Methods

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/).

The trial protocol has been published in open-access form.100

Trial design

The seAFOod polyp-prevention trial was a randomised, blinded, placebo-controlled 2 × 2 factorial trial. The trial was designed to integrate fully into the screening and surveillance phases of the BCSP in England so that participation would not alter routine clinical practice.

Participants were randomised to one of four groups to receive EPA-FFA or EPA-TG (both 2 g of FFA equivalent daily) daily with food, or identical placebo, AND 300 mg of enteric-coated aspirin daily taken with food, or identical placebo, until the day before surveillance colonoscopy (at 12 months). The primary outcome was the number of individuals with one or more colorectal adenomas at the surveillance colonoscopy (the ADRa). Secondary outcomes included the total number of colorectal adenomas per participant, subtype (advanced, conventional, serrated, left and right) of adenomas (ADRa and number), the number of participants reclassified as being at intermediate risk for future surveillance, EPA and other PUFA levels in RBCs and rectal mucosa, dietary fish intake and assessment of the tolerability and safety of EPA (both FFA and TG formulations) alone and in combination with aspirin.

Trial setting and participants

NHS Bowel Cancer Screening Programme recruiting sites

The English BCSP is organised into local centres of a variable number (usually 1–3) of individual BCSP sites (hospitals undertaking endoscopy), which receive referrals for screening colonoscopy after guaiac FOBt analysis at five regional hubs. Participants identified as ‘high risk’ at participating sites were randomised and followed to surveillance colonoscopy at 12 months. The trial was integrated into the BCSP to utilise routine clinical pathways in order to collect quality-assured data from screening and surveillance colonoscopies.

During the trial, 61 BCSP sites were opened. The date of the first participant, first visit (FPFV) was 11 November 2011; the last participant in was 10 June 2016; and the last participant, last visit (LPLV) was 8 June 2017. Despite approval from the BCSP Research Advisory Committee, widespread engagement from SSPrs and Clinical Research Network (CRN)-funded research nurses (RNs) and a trial extension in 2014, the trial did not recruit to target, recruiting 709 participants against a revised target of 755. The sample size calculation remained at 853, based on the predicted effect size of the interventions, giving 80% power. However, recruitment figures achieved prior to the 2014 extension, the limited recruitment period mandated by the funder and the limitations set by the expiry date on the capsule (EPA-TG) investigational medicinal product (IMP) suggested that recruitment of 755 individuals would be feasible in the extended intervention period (see Statistical methods).

The original strategy was to open 15 recruiting sites from two BCSP hubs (North-East and Eastern). This was based on data from the national BCSP database that suggested that each BCSP centre could identify approximately 50 high-risk patients per year. Trial screening data supported this assumption; however, the eligibility rate was much lower than expected, primarily because of a higher than expected number of screened high-risk individuals who were excluded because of the need for repeat colonoscopy or FS to check for adenoma excision within a 3-month window. Analysis of up-to-date BCSP data in early 2012 revealed that the number of cases requiring repeat endoscopy as part of routine BCSP care had increased nationally during the grant application and set-up phases of the trial. It was determined that a second colonoscopy or FS did not significantly alter the overall ADRa at the 12-month surveillance colonoscopy (see Changes to the protocol). Therefore, the protocol was amended (version 4.0, dated 24 May 2012)100 to include these patients, without loss of statistical power.

A trial site expansion strategy was also implemented in 2012, increasing the number of recruiting sites across England, from Cornwall to Cumbria, to 60 (representing ≈50–60% of English BCSP centres). In late 2015, a decision was made to add one further site that had expressed a strong interest to be involved in the trial.

Delays were experienced in gaining NHS trust research and innovation (R&I) approvals as a result of a general lack of Good Clinical Practice accreditation and research training for BCSP staff, many of whom had not previously contributed to a Clinical Trial of an Investigational Medicinal Product (CTIMP). Sites were supported by the Nottingham Clinical Trials Unit (NCTU) and local CRNs to access training. The median time to gain R&I approval at trial sites was 11.5 (range 4–19) weeks. In addition to a site initiation visit, supplementary training was provided to sites by the co-ordinating centre for participating local investigators, RNs and SSPrs via an instructional video.

In February 2014, recruitment was disrupted significantly when the manufacturer of the original capsule IMP (EPA-FFA) was no longer able to provide stock for the trial. Until an alternative capsule IMP could be identified, approved, manufactured and distributed, sites continued to recruit until local stock was exhausted, at which point that site temporarily suspended recruitment. To maximise recruitment during this period, the top eight most active sites were prioritised for allocation of remaining central stock of capsule IMP. Stock management also ensured that all participants completed the intervention phase of the trial using the same EPA formulation (FFA or TG). This strategy enabled the trial to continue recruiting between February and October 2014, after which a new capsule IMP became available. Partly because of this delay, a 36-month extension was approved (in October 2014) by the NIHR EME board in order to complete trial recruitment.

Identification of participants

Individuals identified as ‘high risk’ at screening colonoscopy on the basis of colorectal adenoma number and (endoscopic) size, and confirmed later by the histopathology report, were screened for the trial and approached by a member of the site research team to determine whether or not they were interested in trial participation. For those participants who had a bowel scope FS, summated colorectal adenoma findings were used to define individuals as ‘high risk’.

All ‘high-risk’ participants were given written trial information on discharge by a BCSP SSPr or RN. They were provided with a patient information leaflet (PIL), given a verbal explanation of the trial and given the opportunity to ask questions.

Participants were able to discuss the trial with their family, friends and/or health-care professionals before they attended a routine BCSP outpatient follow-up visit 7–14 days after screening colonoscopy. All participants provided written informed consent.

Eligibility criteria

Patients were eligible for inclusion if they were aged 55–73 years and were a BCSP participant identified as ‘high risk’ (i.e. they had five or more small colorectal adenomas or three or more colorectal adenomas with at least one being ≥ 10 mm in diameter) at the first complete screening colonoscopy. This included participants who were identified as ‘high risk’ at colonoscopy after FOBt screening, or who were deemed ‘high risk’ on the basis of the combined findings from a bowel scope FS and subsequent full colonoscopy. If the first screening colonoscopy was defined as complete, the participant was immediately stratified as ‘high risk’. If the first colonoscopy was incomplete, that individual was required to have a second colonoscopy to complete the initial examination, after which both procedures were added together as the screening colonoscopy result for the purposes of BCSP surveillance and trial eligibility.

Patients were excluded from the trial if:

• They had a requirement for more than one repeat colonoscopy or FS within the BCSP 3-month screening window (see Changes to the protocol).
• They had a malignant change in a colorectal adenoma requiring management by a CRC MDT.
• They were regularly (i.e. more than three doses per week) taking prescribed or OTC aspirin or regularly (i.e. more than three doses per week) taking prescribed or OTC non-aspirin NSAIDs. This was not an exclusion criterion if the drug was self-prescribed and not recommended by a doctor and if the individual was willing to stop taking it for the duration of the trial.
• They had aspirin intolerance or hypersensitivity, including aspirin-sensitive asthma.
• They had had active peptic ulcer disease within 3 months or previous peptic ulcer (and were not on proton pump inhibitor prophylaxis).
• They had a fish or seafood allergy.
• They used or were planning to regularly use (i.e. more than three doses per week) fish oil supplements. This was not an exclusion criterion if the supplements were self-prescribed and not recommended by a doctor and if the individual was willing to stop taking them for the duration of the trial.
• They had a known clinical diagnosis or were a gene carrier of a hereditary CRC predisposition (e.g. FAP).
• They had Lynch syndrome, also known as hereditary non-polyposis colon cancer (HNPCC).
• They had a previous or planned colorectal resection.
• They had known bleeding diathesis or concomitant warfarin therapy or use of any other anticoagulant or antiplatelet agent.
• They had severe liver impairment.
• They had severe renal failure (i.e. creatinine clearance of < 10 ml/minute).
• They currently used methotrexate at a weekly dose of ≥ 15 mg.
• They were not able to comply with trial procedures and IMP use.
• They had a serious medical illness interfering with trial participation.
• They were taking part in another interventional clinical trial.
• They failed to give written informed consent.

Responsibilities for checking eligibility and obtaining informed consent could be delegated to the SSPr or RN according to site approvals, but eligibility was confirmed by the local principal investigator (PI) for all participants.

Changes to the protocol

Soon after FPFV on 11 November 2011, it became apparent from screening log activity at individual BCSP sites that the eligibility rate was ≈15–20%, rather than the 60% that had been predicted in the original recruitment projection. A contributing factor was the higher than expected use of other non-aspirin antiplatelet agents, such as clopidogrel, which was later added as an exclusion criterion in January 2012 (protocol version 3.1, dated 12 January 2012). Another major contributing factor was the larger than expected number of ‘high-risk’ individuals (≈25%) who required a repeat endoscopy after randomisation. Repeat endoscopy was originally an exclusion criterion because the primary end point might be confounded by colorectal adenoma detection and removal at an extra endoscopic procedure between the screening (index) and 1-year surveillance colonoscopies. However, subsequent analysis in April 2012 of 1189 ‘high-risk’ patients who underwent 1-year surveillance colonoscopy in 2010 in 26 BCSP centres did not support this notion. The overall ADRa at surveillance colonoscopy [including those who underwent repeat partial colonoscopy or FS (ADRa 54%), repeat full colonoscopy (ADRa 67%) or no repeat procedure (ADRa 63%)] was 62%, which was consistent with the value (60%) used in the original sample size calculation. Therefore, the protocol was amended to allow recruitment of those individuals who required no more than one repeat endoscopic procedure [either colonoscopy (full or partial) or FS] within a 3-month screening episode window (version 4.0, dated 24 May 2012).100 A number of other changes to details contained within the protocol are documented in the statistical analysis plan (SAP) (version 1.1, dated 24 August 2017). These are listed in Statistical methods.

During trial recruitment, the trial was included as part of a larger MRC-funded programme of research [Systematic Techniques for Assisting Recruitment to Trials (START)] to assess web-based recruitment strategies. A separate protocol was approved for the cluster randomised (by site) substudy of a web-based information tool and the PIL was updated accordingly. Details of this study and the seAFOod trial contribution will be reported separately by the MRC-START study team.

Trial procedures

Trial procedures were performed as shown in Figure 1.

FIGURE 1 The seAFOod trial participant pathway. (a) Participants whose visit 6 date was scheduled from the first complete colonoscopy; and (b) participants whose visit 6 date was scheduled from a repeat full colonoscopy. FFQ, Food Frequency Questionnaire. (Figure not reproduced here; see the original article.)

Baseline visit

Individuals attending a routine outpatient follow-up BCSP appointment to obtain results of the screening colonoscopy were approached. Individuals who were eligible and willing to take part in the trial were asked to provide written informed consent. Demographic information and details of participants’ medical histories were collected and participants were randomised. It was preferred that participants were randomised within 4 weeks of the first complete BCSP screening colonoscopy. However, to maximise recruitment, randomisation was allowed outside this time window as long as it was recorded on the protocol deviation log. A prescription was issued for the supply of IMP for 6 months. The local hospital pharmacy dispensed the trial treatment. A second prescription for a further 6 months of IMP was provided at visit 4.

Biological samples were taken, comprising a blood sample [2 × 6-ml K₂EDTA (ethylenediaminetetraacetic acid) Vacutainer^® tubes; Becton, Dickinson and Company (BD), Franklin Lakes, NJ, USA] and a urine specimen of 5–10 ml. Samples were taken only if the participant provided separate, specific consent for collection of blood and urine.

In addition, participants were asked to complete a pre-treatment Food Frequency Questionnaire (FFQ) so that any change in dietary ω-3 PUFA intake during trial involvement could be determined.

Repeat endoscopy

Participants were informed whether or not a repeat colorectal endoscopic examination was required at the baseline visit. This was one of the following: a second full colonoscopy, a partial colonoscopy (planned incomplete views of the colorectum) or a FS. Participants undergoing one repeat endoscopy within 3 months of the screening colonoscopy remained eligible for the trial. The results of the colonoscopy were collected at visit 3. These participants were in the intervention period of the trial for a maximum of 15 months, rather than 12 months, so an additional 3-month trial prescription was dispensed to cover this period. If a participant who had already undergone a repeat endoscopy had a further repeat procedure scheduled by the local BCSP team, this made him or her ineligible for the trial.

All participants were asked to commence the IMP immediately following consent at the baseline visit (visit 1). Those participants who were due to undergo a repeat endoscopy procedure temporarily stopped IMP 10 days prior to the endoscopic procedure and restarted IMP 4 days after the endoscopy. Participants who underwent a partial colonoscopy or FS had a surveillance colonoscopy 12 months after the first complete screening colonoscopy, as per BCSP guidelines. Participants who underwent a second full colonoscopy had a surveillance colonoscopy date booked as per BCSP team preference, but preferably dated 12 months after the first screening colonoscopy.

Visits 2 and 3: telephone calls at 2 and 12 weeks

Participants were contacted by the SSPr/RN by telephone at 2 and 12 weeks after starting trial treatment. Participants were asked about any symptoms or new medical problems since the last contact and were reminded to take the IMP as directed.

Participants who were due to undergo a repeat colorectal endoscopic procedure between visits 2 and 3 were reminded to discontinue IMP temporarily. Colonoscopic findings at the repeat procedure were collected and recorded in the same way as for the baseline visit.

Visit 4: outpatient visit

At 6 months, participants were invited to attend the BCSP site, at which time mid-treatment blood and urine specimens were collected from those who had provided consent. Participants were asked about any symptoms or new medical problems since the last contact. Any unused trial treatment from the first prescription was collected and counted. Each participant then received a new prescription for trial treatment for a further 6 months.

Visit 5: telephone call

A third telephone contact was conducted with participants at 38 weeks after starting the trial treatment. The SSPr/RN asked about any symptoms or new medical problems since the last contact and reminded the participant to take trial treatment as directed.

Visit 5a: telephone call

An extra telephone contact was conducted for participants who had a repeat full colonoscopy between visits 2 and 3 after starting the trial treatment. The SSPr/RN asked about any symptoms or new medical problems since the last contact and reminded the participant to take his/her trial treatment as directed. The SSPr/RN liaised with the hospital pharmacy about delivery of a third dispensing of IMP to these participants. The relevant participants then received a new prescription for trial treatment for a further 3 months.

Visit 6: surveillance colonoscopy

Participants attended for routine surveillance colonoscopy at 12 months from the date of the screening colonoscopy. Participants took the final dose of trial treatment on the day before surveillance colonoscopy. Blood and urine specimens were obtained, as well as four random biopsies of macroscopically normal rectal mucosa (at least 2 cm from any polyp) at the end of the surveillance colonoscopy.

Colorectal adenoma outcomes at the 12-month surveillance colonoscopy were collected as per usual BCSP practice, including the number, size (maximum dimension in mm from the histopathology report, or the endoscopic size if the adenoma was not retrieved or was removed by hot biopsy), site [proximal to the splenic flexure (right) or at/distal to the splenic flexure (left)], histological type (tubular/tubulo-villous, villous, serrated) and presence of high-grade dysplasia of all colorectal adenomas.

Visit 7: routine post-colonoscopy visit

Participants were seen after surveillance colonoscopy as part of routine BCSP follow-up, during which a second FFQ was completed. Participants had the option to complete the FFQ at visit 6 or over the telephone if they decided to receive colonoscopy results by telephone.

Randomisation

Participants were registered in the trial using a secure web-based randomisation system. Randomisation was based on a computer-generated, internet-based treatment assignment determined by a pseudo-random code using random permuted blocks of randomly varying size, created by NCTU. Participants in the trial were allocated with equal probability to either treatment group. It was planned to stratify by BCSP centre. However, after database lock, it was discovered that BCSP site had been used, rather than BCSP centre. As sites could be associated only with an individual BCSP centre, this still ensured balance between centres.

Participants were randomised to a simple 2 × 2 factorial design (Table 1) to:

• 2 g of EPA-FFA, or an equivalent FFA dose of 90% EPA-TG (2780 mg), daily by mouth, or their identical placebos (capric and capryllic acid medium-chain TGs for both formulations)

in addition to:

• 300 mg of enteric-coated aspirin daily by mouth (as one 300-mg tablet taken with food) or identical placebo.

The sequence of treatment allocations was concealed until interventions had all been assigned and recruitment, data collection and all other trial-related assessments were completed. The actual allocation was not divulged to either the staff at the BCSP site or the participant. The trial prescription produced by the randomisation system referenced specific trial treatment containers. The trial drug prescription was signed by the local PI or a co-investigator, as defined by the site delegation log.

Interventions

The IMPs used in this trial were gastroresistant capsules of 99% pure EPA in the FFA form (EPA-FFA), 90% EPA as the TG conjugate (EPA-TG) in soft gelatin capsules, enteric-coated aspirin tablets and their identical placebos.

In February 2014, the supplier of the EPA-FFA IMP (SLA Pharma AG) disclosed that it could no longer provide further capsule IMP to the trial. The Trial Management Group (TMG), NIHR EME programme and the Trial Steering Committee (TSC) made the decision to continue the trial using an alternative EPA formulation and identical (medium-chain TG) placebo. Close consultation with the Medicines and Healthcare products Regulatory Agency (MHRA) and the Research Ethics Committee (REC) was also undertaken. A substantial amendment (number 14) was approved by the REC on 26 August 2014 and the trial received a Clinical Trials Authorisation (CTA) from the MHRA on 29 August 2014 for a new capsule IMP (EPA-TG). The new formulation maintained FFA equivalence (2000 mg daily) with the previous EPA-FFA formulation by using 5 × 574-mg 90% EPA-TG capsules per day (a total of 2870 mg), taking into account the percentage weight per weight (w/w) content of EPA and the presence of the glycerol backbone in the re-esterified TG.

The CTA required a specific simplified IMP dossier for the 90% EPA-TG, detailing its multistep manufacture, and also a cover document for the existing investigator brochure, which compared the chemical structure, GI absorption, bioavailability and tolerability profiles of the FFA and TG forms of EPA.

Although the 90% EPA-TG capsules that were proposed as a new IMP were publicly available for purchase as a nutritional supplement (from Igennus Healthcare Nutrition), the MHRA requested a programme of stability testing to meet manufacturing QA requirements for a CTIMP (see Appendix 1). A programme of accelerated (30 °C, 65% relative humidity) and standard (25 °C, 60% relative humidity) stability testing of capsules began in July 2014 (performed by ALS Food & Pharmaceutical, Carlisle, UK) and was performed once every 3 months. The core data set comprised the peroxide value (POV), para-anisidine value (pAV) and derived total oxidation (TOTOX) value, as well as a full PUFA analysis, to determine the EPA content. Data from initial accelerated testing at 3 months gave a minimum 12-month shelf life under standard conditions for the trial to continue with the new capsule IMP. Rolling stability testing provided a continuous extension of shelf life until 12 June 2016, when capsule IMP use ceased, as per the maximum approved shelf life (3 years) of the capsule IMP approved by MHRA.

The commercially available 90% EPA-TG capsule was also encapsulated with a differently coloured soft gelatin coat (olive green) to be able to produce an identical placebo because of the difference in appearance of EPA and medium-chain TG oils.

Participants took only one formulation of the EPA, which was either FFA or TG, or its matching placebo. Stocks of the IMP were managed during the transition period from the FFA formulation to the TG formulation to ensure that sites had sufficient stock of EPA-FFA and placebo capsules to manage existing and new participants throughout each individual intervention period.

The trial treatment was taken daily from the date of randomisation to the day before the 12-month surveillance colonoscopy.

Suppliers

SLA Pharma AG supplied EPA-FFA capsules and an identical placebo free of charge. Igennus Healthcare Nutrition supplied the 90% EPA-TG capsules and an identical placebo at cost price. Aspirin and its identical placebo were supplied by Bayer AG (Leverkusen, Germany) free of charge.

Eicosapentaenoic acid-free fatty acid dose

Participants took two 500-mg gastroresistant capsules of 99% pure EPA-FFA (or placebo) twice daily with food, giving a total daily dose of 2 g of EPA-FFA. Previous experience suggested that 2 g of EPA-FFA daily is well tolerated; principal side effects are diarrhoea, abdominal pain, nausea and vomiting, which are generally mild in severity and minimised or avoided by dosing with food or dose reduction to 1 g daily. A dose-reduction algorithm was used for participants experiencing side effects and managed by the local RN or SSPr.

Eicosapentaenoic acid-triglyceride dose

Alternatively, five soft gelatin 574-mg capsules of 90% EPA-TG (or placebo), equivalent to 2780 mg, were taken orally with food each day. It was preferred that three of the EPA-TG capsules were taken with the largest meal of the day and two capsules were taken with a smaller meal. Each capsule contained approximately 516 mg of EPA-TG, which is equivalent to 398 mg of EPA-FFA. Other PUFAs in the formulation included 3.9% (w/w) AA. The 90% EPA-TG did not have pharmaceutical marketing approval [see the project web page: www.journalslibrary.nihr.ac.uk/programmes/eme/0910025/#/ (accessed 25 April 2019)]. Clinical studies have indicated that EPA-TG (usually in a fish oil mixture with other PUFAs) is well tolerated at doses exceeding 2 g per day over periods of up to 6 months. The principal known side effects are the same as indicated for EPA-FFA.101–105

Aspirin

Participants were randomised to one 300-mg enteric-coated aspirin tablet (or placebo), taken orally once a day with food. Aspirin (acetylsalicylic acid) was supplied as 300-mg enteric-coated tablets in accordance with the Summary of Product Characteristics for aspirin (see Appendix 2). These tablets had marketing approval in the European Union.

Placebos

The EPA-FFA and EPA-TG placebos consisted of identical capsules of capric and capryllic acid medium-chain TGs that had previously been used in placebo-controlled trials of EPA.67

The placebo for aspirin consisted of the same excipients as the active formulation of the drug minus the active ingredient.

Discontinuation of treatment

In the event of an adverse drug reaction (ADR), either serious or non-serious, the local PI or attending physician was to take direct and appropriate action to provide care for the participant and to decide whether or not the trial treatment should be discontinued. However, unless there was a clear contraindication, trial treatment was continued, or stopped temporarily.

In all cases, the reasons for discontinuation of trial treatment were recorded in the clinical record file and if the investigator had recorded more than one reason, he or she was to indicate the main reason.

Treatment was discontinued permanently if the participant needed treatment with a contraindicated drug: ≥ 15 mg of methotrexate weekly, any dose of warfarin or any other anticoagulant therapy, any other antiplatelet agent such as clopidogrel, or prescription of aspirin for any other indication. These participants were still followed up for the remainder of the trial.

Prescriptions and accountability

The local PI or delegated pharmacy trial staff member was responsible for ensuring trial treatment accountability, including reconciliation of trial treatment and maintenance of trial treatment records, throughout the course of the trial, in accordance with UK regulatory requirements. On receipt of a delivery of trial treatment, details were checked for accuracy and receipt was acknowledged by signing and dating the documentation provided. In addition, receipt was acknowledged in the web-based system by the local pharmacy team, which had access to the web-based stock control system. Stock did not become available for allocation until it was accepted in the stock control system.

The local hospital pharmacy completed the dispensing process by addition of a participant’s name, subject number, date of dispensing and visit number to each allocated container. This process was repeated again at visits 4 and 5a if required.

Blinding

Participants, SSPr/RNs, local investigators and those assessing the outcomes were all blinded to treatment allocation. The statistical analysis for the trial was also blinded until data were locked, except for independent Data Monitoring Committee (DMC) reports.

Trial treatment

The trial treatment was packaged and labelled in accordance with UK regulatory requirements. The containers were clearly marked and had a unique identification number.

Bulk supplies of EPA-FFA, EPA-TG, aspirin and placebo capsules and tablets were delivered to Stockport Pharmaceuticals (Stockport, UK) for packaging and labelling to allow preparation of blinded supplies.

Blinded supplies were then stored at Stockport Pharmaceuticals for distribution to participating sites under a web-based stock control system reviewed regularly by the NCTU pharmacist.

Unblinding

Access to the sequence of treatment allocations was confined to the NCTU data manager and a central pharmacy, in case of out-of-hours unblinding. In the event of the need to break the code, the date and reason were recorded on the web-based unblinding system. The local hospital pharmacy had access to the web-based unblinding system in normal office hours and out-of-hours access was provided via the sponsor at St James’s University Hospital. The requirement for unblinding was considered low. All participants were given a trial identification card, containing details of the IMPs, which participants were encouraged to show when seeking advice or management from any health professional. Unblinding did not occur during the trial.

End of the trial

Participants left the trial when they completed their routine post-surveillance colonoscopy visit (visit 7).

Cases of failure to receive allocated treatment and withdrawal from follow-up were reported, and the reason(s) for withdrawal (if given) were documented. If a participant did not receive allocated treatment but agreed to remain in the trial, outcome data collection continued in accordance with the protocol. Participants were informed at the start of the trial that data collected up to the point of withdrawal would be retained and used in the final analysis.

Trial withdrawal

Participants could withdraw from the intervention or the trial at any time without giving a reason and without compromising future management. Data collected up to the point of withdrawal were retained for the purposes of the intention-to-treat (ITT) analysis. Participants could withdraw from the intervention only but continue in the trial, thereby completing outcome measures. To maximise primary outcome data collection, BCSP surveillance colonoscopy data were also collected from participants who withdrew from the trial, as per the informed consent.

Outcome measures

Primary outcome

The primary outcome was the number of participants with one or more colorectal adenomas detected at the first BCSP surveillance colonoscopy 12 months after the screening examination (ADRa).

Secondary outcomes

The secondary outcomes were as follows:

• Total number of colorectal adenomas per participant at BCSP surveillance colonoscopy (total MAP).
• Detection of one or more ‘advanced’ (i.e. ≥ 10 mm in diameter, high-grade dysplasia or villous histology) colorectal adenomas at the 12-month BCSP surveillance colonoscopy (advanced ADRa).
• Number of ‘advanced’ colorectal adenomas per participant at the 12-month BCSP surveillance colonoscopy (advanced MAP).
• Detection of one or more conventional adenomas (conventional adenoma end points were defined after database lock) at the first BCSP surveillance colonoscopy (conventional ADRa).
• Number of conventional adenomas (conventional adenoma end points were defined after database lock) per participant at the first BCSP surveillance colonoscopy (conventional MAP).
• Detection of one or more serrated adenomas at the first BCSP surveillance colonoscopy (serrated ADRa).
• Number of serrated adenomas per participant at the first BCSP surveillance colonoscopy (serrated MAP).
• The region of the colorectum (right colon: any part of the colon proximal to the splenic flexure; left colon: the rectum and the colon at/distal to the splenic flexure) in which adenomas are detected at the first BCSP surveillance colonoscopy.
• Reclassification from ‘high risk’ to ‘intermediate risk’ after the first BCSP surveillance colonoscopy (BCSP risk stratification at the first surveillance colonoscopy states that any individual who does not continue to fulfil ‘high-risk’ criteria is classified as ‘intermediate risk’ for further colonoscopic surveillance at 3 years).
• Detection of CRC prior to, or at, the first BCSP surveillance colonoscopy.
• Dietary fish and other seafood intake at baseline and at the end of the trial.
• Red blood cell EPA and rectal EPA levels at baseline, 6 months (RBC only) and 12 months from randomisation.
• Absolute RBC fatty acid (i.e. DHA, AA, EPA-to-AA ratio) levels and difference from baseline at 6 months and 12 months.
• Rectal mucosal fatty acid (i.e. DHA, AA, EPA-to-AA ratio) levels at surveillance colonoscopy.
• Adverse events, including clinically significant bleeding episodes (i.e. haemorrhagic stroke or GI bleeding requiring hospital admission or investigation).

Exploratory outcomes

• Colorectal adenoma size.
• Association between change of RBC EPA level at 12 months and individual number of total colorectal adenomas.
• Association between rectal and RBC EPA levels at 12 months.

Research governance

The trial was conducted in accordance with (1) the recommendations adopted by the 18th World Medical Assembly, Helsinki 1964, amended at the 48th General Assembly, Somerset West, Republic of South Africa, October 1996 (www.wma.net/wp-content/uploads/2018/07/DoH-Oct1996.pdf); (2) the principles of the International Conference of Harmonisation of Technical Requirements for Registration of Pharmaceuticals for Human Use – Good Clinical Practice (ICH-GCP) guidelines (https://ichgcp.net); and (3) the Medicines for Human Use (Clinical Trials) Regulations 2004106 (UK Statutory Instrument 2004/1031) and any subsequent amendments of the Clinical Trial Regulations.

The National Research Ethics Service (NRES) Trent – Trent REC (reference number 10/H0405/90) gave ethics approval for the trial for NHS participants. First, MHRA CTA approval was obtained on 16 March 2011. The trial was approved by the BCSP Research Committee on 13 October 2009.

This project was funded by the EME programme, a MRC and NIHR partnership (project number 09/100/25). The trial was registered (as ISRCTN05926847) in a publicly available database prior to FPFV.

The final, approved protocol was version 6.0, approved on 11 August 2014. There were a number of administrative and procedural changes made to the protocol during the trial, which are outlined in Appendix 3.

Protocol deviations

The protocol defined a protocol violation as:

• > 50% of trial medication returned in total
• inadvertent use of OTC medication containing aspirin, NSAIDs or fish oil for > 2 weeks in total
• surveillance colonoscopy occurring outside the allowed time windows (48–52 weeks after the last complete screening BCSP colonoscopy, or 60–64 weeks for participants undergoing a repeat full colonoscopy within 3 months of initial screening).

The protocol deviation log collected the above violations and any additional protocol deviations (i.e. any deviation from the protocol that occurred during a participant’s time in the trial, whether deliberate or non-deliberate).

Trial oversight

Oversight committees were assembled to ensure the proper management and conduct of the trial, and to uphold the safety and well-being of participants. The general purpose, responsibilities and structures of the committees were described in the protocol.

Trial Management Group

The TMG comprised the chief investigator, members of the NCTU and other visiting members of the wider trial team, including the University of Bradford biobank and lipid analysis team and an expert in human nutrition, as required. This group met regularly, with supplementary meetings as required, to oversee the day-to-day operational aspects of the trial, reviewing progress and resolving issues that arose.

Trial Steering Committee

The TSC was led by an independent chairperson and consisted of members with professional expertise in delivering RCTs, CRC diagnosis and management, primary care gastroenterology and the BCSP. The TSC also had a patient and public involvement (PPI) representative (an individual who had undergone BCSP colonoscopy) during the course of the trial. The TSC monitored, reviewed and supervised the progress of the trial, particularly advising on recruitment and retention, as well as ensuring adherence to the trial protocol. It also monitored blinded data to consider safety and effectiveness indications. The TSC considered reports from the DMC when making recommendations. The chief investigator, trial manager and at least one co-investigator were also in attendance as non-independent members to provide information to the committee. The trial sponsor (University of Leeds) and IMP providers were invited to attend as observers.

The TSC met independently prior to the start of the trial and agreed terms of reference in a charter. The committee met approximately every 6 months thereafter, convening supplementary meetings as required (e.g. related to the capsule IMP switch).

Data Monitoring Committee

An independent DMC, which met approximately every 6–12 months, was established with access to unblinded data to provide independent review and recommendations in the light of potential treatment effect and harm, as well as to assist and advise the TSC and the TMG. The DMC consisted of a chairperson and members with expertise in colorectal surgery and prevention, early detection and treatment of CRC, and statistics. The DMC met prior to the start of the trial and agreed terms of reference in a charter. Only the DMC had access to unblinded data until the final end-point assessment was completed.

Risk assessment and safety monitoring

A risk assessment was conducted as part of protocol development and was monitored regularly throughout the trial for new risks. The main risks to the trial were reliance on the BCSP to recruit to a large multicentre RCT, as the BCSP had not previously hosted CTIMP research and staff were not experienced in recruiting participants. Compliance with trial medication and drug accountability was reliant on participants returning medication and patient reports, as well as staff maintaining detailed records of medication logs.

Recruitment sites were supported by the NCTU, which initially provided training through a site initiation visit, as well as supplementary training, as required. Investigators had access to a research area of the trial website (www.seafood-trial.co.uk) and access to the NCTU trial team for day-to-day queries. To obtain high-quality data, regular central monitoring checks were performed according to the monitoring plan, including review of recruitment, retention and data collection rates by the TMG, TSC and DMC.

Data on AEs and serious adverse events (SAEs) were collected. As agreed by the sponsor, REC, DMC and TSC, the DMC was provided with a listing of all AEs and SAEs, including any deaths (if applicable), at each DMC meeting.

As part of the switch from the EPA-FFA formulation to the EPA-TG formulation, a review was undertaken of the Reference Safety Information (RSI) for the trial. Evidence was provided in a supplement, including MHRA-stipulated RSI for the EPA formulations, version 2.0, dated 22 August 2016 (approved by the MHRA on 3 October 2016) to the investigator brochure version 6, dated 10 January 2014, that the FFA and TG formulations were likely to have a similar pharmacological and risk/AE profile, as well as similar anti-CRC efficacy, in the seAFOod trial population [see the project web page: www.journalslibrary.nihr.ac.uk/programmes/eme/0910025/#/ (accessed 25 April 2019)].

Monitoring

Following an internal assessment of the trial by the quality assurance manager at the NCTU, the trial was assessed as a medium-risk trial requiring low-intensity monitoring. Based on this risk category and the specific risks identified, the monitoring strategy consisted of on-site routine and triggered monitoring visits, central monitoring and review of regular reports by trial oversight committees, as well as assessment and oversight of third-party vendors. On-site visits were conducted by either the trial manager or a trial monitor. Central monitoring was conducted by the NCTU and reviewed regularly by the TMG, the TSC and the DMC. On-site and central monitoring revealed no major common concerns during the trial. Sites that received an on-site monitoring visit at some point during the trial accounted for 68% of randomised participants.

Patient and public involvement

There was PPI throughout the trial. One PPI member was integral to trial design, grant application and production of participant-facing trial materials. He then took no further part in the trial. Subsequent PPI input on the TSC was provided by a different PPI member. Both PPI representatives brought personal experience of the BCSP pathway to their roles. Two PPI representatives from Independent Cancer Patients’ Voice contributed to the production of site and participant results summaries and to the development of the dissemination plan for the trial findings.

Payments to participants

Participants were not paid to participate in the trial. Visits 1, 6 and 7 were planned to co-ordinate with routine BCSP appointments. However, an additional visit at 6 months (visit 4) was required; therefore, participants were offered a reimbursement of travel expenses up to a maximum of £10.

The seAFOod polyp-prevention trial biobank

Blood, urine and rectal mucosa sample collection in the seAFOod polyp-prevention trial is described in detail in an appendix of the trial protocol, which has been published.100

In brief, a blood sample and a urine specimen were obtained at visits 1, 4 and 6. Blood was immediately separated into plasma, leucocytes and RBCs. Rectal biopsies were obtained only at the surveillance colonoscopy at visit 6.

Sample cryovials were stored in either a Liebherr Underbench –20 °C freezer (Liebherr Group, Bulle, Switzerland) supplied by the seAFOod trial or the existing trust freezer facility at –20 °C or colder, in PathoSeal bags (DGP Intelsius Ltd, York, UK). Storage temperature was monitored using a Hanna HI-141 CH Datalogger (Hanna Instruments Ltd, Leighton Buzzard, UK), or a similar system provided by the trust site, and recorded on sample worksheets.

University of Bradford staff, in close collaboration with the NCTU, organised the collection of samples from each BCSP site approximately every 6 months by a specialist courier (CitySprint Health, London, UK). The original proposal was to transfer samples to the central biobank facility every 3 months. However, the expansion of participating trial sites in 2012–13 meant that the collection strategy was revised. BCSP sites were grouped into 10 individual routes by CitySprint Health so that collection from these sites could be scheduled for the same day. Sites were asked to confirm whether or not they required a collection, and the number of PathoSeal bags at that site, to allow CitySprint Health to supply the correct thermal box size. If sites were unable to accommodate a collection on their scheduled route day, individual collections were arranged. The day before collection, CitySprint Health arranged for pre-addressed thermal boxes, which contained dry ice, to be sent to appropriate BCSP sites. The lid of the thermal box was removed only when necessary for adding samples. All sealed PathoSeal bags collected since the last courier sample collection were placed in the thermal box and dry ice was spread around the bags to ensure that they were completely covered. The thermal boxes were delivered to the University of Bradford on the same day or, in the case of routes involving longer distances, the following morning. No units were received with low dry ice levels and all samples were received frozen.

Each BSCP site completed a shipment form that documented the samples being transported. A carbon copy of the shipment form was kept in the site file with the carbon copy of the completed sample worksheets. The original shipment form and the original completed sample worksheets were sent with the samples.

On receipt of each shipment of transportation units at Bradford, a sample tracking form was completed. Contents were then checked against the sample worksheet. Cryovials were transferred into sample boxes (10 × 10 cell cryoboxes, manufactured by Sarstedt, Nümbrecht, Germany) specific to each sample type, and cryoboxes were kept on ice at all times when removed from the freezer. All details were entered onto a separate sample tracking form for each thermal box. Samples were stored in type-specific cryoboxes and racks, minimising disruption when samples were removed for analysis.

All samples were stored in a dedicated –80 °C freezer (New Brunswick™ U570 –80 °C; Eppendorf, Hamburg, Germany), which was connected to the emergency power supply at the Institute of Cancer Therapeutics, University of Bradford, and supported with a CO₂ back-up system (New Brunswick) and Centroller AD11+ (Centroller, Staines, UK) telephone alarm system. Freezer temperature was monitored daily.

The details from the sample tracking form and the sample worksheets were entered into the seAFOod trial biobank sample database. The database allocated a unique Clinical Trials Pharmacology Laboratory (CTPL) tracking identification (ID) to each participant-specific set of samples in a PathoSeal bag (i.e. if a participant had samples taken on three visits, they received three CTPL IDs). The CTPL ID was marked on the sample worksheet and the sample tracking form. The database was stored on the University of Bradford server, in the ‘Secure’ section, accessible only by nominated CTPL staff, and backed up regularly, as described in the University of Bradford’s computer policy. The database was the primary source for sample tracking; however, paper documents were available as a secondary source if necessary, stored in participant and shipment folders in the secure Human Tissue Act-approved laboratory.107

Samples were obtained at BSCP sites between FPFV (i.e. 11 November 2011) and LPLV (i.e. 8 June 2017). The first samples received at Bradford were delivered on 18 July 2012 and the last sample shipment took place on 27 July 2017. In the intervening time, 332 collections were made from BCSP sites, with 1775 individual PathoSeal bags received.

Samples were stored at BSCP sites for between 1 and 696 days (mean 124 ± 92 days, median 115 days). There were 1378 (78%) sample sets stored at BSCP sites for < 6 months. Thirty (2%) sample sets were stored at BSCP sites for > 12 months.

The majority of sample sets (n = 1021; 58%) were stored in BCSP sites at –20 °C (range –16 to –24 °C), with 230 (13%) sample sets stored at –40 °C (range –25 to –69 °C) and 524 (30%) sets stored at or below –70 °C.

One or more biological samples were received from 677 of 709 (95%) randomised seAFOod trial participants. Of the 709 participants, 73% (519) of participants provided full sample sets of blood, urine and rectal mucosa from all three visits. There were 76 participants who provided samples at two visits (visits 1 and 4, n = 49; visits 4 and 6, n = 15; visits 1 and 6, n = 12). Eighty-two participants provided samples at only a single visit (visit 1, n = 75; visit 4, n = 2; visit 6, n = 5).

Overall, a total of 7322 biological samples were received (16,258 sample aliquots):

• 1715 plasma samples (6746 aliquots)
• 1714 leucocyte samples (1714 aliquots)
• 1707 RBC samples (3421 aliquots)
• 1664 urine samples (3309 aliquots)
• 522 rectal biopsies (1068 aliquots).

Compliance with biological sample collection [defined as the proportion of sample sets expected (n = 2127) that were received with at least one sample aliquot] was 80% (blood), 78% (urine) and 74% (rectal mucosa).

Laboratory protocol deviations were monitored carefully by sample worksheets:

• 80% of blood samples were obtained per protocol (centrifugation within 30 minutes and transfer to the freezer within 60 minutes).
• 2% of sample sets were received without worksheets.
• 1% of samples sets were received with insufficient data to assess timing.
• 8% of blood was separated after 30 minutes, but frozen within 1 hour of collection.
• 9% of blood samples took > 1 hour to separate (often related to split hospital sites for endoscopy and sample handling).

A small number of other protocol deviations were noted, including:

• 1% of sample sets that suffered a temperature deviation (but no thaw).
• 1% of sample sets that defrosted at some point.
• 0.5% of sample sets suffered ‘other’ deviations (including wrong anticoagulant blood tube used and biopsies placed in formalin).

Fatty acid measurement and analysis

The methods used are described in detail by Volpato et al.89 In brief, fatty acids were extracted from washed RBC membranes, or rectal mucosal homogenates, using an isopropanol/chloroform method with acid hydrolysis, in order to measure the total membrane/tissue fatty acid pool.89 Liquid chromatography, in combination with electrospray ionisation triple quadrupole tandem mass spectrometry (ESI-MS), was performed with a Waters Alliance™ 2695 High Pressure LC module (Waters, Milford, MA, USA) in combination with a Waters Micromass™ Quattro Ultima triple quadrupole mass spectrometer on derivatised samples, in the presence of internal standard (deuterated ALA), as described by Volpato et al.89 Data are expressed as the percentage of each fatty acid relative to the total fatty acid peak chromatographic area for the ω-3 PUFAs C18:3 (ALA), C20:5 EPA, C22:5 docosapentaenoic acid (DPA) and C22:6 DHA; the ω-6 PUFAs C18:2 linoleic acid (LA) and C20:4 AA; and the monounsaturated C18:1ω-9 oleic acid and saturated fatty acids C18:0 stearic acid and C16:0 palmitic acid.89

Measurement of dietary fish intake

Participants were asked to complete a FFQ at baseline (visit 1) and at the end of the intervention (visit 6 or 7). The validated European Prospective Investigation of Cancer (EPIC) short FFQ was used, which is a semiquantitative, paper-based FFQ that includes 130 food items and supplementary questions about use of fat, milk, cooking methods, salt and supplement use.108 The FFQ measures an individual’s habitual food intake over the preceding 12 months. The primary purpose of the FFQ was to determine if there was any change in dietary marine ω-3 PUFA intake during trial participation. As fish is the primary source of bioactive ω-3 PUFAs EPA and DHA, the reported consumption of fish was analysed before and after intervention with IMP.

Total fish consumption and consumption of oily fish were considered separately. Six FFQ food item variables were used for the analysis of total fish (i.e. fried fish, fish fingers, white fish, oily fish, shellfish, roe). Consumption of oily fish was based on a single food item variable (oily fish). Participants indicated the frequency of food consumption, ranging from ‘never or less than once per month’ to ‘six times per day’. This was recoded into frequency per day (Table 2). The consumption of total fish per day was calculated by summing the reported fish consumption for each of the six fish variables.

The data were used to recode participants into four categories (i.e. never, low, middle or high), based on their weekly consumption of total fish and oily fish (Table 3). Categorisation of weekly fish consumption was based on the Scientific Advisory Committee on Nutrition recommendation to consume two portions of fish per week, one of which should be oily.109

Further analysis of the dietary PUFA intake data is planned and will include full nutrient and food group analysis using FETA (FFQ EPIC Tool for Analysis) software (www.srl.cam.ac.uk/epic/epicffq/; accessed April 2018).

Planned laboratory studies

The nested laboratory studies originally planned for the EME-funded seAFOod trial project are described in the trial protocol version 6.0, dated 11 August 2014 [see the project web page: www.journalslibrary.nihr.ac.uk/programmes/eme/0910025/#/ (accessed 25 April 2019)]. The switch in capsule IMP that occurred in 2014 required a more detailed PUFA analysis than was originally planned to test equivalence of the bioavailability of FFA and TG formulations measured by RBC and rectal mucosal EPA content. Therefore, funding for much of the planned biomarker work was diverted to a more extensive PUFA analysis.

Statistical methods

Changes from the protocol to the statistical analysis plan

A number of changes to details contained within the protocol are documented in the SAP (version 1.1, dated 24 August 2017) [see the project web page: www.journalslibrary.nihr.ac.uk/programmes/eme/0910025/#/ (accessed 25 April 2019)]. These are listed in Table 4.

In addition, the changes documented in Table 5 were made after database lock and release of the treatment codes.

Sample size

The original sample size estimate of 904 participants (to ensure 768 evaluable patients and assuming a 15% loss to follow-up) in trial protocol versions 1–3 was based on a RCT of the same dose and preparation of EPA-FFA in FAP patients,67 a meta-analysis of aspirin RCTs49 and detailed 2007–8 audit data from the South of Tyne and Tees BCSP centres.

For the protocol revision in May 2012 (related to inclusion of patients who required a repeat endoscopic procedure during the 3-month screening window), the sample size was re-estimated using audit data on surveillance colonoscopy in 1189 patients classified as ‘high risk’ in 2010, from the North-East BCSP hub (nine BCSP centres) and the Southern BCSP hub (17 BCSP centres). A total of 930 (78%) patients went straight to surveillance after a single screening colonoscopy. Colorectal ADRa data at surveillance colonoscopy were available for 738 patients and, of these, one or more colorectal adenomas were detected in 465 (63%) patients. Corresponding figures for ‘high-risk’ patients having a repeat partial colonoscopy or FS within 3 months of an initial screening colonoscopy showed that one or more adenomas were detected at first surveillance colonoscopy in 59 out of 110 (54%) patients, and for patients having a repeat full colonoscopy, one or more adenomas were detected in 56 out of 83 (67%). The overall ADRa at first surveillance colonoscopy was 62%, which was consistent with the original estimate of 60%; therefore, the sample size remained unchanged.

To detect a minimum 18% relative reduction in adenoma risk in each two-group comparison [less than the 22% reduction in polyp number compared with placebo in the FAP trial67 and below the absolute reduction in polyp number at 1 year (38%) in aspirin RCTs49] from a 60% adenoma recurrence rate at surveillance colonoscopy to 49%, 678 evaluable ‘high-risk’ individuals were required to be randomised equally to the four treatment groups, with 80% power at a 5% two-sided significance level.

Standard practice for 2 × 2 factorial designs, in the absence of an interaction, bases the sample size estimate on the two-group comparison of treatment versus placebo (and divides the total equally between the four groups). With the sample size of 678 based on this method, there is, in fact, a slight reduction in power (to 75%), which arises if both treatments work, because then the overall comparison for treatment A is not 0.49 versus 0.6, but is 0.445 versus 0.545 (averaging over the placebo and treatment B groups). To keep power at 80% for the above figures, a simulation using Stata^® version 10 (StataCorp LP, College Station, TX, USA) and employing the proposed analysis method indicated that 192 individuals were required per group (a total of 768 evaluable ‘high-risk’ individuals).

In trial protocol versions 1–3, a 15% drop-out rate was assumed. However, feedback from BCSP sites and experience from the first few months of the trial suggested that the drop-out rate of ‘high-risk’ BCSP patients was < 15%. Allowing for a 10% drop-out rate, the proposed sample size increased to 768/0.9 = 853 individuals.

For the purposes of the trial extension granted by the EME board in 2014, we proposed a revised realistic recruitment target of 755, which provided 71% power to detect the same effect size as above, still assuming a 10% drop-out rate.

Analysis plan

A SAP was finalised prior to database lock and release of treatment codes to the statistician. All summaries and statistical analyses were conducted using Stata version 15.0.

Analysis populations

• Intention-to-treat population: all randomised participants with post-randomisation data. Analysis was according to the treatment group to which they were randomised.
• Safety population: all randomised participants who had at least one dose of trial medication. Analysis was according to the treatment they actually received.
• Per-protocol population: all randomised participants not deemed to have a major protocol violation. Major protocol violations that resulted in exclusion from the per-protocol population were –
  • had not taken sufficient medication (participants who had taken ≤ 75% of their expected EPA and/or ≤ 50% of their expected aspirin)
  • found to be ineligible post-randomisation
  • any use of OTC medication containing aspirin, NSAIDs or fish oil for > 2 weeks during the treatment period.

The final composition of the per-protocol population was determined by final data review prior to the treatment codes being revealed.

Analysis of the primary end point was based on the ITT population. The analysis was repeated based on the per-protocol population. This was considered supportive to the primary analysis.

Analyses of all secondary end points, with the exception of AEs, were based on the ITT population.

Summaries of the AEs were based on the safety population.

Missing data

Sensitivity analyses were performed to support the primary analyses. All secondary analyses assumed that the data were missing at random and no imputation was performed.

Primary outcome

The primary outcome was the number of participants with one or more colorectal adenomas detected at the first BCSP surveillance colonoscopy 12 months after the screening examination (the ADRa).

The primary outcome was analysed by an ‘at-the-margins’ approach,110 after first examining whether or not there was any evidence of an interaction between EPA and aspirin. An ‘at-the-margins’ approach analyses all participants randomised to EPA (i.e. EPA + aspirin plus EPA + placebo aspirin, combined) versus all participants not randomised to EPA (i.e. placebo EPA + aspirin plus placebo EPA + placebo aspirin, combined), and all participants randomised to aspirin versus all participants not randomised to aspirin. Given that there was no evidence of an interaction, the log relative risk was estimated using a mixed-effects log-binomial regression model, with BCSP site included as a random effect, and the risk differences and ratios presented. Both interventions were fitted simultaneously and the analysis was adjusted for ‘repeat colorectal endoscopic procedure within 3 months required’ and BCSP site. This was a change to the analyses defined in the SAP, which stated that BCSP centre would be included as a random effect.

Sensitivity analyses for the primary outcome

The following sensitivity analyses were performed on the primary end point:

• Analysis using the per-protocol population.
• Analysis as a multilevel model. Some BCSP centres comprise multiple hospital sites; therefore, a sensitivity analysis was conducted in which both BCSP centre and site were treated as random effects in a multilevel model.
• Multiple imputation of missing primary end-point data.
• Further adjustment of baseline variables with any marked imbalance, if appropriate.
• Investigation of the effect of treatment adherence using complier-average causal effect (CACE) estimation methods.111,112 ITT analysis does not represent the treatment effect of non-compliance with treatment; therefore, CACE analysis was deemed to be important if any treatment effect was directly affected by the level of compliance. The percentage of the required total dose taken by participant from randomisation to first surveillance colonoscopy (both in binary and continuous form) was included in the model as an instrumental variable to estimate such an effect.
• Further adjustment for EPA capsule formulation, that is EPA-FFA (or placebo) or EPA-TG (or placebo).
• Further adjustment for oily fish intake.
• Further adjustment for baseline RBC EPA levels.

Secondary outcomes

• The total number of colorectal adenomas per participant at the 12-month BCSP surveillance colonoscopy was analysed using a Poisson regression model. The incidence rate ratio (IRR) and 95% CIs were presented.
• The number of participants with an ‘advanced’ colorectal adenoma (≥ 10 mm maximum dimension, high-grade dysplasia or villous histology) at the first BCSP surveillance colonoscopy (advanced ADRa) was analysed using a log-binomial regression model. The risk difference and 95% CIs were presented.
• The number of ‘advanced’ colorectal adenomas per participant at the first BCSP surveillance colonoscopy (advanced MAP) was analysed using a Poisson regression model. The IRR and 95% CIs were presented.
• The number of participants with one or more conventional colorectal adenomas at surveillance colonoscopy (conventional ADRa) was derived from the total and serrated adenoma data and was analysed using a log-binomial regression model. Risk difference and 95% CIs were presented.
• The number of conventional colorectal adenomas per participant at surveillance colonoscopy (conventional MAP) was derived from the total and serrated adenoma data and was analysed using a Poisson regression model. The IRR and 95% CIs were presented.
• The number of participants with one or more serrated adenomas at surveillance colonoscopy (serrated ADRa) was analysed using a log-binomial regression model. The risk difference and 95% CIs were presented.
• The number of serrated adenomas per participant at surveillance colonoscopy (serrated MAP) was analysed using a Poisson regression model. The IRR and 95% CIs were presented.
• The region of the colorectum (right colon: any part of the colon proximal to the splenic flexure; left colon: the rectum and the colon at/distal to the splenic flexure) that colorectal adenomas were detected at the first BCSP surveillance colonoscopy was explored, using a Poisson random-effects model with bivariate response (corresponding to adenoma counts in the left and right colon), in which treatment and a baseline adenoma count were independent variables together with random intercepts corresponding to participant and BCSP site. The IRR and 95% CIs were presented.
• The number of ‘high-risk’ participants reclassified as ‘intermediate risk’ after the first BCSP surveillance colonoscopy was analysed using a log-binomial regression model. The risk difference and 95% CIs were presented.
• The number of participants with CRC detected prior to, or at, the first BCSP surveillance were to be summarised descriptively; however, there were no participants with CRC.
• The levels of RBC EPA and rectal mucosal EPA were summarised at baseline (RBC only) and at visits 4 (RBC only) and 6 for those receiving EPA-FFA and those receiving EPA-TG. In addition, the change from baseline to visits 4 and 6 were summarised.
• The levels of DHA, AA and the EPA-to-AA ratio were summarised at baseline (RBC only) and at visits 4 (RBC only) and 6 (RBC and rectal mucosal) for those receiving EPA-FFA and those receiving EPA-TG. In addition, the change from baseline to visits 4 and 6 were summarised for RBC samples.
• Dietary fish and other seafood intake (oily fish and total) at baseline (visit 1) and at the end of the trial (visit 6 or 7) were summarised by treatment group.

Safety analyses

Adverse events were summarised for the safety population, that is all participants who received at least one dose of trial medication. Summaries were based on the IMP that the participant received, irrespective of randomisation.

Both AEs and treatment-emergent ADRs were summarised by system organ class. GI AEs were also summarised by preferred term and by formulation. ADRs were summarised by severity using the preferred term. The worst case (i.e. severe and/or related to trial treatment) was assumed if severity or causality were missing, unless otherwise stated.

Clinically significant bleeding episodes (i.e. haemorrhagic stroke or acute GI bleeding requiring hospital admission or investigation) were identified by the chief investigator using a manual search of the full list of AEs and SAEs.

Common GI AEs associated with ω-3 PUFA and/or aspirin use were presented separately as clinically meaningful symptom categories, which were defined by the chief investigator.

Serious adverse reactions and ADRs that led to trial discontinuation were summarised by treatment group and preferred term.

All treatment-emergent ADRs were listed.

Exploratory analyses

• Colorectal adenoma size was estimated using a multilevel model, adjusting for colorectal adenoma histology type and number of colorectal adenomas per individual.
• Rectal and RBC EPA levels were plotted against the individual number of total, conventional, serrated and left- and right-sided colorectal adenomas by treatment group.
• A correlation between rectal mucosal and RBC EPA levels at 12 months was investigated by graph and the correlation coefficient was calculated.