An international randomised controlled trial to compare TARGeted Intraoperative radioTherapy (TARGIT) with conventional postoperative radiotherapy after breast-conserving surgery for women with early-stage breast cancer (the TARGIT-A trial)

Inclusion criteria

All patients aged ≥ 45 years with operable invasive breast cancer [tumour, nodes, metastasis (TNM) – T1 and small T2 ≤ 3.5 cm, N0–1, M0], confirmed by cytological or histological examination, who were suitable for breast-conserving surgery were eligible. The tumour needed to be clinically suitable for breast conservation on conventional imaging. A MRI scan was not required. Individual centres could restrict entry to a more exactly defined subset of patients, in which case only patients with these characteristics could be entered by that particular centre. For example, centres could at the outset decide to recruit only women aged > 50 years or even only women aged > 65 years. Such treatment policies were predefined in writing and approved by the International Steering Committee (ISC). Patients needed to be available for regular follow-up (according to local policies) for at least 10 years.

Exclusion criteria

• More than one obvious cancer in the same breast as diagnosed by clinical examination, mammography or ultrasonography (MRI not required).

• Bilateral breast cancer at the time of diagnosis.

• Ipsilateral breast had a previous cancer and/or irradiation.

• Patients known to have BRCA gene mutations but testing for gene mutations was not required.

• Lobular cancer or EIC (in EIC ≥ 25% of the tumour is intraductal) on core biopsy or initial pathology (if performed).

• Patients undergoing primary medical treatment (hormones or chemotherapy) as initial treatment with neoadjuvant intent of reducing tumour size were excluded; those given short-duration (up to 4 weeks) systemic therapy could be included.

• Patients presenting with gross nodal disease, considered to be clinically malignant or proven cytologically or by scanning. In general, four or more positive nodes or extranodal spread meant that a patient was not suitable for TARGIT alone and should receive EBRT as well. However, individual centres could decide that anything more than micrometastasis should receive EBRT.

• Patients with any severe concomitant disease that may limit their life expectancy. Previous history of malignant disease did not preclude entry if the expectation of relapse-free survival at 10 years was ≥ 90%.

• Any factor included as an exclusion criterion in the local centre’s treatment policy. This was particularly relevant to patients entered into the postpathology stratum.

• No more than 30 days elapsed between last breast cancer surgery (not axillary) and entry into the trial for patients in the postpathology stratum.

Each patient was given time to consider participation and ask questions before giving consent by signing the patient consent form and randomisation occurred only after fully informed consent was freely given.

The treatments in each of the arms are described in detail in the protocol, which is published in full at www.nets.nihr.ac.uk/projects/hta/076049 (accessed 14 July 2016). Patients in both arms underwent a similar primary surgical procedure. The patients in the conventional arm underwent standard EBRT (according to the centres’ predefined local policy). Patients in the experimental arm underwent TARGIT as described previously in detail. 22,24,36 The papers, presentations and video related to TARGIT are available at www.targit.org.uk (accessed 29 June 2016) and a video demonstrating the operative technique is available at http://goo.gl/iuF9ZR (accessed 29 June 2016).

Protocol amendment

In the initial trial design, randomisation to TARGIT or EBRT was carried out before lumpectomy (prepathology). However, the trial was also firmly rooted in the principles of pragmatism to test a new approach (single-dose TARGIT to the tumour bed followed by EBRT in patients with unforeseen adverse factors). Therefore, when some of the centres planning to join the trial requested us to allow them to give intraoperative radiotherapy (TARGIT IORT) as a second procedure by reopening the wound, we permitted this. This decision facilitated a more stringent selection of patients (tumour pathology was available, hence postpathology) and was logistically easier, allowing enrolment of patients from neighbouring centres who had already had the lumpectomy. We therefore made a protocol amendment on 22 September 2004, obtained ethics approval and added this postpathology stratum to the trial, along with a completely separate randomisation table for such patients.

We specified that postpathology patients should be randomised within 30 days after lumpectomy. If allocated to TARGIT, patients in the prepathology stratum received it concurrently, immediately after surgical excision under the same anaesthesia; patients in the postpathology stratum received it as a subsequent procedure. We planned a separate analysis of the two strata (prepathology vs. postpathology). The rationale for stratification according to the scheduling of radiotherapy was that randomisation to the trial after full pathology had become available might theoretically allow better case selection. Conversely, treatment given at the time of initial lumpectomy could have a greater effectiveness because of its immediacy. Furthermore, the degree of accuracy of placement of the radiotherapy applicator for giving TARGIT by reopening the cavity might be quite different from that achieved at the time of original lumpectomy.

Stratum 1: prepathology entry

Eligible patients’ consent was sought and randomisation was carried out prior to the surgical removal of the tumour. Postoperatively, some patients were found to have characteristics that militated against a single intraoperative dose of radiotherapy (e.g. lobular carcinoma, positive margins at first excision, extensive lymphovascular invasion, multiple involved axillary nodes). In these cases (as per each centre’s pre-specified treatment policy), patients were recommended to have a full course of EBRT, without the tumour bed boost. Grade 3 cancers were not necessarily excluded. This histology was considered in combination with other factors such as presence of lymphovascular invasion, as specified in each centre’s policy document. For example, a patient with a screen-detected 1.5-cm grade 3 oestrogen receptor (ER)-positive tumour would have been eligible and been randomised. However, in rare circumstances she may postoperatively have been found to have extensive lymphovascular invasion and multiple positive lymph nodes. This would mean that if she had received TARGIT she would have needed to have EBRT as well, while remaining in the TARGIT arm of the study. The intraoperatively delivered radiotherapy replaced the boost in this instance. The two arms were compared on the basis of the policy to provide local tumour control, accepting that some (about 15%) of the patients in the TARGIT arm would also receive EBRT. This design tested the ‘real-world’ policy as it was the most likely way that TARGIT would be implemented in the future if proven to be non-inferior and/or less toxic in the trial.

Stratum 2: postpathology entry

Eligible patients were randomised for entry to the trial after they had had their cancer removed with a lumpectomy. If allocated to the TARGIT arm they had further surgery to reopen the wound, with TARGIT delivered to the tumour bed. This stratum was added for logistical reasons, particularly in some centres. Furthermore, this stratum allowed easier operating theatre logistics and a more stringent case selection, although it required a reoperation to reopen the wound to provide TARGIT as a delayed procedure. This was a disadvantage for patients in this stratum, although reopening the wound could have been carried out under local anaesthetic. 38 However, it allowed for the entry of patients who had already received surgery at an outlying centre, a common practice in countries with a large land mass and large rural populations.

Stratum 3: previous contralateral breast cancer

Most patients who require treatment for a metachronous breast cancer are excluded from entry to trials of local treatment. However, the second cancer is usually treated in a manner similar to the first one and intraoperative treatment may be particularly suitable. This trial assessed this stratum only for ipsilateral local control and data were censored for other end-point assessments.

It is important to recognise that the trial was pragmatic and tested two policies rather than two techniques. In other words, the trial did not test TARGIT compared with EBRT but a pragmatic policy of TARGIT (in 100%) ± EBRT (in approximately 15%) compared with EBRT (in 100%), that is, the novel approach was to use TARGIT in potentially eligible patients and add EBRT if an unexpectedly higher risk was found postoperatively. From our Phase I/II studies we had established that the TARGIT ± EBRT protocol was safe22,23 and very effective. 27,28,39

Prior to patient recruitment

Prior to entry of any patients, each centre registered with the ISC and completed a treatment policy document that defined the categories of patients to be entered (e.g. patients aged > 50 years, all N0) together with some details of treatment policy (e.g. fractionation and dose of conventional radiotherapy to be used). Any change to practice during the course of the trial had to be notified to the ISC in writing prior to implementation. This was to enable the ISC to audit the patients entered and to confirm that treatment remained true to the core protocol.

Only clinical centres with the INTRABEAM or those that were able to refer patients to such a centre could enter the trial. Centres with newly acquired equipment were required to consult the TARGIT trial operations office prior to entering patients into the trial. Confirmation of the quality control of the system set-up had to be received at the operations office before randomisation could begin.

Before entering any patients into the trial, centres were expected to submit data for each X-ray source probe and applicator set in use. Each centre was responsible for measuring data for the probe and applicator set and submitted the data supplied by the manufacturer for comparison with measured data together with a copy of the letter of acceptance supplied by Carl Zeiss.

In addition, a minimum of five ‘pilot’ cases (non-randomised patients) were performed in every new centre followed by an audit by a member of the ISC (or an appointed delegate).

Surgery

All patients had wide local excision of the primary tumour following appropriate clinical work-up. No special assessments prior to randomisation were required, although mammography and ultrasound were recommended to try to exclude multifocal disease and to determine as accurately as possible the size of the tumour.

Surgery was carried out according to usual local practice. Complete macroscopic excision of the tumour was required. The aim of the local excision was to achieve the widest margin of excision while maintaining a good cosmetic outcome. The final histological margin needed to be ≥ 1 mm clear of all invasive and in situ disease.

For superficial tumours an ellipse of overlying skin was normally excised. The depth of resection depended on the position of the tumour within the breast and the size of the breast, but in most instances extended to the pectoral fascia.

The protocol specified that in all patients, but especially in those women with impalpable tumours in whom preoperative localisation had been performed, the specimen needed to be well orientated with sutures or clips according to local protocols and to be radiographed intraoperatively. The specimen radiograph was examined in theatre to ensure complete excision of the lesion and to help with the assessment of adequacy of the margins. Further tissue had to be taken (and marked) from a margin if the radiographic abnormality had extended near the margin.

Either a standard sentinel node biopsy or at least level II axillary node clearance had to be performed in all patients. The protocol specified that similar surgical techniques should be employed in all patients regardless of randomisation, wound closure should be performed meticulously (air and water tight) as described22–24 and sutures (if non-absorbable) should remain in place for 14 days. In our pilot study of 300 patients27,28,39 we had found that, with the meticulous wound closure specified, wound healing was not a problem whether TARGIT was delivered during the primary surgery or as a secondary procedure.

Pathological examination

Data from pathological examinations was recorded on the appropriate data collection forms. We recommended that the minimum data as requested on the case record forms were recorded.

Radiotherapy

Targeted intraoperative radiotherapy

Intraoperative radiotherapy was delivered either in the operating theatre immediately after the removal of the tumour or as a subsequent procedure, a short time later.

The concept and the TARGIT technique, which was pioneered by investigators at UCL, London,22–24 allows the patient to potentially receive all of the required radiation in a single fraction before she awakes from surgery22–24,34,37,41–48 (Figure 3). The INTRABEAM device provides a point source of 50-kV energy X-rays at the centre of a spherical applicator. The appropriately sized (1.5–5 cm diameter) applicator is placed in the tumour bed using meticulous surgical technique including a carefully inserted purse-string suture that ensures that breast tissues at risk of local recurrence receive the prescribed dose while skin and deeper structures are protected. Radiation is delivered to the tumour bed over 20–45 minutes. The surface of the tumour bed typically receives 20 Gy, which attenuates to 5–7 Gy at a depth of 1 cm.

FIGURE 3.

The TARGIT technique: (a) The INTRABEAM device; (b) a schematic diagram showing how the spherical applicator fits into the tumour bed; (c) IORT being delivered in the operating theatre; and (d) a close-up of the spherical applicator in the tumour bed. Reproduced with permission from Vaidya et al. 2 Copyright © Vaidya et al. Open Access article distributed under the terms of CC BY-NC-ND.

The procedure has been described previously. 22–24 The papers, presentations (see Appendix 3) and video related to TARGIT are available at www.targit.org.uk (accessed 29 June 2016) and a video demonstrating the operative technique is available at http://goo.gl/iuF9ZR (accessed 29 June 2016).

All patients should receive a prophylactic dose of antibiotic just before skin incision. The device and the arm of the stand are wrapped in a sterile clear plastic cover. The individual applicators are sterilised prior to the theatre session. The size of the sphere is determined at surgery by the surgeon and/or the radiation oncologist. An appropriately sized INTRABEAM sphere fits comfortably without tension in the surrounding tissue so that the skin and subcutaneous tissues can be gathered with a purse-string suture over the sphere. Any other technique to assist this apposition may also be used. The surgeon and radiation oncologist should choose the largest possible suitable applicator to ensure that the highest dose is delivered to the tumour bed tissue.

It is essential that complete haemostasis is achieved before insertion of the applicator sphere, because even a small ooze of blood can distort the cavity around the sphere and significantly change the target dose. The applicator sphere is inserted into the surgical cavity and a deep surgical purse-string suture is inserted in the breast tissues to bring together the target breast tissue so that it applies well to the surface of the INTRABEAM applicator sphere and holds it in place during treatment. The skin, but not the breast tissue, should usually be everted and held away from the delivery device by surgical sutures to prevent direct contact with the sphere. One patient in the pilot series did develop an area of skin necrosis. It is important to keep the skin at a distance of at least 1 cm from the applicator.

If necessary, protective caps (made from tungsten-impregnated rubber and available from Carl Zeiss) may be fashioned by the surgeon to protect deep or superficial structures. If the deep margin of excision is such that the left anterior descending branch of the coronary artery could receive a significant radiation dose, then the surface of the applicator sphere should be covered with a protective cap at the chest wall. However, in most patients the normal thickness of the chest wall (muscle and rib cage) provides adequate shielding and such a protective cap is not required. Sometimes the superficial skin flap may require protection with a 0.5-cm thick, cut piece of wet gauze. Care must be taken, however, not to inadvertently shield the areas of tissue that require radiation treatment. The shaft does not emit radiation so wet gauze should not be placed between the shaft and the skin. The anterior surface of the tumour bed should be a relatively thick skin flap so that it can receive radiation from inside without the applicator being too close to the dermis or if this skin is too thin it can be excised as a skin ellipse to get clear margins.

Radiation protection shielding material should also be used to cover the would around the radiation device; this significantly reduces the dose to the operating theatre staff to very low levels.

The protocol allowed for two dose prescriptions, each of which was equivalent to 20 Gy at the surface of the applicator. Each participating site decided on one method and used this procedure for all patients for the duration of the trial.

• Alternative A. A dose of 20 Gy at the surface of the applicator (in water) was prescribed by the radiation oncologist and delivered to the breast tissue. This takes approximately 30 minutes, depending on the size of the applicator.

• Alternative B. A dose of 6 Gy at 1 cm (in water) was prescribed by the radiation oncologist and delivered to the breast tissue. This also takes approximately 30 minutes, depending on the size of the applicator. This dose is equivalent to 5 Gy at a 1-cm depth for adipose tissue.

Using prescription A ensured that larger tumours received a slightly higher dose in the tumour bed (> 6 Gy at 1 cm and beyond). Previous versions of the TARGIT protocol recommended a dose of 5 Gy at 10 mm in adipose tissue, which is equivalent to 6 Gy at 10 mm in water. Rules in some countries such as Germany dictate that the prescription should always be at the highest dose delivered. Therefore, we adopted alternative A but at the same time, as the dose at the surface cannot be physically measured (but rather estimated), we kept the option of using alternative B, prescribing the dose at 1 cm, because the dose delivered by both approaches will typically be very similar. With the publication of further research40,45 it was later recommended that 20 Gy at the applicator surface should be adopted by all new sites. Such a dose prescription is arguably more logical as it ensured that tumour beds from larger tumours received a higher dose than small tumours.

During the radiation treatment, the anaesthetist, clinician and physicist could remain in the room. To avoid unnecessary exposure we recommended that as many people as possible vacated the operating theatre and those remaining either wore a lead apron or remained behind a shielded screen. No modifications to the operating theatre were required.

After completion of radiation, the conforming stitches are removed. Strict haemostasis needs to be obtained following the removal of the INTRABEAM device. The skin is sutured meticulously to achieve a water-tight closure and a good cosmetic result. If non-absorbable sutures are used they are left in situ for 14 days and, if absorbable sutures are used, Steri-Strips covering the entire wound are left in place for 14 days.

Adjuvant systemic therapy

Following completion of TARGIT, patients were recommended appropriate adjuvant therapy according to local practice or trial protocols. For all trial patients, the sequencing of these other therapies was not governed by this protocol, but careful consideration needed to be given for patients randomised to TARGIT but needing EBRT and adjuvant chemotherapy. The policy for such treatments needed to be declared in advance in the treatment policy document. It was recommended that even the postpathology TARGIT was delivered before beginning chemotherapy.

Adverse events

Details of management of adverse events, definitions of suspected serious adverse reactions and suspected unexpected serious adverse reactions and the reporting requirements are provided in the protocol. The ISC and the DMC reviewed data on adverse events and complications. Acute and late radiation morbidity was graded according to the Radiation Therapy Oncology Group (RTOG) criteria and ‘pain due to radiation’ according to the common toxicity criteria. 50 These were the only expected adverse events but other adverse events were also recorded.

Proposed sample size

The main objective of the trial was to determine whether or not the use of IORT gave rates of local control that were not inferior to those obtained using EBRT.

This was a non-inferiority trial with a one-sided design and we selected a clinically meaningful non-inferiority margin, δ₀ > 0. We were interested in testing H₀: δ = δ₀ compared with H₁: δ < δ₀ using a one-sided level-α test with power 1 − β to reject H₀ when δ = 0.

We defined the power calculations for this trial using absolute values for local breast relapse. Estimates were obtained from published data. The Oxford Overview of radiotherapy in breast cancer showed a baseline rate of about 7.8% at about 10 years. 6 However, most of these trials began in or before 1985 and the outcomes have much improved since then such that many papers are reporting much lower rates with conventional radiotherapy. Hence, we had originally considered a baseline relapse rate of 6% at 5 years and therefore for a power of 80% to detect an absolute increase or decrease in relapse rate of 2.5% (the non-inferiority margin) we would need 2232 patients. If the absolute rate of recurrence is lower, for example 4%, as may be expected from recently announced trial results,49,51–53 slightly fewer (n = 2153) patients would allow an 80% chance of detecting a difference of 2.1% (the non-inferiority margin). Although the odds ratio detectable for an even lower (than 4%) background recurrence rate may not be as low, the absolute difference in recurrence rates would be very low (< 2%) and clinically acceptable. We therefore maintained the original accrual goal of a total of 2232 patients to have adequate power to meet the primary objective. It is possible that the final recurrence rate is low and the numbers accrued will give substantially more than adequate power.

The sample size was calculated for the main end point – local recurrence – for the whole patient population. At the time of the funding application to HTA in 2007–8, 70% of patients were in the single-procedure stratum and 25% in the two-procedure stratum, with < 5% in the contralateral stratum. The main (ITT) analysis was planned to be performed on the whole population and stratified analyses in the two main strata. We expected to yield a meaningful answer even in the smaller postpathology (two-procedure) stratum, in which we initially expected to have about 600 very-low-risk patients. The reasoning for this is as follows. The background recurrence rate in this stratum was expected to be very low, possibly as low as 1–2%. As we can clinically accept the original absolute non-inferiority margin of 2.5%, the nominal statistical hazard is high (e.g. 1.2% vs. 3.7%). Therefore, the statistical power (one-sided log-rank, 80% power, 95% confidence) of these 600 patients would be adequate for a meaningful result.

This rationale for power analysis also holds true if we have a significantly lower overall recurrence rate, in which case the study will have more than adequate power to confidently dismiss a clinically significant difference in local recurrence rate for establishing non-inferiority.

We alluded to this concept in our original application and were interested to see that it has been used in the recently published results of the START trial. 49

Therefore, once we had recruited > 800 patients with > 3 years of follow up, we planned to perform a futility analysis, which would be able to inform us whether or not the difference between the two groups will ever reach a clinically significant level.

Early stopping of the trial because of demonstration of superiority, or non-inferiority, or for significant safety reasons was within the remit of the DMC.

Our plan was to undertake one formal interim analysis for efficacy when half the expected number of events had been reached using an O’Brien–Fleming rule with a stringent p-value of < 0.001. Thus, stopping early for futility would follow an analysis which demonstrates that the 99.9% CIs fall outside the lower range defined for equivalence (i.e. the upper end of the 99.9% CI of 5-year local recurrence-free survival rate is < 91.5%).

If, however, any of the following were demonstrated at p < 0.01 the DMC would consider recommending early stopping:

• a significant increase in grade 4 skin or rib fractures (as a sign of radionecrosis)

• delays in wound healing, which after detailed review by the DMC as to time course and severity were considered clinically significant.

Early stopping could be applied to specific strata, with other strata allowed to continue recruitment. In addition, early stopping would be applied depending on the results of the futility analysis. However, by that time, most of the target recruitment would have already occurred.

When the original sample size of 2232 was calculated, we based our estimate of the 5-year local recurrence rate of 6% on the literature available in 1999. 6,54 We chose the non-inferiority margin as an absolute difference of 2.5% because this seemed clinically acceptable to physicians and patients. However, during the past decade recurrence rates have substantially reduced. The recurrence rate in the control group of our trial was 0.95% at 3 years. It would be logical to extrapolate the 5-year local recurrence rate to 1.5%, which is not unexpected. For example, in the UK START trial,55 patients had a worse prognosis (e.g. 36% had a tumour size of > 2 cm vs. 14% in TARGIT-A and 22% were node positive vs. 17% in TARGIT-A) and were treated a few years before the patients in our trial. In the START trial the recurrence rate at 5 years was 2.3%. Therefore, the estimate of 1.5% for our trial is not unrealistic.

Statistical analysis

The major end point was the incidence of local recurrence in the conserved breast. This was compared on an ITT basis (i.e. all randomised patients were analysed) and the log-rank test was used. This test allows for the hazards to be non-linear. This was performed once the baseline data had been compared to test the randomisation and to define whether or not any stratified analyses were required. In addition, ratios of radiological lesion size to clinical and pathological size, and ratios of specimen weights in the two arms of the trial, were compared to ensure that the extent of the surgical procedure was similar in both groups. The baseline comparisons between the two groups were carried out using the chi-square test for categorical variables and an independent two-sample t-test for all continuous variables. Statistical significance was defined at the p < 0.05 level in the first analysis and at the p < 0.01 level in the second analysis. We planned to use Kaplan–Meier curves and proportional hazard regression models to account for time to event and censoring of the data. Standard tests for non-inferiority were performed with the margin of non-inferiority set at 2.5% absolute difference in local recurrence at 5 years. We analysed the non-inferiority statistic by calculating the difference in binomial proportions of local recurrences in the conserved breast between the two randomised groups (TARGIT vs. EBRT).

All reports of local and regional recurrence and death were checked (before the data were unblinded, thus masking the randomised allocation), to ensure that they were correct.

To assess stability over time, we also calculated this statistic for the mature cohort (n = 2232), reported in 2010, and for the earliest cohort (excluding the last 4 years of enrolment; n = 1222), who had a median follow up of 5 years. We calculated the z-score and P_{non-inferiority} using established methods53–55 for the whole cohort and the two pre-specified strata – prepathology and postpathology.

Early complications were reported in 20101 and complications arising > 6 months after randomisation were reported in 2014. 2

To address the issue of follow-up, we charted the absolute differences in the 5-year Kaplan–Meier estimates of local recurrence in the conserved breast and overall mortality for patients with prepathology randomisation in the whole trial along with those for the mature cohort reported in 2010, who had a longer follow-up (median 3 years 8 months, maximum 12 years) and the earliest cohort.

A patient was deemed to have adequate follow-up if they had at least 5 years of follow-up or if they were seen within the year before database lock. Patients were censored when they were last seen or withdrawn from the trial. The database (customised Microsoft Access^® 1999 onwards; Microsoft Corporation, Redmond, WA, USA) as validated on 29 June 2012 was used for this analysis, with 1 June 2012 as a reference date. SAS (version 9.3; SAS Institute Inc., Cary, NC, USA), Microsoft Excel^® 2011, Stata (version 12.0; StataCorp LP, College Station, TX, USA) and IBM SPSS Statistics (version 20.0; IBM Corporation, Armonk, NY, USA) were used for data compilation, validation and analysis. Kaplan–Meier graphs were displayed as recommended by Pocock et al. 56 and a log-rank test was used to compare the difference between the survival function and to obtain p-values (significance level set at p < 0.01 for local recurrence and p < 0.05 for survival).

The 2010 analysis was prompted by the DMC at the March 2010 ISC meeting. The DMC felt that the data needed to be made public. The results of the first analyses were therefore submitted to an American Society of Clinical Oncology (ASCO) meeting and to The Lancet. The Lancet editors chose to put the manuscript through their fast-track rigorous review process and published it online1 on the same day as the ASCO presentation on 5 June 2010.

In 2010 it was also recommended by the DMC that the next analysis should be performed after 2 more years, in 2012. The final 2012 analysis was presented to the ISC and DMC in September 2012 and submitted with their recommendation to the San Antonio Breast Cancer Symposium (SABCS) as a late-breaking abstract, where it was accepted and presented on 6 December 2012. The manuscript was submitted to The Lancet in 2013 and published online in November 2013 and in print in February 2014. 2

Planned subgroup analysis

Although most patients recruited in the trial were good-prognosis patients, there was a substantial number of ‘high-risk’ cases, for example approximately 500 cases were node positive, or grade 3, because of the broad inclusion criteria. When the TARGIT approach is applied to normal clinical practice, patient selection will be crucial in ensuring that the results of day-to-day practice reflect the results obtained in the clinical trial.

Evidence from randomised trials and laboratory research suggested that progesterone receptor (PgR) status may be a predictive factor for local recurrence. The 2011 Oxford overview3 had found for the first time that the hormone receptor status of the tumour predicted benefit from radiotherapy. The proportional reduction in recurrence as a result of radiotherapy was higher in patients with ER-positive tumours (nearly two-thirds reduction from 22% to 8.7%) than in patients with ER-negative tumours (about one-third reduction from 43.8% to 28.9%). Furthermore, molecular analysis of tumours in the ELIOT study presented at the European Breast Cancer Conference EBCC-8 in March 2011 and subsequently published35 suggested that IORT was less effective in patients with non-luminal A (hormone receptor-negative) tumours. Evidence from Prat et al. 57 suggested that tumours that are PgR negative [even though ER positive and human epidermal growth factor receptor-2 (HER2) negative] should not be classified as luminal A as they may represent a more aggressive tumour type [IMProving care And Knowledge through Translational research (IMPAKT) meeting in Brussels, May 2012, and published later57]. Furthermore, three additional studies from Australia and the USA58–60 have suggested that PgR status is an independent predictor of local recurrence following radiotherapy, perhaps because PgR status is associated with more infiltrating and irregular margins and would warrant a wider radiation field. Hence, we considered testing whether or not PgR status was a predictive factor for local recurrence. Another reason for this was that most patients in the trial were ER positive (< 10% were ER negative) and there were very few events overall; therefore, an effect of ER status on local recurrence would not be discernible. On the other hand, PgR positivity is an expression of a functional ER receptor and, as not all ER-positive cases are PgR positive, many more cases would be PgR negative and this would make the analysis by PgR status more meaningful.

In view of these findings, we hypothesised that hormone sensitivity might be predictive of response to TARGIT and therefore, before the data were unblinded for this analysis, we planned to analyse whether or not the response to radiotherapy in the TARGIT-A trial was dependent on hormone receptor responsiveness, using PgR status as a marker.

We also tested whether or not patient age and certain other tumour factors, as defined below, were predictive of response or prognostic of outcome. This analysis of patient and tumour factors may help select patients in whom partial breast irradiation (PBI) approaches such as TaRGIT achieve the best results. These analyses may help refine current guidelines that were mainly based on a presumed risk of recurrence rather than effectiveness of radiation, for which only randomised evidence can be relied on.

A substantial number (18%) in each stratum were PgR negative (554 among all patients and 535 among those who had breast-conserving therapy). PgR status was unknown in only 88 patients among the 3104 patients in whom ER status was known.

As described above, we hypothesised that PgR negativity, as a surrogate marker of functional ER status, might be a predictor of poor response to radiotherapy. Analyses were performed by plotting Kaplan–Meier curves and tested for significance using the log-rank test and using a p-value of 0.05 as the boundary for statistical significance.

1. We first analysed whether or not PgR status affected the primary and secondary end points of local recurrence and survival (deaths from breast cancer and deaths from other causes) in the whole trial population.

2. As per the updated main results, TARGIT should ideally be used concurrently with the primary excision of the tumour, that is, as used in the prepathology stratum. Therefore, within the prepathology stratum we analysed the difference in local control and survival (deaths from breast cancer and deaths from other causes) between the TARGIT arm and the EBRT arm in PgR-negative and PgR-positive cases.

3. We also subsequently analysed in all patients whether or not certain patient or tumour characteristics (randomisation arm, timing of randomisation/delivery of TARGIT, age, tumour grade, presence of DCIS, margin positivity, ER, PgR and HER2 status, whether screen detected or not, presence of lymphovascular invasion, node positivity) influenced the effect of radiotherapy using the Cox proportional hazards model.

The effect of timing of randomisation and delivery of radiotherapy in relation to lumpectomy: the prepathology and postpathology strata

Figure 11 shows the Kaplan–Meier graphs for the prepathology and postpathology strata.

FIGURE 11.

Primary and secondary outcomes for the two strata defined by timing of randomisation and delivery of TARGIT: (a and c) prepathology (n = 2298) (randomised before lumpectomy and TARGIT given concurrently with lumpectomy); (b and d) postpathology (n = 1153) (randomised after lumpectomy and TARGIT given by reopening the wound). Kaplan–Meier plots for local recurrence in the conserved breast in patients who had breast-conserving surgery (a and b) and for deaths in all patients (c and d). Reproduced with permission from Vaidya et al. 2 Copyright © Vaidya et al. Open Access article distributed under the terms of CC BY-NC-ND.

In the prepathology stratum (n = 2298), that is, when TARGIT was delivered during the initial lumpectomy, the risk of local recurrence in the conserved breast was much the same for TARGIT as for EBRT [TARGIT 2.1% (95% CI 1.1% to 4.2%) vs. EBRT 1.1% (95% CI 0.5% to 2.5%); p = 0.31]. In total, 17 patients died from breast cancer in the TARGIT group and 15 patients died from breast cancer in the EBRT group [TARGIT 3.3% (95% CI 1.9% to 5.8%) vs. EBRT 2.7% (1.5% to 4.6%); p = 0.72]; the equivalent figures for non-breast-cancer mortality were 12 and 27, respectively [TARGIT 1.3% (95% CI 0.7% to 2.8%) vs. EBRT 4.4% (95% CI 2.8% to 6.9%); p = 0.016]. Thus, in absolute terms, there were four additional local recurrences but 13 fewer deaths in the prepathology stratum.

In the postpathology stratum (n = 1153), that is, when TARGIT was delivered as a delayed procedure by reopening the lumpectomy cavity, the difference between the two groups in local recurrence in the conserved breast was larger than 2.5% [TARGIT 5.4% (95% CI 3.0% to 9.7%) vs. EBRT 1.7% (95% CI 0.6% to 4.9%); p = 0.069]. In total, three patients died from breast cancer in the TARGIT group and one patient died from breast cancer in the EBRT group [TARGIT 1.2% (95% CI 0.4% to 4.2%) vs. EBRT 0.5% (95% CI 0.1% to 3.5%); p = 0.35]; the equivalent figures for non-breast-cancer mortality were five and eight, respectively [TARGIT 1.58% (95% CI 0.62% to 3.97%) vs. 1.76% (95% CI 0.7% to 4.4%); p = 0.32]. Thus, in absolute terms, there were eight additional local recurrences and one less death in the postpathology stratum.

Figures 12–14 show more detailed results for the prepathology stratum. Figure 12 shows the primary (local recurrence in the conserved breast) and secondary (deaths) outcomes. Figure 13 shows the differences in 5-year estimates for these outcomes for the whole cohort, the mature cohort and the earliest cohort. This figure demonstrates the stability of the results with longer follow-up and the trade-offs between the two outcomes. Figure 14 shows the 10-year Kaplan–Meier graph for the prepathology stratum, which shows that the small difference in local recurrence does not increase with longer follow-up.

FIGURE 12.

Primary and secondary outcomes for the prepathology stratum (n = 2298). Kaplan–Meier plots for (a) local recurrence in the conserved breast in patients who had breast-conserving surgery; (b) breast cancer deaths in all patients; and (c) non-breast-cancer deaths in all patients. Reproduced with permission from Vaidya et al. 2 Copyright © Vaidya et al. Open Access article distributed under the terms of CC BY-NC-ND.

FIGURE 13.

Absolute differences in 5-year Kaplan–Meier estimates of local recurrence in the conserved breast and overall mortality (TARGIT – EBRT) for the prepathology stratum in the whole cohort, the mature cohort and the earliest cohort. All patients in the earlier cohorts are included in later cohorts. Median follow-up for the whole cohort was 2 years 4 months, median follow-up for the mature cohort was 3 years 8 months and median follow-up for the earliest cohort was 5 years. Reproduced with permission from Vaidya et al. 2 Copyright © Vaidya et al. Open Access article distributed under the terms of CC BY-NC-ND.

FIGURE 14.

Prepathology stratum: Kaplan–Meier plot of local recurrence in the conserved breast up to 12 years.

One possible concern might be that the addition of EBRT was different between the two strata. If prespecified unsuspected adverse factors were found when the full pathology report was available, whole-breast radiotherapy was recommended. This was more common in the prepathology stratum than in the postpathology stratum (21% vs. 3%). To assess whether or not more frequent use of additional EBRT could be the cause of better outcomes in the prepathology stratum, we performed an exploratory non-randomised comparison of tumour characteristics and primary and secondary outcomes between prepathology and postpathology patients who received TARGIT alone.

Patients in the super-selected postpathology stratum had a much better prognosis than patients in the prepathology stratum, as seen from the tumour size, tumour grade, lymph node status and breast cancer survival (Table 11). Despite this, the 5-year local control in prepathology cases who received TARGIT alone appears to be better than that in the postpathology cases who received TARGIT alone (2.7% vs. 5.9%) and similar to that in the whole prepathology stratum (2.1%).

These data suggest that TARGIT during lumpectomy is effective and that the accurate and timely application of radiation to the fresh tumour bed avoiding spatial and temporal miss rather than the addition of EBRT in higher-risk cases is responsible for the better results in the prepathology stratum.

Analysis of the earliest cohort of over 800 patients randomised in the first 8 years in the prepathology stratum of the TARGIT-A trial, and who have a median follow-up of 5 years

We paid special attention to the earliest cohort of over 800 patients randomised in the first 8 years in the prepathology stratum of the TARGIT-A trial. This cohort had a median follow-up of 5.01 years. As well as being a conventional yardstick, this 5-year point happens to be considerably in excess of the known peak hazard of local recurrence. In addition, the effect of radiotherapy on local recurrence is limited to the first 5 years. From the results of the main analysis it was determined that giving TARGIT during lumpectomy should be the preferred option and hence we analysed these earliest patients.

This section provides details of the analysis for this cohort of patients who were randomised in the first 8 years of the TARGIT-A trial, that is, from 2000 to 2008. All patients randomised in this period in the prepathology stratum were included. These patients happen to have the longest follow-up and have not been ‘cherry-picked’.

In total, there were 817 patients in the prepathology stratum of the earliest cohort. For local recurrence in the conserved breast, TARGIT was non-inferior to EBRT (P_{non-inferiority} = 0.0091). The Kaplan–Meier plots are shown in Figure 15.

FIGURE 15.

Kaplan–Meier plots for the earliest cohort of patients randomised in the first 8 years of the trial: (a) local recurrence in the conserved breast; (b) breast cancer deaths; and (c) non-breast-cancer deaths.

The 5-year estimated risk was not statistically different between groups for local recurrence [TARGIT 1.8% (95% CI 0.84% to 4.2%) vs. EBRT 0.84% (95% CI 0.3% to 2.6%); p = 0.32] or deaths from breast cancer [TARGIT 3.9% (95% CI 2.3% to 6.7%) vs. EBRT 3.0% (95% CI 1.7% to 5.4%); p = 0.34]. However, there were significantly fewer deaths from causes other than breast cancer with TARGIT [1.9% (95% CI 0.9 to 3.9)] than with EBRT [5.1% (95% CI 3.2% to 8.0%); p = 0.04.]

It should be noted that the number needed to prove non-inferiority was calculated to be 5851 and, therefore, this earliest cohort of 817 patients had enough power to draw reliable conclusions.

From this analysis it is clear that, if the trial had been stopped after 8 years of recruitment, there would have been an adequate number of patients for testing the hypothesis with a median follow-up of 5 years. The patients who were recruited more recently should not be perceived as weakening the result because the Kaplan–Meier analysis as well as the cohort analysis for non-inferiority (see Table 8) takes it all into account.

Effect of progesterone receptor status

Figure 20 shows the Kaplan–Meier plots, 5-year risks with CIs and significance levels for all patients, that is, both prepathology and postpathology strata combined.

FIGURE 20.

Local recurrence (a and b) and overall mortality (c and d) for the pre-specified subgroups defined by PgR status.

Among patients who had PgR-positive tumours, there was no significant difference in the primary outcome of local recurrence in the conserved breast at 5 years between TARGIT and EBRT [13/1204 vs. 10/1203, 5-year risk 2.3% (95% CI 1.3% to 4.3%) vs. 1.5% (95% CI 0.75% to 3.0%); p = 0.51] and there were significantly fewer deaths with TARGIT than with EBRT [22/1232 vs. 37/1230, 5-year risk 2.7% (95% CI 1.6% to 4.5%) vs. 4.9% (95% CI 3.3% to 7.1%); p = 0.0487].

Among patients who had PgR-negative tumours, local recurrence in the conserved breast was higher in the TARGIT arm than in the EBRT arm [9/276 vs. 1/259, 5-year risk 7.0% (95% CI 3.5% to 13.6%) vs. 0.5% (95% CI 0.1% to 3.7%); p = 0.017]. Overall mortality was similar between the groups [13/289 vs. 11/265, 5-year risk 8.1% (95% CI 4.4% to 14.7%) vs. 6.3% (95% CI 3.2% to 12.3%)].

Effect of progesterone receptor status in the prepathology stratum when targeted intraoperative radiotherapy was given concurrently with lumpectomy

Figure 21 shows the Kaplan–Meier plots, 5-year risks with CIs and significance levels for the prepathology stratum. Figure 22 shows the local recurrence and overall mortality in cohorts of patients with increasing follow-up periods. This demonstrates that the results remain stable with longer follow-up periods, including in the earliest cohort of 636 patients with a median follow-up of 5 years.

FIGURE 21.

Effect of PgR status on primary and secondary outcomes in the prepathology stratum (TARGIT given concurrently with lumpectomy): (a)–(c) PgR positive; (d)–(f) PgR negative. (a) Local recurrence in the conserved breast – events: TARGIT n = 4, EBRT n = 5; (b) deaths from breast cancer – events: TARGIT n = 8, EBRT n = 9; (c) deaths from other causes – events: TARGIT n = 10, EBRT n = 22; (d) local recurrence in the conserved breast – events: TARGIT n = 5, EBRT n = 1; (e) deaths from breast cancer – events: TARGIT n = 8, EBRT n = 5; and (f) deaths from other causes – events: TARGIT n = 2, EBRT n = 5.

FIGURE 22.

Local recurrence in the conserved breast and overall mortality for PgR-positive cases in the prepathology stratum.

When TARGIT was given concurrently in patients who had PgR-positive tumours (n = 1625), local recurrence in the conserved breast [1.4% (95% CI 0.5% to 3.9%) vs. 1.2% (95% CI 0.5% to 2.9%)] and breast cancer mortality [1.78% (95% CI 0.7% to 4.4%) vs. 1.98% (95% CI 0.94% to 4.2%)] were similar for TARGIT and EBRT and mortality from other causes was significantly lower by 2.9% in the TARGIT arm [1.6% (95% CI 0.74% to 3.4%) vs. 4.5% (95% CI 2.8% to 7.3%); p = 0.04], leading to a 3.1% reduction in overall mortality [3.3% (95% CI 1.8% to 6.0%) vs. 6.4% (95% CI 4.3% to 9.6%); p = 0.083].

For patients with PgR-negative tumours (n = 366), local recurrence in the conserved breast was numerically, but not statistically, higher in the TARGIT arm than in the EBRT arm. There was no difference between the two arms in breast cancer mortality, non-breast-cancer mortality or overall mortality.

Thus, when TARGIT is delivered at the time of primary surgery, the chance of remaining free from local recurrence at 5 years possibly drops from 99% to 98% (p = 0.31), but the chance of staying alive possibly increases from 93.1% to 95.4% (p = 0.12). The p-value expresses the probability of the difference we found being observed if there was no real difference between the two treatments. Alternatively, one can say that there is no difference in local recurrence or mortality, with a large benefit of having the treatment in one sitting.

Cox model

The following factors were entered into the Cox model for the primary outcome of local recurrence in the conserved breast: randomisation arm, timing of randomisation/delivery of TARGIT, age, whether screen detected or not, tumour grade, presence of DCIS, margin positivity, ER, PgR and HER2 status, lymphovascular invasion and node positivity.

We found that age, whether screen detected or not, tumour grade, ER status, HER2 status, lymphovascular invasion and nodal status did not appear to have any influence on local recurrence, whereas timing of randomisation and delivery of TARGIT [hazard ratio (HR) for delayed TARGIT 2.26, 95% CI 1.07 to 4.78; p = 0.033] were significant predictors of local recurrence. PgR status was a significant predictor of local recurrence overall (HR for PgR negative 2.46, 95% CI 1.11 to 5.47; p = 0.027) and in the TARGIT arm (HR for PgR negative 3.14, 95% CI 1.29 to 7.61; p = 0.011). Margin status was significant overall (HR for positive first margin 2.68, 95% CI 1.08 to 6.68; p = 0.034) and in the EBRT arm (HR for positive first margin 5.17, 95% CI 1.50 to 17.8; p = 0.009), but not in the TARGIT arm. Fitting a separate regression model is equivalent to fitting an interaction of treatment arm with all of the independent risk factors in the model and also assumes that the variance in both treatment groups is similar. We also fitted a full model with a treatment by PgR status interaction, giving an overall Wald statistic p-value of 0.08.

For those patients who received only TARGIT during lumpectomy, multivariate analysis showed that only PgR status (HR for PgR negative 5.57, 95% CI 1.23 to 25.21; p = 0.026) remained significant. Other factors, such as younger age, grade (grade 3), tumour size > 2 cm or node positivity, did not influence the difference between the two randomised arms: the 5-year rates of local recurrence in the conserved breast were as follows: for 263 patients with age ≤ 50 years [events 3/145 vs. 3/118, TARGIT 2.9% (95% CI 0.94% to 8.64%) vs. EBRT 6.2% (95% CI 1.99% to 18.47%); p = 0.70]; for 436 grade 3 cancers [events 2/20 vs. 2/216, TARGIT 1.5% (95% CI 0.37% to 5.87%) vs. EBRT 1.7% (95% CI 0.42% to 6.45%); p = 0.98]; for the 363 patients with tumour ≥ 2 cm [events 1/173 vs. 1/190, TARGIT 0.75% (95% CI 0.1% to 5.2%) vs. EBRT 0.86% (95% CI 0.1% to 6.0%); p = 0.95]; for the 472 node positive patients [events 1/245 vs. 2/227, TARGIT 0.72% (95% CI 0.1% to 5.0%) vs. EBRT 1.3% (95% CI 0.3% to 5.2%); p = 0.56].

Forest plot

The forest plot in Figure 23 shows that TARGIT was non-inferior to EBRT in every subgroup apart from the ER-negative and PgR-negative subgroups.

FIGURE 23.

Forest plot showing the non-inferiority statistic for each of the subgroups. The green shaded area is the non-inferiority margin and each of the green squares is the local recurrence rate in terms of binomial proportions along with the 90% CI. From this plot it appears that only ER and PgR status affects whether or not there is a difference between TARGIT and EBRT. LVI, lymphovascular invasion.

Background

In 1996 we reported that 63% of specimens of mastectomy performed for a unifocal cancer harbour other cancer foci, with 80% of these foci in other quadrants. 13 In contrast, local recurrence after a lumpectomy occurs mainly at the site of the original tumour. Therefore, we hypothesised that the cancer foci in other quadrants remain dormant even in the absence of radiation treatment to the whole breast. This academic insight led us to develop the TARGIT technique and the INTRABEAM device. In the TARGIT-A randomised trial (n = 3451) we compared risk-adapted TARGIT vs. whole-breast radiotherapy. 2

Method

Randomisation occurred either before surgery (prepathology stratum; TARGIT given during lumpectomy) or after surgery (TARGIT given as a delayed procedure). In the prepathology stratum we assessed the number of local recurrences that occurred in other quadrants in the patients who received only TARGIT and in the patients who received EBRT. We also estimated the number of cancer foci in quadrants other than that with the original tumour.

Results

In total, 94.4% of cases in the TARGIT-A trial did not undergo a preoperative MRI scan. A total of 793 patients in the prepathology stratum randomised to TARGIT had only TARGIT as their radiotherapy. With 2098 women-years of follow-up, there were nine recurrences in the conserved breast. The 5-year local recurrence rate in those who received TARGIT alone was 2.7% (95% CI 1.3% to 5.5%), which was not different from that in the whole prepathology cohort randomised to TARGIT [2.1% (95% CI 1.1% to 4.2%)]. In these 793 patients, one would expect 63% (i.e. 500) to have additional foci of cancer in their breasts and 80% of these (i.e. 400) should be in quadrants other than the index quadrant. In reality, after 2098 women-years of follow-up, seven patients had recurrence in the scar, six had new contralateral cancers and two had cancers growing in other quadrants, implying that the remaining 398 foci had remained dormant. Among 935 patients who received whole-breast radiotherapy the same number of cancers (n = 2) grew in other quadrants and there were five new contralateral cancers. In the randomised comparison, there were two local recurrences in other quadrants in each of the randomised arms of the trial.

Conclusion

Cancer foci in breast that are away from the site of the primary tumour remain dormant and behave no differently from those in the contralateral breast. They also appear to be unaffected by whole-breast radiotherapy. This analysis provides level 1 evidence supporting PBI.

Background

The Z–11 trial68 found that, even when axillary clearance is not performed after finding one or two positive sentinel nodes, it does not affect local control, despite 23% of patients in the axillary clearance (control) arm (and, therefore, those in the experimental arm as well) having positive nodes. However, every patient in this trial received whole-breast radiotherapy and it has been suggested that inadvertent non-therapeutic irradiation of the lower axilla that occurs with tangential fields of conventional whole-breast radiotherapy could be responsible for controlling the growth of the minimal residual axillary disease. Therefore, questions are being raised whether following the concept of sentinel node biopsy or not dissecting the axilla after one to two positive nodes is applicable to patients receiving PBI.

Methods

We compared the risk of axillary recurrence in patients with negative sentinel node biopsy and those with one to two positive nodes as per treatment received – whether they received EBRT or TARGIT only within the updated TARGIT-A trial. 2 It is important to note that this was a comparison as per treatment received rather than as per randomised allocation (Figure 24).

FIGURE 24.

Comparison of the two non-randomised groups for the outcome of local recurrence, with or without EBRT.

Results

Overall, 3375 patients underwent breast-conserving surgery. In total, 1222 patients had a median follow-up of 5 years and all patients had a median follow-up of 2 years 5 months, giving 9491 women-years of follow-up.

The trial patients had a generally good prognosis but a substantial number (> 1200) were aged ≤ 60 years and > 500 were node positive and/or grade 3.

In total, 91% of patients had a sentinel node biopsy, 90% of whom had < 10 nodes removed if this was negative. Eleven patients had axillary recurrence, one of whom had previously undergone axillary clearance with the other 10 having a negative sentinel node biopsy. The risk of axillary recurrence was similar whether the patients received EBRT [6/1762, 5-year risk 0.82% (95% CI 0.34% to 2.02%)] or not [5/1613, 5-year risk 0.68% (95% CI 0.28% to 1.6%); HR 0.84 (95% CI 0.26 to 2.74); p = 0.8]. The results were similar if only patients with one or two positive axillary nodes were included (EBRT 1/255 vs. no EBRT 0/127) (Figures 25 and 26).

FIGURE 25.

Numbers of patients and axillary recurrences in the EBRT and no EBRT groups (non-randomised comparison).

FIGURE 26.

Kaplan–Meier plot for axillary recurrence in patients who were given EBRT and patients who were not given EBRT.

Conclusion

Omitting whole-breast radiotherapy after a sentinel node biopsy in this good-prognosis population was not associated with an increased axillary recurrence rate.

Background

The TARGIT-A randomised trial tested the outcomes of breast cancer treatment between traditional EBRT given over several weeks versus a risk-adapted approach using single-dose TARGIT. In the prepathology stratum TARGIT was given concurrently with lumpectomy. If on postoperative histopathological examination of the operative specimen any unsuspected high-risk factors were found, EBRT was added in 15–20% of cases. The main results showed that for the for local control, compared with EBRT, TARGIT achieved non-inferior local control in all cases and identical local control when given concurrently with lumpectomy in ER-positive, PgR-positive cases. Interestingly, the patients in the TARGIT arm had a significantly lower risk of dying from non-breast-cancer causes such as heart attacks and other cancers. We investigated whether or not this difference could be fully explained by the avoidance of the effects of EBRT on the heart and other organs in the TARGIT arm.

These data were first presented at the March 2013 St Gallen Symposium (poster presentation) and in September 2013 at the ASTRO Annual Meeting (poster presentation). 69,70

Methods

1. We compared cardiac deaths for left- and right-sided breast cancer.

2. We estimated cardiac deaths based on age, sex and follow-up.

3. We compared survival between the two randomised arms in the prepathology stratum, but limited to those who had received EBRT. Therefore, the control arm included patients who were allocated EBRT and the experimental arm included patients who were allocated TARGIT and received EBRT in addition; because both groups received EBRT, any difference that was found between the groups would be attributable to TARGIT (Figure 27).

FIGURE 27.

Non-randomised comparison between those who received TARGIT + EBRT and those who received EBRT for deaths from non-breast-cancer causes: (a) schema of the analysed cohorts of patients; and (b) Kaplan–Meier plot depicting deaths from causes other than breast cancer in prepathology patients who received EBRT – non-randomised comparison.

Results

In the whole trial there was a highly significant reduction in non-breast-cancer mortality in the TARGIT arm (HR 0.47; p = 0.0086):

1. There were two cardiac deaths in the TARGIT arm and eight in the EBRT arm, with similar numbers in each arm for left- and right-sided breast cancer (1/775 vs. 4/795 for left; 1/776 vs. 4/746 right).

2. Among the 1730 patients randomised to receive EBRT, the number of cardiac deaths (n = 8) was not higher than the 12 estimated based on age, sex and follow-up period.

3. There were no deaths from non-breast-cancer causes in the TARGIT + EBRT group compared with 24 in the EBRT group. Thus, there were significantly fewer deaths in those patients who received TARGIT (0/218 vs. 24/892; log-rank p = 0.012) (see Figure 27).

Conclusion

One might expect irradiation for left-sided cancers to result in higher cardiac toxic effects. However, the ratio of cardiac risk of left- to right-sided cancers is small (1.34 according to Darby et al. 71). Furthermore, Darby et al. 71 recorded no significant effect of laterality on cardiac toxic effects per Gy. With modern radiotherapy designed to reduce the cardiac dose, the absolute difference in cardiac deaths between those with left-sided breast cancer and those with right-sided breast cancer is likely to be even lower and undetectable with few events. Note that an absence of an effect of laterality on cardiac deaths cannot be used to imply that there is no cardiac toxicity, because the difference in the effect between the two sides is not likely to be large. Finally, with eight cardiac deaths a ratio of 1.34 would not be detectable (e.g. it would need to be distributed as 3.5 vs. 4.7, with even 3 vs. 5 already being a ratio of 1.7). Therefore, an absence of a difference between the left and the right sides should not be interpreted as an absence of cardiac toxicity.

A risk similar to that in the age-matched population should be interpreted with the knowledge that patients in clinical trials are generally in better general health than the normal population and we should therefore have expected a lower risk of cardiac mortality in the trial population.

The most interesting finding was that there were significantly fewer deaths among patients who were allocated to TARGIT + EBRT compared with those allocated to EBRT. In this comparison the only difference between the two groups was the delivery of TARGIT in one group, because both groups received EBRT, albeit with a slightly higher dose – the boost – in the EBRT group. It suggests that the difference in non-breast-cancer mortality could arise from an increase in mortality from EBRT toxicity, as one would think conventionally, as well as a reduction in mortality because of a potential beneficial effect of TARGIT. The latter is a radically new hypothesis that the local effect of TARGIT on the tumour bed by inhibiting cancer-stimulating cytokines17 may spill over to reduce the systemic inflammatory response to trauma and have significant long-term systemic beneficial effects that might be protective against cardiac and cancer mortality (Figure 28). This remarkable finding, albeit a non-randomised comparison, could potentially have far-reaching implications. It is fortuitous that it will be tested in the HTA programme-funded TARGeted Intraoperative radioTherapy Boost (TARGIT-B) randomised trial (ISRCTN43138042), which will compare TARGIT boost with EBRT boost, with both randomised arms receiving whole-breast radiotherapy (Figure 29).

FIGURE 28.

Targeted intraoperative radiotherapy abrogates the tumour-stimulating effect of surgical wound fluid. AgRP, Agouti-related protein; EGFR, epidermal growth factor receptor; FAS/TNFRSF6, FAS receptor or tumour necrosis factor receptor superfamily member 6; FGF-4, fibroblast growth factor 4; Flt-3, Fms-like tyrosine kinase 3; G-CSF, granulocyte colony-stimulating factor; GRO, growth-regulating oncogene; HGF, hepatocyte growth factor; IGFBP-6, insulin-like growth factor-binding protein 6; IL, interleukin; MCP-1/-2, monocyte chemoattractant protein 1/2; Mip-1a/1d, macrophage inflammatory protein 1a/1d; PDGF-BB, platelet-derived growth factor BB; RANTES, regulated on activation, normal T cell expressed and secreted; sTNFR-I/-II, soluble tumour necrosis factor receptor I/II; Tie-1/-2, tyrosine kinase containing immunoglobulin and epidermal growth factor homology domain-1 and 2; uPAR, urokinase-type plasminogen activator receptor; VEGF-R3, vascular endothelial growth factor receptor 3.

FIGURE 29.

Could TARGIT be protective against heart attacks and deaths from other cancers? This hypothesis will be tested in the TARGIT-B trial.

Background

In the TARGIT-A trial protocol,36 a positive margin leads to a recommendation of additional EBRT after TARGIT. The core protocol defined 1 mm as an adequate tumour-free margin but individual centres were allowed to pre-specify a wider margin. In all German centres EBRT was recommended after TARGIT if the margin was < 10 mm. We analysed the impact of such a decision on treatment received and local recurrence.

Methods

All German cases were planned to be treated in the prepathology stratum, that is, randomisation before excision of the tumour and TARGIT delivered during lumpectomy.

Therefore, we analysed in the prepathology stratum the proportion of cases in whom additional EBRT was actually given after TARGIT, as well as the primary (local recurrence in the conserved breast) and secondary (breast cancer and non-breast-cancer mortality) outcomes. We compared these values for the German cohort with the equivalent values for the rest of the trial population in a post hoc exploratory non-randomised analysis.

Results

Additional EBRT was given in nearly twice the number of patients in the TARGIT arm in the German centres as in the TARGIT arm in the rest of the trial population [31.3% (96/306) vs. 17.4% (123/706)]. However, the 5-year local recurrence rate in the German cohort was the same as in rest of the trial population [German 2.6% (95% CI 0.87% to 7.8%) vs. rest of the trial population 1.9% (95% CI 0.81% to 4.5%)]. Furthermore, there was no significant difference between the German cohort and rest of the trial population in breast cancer mortality [2.5% (95% CI 0.64% to 9.7%) vs. 3.3% (95% CI 1.7% to 6.4%)] or non-breast-cancer mortality [0.78% (95% CI 0.11% to 5.4%) vs. 1.7% (0.8% to 3.7%)].

Conclusion

The greater use of additional EBRT as seen in the German cohort prompted by the higher limit of a 10-mm tumour-free margin may not be necessary and a policy of adding EBRT after TARGIT when the margin is < 1 mm appears to be appropriate.

Patients

The analysis was based on costs and outcomes for the 817 patients randomised in the ‘earliest cohort’ in the prepathology stratum of the TARGIT-A trial. Several issues were considered when deciding which cohort of patients to include in the cost–utility analysis:

1. We did not include the postpathology stratum from the earliest cohort because the results in this group were less favourable than those of the prepathology stratum. Hence, it is highly unlikely that TARGIT would be adopted in clinical practice for this group. As patients from this stratum were not included in the analysis, the results cannot be applied to this group.

2. The number of participants needed to prove non-inferiority was calculated to be 585 and therefore the earliest cohort of 817 patients had enough power to draw reliable conclusions.

3. The earliest cohort was randomised between 2000 and 2008 and the average follow-up was 5 years, permitting a reasonable follow-up period without a large number of missing data. The complete prepathology stratum from TARGIT-A consisted of 2298 patients, with an average follow-up of 2 years 4 months. Hence, by including the full cohort we would have substantially increased the proportion of missing data in the sample if we wanted to use a 5-year time horizon or we would have had to use a shorter time horizon.

We therefore balanced the number of patients in the whole cohort compared with the number in the earliest cohort against the duration of follow-up in the two cohorts against the fact that the earliest cohort had enough statistical power to draw reliable conclusions and decided to base our analysis on the 817 patients randomised in the earliest cohort of the TARGIT-A trial in the prepathology stratum. In this cohort, as in the mature cohort in the prepathology stratum and all patients in the prepathology stratum, TARGIT was non-inferior to EBRT with respect to local recurrence and the 5-year estimated risks of local recurrence were not statistically different between the treatment groups.

Overview of the cost–utility analysis

We undertook a cost–utility analysis to compare the costs and outcomes associated with TARGIT compared with EBRT in the prepathology stratum of the TARGIT-A trial. The outcome measure was QALYs, which combine length of life and quality of life, consistent with NICE guidelines. 75 Cost-effectiveness was expressed as incremental net monetary benefits. 75 The analysis took a UK NHS and personal social services (PSS) perspective. 75 Resource use data were included from all participating centres and UK unit costs were applied. Costs are presented in 2013/14 UK pounds. The time horizon was 5 years, reflecting the average follow-up in the earliest cohort in the prepathology stratum of the TARGIT-A trial. Extrapolation beyond the end of the trial using decision-analytical modelling was not undertaken because the within-trial analysis found no evidence of significant differences in QALYs between the groups. This probably reflects the main finding from the TARGIT-A trial that TARGIT was non-inferior to EBRT with regard to local recurrence. Although there was some evidence of differences in costs, these differences were accrued during the first year, with no evidence of significant differences in costs beyond the first year. Hence, the 5-year time horizon was long enough to reflect all important differences in costs or outcomes between the two treatments. An annual discount rate of 3.5% was applied to costs and outcomes. 75

Resource use and costs

Utilities and quality-adjusted life-years

The outcome measure in our cost–utility analysis was QALYs, which combine length of life and quality of life, the latter being measured by utility scores. A utility score of 1 represents full health and a score of 0 denotes death; negative values represent states worse than death.

Utility data were not collected in the TARGIT-A trial. Patient-level data on the timing of events were collected and for every patient we created a data set describing the health state that they were in during every day of the 5-year time horizon. Utility values from published sources were then applied to each health state. These were used to construct five 1-year utility profiles for every patient covering the 5-year time horizon. QALYs for every patient for each year were calculated as the area under the utility profile for that year. These were discounted and summed across all 5 years to calculate QALYs per patient over the whole period.

The health states included in the cost–utility analysis were recurrence free, local recurrence, distant recurrence, breast cancer death and non-breast-cancer death. A review of the Cost-Effectiveness Analysis Registry86 was undertaken using the search term ‘breast cancer’ to identify studies reporting relevant utility scores and 1291 results (utility scores) were identified. Picot et al. 72 recently undertook an extensive literature search of studies providing utility values for such patients and identified nine suitable studies. The criteria for the values that they selected in their analysis were that they would ideally be based on EQ-5D scores, would ideally have been derived from UK patients and these patients would ideally reflect the younger age range of patients in the TARGIT-A trial. The values that they selected, from studies by Turnbull et al. 87 and Lidgren et al. ,88 were as follows:

• recurrence free in first year: 0.7728

• recurrence free after first year: 0.8112

• local recurrence: 0.8112

• recurrence free after local recurrence: 0.8112

• distant recurrence: 0.658.

We used these values in our base case. The values imply that the utility associated with local recurrence is the same as the utility associated with being recurrence free after the first year and the utility associated with being recurrence free after local recurrence. We undertook a sensitivity analysis using values from an alternative study by Hayman et al. ,89,90 which have been used in previous studies, as follows:

• recurrence free: 0.92

• local recurrence: 0.87

• recurrence free after local recurrence: 0.92

• distant recurrence: 0.70.

Patients who died in the TARGIT-A trial (either from breast cancer or from other causes) were assigned a utility value of 0 at their date of death until the end of the 5-year time horizon.

In the cost–utility analysis we did not incorporate utility losses associated with additional procedures, chemotherapy, mastectomy or complications. Given the low incidence of these events, that they were evenly distributed between treatment groups and that the time period affected is likely to be short, this is unlikely to affect the QALYs associated with each treatment group. We also did not include any utility losses associated with EBRT. Therefore, this would make our estimates more conservative because such an omission would work against TARGIT.

Missing data

There were some missing data on patient follow-up, meaning that for some patients we did not know whether or not they had experienced events. This affected both the total costs incurred by each patient and the total QALYs. Multiple imputation was used to impute missing data separately for costs in years 1–5, total costs, QALYs in years 1–5 and total QALYs. The following variables were included in the imputation models as additional explanatory variables: cost of EBRT, cost of the index procedure, cost of additional procedures, cost of chemotherapy, cost of mastectomy, whether or not the patient had each type of complication, age at randomisation, tumour size in millimetres at randomisation, ER status at randomisation, PgR status at randomisation, contralateral cancer or not, whether the cancer was screen detected or not, study centre, year of randomisation and treatment allocation. We used multivariate normal regression to impute missing values and generated 20 imputed data sets. We repeated the multiple imputation several times using different random number seeds to investigate whether or not the conclusions of the analysis changed.

Statistical methods

Mean costs, outcomes and net monetary benefits were compared between all patients randomly assigned to EBRT and TARGIT, irrespective of which treatment was administered and whether or not patients received additional therapies of either type. We calculated differences in mean costs and QALYs and incremental net monetary benefits between groups. Net monetary benefits for EBRT and TARGiT were calculated as the mean QALYs per patient multiplied by the maximum willingness to pay for a QALY minus the mean cost per patient. Incremental net monetary benefits were calculated as the difference in mean QALYs per patient with TARGIT compared with EBRT multiplied by the maximum willingness to pay for a QALY minus the difference in mean costs per patient. We used the cost-effectiveness threshold range recommended by NICE of £20,000–30,00075 as the lower and upper limits of the maximum willingness to pay for a QALY. If the incremental net monetary benefit is positive (negative) then TARGIT (EBRT) is preferred on cost-effectiveness grounds. The QALYs gained and incremental costs were adjusted for age at randomisation, tumour size in millimetres at randomisation, ER status at randomisation, PgR status at randomisation, contralateral cancer or not, whether the cancer was screen detected or not, study centre and year of randomisation. For each of the 20 imputed data sets we ran 1000 bootstrap replications and combined the results using equations described by Briggs et al. 91 to calculate standard errors (SEs) around mean values accounting for uncertainty in the imputed values, the skewed nature of the cost data and utility values and sampling variation. SEs were used to calculate 95% CIs around point estimates. A similar analytical approach has been used previously. 92

Sensitivity analyses

We undertook deterministic sensitivity analyses to evaluate the impact of uncertainty in the following components. In each case the changes made were applied one at a time to the base case.

• No adjustment for age at randomisation, tumour size in millimetres at randomisation, ER status at randomisation, PgR status at randomisation, contralateral cancer or not, whether the cancer was screen detected or not, study centre and year of randomisation.

• Complete case analysis without imputing missing values.

• Complete case analysis without imputing missing values plus with no adjustment for age at randomisation, tumour size in millimetres at randomisation, ER status at randomisation, PgR status at randomisation, contralateral cancer or not, whether the cancer was screen detected or not, study centre and year of randomisation.

• EBRT costs based on number of fractions received in the trial [mean (SD) number of fractions administered per patient who received EBRT in the trial was 23 (5)].

• No EBRT boost.

• Costs of EBRT per fraction of £101 and £154, based on the lower and upper values of the IQR of the NHS reference costs. 76

• Costs of TARGIT of £1300, £1500, £1700, £1900, £2100, £2300, £2500 and £2700. The value of £1300 corresponds to the minimum value in Picot et al. ,72 in which the capital and one-off costs were annualised using a device lifetime of 10 years and these costs and the annual costs were assigned to individual treatments assuming that each device was used to undertake 631 procedures per year. The value of £2500 corresponds to the maximum value in Picot et al. ,72 with a device lifetime of 5 years and 100 procedures per year.

• Percentage of patients using NHS transport for EBRT of 0% (no patients use NHS transport) and 30%.

• Health states valued using utilities from Hayman et al. 89,90

A cost-effectiveness acceptability curve93 showing the probability that TARGIT was cost-effective compared with EBRT at a range of values for the maximum willingness to pay for a QALY was generated based on the proportion of the bootstrap replications across all 20 imputed data sets with positive incremental net monetary benefits. 94 The probability that TARGIT was cost-effective at a maximum willingness to pay for a QALY of £20,000 and £30,000 was reported, based on the proportion of bootstrap replications with positive incremental net monetary benefits at these values.

Resource use and costs

In total, 15.2% of patients randomised to TARGIT also received EBRT (Table 13). We assumed that every patient receiving EBRT received 15 fractions. In total, 38% of patients randomised to EBRT also received an EBRT boost [mean (SD) 5 (2) fractions]. We assumed that 13.5% of all EBRT patients used NHS transport to travel to hospital for their radiotherapy treatment. The mean (median) number of nights in hospital for the initial procedure was 4 (3) for both TARGIT and EBRT patients. A total of 19% of EBRT patients received additional procedures, compared with 12% of TARGIT patients. In total, 20% of EBRT patients received chemotherapy and 4% had a mastectomy; for TARGIT the figures were 23% and 3%, respectively. The incidence of complications was low in both treatment groups. The number of events for TARGIT and EBRT were local recurrences (6 vs. 3), distant recurrences (21 vs. 18), breast cancer deaths (13 vs. 11) and non-breast-cancer deaths (7 vs. 18).

Accounting for missing data using multiple imputation, mean total costs per patient (95% CI) were £11,840 (£11,422 to £12,259) in the EBRT group (n = 416) and £11,404 (£10,800 to £12,008) in the TARGIT group (n = 401; Table 14). The mean radiotherapy cost per patient (summing the cost of TARGIT plus EBRT plus EBRT boost plus NHS transport for EBRT) was £3373 in the EBRT group and £2307 in the TARGIT group. Other costs were similar for EBRT and TARGIT. Values were similar for complete cases (Table 15).

Quality-adjusted life-years

Accounting for missing data using multiple imputation, mean QALYs per year were similar for the two groups and there was a decline over time. Mean QALYs per patient (95% CI) fell from 0.810 (0.808 to 0.812) in the EBRT group in year 1 to 0.657 (0.640 to 0.674) in year 5. In the TARGIT group the values were 0.811 (0.810 to 0.811) and 0.674 (0.660 to 0.689), respectively. Mean total QALYs per patient over the 5-year period were 3.663 (3.614 to 3.713) in the EBRT group and 3.704 (3.664 to 3.744) in the TARGIT group (see Table 14). QALYs were similar for complete cases (see Table 15).

Cost–utility analysis

Accounting for missing data using multiple imputation, the mean net monetary benefits for EBRT and TARGIT were £61,426 (95% CI £60,299 to £62,544) and £62,678 (95% CI £61,542 to £63,762) at a maximum willingness to pay for a QALY of £20,000 and £98,059 (95% CI £96,470 to £99,644) and £99,720 (95% CI £98,228 to £101,147) at a maximum willingness to pay for a QALY of £30,000 (see Table 14).

In the base-case analysis TARGIT was less costly than EBRT (mean incremental cost –£685) and produced slightly more QALYs than EBRT (mean QALYs gained 0.034; Table 16). The difference in costs between the two groups was statistically significant (mean incremental cost for TARGIT vs. EBRT –£685, 95% CI –£1131 to –£63) but the difference in QALYs was not (mean QALYs gained 0.034, 95% CI –0.026 to 0.095). The incremental net monetary benefit for TARGIT compared with EBRT was positive indicating that TARGIT was cost-effective: at a maximum willingness to pay for a QALY of £20,000 or £30,000 the mean incremental net monetary benefit was £1363 and £1730 (see Table 16). The incremental net monetary benefit was not significantly different from zero at a maximum willingness to pay for a QALY of £20,000 (mean £1363, 95% CI –£66 to £2838) or £30,000 (mean £1730, 95% CI –£284 to £3740). However, the incremental net monetary benefit for TARGIT compared with EBRT was borderline significantly different from zero: at a maximum willingness to pay for a QALY of £20,000 the 90% CI was £175 to £2818 and at £30,000 it was £38 to £3746. In a hypothesis test, this would indicate that against a null hypothesis the incremental net monetary benefit equals zero; the p-value for rejecting the null hypothesis would be between 0.05 and 0.1.

We repeated the analysis several times using alternative versions of the multiple imputation process using different random number seeds to investigate whether or not the conclusions of the analysis changed; in every case the results were qualitatively the same.

Sensitivity analyses

In all but one of the scenarios tested in the deterministic sensitivity analysis TARGIT was less costly than EBRT (Table 17). The exception was when the cost of TARGIT was £2700 per patient, which is higher than the maximum value in Picot et al. 72 (£2500). The costs were statistically significantly lower for TARGIT compared with EBRT (the 95% CI did not cross zero) when EBRT costs were based on the number of fractions received in the trial, the unit cost per fraction of EBRT was £154 (the upper quartile unit cost in the NHS reference costs76), the cost of TARGIT was ≤ £1900 per patient and the alternative utility values were used.

In every case the QALYs gained were small, positive and non-significant. Note that these were unlikely to change given that the parameters varied in the deterministic sensitivity analysis were mainly cost parameters.

In all cases tested the incremental net monetary benefits for TARGIT compared with EBRT were positive at a maximum willingness to pay for a QALY of £20,000 and £30,000. The incremental net monetary benefits were significantly greater than zero (the 95% CI did not cross zero) when EBRT costs were based on the number of fractions received in the trial, the unit cost per fraction of EBRT was £154 and the cost of TARGIT was ≤ £1700 per patient at a maximum willingness to pay for a QALY of £20,000 or ≤ £1300 per patient at a maximum willingness to pay for a QALY of £30,000.

The probability that EBRT is cost-effective is equal to 1 minus the probability that TARGIT is cost-effective at each value of the maximum willingness to pay for a QALY. The cost-effectiveness acceptability curve shows that, at a maximum willingness to pay for a QALY of £20,000 (£30,000), the probability that TARGIT was cost-effective was 0.965 (0.950) in the base case (Figure 30 and see Table 17). In the deterministic sensitivity analyses the probability that TARGIT was cost-effective at a maximum willingness to pay for a QALY of £20,000 was > 0.75 in every case. At a maximum willingness to pay for a QALY of £30,000 the probability that TARGIT was cost-effective was > 0.80 in every case.

FIGURE 30.

Cost-effectiveness acceptability curve showing the probability that TARGIT is cost-effective compared with EBRT at different values of the maximum willingness to pay for a QALY.

Potential budget impact

The cost savings per patient found in our base case could translate into cost savings per year for the NHS if TARGIT was carried our routinely instead of EBRT in eligible patients. The latest available evidence suggests that in 2011 there were 49,936 new cases of breast cancer in the UK. 95 Figures from Germany96 based on 1108 new cases of breast cancer treated at a single centre between 2003 and 2009 suggest that 258 patients (23.3% cases) would have met the eligibility criteria for participation in the TARGIT-C trial (ClinicalTrials.gov NCT02290782),97 which has similar but more restrictive inclusion and exclusion criteria than the TARGIT-A trial (e.g. age ≥ 50 years rather than ≥ 45 years, tumour size ≤ 2 cm rather than ≤ 3.5 cm). This conservatively suggests that around 49,936 × 23.3% = 11,600 patients may be eligible for TARGIT in the UK each year. Applying the cost saving per patient in our base case to this estimate suggests that the NHS might save around 11,600 × –£685 = £8 million a year.

Figures from France98 based on two cohorts of patients between 1980 and 2008 indicate that, across a combined total of 12,025 patients receiving breast-conserving surgery, 5545 patients (46%) would have been eligible for TARGIT according to the eligibility criteria of the TARGIT-A trial. Approximately 58% of newly diagnosed patients with breast cancer in the UK undergo lumpectomy. 99 Therefore, according to these figures, around 49,936 × 58% × 46% = 13,300 patients may be eligible for TARGIT in the UK each year. Applying the cost saving per patient in our base case to this estimate suggests that the NHS might save around 13,300 × £685 = £9.1 million a year.

Combined, these calculations suggest that if TARGIT was carried our routinely instead of EBRT in eligible patients the potential cost savings to the NHS would be around £8–9.1 million each year.

Summary

We undertook a cost–utility analysis comparing TARGIT versus EBRT in the prepathology stratum of the earliest cohort of the TARGIT-A trial. In our base case TARGIT was statistically significantly less costly than EBRT, produced similar QALYs, had a positive incremental net monetary benefit that was borderline statistically significantly different from zero and had a probability of > 90% of being cost-effective. Although there appears to be some uncertainty about the statistical significance of the differences in costs and whether or not the incremental net monetary benefit is different from zero, the appears to be little uncertainty in the point estimates, based on deterministic and probabilistic sensitivity analyses.

Comparison with other studies

Alvarado et al. 73 found that TARGIT dominated EBRT (was less costly and more effective) in that it resulted in a QALY gain of 0.00026 compared with EBRT and cost US$5191 less. Based on their analysis using TARGIT-A trial data, Shah et al. 74 reported that use of TARGIT was associated with cost savings of US$3.6–4.3 million per 1000 patients compared with EBRT. Neither study reported CIs around the point estimates and so it is unclear if the QALYs gained or cost savings were significantly different from zero. Our results are qualitatively similar to those of Alvarado et al. 73 in that based on the point estimates in our base case we also found a cost saving for TARGIT and a small QALY gain compared with EBRT. Our findings are also qualitatively similar to those of Shah et al. 74 in that we also found a cost saving with TARGIT compared with EBRT. However, given that both studies were US based it is difficult to draw close comparisons.

Picot et al. 72 found that TARGIT produced a small cost saving compared with EBRT and a small QALY loss; the authors’ conclusion was that EBRT was associated with more QALYs than TARGIT at a broadly similar overall cost. The point estimates of the costs saved per QALY lost were < £20,000, indicating that TARGIT was not cost-effective (in cases in which an intervention is less costly and less effective than the comparator then for it to be cost-effective the incremental cost-effectiveness ratio must lie above the threshold value). CIs around the cost and outcome differences and the incremental cost-effectiveness measures were not reported and so it is difficult to make a full comparison of the findings. Other than the use of patient-level data, the main differences between the study by Picot et al. 72 and the present study were the time horizon and the range of costs included. Picot et al. 72 modelled costs and outcomes using a time horizon of 40 years, whereas the time horizon in the present study was 5 years based on the average length of follow-up in the trial. We did not extrapolate beyond the end of the trial because the within-trial analysis found no evidence of significant differences in QALYs between the groups and, although there was some evidence of differences in costs, these differences were all accrued during the first year. In terms of costs included, there were several differences between the present study and that by Picot et al. 72 During the radiotherapy treatment period the present study included the cost of EBRT boost and NHS transport costs for EBRT, which were not included in the study by Picot et al. 72 More generally, the total cost per patient in the study by Picot et al. 72 over the 40-year period, based on the costs included in the analysis, was around £2300 in both groups. In our study the total cost per patient over the 5-year period, based on the costs included in the analysis, was around £11,600 in both groups, suggesting large differences in the range of cost components included.

Strengths and limitations

The main strength of our analysis is that it is based on a large international multicentre randomised trial with detailed information on resource use and events for a median follow-up period of 5 years.

There are several limitations. First, the time horizon was 5 years. Extrapolation beyond the end of the trial using decision-analytical modelling was not undertaken because the within-trial analysis found no evidence of significant differences in QALYs between groups during the 5-year period. This probably reflects the main finding from the TARGIT-A trial that TARGIT was non-inferior to EBRT with regard to local recurrence. Although there was some evidence of differences in costs these differences were all accrued during the first year; there was no evidence of significant differences in costs beyond the first year. Hence, the 5-year time horizon was long enough to reflect all important differences in costs or outcomes between the two treatments. Although local recurrence (and other events) are likely to continue to occur over a patient’s lifetime, the evidence from the TARGIT-A trial is that TARGIT is non-inferior to EBRT. Hence, taking a longer time horizon is unlikely to have affected the results of the incremental analyses.

Second, utility data were not collected in the TARGIT-A trial. We therefore applied utility values from published sources to the health states experienced by patients in the trial. The utility values that we applied may not reflect the values of patients in the study. Given the relatively small number of events, and that the numbers of events were largely not different between the two groups, the QALY differences between the two groups may not be expected to change much with alternative utility values. This is borne out by our sensitivity analysis, which showed that the results did not change appreciably when we used alternative values. We did not incorporate utility losses associated with additional procedures, chemotherapy, mastectomy or complications in our analysis. Given the low incidence of these events, that they were evenly distributed between treatment groups and that the time period affected is likely to be short this is unlikely to affect the QALYs associated with each treatment group. We also did not include any utility losses associated with EBRT. Therefore, this would make our estimates more conservative because such an omission would work against TARGIT.

Third, the dose of EBRT administered to patients in the TARGIT-A trial does not reflect current UK treatment guidelines. This reflects the multinational nature of the trial, plus that it began recruiting patients in 2000 when treatment recommendations were different. We accounted for this in our base case by assuming that all patients in the TARGIT-A trial who received EBRT received a fixed number of 15 fractions.

Fourth, the analysis took a NHS/PSS perspective on costs. A wider perspective (e.g. societal) could have been taken to measure costs, including impacts on the rest of society, patients, families and businesses. If a wider perspective was taken this should include the additional costs borne by patients and families in terms of time and travel costs associated with additional radiotherapy visits for EBRT compared with TARGIT. If these costs were included it is likely that the cost savings attributable to TARGIT would be greater than demonstrated. Taking the example of the transport costs, we used the figure of 13.5% for the proportion of patients for whom the NHS paid for transport for radiotherapy visits for EBRT. Assuming that the same cost is paid out of pocket by the remaining patients, the difference in costs between TARGIT and EBRT would be increased by £877 to £1562 per patient, taking the total saving to the UK national economy to between 11,600 × £1562 = £18.1 million and 13,400 × £1562 = £20.9 million each year. These are crude estimates and further research to evaluate the wider impacts of TARGIT, including on other costs to the rest of society, would be useful.

Non-inferiority for local recurrence

Our primary outcome was non-inferiority in terms of local recurrence. We tested for non-inferiority of the TARGIT technique for local control within a risk-adapted design. When we designed the trial in 1999, we set the margin of non-inferiority to be an absolute difference in the local recurrence rate of 2.5%. This is a stringent and relatively conservative margin compared with that in contemporary studies. For example, the only other trial of IORT – the ELIOT trial35 – was set up with the assumption that a difference of 7.5% in local recurrence between the experimental and the control groups would result in the two treatments being considered as equivalent. The Cancer and Leukemia Group B (CALGB) study11 found that there was a 7% difference in local recurrence when radiotherapy was omitted in low-risk (T1N0) patients aged > 70 years and concluded that this difference was not significant. Furthermore, two patient preference studies conducted in parallel with the main trial have also corroborated that, given the convenience of a single treatment session, the median increase in local recurrence that patients would be ready to accept is 2.5%. 100,101 Most importantly, analyses by the Early Breast Cancer Trialists’ Collaborative Group (EBCTCG) indicate that an absolute increase in local recurrence of < 10% does not result in any increase in mortality and therefore we did not feel that there were any ethical concerns in setting up this study. 4–6 We were also certain that the trial was well powered to detect a difference of ≥ 2.5% and therefore that if the results showed non-inferiority they could be relied on. In fact, we provided detailed calculations of how, in view of the recurrence rates seen in the trial, 585 patients would have provided sufficient power for the trial.

Although non-inferiority trials are becoming more common, especially when cure rates are high and a reduction in treatment toxicity without a loss of efficacy is desirable, the concept of non-inferiority can be difficult to grasp. The TARGIT-A trial was such a non-inferiority trial, which meant that, even if there was a statistically significant difference between the groups, as long as it was less than the pre-specified value (2.5% in the TARGIT-A trial), the two groups would be judged to be non-inferior to each other. In these circumstances, less expensive, less toxic or more convenient treatment becomes the preferred choice (Figure 31).

FIGURE 31.

The meaning of non-inferiority: 10 examples of different scenarios that might occur in a randomised trial testing non-inferiority between two treatments. The dots represent the absolute difference and the lines the CI. The blue circle includes two of the trial results: on the left is the TARGIT prepathology stratum (difference 0.37%) and on the right is the earliest cohort of the whole trial (n = 1222), which had a median follow-up of 5 years (difference 1.13%) (see Table 8). 2,102

The pre-specified non-inferiority boundary of a 2.5% absolute difference in local recurrence is very conservative and has been validated in patient preference studies. 101,103,104 It is much smaller than the 7.5% margin in the ELIOT trial105 and lower than the difference considered ‘acceptable’ by the CALGB (5% difference)11 and Post-operative Radiotherapy In Minimum-risk Elderly (PRIME) II (3% difference)106 studies. At this boundary, TARGIT was non-inferior to EBRT (P_{non-inferiority} < 0.00001). Within the trial of 3451 patients, the first 1222 patients had a median follow-up of 5-years; the safety, efficacy and non-inferiority results of these 1222 patients were similar to those seen in all patients (see Table 8). 2

We found that, when all patients were analysed together, the risk of local recurrence for TARGIT was non-inferior to that for EBRT (5-year risk: TARGIT 3.3% vs. EBRT 1.3%). We had pre-specified that the significant p-value for difference for the log-rank test would be < 0.01 for local recurrence, as this was the second such analysis. Therefore, the 2% absolute difference was not only within the non-inferiority margin but also, as the p-value was 0.04, strictly speaking, it was not statistically significant.

We used the standard method of using binomial proportions to calculate the non-inferiority statistic, rather than using single 5-year point estimates, which do not represent absolute events and can lead to erroneous conclusions, especially when the event rate is low. The standard method that we used makes a comprehensive assessment of the whole follow-up period, which stretches from the day of randomisation to the longest follow-up of more than 12 years and includes all events in that period. To address the issue of follow-up, we repeated the analysis by restricting the analysis to cohorts randomised early in the trial, including the earliest 1222 patients; these patients had a median follow-up of 5 years. The multiple comparisons issue does not apply here as we have not chosen one of several comparisons to make our conclusions but rather used all three comparisons to draw an informed conclusion.

When analysed according to the timing of randomisation/delivery of TARGIT, we found that, whereas the trial as a whole as well as the prepathology stratum confirmed the non-inferiority of TARGIT, the difference in local recurrence in the postpathology stratum was > 2.5%. Therefore, we have recommended that TARGIT should be used during the initial lumpectomy, as in the prepathology stratum (note that this is a stratum not a subgroup), in which the difference between the two treatments was undoubtedly not statistically significant (p = 0.31). Finally, survival without local recurrence (in which deaths are not censored, in line with recommendations by the US Food and Drug Administration62 and elsewhere63) was not statistically different between the two arms (see Figures 18 and 19).

Toxicity

There was no increased wound-related toxicity with TARGIT and there was reduced radiotherapy-related skin toxicity. This was not negated by the EBRT boost delivered to 15% of the TARGIT arm patients; toxicity was the same whether such an EBRT boost was applied to TARGIT or EBRT patients. 107

There were significantly fewer grade 3 or 4 radiotherapy-related complications in the TARGIT arm; however, one should note that the incidence of these complications was very low in both arms of the trial. This, and the low local recurrence rate reported, is a testament to the high-quality radiotherapy given in the EBRT arm.

Studies conducted parallel to the TARGIT-A trial in individual centres have shown that TARGIT delivers a better cosmetic outcome,108 lower short-term and long-term skin toxicity and an overall better quality of life. 107,109,110

Mortality

Although the number of breast cancer deaths did not differ significantly between groups, there were significantly fewer deaths from other causes in the TARGIT arm of the trial (number of deaths from causes other than breast cancer: TARGIT 17, EBRT 35). In particular, there were far fewer deaths from cancers other than breast cancer (TARGIT 8 vs. 16) or from cardiovascular disease (EBRT 2 vs. 11).

This early decrease in cardiovascular deaths was somewhat surprising as it was previously believed that any EBRT-related increase in cardiovascular disease would not be apparent until 7–10 years after radiation. 3 On the other hand, given the size of the trial, it is unlikely that there was a significant imbalance in baseline comorbidities and, statistically, the probability of observing the difference we observed if, in reality, there was no real difference in non-breast-cancer mortality between the randomised arms, is very low (p = 0.0086). The difference is also unlikely to be caused by poor EBRT delivery, given the low rates of local recurrence (1.2%) and grade 3/4 radiotherapy toxicity (2.1%) among patients receiving EBRT. These results therefore deserve a search for an explanation, rather than dismissal.

Indeed, a large recent study from a group from Oxford, UK, published in the New England Journal of Medicine, has shown that EBRT toxicity appears within the first 5 years. 71 The authors found an increase in cardiac mortality of 16.3% per gray in the first 5 years, 15.5% in the second 5 years, 1.2% in the years 10–19 and 8.2% after 20 years. As the effect appears to be the largest in the first 10 years – more than twice that of the average – the yardstick should be based on the period of follow-up and it would be wrong to use the average value of a 7.4% increase in cardiac mortality per gray as a yardstick. The higher increase in the first 5 and 10 years is consistent with our findings as they fall within each others’ 95% CIs.

It is possible that the TARGIT-A trial may have uncovered a very small increase in non-breast-cancer mortality associated with EBRT that has previously been masked by breast cancer deaths. The 5-year risk of breast cancer mortality was 2.2% in TARGIT-A, compared with, for instance, 30% in the Cancer Research Campaign (CRC)1 trial111 (Figure 32). The absolute difference in non-breast-cancer mortality in the TARGIT-A trial was just 2.1% (TARGIT 1.4% vs. EBRT 3.5%) and approximately one-third of this increase (0.6%) was from cardiac causes.

FIGURE 32.

(a) The high breast cancer mortality rate (70% at 5 years) in older trials such as the CRC trials could have masked the deaths from causes other than breast cancer in the early years, which were revealed only when the breast cancer mortality lines started flattening. (b) In the TARGIT-A trial, however, the breast cancer mortality rate was very low (2.1%), which would allow the early, albeit small in absolute terms, increase in non-breast-cancer mortality to be unmasked. Part (a) is reproduced from Postoperative radiotherapy and late mortality: evidence from the Cancer Research Campaign trial for early breast cancer, Haybittle JL, Brinkley D, Houghton J, et al. , vol. 298, pp. 1611–14,111 with permission from BMJ Publishing Group Ltd.

As can be seen in Figure 32, the difference in deaths from other causes starts becoming apparent only when the deaths from breast cancer start levelling off. The slope of the curve denoting breast cancer deaths at 10 years in the CRC1 trial is similar to that seen in the TARGIT-A trial from the beginning, which has probably allowed us to see the small difference in non-breast-cancer morality between those who received some irradiation to the heart and other organs and those who did not. This idea may well explain the recently published results of a meta-analysis of randomised trials comparing PBI and whole-breast irradiation. 112 The meta-analysis demonstrates a reduction in non-breast-cancer and overall mortality with PBI compared with whole-breast irradiation.

Alternatively, we believe that TARGIT may in fact be protective against deaths from cardiovascular causes and other cancers,69,70,113 perhaps through a reduction of the systemic inflammatory response that potentiates cardiovascular diseases or cancers; this, in combination with the reduced radiation toxicity, may be the cause of the difference that we have observed here. The first hints of this may be found in the finding of a significantly lower non-breast-cancer mortality hazard in patients who received TARGIT + EBRT than in those receiving EBRT alone. It is worth keeping in mind, however, that this comparison was non-randomised and the numbers were small.

Further research into the potential positive effects of TARGIT given alongside EBRT is currently under way in the TARGIT-B trial, which is looking at the impact of TARGIT given as a radiotherapy boost.

Trial design: experimental treatment = a risk-adapted approach

The TARGIT-A trial was a pragmatic trial that tested two radiotherapy policies for non-inferiority in local recurrence after breast-conserving surgery: risk-adapted radiotherapy and conventional standard radiotherapy given daily over several weeks.

Following breast-conserving surgery, the control arm received EBRT over several weeks according to local treatment guidelines. This conventional treatment does not normally vary with tumour characteristics and everyone receives the same dose of whole-breast radiation, although there could be variation between centres. There could also be some variation in terms of delivery or omission of the tumour bed boost depending on the age of the patient.

The experimental arm, on the other hand, aimed to deliver all radiotherapy in a single dose in the majority of patients, using the TARGIT technique and the INTRABEAM device. TARGIT was given either at the time of lumpectomy (the prepathology stratum, 2298 patients) or as a second procedure several days after the initial lumpectomy, by reopening the wound (the postpathology stratum, 1153 patients).

The risk-adapted design meant that, if after TARGIT adverse features were found in the final histopathological specimen, the trial protocol mandated the addition of EBRT to the rest of the breast (excluding a tumour bed boost). Individual centres decided on the set of histological adverse features before randomisation began, in addition to the factors stipulated in the core protocol: the presence of positive margins, evidence of an invasive lobular carcinoma or an EIC.

Such additional EBRT was needed in 15% of cases, as per our original estimations. This proportion was higher in the prepathology stratum, with one-fifth of the patients receiving TARGIT at the time of their primary surgery also being given EBRT because of adverse prognostic factors as determined in the multidisciplinary discussion of the full histopathology after surgery. This adaptation of the treatment based on risk should ensure that the experimental group in our clinical trial reflects how TARGIT would be delivered in the real world; meanwhile, the patients in the EBRT arm provided a good control group, representing how the treatment is currently given, according to local guidelines. The trial should therefore tell us how the outcome for patients would change (or not) if TARGIT were to be introduced into routine clinical practice. We believe that this is the only method of accurately reflecting practice in the real world, without sacrificing statistical validity; it may be disappointing for the women who needed supplemental EBRT but it is worth remembering that TARGIT allows the large majority of patients to avoid several weeks of EBRT.

We believe that this type of pragmatic non-inferiority design is a strength of the trial, although at the same time it could make it more difficult for purists to interpret the results. This is quite contrary to how patients and the lay public interpret the results – they have consistently felt that the small statistically non-significant difference in local recurrence was really not a concern, particularly compared with the great benefit and convenience of patients being able to have all of their treatment in one sitting100,101,103,104,114,115 [see http://goo.gl/j3jyM1, http://goo.gl/x6Sk80, http://goo.gl/j8xwHZ, www.bbc.co.uk/programmes/p01lhjm2, www.telegraph.co.uk/women/womens-health/10439775/A-revolution-in-breast-cancer-therapy.html, www.bbc.co.uk/news/health-28485504, http://goo.gl/dJQKAF, www.dailymail.co.uk/health/article-1378267/Me-operation-Targeted-intraoperative-radiotherapy.html and www.wsj.com/articles/alternative-way-to-treat-early-stage-breast-cancer-with-radiation-1440448587 (accessed 5 July 2016)]. The observed reduction in non-breast-cancer mortality was an added advantage.

Accrual, power and follow-up duration

The current analysis represents mature data from a large number of patients. Fifteen years ago our original power calculations expected a background local recurrence rate of 6% and, if we accept a non-inferiority margin of 2.5%, the sample size required to achieve 80% power and 95% confidence would then have been 2232 patients. The local recurrence rate measured in the EBRT group of the TARGIT-A trial at 5 years was in fact just 1.5% and, given this lower value, we needed only 585 patients to achieve 80% power and 95% confidence (Table 18). 1,116 It has been well documented that, in the first 5 years after treatment, breast cancer patients face the highest risk of local recurrence. The peak hazard of local recurrence is between 2 and 3 years after surgery, as seen in many studies, including the Oxford overview4,117,118 (Figure 33).

FIGURE 33.

Two large studies118,119 demonstrating that the risk of local recurrence is mainly in the first few years after surgery and flattens after 5 years. (a) Reprinted from Cancer Epidemiology, Biomarkers and Prevention: a publication of the American Association for Cancer Research, cosponsored by the American Society of Preventive Oncology, 2012, vol. 21, Cheng L, Swartz MD, Zhao H, et al. , Hazard of recurrence among women after primary breast cancer treatment – a 10 year follow up using data from SEER-Medicare,118 with permission from AACR. (b) Reproduced from N Engl J Med, Fisher B, Anderson S, Bryant J, et al. , Twenty-year follow-up of a randomized trial comparing total mastectomy, lumpectomy, and lumpectomy plus irradiation for the treatment of invasive breast cancer, vol. 347, pp. 1233–41. 119 Copyright © 2002 Massachusetts Medical Society. Reprinted with permission from Massachusetts Medical Society.

Esserman et al. 120 have recently applied a decision-analytic framework to conclude that the median follow-up of the TARGIT-A trial is sufficient to strongly recommend wide adoption.

Indeed, a large recent study with a 20-year follow up, conducted by Wickberg et al. ,121 has shown that, whereas radiotherapy decreases the number of recurrences in the first 5 years, it ceases to have any effect after the first 5 years; thus the rate of local recurrence after the first 5 years following radiotherapy is similar to that in patients who received only lumpectomy. This is best demonstrated in Figure 34, which depicts how the lines representing local recurrence are parallel after 5 years and most of the hazard of not taking radiotherapy is in the first 5 years. 122

FIGURE 34.

A large Swedish study121,122 with a long follow-up demonstrates that, although the local recurrence risk persists even after 5 years, the effect of radiotherapy on reducing that risk is seen only in the early years. This is seen most clearly in (c), with the risk reduction from radiation being mainly in the first 5 years, with very little effect thereafter. Part (c) is drawn using data from Wickberg et al. 122 (a) Local recurrence with and without radiotherapy over 25 years;121 and (b) hazard of not taking radiotherapy is mainly in the first 5 years;122 and (c) hazard EBRT vs. no- EBRT. 122 (a) Reprinted with permission from Wickberg et al. , J Clin Oncol, Vol. 32, Issue 8, 2014: 791–7. 121 © (2014) American Society of Clinical Oncology. All rights reserved; and (b) reprinted with permission from Wickberg et al. , J Clin Oncol, Vol. 32, Issue 29, 2014: 3340. 122 © (2014) American Society of Clinical Oncology. All rights reserved.

The first patient in the TARGIT-A trial was randomised on 25 March 2000 and the last patient > 12 years later on 25 June 2012. This report presents data from all of the patients randomised in this period: 3451 women in 33 centres across 11 countries. Of these 3451 patients, the first 1222 (called the earliest cohort in our 2014 paper), recruited in the first 8 years of the trial, have a median follow-up of 5 years and the first 2232 patients, originally deemed necessary for our power calculations (called the mature cohort in our 2014 paper2), have a median follow-up of close to 4 years.

Figure 35a plots each patient’s follow-up according to when they were recruited; one can see that the rate of accrual was slow initially and increased over the last few years of the trial, giving us a misleadingly lower total median follow-up. This is already accounted for in our statistical tests: Kaplan–Meier curves and the log-rank test. These give accurate calculations for risk at 5 years for the entire trial. We have also conducted our main analyses for patients randomised in the first 8 years of the trial, as these patients have a median follow-up of 5 years. The results for these 817 patients in the prepathology stratum (more than the 585 needed for statistical power) matched our overall results.

FIGURE 35.

The accrual and follow-up duration of patients in the prepathology stratum. (a) Each vertical green line represents one patient and the location of the green line on the x-axis represents the date of randomisation. The height of the green line represents the follow-up duration. The blue line represents the cumulative number of patients accrued in the trial (during 2001–2 the line was flat with no accrual because of the chief investigator working in a different centre); and (b) average accrual per month through the 12 years of the trial.

We therefore believe that the trial is sufficiently powered to demonstrate that TARGIT is non-inferior, particularly in the prepathology stratum, that is, when TARGIT is given simultaneously with the initial lumpectomy, which is the approach that we believe is preferable, based on current data, to using TARGIT as a second procedure by reopening the initial lumpectomy wound.

Two strata, two distinct treatments

The original trial design set out that patients should be randomised before the initial lumpectomy and, if allocated to TARGIT, this should be delivered at the time of surgery. When the Australian centre joined, they found it logistically difficult to give TARGIT during the first procedure. They suggested that they would like to give TARGIT as a second procedure by reopening the wound. This required us to add a new stratum with its own randomisation table: the postpathology stratum. The new stratum allowed patients to be randomised after lumpectomy. Although this allowed more stringent case selection, and simplified the logistics of arranging the theatre slot and staffing for each case, it also necessitated the reopening of the surgical wound. Thus, the difference between the postpathology stratum and the original prepathology stratum was in the timing of randomisation in relation to removal of the cancer either as a lumpectomy or an excision biopsy. The two strata differed in two major aspects: (1) in the postpathology stratum case selection could be more stringent as the tumour had already been examined and (2) in the postpathology stratum the experimental intervention was practically very different from that in the prepathology stratum: radiation was delivered to a reopened wound, with an accuracy that may not be as good as when it is delivered immediately after lumpectomy, and it was delivered to tissues that were fibrous scar rather than a pliable fresh wound.

A separate analysis was planned for these two strata of patients from the outset.

Randomisation in the TARGIT trial was carried out before lumpectomy in 2298 patients, or after the excision of the cancer in 1153 patients, in the prepathology and postpathology strata, respectively, whose treatment, selection and ultimately, the outcomes differed.

Having two strata allowed more centres to participate and we believed that the more stringent criteria would give us a nested cohort with a better cancer prognosis in whom the results might be even better. However, we did not realise that the immediacy may be so important and that irradiating the scarred tumour bed may not be as effective as giving radiation to the fresh tumour bed. Therefore, although the better case selection might have allowed for a better outcome, the different mode of delivery of radiation may have jeopardised local control, albeit by a small amount.

When analysed together, local recurrence with TARGIT was non-inferior to EBRT and there was a trend towards lower mortality with TARGIT. There was no significant difference in deaths from breast cancer but there was a significant reduction in the number of deaths from cardiovascular causes and other cancers in the TARGIT arm compared with the EBRT arm.

Analysis of the prepathology stratum also found similar results: non-inferiority for local recurrence, no significant difference in breast cancer mortality and a significant reduction in non-breast-cancer mortality.

However, in the postpathology stratum, the difference in local recurrence between TARGIT and EBRT was more than the pre-specified 2.5% non-inferiority margin. There was no difference in mortality. This relatively reduced efficacy of postpathology delivery of TARGIT may be attributable to the need to reopen the surgical wound (increasing trauma), the reduced accuracy in placement of the applicator, the radiation being given to scar tissue or missing a critical temporal window in radiation delivery; the median time between the primary surgery and TARGIT delivery in the postpathology stratum was 37 days.

Statistical tests comparing the two strata were not planned, as such comparisons would be non-randomised. Nevertheless, we believe that the difference observed is the result of the difference in TARGIT timing, as the two EBRT groups were similar in breast cancer recurrence (1.1% prepathology vs. 1.7% postpathology; p = 0.49) and non-breast-cancer mortality.

The postpathology stratum has therefore highlighted the importance of temporal precision in the delivery of radiation using this technique; the results from the postpathology stratum are not favourable, in terms of a higher local recurrence but, reassuringly, there was no decrease in mortality.

When we assessed whether or not more frequent use of additional EBRT could be the cause of better outcomes in the prepathology stratum (21% vs. 3% in postpathology stratum), we found that patients in the super-selected postpathology stratum had a much better prognosis than those in the prepathology stratum, as expected. Despite this worse prognosis, the 5-year local control in prepathology cases who received TARGIT alone appears to be better – not worse – than those in the postpathology stratum who received TARGIT alone (2.7% vs. 5.9%, respectively) and similar to that in the whole prepathology stratum (2.1%). These data suggest that delivery of TARGIT IORT at the time of initial lumpectomy leads to the accurate and timely application of radiation to a well-vascularised, undisturbed, fresh tumour bed that ensures temporal and spatial accuracy rather than that the addition of EBRT in higher-risk cases is responsible for the favourable results in the large prepathology stratum.

Furthermore, it has been reported that surgical wounds promote tumorigenesis,123 possibly via a T-cell-dependent mechanism. 124 Delivered at the time of surgery, TARGIT has been shown to reduce the motility of the wound fluid, thereby perhaps reducing recurrence through a beneficial effect on the tumour microenvironment, that is, it abrogates the stimulation of motility, proliferation and invasiveness. 17,125

Recent translational work126,127 has identified increased levels of microRNA-223 when TARGIT is delivered to the fresh surgical wound, which reduces the malignant potential of cancerous cells. TARGIT appears to act through an epidermal growth factor receptor (EGFR) pathway. A possible mechanism is shown in Figure 36. These data suggest that giving TARGIT at the time of lumpectomy would be preferable to giving it as a second procedure.

FIGURE 36.

Possible molecular biological explanation of the effectiveness of TARGIT during lumpectomy particularly in PgR positive cases. miRNA, microRNA.

Risk-adapted approach

About one-fifth of the patients receiving TARGIT at the time of their first surgery were also given EBRT; this was indicated if pathology conducted after lumpectomy showed any adverse prognostic factors as determined in the treatment policy document and the multidisciplinary discussion of the full histopathology after surgery. Overall, that is when postpathology stratum is included, 15% of patients randomised to the risk-adapted TARGIT arm were given both TARGIT and EBRT.

We believe that a design in which the experimental arm was a risk-adapted approach was the best method of accurately reflecting practice in the real world, without sacrificing statistical validity; it may be disappointing for the women who needed supplemental EBRT but it is worth remembering that TARGIT allows the large majority of patients (80–85%) to avoid several weeks of EBRT.

Is ‘no radiotherapy’ an option? Only if one is prepared to take the risk of local recurrence

Sceptics have suggested that these 15% of patients who received both TARGIT and EBRT were indeed those very patients who would otherwise have recurred without radiation, but who did not do so because they received EBRT. The remaining 85% of patients, so the sceptics argue, were therefore those who would never have recurred even without radiation and TARGIT did not really prevent any recurrences, only the added EBRT did. If this were true, the trial guidelines detailing when EBRT was to be added have extraordinarily managed to identify all patients who might recur, based on tumour characteristics alone! Obviously, such an algorithm for case selection is currently impossible.

Although this would be fortunate, it is clearly not the case. Other studies testing the benefits of EBRT over no radiotherapy have enrolled patients whose tumour characteristics were far better than those in the TARGIT arm who did not receive EBRT have found that entirely omitting radiotherapy results in a significantly higher recurrence rate. 11,106,129 Additionally, as the postpathology stratum usually did not need additional EBRT, as this stratum was randomised after lumpectomy and pathology, these patients provided an internal control. As seen in Table 11, the tumour characteristics in this postpathology stratum were much better than those in the prepathology TARGIT arm who did not receive EBRT, but recurrence was much higher, again indicating that TARGIT, delivered at the correct time, is particularly good at avoiding local recurrence. 5,130

A recently published CALGB randomised clinical trial,11 which has addressed this even in the very best prognosis patients (age >70 years, T1N0M0, ER positive), found that giving ‘no radiotherapy’ leads to a significantly higher local recurrence rate (9%, i.e. one in every 11, vs. just 2% with radiotherapy). In a more recently reported PRIME II trial106 with super-selected patients with an extremely good prognosis (much better than that of the TARGIT-A trial patients), there was a statistically significant reduction in disease-free survival when radiotherapy was omitted and local recurrence without radiotherapy was 4.1%. Even in these highly selected good-prognosis patients, the subgroup of patients who were ER positive had a local recurrence rate of 3.2% without compared with 0.8% with radiotherapy, which is relatively much higher than that achieved by TARGIT during lumpectomy in ER-positive, PgR-positive cancers (1.4% with TARGIT vs. 1.2% with EBRT). If such patients were treated by TARGIT IORT initially, they could avoid the need for EBRT with its associated risks, inconvenience and costs as well as the increased risk of recurrence associated with completely withholding radiotherapy (see Table 20).

Other trials

Several other systems of delivering a targeted dose of radiation to the affected area of the breast have been trialled; these have been reviewed elsewhere. 112,131 It is worth keeping in mind that, of these, the TARGIT-A trial alone took a truly risk-adjusted approach. A recent meta-analysis of the TARGIT-A trial and the GEC-ESTRO (Groupe Européen de Curiethérapie ESTRO) trial of PBI with interstitial wires has found that the local control with PBI is non-inferior to whole-breast irradiation and there is a reduction in non-breast-cancer mortality. 132 Importantly, the overall survival benefit of using TARGIT IORT instead of traditional EBRT has been confirmed in the meta-analysis of all randomised trials of PBI versus whole-breast irradiation. 112

The ELIOT trial,35,133 which recently published results indicating poorer outcomes with IORT using the NOVAC-7 (New Radiant Technology) device than with EBRT,35 was the only other randomised trial testing intraoperative partial breast radiotherapy. Their technique used 4- to 12-MeV electrons to deliver 21 Gy over 3–5 minutes; this seems comparable to TARGIT’s 50 kV over 20–30 minutes to deliver 20 Gy to the tumour bed surface, a dose that attenuates to 5–7 Gy at a 1-cm depth. However, the preparation for delivery is fundamentally different. Whereas the TARGIT technique delivers radiotherapy from within the breast, with the radiation source sitting in the tumour bed – where the tumour was before its excision – and radiation given directly to the well-vascularised tissues of the tumour bed, in the ELIOT trial radiotherapy was delivered from outside the body; this therefore requires the extensive mobilisation of the mammary gland, insertion of a prepectoral lead shield and the apposition of the edges of the tumour bed. Such dissection could in theory lead to relative ischaemia, at least in the short term, of the edges of the mobilised breast gland, the very tissues that are being immediately irradiated. The higher (albeit low in absolute terms) recurrence rate seen in the ELIOT trial might therefore be because of the lower oxygen saturation of the tissues being irradiated as it is well known that tissue ischaemia reduces the effectiveness of radiation. 134–141 In addition, theoretically, some at-risk tissue such as the retracted skin and subcutaneous breast tissue and the prepectoral tissue behind the shield would not receive any radiation while exposing more raw surface to tumour seeding.

Hormone sensitivity

We used PgR status as a surrogate marker for a functionally active ER receptor and found that PgR status appears to predict response to radiotherapy. Patients whose tumours were less hormone sensitive, as shown by a negative PgR status, had a significantly higher risk of local recurrence overall and in the TARGIT arm.

Interestingly, however, no such difference was observed in the EBRT arm of our trial. This could be because of the small number of events in the EBRT arm (n = 11) or because of the nature of PgR-negative tumours: the more infiltrative margins57–59 should perhaps most appropriately be irradiated by the wide and uniform field of EBRT. The smaller volume of tissue irradiated with partial breast radiotherapies could result in a failure to wholly irradiate the cancerous cells. Furthermore, the EGFR pathway by which TARGIT increases mi223 levels and therefore reduces tumour growth is PgR dependent; in ER-positive, PgR-positive patients, this may contribute to the clinical effectiveness of appropriately timed TARGIT given during lumpectomy, but with a hormone-insensitive tumour this mechanism of local recurrence reduction would not be in place. This may explain why, even though among PgR-positive cases (n = 2462) there was no significant difference between the TARGIT and the EBRT arms (2.3% vs. 1.5%; p = 0.51), in the PgR-negative cases (n = 554) local recurrence was significantly higher in the TARGIT arm (7.0% vs. 0.5%; p = 0.017).

It is possible that with more events the TARGIT and EBRT local recurrences for PgR-negative cases may equalise and that the current finding is only a mere chance finding. However, the decision to analyse by PgR status was made before unblinding of the database rather than while analysing multiple other factors. The plan for this analysis was prompted by results from other studies, for example molecular analysis of the tumours in the ELIOT trial35 had revealed IORT to be less effective with hormone receptor-negative (non-luminal A) tumours and evidence from Perou’s group (Prat et al. 57) presented in 2012 suggested that PgR-negative tumours in particular should not be classified as luminal A at all. Furthermore, statistically, both separate models and an interaction model produced statistically similar results for PgR status, hence minimising the probability that it is a chance finding.

Margin status

We found that margin status was a predictive factor for local recurrence in the EBRT arm (HR 5.17) and overall (HR 2.68). However, margin status did not significantly affect recurrence in the TARGIT arm; although this could be a chance finding (although the p-value = 0.034), it could also be because of the temporal and spatial accuracy of radiotherapy delivery that TARGIT facilitates. It is worth remembering that, although in both arms of the trial > 90% of patients with initial positive margins did undergo re-excision, only 25% of re-excision specimens had tumour present on histology, perhaps indicating inaccurate removal. Alternatively, the proliferation of the remaining tumour cells until the start of EBRT may be responsible for the local recurrence and delay does not occur when TARGIT is given immediately after lumpectomy under the same anaesthetic, so margin positivity did not come out as a factor predicting outcome after TARGIT IORT.

Forest plot

The forest plot in Figure 23 clearly shows how it is mainly ER and PgR receptor status that affected the difference between the two arms. Other factors such as age, grade, tumour size or nodal status did not have an effect; for example, TARGIT was as effective as EBRT in younger patients or those with larger or grade 3 or node-positive tumours.

Suggested modification to American SocieTy for Radiation Oncology and European SocieTy for Radiotherapy & Oncology criteria

The results of the trial suggest that effective cancer control, a reduction in toxicity and improved patient quality of life can be achieved by switching suitable patients to TARGIT, delivered at the time of surgery.

The suitability criteria for PBI that have been suggested by ASTRO142 and the European SocieTy for Radiotherapy & Oncology (ESTRO)143 might be refined based on the TARGIT-A trial results. The ASTRO and ESTRO guidelines would normally exclude many of the patients with ‘adverse’ factors who were in fact eligible for the TARGIT-A trial’s risk-adapted approach, and in the TARGIT-A trial it was found that these ‘adverse’ factors do not predict a worse outcome with TARGIT compared with EBRT. One exception is hormone sensitivity, with TARGIT appearing to work best for patients who are ER and PgR positive.

Table 19 shows how the suitability criteria for PBI may be altered using the data from the TARGIT-A trial.

Executive summary of the analysis provided to the National Institute for Health and Care Excellence

1. This document provides all of the additional analyses requested by NICE. Following the initial request, NICE and the TARGIT-A trial investigators met on 9 December 2014. The TARGIT-A investigators gave a response on 18 December. NICE responded on 16 February 2015, which was followed by a meeting on 3 March to discuss and confirm that all of the required analyses had been included.

2. It is well established that the peak hazard of recurrence is in the first 2–3 years. Most importantly, the effect of radiotherapy on local recurrence is limited to the first 5 years, with most of the radiotherapy effect being seen in the first 2–3 years.

3. The results in the TARGIT-A trial were obtained despite the fact that the eligibility criteria and cases in the trial were not limited to very-low-risk cases (between 450 and 550 patients had a grade 3 tumour or a tumour size > 2 cm or node-positive disease).

4. With over 1200 patients with a median follow-up of 5 years, there were very few events for local recurrence (n = 34), suggesting that both treatments work very well; on the other hand, the number of events for local recurrence-free survival was substantial (n = 120) and similar with TARGIT and EBRT, giving sufficient data to change clinical practice.

5. The Kaplan–Meier curves for local recurrence, survival without local recurrence, overall survival, breast cancer survival and disease-free survival all demonstrated that the lines representing TARGIT and EBRT with their 95% CI overlap each other.

6. On the other hand, the 95% CIs for the TARGIT and EBRT curves for non-breast-cancer survival do not overlap, demonstrating the previously published statistically significant difference.

7. Absolute difference between Kaplan–Meier curves:

   1. the appropriate method of calculating the difference between Kaplan–Meier curves was by using the integrated difference of the two survival functions to quantify the difference between the Kaplan–Meier curves

   2. for the primary outcome of local recurrence, the difference between the Kaplan–Meier curves for TARGIT and EBRT from 0 to 5 years was as follows:

      • for the whole trial 0.62% (95% CI 0.007% to 1.24%)

      • for the prepathology stratum 0.3% (95% CI –0.4% to 1.03%).

8. Data from further follow up: as of October 2014, the number of patients with a minimum follow-up of 5 years in the whole trial was 1116. For the prepathology stratum this number was 776, with 15 new events of local recurrence in addition to the previous 16. We were told that unplanned unblinding for new analysis was not necessary. Therefore, we have remained blind to the apportioning of the new events and have provided a blinded analysis. In the most plausible hypothetical scenario, weighted against TARGIT the new local recurrences might be distributed as TARGIT 10, EBRT 5. In this case, the difference between the binomial proportions of the two arms would be 0.83 (90% CI 0.0 to 1.6, 95% CI –0.1 to 1.8; P_{non-inferiority} = 0.00038) and TARGIT would remain non-inferior to EBRT.

9. There was no significant difference between TARGIT during lumpectomy and EBRT by conventional log-rank test (p = 0.31) and TARGIT was non-inferior to EBRT using the standard test for non-inferiority (P_{non-inferiority} < 0.00001) and the 5-year survival without local recurrence was 93.9% (95% CI 90.9% to 95.9%) for TARGIT and 92.5% (95% CI 89.7% to 94.6%) for EBRT (p = 0.35).

10. Mortality cannot be ignored when a difference is generated within a randomised trial (TARGIT 29 vs. EBRT 42 deaths), particularly when the number of events is several times higher than the number of local recurrence (71 deaths vs. 16 local recurrences) and death is clearly an important outcome. If a statistically significant difference is found at the time of analysis initial ‘power’ calculations are not relevant.

11. Cosmetic outcome and quality of life have been shown to be better with TARGIT than with EBRT as per the published randomised data from the TARGIT-A trial.

12. In conclusion, using three quite separate statistical methodologies, the risk-adapted approach using single-dose TARGIT IORT given during lumpectomy is found to provide breast cancer control that is not inferior to that seen with several weeks of conventional radiotherapy; it also provides a more cost-effective treatment that is more convenient for patients.

13. Figure 37 provides a pictogram to help patients and doctors make a shared well-informed decision.

FIGURE 37.

A pictogram to help patients and doctors to make a shared, well-informed decision.

Important considerations about the use of radiotherapy for breast cancer

Breast cancer surgery has evolved from being very radical (e.g. radical mastectomy and axillary clearance) to being more individualised and precise (lumpectomy and sentinel node biopsy).

Even though randomised evidence supported less radical treatment, there was strong initial opposition to its adoption only a few decades ago. The TARGIT-A trial has provided evidence that radiotherapy also does not need to be radical and should be more precisely and individually optimised. The opposition to its adoption appears to be similar to that shown by strong proponents of radical mastectomy towards those who wanted to spare women this mutilating operation. Some important and relevant points about the natural history of breast cancer have been elucidated via the results of several randomised controlled trials:

1. The peak hazard of recurrence is in the first 2–3 years and, more importantly, the effect of radiotherapy on local recurrence is limited to the first 5 years, with most of the radiotherapy effect being seen in the first 2–3 years. 4,119,121,122 This is clearly seen from the figures below. Thus, local recurrence occurring between 5 and 25 years is no more frequent in non-irradiated patients than in those who have received radiotherapy. 4,119,121,122

   [Parts of Figures 33 and Figures 34 that were included in the correspondence are reproduced here for ease of reading.] These figures demonstrate how the reduction in local recurrence by radiotherapy occurs only in the first 5 years, with most of the effect already seen in the first 2–3 years. Figure 33b (top figure) is the Kaplan–Meier plot from the landmark NSABP B06 trial by Fisher et al. 119 of radiotherapy compared with no radiotherapy after lumpectomy. Figure 34a (middle figure) is the Kaplan–Meier plot taken from the Swedish trial by Wickberg et al. 121 of radiotherapy compared with no radiotherapy after lumpectomy with a 25-year follow-up. Figures 34b and c (bottom figures) are also based on the data from the same Swedish trial by Wickberg et al. 122

   Therefore, the available follow-up in the TARGIT-A trial provides enough information to enable its use in clinical practice.

2. When the reduction in local recurrence because of radiotherapy is < 10%, there is no discernible benefit for survival from breast cancer.

3. The detrimental effect of conventional whole-breast radiotherapy on non-breast-cancer deaths (e.g. from cardiac disease/other cancers) becomes more important when deaths from breast cancer are few and this effect starts in the first few years.

4. It has been suggested that TARGIT treatment is as good as ‘no radiotherapy’. Table 20 provides the results of randomised controlled trials testing the effect of completely omitting radiotherapy. As can be seen clearly in the table, one in every 17–25 of even the most stringently selected low-risk patients would have a local recurrence if radiotherapy were omitted. 11,106,129 On the other hand, when TARGIT is given during lumpectomy local recurrence is rare – one in 48, which reduces to one in 71 when just a single selection criterion (ER positive) is applied.

5. Importantly, TARGIT-A trial eligibility was not limited to ‘good prognosis’ cases. In fact, 85% of patients in the TARGIT-A trial were aged < 70 years and there were a large number of patients in each of the adverse prognostic groups such as node positive (n = 502), ER or PgR negative (n = 554), grade 3 (n = 459) or tumour size > 2 cm (n = 397); > 60% of cases in the TARGIT-A trial would be considered ‘unsuitable’ or ‘cautionary’ by the ASTRO criteria for PBI;172 and only 17.5% of patients in the TARGIT-A trial prepathology stratum would have been eligible for the PRIME II trial – all others (82.5%) had ‘worse’ prognosis cancers. The results in the trial were obtained despite the fact that the eligibility criteria and cases in the TARGIT-A trial were not limited to very-low-risk cases.

6. In a non-inferiority trial, if the difference between the treatments being tested (and its upper confidence limit) is less than a preset non-inferiority margin the treatments are considered non-inferior, even if the difference is ‘statistically significant’ using a log-rank test. The pre-specified non-inferiority boundary of a 2.5% absolute difference in local recurrence is very conservative and has been validated in patient preference studies. 101,103,104 At this boundary, TARGIT is non-inferior to EBRT (P_{non-inferiority} < 0.00001) (Figure 38). Furthermore, we have recommended that TARGIT should be used during the initial lumpectomy, as in the prepathology stratum (note that this is a stratum not a subgroup), when the difference between the two treatments was undoubtedly not statistically significant (p = 0.31).

FIGURE 38.

The meaning of non-inferiority: 10 examples of different scenarios that might occur in a randomised trial testing non-inferiority between two treatments. The dots represent the absolute difference and lines the CIs. The blue circle includes two of the trial results: on the left is the TARGIT prepathology stratum (difference 0.37%) and on the right is the earliest cohort of the whole trial (n = 1222), who have a median follow-up of 5 years (difference 1.14%) (see table 3 in Vaidya et al. 2). 2,102

Whole-study population: local recurrence

1. Local recurrence:

   1. The absolute number of local recurrence events (n) (Table 21).

   2. A Kaplan–Meier analysis including all patients from TARGIT-A for each treatment group showing the cumulative risk of local recurrence over time using the latest available follow-up data. The requested figures for this survival analysis are given below, with or without censor lines and 95% CI curves (Figures 39–43). Figure 44 shows the 5-year survival without local recurrence. The numbers of patients at risk of local recurrence in each treatment group at yearly intervals are reported below the plot.

   3. The absolute difference in the Kaplan–Meier estimate of the 5-year risk of local recurrence between treatment groups and the 95% CI around that difference.

      • In the paper,2 we reported the 90% CI of the difference in binomial proportions, not of the difference in Kaplan–Meier estimates (for non-inferiority testing the convention is to use the 90% CI rather than the 95% CI105,173–175). The difference between the binomial proportions of local recurrence for TARGIT and EBRT was 0.72% (90% CI 0.15% to 1.30%) (95% CI 0.05% to 1.40%).

      • In the presence of censoring, the Kaplan–Meier point estimate at a particular time point, for example 5 years, is not a simple binomial proportion. Therefore, it is inappropriate to apply the simple formula normally used to calculate the SE (and CI) of a difference between binomial proportions, namely SE12+SE22, to calculate differences between such point estimates.

      • Furthermore, when looking at Kaplan–Meier curves, the right-hand end of the curve is the one with the most uncertainty and with the widest CIs. These values, that is, 5-year point estimates, should not be used to calculate the CI of the difference or for testing non-inferiority.

National Institute for Health and Care Excellence agreed that the differences in 5-year Kaplan–Meier estimates are not to be used for assessing non-inferiority, but are provided for completion. For local recurrence the difference is 2% (95% CI –0.14% to 4.14%) (90% CI 0.18% to 3.82%). For survival without local recurrence the difference is 0.64% (95% CI –2.09% to 3.37%) (90% CI –1.69% to 2.97%).

The appropriate method for calculating the difference between Kaplan–Meier curves is by using the integrated difference of the two survival functions to quantify the difference between the Kaplan–Meier curves. 176 The difference between the Kaplan–Meier curves for TARGIT and EBRT from 0 to 5 years is 0.62% (95% CI 0.007% to 1.24%).

Whole-study population: survival

1. Survival (whole trial).

   1. The absolute number of deaths (n) (Table 22).

   2. The number of patients with different causes of death (n) (Table 23).

   3. Kaplan–Meier analysis of mortality:

      A Kaplan–Meier analysis including all patients from TARGIT-A for each treatment group showing the cumulative risk of death over time using the latest available follow-up data. The two requested figures for this survival analysis are given below, with or without 95% CI curves (Figures 44 and 45).

   4. The absolute difference in the Kaplan–Meier estimate of the 5-year risk of overall mortality between treatment groups (TARGIT and EBRT) in the whole-study population and the 95% CI around that difference. Note that the caveats mentioned in Whole-study population: local recurrence about using 5-year estimates to calculate the difference between treatments and its CI should be read before using these figures. The difference between 5-year Kaplan–Meier estimates is –1.38% (95% CI –3.67% to 0.91%) (90% CI –3.3% to 0.57%). Using the integrated difference of the two survival functions to quantify the difference between the Kaplan–Meier curves,176 the difference between the Kaplan–Meier curves for TARGIT and EBRT from 0 to 5 years is –0.85% (95% CI –1.75% to 0.04%).

1. Breast cancer mortality (whole trial).

   1. The absolute number of breast cancer deaths (n) (Table 24).

   2. Kaplan–Meier analysis of breast cancer deaths.

      A Kaplan–Meier analysis including all patients from TARGIT-A for each treatment group showing the cumulative risk of breast-cancer death over time using the latest available follow-up data. The two requested figures for this survival analysis are given below, with or without 95% CI curves (Figures 46 and 47).

   3. The absolute difference in the Kaplan–Meier estimate of the 5-year risk of breast cancer mortality between treatment groups (TARGIT and EBRT) in the whole-study population and the 95% CI around that difference. Note that the caveats mentioned in Whole-study population: local recurrence about using 5-year estimates to calculate the difference between treatments and its CI should be read before using these figures. The difference between 5-year Kaplan–Meier estimates is 0.67% (95% CI –1.01% to 2.35%) (90% CI –0.76% to 2.10%). Using the integrated difference of the two survival functions to quantify the difference between the Kaplan–Meier curves,176 the difference between the Kaplan–Meier curves for TARGIT and EBRT from 0 to 5 years is 0.15% (95% CI –0.71% to 0.42%).

1. Non-breast-cancer mortality (whole trial).

   1. The absolute number of non-breast-cancer deaths (n) (Table 25).

   2. Kaplan–Meier analysis of non-breast-cancer mortality:

      A Kaplan–Meier analysis including all patients from TARGIT-A for each treatment group showing the cumulative risk of non-breast-cancer death over time using the latest available follow-up data. The two requested figures for this survival analysis are given below, with or without 95% CI curves (Figures 48 and 49).

   3. The absolute difference in the Kaplan–Meier estimate of the 5-year risk of non-breast-cancer mortality between treatment groups (TARGIT and EBRT) in the whole-study population and the 95% CI around that difference. Note that the caveats mentioned in Whole-study population: local recurrence about using 5-year estimates to calculate the difference between treatments and its CI should be read before using these figures. The difference between 5-year Kaplan–Meier estimates is –2.08% (95% CI –3.70% to –0.46%) (90% CI –3.46% to –0.70%). Using the integrated difference of the two survival functions to quantify the difference between the Kaplan–Meier curves,176 the difference between the Kaplan–Meier curves for TARGIT and EBRT from 0 to 5 years is –0.72 (95% CI –1.42 to –0.02).

1. Tabulation of the number of patients with at least 5 years of follow-up data (Table 26).

Prepathology stratum: local recurrence

1. Local recurrence.

   1. The absolute number of local recurrence events (n) (Table 27).

   2. A Kaplan–Meier analysis including all patients in the prepathology stratum of the TARGIT-A trial for each treatment group showing the cumulative risk of local recurrence over time using the latest available follow-up data. The requested figures for this survival analysis are given below, with or without censor lines and 95% CI curves (Figures 50–53). Figure 54 shows the 5-year survival without local recurrence. The numbers of patients at risk of local recurrence in each treatment group at yearly intervals are reported below the plot.

   3. The absolute difference in the Kaplan–Meier estimate of the 5-year risk of local recurrence between treatment groups and the 95% CI around that difference.

      • In the paper,2 we reported the 90% CI of the difference in binomial proportions, not of the difference in Kaplan–Meier estimates (for non-inferiority testing the convention is to use the 90% CI rather than the 95% CI105,173–175).

      • The difference between binomial proportions of local recurrence for TARGIT and EBRT was 0.37% (90% CI –0.23% to 0.97%) (95% CI –0.33% to 1.07%).

      • In the presence of censoring, the Kaplan–Meier point estimate at a particular time point, for example 5 years, is not a simple binomial proportion. Therefore, it is inappropriate to apply the simple formula normally used to calculate the SE (and CI) of a difference between binomial proportions, namely SE12+SE22, to calculate differences between such point estimates.

      • Furthermore, when looking at Kaplan–Meier curves, the right-hand end of the curve is the one with the most uncertainty and with the widest CIs. These values, that is, 5-year point estimates, should not be used to calculate the CI of the difference or for testing non-inferiority.

National Institute for Health and Care Excellence agreed that the differences in 5-year Kaplan–Meier estimates are not to be used for assessing non-inferiority, but are provided for completion. For local recurrence the difference is 1% (95% CI –0.68% to 2.68%) (90% CI –0.43% to 2.43%). For survival without local recurrence the difference is –1.32% (95% CI –4.74% to 2.10%) (90% CI –4.24% to 1.60%).

If the difference in Kaplan–Meier estimates is being used to assess the difference between treatments (despite the concerns about using these), then although the upper confidence limit of the 95% CI of the difference in local recurrence is 2.68%, the absolute difference in overall mortality between TARGIT and EBRT is –2.33% (95% CI –5.48% to 0.82%), favouring TARGIT. The lower confidence limit of the 95% CI for overall mortality is –5.48 and this is a very much larger value than for local recurrence and is in the opposite direction – for a much more important outcome. It would mean that, although at worst the difference in local recurrence might be 2.68% favouring EBRT, the difference in overall mortality might be 5.48% favouring TARGIT. One cannot consider one and ignore the other value.

In addition, while discussing the actual difference between recurrence rates (not for non-inferiority determination), both the 95% confidence limits need to be looked at, not just one. If the upper 95% confidence limit is considered, that is, 2.68%, as the worst-case scenario, then the lower 95% confidence limit (–0.68%) should also be considered as it has exactly the same chance of occurring.

The chance of being alive without local recurrence is the most relevant statistic (there cannot really be a local recurrence if the patient is no longer alive). For this statistic, the two treatments have very similar outcomes; the number of events is also large and similar in the two arms [TARGIT 57 vs. EBRT 59 for the whole trial (n = 3451) and TARGIT 36 vs. EBRT 45 for the prepathology stratum (n = 2298): the number of events without excluding those who had a mastectomy were 120 (59 vs. 61) in the whole trial and 85 (TARGIT 38 vs. EBRT 47) in the prepathology stratum].

The appropriate method of calculating the difference between Kaplan–Meier curves is by using the integrated difference of the two survival functions to quantify the difference between the Kaplan–Meier curves. 176 The difference between the Kaplan–Meier curves for TARGIT and EBRT from 0 to 5 years is 0.3% (95% CI –0.4% to 1.03%).

Prepathology stratum: survival

1. The absolute number of deaths (n) (Table 28).

2. The number of patients with different causes of death (n) (Table 29).

3. Kaplan–Meier analysis of survival:

   A Kaplan–Meier analysis including all patients in the prepathology stratum of the TARGIT-A trial for each treatment group showing the cumulative risk of death over time using the latest available follow-up data. The two requested figures for this survival analysis are given below, with or without 95% CI curves (Figures 55 and 56).

4. The absolute difference in the Kaplan–Meier estimate of the 5-year risk of overall mortality between treatment groups (TARGIT and EBRT) in the whole-study population and the 95% CI around that difference. Note that the caveats mentioned in Whole-study population: local recurrence about using 5-year estimates to calculate the difference between treatments and its CI should be read before using these figures. The difference between the 5-year Kaplan–Meier estimates is –2.33% (95% CI –5.48% to 0.82%) (90% CI –5.02% to 0.36%). Using the integrated difference of the two survival functions to quantify the difference between the Kaplan–Meier curves,176 we found that the difference between the Kaplan–Meier curves for TARGIT and EBRT from 0 to 5 years is –1.43% (95% CI –2.66% to –0.2%).

1. Breast cancer mortality (prepathology stratum).

   1. The absolute number of breast cancer deaths (n) (Table 30).

   2. Kaplan–Meier analysis of survival:

      A Kaplan–Meier analysis including all patients in the prepathology stratum of the TARGIT-A trial for each treatment group showing the cumulative risk of breast cancer death over time using the latest available follow-up data. The two requested figures for this survival analysis are given below, with or without 95% CI curves (Figures 57 and 58).

   3. The absolute difference in the Kaplan–Meier estimate of the 5-year risk of breast cancer mortality between treatment groups (TARGIT and EBRT) in the whole-study population and the 95% CI around that difference. Note that the caveats mentioned in Whole-study population: local recurrence about using 5-year estimates to calculate the difference between treatments and its CI should be read before using these figures. The difference between the 5-year Kaplan–Meier estimates is 0.66% (95% CI –1.72% to 3.04%) (90% CI –1.36% to 2.68%). Using the integrated difference of the two survival functions to quantify the difference between the Kaplan–Meier curves,176 we found that the difference between the Kaplan–Meier curves for TARGIT and EBRT from 0 to 5 years is –0.34% (95% CI –1.17% to  0.49%).

1. Non-breast-cancer mortality (prepathology stratum).

   1. The absolute number of non-breast-cancer deaths (n) (Table 31).

   2. Kaplan–Meier analysis of survival:

      A Kaplan–Meier analysis including all patients in the prepathology stratum of the TARGIT-A trial for each treatment group showing the cumulative risk of non-breast-cancer death over time using the latest available follow-up data. The two requested figures for this survival analysis are given below, with or without 95% CI curves (Figures 59 and 60).

   3. The absolute difference in the Kaplan–Meier estimate of the 5-year risk of non-breast-cancer mortality between treatment groups (TARGIT and EBRT) in the whole-study population and the 95% CI around that difference. Note that the caveats mentioned in Whole-study population: local recurrence about using 5-year estimates to calculate the difference between treatments and its CI should be read before using these figures. The difference between the 5-year Kaplan–Meier estimates is –3.07% (95% CI –5.25% to 0.89%) (90% CI –4.93% to 1.21%). Using the integrated difference of the two survival functions to quantify the difference between the Kaplan–Meier curves,176 we found that the difference between the Kaplan–Meier curves for TARGIT and EBRT from 0 to 5 years is –1.11% (95% CI –2.05% to –0.17%).

1. Tabulation of the number of patients with at least 5 years of follow-up data (Table 32).

New analysis

1. During the meeting of 8 December 2014, we presented an additional analysis of the data published in The Lancet,2 namely that the disease-free survival of the patients in the TARGIT and EBRT arms is identical for the whole trial (p = 0.78) and for the prepathology stratum (p = 0.68), with the Kaplan–Meier curves overlapping each other (Figure 61).

FIGURE 61.

Ten-year Kaplan–Meier estimates for disease-free survival for (a) all patients; and (b) the prepathology stratum [TARGIT 91.6% (95% CI 88.7% to 93.8%) vs. EBRT 90.1% (95% CI 86.8% to 92.6%) at 5 years and TARGIT 81.3% (95% CI 71% to 88%) vs. EBRT 71.2% (95% CI 49% to 85%) at 10 years].

1. As it was made clear to us that the NICE committee did not expect us to unblind the trial at this moment, we presented new updated data resulting from increased follow-up as of 1 October 2014, using the total number of events without unblinding the trial.

2. Further follow up: at the time of this analysis, the median follow-up was 4 years, which means that a very large number of patients (n = 1725) had at least 4 years of follow-up or longer. This number is large compared with the majority of breast cancer trials.

3. New events: with additional follow-up since the last data lock in June 2012, there was a total of 15 new local recurrences in the prepathology stratum. This was in addition to the 16 already reported in The Lancet,2 out of a total number of 2298 patients.

4. Remaining blind to the randomisation arm, we presented two hypothetical scenarios – one worst-case scenario and one less extreme but still weighted against TARGIT.

5. For the 15 total new events of local recurrence, rather than an even split of seven compared with eight or eight compared with seven, remaining blind to the randomisation arm we modelled the events as in Tables 33 and 34.

6. For death, the secondary end point, there were 28 new events and these would need to have occurred in a ratio of 20 in the TARGIT arm to 8 in the EBRT arm to equalise the total number of deaths between the two treatments (Table 35). As the initial observation was 29 deaths in the TARGIT arm and 42 in the EBRT arm, the probability of such a drastic reversal is low (p = 0.008) and so the difference in deaths is likely to remain in favour of TARGIT.

   Although NICE clarified that mortality was not the remit of this committee, we are compelled to emphasise the fact to your attention because it arose during the discussion. It is important to recognise that, when a study actually finds a difference, the question of power is not relevant any more. Then, the probability of observing the difference favouring TARGIT that was observed, if there was no real difference between the two groups is given by the p-value, which in the case of the TARGIT-A trial was 0.099 for all deaths and 0.0086 for deaths from causes other than breast cancer. We believe we cannot ignore these randomised data, particularly when deaths were more frequent than recurrences (n = 88 vs. n = 34) and non-breast-cancer deaths (n = 52) were more frequent than breast cancer deaths (n = 36) or local recurrences (n = 34).

7. We were asked by the NICE committee whether or not we would offer INTRABEAM to patients in the control arm of the TARGIT-A trial who have already received EBRT if NICE gives a positive response. In fact, the surgery and radiation treatment of patients in the TARGIT-A trial (whether INTRABEAM or whole-breast radiotherapy) has already been long completed (the trial closed in June 2012). Therefore, all those who received EBRT in the TARGIT-A trial cannot and will not be offered an operation to reopen their wound and give INTRABEAM.

8. Please note that, as published in The Lancet,2 there is no significant difference between TARGIT and EBRT by conventional log-rank test (p = 0.31) and TARGIT was non-inferior to EBRT using the standard test for non-inferiority (P_{non-inferiority} < 0.00001).

The TARGIT-A trial investigators offered to supply the raw data as long as all of the governance, consent, custody, data access and security issues were looked after appropriately. NICE declined, saying that it was not necessary to supply the raw data because all of the analysis requested by the committee had been satisfactorily supplied.

Table 36 shows the results of the four different methods of analyses and the raw numbers, which essentially show very similar results.

In conclusion, using four quite separate statistical methodologies, the TARGIT-A trial has demonstrated that the risk-adapted approach using single-dose TARGIT IORT given during lumpectomy provides breast cancer control that is not inferior to several weeks of conventional radiotherapy.

Acknowledgement

We thank Dr Hajime Uno PhD (Department of Biostatistics, Harvard University, Boston, MA, USA) for providing the software code and for independently verifying our results and our interpretation.

Aim and rationale

The latest analysis of the TARGIT-A trial includes a very large number of patients (n = 1222) with a median follow-up of 5 years and our analysis suggests that the results remain stable for cohorts of patients with increasing periods of follow-up up to 5 years. However, for the whole trial, the median follow-up is 2.6 years and one of the barriers to widespread adaptations of the new treatment appears to be the perception that we should undertake 5-year follow-up of all of the patients in the trial. This will mean that a substantial number will also have 10-year follow-up, a milestone that is now considered essential in many trials. Importantly, follow-up up to 10 years was stipulated in the original protocol.

Work package 1 will deliver this.

Primary and secondary objectives

To evaluate in the longer term the primary outcome of local recurrence in the conserved breast and secondary outcomes of the main trial, including complications/side effects and breast cancer and non-breast-cancer survival.

Aim and rationale

This work package is designed to investigate whether the two methods of radiotherapy utilised in the TARGIT trial are associated with a longer-term difference in quality of life, shoulder and arm function or lymphoedema.

Primary and secondary objectives

To evaluate the longer-term side effects (particularly arm and shoulder function and lymphoedema) and quality of life in patients randomised within the TARGIT trial to one of two methods of radiotherapy after surgical treatment of early breast cancer.

Primary and secondary objectives

To evaluate the costs and benefits experienced by patients in the trial using the following tools:

• European Quality of Life-5 Dimensions five-level questionnaire (EQ-5D 5L)

• cost data:

  • capital costs associated with the TARGIT equipment (completed)

  • NHS contacts for receipt of radiotherapy (completed)

  • NHS contacts for treating the side effects of radiotherapy

  • costs incurred by patients and families.

Detailed description

This will be a cross-sectional analytical study to determine what increased risk of local recurrence women who have completed postoperative radiotherapy for early breast cancer are prepared to accept in return for the increased convenience and altered toxicity profile of TARGIT and the factors that influence these preferences, to guide women and their doctors in making choices about radiotherapy after surgery for early breast cancer.

This work package is based on a study successfully run in Sir Charles Gairdner Hospital, Perth, WA, Australia. It will address the patient preference subprotocol of TARGIT-A.

Aim and rationale

Patient preferences for adjuvant radiotherapy in early breast cancer, a substudy of the TARGIT-A trial.

Aim and rationale

The conventional treatment of early breast cancer involves surgical excision of the tumour, surgery to the axillary lymph nodes and EBRT to the conserved breast tissue. Although this is an effective treatment with a low rate of local recurrence of cancer, recent developments have explored alternatives to EBRT.

The TARGIT trial studied the use of single-dose IORT compared with conventional radiotherapy. The trial has demonstrated equivalence with both study arms having a low local recurrence rate and acceptable acute toxicity.

This work package is designed to investigate whether the two methods of radiotherapy utilised in the TARGIT trial have adversely affected the heart.

Primary and secondary objectives

To evaluate the cardiac and respiratory outcomes in patients randomised within the TARGIT trial to one of two methods of radiotherapy after surgical treatment of early breast cancer.

NICE to recommend new breast cancer radiotherapy treatment alongside further research

1. Radiotherapy preparation for targeted intraoperative radiotherapy using the INTRABEAM device

• X68.2 – Preparation for intracavitary brachytherapy (as currently used).

This is because the treatment involves irradiation of tissues immediately next to the applicator within a surgically prepared cavity (tumour bed) following wide local excision of breast cancer. The correct size of the applicator is chosen by measuring the size of the cavity.

As there is no specific code in the book at present, this code is chosen because it best approximates the TARGIT procedure. However, the cavity that is prepared is different from the physiological cavities that this code is usually meant for (e.g. thoracic, abdominal, pelvic, oral cavity).

In the next edition of OPCS-4, it would be preferable to describe this new surgical technique with a new code, for example X68.4 – Preparation for giving targeted intraoperative radiotherapy using INTRABEAM. Until then, X68.2 should be used.

2. Surgical preparation for targeted intraoperative radiotherapy using the INTRABEAM device (including insertion of the applicator)

The codes that we would suggest would be:

• B37.8 – Other specified other operations on breast, plus

• Y258 – Other specified suture of organ not otherwise classified (NOC), plus

• Y02.8 – Other specified placement of prosthesis in organ NOC.

This is because, after wide local excision of breast cancer, the tumour bed needs to be prepared. First, complete haemostasis needs to be achieved as small amounts of oozing blood can affect the treatment. Then, a carefully placed purse-string suture that is taken in such a manner that the tissues closely appose the spherical applicator without bringing the dermis/skin or the ribs too close to the applicator surface. The applicator is attached to the X-ray source, which is covered with a sterile drape, and is then placed within the tumour bed. The purse string is then tightened. After delivery of the radiation is complete, the purse string is cut and the applicator is removed and the tumour bed inspected. The breast is then closed in the usual manner.

In the next edition of the OPCS-4, it would be preferable to describe this new surgical technique with new codes:

• B37.6 – Preparation of tumour bed for giving targeted intraoperative radiotherapy (TARGIT) with INTRABEAM, plus

• Y02.3 – Placement of INTRABEAM applicator for giving targeted intraoperative radiotherapy (TARGIT).

Until then, B37.8 plus Y25.8 plus Y02.8 should be used.

3. Delivery of targeted intraoperative radiotherapy using the INTRABEAM device

The codes should be:

• X65.2 – Delivery of a fraction of intracavitary radiotherapy (as currently used), plus

• Y35.8 – Other specified introduction of removable radioactive material into organ NOC, plus

• Y89.8 – Other specified brachytherapy, plus

• Z15.– – Breast (as currently used).

This is because, after the applicator is placed in the tumour bed, single-dose radiation is delivered to the tumour bed. This takes between 15 and 35 minutes depending on the size of the applicator/tumour bed. The radiation source is X-rays, which are emitted only after the radiation is switched on. Once the radiation has been delivered, the radiation is switched off.

First, the word ‘fraction’ in X65.2 might imply that this treatment is part of a course of several fractions (5–15); however, TARGIT using INTRABEAM is always given as a single dose and therefore, ideally, a new code should be introduced in the next edition of the coding manual, for example X69.1 – Delivery of single dose of targeted intraoperative radiotherapy (TARGIT) using INTRABEAM.

Second, this is not strictly a ‘radioactive material’ as specified in Y35.8; however, as there is no code to describe the introduction of a X-ray source into an organ and, in reality, the source does emit radiation when switched on, this code should be used until a new code is introduced in the next edition of the OPCS-4 to describe the procedure. This could be achieved by removing the word ‘radioactive’ from the three category heading at Y35, for example Y35.5 – Introduction of low-energy/low-dose rate X-ray source – INTRABEAM – into breast.

Third, the additional code of Y89.8 should be used to specify that this is a brachytherapy (internal radiotherapy) that is neither high dose nor pulsed dose until a new code is introduced in the update of the OPCS-4, for example Y89.3 – Single low-dose brachytherapy treatment with INTRABEAM.

Until then the X65.2 plus Y35.8 plus Y89.8 plus Z15.– codes should be used.

Ideally, there should be a single new code for the procedure of ‘Wide local excision of breast cancer + TARGIT-IORT’ because the latter cannot be given without initially removing the tumour. This will allow easier coding the classification.