Multi-centre randomised controlled trial examining the cost-effectiveness of contrast-enhanced high field magnetic resonance imaging in women scheduled for wide local excision (COMICE)

Article history

The research reported in this issue of the journal was commissioned by the HTA programme as project number 99/27/05. The contractual start date was in June 2001. The draft report began editorial review in October 2008 and was accepted for publication in June 2009. As the funder, by devising a commissioning brief, the HTA programme specified the research question and study design. 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 referees for their constructive comments on the draft document. However, they do not accept liability for damages or losses arising from material published in this report.

Declared competing interests of authors

None

Copyright statement

© 2010 Queen’s Printer and Controller of HMSO. This journal may be freely reproduced for the purposes of private research and study and 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: NETSCC, Health Technology Assessment, Alpha House, University of Southampton Science Park, Southampton SO16 7NS, UK.

2010 Queen’s Printer and Controller of HMSO

Overview

This report is divided into seven chapters. The first, the introduction, sets out the complexity of the area and outlines the need for this trial. The second chapter describes the methodology of the main trial and the sub-studies [health economic, quality of life (QoL) and non-schedule standardised interview (NSSI) (well-being)], and the third chapter presents the results of the main trial. The QoL substudy findings and the NSSI study results are presented in the fourth and fifth chapters respectively, and the health-economic analysis results are presented in the sixth chapter. The final chapter provides a discussion of the empirical findings and recommendations for future research.

Triple assessment of the breast

Patients with a screen detected abnormality of the breast, or who present symptomatically, will typically undergo triple assessment. Triple assessment involves clinical examination of the breasts; radiological assessment using X-ray mammography and ultrasound (USS); and pathological assessment either by fine needle aspiration cytology (FNAC) or core biopsy of the suspicious lesion. X-ray mammography relies on the detection of abnormal microcalcifications, focal asymmetric densities and the presence of architectural distortion, created by the variable absorption of X-rays by normal and abnormal tissues. However, it is not diagnostic if the typical characteristics of a malignant process are absent or if the lesion is an encapsulated fat-containing lesion defining a benign process.

Malignant lesions are more difficult to detect in the mammographically dense breast because of technical factors, including reduced image contrast, unsharpness and the similarity in density between cancer and normal fibroglandular elements. This issue may be reduced in the future with the use of full-field digital mammography, which is currently replacing conventional film screen mammography. Large international multicentre studies1 have demonstrated an equivalent or superior detection rate of breast cancer by digital mammography in comparison with conventional mammography, especially in dense breasts, premenopausal and perimenopausal women, and in women under 50 years of age.

Clinical and surgical management of early breast cancer

It is generally accepted that a patient’s best chance of a successful treatment outcome is the accurate identification of the cancer burden present. This means the identification of all tumour foci present and their location and extent. Failure to detect the additional tumour burden provided by multiple small foci may understage the disease present and deny the patient the opportunity of adjuvant therapies if the contribution of the smaller foci is ignored. However, if tumour extent can be delineated accurately, breast conservation, even for those with macroscopically multiple synchronous ipsilateral tumours, is an effective treatment. 2 Local recurrence after conservation surgery usually results from growth of residual cancer adjacent to the excised primary tumour or from multicentric disease. 3–6 Complete local excision, confirmed histologically, is essential to ensure that the risk of local recurrence is minimal. It is now recognised that for patients with multicentric disease detected prior to surgery, breast conservation surgery may still be appropriate if all clinically and radiologically apparent abnormalities are removed, clear margins of resection are achieved, and there is no extensive intraductal component present.

Whilst conventional triple assessment has a high sensitivity for the diagnosis of symptomatic breast cancer, it has limitations in defining the extent of disease present within the breast. For example, Van Goethem et al. 7 reported on the results of 67 preoperative breast examinations carried out to predict the extent of cancer in patients with dense breast tissue or to determine whether dense breast parenchyma would lead to false-positive or inconclusive examinations. The sensitivity values for detection of the index lesion were 83.0% for X-ray mammography (XRM) and 70.8% for USS, with XRM underestimating the extent of cancer present in 37% and USS in 40% of cases. The detection rates for multifocal or multicentric disease (in 20/67 patients) were 35.0% XRM and 30.0% USS, and the false-positive rates were 12.5% and 14.0%, respectively.

The selection of patients for wide local excision (WLE) is dependent on the clinical and imaging findings, namely the site and size of the tumour relative to the breast size. Important excluding factors include: lesions greater than 4 cm in diameter; multifocal or multicentric disease; an extensive in situ component; widespread lymphovascular invasion on biopsy; and centrally placed tumours in small breasts. The Milan II trial,8 which compared quadrantectomy versus WLE, both followed by radiotherapy, demonstrated that although cosmesis was improved in the WLE group, this was at the expense of a marked increase in loco-regional recurrence (18.6% versus 7.4% 10-year crude cumulative incidence) due to increased incidence of positive excision margins (16% versus 4% in the quadrantectomy group).

The aim of WLE is to remove the palpable lesion with a 1-cm margin of surrounding normal tissue. Using fingers as a guide, the surgeon makes an incision circumferentially, a finger’s breadth away from the palpable mass and continues this posteriorly through the breast tissue until the pectoral fascia is reached. If the lesion is not palpable, the mass is located by wire localisation and incision made parallel to the long axis of the wire, and extended to encompass the lesion within a cylinder of tissue. After excision the specimen is marked in three axes and X-rayed to demonstrate the location of the tumour with respect to surrounding excision margins. If the tumour lies at the edge of the specimen and appears incompletely excised, further shavings from the excision cavity are obtained and appropriately marked for pathological verification. It must be noted that at some institutions this is carried out routinely, regardless of whether or not the excision is complete.

The positivity of the margins and the width of the negative margin correlate with the likelihood of residual cancer, whether invasive or in situ, remaining within the breast post WLE. As radiotherapy reduces the ipsilateral breast tumour recurrence rate by a factor of four, resulting in a 75% local control rate, the concept of ‘close’ margins, with tumour extending to within 1–2 mm of the edge of the specimen, has gained increasing acceptance. 9–12 However, it is likely that the residual tumour burden decreases with increasing margin width, and, as a consequence, the margin width deemed acceptable varies between centres and will drive the need for further local treatments.

Reoperation rates

In 2006–7 some 17% of women in the UK with a screen detected pathologically proven tumour underwent more than one therapeutic operation. 13 This value ranged within the UK from 13% to 21%, but was similar for invasive and non-invasive cancers at 16% and 17%, respectively.

In the UK, 9% (range 6–15%) of patients with invasive cancer with a B5b (invasive malignancy) preoperative diagnosis, who were initially treated with a conservative operation, had a repeat conservative operation to clear positive margins. Patients with invasive cancers with a C5 (malignant) cytology-only preoperative diagnosis who were initially treated with conservative surgery, had a repeat operation rate of 12% (range 8–22%) to clear involved margins, and 16% of patients (range 11–24%) with non-invasive and micro-invasive cancer with a B5a (in situ malignancy) preoperative diagnosis had a repeat operation. Patients with invasive cancers with a B5a preoperative diagnosis, treated initially with conservative surgery, had a repeat operation rate of 31% (range 19–54%). Overall, 12% of all patients with breast cancer, who had a preoperative diagnosis, treated initially by conservative surgery, had repeat conservation surgery for positive margins.

Additionally, in the UK, 6% of patients with invasive cancer with a B5b preoperative diagnosis, 7% of patients with invasive cancers diagnosed by C5 cytology alone, 10% (range 3–15%) of patients with non-invasive cancer with a B5a diagnosis and 20% (range 11–30%) of patients with invasive cancer with a B5a diagnosis underwent mastectomy for positive margins following initial conservative surgery. Overall in the UK, 7% of all patients with a preoperative diagnosis, treated initially by conservation surgery, subsequently underwent mastectomy to achieve tumour clearance.

Thus in 2006/7, 19% of all patients with breast cancer, who had a preoperative diagnosis and who were initially treated by conservative surgery, had repeat therapeutic procedures (conservative surgery or mastectomy) to achieve clear margins.

The corresponding data from the Association of Breast Surgeons at the British Association for Surgical Oncology (ABS at BASO) for screen detected malignancies for 2001–2 showed that of the 5287 patients with invasive cancer with a preoperative B5b core biopsy, 624 (12%) underwent a repeat therapeutic operation. This varied from 8% to 17%. In the group of patients with invasive cancer diagnosed preoperatively by cytology alone, 15% (range 2–30%) underwent a repeat therapeutic operation. In the group of patients with invasive cancer with a preoperative B5a core biopsy, 41% (range 27–62%) underwent a repeat therapeutic operation.

In the UK as a whole, 14% of patients with invasive cancer and 20% of patients with non-invasive cancer underwent more than one therapeutic operation in 2001–2. For patients with invasive cancers, a repeat therapeutic operation was necessary for 12% of those with a B5b preoperative core biopsy sample, 15% of those with a preoperative diagnosis by fine needle cytology alone, and 41% of those with a B5a preoperative core biopsy.

Thus in 2001–2, at the time of initiation of the COMICE trial, overall a total of 14.2% of women underwent a repeat therapeutic operation for positive margins post WLE.

The figures quoted above relate to patients with a screen detected malignancy. The UK NHS Breast Screening Programme (NHS BSP) invites women between the ages of 50 and 65 years to attend for screening every 3 years, and screening is available for women over 65 years of age if they self-refer. Detection rates are not available for the self-referring population of patients who present with symptoms that are subsequently found to be due to malignancy. Studies suggest that there is a difference in characteristics between screen detected and symptomatic tumours. Screen detected cancers are significantly more frequently grade I, less than 10 mm in diameter and node negative, whereas symptomatic cancers are more frequently grade III, greater than 20 mm in diameter and exhibit lymphovascular invasion. Screen detected cancers favour breast-conserving surgery and are associated with a reduced requirement for adjuvant chemotherapy. 14

Requirement for the COMICE trial

This trial was developed in response to an open call from the NHS Health Technology Assessment programme for research and development in the area of magnetic resonance imaging (MRI) in patients with newly diagnosed breast cancer. One of the objectives of the NHS BSP is to reduce the reoperation rate for patients with screen detected primary breast cancers to below 10%, whilst achieving a good cosmetic result by minimising the volume of tissue removed. As a consequence of the high reoperation rate, the known problems associated with X-ray mammography and USS, and with reference to the available literature on the results of MR breast imaging, the COMICE trial proposed to determine the comparative effectiveness of the addition of MRI to conventional triple assessment to reduce reoperation rates in women with primary breast cancer treated by WLE. The trial hypothesis was that inclusion of three-dimensional MRI data with conventional triple assessment would aid the localisation of tumour within the breast and enable the surgeon to achieve a higher complete tumour excision rate.

Magnetic resonance imaging: review of literature

Magnetic resonance imaging provides high-resolution soft tissue detail in any plane desired and produces both morphological and functional information. There have now been a number of studies examining the role of MRI of the breast in preoperative and problematic cases, which have shown a good correlation between histological and MR measurement of invasive tumour size (r = 0.93) compared with mammographic measurement of tumour size (r = 0.59). In 1993, Harms et al. 15 used a RODEO (Rotating Delivery of Excitation Off-resonance) technique to demonstrate a good correlation between MRI findings and histopathology of lesion margins in patients who have undergone lumpectomy. This work was confirmed 3 years later by Davies et al. 16, who used a three-dimensional fast spoiled gradient echo (FSPGR), contrast-enhanced, fat-suppressed sequence and demonstrated an excellent correlation (r = 0.98; standard error = 0.34) between the maximum cancer diameter measured by MR and histopathology. This was particularly evident for the largest cancer diameters. This compared with poorer correlation coefficients and larger standard errors for mammography and USS at 0.46 and 0.45, and 1.04 and 0.78, respectively. Similar data was presented by Ando and colleagues,17 who demonstrated a good correlation between direct invasion of mammary tissue, satellite nodule formation and intraductal tumour extension with histopathology. In the study by Van Goethem,7 mentioned previously, the comparative sensitivity of MRI was 98% and tumour size was only underestimated by 12.5%. The detection rate for multicentric disease was 100%, although the false-positive rate was elevated, with respect to XRM and USS, at 23%. The authors concluded that MRI was more accurate than the other modalities in assessing tumour extent and multifocality in patients with dense breasts, but cautioned that coexisting benign disease could lead to false-positive examinations.

A number of studies have now reported that MRI is more accurate than X-ray mammography in depicting multicentric and multifocal disease, intraductal extension associated with invasive cancer and tumour infiltration of the nipple retro-areolar complex. In a comparative study of mammography, USS and MRI, Boetes et al. 18 reported underestimation of tumour size by 14% and 18%, respectively, for mammography and USS, while MRI showed no significant difference in size compared with that found at pathological evaluation. MRI also detected all additional tumour foci found at subsequent histopathology compared with detection rates of 31% and 38% by mammography and USS, respectively. Hata et al. 19 also examined the ability of dynamic contrast-enhanced (DCE)-MRI to detect intraductal spread of tumour in comparison with USS and mammography. The sensitivity, specificity and accuracy of detection for intraductal spread by DCE-MRI were 66.7%, 64.2% and 65.6%, respectively. Corresponding results for mammography were 22.2%, 85.7% and 50.0%, and 20.6%, 85.2% and 50.0% for USS, suggesting that DCE-MRI offers a benefit over other imaging modalities for loco-regional staging.

Magnetic resonance imaging: techniques and protocols

Magnetic resonance imaging allows the acquisition of high-resolution anatomical and morphological information, as well as functional information. Functional information can be acquired using DCE-MRI, which refers to the acquisition of data before, during and after intravenous contrast agent administration. As the contrast agent enters the tissue under investigation, the T1 and T2 relaxation times of tissue water decrease over time to an extent mostly determined by the concentration of contrast agent present. This technique is employed to examine neoangiogenically induced vascular changes, which result in the proliferation of abnormally leaky microvessels. DCE-MRI methodology is based on the rapid diffusion of a small molecular weight contrast agent through the fenestration present in these abnormal microvessels. Comparative studies have demonstrated that the signal intensity changes relate to the vascular density within the lesion, and that the rate of enhancement is determined by the vascular fenestrations and functional permeability20–22 and by the interstitial environment, which influences the diffusibility and temporal retention of the contrast agent. By examining the signal intensity time curves, physiological parameters that relate to tissue perfusion, microvascular vessel wall permeability and the extravascular–extracellular volume fraction can be extracted, which may aid characterisation of the underlying pathology.

The ideal DCE-MRI sequence would encompass the following parameters: excellent temporal resolution (< 30 seconds) to optimally define the signal intensity changes over time (the signal–intensity time curve); a volumetric as opposed to a two-dimensional acquisition, allowing the use of thinner slices with no interslice gap to minimise partial volume averaging and inflow effects; isotropic spatial resolution (< 1 mm in plane); excellent uniform fat/water suppression throughout the volume of interest; high sensitivity to the contrast agent with a good dynamic range; and the capability to image both breasts in their entirety in one pass. Such stringent technical requirements are not currently feasible at 1.5 tesla (T) and, as a consequence, compromises must be made either to the temporal or spatial resolution employed or to extent of the breast coverage obtained.

Currently, two approaches have been used to examine contrast uptake characteristics of breast tissue at 1.5 T. Two-dimensional dynamic imaging allows rapid data acquisition at a limited number of slice locations, and hence good delineation of the signal–intensity time curves, and is suitable for investigation of equivocal lesions on mammography/USS. Alternatively, three-dimensional imaging provides the complete coverage of both breasts required for screening, but with the penalty of decreased temporal resolution and hence poorer delineation of the signal–intensity time curves. Contrast uptake data must be viewed together with morphological information, which may provide additional insight into the nature of the abnormality. For example, in the presence of rapid contrast uptake the presence of spiculation of a mass and rim enhancement is highly suggestive of malignancy, whereas a lobulated lesion with internal septations is suggestive of a fibroadenoma. The American College of Radiologists Breast Imaging Reporting and Data System (ACR BI-RADS)-MRI Lexicon23 advocates the use of both lesion architecture and enhancement characteristics, and provides a simple descriptor for reporting of MRI findings.

The trade-off between spatial and temporal resolution in DCE-MRI has been investigated by Kuhl. 24 She examined 30 patients with 54 enhancing lesions (26 malignant, 28 benign) at 1.5 T on two separate occasions. A standard dynamic protocol was employed, using a matrix size of 256 × 256 and a 69-second acquisition time, followed on a separate day by a modified dynamic protocol using a matrix of 400 × 512 and 116-second acquisition time. Significant difference between benign and malignant lesions, determined using a generalised linear model, were lost using the modified dynamic protocol, although kinetic information from the signal–intensity time curves was preserved and delineation of lesion margins and internal architecture was superior. Receiver operator characteristic curve analysis demonstrated a significantly larger area under the curve (0.945 versus 0.877, p < 0.05) for results obtained using the modified dynamic protocol. They concurred that increased spatial resolution increased diagnostic confidence and accuracy and that because of the overlap in contrast kinetics between benign and malignant lesions, the loss of temporal resolution was of no consequence in characterisation of primary lesions for the individual patient.

The diagnostic performance of dynamic contrast-enhanced parameters and morphological features has been further investigated by Goto et al. 25 High temporal resolution dynamic three-dimensional gradient echo (6.8 seconds) and high spatial resolution T1-weighted FSPGR sequences (in-plane resolution of 0.68 × 0.68 × 1.0mm) were carried out on 190 patients with a positive diagnosis of malignancy on mammography, USS or both, with a total of 204 lesions (144 malignant) and compared for diagnostic performance. The sensitivity and specificity of the morphological criteria were significantly greater than the enhancement criteria (p = 0.0012 and p = 0.0003, respectively). Statistically significant differences in morphological criteria were also reported between benign and malignant lesions. They suggested that signal intensity–time curves may not be required to diagnose malignant breast lesions. However, despite excellent temporal resolution, kinetic analysis was only performed subjectively, with signal–intensity peak by the 18th of 28 frames being defined as positive for malignancy and continuous increase in signal intensity through all 28 frames as negative. Additionally, slice thickness for the DCE-MRI examination ranged from 3 to 6 mm, depending on breast size, with a matrix of 196 × 256 and a field of view of 35 cm, limiting the diagnostic potential of the DCE-MRI examination.

In some reports the combination of functional and morphological data has given added diagnostic value. The importance of morphological information was studied in a report from Gibbs et al. ,26 who examined the role of MRI in differentiating less than 1 cm diameter benign from malignant lesions, using a high temporal resolution dynamic two-dimensional FSPGR technique (11 seconds) and high spatial resolution post-contrast T1-weighted imaging. Radiological assessment of the post-contrast data provided a diagnostic accuracy rate of 69%, compared with the exchange rate constant calculated from the DCE-MRI data, which revealed a diagnostic accuracy rate of 74%. However, when the information was combined in a logistic regression model, a diagnostic accuracy of 92% was obtained. This would suggest that the morphological features of small lesions are not adequate in isolation for good diagnostic accuracy.

Ultimately, the characteristics of larger mass lesions compared with small localised abnormalities or areas of diffuse regional enhancement are different and consequently the composition of the patient cohort has a huge impact on the diagnostic accuracy of each study reported. Summarising the available information, the protocols used for loco-regional staging of known or suspected malignancy and those for characterisation of equivocal lesions and screening have different requirements. The latter clinical scenario is potentially diagnostically more challenging, and ‘high-quality’ functional and morphological information is required to achieve clinically useful accuracy rates.

Role of magnetic resonance imaging for screening

There are now a number of reports advocating the use of MR breast imaging in screening women with BRCA1, BRCA2 or TP53 gene mutation, or those with a high risk of developing breast cancer from the family history. In general, this patient group is most likely to present with small localised abnormalities or areas of diffuse regional enhancement. The UK multicentre study (MARIBS) of such a patient group, conducted over a 7-year period, utilised a three-dimensional volume acquisition repeated at 90-second intervals to generate functional data and ensure whole breast imaging, followed by a high spatial resolution fat-suppressed sequence for morphology. 27 This protocol resulted in an overall sensitivity and specificity of 77% [confidence interval (CI) 60% to 90%] and 81% (CI 80% to 83%), respectively, for MRI, compared with 40% (CI 24% to 58%) and 93% (CI 92% to95%) for mammography, representing a highly significant improvement in detection by MRI (p < 0.01), but significantly poorer specificity (p <0.001). The improved detection rate of MRI compared with mammography was even more apparent for patients with the BRCA1 mutation or first-degree relatives of patients with BRCA1 mutation, in whom the sensitivity was 92% (CI 64% to 100%) compared with 23% (CI 5% to 54%), p < 0.004, for mammography.

In 2005, Kuhl,28 using a similar three-dimensional protocol, reported on a surveillance cohort study of 529 asymptomatic BRCA patients (proven or suspected from their family history) with a lifetime risk of breast cancer of greater than 20%. In total, 1542 surveillance rounds were performed, with a mean follow-up period of 5.3 years. They detected 43 breast cancers [34 invasive, nine ductal carcinoma in situ (DCIS)], resulting in sensitivity and specificity values of 91% and 97.2%, respectively, for MRI, compared with 33.0% and 96.8% for mammography, and 40.0% and 90.5% for USS. Trecate et al. ,29 in a smaller study of 116 patients screened annually using mammography, USS and MRI over a 5-year period, detected 12 cancers of which six were only detected by MRI. In this study there were three false-positive results for MRI but no false-positive results for other techniques.

Role of magnetic resonance imaging for DCIS

Dynamic contrast-enhanced magnetic resonance imaging has also been employed to investigate suspicious microcalcifications on mammography, present either in isolation or associated with a breast mass. In a study of 88 patients, DCE-MRI was used to investigate women recalled following screening mammography for further evaluation of microcalcifications. 30 Dynamic contrast-enhanced imaging data acquired using a high temporal resolution (11 seconds) FSPGR sequence, was analysed using a two-compartment modelling technique, resulting in sensitivity, specificity, positive predictive value, negative predictive value and accuracy values of 80.0%, 82.4%, 57.1%, 93.3% and 81.8%, respectively. These compared with corresponding values of 75.0%, 89.7%, 68.2%, 92.4% and 86.4% if the data were examined by a radiologist using empirical data and lesion morphology together, indicating the benefit of functional information in this patient group.

Bazzocchi et al. 31 used a three-dimensional gradient echo dynamic sequence acquired coronally at a temporal resolution of approximately 60 seconds, to investigate 112 patients with mammographically detected microcalcifications with BI-RADS pattern 4 or 5. All subsequently underwent surgical resection and the findings were compared. Analysis of microcalcifications, either alone or in association with a mass, resulted in sensitivity values of 80% and 97%; specificity values of 79% and 33%; positive predictive values of 86% and 82%; negative predictive values of 71% and 75%; and accuracy values of 80% and 82%, respectively. In a small study of only 14 patients with pure DCIS, Mariano et al. used an intensity-modulated parametric mapping technique with the data categorised according to morphological and kinetic criteria from the ACR BI-RADS-MRI Lexicon. 23 Using morphological criteria, 71% of cases were correctly classified, with regional enhancement pattern being most prominent, whereas parametric mapping classified 86% of cases, correctly identifying all intermediate and high-grade DCIS cases.

Diagnostic accuracy of magnetic resonance imaging

The studies detailed above have shown a strong correlation between post-contrast, fat-suppressed images and histopathology in patients with mammographically detected, biopsy-proven malignancy. However, with the detection of increasingly small lesions, the specificity of MRI becomes crucial to patient management. For example, Kramer et al. ,32 using a multiple three-dimensional acquisition technique at 90-second intervals, reported a sensitivity of 89% for detection of malignancy, but 17% of women had an incorrect diagnosis of multicentric disease. Similarly, Balen et al. 33 commented on inappropriate mastectomy in up to 28% of patients. This study utilised three-dimensional imaging of the breast at between 60- and 80-second intervals following bolus contrast agent injection. Studies such as these rely on detection of contrast uptake by image subtraction and empirical techniques and it is possible that the reduced temporal resolution, and therefore poor delineation of the contrast uptake curve, may have contributed to the false-positive results. This problem is particularly true of screening trials when spatial and temporal resolution is sacrificed for whole breast coverage. Indeed, the screening trials of women with gene mutations, or those with a high risk of developing breast cancer from the family history,27 have recommended that second-look USS and/or biopsy confirmation of MR findings is obtained before clinical management is changed from WLE to mastectomy.

Role of magnetic resonance imaging in therapeutic management

Dynamic contrast-enhanced MRI, by detecting the neovascularisation induced by malignant lesions, has already been used to determine the therapeutic approach. Tan et al. 34 examined 83 patients who were scheduled for breast conservation therapy and found management to be definitively altered in 18%, with 13% of women undergoing additional surgery. Fischer et al. 35 investigated 463 women with 548 cancers, and reported a change in management in 14.3% of women, due to detection of more extensive or multicentric disease. Neither study detected factors that were predictive of alteration in outcome from the patient or tumour characteristics, mammographic results or the timing of MRI.

The MRI protocol developed for COMICE is pragmatic and was based on the available technology in the UK in 2001. As a consequence, the majority of examinations were performed at 1.5 T, using a protocol which provides a temporal resolution of 45 seconds and an in-plane spatial resolution of 1.2 mm × 1.1 mm and a through plane resolution of 4mm. The functional data was complimented by a high-resolution post-contrast T1-weighted sequence (0.4 mm × 0.4 mm in plane; 2.5 mm through plane), acquired either with fat-suppression or contrast enhancement later assessed by image subtraction to provide morphological information.

Of particular relevance to this study is the transfer of imaging information, concerning tumour size and location, to the surgeon. It must be remembered that X-ray mammograms are obtained with the patient erect and the breast compressed, USS is performed with the patient supine and the breast variably compressed by the high-frequency probe, whilst for MRI the patient is scanned prone, with the breasts dependent and the arms outstretched beside the head. Surgery is performed with the patient supine and the arm on the affected side abducted to approximately 90 degrees for access. Thus the patient is variably positioned for all imaging techniques and indeed for surgery, and reliance is placed on the surgeon utilising the images obtained to aid excision. Unlike other surgical specialties, stereotactic localisation of the tumour is not carried out.

Influence of magnetic resonance imaging on quality of life

Unfavourable side effects on health-related quality of life (HRQoL) (or health status) may arise from the process of screening itself, such as pain, discomfort and feelings of anxiety and distress. Several studies have shown that recall because of a false-positive mammogram causes adverse emotional, physical and social effects. 36–39 It is already established that breast MRI can cause significant anxiety before, during and for some weeks after the scan. Some studies have shown that the distress caused by MRI is comparable to that caused by elective surgery, and others have found that MRI could not be completed because of anxiety in up to 5% of patients. 40 In a study of 616 women undergoing annual breast MRI because of high genetic risk, MRI-related distress was shown to be greater following breast MRI than X-ray mammography, and persisted in some women for at least 6 weeks after the scan. 41 Anxiety may be related to multiple factors, including aspects of the procedure (e.g. confinement and noise) as well as context (e.g. fear that cancer may be discovered). 40,42–45

The COMICE trial sought to elicit if the addition of MRI to triple assessment alone had any impact, negative or positive, on generic HRQoL and distress among women undergoing treatment for breast cancer.

In the main QoL study, two standardised questionnaires, the Hospital Anxiety and Depression Scale (HADS)46 and the Functional Assessment of Cancer Therapy (breast cancer version) (FACT-B)47,48 were used to evaluate important generic parameters of HRQoL. The FACT-B questionnaire is a 44-item self-report instrument, designed to measure multidimensional QoL in patients undergoing therapy for breast cancer and has been used extensively in oncology clinical trials, but not specifically in those patients with newly diagnosed breast cancer. Thus to supplement the information obtained from these standardised measures, an NSSI was developed and validated to assess the self-reported psychosocial effects of specific aspects of participation in the COMICE trial, for example reaction to randomisation and the extent to which the various investigations caused distress.

Describing NSSI methodology, Brown and Rutter49 state: ‘In contrast to most research interviews, the wording and ordering of questions are not rigidly laid down in advance. The idea is rejected that standardisation can be achieved by the use of identically worded questions in the same sequence. Some questions may be given, but the interviewer relies much more on a list of information required. It is his job to inquire into each area … until he is satisfied he has obtained the material. In a certain sense, the schedule may be said to be a questionnaire addressed to the interviewer and not the informant.’ The NSSI, therefore, is a flexible quantitative interview method, whereby the interviewer can be flexible about the order and exact wording of questions, as well as in the use supplementary questions for clarification.

Members of the Trial Management Group have previously developed NSSIs for use in two studies. The first evaluated the lifetime care pathways of preschool children with special needs who had been referred to a multidisciplinary assessment centre. 50 The views of referrers, recommenders and parents were sought. The second study used an NSSI to obtain the views of parents about various aspects of their children’s behaviour and family relationships. 50

Health economic evaluation

The addition of MRI clearly results in a larger upfront use of health-care resources. However, it is unclear if its addition to conventional triple assessment will result in benefits in terms of better patient outcomes and lower NHS resource use in the future. The key issues with regard to cost-effectiveness from an NHS perspective include: the relative accuracy rates for depicting tumour margins; the uncertainty surrounding the identification of multicentric disease preoperatively; determination of the risk factors for referral for MRI; the impact of MRI on clinical management and patient’s QoL; and the medium-term ipsilateral breast tumour recurrence rate.

A review of the literature found no previous cost-effectiveness analyses that have addressed the question of whether the addition of MRI to the routine techniques is worthwhile. The aim of the economic analysis was to compare the costs and consequences (in terms of HRQoL) of the two alternative imaging strategies considered in the COMICE trial. The economic evaluation was conducted from the perspective of the NHS. The costs considered were those relating to NHS and Personal Social Services resource use, while patient outcomes were intended to be measured in HRQoL using the EuroQol 5 Dimensions (EQ-5D) questionnaire, a standardised instrument for measurement of health outcome.

The COMICE randomised controlled trial sought to determine the potential benefits of the addition of MRI to the routine techniques employed for loco-regional staging of primary breast cancer. The cost-effectiveness of MRI in this clinical setting is unknown and the economic analysis of this trial intended to answer whether the addition of MRI is worthwhile from the perspective of the NHS.

Summary

• In 2001–2, at the time of initiation of the COMICE trial, overall a total of 14.2% of women underwent a repeat therapeutic operation for positive margins post WLE. This exceeds the NHS BSP target of 10%.

• Conventional triple assessment has a high sensitivity for the diagnosis of symptomatic breast cancer, but it has limitations in defining the extent of disease present within the breast.

• Magnetic resonance imaging of the breast in preoperative and problematic cases, has shown a good correlation between histological and MR measurement of invasive tumour size.

• The trial hypothesis was that inclusion of three-dimensional MRI data with conventional triple assessment would aid the localisation of tumour within the breast and enable the surgeon to achieve a higher complete tumour excision rate.

• Health economic assessment addressed the question of whether a larger upfront use of health-care resources from the addition of MRI to routine techniques is worthwhile. The trial also sought to elicit the impact on generic HRQoL and the level of distress experienced.

Objectives

The overall aim of this randomised controlled trial was to determine the potential benefits to the patient and to the NHS of the addition of MRI to the routine techniques employed for loco-regional staging of primary breast cancer.

The primary objective of the study was to evaluate the role of MRI with respect to the repeat operation (conservative surgery) or mastectomy rates following primary excision between those planned by conventional triple assessment, and those planned by a combination of triple assessment and MRI. This included the rates of pathologically avoidable mastectomy at initial operation. An economic evaluation of the cost-effectiveness from an NHS perspective between the two arms also formed part of the primary objective.

The secondary objectives of the study included:

1. An investigation of the factors associated with differences in imaging findings and histopathology, which may influence referral for MRI.

2. Evaluation and comparison of the accuracy of loco-regional staging by X-ray mammography, USS and MRI, with reference to the tumour extent determined by histopathology of the resected specimens.

3. Observation of the percentage of patients in whom a change in clinical management was proposed after MRI.

4. Comparison of subsequent chemotherapy/radiotherapy/additional adjuvant therapy interventions between patients receiving MRI and those receiving no MRI.

5. Follow-up of MRI-only-detected lesions that were < 5 mm in diameter at diagnosis, or ≥ 5 mm in diameter, but which were negative on biopsy.

6. Determination of the ipsilateral breast tumour recurrence rate for both groups.

7. An assessment of QoL and patient satisfaction, with management decisions based on either triple assessment or triple assessment combined with MRI.

Trial design

COMICE was a multicentre, randomised, controlled, open, fixed-sample, parallel group trial with equal randomisation, in women with biopsy-proven primary breast cancer, who were scheduled for WLE following triple assessment [clinical, radiological (X-ray mammography and breast USS) and pathological (FNAC/core biopsy)]. Patients were randomised to receive MRI or no MRI. A pragmatic approach to trial design was chosen so that results could be generalisable in clinical practice and to reduce unnecessary protocol-driven trial costs.

The main trial design was also supplemented with a qualitative study, the Well-Being study, involving a sample of 100 patients, in order to assess patients’ subjective and objective experiences of the treatment process and the care pathway. This supplemental study included the development and validation of an NSSI to assess the self-reported psychosocial effects of specific aspects of trial participation. Although the trial was referred to as the Well-Being study at sites, and was introduced to patients as such, from here on in, the report will refer to the Well-Being study as the NSSI study as this better represents the nature of the research.

Eligibility

End points

Trial conduct

Informed consent and randomisation

Invitation to participate in the COMICE trial was made at the time at which treatment options were discussed and agreed with the patient. Whilst at the outpatient clinic, women scheduled for WLE were invited to participate in the study by the consultant breast surgeon or the consultant radiologist, and were subsequently given further information, including the patient information sheet (Appendix 6), by the research nurse. Wherever possible, patients were given at least 24 hours to consider participation in the trial. If the patient wished to participate in the trial, the research nurse arranged an appointment to obtain written consent (see Appendix 7 for a copy of the informed consent form). Once eligibility and written informed consent had been confirmed, the research nurse randomised the patient using the CTRU central automated 24-hour randomisation system, to receive either MRI or no MRI. Authorisation codes provided by the CTRU were required to access the service. Since randomisation was performed via the independent, central CTRU system, allocation concealment was maintained.

Randomisation was performed using minimisation incorporating a random element (dynamic allocation using a pre specified computer generated algorithm incorporating an element of randomness). The integrity of the randomisation system was tested on a regular basis. To ensure balanced treatment groups with respect to prognostic factors, the following minimisation factors were incorporated, as recorded at the time of randomisation:

• consultant breast surgeon

• patient’s age (< 50 years versus ≥ 50 years)

• breast density (ACR BI-RADS group 1 (type 1 only) versus ACR BI-RADS group 2 (type 2, 3 or 4).

Homogeneously or heterogeneously structured dense fibroglandular tissue in a large percentage of the entire breast volume is the only mammographic or USS finding to date that has helped define a subgroup of patients with multifocal or multicentric disease detected by MRI alone. 35,52 The definition of breast density according to the mammographic pattern followed the criteria stated in the ACR BI-RADS, as follows: type 1 – almost entirely fatty breast tissue; type 2 – scattered fibroglandular tissue; type 3 – heterogeneously dense breast tissue; and type 4 – extremely dense breast tissue. 23

Blinding

COMICE was an open trial. Since patients either received MRI or no MRI, the nature of the trial prevented masking the randomised intervention.

Assessments/interventions

For details of the stages in the trial process, see the trial flow diagram in Figure 1.

FIGURE 1.

Trial flow diagram.

Data collection

Clinical and resource use data generated by all centres was collected on study CRFs, which were monitored and computerised by the CTRU. Details can be found in Appendix 8.

Using detailed case report forms, information on health-care resource utilisation of the patients in both trial arms were collected at randomisation and during follow-up. These have been supplemented with clinical expert opinion and other additional data where appropriate.

Quality of life

At randomisation, patients were asked if they were willing to take part in the QoL study, in order to evaluate the impact of the investigations and treatment. QoL participation was not compulsory, in order to avoid jeopardising recruitment in to the main MRI study. The FACT-B was used to evaluate the impact on physical, social, emotional and functional well-being, and breast cancer concerns. The FACT-B comprises the FACT-General48 and 10 specific items related to breast cancer. 47 Anxiety and depression were assessed using the HADS. 46 The EQ-5D was also used to provide a description of patients’ HRQoL and HRQoL weights, based on the preferences of a sample of the UK population. 61–63 The EQ-5D is a standardised non-disease-specific instrument that describes and values HRQoL, and provides a single index value for a number of different health states. It is applicable to a wide range of health conditions and treatments, and provides a simple descriptive profile and a single index value for a patient’s health status. Its descriptive system consists of five dimensions (mobility, self-care, usual activity, pain/discomfort and anxiety/depression), with each dimension having three different levels (no problem, some problem or extreme problem). All questionnaires were administered together, with the EQ-5D appearing first, the HADS second and the FACT-B third.

Early in the course of the COMICE trial, question GE3 ‘I am losing hope in the fight against my illness’ on the FACT-B was removed from the trial questionnaires, as there was some evidence that it had caused distress to a few patients in the trial and was not particularly relevant to this recently diagnosed patient population. This does not affect the scalar structure of the questionnaire, and pro-rating was used to compensate for the removal of this item.

Patients were asked to complete QoL questionnaires at the following times: baseline, 8 weeks post randomisation, and at 6 and 12 months post initial surgery. At the start of recruitment in December 2001, and until February 2004, post-randomisation questionnaires were administered at 4 weeks post initial surgery and further questionnaires were administered at 4 weeks post repeat operation (if appropriate). However, it was difficult for the CTRU to obtain initial surgery dates in time to send questionnaires, and some patients underwent repeat operation before their 4 weeks-post-initial-surgery questionnaire was due. As a consequence, the timing of questionnaires was changed, to be administered at 8 weeks post randomisation for all patients thereafter to encapsulate the time period of the initial surgery and any subsequent surgery performed. Post-surgery time points were selected as it was felt necessary to standardise the assessments around the time of surgery, and avoid the possibility of assessments coinciding with actual time of surgery.

Baseline QoL questionnaires were completed by patients in clinic, after written informed consent had been given, and prior to randomisation (or knowledge of randomisation outcome). QoL was then assessed at 8 weeks post randomisation, and at 6 and 12 months post initial surgery by sending questionnaires to the patient’s home address (by the CTRU), after the patient’s current health status had been checked with the relevant research nurse. Patients who did not respond within 2 weeks of the initial questionnaire being sent were sent a reminder letter; however, if two consecutive sets of questionnaires were not returned, no further questionnaires were sent. A letter of thanks was also sent to patients who returned a set of completed questionnaires.

Data quality and monitoring

Data management and monitoring were conducted according to the MRC Guidelines for Good Clinical Practice in Clinical Trials64 and CTRU Standard Operating Procedures. Data management practice included verification, database validation and formal data checking following data entry. All missing and ambiguous data were chased until resolved or confirmed as unavailable.

Statistical methods

Non-schedule standardised interview substudy

Quality assurance

A total of 45 centres participated in the trial, using 41 MR scanners. This large network of centres was necessary to achieve trial completion in a timely manner. All centres were NHS hospitals with functioning breast cancer multidisciplinary teams. Participating radiologists were members of those teams and routinely used imaging for breast care. A separate quality assurance process was undertaken to ensure that MR scans were being completed in accordance with the technical requirements of the trial protocol, and that scan interpretation was reasonable and consistent across the network.

The re-reading process was undertaken by an experienced breast radiologist, external to the trial, who was vetted and approved by the DMEC. This radiologist, blinded to original findings, advised on the compliance of the scans to the technical protocol, and then re-reported them using the trial forms. Trial staff compared these re-reports to the original radiologist’s reports.

Where variations, such as technical failure, non-identification of lesions or identification of additional lesions were detected, these were referred to the chief investigator for a third reading of the scan. In cases where the re-reported scan and the report from the chief investigator concurred, and were at variance to the original radiologist report, then that scan was considered to have been misreported. If technical failures were confirmed, or scans were considered to have been misreported, then up to five additional scans, if available, were re-read.

The process of scan selection for the quality assurance process was divided into two components. These were:

• Initial assessment The quality assurance process involved using a questionnaire to assess radiologist experience. ‘Less experienced’ radiologists were defined as those who had read fewer than 50 MR scans, or had been reading breast MR scans for less than 2 years. The initial two MR scans from ‘experienced’ radiologists, and the initial four, from those classified as ‘less experienced’, were re-read for quality assurance.

• Ongoing assessment After this initial assessment, a random sample of scans from each centre was re-read to ensure ongoing consistency. Participating centres were defined as either ‘large’, recruiting at least 12 patients per year, requiring one in every 10 scans to be re-reported, or ‘small’, recruiting fewer than 12 patients per year, and requiring one in every five scans to be re-reported.

Summary of changes to the protocol

The following protocol amendments were submitted and approved by the HTA: clarification of the definition of multifocal/multicentric lesions; updated information to detail that indeterminate tumours and highly suspicious lesions should be treated as per local protocol; modification of the patient information sheet; removal of question GE3 ‘I am losing hope in the fight against my illness’ from the FACT-B questionnaire; reduction of the quantity of health economic information recorded on the initial surgery CRF; and amendment to the QoL questionnaire schedule.

The following amendments to end point definitions and to the follow-up schedule were also agreed by the TSC and incorporated in the statistical analysis plan, prior to any analysis being performed: the definition of the primary end point has been updated to incorporate pathologically avoidable mastectomies at initial surgery; the secondary end point ‘risk factors for referral for MRI’ has been clarified to be ‘factors associated with differences in imaging findings’; the follow-up schedule has been amended due to the extension to recruitment, to follow up all patients for at least 1 year post randomisation, and, consequently, ipsilateral breast tumour recurrence has been summarised at 1 and 3 years post randomisation. The exclusion criteria were also amended to exclude patients scheduled to receive chemotherapy to any site prior to their breast surgery, and the timing of repeat MR scans was updated to coincide with 12 months post radiotherapy.

Participant flow

Trial conduct

A CONSORT (Consolidated Standards of Reporting Trials)73 flow diagram of trial progress is presented in Figure 2.

FIGURE 2.

CONSORT diagram. Please note that data regarding numbers of patients assessed for eligibility were not complete for all centres.

No patients withdrew their consent to use data already collected during the trial; however, one patient in the MRI group withdrew from trial follow-up before undergoing their surgery therefore this patient was classed as being lost to follow-up. In total there were 10 patients who were lost to follow-up at the time of analysis (four patients in the MRI group, six patients in the no-MRI group). In the MRI arm, one patient withdrew from follow-up as detailed above, and three patients had missing data at the time of analysis. In the no-MRI arm, two patients moved away and could not be followed up, and four patients had missing data at the time of analysis. Analysis of the primary end point and shorter-term end points was conducted once all patients had been followed up for at least 6 months. Analysis of longer-term end points and health economic analysis was conducted once all patients had been followed up for at least 12 months.

Baseline data

Baseline characteristics, including minimisation details, patient characteristics and clinical details of the ITT population are displayed in Tables 2–5. Corresponding details for the per-protocol population are displayed in Appendix 14. The two groups were well balanced with respect to baseline characteristics, and were very similar between the ITT and per-protocol populations.

The minimisation factors consultant surgeon, age and breast density were well balanced across the two arms for the ITT populations, with the majority of patients aged 50 or over (77.0%), and with breast density 2, 3 or 4 (87.2%). Patients were recruited to the COMICE trial by 107 consultant surgeons, with 60 (56.1%) recruiting fewer than 10 patients, and 47 (43.9%) recruiting 10 or more patients. The number of patients randomised to each of the interventions was well-balanced according to the number of patients each consultant surgeon had recruited.

The majority of patients in the ITT population were randomised between 2004 and 2006. The median age of patients at randomisation was 57 (range 27–86). At the time of randomisation, 31.9% of patients were employed full time, 23.1% were employed part time, and 32.6% were retired. All other patients were either unable to work due to illness/disability, were unemployed, or were students. Employment status was missing for 12 patients (0.7%). At the time of randomisation, 1139 patients (70.2%) were post-menopausal, with 60.8% of patients currently or previously taking the contraceptive pill/slow release injection. Of those patients taking the contraceptive pill/slow release injection at the time of randomisation, the median time patients had been taking it was 14 years (range 1–32) (MRI: 13 years, range 1–30; no MRI: 15 years, range 1–32). For those patients previously taking the contraceptive pill/slow release injection, the median time that patients had been taking it was 6 years (range < 1–35) (MRI: 6 years, range 1–30; no MRI: 5 years, range < 1–35). Five hundred and seventy-two patients (35.2%) were either currently using HRT (6.7%) or had previously used HRT (28.5%) at the time of randomisation. Of those patients using HRT at the time of randomisation, the median time that patients had been using it was 8 years (range < 1–32) (MRI: 8 years, range < 1–23; no MRI: 7 years, range 1–32). For those patients previously using HRT the median time that patients had been taking it was 6 years, range < 1–30 (MRI: 7 years, range < 1–27; no MRI: 6 years, range 1–30).

Cancer was identified through screening for 847 patients (52.2%) in the ITT population, and the method of confirming primary breast cancer was fine needle aspiration for 146 patients (9.0%), core biopsy for 1260 patients (77.6%) and fine needle aspiration and core biopsy for 204 patients (12.6%). One patient had a confirmatory histological sample taken after randomisation. Only 17 patients (1%) received preoperative neoadjuvant therapy, and, of these, 10 (58.8%) received tamoxifen, three (17.6%) were presribed anastrozole, and four (23.5%) other therapy, namely letrozole (three patients) and FEC (one patient, who received a combination of docetaxol, epirubicin and cyclophosphamide). An additional patient underwent eight cycles of chemotherapy prior to her surgery, but this was not documented at the time of initial assessment prior to randomisation.

Details of the mammography and USS findings for the ITT population are given in Appendix 13, as well as the MRI findings for the ITT population. The corresponding summaries of baseline data and MRI findings for the per-protocol population can be found in Appendix 14. Mammography and USS findings were similar between the two arms.

Detailed surgery characteristics are displayed in Appendix 13. The randomising consultant surgeon was present at initial surgery for 1077 patients (66.4%). In total, 1537 patients (94.7%) underwent a WLE at initial surgery, 68 (4.2%) underwent a mastectomy, one (0.1%) underwent a quadrantectomy and mini flap, and two patients underwent other surgery (reduction mammoplasty and segmentectomy). Median time from randomisation to surgery was 13 days (range 1 to 243), and was similar between the two arms. Two patients underwent surgery before randomisation and one patient had eight cycles of neoadjuvant chemotherapy prior to surgery. Axillary surgery was performed on 92.5% of patients and a clear margin was obtained for 94.5% of patients. Twelve patients (0.7%) in total underwent a WLE of the contralateral breast and one patient underwent a mastectomy of the contralateral breast (in the MRI arm).

Pathological findings are displayed in Tables 6–8, and it can be seen that the findings are very similar between the two arms. Additional findings (Appendix 13, Table 43, Pathology: predictive markers) are given in Appendix 13. The median weight for WLE specimens was 52.8 g (range 5.0–770.0) and for mastectomies was 842 g (range 217–2415). In total, 1154 patients (71.1%) had carcinoma in situ (CIS), 1466 patients (90.3%) had invasive carcinoma and 91 patients (5.6%) had DCIS alone. An additional patient had lobular carcinoma in situ (LCIS) alone. Of the 1466 patients with invasive carcinoma, 1114 patients (76.0%) had ductal NST, and 133 (9.1%) had lobular carcinoma. 1244 patients (84.9%) had localised disease and 179 patients (12.3%) had multifocal or multicentric disease. Extent of disease was not assessable for 16 patients (1.1%). Nodes were examined for 1470 patients (90.6%). Details of predictive markers are given in Appendix 13. Overall, 1242 patients were ER positive (76.5%) and 225 (13.9%) were ER negative; 823 patients (50.7%) were PR positive and 331 (20.4%) were PR negative; while the HER2 status was known in 665 (41%) patients, of whom 444 (66.8%) had a score of 0.

Secondary end points

Local recurrence-free interval

Local recurrence-free interval was calculated as the time from randomisation to the date of local recurrence, or death due to breast cancer. There were four patients who did not undergo surgery; therefore, these patients were censored at randomisation. There was one further patient who had a local recurrence; however, only the year and month were known, therefore this patient was assumed to have had a local recurrence on the 15th of the month. Patients with missing follow-up data, or who were alive and local recurrence-free at the time of analysis, were censored at the last date they were known to be alive and local recurrence free.

The median length of follow-up of all patients was 2.1 years (range 0.0–5.7), and this did not differ between the two arms. Figure 3 displays Kaplan–Meier curves of local recurrence-free intervals. Local recurrence-free interval rate at 1 year post randomisation was 99.87% (95% CI 99.05% to 99.98%) for patients randomised to MRI, compared with 99.73% (95% CI 98.93% to 99.93%), for patients randomised to no MRI. At 3 years, local recurrence-free interval rate was 93.90% (95% CI 90.94% to 95.92%) for patients randomised to MRI, compared with 96.46% (95% CI 93.88% to 97.96%) for patients randomised to no MRI. It was noted that there were more deaths due to breast cancer in the MRI arm (16/30, 53.3%) than in the no MRI arm (4/15, 26.7%), which explains the differences in local recurrence-free interval between the arms (since the number of local recurrences is similar between the two). These excess deaths in the MRI arm had previously been acknowledged by the independent DMEC and were thought to be due to chance.

FIGURE 3.

Local recurrence-free interval: Kaplan–Meier curves.

Cox’s proportional hazards model was fitted to adjust for the minimisation factors. The hazard ratio for MRI versus no MRI was 2.02 (95% CI 1.09 to 3.75), indicating an increased risk of local-recurrence or death due to breast cancer in the MRI arm compared with the no-MRI arm, reflecting the previous results.

Adverse events

Adverse events relating to the intervention were not routinely collected.

Quality assurance findings

In total, 171 MR scans (21% of all scans) were requested for re-reading. Eighteen of these could not be recovered, primarily due to local archiving problems. Of the remaining 153 scans, 12 (7.8%) were non-compliant with the technical scanning protocol, and five (3.2%) were considered as misreported. Of the misreported MR scans, three were based on technically non-compliant scans, and two used technically compliant scans. Findings are summarised in Table 22.

Irretrievable scans, unreadable scans and technical failures were concentrated at six small centres, which accounted for 43 (5.2%) scans undertaken in the trial. The remaining 39 centres, accounting for 94.8% of scans conducted during the trial, reported no technical failures, and had a misreporting rate of 1.4%. The performance of the above mentioned six centres and of the remaining 39 centres is summarised in Table 23.

Problems were particularly concentrated at one centre, which accounted for 24 of the total scans undertaken for the trial. Because only a sample of patient scans were re-reported, no patient data was excluded from the trial on the basis of the findings of the QA, as this could introduce bias. However, the primary end point analysis was conducted both including and excluding the most problematic centre, which accounted for 24 scans in the trial. The results from the analysis excluding this centre were consistent with those of the main ITT analysis.

Clinical results summary

• There was no evidence of a difference in the reoperation rate between the MRI and no-MRI groups. In the primary ITT analysis, 18.8% of patients in the MRI group underwent a reoperation, compared with 19.3% in the no-MRI group, with an odds ratio of 0.96 (95% CI 0.75 to 1.24) and p-value of 0.7691.

• Overall, the best agreement between all imaging modalities and histopathology, with respect to tumour size and extent of disease, was found in patients who were aged over 50, had ductal tumours NST and who were node negative.

• The sensitivity and positive predictive values of MRI for determining patient management were 50.0% and 61.8%, respectively, and of the 58 patients in the MRI arm who underwent a mastectomy, 16 (27.6%) were classed as being pathologically avoidable.

• Exploratory analyses considering the level of agreement in size of tumour between histopathology and each imaging method identified all imaging methods to have borderline moderate to fair agreement with histopathology (weighted kappa statistics range 0.3803–0.4767). In general, the imaging methods tended to upstage smaller tumours and downstage larger tumours.

• Additional findings in the contralateral breast were identified for 62 patients (7.6%) via MRI, resulting in 12 patients (19.4%) undergoing WLE and one patient (1.6%) undergoing mastectomy.

• There were no significant differences in the proportion of patients receiving chemotherapy, radiotherapy or additional adjuvant therapies between the groups. Overall, within 6 months of initial surgery, 29.1% of patients received chemotherapy, 57.9% of patients received radiotherapy, and 51.3% of patients received additional adjuvant therapies.

• None of the 25 patients with MR-only-detected < 5-mm lesions had a clinically significant lesion evident at their 12-month repeat MR scan. Of the 66 patients with ≥ 5-mm biopsy-negative lesions, only three had potentially clinically significant lesions at their repeat MR scan; however, this was based on overall lesion score as these lesions were not biopsied.

Assessment of compliance and missing data

Table 24 displays summaries of questionnaire timing for all patients in the QoL population. At 8 weeks post randomisation, the median time from randomisation to questionnaire completion was 8.4 weeks (range 1.6–25.4). At 6 months post surgery, the median time from surgery to questionnaire completion was 6.0 months (range 1.1–9.8). At 12 months post surgery, the median time from surgery to questionnaire completion was 12.0 months (range 8.2–20.9). Additionally, Table 25 displays the percentage of expected questionnaires that are missing for each QoL time point (i.e. baseline, 8 weeks post randomisation, and 6 and 12 months post initial surgery). The number of expected questionnaires excludes deceased patients and, at 6 and 12 months post surgery, patients who did not undergo surgery, but includes patients who withdrew or were withdrawn from the QoL study for any reason post randomisation. Compliance is good with rates being similar between the two arms at all time points. At each time point there were between 11% and 15% of patients who did not complete a whole QoL form. Compliance decreases over time but even at 1 year post initial surgery, compliance is high at 86.9% and 84.2% in the MRI and no-MRI arms, respectively. Reasons for missing questionnaires are not available.

Missing data were assessed in detail and overall there were very few missing data for individual QoL questions and QoL subscales, at most 5% for any single subscale score. Additionally, mean QoL scores were grouped by the timing of patients’ last assessments, and showed little difference in mean scores according to last assessment, indicating that missing data does not appear to be influenced by patients QoL. Summary tables and figures of missing data assessments are not given here. Data were deemed to be missing at random. In addition, the proportions of missing data at each time point were almost identical between the two arms. Since multilevel modelling was used, which accounts for missing data at the time point not of interest, and given the proportions of missing data were the same between the two arms, imputation was not deemed necessary for this analysis. Data summaries are presented for available case data.

Only prerandomisation assessments were included as baseline measurements (or assessments completed post randomisation, but before the patient was informed of their allocation result). Data were analysed using a time frame of ± 14 days around the expected date of completion of the questionnaire at 8 weeks post randomisation, a time frame of ± 28 days around the expected date of completion of the questionnaires at 6 months and ± 56 days around the expected date of the 1 year-post-surgery questionnaire. Table 26 displays the number and percentage of questionnaires received that were completed in the relevant time windows. Compliance within the time windows is very good at baseline and 6 and 12 months post initial surgery; however, it is lower at 8 weeks post randomisation, with only 66.7% and 66.3% of patients completing within 6–10 weeks post randomisation in the MRI and no-MRI arms, respectively. Using a time frame of ± 21 days rather than ± 14 days, questionnaires were completed in the window for 1035 patients (81.0%) [MRI: 522 (80.8%); no MRI: 513 (81.2%)]. In order to establish whether to conduct sensitivity analyses using the ± 21 days time window, data were summarised using median and mean QoL scores, with corresponding 95% CIs for the means for each time window. Results were almost identical for the two time windows, therefore sensitivity analyses were not deemed necessary and the original time window of ± 14 days was used.

Data summaries

Table 27 displays QoL summaries for those patients who completed questionnaires within the relevant time frames.

Overall, QoL scores were similar between the two treatment groups, with QoL decreasing minimally between baseline and 8 weeks post randomisation, then recovering at between 6 and 12 months post initial surgery.

At each time point, physical, social/family and functional well-being scores were above the normative data described by Webster and colleagues75 for a general US population. Physical and functional well-being decreased between baseline and 8 weeks post randomisation, then recovered between 8 weeks post randomisation and 12 months post initial surgery, whereas social/family well-being and breast cancer concerns scores did not change. Emotional well-being improved between baseline and 8 weeks post randomisation, then did not change.

HADS anxiety and depression scores were also similar between the treatment groups, with median anxiety scores of 7 at baseline, which then decreased between baseline and 8 weeks post randomisation and then stabilised. Depression scores were low at baseline and did not change thereafter.

In addition to the summary scores presented in Table 27, the HADS questionnaire was also summarised categorically, with anxiety or depression scores of 7, indicating that a patient shows normal levels of anxiety or depression, a score of 8–10 indicating that the patient shows borderline levels of anxiety or depression, and a score of greater than 10 indicating that the patient shows levels of anxiety or depression that are probably clinically significant. 46 Table 28 displays these results. Considering anxiety, at baseline the two treatment groups were similar and the percentage of patients who had clear signs of anxiety was moderate [MRI: 123/647 (19.0%); no MRI: 137/635 (21.6%)]. At 8 weeks post randomisation, although the groups were also similar, the percentage of patients who had clear signs of anxiety decreased [MRI: 48/431 (11.1%); no MRI: 53/419 (12.6%)]. At 6 months post initial surgery, 44/563 (7.8%) patients randomised to MRI had probably clinically significant signs of anxiety, compared with 61/547 (11.2%) patients randomised to no MRI; at 12 months post surgery 57/614 (9.3%) patients randomised to MRI had probably clinically significant signs of anxiety, compared with 67/589 (11.4%) patients randomised to no MRI.

The percentage of patients who had probably clinically significant signs of depression was less than 5%, at all assessment times. At baseline, 17/647 patients (2.6%) randomised to receive an MRI and 11/635 patients (1.7%) randomised to no MRI had probably clinically significant signs of depression. These percentages increased slightly at 8 weeks post randomisation and 6 months post initial surgery, and decreased again at 12 months post initial surgery.

Since the number of patients completing QoL questionnaires at each time point varies, the summary statistics cannot be validly compared over time, as a different subset of patients was observed at each point. QoL scores for subsets of patients with differing lengths of complete follow-up were calculated and showed similar profiles to those presented above; therefore, additional summaries are not presented.

Treatment comparisons at each time point

Mean adjusted for baseline QoL scores and 95% CIs, and corresponding differences between the trial arms at each time point, were calculated; however, results were very similar to the non-adjusted results presented above and are therefore not displayed.

Subsequent to developing the protocol for the COMICE trial, Eton et al. 76 established the minimally important difference (MID) in QoL measured using the FACT-B. This corresponds to the minimum difference in QoL that is both clinically and statistically significant, as it was established using both distribution- and anchor-based methods. In their paper, the authors outline a difference of 5–6 points to be the MID for the TOI. As can be seen by considering the TOI scores in Table 27, differences between the trial arms at each time point are minimal and do not reach the MID. Overall, QoL was found to be very similar between the two arms, with only minor changes in QoL over time.

Quality of life results summary

• The QoL substudy population was representative of the population on which clinical inferences were made.

• Overall, QoL scores were similar between the two treatment groups, with QoL decreasing minimally between baseline and 8 weeks post randomisation, then recovering at between 6 and 12 months post initial surgery.

• Differences between the two trial arms at each time point for the TOI score were minimal and did not reach the minimal important difference of 5–6 points.

• HADS anxiety and depression scores were also similar between the treatment groups, with median anxiety scores of seven at baseline, which then decreased slightly between baseline and 8 weeks post randomisation and then stabilised.

• Depression scores were low at baseline and did not change thereafter.

Resource use

The economic analysis was carried out once all patients had been followed up for at least 12 months. Details of some of the key resource use from the initial surgery can be found in Appendix 17 (Summary of key resource use from initial surgery). It should be noted that we have attempted not to replicate information on resource use that is contained in the clinical analysis in earlier chapters. The times in the anaesthetic room, theatre, recovery room and for axillary surgery were broadly similar across the two arms, although these were all marginally higher in the MRI arm. This was as expected, as more patients in this arm received surgery other than WLE surgery. Whilst marginally more patients in the no-MRI arm experienced complications, the number of patients who were returned to theatre was very similar in both arms. Similarly, the number requiring fluid replacement was very similar in the two arms. Only one patient was placed in a high-dependency ward following surgery.

Summaries of the resource use in the 12-month period following initial surgery are also given in Appendix 17 (Table 58). Resource use in the trial is broadly similar across the two arms. Slightly more patients in the MRI arm experienced complications between leaving hospital and their first post-operative follow-up. However, even though the number of patients who experienced complications was larger, the numbers who were admitted to hospital due to these complications was lower than in the no MRI arm of the trial. More patients in the no MRI arm required a repeat operation than those in the MRI arm. In contrast to the first post operative follow-up, more patients in the no MRI arm experienced complications after leaving hospital, whilst more in the MRI arm were hospitalised as a result. However, as with the first postoperative follow-up, the numbers are similar. At the 6-month follow-up, slightly more patients in the no-MRI arm had undergone an oophorectomy, but more patients in the MRI arm had undergone further surgery. At the 12-month follow-up, more patients in the MRI arm had been readmitted to hospital, or undergone an oophorectomy or other surgery. More patients in the MRI arm had received chemotherapy but less had received radiotherapy; however, the differences between the two arms are only marginal.

Unit costs

Unit costs at 2006–7 prices were used to value the resource use measured in the trial. These were taken to be long-term average costs. Details of some of the unit costs for key resource use in the trial can also be found in Appendix 17 (Prices and unit costs of major resource items). As explained in the methods chapter of this report, it is the incremental cost is of interest in this trial. As such when resource use was considered to be broadly equivalent across the two arms, the resources were not costed (for example, conventional triple assessment was not costed).

Missing data

Table 29 below describes the extent of missingness of the data once the resource use data was costed and aggregated into various cost categories. As can be seen from the table, certain components were subject to a large amount of missing data, in particular the cost of initial surgery, which is an important cost contributor but which had 44.79% of observations missing. Table 29 also describes the extent of missingness of the EQ-5D scores at baseline, 8 weeks post randomisation, and at 6 and 12 months post initial surgery. The dearth of cost of initial surgery data is due to a combination of missing forms, for which no data on costs were available, and also due to poor recording of times in the various aspects of surgery. Poor recording of times resulted in all the other costs accounted for in initial surgery being ignored as aggregation into the component was only possible when data on all of the resource usage of initial surgery was available. This data was primarily completed by the theatre staff or surgeons, and as such resulted in relatively poor completion in comparison with other data, which was routinely collected by research staff. The missing data in the other cost components and the EQ-5D data was largely a result of missing forms.

Imputed cost data

As a result of the missing data problems described above, various cost components were imputed using multiple ICE (this is described in the methods section of this report). Table 30 details the mean costs, standard errors and 95% CIs for the various aggregated cost categories. The CIs have been calculated assuming that the cost components follow a gamma distribution.

As can be seen from the table, the mean cost of initial surgery was higher in the MRI arm, as expected given the longer duration of the surgery and more non-WLE surgeries being conducted. The cost of staying in hospital following the initial surgery was also higher in the MRI group, reflecting the slightly longer mean length of stay than the non-MRI arm (3.55 nights compared with 2.86 nights). The MRI cost was larger in the MRI arm, as was expected. Due to the larger number of patients in the non-MRI arm who received a repeat operation, the costs were higher. However, the CIs of the mean estimates for both arms do overlap. The mean cost of follow-up at 6 months is very similar in both arms (£102.06 in the MRI arm compared with £105.02 in the non-MRI arm). The mean cost of follow-up at 12 months is higher in the MRI arm although, again, the CIs overlap. The mean cost of chemotherapy is higher in the MRI arm, but this partially offset by a lower mean cost of radiotherapy. GP costs are broadly similar between the arms at 8 weeks post randomisation, and 6 and 12 months post initial surgery.

The mean total cost of the resources measured in this cost analysis was higher in the MRI arm than in the non-MRI (£5508.40 compared with £5213.50, resulting in an incremental cost of £294.90). This is largely a result of the extra cost of MRI (an additional £236.45 for the MRI arm compared to the non-MRI arm) as the other cost categories are reasonably balanced between arms. Given the lack of difference in clinical end points in the trial, the similar total costs are as expected with the difference appearing to be driven by the one clear resource use difference between the two arms – the use of MRI.

Regression analyses on imputed cost data

Table 31 presents the regression of total cost on treatment arm, age, BMI and whether the patient has had a recurrence at 12 months post initial surgery. This has been performed using a generalised linear model with an identity link function and assuming a gamma distribution. The regression allows us to look at the effects of age, BMI, recurrence status and treatment arm have on total costs. An example of how these results can be interpreted is presented below.

For example, a 60-year-old in the MRI arm who had not suffered a recurrence and with a BMI of 30 would have the following cost: total cost = (292.35 × 1) + (–109.71 × 60) + (2882.67 × 0) + (55.29 × 30) + 10050 = £5418.64.

Table 31 suggests the mean additional cost per patient of the MRI arm is £292.35 after controlling for the other variables included as regressors in the model. However, this is not statistically significant at a 5% level. The negative coefficient on age suggests that total costs decrease with a patient’s age. The positive coefficient on BMI suggests that a patient’s total cost increases with their BMI. The results also suggest that patients who have had a recurrence have significantly higher costs than those who do not (£2882.67 more).

Imputed EQ-5D scores

As with the cost components, the large extent of missing data (see Table 29) in the EQ-5D scores made it necessary to impute the missing observations using multiple ICE. The EQ-5D mean scores, standard errors and 95% CIs are detailed in Table 32, by trial arm, for baseline, 8 weeks post randomisation, and 6 and 12 months post initial surgery.

The mean score in both arms fell from baseline to 8 weeks post randomisation and then increased at 6 months post initial surgery and again at 12 months post initial surgery. The scores are very similar between the arms at each time point, suggesting that there is no difference in HRQoL. Given the lack of difference in clinical end points between the two arms, the similar EQ-5D scores are as expected.

Regression analyses on imputed EQ-5D scores

Table 33 details the results of an ordinary least squares regression of the EQ-5D score at 12 months post initial surgery in the MRI arm, baseline EQ-5D, age, whether the patient had had a recurrence by 12 months and BMI. The coefficient on the baseline EQ-5D score is positive and statistically significant, which is expected given that other studies have shown that current HRQoL is related to previous HRQoL. The positive coefficient on age suggests that HRQoL increases with age, which is counterintuitive; however, the coefficient is not statistically significant at a 5% significance level. The negative coefficient on recurrence suggests that, on average, those patients who have had a recurrence will have a lower HRQoL than those who have not, which would be expected. However, again the coefficient is not statistically significant at a 5% significance level. This may be due to the small proportion of individuals who experienced a recurrence and, as such, the trial was not powered to identify this effect. The results also suggest that patients with a higher BMI, i.e. those who are more obese, will have a lower HRQoL, which would be expected.

The negative coefficient on the MRI arm suggests that after controlling for baseline EQ-5D, age, whether the patient had had a recurrence and BMI the mean score is lower for patients in the MRI arm, although the size of the coefficient suggests there is no difference between the arms. However, the coefficient is very small and is not statistically significant at a 5% level of significance and, as such, it would appear that there are no differences between the treatment arms. This result is consistent with the clinical analyses, in which no differences in primary end points were found between the arms.

Economic evaluation results summary

• The economic evaluation found that 12 months after initial surgery there was no statistically significant difference in HRQoL, as measured by the EQ-5D, between the two trial arms once baseline HRQoL and other covariates were controlled for. The nominal values of the point estimates of the mean changes between baseline and 12 months were also very similar.

• These results are consistent with the clinical and QoL findings that there is no difference between the trial arms in terms of outcomes.

• The economic evaluation did suggest that there may be a cost difference between the two trial arms, with the MRI arm having a larger mean resource cost per patient (£5508.40 compared with £5213.50), although the difference was not statistically significant once other covariates had been controlled for.

Clinical and sociodemographic characteristics

Forty-six (46%) of the 100 participants taking part in the NSSI study were randomised to receive MRI. Three patients who were randomised failed to have the scan. In one case, the scanner was undergoing repair, in the second case, the patient declined the scan and in the final case the patient was unable to fit into the scanner.

The median age of respondents was 59 years (range 37–82). Ninety-four women had breast conservative surgery (WLE or quadrantectomy) and six had unilateral mastectomy. The median number of nights in hospital postoperatively was two (range 1–10) and 15 (15%) of the women reported postsurgical complications.

Eighty-nine women expressed satisfaction with the shape of their breasts following surgery. Forty-three said they were very satisfied and 46 said they were satisfied. Ten women expressed dissatisfaction with their shape, two of whom were very dissatisfied.

In terms of adjuvant treatment, the women were having, or had had, the following treatments: chemotherapy – 38%; radiotherapy – 92%, hormone therapy – 78% and trastuzumab – 5%. The self-reported median waiting time post surgery for chemotherapy was 4 weeks (range 2–20 weeks) and for radiotherapy 12 weeks (range 2–36 weeks). Ten per cent of those receiving radiotherapy thought it had been given too late.

Response to randomisation

Fifty-five percent of the patients were pleased with the outcome of the randomisation process, although the majority of this group was randomised to have MRI. All of the patients randomised to have MRI were pleased with the outcome, whereas only 66% of patients in the ‘no scan’ group were pleased with the outcome (chi-squared test for trend = 53.35, p < 0.0001) (Table 34).

The majority of patients who had MRI were reassured by it. Two patients said it made them anxious and 16 were indifferent to the decision. Not having the scan made 18 of the patients anxious, and 32 said they were indifferent (chi-squared test for trend 33.397, p < 0.0001) (Table 35).

Procedural distress

Of those patients who received an MR scan, 34% indicated that they had found this ‘distressing’, and 14% found it ‘very distressing’. In contrast, 1% of those having USS found this distressing (and she found it very distressing), 56% found biopsy distressing (14% very distressing), and 30% found mammography distressing (4% very distressing). The type of distress reported by the women is shown in Table 36 (some women identified more than one type of distress).

Perceived choice of surgery

Thirty (30%) of the women said that they had been given a choice of operation and 26 (86%) found this helpful. Sixty-nine women said they had not been given a choice and of these 64 (92%) said that that had been helpful. One could not remember whether she had been given a choice. There was no significant difference between those given a choice and those not in terms of how helpful this was perceived to be (Fisher’s exact p = 0.448).

Of the 98 women expressing an opinion 28 (93%) of the women who were given a choice of surgery felt that they had made the right decision about their operation. Sixty-three (92%) of the women who were not given a choice felt that the operation they had was the right one for them (Table 37). The proportion of women considering that they had received the right operation for them was unrelated to having been given a choice (Fisher’s p = 1.000).

NSSI summary of results

• One hundred women from a range of recruiting centres participated in a non-schedule standardised interview study to evaluate aspects of their experience of treatment and study participation.

• All of the patients randomised to have MRI were pleased with the outcome of randomisation compared with 66% of patients in the no-MRI group.

• The majority of patients who had MRI were reassured by having it.

• Thirty-four per cent of patients randomised to MRI indicated that they had found this ‘distressing’, and 14% found it ‘very distressing’, whereas 1% found USS distressing, 56% found biopsy distressing and 30% found mammography distressing.

Trial outcomes

Summary

The results of the COMICE trial are important from both a health economic aspect, and also from a patient burden aspect. MRI is an expensive procedure. The findings of this trial are of benefit to the NHS as they show that this additional procedure may not be necessary in this population of patients, thus additional funding may not required and potential increases in waiting lists for MRI may be avoided. Furthermore, not requiring an MR scan may relieve the burden of an additional hospital visit or a delay in the care pathway due to the availability of an MRI slot.

The COMICE study is the first large pragmatic trial evaluating the effectiveness of MRI of small breast lesions, suitable for WLE. The results have shown that the addition of MRI to triple assessment in women with breast tumours deemed suitable for treatment by WLE does not result in a reduction in the subsequent reoperation rates. However, this does not necessarily reflect a poor correlation between tumour extent as assessed by MRI and histopathology. This correlation is in general better than that obtained for mammography and USS, particularly for larger lesions and although the kappa statistic is only in the fair to moderate range, allowance must be made for inaccuracies in histopathological reference measurements, due to the orientation of sectioning of tissue blocks and the inability to correct for tissue shrinkage that occurs during processing and mounting. As detection of intraductal tumour is limited macroscopically, use of fresh specimens would not have been a feasible alternative. The improved sensitivity of MRI over mammography and USS is reflected in the detection of increased lesion size or multifocal/multicentric disease in 39 women, who following MRI underwent mastectomy appropriately.

A small percentage of patients underwent a pathologically avoidable mastectomy due to the suboptimal specificity of MRI, emphasising the requirement for biopsy of lesions that might result in an alteration to the planned surgical procedure.

The extensive QoL analysis performed showed no differences between the trial arms. It demonstrated high levels of satisfaction and reassurance in those randomised to receive MRI, despite the reported levels of distress secondary to the procedure, which were comparable to those reported for mammography and less than those for breast biopsy.

Several studies have examined the importance of tumour-free surgical margins after breast-conserving therapy. 3–6 Ideally, the tumour, along with a margin of at least 10 mm of normal-appearing tissue, is resected to attempt to remove any microscopic cancer. The minimum cosmetically acceptable tumour-free margin in relation to the risk of local or distant recurrence has been debated in many studies. 87,88 For image-guided ablation successful treatment of the entire tumour relies on accurate tumour volume delineation using imaging for both tumours targeting and monitoring the ablation procedure. Although the results of COMICE did not result in a reduction in the reoperation rate, there was a fair to moderate correlation of imaging with histopathology. The audit of the NHS BSP by ABS at BASO also indicates that, at best, there has been no reduction in the reoperation rate following WLE over the past 7 years. This is despite advances in imaging techniques and an experienced workforce. Thus, considerations must be given to the investigation of alternative treatments, potentially image-guided ablation, for the treatment of small tumours.

Implications for practice

The addition of MRI to triple assessment in women with small breast tumours does not result in a reduction in reoperation rates.

Preoperative biopsy of MRI-only-detected lesions prior to surgery is likely to minimise the incidence of inappropriate mastectomy.

Research recommendations

Contribution of authors

Professor Lindsay Turnbull, (Centre for MR Investigations, University of Hull) was Chief Investigator for the trial, a grant co-applicant who contributed to the conception, design, conduct and monitoring of the trial, and who contributed to the analysis, the interpretation of results, the drafting of the report and approved the final version.

Sarah Brown (Senior Medical Statistician, Clinical Trials Research Unit, University of Leeds) conducted the analysis of the clinical data, participated in the data monitoring, interpretation of results, contributed to the drafting of the report and approved the final version.

Catherine Olivier (Senior Trial Manager, Clinical Trials Research Unit, University of Leeds) contributed to the conduct of the trial, was instrumental in collection and verification of all data, participated in the interpretation of results, the drafting of the report and approved the final version.

Dr Ian Harvey (COMICE Project co-ordinator, University of Hull) led the quality assurance review, participated in centre enrolment, data monitoring, contributed to the drafting of the report and approved the final version.

Professor Julia Brown (Director, Clinical Trials Research Unit, University of Leeds) was the Supervising Statistician for the trial, and took a lead role in protocol development and implementation, contributed to data monitoring, the analysis, interpreting the data and approving the final version of the report.

Professor Phil Drew (Honorary Consultant Surgeon, Royal Cornwall Hospitals NHS Trust), a grant co-applicant who contributed to the design of the trial, participated in the data collection, data monitoring, the interpretation of results, contributed to the drafting of the report and approved the final version.

Professor Andy Hanby contributed to the design of the trial, data collection, data monitoring, contributed to the drafting of the report and approved the final version.

Andrea Manca (Senior Research Fellow, Centre for Health Economics, University of York) contributed to the design of the trial and oversaw the economic evaluation, and contributed to drafting sections of the report describing the methods and results of the economic evaluation analysis, and approved the final version of the report.

Vicky Napp (Operations Director, Clinical Trials Research Unit, University of Leeds) who was CTRU Principal Investigator for the project, contributed to the design of the trial, trial set-up, the conduct of the trial, chaired meetings on a regular basis to review and manage the progress of the project, contributed to analysis and approved the final version of the report.

Professor Mark Sculpher (Professor of Health Economics, Centre for Health Economics, University of York) was a grant co-applicant who contributed to the design of the trial and oversaw the economic evaluation, and contributed to drafting sections of the report describing the methods and results of the economic evaluation analysis, and approved the final version of the report.

Professor Leslie Walker (Director of Institute of Rehabilitation, University of Hull) was a grant co-applicant and was the lead for the NSSI (Well-Being) Sub-Study contributing to the design of the trial, data monitoring, interpretation of Quality of Life results, drafting of the report and approved the final version.

Simon Walker (Research Fellow, Centre for Health Economics, University of York) conducted the economic evaluation analysis and interpretation of results, contributed to drafting sections of the report describing the economic evaluation analysis, and approved the final version of the report.

Disclaimers

The views expressed in this publication are those of the authors and not necessarily those of the HTA programme or the Department of Health.