Coronary artery bypass grafting in High RISk patients randomised to Off Pump or On Pump Surgery - the CRISP trial

Article history

This issue of Health Technology Assessment contains a project originally commissioned by the MRC but managed by the Efficacy and Mechanism Evaluation Programme. The EME programme was created as part of the National Institute for Health Research (NIHR) and the Medical Research Council (MRC) coordinated strategy for clinical trials. The EME programme is funded by the MRC and NIHR, with contributions from the CSO in Scotland and NISCHR in Wales and the HSC R&D, Public Health Agency in Northern Ireland. It is managed by the NIHR Evaluation, Trials and Studies Coordinating Centre (NETSCC) based at the University of Southampton. The authors have been wholly responsible for all data collection, analysis and interpretation, and for writing up their work. The HTA editors and publisher have tried to ensure the accuracy of the authors’ report and would like to thank the reviewers for their constructive comments on the draft document. However, they do not accept liability for damages or losses arising from the material published in this report.

Copyright statement

© Queen’s Printer and Controller of HMSO 2014. This work was produced by Rogers et al. under the terms of a commissioning contract issued by the Secretary of State for Health. This issue may be freely reproduced for the purposes of private research and study and extracts (or indeed, the full report) may be included in professional journals provided that suitable acknowledgement is made and the reproduction is not associated with any form of advertising. Applications for commercial reproduction should be addressed to: NIHR Journals Library, National Institute for Health Research, Evaluation, Trials and Studies Coordinating Centre, Alpha House, University of Southampton Science Park, Southampton SO16 7NS, UK.

Background and rationale

Despite advances in medical therapy and percutaneous coronary interventions (PCIs) there is good evidence that coronary artery bypass grafting (CABG) offers superior survival and freedom from repeat intervention in patients with multivessel coronary artery disease (CAD). 1–5 For example, in the published New York State registry of almost 60,000 patients, after risk stratification for cardiac and non-cardiac comorbidity, there was a significant reduction in mortality (absolute difference of 5%) and a sevenfold reduction in the need for repeat interventions at 3 years in patients undergoing CABG rather than PCI using stents. 2 Predictions that drug-eluting stents will significantly reduce the need for CABG are premature because, although these stents reduce the incidence of restenosis compared with bare metal stents, three large meta-analyses have shown that they do not improve survival or reduce the incidence of subsequent myocardial infarction (MI). 6–8 There are two reasons why CABG is likely to remain a superior treatment to PCI over the longer term: (1) CABG protects whole zones of proximal myocardium (as the graft is placed to the midcoronary vessel beyond all proximal disease);9 and (2) PCI frequently results in incomplete revascularisation, which adversely affects survival proportional to the incompleteness of revascularisation. 10 Currently around half a million patients worldwide undergo CABG each year. There is a real possibility that these numbers will increase with a growing elderly population, an increasing epidemic of diabetes and obesity which all predispose to the development of CAD, and an increasing realisation that PCI may merely delay definitive treatment.

Conventional CABG uses cardiopulmonary bypass (CPB) (‘on-pump’) to support the circulation while the heart is temporarily stopped. CPB causes a systemic inflammatory response syndrome, which leads to multiorgan dysfunction, and, although mild and reversible in most, can contribute to mortality and overt morbidity, particularly in higher-risk patients. 11–19 Evidence from randomised controlled trials (RCTs) in low-risk populations shows that ‘off-pump’ CABG (OPCABG) is at least as safe as ‘on-pump’ CABG (ONCABG) in terms of mortality and that it reduces several aspects of morbidity but may lead to a higher need for subsequent reintervention. 11–14

However, the exclusion of high-risk patients from these RCTs is of key importance because there are consistent findings from large observational studies that OPCABG appears to reduce mortality and morbidity in such patients. 15–19 These studies, summarised in Table 1 , have used propensity scoring and/or logistic regression to take account of different baseline characteristics in the OPCABG and ONCABG groups but are still prone to all the limitations of non-randomised studies.

Only 15–20% of all CABG in Europe and the USA are performed as OPCABG owing to concerns that it may result both in fewer grafts and in lower graft patency. The Prague-4 RCT of 400 patients in a single centre reported similar 30-day clinical outcomes but a reduction in 1-year saphenous vein graft patency (49% in OPACBG group vs. 59% in ONCABG group) in the OPCABG group. 20 In contrast, in the Surgical Management of Arterial Revascularisation Therapies trial, a single-centre, single-surgeon RCT of 197 patients, Puskas et al. 21 reported 1-year angiographic graft patencies of 94% for OPCABG (mean of 3.2 grafts) and 96% for ONCABG (mean of 3.4 grafts). In the Beating Heart Against Cardioplegic Arrest Studies,22 two single-surgeon RCTs of 401 patients in total, 7-year follow-up has shown graft patency of 86.2% and 85.4%, respectively.

Past research

Aims and objectives

The CRISP trial was set up to address the limitations highlighted in the meta-analyses, namely to test the hypothesis that OPCABG in high-risk patients reduces mortality and morbidity, without causing a higher risk of reintervention. It complemented the CORONARY trial, which recruited predominantly lower-risk patients. Overall, 5.6% of CORONARY trial participants had impaired LV function (impairment was defined as LV function < 35%) and only 17.7% had a European system for cardiac operative risk evaluation (EuroSCORE) of > 5. 40

This report describes the results of the CRISP trial. The trial closed early, on the grounds of futility, after less than 2% of the target sample size had been reached. The challenges faced and the outcomes for the small cohort of patients recruited are described.

Study design

The CRISP trial was a designed as an international, multicentre, open, parallel-group RCT of isolated OPCABG versus ONCABG in high-risk patients with an additive EuroSCORE of ≥ 5. The study received research ethics approval (reference 08/MRE00/58) and was registered (reference ISRCTN29161170).

The preferred method of randomisation when CRISP was set up was expertise based, i.e. patients were randomised to surgery carried out by an experienced off-pump surgeon or by an experienced on-pump surgeon. Evaluating surgical interventions using an expertise-based trial design was first proposed in 1980,43 but was rarely used until more recently. 44 The advantages of an expertise-based design have been discussed in detail by Devereaux et al. ,45 Cook46 and in the orthopaedic setting by Scholtes et al. 47 The rationale for choosing an expertise-based design for the CRISP trial was as follows: individual surgeons, because of their training and experience, are generally more proficient in a particular technique and so are likely to use primarily a single surgical approach. This could compromise the validity of a conventional RCT as the surgical expertise may be skewed toward the technique which is best established, most widely used or easiest to perform; a conventional RCT also has limited applicability since, by design, only surgeons experienced in OPCABG can take part. Surgical procedures that require a ‘learning curve’ are clearly disadvantaged as a minimum number of cases need to be performed and considerable experience is needed before a surgeon feels at ease with both techniques. Unless participating surgeons have expertise in both procedures, there is also a potential for differential crossover in the two arms of the trial (i.e. more crossovers in one direction than the other). OPCABG is less frequently performed than ONCABG, technically more demanding and may have a more prolonged ‘learning curve’. Previous conventional RCTs have been criticised for recruiting ‘inexperienced’ OPCABG surgeons, resulting in poor OPCABG results with an excess of graft occlusion and not the best ONCABG surgeons. 48 Expertise-based randomisation was chosen to avoid these problems. The surgeon eligibility criteria for participation in the CRISP trial are described in Settings.

Participants

Settings

Patients were recruited to the CRISP trial from specialist cardiac surgery centres in the UK and Kolkata, India.

The preferred method of randomisation for CRISP was expertise-based randomisation (see Study design). Surgeons at participating centres using this preferred method were eligible to join CRISP if they had a stated preference for either OPCABG or ONCABG and were approved by the TSC as being sufficiently experienced in their preferred technique (i.e. at least 100 operations).

If, after detailed discussion with the research team, it was agreed that expertise-based randomisation was not possible at a centre, stratified within-surgeon randomisation was used. Centres and surgeons that planned to use within-surgeon randomisation required approval from the TSC (prior to the randomisation criteria being relaxed part-way through the trial; see Study design). The surgeons concerned were required to provide evidence that they have expertise in both techniques (at least 100 operations carried out using each method) and that they used both techniques with similar frequency.

Interventions

Trial patients were randomised to

1. CABG without CPB, i.e. OPCABG on the beating heart, via a median sternotomy incision, or

2. CABG with CPB, i.e. ONCABG on a chemically arrested heart, via a median sternotomy incision.

The anaesthetic technique and method of myocardial protection used were in accordance with established local protocols. These aspects were not specified in the trial protocol as there is a consistent 30-day mortality of around 2% for CABG across most UK centres, suggesting that minor differences in anaesthetic technique and methods of myocardial protection do not have a major influence on perioperative mortality. Surgical details were recorded on the case report form (CRF).

The only requirement was that the centre/surgeon followed the randomisation allocation. If it proved necessary to convert from OPCABG to ONCABG during the operation, this was recorded on the CRF.

Outcomes

Sample size

The study sample size was set at 5418 patients (2709 per group). Pooled data collected from Bristol and Oxford cardiac databases were used to inform the sample size calculation. The data suggested an expected incidence of the composite primary outcome of 9.3% for patients with a preoperative EuroSCORE of ≥ 5. As all patients randomised to a given surgeon under expertise-based randomisation will have had their operations using the same technique, they cannot be regarded as independent of each other. Assuming that 80 surgeons would take part in the trial, the resultant intraclass correlation coefficient (ICC) was estimated from data from Bristol and Oxford cardiac databases to be 0.005. Using these assumptions, a sample size of 5418 patients had 90% power to detect a 30% reduction in RR with 5% significance (two tailed).

The DMSC periodically reviewed the safety data. At the start of the trial, two interim analyses of clinical outcomes were proposed: (1) when 50% of participants had been followed up to 30 days and (2) when 50% of participants had completed the trial (i.e. had been followed up for 12 months after surgery, the end of follow-up according to the original trial design). It was proposed that the trial should continue as planned unless there was a statistically significant difference between the two surgical approaches, with p ≤ 0.001. These interim analyses were not undertaken owing to the premature termination of the trial (see Chapter 3, Decision to close the trial early).

Randomisation

Randomised treatment allocations were internet based and generated by Sealed Envelope Ltd, London, UK. 55 Allocations were stratified by centre and cohort-minimisation used to minimise imbalance of key prognostic factors (age, sex, urgency of operation, poor LV function, impaired renal function, previous stroke, redo CABG and significant pulmonary disease) across the OPCABG and ONCABG groups. Patients were randomly assigned in a 1 : 1 ratio.

Using an internet-based randomisation system ensured that allocations were concealed until all data that could uniquely identify the patient, confirm eligibility and establish stratification and cohort minimisation groups were entered. Access to the system was password protected and only available for designated site staff. Randomisation was carried out after the trial co-ordinator or research nurse had obtained written informed patient consent. The timing of expertise-based randomisation was carefully chosen to leave enough time to schedule the surgery, but also to minimise the time between randomisation and surgery and, therefore, reduce the possibility of outcome events or cancellation of surgery occurring in this period. Within-surgeon randomisation was usually carried out as close as possible to surgery. Any patients who were unexpectedly rescheduled retained their study numbers and randomised allocation and every effort was taken to ensure the rescheduled operation was carried out by an appropriate surgeon participating in the trial, according to the randomisation method used and the assigned allocation.

Blinding

It was not possible to blind the surgeons or those involved in the postoperative care of the patients. However, at most centres, postoperative care follows strict protocols that are not ONCABG or OPCABG specific. Patients were not explicitly informed of their allocation and the external signs of surgery were similar for both groups. The careful choice of objective, clinically defined primary outcome components should minimise bias. In addition, the individuals undertaking the adjudication of MI data were masked to the treatment allocation.

Data collection

Data collection was performed both while the patient was under the care of the cardiac unit and again at their standard 4–8 week postoperative outpatients appointment to identify any elements of the primary outcome and/or adverse events (AEs). Data were collected from clinical records by research nurses and/or clinical trial co-ordinators. Purpose-designed CRFs were used to record data at each stage of a patient’s journey through the trial (see Appendix 4 ), with the key data collection points being pre surgery, the period from surgery to discharge and the routine 4–8-week follow-up appointment. Completed CRFs were then entered into the trial database via a password-protected web–based interface.

A bespoke trial database was designed using SQL server (2008). The database was intended to act as both a data storage facility and a trial management resource. For example, the database issued reminders when 4–8 week postoperative assessments were due, managed payment schedules to sites and provided facilities for tracking the progress of serious adverse event (SAE) reporting. Owing to the intended large sample size, a considerable amount of data validation was applied to the database. The validation rules were determined as a result of detailed discussions between clinical trial co-ordinators, research nurses, statisticians and database developers working on the study and were refined following any feedback from sites. Validation broadly included rules such as (1) the correct ordering of any dates and times, e.g. the date and time of CICU, high-dependency unit (HDU) or ward admission must be after the operation end date and time but prior to hospital discharge; (2) agreement of data on postoperative complications between the study CRFs and SAE forms for sponsor reporting, e.g. if there is an AE classified as serious on the CRFs an SAE form should be completed; (3) patient details (e.g. sex, age) and stratification/cohort minimisation data entered on the study CRFs should match that entered on the internet-based randomisation system; and (4) miscellaneous validation of related data recorded on different CRF pages, e.g. if the patient is recorded as being reintubated twice, there should be two reintubation and re-extubation dates and times entered on the relevant CRF. See Appendix 4 , Figure 1 , for an example of a message to the user if validation rules were not met.

Statistical methods

Analyses of the primary and secondary outcomes were carried out on the basis of intention to treat (ITT). The analysis (ITT) population consisted of all randomised patients excluding patients who died prior to surgery, patients who withdrew prior to surgery as it was decided not to perform surgery and patients who withdrew at any time and were unwilling for any data collected to be used. Continuous variables were summarised using the mean and standard deviation (SD) [or median and interquartile range (IQR) if the distribution was skewed] and categorical data were summarised as a number and percentage. All treatment comparisons are presented as effect sizes with 95% CIs, and p-values of < 0.05 from likelihood-ratio tests have been considered statistically significant. However, as the trial was stopped early, it was very underpowered to detect clinically important differences.

It was intended to adjust all formal comparisons of OPCABG versus ONCABG for surgeon and the factors used in the cohort minimisation. However, owing to the reduced sample size and resultant low event rates of some of the cohort minimisation factors, all models were adjusted for age, sex and operative priority as fixed effects and surgeon as a random effect. All underlying model assumptions were checked using standard methods (e.g. residual plots, tests for normality or for proportional hazards). Outlying observations that meant models did not fit the data adequately were excluded from analyses and are indicated in table footnotes.

The primary analysis is the proportion of patients experiencing the composite outcome of death or major morbidity (CRISPSw) up to 30 days and has been analysed using logistic regression with the treatment effect reported as an OR. Component events have been presented separately by occurrence pre or post hospital discharge. The duration of CICU stay and hospital stay were analysed as time to event outcomes, with patients who die prior to CICU/hospital discharge censored at the time of their death. Comparisons were performed using Cox proportional hazards models and treatment comparisons are presented as HRs. The validity of the assumption of proportional hazards was tested and, if violated, a model with a time-dependent covariate (the interaction term between the treatment and the survival time) was used. Random effects terms were fitted via the use of shared frailty terms. 56

For all QoL data, standard rules have been used to derive outcome measures. Rose angina and CCS angina class both result in ordinal outcomes ranging from no angina symptoms to ordered grades of angina symptoms. EQ-5D data are in two sections, the first consisting of five ordinal questions (which, for the patients who used the standard EQ-5D questionnaire, is converted into an EQ-5D single summary index) and the second a visual analogue scale. Finally, data from CROQ questionnaires are used to derive seven continuous scores, including an overall ‘core total’ score.

Rose and CCS angina class data at 4–8 weeks post surgery have been dichotomised into any angina symptoms versus no angina symptoms. Treatment groups have been compared using logistic regression, adjusting for the appropriate preoperative angina class as a categorical outcome, with treatment effects reported as ORs. Formal statistical comparisons of treatment effects have been performed only if > 10 patients in total experience the angina outcome (with at least one event in each treatment group). Responses to the five EQ-5D ordinal questions have been tabulated but no formal analyses undertaken (see Appendix 2 , Table 21 ). EQ-5D single summary index and visual analogue scale data and the CROQ core total score have been analysed using linear mixed effects methodology. Pre and postoperative values were modelled jointly to avoid the necessity to either exclude cases with missing preoperative measures or to impute missing preoperative values. Multivariate normal models were fitted incorporating separate parameter estimates for the mean baseline response and for each treatment at the 4–8 week time point (i.e. saturated model with time fitted as a categorical variable).

Safety data have been reported on the safety population, defined as all randomised patients excluding patients who withdrew prior to surgery, as it was decided not to perform surgery, and patients who withdrew at any time and were unwilling for any data collected to be used. Expected events (i.e. listed in the study protocol as expected prior to hospital discharge following cardiac surgery) and unexpected events (any event not listed in the protocol occurring before discharge and any event occurring after hospital discharge) have been tabulated separately (see Tables 15 – 17 ), with events that meet the criteria (prolonged an ongoing hospitalisation/resulted in hospitalisation, resulted in death, was life-threatening or resulted in persistent or significant disability/incapacity) of a SAE identified. Events have been presented and grouped by the treatment received, rather than the treatment allocated, and no formal comparisons between treatment groups have been made.

No formal corrections have been made for multiple testing, but the number of statistical comparisons has been limited and our interpretation of the results takes into account the magnitude and consistency of effect estimates. No subgroup or sensitivity analyses were performed. A planned subgroup analysis to compare the treatment effect in patients with an additive EuroSCORE of < 8 and ≥ 8 was planned but not performed owing to the early termination of the trial.

Missing data in all tables are indicated by footnotes. There were no missing data for the primary outcome or the time to event outcomes. Missing data for QoL outcomes were infrequent (< 5%) and, therefore, cases with missing postoperative values have been excluded from analyses. For cases with complete postoperative but missing preoperative data, the joint modelling of continuous data avoids the necessity to impute such data under the assumption that data are missing at random, but for categorical data the most common category across both treatment groups has been imputed. Owing to the low levels of missing data, it was judged that more complex missing data approaches (e.g. multiple imputation) would be unlikely to recover any additional information.

All statistical models were fitted in Stata version 12.0 (StataCorp LP, College Station, TX, USA). All other analyses and data management were performed in SAS version 9.2 (SAS Institute Inc., Cary, NC, USA).

Health economics

Given the early cessation of the trial (see Chapter 3, Decision to close the trial early), unit cost estimates for valuing resource utilisation data had not yet been collected. This, plus the small sample sizes at trial cessation, precludes the calculation of the costs associated with each method of CABG, as well as estimates of cost-effectiveness. Resource utilisation data reported for each arm of the trial are, therefore, limited to key items consumed during the index hospital admission for surgery, including duration of operation, duration on ventilation, time in CICU, time in HDU and time on a ward.

Following general guidance issued by the National Institute for Health and Care Excellence, continuous data are presented using mean and SDs for each group. Differences between groups are presented using the mean difference (MD) and 95% (bootstrapped) CI for the difference.

Centres

The CRISP trial planned to recruit patients from 40 centres, 20 in the UK and 20 overseas. In the recruitment period from October 2009 to March 2011, patients were recruited from eight centres in the UK and one centre in India. A total of 39 surgeons participated: 19 ONCABG specialists and 20 OPCABG specialists. The number of surgeons at each centre ranged from two to nine ( Table 2 ). The proportion of consultant surgeons at a centre participating in CRISP ranged from 20% to 100%.

In addition to the nine participating centres, a further five UK centres (University College London; Sussex Cardiac Centre, Brighton; The Cardiothoracic Centre, Liverpool; Nottingham University Hospital; and South Tees Hospital, Middlesbrough) had the necessary approvals in place to start but had not recruited any trial participants before the study closed to recruitment in March 2011 at the request of the funder (see Decision to close the trial early). Two UK centres, in Edinburgh and Cardiff, and 10 overseas centres were at various stages of the research approvals process when the study closed (see Appendix 1 ).

Screened patients

A total of 787 patients were assessed for potential inclusion in the trial. Six hundred and eighty-one were excluded: 523 were ineligible, 82 were eligible but not approached, 74 were approached but did not consent and two were omitted for other reasons. The numbers of patients screened, found to be ineligible, not approached, did not consent and randomised are given by centre in Table 3 and demonstrate different proportions of ineligible patients between centres (range 0% to 76%). This reflects the fact that some centres did not screen all potential patients.

The majority of ineligible patients [493 out of 523 (94%)] had an additive EuroSCORE of < 5. Other reasons for ineligibility, non-approach and non-consent are given in Figure 1 . Reasons for eligible patients not being approached included (1) cancellations and transfers to another surgeon’s list, (2) a decision not to operate, (3) time constraints and (4) a surgeon’s decision.

FIGURE 1.

Flow of participants. a, The exclusions section is incomplete as not all sites have entered full screening data; b, some patients may be ineligible for more than one reason; c, one further patient (not included on flowchart) withdrew pre surgery but was happy for data collection to continue; therefore, the patient is included in all applicable analyses (for details of all withdrawals see Table 8 ); and d, for further details see Table 9 .

The main reason given for patients declining to take part was personal reasons, followed by the patient having a preference for a specific surgeon.

Even at the Bristol and Oxford centres, where the screening data were most complete, 75 and 39 eligible patients, respectively, were identified each year on average: significantly fewer than the average 300 eligible patients identified retrospectively from an institutional database of all cardiac procedures over the same period in Bristol. The main reasons for the deficit were (1) not all surgeons were participating in CRISP, (2) only willing OPCABG surgeons could participate if logistical problems (e.g. time constraints or surgeon unavailability) required a within-surgeon allocation, (3) other trials were recruiting from the same pool of patients in the same time period (although CRISP was prioritised over other trials in Bristol).

Recruitment

A total of 106 eligible patients were recruited into the study from October 2009 to March 2011. Patient follow-up was completed in June 2011. The study was closed to recruitment in March 2011 at the request of the funder (see Decision to close the trial early).

Recruitment pathway

The logistics of identifying eligible patients, recruiting them into the trial and organising the surgery within an expertise-based allocation framework was recognised as the key challenge for participating centres. It was acknowledged that the recruitment pathway could vary between centres in order for them to meet this challenge while continuing to work and operate within national and local protocols. The recruitment pathway envisaged before commencement of the trial, modelled on the recruitment pathway at the Bristol centre, is described in Figure 2 .

FIGURE 2.

Recruitment pathway (Bristol model).

When presenting the study at site initiation visits it became apparent that this exact model would not be applicable at all centres. The model developed at Wolverhampton, where the majority of referrals are to a named surgeon, is shown in Figure 3 .

FIGURE 3.

Recruitment pathway (Wolverhampton model).

Recruitment rate

When the CRISP trial was designed, it was estimated that each centre would recruit at least six patients per month. This estimate was based on data from the Bristol and Oxford centres, where between 200 and 300 eligible patients underwent CABG each year. Based on previous trials, it was anticipated that 40% of eligible patients would be recruited,22 which would have resulted in an annual recruitment rate of between 80 and 120 patients per year. This target was not met at any participating centre. Two centres (Blackpool and Bristol) recruited five patients in 1 month and three centres (Blackpool, Wolverhampton and India) each recruited four patients in 1 month. The number of patients recruited by month and centre is shown in Table 4 and cumulative predicted and actual recruitment is shown in Figure 4 .

TABLE 4 - Number of patients recruited by month and centre

Centre	Month of randomisation																	Total
	2009		2010												2011
	November	December	January	February	March	April	May	June	July	August	September	October	November	December	January	February	March
Basildon	0	0	1	3	0	0	0	0	0	0	2	0	0	0	0	0	0	6
Blackpool	0	0	0	0	0	0	4	5	2	2	2	0	1	0	3	1	1	21
Bristol	1	1	1	3	0	1	2	1	2	1	3	1	5	0	1	2	1	26
King’s College	0	0	0	0	0	0	0	0	0	0	1	1	0	0	1	1	1	5
Oxford	0	0	0	0	3	3	3	1	1	1	3	1	2	1	0	0	0	19
Papworth	0	0	0	1	0	0	1	0	0	0	2	1	0	0	0	0	0	5
Sheffield	0	0	0	0	0	0	0	1	1	2	1	0	0	1	0	1	0	7
Wolverhampton	0	0	0	0	0	0	0	0	0	0	4	2	0	3	0	2	0	11
India	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	2	4	6
Total	1	1	2	7	3	4	10	8	6	6	18	6	8	5	5	9	7	106

FIGURE 4.

Predicted and actual recruitment. The predicted number of patients assumes six patients recruited per centre per month (predictions in study protocol).

Recruited patients

Screening data are compared between ineligible, eligible but non-consenting and randomised patients in Table 6 . Ineligible patients were on average younger, less likely to be female and less likely to have preoperative conditions that result in higher additive EuroSCORE, e.g. chronic pulmonary disease, extracardiac arteriopathy, unstable angina or recent MI.

Differences in randomisation practices between centres are shown in Table 7 . There was wide variation in the proportion of patients randomised using expertise-based randomisation and the median times from randomisation to surgery, although the numbers of randomised patients per centre are small. Overall, patients were randomised earlier using expertise-based randomisation (median 17.5 days prior to surgery, IQR 7–42 days) than using within-surgeon randomisation (median 3.5 days, IQR 1–16 days).

The numbers of urgent and elective patients recruited varied across centres (see Table 7 ). In Blackpool and the centre in India, the patients were predominantly urgent cases (20 out of 21 in Blackpool and 6 out of 6 in India), while in Oxford and Wolverhampton the majority were elective (15 out of 19 and 9 out of 11, respectively). At the other centres, similar numbers of elective and urgent cases were recruited.

Patient withdrawals

Eight of the 106 randomised patients were excluded from the analysis population: six patients withdrew prior to surgery and two patients died prior to surgery. Therefore, 98 patients underwent surgery and have been included in the principal analysis population, 49 in the OPCABG group and 49 in the ONCABG group (see Figure 1 ).

Five patients were withdrawn because it was decided that surgery was no longer required and one patient withdrew on the day of randomisation with no further details given. A further patient (OPCABG group) also withdrew their consent preoperatively owing to anxiety that they were not randomised to ONCABG; however, they were happy to be followed-up and for their data to be used and so remained in the analysis cohort. Table 8 shows a summary of withdrawals; for full details, see Appendix 2 , Table 20 .

Protocol deviations

There were 21 protocol deviations in 19 patients ( Table 9 ). Four patients randomised to OPCABG did not receive their allocation and there were no crossovers in the ONCABG group. Reasons for not receiving the allocated treatment were (1) development of ST segment on ECG during manipulation of the heart, (2) unplanned additional procedure required, (3) VF/VT arrest and (4) myocardial ischaemia with ST changes and low blood pressure. Other types of protocol deviations were (1) patient did not meet eligibility criteria (n = 4), (2) the operating surgeon was not on the approved list of trial surgeons (n = 6), (3) expertise-based randomisation was used but the surgeon was not an expert in the allocated surgery type (n = 6), and (4) within-surgeon randomisation was used with an ONCABG surgeon (n = 1). Data on all patients for whom there was a protocol deviation were included in the trial analyses on an intention-to-treat basis.

Patient follow-up

Follow-up data 4–8 weeks after surgery were obtained for all 98 patients in the principal analysis population: 87 patients attended their follow-up visit, two patients died prior to their visit and nine did not attend but data were retrieved from their clinical notes and/or general practitioners.

Numbers analysed

Ninety-eight patients in the principal analysis population were included in all tables of demographic and operative characteristics and analyses of the primary outcome and duration of CICU/hospital stay. Ninety-seven patients were included in QoL analyses: (1) 90 patients with both preoperative and 4–8 weeks postoperative data, (2) six patients with preoperative data only and (3) one patient with postoperative data only. One hundred patients were included in the safety analyses: the 98 patients in the principal analysis population plus the two patients who died prior to surgery.

Baseline data and operative characteristics

Patient demographics and preoperative characteristics are presented in Table 10 . The median EuroSCORE was 6 (IQR 5–8), the median age 77.1 years (IQR 71.9–80.6 years) and 23 patients (23%) were female. Most patients (95%) had good or moderate LV function and low proportions of patients (< 15%) experienced the remaining EuroSCORE components, with the exception of extracardiac arteriopathy (32%) and recent MI (49%). The majority of patients were non-diabetic (76%), were past or current smokers (62%) and had triple-vessel disease (77%). Approximately half (45%) of procedures were classified as urgent. Characteristics were generally similar between the two groups. However, more patients in the OPCABG group than in the ONCABG group had New York Heart Association (NYHA) functional classification of heart failure as Grade I (46% vs. 20%, respectively) and no patients in the OPCABG group had an abnormal heart rhythm or a pacemaker, but five patients in the ONCABG group had an abnormal heart rhythm and four had a pacemaker. Conversely, slightly more patients in the OPCABG group had > 50% disease in the left main stem and hypertension requiring treatment (39% vs. 27% and 84% vs. 76%, respectively). In addition, angiotensin-converting enzyme (ACE) inhibitor/angiotensin receptor blocker (ARB) II and beta blocker use was more common in the OPCABG group (82% vs. 65% and 84% vs. 65%, respectively). Finally, the average heart rate was lower in the OPCABG group [median 63 beats per minute (b.p.m.) (IQR 58–72 b.p.m.) vs. median 70 b.p.m. (IQR 60–85 b.p.m.)].

Operative characteristics are given in Table 11 . Fewer patients had three or four grafts in the OPCABG group (63% vs. 79%, respectively). Use of the partial aortic clamp was more frequent in the ONCABG group than in the OPCABG group (88% vs. 65%, respectively) and use of a cell saver more frequent in the OPCABG group than in the ONCABG group (78% vs. 47%, respectively). More patients in the ONCABG group were paced than in the OPCABG group (27% vs. 14%, respectively) and more received red blood cell transfusions (31% vs. 16%, respectively). The blood product activated factor VII, which may or may not be given postoperatively, was not given to any participants in this cohort. There were no clear differences in terms of the grafts used and there were no deaths during surgery. The duration of the operation measured from the start of the operation (knife to skin) to the end of the procedure (patient leaves theatre) was slightly shorter in the OPCABG group (median 3.2 hours, IQR 2.7–3.9 hours) than in the ONCABG group (median 3.4 hours, IQR 3.0–4.2 hours). The MD was 0.22 of an hour (approximately 13 minutes).

Primary outcome

In both the OPCABG and ONCABG groups, 6 out of 49 (12%) patients experienced the primary outcome in the first 30 days ( Table 12 ). The estimated treatment effect, adjusted for age, sex, operative priority and surgeon, was OR = 1.07 (95% CI 0.27 to 4.14; p = 0.93).

The most commonly occurring component of the primary outcome was MI (occurring in six patients) and the rarest were death and sternal wound dehiscence (experienced by one patient each). All but one of the constituent events occurred prior to discharge from hospital following cardiac surgery.

Secondary outcomes

Resource use

Resource-use data are summarised in Table 14 . On average, patients randomised to ONCABG spent 0.22 of an hour (approximately 13 minutes) longer in surgery than patients randomised to OPCABG. Time on ventilation after surgery, measured from the time the operation ended to the time the patient was extubated, was longer for patients in the ONCABG group (median 7.1 vs. 5.7 hours). On average, patients randomised to ONCABG also spent longer in the CICU (median 27.7 vs. 26.0 hours), although this difference was not statistically significant ( Figure 5 ). Of those admitted to HDU, the stay was, on average, 37.3 hours (1.6 days) longer in the ONCABG group. In total, six patients were not admitted to a ward; one (in the ONCABG group) had died postoperatively but prior to hospital discharge. On average, of the patients admitted to a ward, length of stay was again longer in the ONCABG group. After surgery, patients randomised to ONCABG spent longer in hospital than patients randomised to OPCABG (median 8 vs. 7 days, Figure 6 ).

TABLE 14 - Resource use in the period from surgery to 6–8 weeks after surgery

Confidence intervals for MDs are bootstrapped. HRs are from Cox proportional hazards models, with age and sex included as fixed effects, operative priority as a fixed effect (duration of CICU stay) or a time-dependent covariate (duration of hospital stay) and a shared frailty term for operating surgeon.

The large SD in the OPCABG group was due to the presence of outliers.

The median and mean values differ because the distributions are highly skewed.

Initial ventilation time only, excluding any further periods of ventilation. Four patients were re-intubated: one patient was re-intubated for 9.6 hours duration and one patient was re-intubated for 521 hours duration (OPCABG group); one patient was re-intubated for 52.5 hours duration and one patient was re-intubated twice, firstly for 58 hours then for 67.4 hours (ONCABG group).

Initial CICU admission only, excluding any further periods of readmission to CICU. Three patients were readmitted to CICU: one patient was readmitted for 27 days (in the OPCABG group) and two patients were readmitted for 2 days each (both in the ONCABG group). The time to discharge was treated as a censored observation for one patient (in the ONCABG group) who died in CICU (length of stay = 253 hours).

Subset of patients admitted to HDU/ward. One patient was readmitted to HDU for 1 day, one patient for 3 days and one patient for 4 days (all in the ONCABG group). Two patients were readmitted to the ward for 4 days each (both in the ONCABG group).

Estimated from operation date and hospital discharge date. The mean values for postoperative total inpatient stay for each group calculated using these dates are slightly greater than equivalent mean values generated by adding the individual CICU, HDU and ward components (based upon hours). The difference between the two groups, however, is similar regardless of the calculation approach.

Data on duration or reason for reoperation were not collected.

Data missing for 22 patients, as this question was not included on early versions of the CRFs.

Three patients with missing data (one in the OPCABG group and two in the ONCABG group). Note that these are patients who did not attend the 4- to 8-week visit.

The three unplanned procedures in the OPCABG arm were (1) ligation of right lung bulla, (2) supra pubic catheter and (3) mitral valve repair and left arterial maze, and left appendage axis.

Resource	Randomised to OPCABG (N = 49)		Randomised to ONCABG (N = 49)		Effect (95% CI)a
	n	%	n	%
Intraoperative
Duration of surgery (hours)b
Median (IQR)	3.2 (2.7–3.9)	–	3.4 (3.0–4.2)	–
Mean (SD)	3.39 (1.18)	–	3.61 (0.86)	–	MD −0.22 (−0.601 to 0.209)
Postoperative
Red blood cells used	25	51	25	51
If yes, median (IQR) units	1.0 (1.0–2.0)	–	2.0 (2.0–4.0)	–
Plasma used	7	14	3	6
Platelets used	9	18	5	10
Any haemostatic agents used	16	33	17	35
Tranexamic acid	10	20	11	22
Activated factor VII	0	0	0	0
Other haemostatic agent	6	12	8	16
Duration of ventilation (hours)c , d
Median (IQR)	5.7 (4.9–11.3)	–	7.1 (4.9–14.3)	–
Mean (SD)	12.0 (23.2)	–	17.5 (36.4)	–	MD −5.48 (−18.13 to 6.36)
CICU stayc , e
Median hours (IQR)	26.0 (21.3–65.1)	–	27.7 (20.7–66.5)	–	HR 1.15 (0.69 to 1.91)
Mean hours (SD)	45.9 (49.4)	–	55.1 (58.8)	–	MD −9.20 (−30.2 to 11.7)
Mean days (SD)	1.91 (2.06)	–	2.29 (2.45)	–	MD −0.38 (−1.26 to 0.48)
Admitted to HDU	29	59	27	55
HDU stayc , f
Median hours (IQR)	41.0 (25.8–72.0)	–	48.8 (29.0–100)	–
Mean hours (SD)	57.95 (52.03)	–	95.23 (145.07)	–	MD −37.28 (−99.2 to 8.86)
Mean days (SD)	2.41 (2.17)	–	3.97 (6.04)	–	MD −1.55 (−4.36 to 0.38)
Admitted to ward	48	98	44	90
Ward stayf
Median hours (IQR)	98.0 (70.9–139)	–	94.8 (72.5–143)	–
Mean hours (SD)	110.0 (56.37)	–	136.7 (126.9)	–	MD −26.7 (−68.5 to 15.1)
Mean days (SD)	4.58(2.35)	–	5.69 (5.29)	–	MD −1.11 (−2.97 to 0.47)
Hospital stayg
Median days (IQR)	7 (6–9)	–	8 (6–10)	–	HR 1.26 (0.81 to 1.95)
Mean days (SD)	8.49 (4.98)	–	10.12 (7.39)	–	MD −1.63 (−4.03 to 0.83)
Reoperationh	0	0	4	8	MD −0.082 (−0.159 to −0.005)
Other unplanned procedurei	3	8	0	0	MD 0.079 (−0.007 to 0.165)
Medications at discharge
ACE inhibitors/ARB II	26	53	22	45
Beta blockers	38	78	35	71
Calcium antagonists	5	10	0	0
Statins	45	92	47	96
Aspirin/clopidogrel	47	96	47	96
Medications at 4–8 weeks
ACE inhibitors/ARB II j	24	50	26	55
Beta blockers j	35	73	34	72
Calcium antagonists j	7	15	2	4
Statins j	42	88	43	91
Aspirin/clopidogrel j	46	96	42	89

FIGURE 5.

Duration of CICU stay.

FIGURE 6.

Duration of postoperative hospital stay.

Adverse events and postoperative complications

Main findings: study conduct

The main findings of the CRISP trial are that expertise-based randomisation is challenging to implement and make work in a tertiary surgical setting. For a range of logistical reasons, the trial failed to recruit in time and to target, and the proposal to extend the primary outcome to include (1) reoperation for bleeding, (2) low cardiac output, (3) new onset of atrial arrhythmia and (4) AKI, and thereby reduce the study size, was not accepted by the funder. The trial was closed prematurely on the grounds of futility and also because of the perceived overlap between CRISP and the Canadian-led CORONARY trial.

Some of the challenges faced in CRISP were due to the context and nature of the service provision in the UK. Cardiac surgery is a tertiary service. As a consequence, patients are referred from a large geographical area and a significant proportion of referrals are urgent inpatients waiting in neighbouring ‘feeder’ hospitals for a suitable surgical slot to become available. The information provided at referral was often limited, making the assessment of eligibility for the trial by a research nurse or co-ordinator difficult. CRISP was marketed as a trial in high-risk patients. It was therefore important that only patients likely to be eligible were contacted, to avoid undue stress to patients at lower risk of complications.

Optimising the recruitment pathway was difficult, and the challenges varied according to how the local service was organised. Elective patients were usually seen at least once before surgery in an outpatient referral and/or preoperative assessment clinic. These contacts provided opportunities for the local research team to engage with potential participants, discuss the trial and seek consent, but often patients were unwilling to take part because they either wished to stay with the surgeon they met at the first appointment or wanted the surgeon to decide which type of surgery was best for them. Frequently, the need for surgery was not discussed until this first appointment so contacting a patient in advance of this was not considered appropriate. Urgent patients presented a different challenge. In the centres with a policy of transferring patients to the cardiac centre 2 or 3 days before surgery, the recruitment window was adequate and expertise-based randomisation was achievable, provided experts in both ONCABG and OPCABG were available to carry out the surgery. In centres where the policy was to transfer the patient as close to surgery as possible, recruitment and expertise-based randomisation was severely hampered. The CLRN was not long established when CRISP was set up and support from CLRN research nurses working at ‘feeder’ hospitals to facilitate recruitment was not forthcoming. This may not be the case now. Research governance issues were also a limiting factor, the concept of the research passport was not working well at that time and the need to identify local principal investigators at hospitals where the study was not taking place and in a speciality that was not theirs proved impossible.

Some of these issues were relevant to the context and setting in which CRISP was based only, but others were not. The availability of an expert surgeon to carry out the operation within a time scale that does not breach local and national targets for treatment applies to any surgical trial using expertise-based randomisation. The allocation of patients to surgeons through a system of named referrals, or via a generic pool, and the willingness or otherwise of surgeons to work together and ‘share’ their patients is a challenge and potential barrier to recruitment into any trial using expertise-based randomisation. The majority of surgeons continue to work autonomously, but this is gradually changing with the appointment of a clinical director or a chief of service; however, this is by no means widespread, particularly in the UK. When a patient is referred directly to an individual surgeon, that surgeon becomes responsible for that patient. Surgeons are often reluctant to transfer the patient to another surgeon, especially after meeting the patient and the ‘doctor–patient bond’ has formed. In addition, there continues to be a strongly held belief that the length of a surgeon’s waiting list reflects his or her surgical ability. Similarly, understanding the recruitment pathway and optimising when and how to introduce the trial to patients to ensure that surgeon preferences do not influence patient decisions is relevant to all surgical trials.

Many of these barriers to recruitment have been encountered previously. Ross et al.,58 in 1999, identified time constraints, lack of staff and training, worry about the impact on the doctor–patient relationship, concern for patients, loss of professional autonomy, difficulty with the consent procedure and lack of rewards and recognition as the key clinical-based barriers, while, for patients, the main barriers were the additional demands of the trial, patient preferences, worry caused by uncertainty and concerns about information and consent. A survey from 2011 of centres recruiting to three trials in head and neck surgery, all of which were significantly delayed and behind target, identified patient and surgeon preferences, insufficient time in the NHS clinic, lack of research nurse support, insufficient funding for excess treatment costs and delays in the approval process as the key barriers. 59 Complex recruitment pathways involving staff across different specialties/centres have also hindered recruitment in other trials. 60

In addition, for a trial such as CRISP to recruit successfully in the UK health-care setting, there has to be an agreement when the research is funded that a centre as a whole will participate in the study. The surgical autonomy needs to be broken down and the structure of the NHS, with consultants responsible for their own patients, is a stumbling block that is not limited to expertise-based recruitment. Surgeons need to work together and there need to be improved links between those responsible for service delivery and for the research. In the UK, the NHS is under huge pressure to deliver services and treatment to target, while at the same time reducing costs. Expertise-based recruitment, with a limited number of surgical experts, will almost inevitably lead to longer waiting times for some patients. For it to be implemented successfully in a surgical trial, the service providers and the health-care commissioners need to be committed to the research and be prepared to allow some flexibility in the targets in order for the research to succeed. Similarly, research needs to be considered an integral part of the service provision of a hospital; strategies for reducing hospital-based costs often impact on research. For example, patients are increasingly spending less time in hospital before their surgery and so the opportunities for recruitment are restricted. This was a particular problem for high-risk urgent in-hospital transfers (ideal candidates for the CRISP trial) as these patients will not have attended the cardiac centre previously and so there were no opportunities for earlier recruitment. Similarly, there needs to be a greater flexibility in the implementation of the research governance framework in NHS hospitals and within the CLRN. The need for local principal investigators at ‘feeder’ hospitals and the unwillingness of CLRN nurses at these hospitals to facilitate recruitment caused particular frustration.

Main findings: study results

The CRISP trial did not find statistically significant differences between the OPCABG and ONCABG groups owing to the limited power (< 2% of the target number of patients was recruited). However, the question that the trial set out to address remains important. The Cochrane review, published in 2012,34 acknowledged that mainly patients with low risk of postoperative complications were enrolled in the 86 trials reviewed and patients with three-vessel coronary disease and impaired LV function were under-represented.

The two largest trials to compare ONCABG and OPCABG, the ROOBY37 and the CORONARY40 trials, have been published since the CRISP trial began. The ROOBY trial, which contributed 2203 patients to the Cochrane review, has been severely criticised. The operative experience of the surgeons in the OPCABG group was substantially less than that of the ONCABG surgeons (median of 50 patients per surgeon), which was reflected in a high conversion rate from OPCABG to ONCABG (12%), a significant proportion of patients receiving fewer grafts than planned (18% OPCABG vs. 11% ONCABG),37 significantly lower patency rates (arterial conduits: 85.8% vs. 91.4% and saphenous vein grafts: 72.7% vs. 80.4%) at 1 year and fewer patients with effective revascularisation (50.1% vs. 63.9%) with OPCABG compared with ONCABG. 61 The trial also recruited predominantly low-risk patients.

The CORONARY trial, the largest trial to date, recruited a higher proportion of higher-risk patients than the ROOBY trial, although < 20% of participants had a EuroSCORE of > 5. 40 This compares with 74% of patients recruited to CRISP. The participating surgeons were also more experienced than those recruited to the ROOBY trial: all surgeons were required to have > 2 years’ experience and have completed > 100 procedures involving their preferred technique. Trainees were not allowed to be the primary surgeon for any procedure. This experience threshold was consistent with that used in CRISP.

The Cochrane meta-analysis has been updated to include the results from the CORONARY and CRISP trials. The results, for all-cause mortality, MI, stroke and renal failure are summarised in Table 18 . The RR of death and MI reduced from 1.24 (95% CI 1.01 to 1.53) to 1.18 (95% CI 0.98 to 1.40) and from 1.00 (95% CI 0.79 to 1.26) to 0.96 (95% CI 0.82 to 1.12) respectively, while the RR of a stroke and a renal complication increased from 0.76 (95% CI 0.54 to 1.06) to 0.80 (95% CI 0.61 to 1.06) and from 0.86 (95% CI 0.62 to 1.20) to 0.92 (95% CI 0.70 to 1.21), respectively.

The Cochrane review identified three trials in high-risk patients: the BBS trial, which recruited 341 patients with a EuroSCORE of ≥ 5 and triple-vessel disease;38,62 a trial by Carrier and colleagues, which recruited 65 patients with at least three of the following criteria: age > 65 years, high blood pressure, diabetes, creatinine > 133 mol/l, LV ejection fraction < 45%, chronic pulmonary disease, unstable angina, congestive heart failure, repeat CABG, anaemia and carotid atherosclerosis;41 and a study in 128 patients with a ST-segment elevation MI. 42 The data from these trials, plus CRISP, have been combined in meta-analyses, the results of which are shown in Figures 7 – 10 . Part (a) of each figure is restricted to early outcomes (30 days or hospital discharge) and part (b) includes outcomes across the full follow-up period of each study. The BBS trial and the trial in patients with a ST-segment elevation MI reported cardiac-related mortality outcomes to 3 years, while CRISP and the trial by Carrier et al. 41 reported outcomes to 30 days only. It was not possible to include the CORONARY trial results in these meta-analyses and the data were not reported for the individual components of the trial’s composite outcome for the subgroup of high-risk patients. In contrast to the Cochrane review, these analyses suggest a lower risk of death with OPCABG in the early postoperative period (RR 0.46, 95% CI 0.20 to 1.04; p = 0.06) and a comparable risk overall (RR 0.90, 95% CI 0.32 to 2.58; p = 0.85). The risk of an MI was also reduced in the early postoperative period (RR 0.59, 95% CI 0.33 to 1.06; p = 0.077). No differences in the risk of a stroke or of renal complications were found.

FIGURE 7.

Meta-analysis of trials in high-risk patients. a, Death up to day 30/hospital discharge; and b, death up to 3 years.

FIGURE 8.

Meta-analysis of trials in high-risk patients. a, MI up to day 30/hospital discharge; and b, MI up to 3 years.

FIGURE 9.

Meta-analysis of trials in high-risk patients. a, Stroke up to day 30/hospital discharge; and b, stroke up to 3 years.

FIGURE 10.

Meta-analysis of trials in high-risk patients. a, Renal complications up to day 30/hospital discharge; and b, renal complications up to 3 years.

The BBS and CORONARY trials both reported the results of a composite primary outcome at 30 days in high-risk patients and the composites varied across studies. The BBS trial38 used death, MI, cardiac arrest, low cardiac output, stroke and coronary reintervention, while the CORONARY trial40 used death, MI, stroke and new renal failure requiring dialysis. These compare with the CRISP composite of death, new renal failure, MI, stroke, prolonged ventilation and sternal wound dehiscence. The composites from these studies were combined in a meta-analysis and the results are summarised in Table 19 . As anticipated, the pooled estimate reflects the estimate from the large CORONARY trial, but with a narrower CI.

One possible reason for the lack of compelling evidence of a difference between OPCABG and ONCABG in the recent trials is that over time techniques in ONCABG have improved. Different methods of cardioplegia and body temperature cooling have been introduced to reduce myocardial injury and systemic inflammatory response during surgery and a miniaturised CPB circuit has been developed that is associated with a non-significantly reduced risk of adverse outcomes. 63 There may also have been ill-defined temporal improvements in care across both techniques.

Strengths and limitations

Despite the failure of CRISP to recruit to target, the options to improve recruitment were thoroughly tested. There was a strongly held view that the expertise-based randomisation was the key barrier to successful recruitment, but, when we attempted to change to a within-surgeon allocation, many of the OPCABG experts were no longer willing to participate. A survey of orthopaedic surgeons similarly found a strong preference for expertise-based randomisation. 64 We believe that expertise-based randomisation is the only way to evaluate established surgical procedures where there are strongly held preferences but collective equipoise. Furthermore, it avoids the problem of differential expertise bias,65 can protect against crossover as a result of unfamiliarity or less experience with one surgical method and allows for greater surgeon participation. In addition, an expertise-based design provides the participant with the assurance that the surgery will be carried out by a surgeon who has both the appropriate expertise and is comfortable carrying out the procedure. Expertise-based randomisation has been used successfully in other areas, for example in studies comparing coronary angioplasty and CABG66–69 and in orthopaedic surgery. 69 However, we have to recognise that it may not be feasible in a tertiary referral setting, when the referral information to determine patient eligibility is often inadequate, surgeon availability is limited and there is an imbalance in the numbers of surgical experts at a centre.

The trial was methodologically strong; the risk of bias was minimised through concealed allocation and objective definitions for the primary end points. There was a blinded review of the blood results and preoperative and postoperative ECGs of all patients and a postoperative MI defined on consensus of the adjudicators. The database used to collect the data was robust and included extensive within-CRF and cross-CRF validation. The screening data were incomplete for most centres, as indicated by the wide variation in the proportion of screened patients recruited, and this is a weakness that was recognised by the TSC.

Despite the poor recruitment, the CRISP patients reflect the population the trial was designed to study. Using data from the Bristol cardiac surgery database, we compared the characteristics of the CRISP patients with 3364 eligible isolated CABG patients with a EuroSCORE of ≥ 5 who had undergone an operation between April 1997 and August 2012 in Bristol. The cohorts were of similar age (median 77 vs. 74 years) and sex mix (23% vs. 28% female) and comorbidities occurred with similar frequency (diabetes 24% vs. 26%, previous MI 70% vs. 68%, previous stroke 8% vs. 6%, median EuroSCORE was 6 in both cohorts). In addition, similar proportions had triple-vessel disease (77% vs. 73%) and > 50% disease in left main stem (33% vs. 30%). However, there was a lower proportion of patients with poor ejection fraction (5% vs. 11%) and the proportion of patients requiring surgery urgently was lower in the CRISP study (45% vs. 66%), which is reflective of the recruitment difficulties.

The final study size is a clear weakness: the study has low power to detect significant differences between the groups and the value of the trial data is their contribution to meta-analyses. However, the approach to the analysis of the data was strong. An analysis plan was prepared in advance of any comparative analyses of the study data and the number of statistical tests carried out was restricted. Formal statistical comparisons of treatment effects were only carried out if > 10 patients in total experienced the outcome (see Appendix 5 ), to minimise the probability of a type 1 error.

We chose to use an additive EuroSCORE of ≥ 5 as a marker of ‘high risk’. All scoring systems have their limitations and this score is strongly influenced by age (one point for every 5 years from 60 years onwards) and less by a participant’s comorbidity. As a consequence, CRISP recruited more elderly patients than the CORONARY trial (median age 77 vs. 68 years, respectively) and many fewer diabetic patients (24% vs. 47%, respectively), although, in both trials, only 5% of patients had poor LV function. The question of which treatment option is most effective, ONCABG or OPCABG, remains an important question in the large group of patients with poor LV function that cannot be answered by either trial.

Lessons for the future

If we were setting up the CRISP trial now there are many things that we would do differently. First, we would design the trial in two phases, with a feasibility phase followed by a main trial phase. This design is being used in other surgical areas and is an attractive option for funders of difficult-to-do trials.

Second, we would include a qualitative research element, which would involve researchers interviewing the research teams at the study centres in order to gain a full understanding of the recruitment pathway, barriers to recruitment (including a willingness or otherwise to work together and share patients) and the extent of the equipoise. Through feedback and training, the study team (including the surgeons) would be taught how to present the trial in an unbiased way to minimise the number who decline to take part. The strength of the bond formed between surgeon and patient at that first referral would also be explored through interviews with patients who did and did not agree to take part. This approach has been used very successfully in the POTECT trial of surgery versus radiotherapy versus medical management in men with localised prostate cancer. 70 A total of 1500 men were recruited to a trial that many strongly believed would never succeed. Failure to meet the recruitment target is a common problem71 and qualitative methods have been recommended as the most effective for identifying and overcoming barriers to clinician recruitment activity and increasing recruitment. 72

Third, we would focus recruitment equally towards UK and overseas centres from the start. Many of the barriers to recruitment experienced in the UK may not be such a problem overseas. Although the centre in India was actively participating for only a short period before CRISP closed, it recruited six patients in 3 weeks, which was more than was achieved in any UK centre. The CORONARY trial, which successfully recruited 4752 patients at 79 centres, recruited only a small number of patients from the UK (227 patients, < 5%). The biggest contributors were India and China (1307 and 781 patients, respectively).

Future research

The answer to the question of whether OPCABG offers an additional benefit over ONCABG in a high-risk population is unclear. The trial evidence in high-risk patients suggests the outcomes are similar, although the collective evidence across all trials suggests the risk of death is higher with OPCABG (RR 1.18; p = 0.077). Possible reasons for this are fewer grafts, a greater need for subsequent revascularisation and worse patency. Despite recruiting more than 15,000 patients into trials of OPCABG versus ONCABG, the views of members of the surgical community are polarised. A qualitative evaluation of the reasons behind the views held by the advocates of the two techniques, and in particular what evidence would need to be presented in order to change individual practice, is an area for future research.

One possible explanation for the polarisation is the belief that ‘it’s in the surgeon’s hands’. If the surgeons are true ‘experts’ then one might anticipate no difference in outcomes between the two methods. Surgeons that use both techniques, albeit one perhaps slightly more frequently than the other, are likely to be less committed to OPCABG than surgeons who use OPCABG exclusively, and this may be reflected in the results. One way to test this hypothesis would be an individual patient data meta-analysis of the trial data, classifying patients according to the characteristics/experience of the surgeon.

Contribution of authors

• Dr Chris A Rogers designed the study with Professors DP Taggart, DG Altman, GD Angelini, A Gray and BC Reeves. She led the team at the Bristol Clinical Trials and Evaluation Unit, which led and managed the trial, and oversaw the analyses and their interpretation. She drafted the report with Miss K Pike.

• Miss Katie Pike advised on the design of the CRFs and the trial database. She wrote the statistical analysis plan and carried out the statistical analyses under the guidance of Dr Rogers. She drafted the report with Dr Rogers.

• Dr Helen Campbell contributed to the design of the resources use components of the CRFs and analysed some resource use data. She designed the five-level versions of the EQ-5D questionnaires and analysed the EQ-5D data.

• Professor Barnaby C Reeves designed the study with Professors DP Taggart, DG Altman, GD Angelini, A Gray and Dr Rogers. He helped to draft the discussion of the trial.

• Professor Gianni D Angelini designed the study with Professors DP Taggart, DG Altman, A Gray, BC Reeves and Dr Rogers. He actively promoted the trial amongst his clinical colleagues, particularly those based in European centres. He reviewed a draft of the report.

• Professor Alastair Gray designed the study with Professors DP Taggart, DG Altman, GD Angelini, and BC Reeves and Dr Rogers and led the health economic analyses. He reviewed a draft of the report.

• Professor Doug G Altman designed the study with Professors DP Taggart, GD Angelini, A Gray, BC Reeves and Dr Rogers. He advised on the use of expertise-based randomisation.

• Dr Helen Miller was a trial manager at Bristol Clinical Trials and Evaluation Unit. She liaised with sites about the protocol and queries, carried out visits to sites and facilitated the closure of the trial.

• Miss Sian Wells was a trial manager at Bristol Clinical Trials and Evaluation Unit. She designed the trial CRFs and database, liaised with sites about the protocol and queries, and carried out visits to sites.

• Professor David P Taggart designed the study with Professors DG Altman, GD Angelini, A Gray, BC Reeves and Dr Rogers. He was chief investigator and the principal applicant in the effort to secure funding. He actively promoted the trial amongst his clinical colleagues. He reviewed a draft of the report.

Disclaimers

This report presents independent research funded by the National Institute for Health Research (NIHR). The views and opinions expressed by authors in this publication are those of the authors and do not necessarily reflect those of the NHS, the NIHR, the MRC, NETSCC, the HTA programme, the EME programme or the Department of Health.

UK Centres

Non-UK Centres

Withdrawals

The European Quality of Life-5 Dimensions responses

Example of database validation

CRISP case report forms

Trial Steering Committee

Professor William Wijns (chairperson)

Dr Jonathan Cook

Professor John Dark

Mr Neville Jones (patient representative)

Dr Belinda Lees

Mr Patrick Magee

Professor John Pepper

Data Monitoring and Safety Committee

Professor Tom Treasure (chairperson)

Dr Tim Clayton

Professor Desmond Julian

Professor Paul Sergeant