c) placebo EPA + aspirin; and (d) placebo + placebo. (Figure not reproduced here; see the original article.)

For the number of participants with at least one right-sided colorectal adenoma (right ADRa), the risk differences were 6% (95% CI –1.9% to 13.9%) for EPA versus no EPA and –3.1% (95% CI –11% to 4.7%) for aspirin versus no aspirin. The mean incidence rate ranged from 0.7 to 1.0 right colorectal adenomas per person per year. IRRs were 1.02 (95% CI 0.85 to 1.22) for EPA versus no EPA and 0.73 (95% CI 0.61 to 0.88) for aspirin versus no aspirin (see Table 25). The distribution of individual right-sided colorectal adenoma counts in each treatment group is shown in Figure 16.

FIGURE 16 Distribution of the number of right colorectal adenomas per participant at 12 months by trial group. (a) EPA + aspirin; (b) EPA + placebo aspirin; (c) placebo EPA + aspirin; and (d) placebo + placebo. (Figure not reproduced here; see the original article.)

Following the 12-month surveillance colonoscopy, the majority of BCSP patients were reclassified as being at intermediate risk: 146 (91%) participants were reclassified as being at intermediate risk at follow-up in the EPA + aspirin group, compared with 128 (84%) in the EPA + placebo aspirin group, 140 (86%) in the placebo EPA + aspirin group and 147 (90%) in the placebo EPA + placebo aspirin group (Table 26). The risk differences for EPA versus no EPA and aspirin versus no aspirin were small (–0.2% and 0.9%, respectively).

Adjusted IRRs and 95% CIs for the secondary MAP data are summarised in Figure 17. Risk differences and 95% CIs for the secondary ADRa data are summarised in Figure 18.

FIGURE 17 Forest plots for secondary MAP analysis. (a) EPA vs. placebo; and (b) aspirin vs. placebo. Adjusted by site as a random effect and by repeat colonoscopy at baseline. (Figure not reproduced here; see the original article.)

FIGURE 18 Forest plots for secondary ADRa analysis. (a) EPA vs. placebo; and (b) aspirin vs. placebo. Adjusted by site as a random effect and by repeat colonoscopy at baseline. (Figure not reproduced here; see the original article.)

Although potential trends may have been observed in the secondary colorectal adenoma data, consideration needs to be given to the large number of analyses that were undertaken (and, therefore, the potential for spuriously significant results), the small sample numbers for some of the colorectal adenoma subtypes and the potential dependencies between the variables.

Number of participants with colorectal cancer detected prior to or at first surveillance colonoscopy

There was no report of any CRC detected at the 12-month surveillance colonoscopy or during the intervention phase of the trial.

Dietary fish intake during trial participation

Approximately 80% of participants completed a FFQ at 12 months, with similar percentages in each of the four treatment groups (Table 27). The total fish and oily fish intakes at 12 months were similar across the four treatment groups (see Table 27), as was the proportion of individuals changing intake level between baseline and the end of trial participation (Table 28). Approximately 50% of each treatment group remained at the same level of oily fish intake during the trial.

Safety

The safety population comprised 697 participants who received at least one dose of EPA or aspirin, or their respective placebos (see Table 15).

Overall, there were no safety concerns in participants receiving either EPA or aspirin. There was an excess of AEs and SAEs in participants receiving EPA + placebo aspirin (Table 29). In this group, 13 participants reported five or more (87 in total) AEs, which resulted in this imbalance (see Table 29). A larger proportion of participants receiving EPA + placebo aspirin reported at least one ADR, with 57 (32%) individuals reporting 119 ADRs. This excess was contributed to by five participants who reported five or more ADRs. Fewer SAEs were reported in the EPA + aspirin group. Five (3%) participants reported at least one SAE, compared with 7% who reported at least one SAE in all the other treatment groups. Only nine SAEs were felt to be related to trial medication.

The majority of AEs and ADRs were mild in severity in each of the treatment groups.

Gastrointestinal disorders were the most frequently reported AE (Table 30), with 209 participants reporting at least one GI AE. A larger number of GI AEs were reported in the EPA + placebo aspirin group, contributed by an excess of individuals who reported multiple GI AEs. However, the distribution of GI AEs was similar across the other three groups (Table 31).

There was an excess of diarrhoea, abdominal pain and nausea in those allocated active EPA (Table 32). The excess of mild to moderate diarrhoea was most prominent in the group receiving EPA alone. There did not appear to be any consistent differences in the reporting of GI AEs between individuals receiving EPA-FFA and those receiving EPA-TG (see Table 32).

Post database lock and code release, the chief investigator summarised all GI AEs according to symptoms that are commonly associated with fish oil intake, categorised as diarrhoea, upper GI symptoms, lower abdominal symptoms, eructation/halitosis and other. These were summarised by the treatment group and EPA formulation that the participants received. There were no notable differences between the formulations in each trial treatment group (see Appendix 5).

Similar to the profile of AEs, there were more ADRs reported in the EPA + placebo aspirin group than in the other three groups (Table 33). The most commonly reported ADRs across all treatment groups were GI disorders. The distribution of GI ADRs across the four groups was similar (Table 34) (see Appendix 4, Tables 41 and 43).

There were a small number of SAEs reported (Table 35) (see Appendix 4, Table 42). The most frequently reported SAEs were cardiac and GI disorders. Among the nine cardiac disorders, five were episodes of atrial fibrillation (AF) and all of these were reported in the EPA + placebo aspirin group. Three were reported by one participant and two additional participants reported one each. Three myocardial infarctions were reported by participants: one in the EPA + placebo aspirin group, two in the placebo EPA + aspirin group and one in the placebo + placebo group. One participant in the placebo EPA + aspirin group had an arrhythmia.

Eight ADRs led to trial discontinuation for four participants, one from each of the four trial groups. The eight ADRs came from six preferred term names (Table 36).

Clinically significant acute upper gastrointestinal bleeding episodes

No haemorrhagic strokes were reported during the trial. A manual search of AEs and SAEs revealed six acute upper GI bleeding events that were considered by the chief investigator to be of clinical significance. These were ‘oesophageal haemorrhage’ (in the placebo EPA + aspirin group), ‘gastro-oesophageal reflux disease’ (in the placebo EPA + aspirin group) and ‘alcohol withdrawal syndrome’ (in placebo EPA + placebo aspirin group), which were all reported as SAEs, and ‘gastric haemorrhage’ (in the EPA + placebo aspirin group) and two cases of ‘melaena’ (in the EPA + placebo aspirin and placebo EPA + aspirin groups), which were recorded as AEs. It is possible that one of the cases of melaena (in a participant receiving EPA + placebo aspirin) was a SAE because the participant had been hospitalised, but this event was recorded specifically as an AE by the site. All tables reflect this categorisation.

Deaths

One death was reported during the trial. During contact made to arrange visit 5, site staff were made aware that a participant had died from bladder cancer, which was deemed to be unrelated to the intervention.

Protocol deviations

Protocol deviations were reported for between 64% and 73% of participants in each treatment group. Most deviations were judged to be minor, with the majority being trial visits outside the time window. The number and types of deviations appeared similar between treatment groups.

There were 11 deviations related to randomisation error (Table 37). Ten of these were as a result of participants being ineligible, and one participant was randomised prior to consent. None of these errors occurred in participants who were randomised to the placebo + placebo group. There was no reason to suspect that the deviation was related to the treatment group to which the participant had been randomised.

Exploratory analyses

Colorectal adenoma size

Given the decrease in total colorectal adenoma number associated with EPA and aspirin treatment, colorectal adenoma size was analysed consistent with size analysis in previous RCTs of chemoprevention agents in FAP patients.55,67 Colorectal adenoma size was summarised within each participant across each treatment group. Analyses were based on the within-participant mean value (Table 38).

The adjusted mean difference between the EPA and no EPA groups was –0.47 mm (i.e. the mean adenoma size was smaller in those receiving EPA than in those not receiving EPA). The mean adjusted difference between the aspirin and no aspirin groups was 0.42 mm (i.e. the mean adenoma size was bigger in those receiving aspirin than in those not receiving aspirin). There was no statistically significant difference between EPA and placebo users, or between aspirin and placebo users.

Relationship between individual colorectal adenoma number and eicosapentaenoicacid levels

The relationship between the change in RBC EPA levels at 12 months from baseline and secondary outcomes, according to EPA factorial margin, was also investigated descriptively by plotting individual values. There was no evidence of a clear relationship between individual increase in RBC EPA level and total colorectal adenoma number (Figure 19). This was also the case for the rectal EPA level at 12 months. The fact that there were three ‘outlier’ individuals in the placebo EPA group who had a large increase in RBC EPA level during the intervention phase suggests that ‘contamination’ by own-use ω-3 PUFA intake may have occurred in these cases (see Figure 19).

FIGURE 19 Change in RBC EPA levels at 12 months from baseline against total number of colorectal adenomas per participant by EPA treatment groups. (a) Placebo EPA; and (b) active EPA. (Figure not reproduced here; see the original article.)

Relationship between red blood cell and rectal mucosal eicosapentaenoic acid levels in participants

The relationship between RBC and rectal mucosal EPA levels at 12 months was of moderate strength, with a correlation coefficient of 0.455 (Figure 20).

FIGURE 20 Scatterplot of the individual rectal mucosal EPA level against the corresponding RBC EPA level at 12 months. (Figure not reproduced here; see the original article.)