First, a significant part of our patient population is aged 65 or older. These patients would also fit all criteria for completely omitting breast radiotherapy as attempted in clinical trials [29,30,31]. These trials showed that omitting WBI resulted in no difference in overall survival in this select population. Yet, a significant increase in IBTR was demonstrated in all trials with a risk difference of about 6% at 10 years in favour of WBI. Thereafter, some guidelines were modified to include RT omission in elderly patients [4, 32]. Some might thus think that RT for elderly patients with early stage breast cancer is no longer needed. However, it should be noted that while WBI is most likely an overtreatment for most of this patient population, APBI was never tested as a comparator in these clinical trials. We would argue that APBI and IORT techniques in particular could well become the best treatment modality for these patients. Indeed, these immediate, one-shot RT treatments still allow for a strong reduction in treatment burden while yielding strong IBTR risk reduction as discussed below.

Our results are aligned with those of the main external and brachytherapy APBI trials [11, 13,14,15, 33,34,35]. While IMPORT LOW has lower five-year IBTR rates than our study, the populations are not similar thus precluding any direct comparison. For example, 23% of our trial population received chemotherapy, 20% had grade 3 tumours against only 5% and 10%, respectively, in IMPORT LOW. The same goes for the GEC-ESTRO trial population. Additionally, we have a relatively similar IBTR rate to both RAPID and the Florence trial, which had populations that more closely resembled our study population. This is particularly significant given that contrary to external APBI, IOERT has insignificant irradiation of organs at risk.

Compared to the populations of the TARGIT-A and ELIOT trials, our study population has far fewer node positive tumours of a size greater than 2 cm or lobular breast cancers [22, 36, 37]. This was expected given the inclusion criteria of this study. Population grade was comparable to both trials. There were more Luminal B-like tumours in ELIOT than in our study. Although there is no molecular subtype data available for TARGIT, surrogate markers (such as ER/PR positivity and tumour grade) are somewhat similar. We included younger patients than both studies given enrolment was open to patients as young as 40.

Interestingly, the results of this study closely resemble those of the TARGIT-A clinical trial IORT arm, one of the two main trials to have compared IORT to WBI. TARGIT-A outlined an IBTR 5-year recurrence risk of 3.3% and overall survival of 96.1%. These results are comparable to the 2.7% and 95.9% rates found by this study [38]. Although the intraoperative radiotherapeutic technique used in our study is different to the one used in TARGIT-A and any comparison should therefore be made carefully, it is interesting to note that similar IBTR and OS outcomes speak in favour of our technique being as safe and efficacious as that of the TARGIT trial when applied to selected populations. It should also be noted that a 2.7% IBTR rate at five years is well within the acceptable non-inferiority margin used in clinical trials comparing APBI with WBI.

Our study also seems to have yielded better IBTR rates than those found in the ELIOT trial (4.4% vs 2.7%). As the IORT technique used in this study is the same as the one in used in the ELIOT trial (IOERT), this is of particular interest. Crucially, differences in lymph node metastasis (1% vs 0.2% in our study) and in distant metastasis (respectively 4.5% vs 1.7%), 5-year rates can also be noted.

We have multiple hypotheses to explain these differences in outcome:

First, the ELIOT trial began enrolment more than a decade before the first patients in our study were treated. During that period, chemotherapy treatment guidelines changed, adopting, among others, taxanes and Trastuzumab into regular chemotherapy regimens for breast cancer. The additional efficacy of these new chemotherapy regimens might have contributed to better results in more aggressive breast cancer subtypes/grade and brought IBTR risk closer to baseline favourable subtype/low grade breast cancer in our study.

Second, investigators in the ELIOT trial enrolled patients with more advanced tumours than in our cohort. Most notably, ELIOT enrolled larger size (up to 2.5 cm), node positive cancers in their trial. Out-of-quadrant dissemination and subsequent IBTR could therefore be more likely. Similarly, lobular breast cancers were enrolled although they are more likely to be multifocal and therefore again risk increasing the IBTR rate [39].

Third, although our radiotherapy technique is similar to ELIOT’s, it is highly operator-dependent. Many variables account for a successful IOERT treatment. Most notably, the treatment critically relies on applicator size and surgical technique to bring the bordering breast tissue inside the irradiation field [40, 41]. The median applicator size in this study is 55 mm compared to 40 mm in ELIOT [42]. In fact, given that the median tumour size in this study is smaller than in the ELIOT trial, the difference should be even greater. Table 4 summarizes the different applicator sizes of both ELIOT and our study. Systematic applicator under sizing has a profound impact on radiotherapy treatment. Out of foci recurrence, for example, has been associated with smaller applicator sizes [43, 44]. In the case of this study, a 15 mm increase in applicator size increases the treated surface by a factor of about 1.89. We would argue this is probably the most important difference between the two studies, and it probably explains most of the difference between ELIOT’s higher IBTR rate and this study.

Much thought had gone into creating a protocol for choosing an appropriate applicator size for patients. We based ours on studies by Holland and Faverly et al. [45, 46] These studies focused on post-mastectomy specimens in the 1980s and showed that a significant number of patients had residual tumour at more than 1 cm from tumour edge. The number of patients having residual tumour tissue dropped by almost half if the surgical margin was increased to 3 cm. We therefore chose—albeit somewhat arbitrarily—to have a total tumour margin of at least 3 cm. This was usually composed of a 1 cm surgical microscopic tumour-free margin and a radiotherapy margin of at least 2 cm. We also introduced a proportional increase in the radiotherapy margin size according to tumour size (Additional file 1: Appendix A). For a 5-mm tumour, for example, the radiotherapy margin was at 2.25 cm and the total margin (surgical + radiotherapy) at 3.25 cm. In the case of a 20-mm tumour, however, these margins would increase to 3 and 4 cm, respectively (Additional file 1: Appendix A).

ELIOT trialists had undertaken a stratified post hoc analysis of factors associated with IBTR [21]. The trial showed on univariate analysis that among others, Luminal B and triple negative molecular subtypes and tumour grade were very strong factors associated with IBTR. Our study, however, has found that while IBTR is strongly associated with patients meeting genetic testing criteria and presenting with plurifocal breast cancer (factors we therefore consider to be relative contraindication to IOERT), grade and molecular subtype were not significantly related to IBTR. In all likelihood, there probably is a slightly increased risk of IBTR with more aggressive cancer subtypes, with increased tumour grade, or with proliferative index, but all these factors are probably also responsible for increased tumour size, and this multicollinearity would perhaps explain part of the non-significance of those variables. As bigger tumours are more likely to have out-of-quadrant foci, tumour size being on the verge of significance in our multivariate analysis therefore seems quite self-explanatory.

Given the results of this study, we hypothesize that tumour size is still a very significant factor of IBTR and that rather than contraindicating some aggressive molecular subtypes or tumour grades, it would be more beneficial to take into account both tumour size and then secondary tumour characteristics in potential guidelines [47, 48]. It should be noted that the post-hoc analyses used in the ELIOT trial to try to find factors associated with IBTR were just exploratory in nature, as were those underpinning our study. Moreover, factors found to be linked to IBTR are very inconsistent across different APBI trials with some finding no link between molecular types or grades and IBTR [16, 49]. Thus, guidelines based on these analyses should always be prospectively assessed.

One main study limitation is that cosmetic evaluations were progressively discontinued given the difficulty in collecting quality data. In total, about 855 patients benefited from a summary long-term skin toxicity and cosmesis assessment. Overall treatment for evaluated patients was demonstrated to be well-tolerated for acute and late toxicity as well as breast cosmesis in keeping with the previously published literature [17, 50,51,52].

This study provides real-life data proving that IOERT using our inclusion criteria yields a low IBTR rate similar to the successful TARGIT IORT clinical trial. It is therefore the conclusion of this study that IOERT guidelines are probably too restrictive and that they could be opened up to allow inclusion of all molecular subtypes/tumour grade, with some restrictions based on tumour size.

Most importantly, this study seeks to highlight the fact that IOERT results in extremely low irradiation to organs at risk due to pectoral wall shielding. Given the significant improvements in breast cancer survival in the past decades, the authors share Vaidya et al.’s opinion that reducing treatment toxicity is now of the upmost importance [18]. Since radiotherapy-related morbidity and mortality is mediated by the irradiation of organs at risk, it is understood that drastically reducing the dose given to those organs with IOERT will reduce non-subcutaneous tissue-related toxicity [19, 20]. It follows that when comparing IOERT to WBI, additional local recurrences can be tolerated, if the patient wills it, as long as they do not affect breast cancer associated mortality. This is especially true as local recurrence does not seem to be as strong an indicator of poor prognosis in IORT as it has been in WBI [53]. Such a trade-off is therefore justified in light of the net gain in overall mortality that can be expected on the basis of lower non-breast cancer mortality [18]. We are of the opinion that overall we should take care not to harm patients by overtreating them with WBI when other IORT treatments are available and have demonstrably no negative impact on breast-cancer mortality.

Nevertheless, we must stress the importance of appropriate and exhaustive patient workup with MRI imaging by a specialized breast radiologist, timely pathological examination of the core biopsy and tumour frozen section as key elements contributing to these good results. If, at any point during patient workup, there was some lingering doubt about IOERT suitability, we would not proceed with the technique.