A systematic review and meta-analysis of intraoperative electron radiation therapy delivered with a dedicated mobile linac for partial breast irradiation in early breast cancer

Graphical abstract

Highlights

• •PBI is valid alternative to WBI in patients at low-risk of local relapse.
• •PBI is delivered by means of various techniques, one of which is IOeRT.
• •After IOeRT, an unexpectedly high LR rate was observed in the only phase III RCT.
• •Patient selection impacts on LR rates after IOeRT.
• •With appropriate patient selection IOeRT outcomes overlap with other RT techniques.

1. Introduction

Breast conserving therapy (i.e. breast conserving surgery (BCS) followed by whole breast irradiation (WBI)) was, for decades, the standard of care in early breast cancer. A valid alternative to WBI in patients who are at low-risk of local relapse is partial breast irradiation (PBI) [[1], [2], [3], [4]], which is performed by means of various techniques (brachytherapy, external beam radiotherapy (EBRT), intraoperative RT (IORT)).

Compared with other PBI techniques, IORT provides the following advantages: 1) it is administered immediately after tumour removal so that surgery and radiation therapy are completed in a one-step session; 2) the intra-operative direct view of the tumour bed lowers the risk of geographic miss; 3) the skin receives a lower RT dose; 4) organs at risk (OARs) also receive a reduced dose and 5) there is no RT delay until after chemotherapy administration. IORT requires, however, close collaboration between the breast surgeon and the radiation oncologist as surgeons may opt for a small incision for cosmetic reasons but need to be aware this may preclude the insertion of appropriately-sized collimators for adequate coverage of the target volume. For radiation oncologists, the more patients that are treated with IORT, the shorter the waiting list for EBRT. From the patient's point of view, IORT improves treatment compliance by removing common inconveniences related to daily radiation sessions such as issues with transportation, work, family duties and (indirect) costs.

IORT is delivered by means of orthovoltage (i.e. with a dedicated device which produces low-energy (up to 50 KV) photon) [5], inherently limiting adequate dose coverage of the target volume, or by electrons (IOeRT) as produced by conventional linear accelerators (linacs) or mobile linacs that are dedicated to operating theatres. Despite the cost of a mobile linac which might have, so far, precluded its widespread use in many radiation-oncology centres, IOeRT is easy to perform as it does not impact on the routine workflow of the radiation oncology centre. Without dedicated facilities, patients have to be transported under anesthesia from the operating suite to the radiation oncology centre where sterility has to be ensured. Whether IOeRT is, however, delivered by a conventional or dedicated mobile machine, physicians, physicists and technicians still have to be assigned to its sessions.

Some specific concerns have arisen with outcomes after IOeRT. Depending on the patient's characteristics and follow-up, local relapse rates even reached an unexpectedly high 10% or more, according to some reports that were prevalently retrospective, prospective observational or phase II studies [[6], [7], [8], [9], [10], [11], [12], [13], [14]]. Other concerns were raised about side effects and cosmetic outcomes.

The present meta-analysis and systematic review of outcome parameters in early breast cancer patients who were treated with IOeRT, as delivered by mobile linacs, was conducted to address these concerns and establish whether IOeRT was a valid option for patients with an indication to PBI.

2. Methods

2.1. Search strategy

A search encompassing the period from January 01, 1980 and April 30, 2023 was conducted in 5 databases: PubMed, Scopus, Web of Science, Science Direct and Cochrane. The search strategy was developed by CA, PP and SB who are experts in breast cancer treatment and the search was conducted by CA, SB, FC, VE, IP, PP. Mesh were “Accelerated partial breast irradiation AND breast cancer AND intraoperative radiotherapy”, “Accelerated partial breast irradiation AND breast cancer AND IORT”, “APBI AND breast cancer AND IORT”, “Accelerated partial breast irradiation AND breast cancer AND IOERT”, “APBI AND breast cancer AND IOERT”.

2.2. Study selection

Inclusion criteria were 1) full papers in English reporting on 10 or more patients, 2) IOeRT delivery of PBI, 3) original research, 4) clinical studies and 5) outcomes.

Exclusion criteria were IOeRT delivery of a boost and administering PBI with electrons as EBRT.

Titles and abstracts were screened independently by FC and VE to identify which studies delivered IOeRT as PBI by means of a mobile linac. Duplicates were eliminated. FC and VE independently screened the full papers which had been identified from their abstracts, excluding those that did not satisfy inclusion criteria.

When PBI was compared with WBI, only cohorts that underwent PBI as delivered by IOeRT were included. When multiple articles reported on outcomes in the same population at different follow-up times, only the most recent was included. Any dispute was resolved by a third reviewer (IP).

The following data were collected: study type (randomized controlled, prospective, retrospective) and quality; demographics (e.g., number of patients, median age and accrual period); disease status (e.g., tumour stage, median tumour size and nodal stage, histology); dosimetry (prescribed dose and reference isodose, collimator size); rates and probabilities of recurrence and survival; side effects and cosmesis.

2.3. Quality assessment

Each study was assessed by a quality score system using the GRADE approach [15].

Studies were scored as follow: 3 points for randomized controlled trials (RCT); 2 points for prospective studies; 1 point for retrospective studies. Quality scores were allotted as follow: +1: studies with ≥100 patients; +1: fully described cohorts; +1: data extraction ease; -1: studies with <100 patients; −1: poorly described cohorts (e.g., without mean age); −1: data extraction difficulties (as encountered by just one assessor). Scores ranged from a maximum of 6 (e.g., a well described RCT, with >100 patients and no problems in extracting data) to −2 (e.g., a retrospective study, with a small, poorly described cohort, and difficulties in extracting data). A +2 score was the cut-off as > 2 indicated good quality while ≤2 meant poor quality.

2.4. Endpoints

The primary endpoint of this meta-analysis and systematic review was to evaluate the local recurrence, rate, i.e., recurrence within the ipsilateral breast. Secondary endpoints were rates of: nodal recurrence, ipsilateral second breast cancer, any recurrence (i.e., local, nodal, ipsilateral second breast cancer, distant metastases) and relapse-related mastectomy. Also included among the secondary end-points were: 5-year probabilities of local recurrence and overall survival (OS), acute and late side effects, cosmetic results.

2.5. Statistical analysis

A random-effect model [16] incorporating heterogeneity of effects was applied to combine extracted data. The Cochrane Q test and I² statistics, which describe the heterogeneity-linked percentage of total variation across studies, evaluated study heterogeneity [17]. Heterogeneity was considered significant if p < 0.10 and I² was >50 %. Begg's and Egger's tests assessed publication bias [18,19].

Meta-analysis calculations were performed using Stats Direct statistical software v.2.7.2, StatsDirect Ltd Merseyside, UK, 2008.

3. Results

3.1. Identification and description of studies

The database searches yielded 2249 papers. After eliminating duplicates, 1659 articles remained but 1575 did not meet the selection criteria; 66/84 remaining papers were excluded for not meeting clinical and outcome criteria. Consequently, only 18 articles that were published in the period 2010–2023 were suitable for inclusion in this meta-analysis and systematic review [[6], [7], [8], [9], [10], [11], [12], [13], [14],[20], [21], [22], [23], [24], [25], [26], [27], [28]]. Fig. 1 illustrates the PRISMA Flowchart. Table 1 reports the quality of the 18 studies. Three studies [10,23,28] with a GRADE score of 2 or less, indicating poor quality, were excluded from the meta-analysis. They were used for the systematic review.

Fig. 1.

PRISMA criteria flowchart.

Table 1.

Study Quality Assessment as scored by GRADE^a.

Author	Year	Journal	Study type	Population	Data description	Extraction	Study Quality
Veronesi U. et al. [6]	2010	Breast Cancer Research and Treatment	Prospective Observational Trial	1	1	1	5
Sawaki M. et al. [20]	2012	Breast Cancer	Prospective phase I-II trial	−1	1	1	3
Osti M. et al. [21]	2013	Anticancer Research	Prospective observational trial	1	1	1	5
Struikmans H. et al. [22]	2016	Strahlentherapie und Onkologie	Prospective Observational trial	−1	1	1	3
Avci G.G. et al. [23]	2018	Journal of Cancer Research and therapeutics	Retrospective trial	−1	1	1	2
Guenzi M. et al. [7]	2018	Frontiers in Oncology	Retrospective trial	1	1	1	4
Jacobs DHM et al. [24]	2018	Breast Cancer Research and Treatment	Prospective Observational trial	1	1	1	5
Sorrentino L. et al. [25]	2018	Elsevier/The Breast	Prospective Observational trial	1	1	1	5
Sawaki M. et al. [26]	2019	Annals of Surgical Oncology	Prospective Phase II trial	1	1	1	5
Williams V.L. et al. [8]	2019	Brachiterapy	Prospective Phase I-II trial	1	1	1	5
Kawamura M. et al. [27]	2020	Journal of Radiation Research	Prospective Phase I-II trial	−1	1	1	3
Cernusco N.L.V. et al. [9]	2021	Clinical Breast Cancer	Prospective Phase II trial	1	1	1	5
Hashemi S. et al. [10]	2021	International Journal of Radiation Oncology	Prospective Observational trial	1	−1	−1	1
Orecchia R. et al. [11]	2021	Lancet Oncology	Prospective Randomized Controlled trial (Phase III)	1	1	1	6
Jacobs DMH et al. [12]	2022	International Journal of Radiation Oncology	Prospective Observational trial	1	1	−1	3
Philippson C. et al. [13]	2022	Breast Cancer Research	Prospective Observational trial	1	1	−1	3
Zangouri V. et al. [14]	2022	BMC Surgery	Retrospective trial	1	1	1	4
Nafissi N. et al. [28]	2023	BMC Cancer	Retrospective trial	−1	−1	−1	−2

^aGRADE: Guyatt G, Oxman AD, Akl EA, Kunz R, Vist G, Brozek J, Norris S, Falck-Ytter Y, Glasziou P, DeBeer H, Jaeschke R, Rind D, Meerpohl J, Dahm P, Schünemann HJ. GRADE guidelines: 1. Introduction-GRADE evidence profiles and summary of findings tables. J Clin Epidemiol. 2011 Apr; 64(4):383-94. https://doi.org/10.1016/j.jclinepi.2010.04.026. Epub 2010 Dec 31. PMID: 21195583.

Overall, 5474 patients were included. As 9 patients had simultaneous bilateral breast cancer and both were treated with IOeRT, the total number of lesions was 5483. Median age was reported in 8 studies [6,8,13,14,21,23,24,27], ranging from 56 to 68 years (study median 64.5 years). Ten studies [6,8,13,14,20,21,23,24,26,27] reported lowest and highest ages (median lowest age 50 years, range 33–59; median highest age 82.5 years, range 62–91). Five studies [14,20,22,25,26] reported mean ages (65.1 years, range 56.4–69.3).

In the surgery session, patients underwent BCS and IOeRT. Axillary management mainly consisted of sentinel lymph node biopsy (SLNB), generally followed by axillary lymph node dissection (ALND) when lymph node metastases were discovered. Axillary dissection as first approach was less common.

Fourteen papers [6,[8], [9], [10], [11], [12], [13], [14],20,21,23,[25], [26], [27]] included a total of 82 patients with ductal carcinoma in situ and 13 studies [[6], [7], [8], [9], [10], [11],13,14,20,21,23,25,26] enrolled a total of 422 patients with lobular infiltrating carcinoma or mixed histology (ductal and lobular). Pathological tumour and nodal stages, such as tumour biopathological characteristics are reported in Table 2.

Table 2.

Tumour and Nodal Pathological Stage, Tumour Biopathological Characteristics and IOeRT Parameters (Patients = 5474; lesions = 5483).

	N° studies [REF]	N° lesions	%
T stage **
pT0	7 [6,7,9,11,20,21,23]	3	0.054
pT1	14 [6–9,11-14,20,21,23,25–27]	4514	82.32
pT2 - pT3	14 [6–9,11-14,20,21,23,25–27]	494	9.00
pT3	10 [6–9,13,20,23,25–27]	3	0.054
Not Reported		390	7.11
N stage**
pN0	14 [6–9,11-14,20,21,23,25–27]	4309	78.58
pN1 – pN2	14 [6–9,11-14,20,21,23,25–27]	790	14.40
pN2	11 [6,8,9,11,13,14,20,23,25–27]	175	3.19
Not Reported		384	7.00
Tumour Biopathological Characteristics
ER +	12 [6,7,8,9,10,11,12,13,14,22,23,25]	4512	–
ER	9 [6,7,9–11,13,14,22,25]	423	–
PgR +	9 [6–9,11,13,14,22,25]	3363	–
PgR	9 [6,7,9,11,13,14,22,23,25]	970	–
ER and/or PgR +	3 [20,21,27]	164	–
ER and/or PgR	3 [20,21,27]	11	–
Not Described	2 [26,28]	–	–
Ki67 < 14 %	2 [6,25]	758	–
Ki67 ≥ 14 %	2 [6,25]	1198	–
Ki67 </≤ 20 %	4 [7,11,13,23]	1259	–
Ki67 >/≥ 20 %	4 [7,11,13,23]	606	–
Ki67 ≤ 30 %	1 [10]	176	–
Ki67 > 30 %	1 [10]	101	–
Not Described	10 [8,9,12,14,16,20,12,22,26,27]	–	–
HER2 +	12 [6–11,13,14,20,22,24,25,27]	426	–
HER2	14 [6–14,20,22,23a,25,27]	4516	–
Not Described	3 [21,26,28]	–	–
IOeRT parameters	N° studies	Median	Range
RT dose (Gy)b	18 [6–14,20-28]	21 Gy	16–23.3 Gy
Isodose (%)	14 [6–9,12-14,20–23,25,27,28]	90 %	80–100 %
Collimator size min. (cm)	7 [7–9,12,14,21,23]	4 cm	2 – 4 cm
Collimator size max. (cm)	7 [7–9,12,14,21,23]	7 cm	6 – 8 cm

^aIn this study in 3 patients HER2 score was reported as 2+. No further evaluation was made.

^bIn 3 studies [6,20,27] a dose-escalation approach was followed (min 16, max 21 Gy).

All 18 studies reported the delivered dose (median 21 Gy, range 16–23.3). In 3 studies [6,20,27] a dose escalation strategy was adopted. One study [6] started from 16 Gy and reached 19 Gy; another 2 [20,27] started from 19 Gy and reached 21 Gy. Fourteen studies [[6], [7], [8], [9],[12], [13], [14],[20], [21], [22], [23],25,27,28] reported the reference isodose. In 7 [6,7,12,13,20,25,27], the dose was prescribed at 90 % isodose, in 2 [21,23] at 100 % isodose, in 1 [9] at 80 % isodose and in 3 [8,14,28] at varying isodoses, ranging from 90 % to 100 %. One study [22] prescribed 23.3 Gy at the 100 % isodose. Seven studies reported collimator size [[7], [8], [9],12,14,21,23]. It ranged from a median minimum of 4 cm (range 2–4) to a median maximum of 7 cm (range 6–8) Table 4 (Table 2).

Table 4.

Late side effects of IOeRT.

Side Effects	Study		Events	% [Events/No patients (1101)]
	5		87	7.90
Breast pain	[23,28]		10	0.90
Dimpling	[13]		2	0.18
Fat necrosis	[23]		3	0.27
Fibrosis	[23,27]		5	0.45
Fibrosis/Fat necrosis	[28]		4	0.36
Hypertrophic scarring	[13,27]		43	3.90
Oedema	[13]		3	0.27
Subcutaneous tissue toxicity	[20a]		17	1.54
Side Effect Grade			Events/Total late events (87)	%
G1 - G2	4 [13,20,23,27]		70	80.46
G3 - G4	3 [20,23,27]		0	0
Not Reported Grade			17	19.54

^aSome patients developed more than 1 side effect.

Administration of adjuvant systemic therapy was not always reported in all studies but, when it was, it was prescribed according to the rules, protocols, national or international guidelines adopted in the various centres. Endocrine therapy was given to most patients, with only a minority receiving chemotherapy, associated or not with endocrine therapy.

Fourteen studies [[6], [7], [8], [9],[11], [12], [13], [14],[20], [21], [22], [23],26,27] reported median follow-up time, ranging between 26 and 149 months (study median 47.8 months). Nine studies [[6], [7], [8],14,20,21,23,26,27] reported only the range of follow-up times, ranging from a median minimum 23.1 months (range 1–65) to a median maximum 78 months (range 40–126). Two papers [10,25] reported mean follow-up time (26.1 and 34.5 months).

3.2. Outcomes

When results were presented as intention to treat or per-protocol, data from the intention to treat analysis was used.

Figs. 2–8 illustrate the meta-analysis plot for each outcome parameter, together with the results of the publication bias analysis.

Fig. 2.

Local Recurrence: Panel A) Meta-analysis plot; Panel B) Publication bias analysis.

Fig. 8.

5-year probability of overall survival: Panel A)Meta-analysis plot; Panel B) Publication bias analysis.

3.3. Oncological

Thirteen studies [[6], [7], [8], [9],[11], [12], [13], [14],20,21,[25], [26], [27]] reported 203 local recurrences. The pooled proportion was 3.8 % (95 % CI: 2.4–5.6 %) (Fig. 2). The I² was 85.3 % (p < 0.0001).

Nine studies [6,[11], [12], [13], [14],20,[25], [26], [27]] reported 47 nodal recurrences. The pooled proportion was 1.1 % (95 % CI: 0.5–1.8 %) (Fig. 3). The I² was 66.0 % (p < 0.0027).

Fig. 3.

Nodal Recurrence: Panel A) Meta-analysis plot; Panel B) Publication bias analysis.

Four papers [6,12,21,27] reported 35 ipsilateral second breast cancers. The pooled proportion was 1.9 % (95 % CI: 0.8–3.4 %) (Fig. 4). The I² was 54.5 % (p < 0.086).

Fig. 4.

Ipsilateral second breast cancer: Panel A) Meta-analysis plot; Panel B) Publication bias analysis.

Thirteen papers [[6], [7], [8], [9],[11], [12], [13], [14],20,21,[25], [26], [27]] described 436 recurrences of any type. The pooled proportion was 7.2 % (95 % CI: 4.4–10.5 %) (Fig. 5). The I² was 93.2 % (p < 0.0001).

Fig. 5.

Recurrences of any type: Panel A) Meta-analysis plot; Panel B)Publication bias analysis.

Six papers [6,8,9,11,21,27] observed 70 relapse-related mastectomies. The pooled proportion was 2.6 % (95 % CI: 1.7–3.7 %) (Fig. 6). The I² was 49.9 % (p = 0.076).

Fig. 6.

Relapse-related mastectomies: Panel A) Meta-analysis plot; Panel B) Publication bias analysis.

The pooled proportion of 5-year probability of local recurrence, as reported in 4 papers [9,[11], [12], [13]], was 6 % (95 % CI 3–9%) (Fig. 7). The I² was 90.1 % (p < 0.0001). Two studies reported longer follow-up data. In 1 [27] the 10-year local recurrence rate was 8 %. In the other [11] ipsilateral breast tumour recurrence rates at 10 and 15 years were, respectively, 8.6 % and 13.7 %.

Fig. 7.

5-year probability of local recurrence: Panel A) Meta-analysis plot; Panel B) Publication bias analysis.

The pooled proportion of 5-year OS, as reported in 4 studies [9,[11], [12], [13]], was 96.0 % (95 % CI 94.1–97.1 %) (Fig. 8). The I² was 62.6 % (p = 0.046). In 3 studies this outcome was reported at 6 years (90.2 %) [7], at 10 years (89.7 %) [6] and at 15 years (83.3 %) [11].

3.4. Side effects

A meta-analysis could not be conducted, as more than 1 side effect was reported in some patients. Therefore a descriptive analysis was performed. It consisted of the maximum grade of any side effect, independently of alleviation or resolution. Overall, 981 acute events such as hematoma, liponecrosis, breast pain, oedema, wound infection were reported in 10 studies [6,9,13,20,21,[23], [24], [25], [26], [27]]. A total of 87 late side effects such as fluid collection, atrophy, fat necrosis, oversensitivity were described in 5 studies [13,20,23,27,28]. Table 3, Table 4 list type and grade of acute and late side effects, respectively.

Table 3.

Acute side effects of IOeRT.

Side Effects	Study	Events	% [Events/No patients (3844)]
	10	981	25.52
Abscess	[23]	2	0.05
Delay of scarring	[13,21°]	18	0.47
Fatigue	[24a]	65	1.70
Hematoma	[6a,9,13,20a]	134	3.48
Hemorrhage	[26]	6	0.15
Lyponecrosis	[6a,20a,21°,26]	103	2.68
Seroma	[6a,23]	239	6.21
Seroma + Lyponecrosis	[25]	16	0.41
Mastitis	[23]	2	0.05
Oedema	[6a,9,21°]	29	0.75
Pain	[6a,20a,21°,24a,26]	67	1.74
Pain + Oedema	[25]	4	0.10
Skin retraction	[6a,21°]	20	0.52
Fibrosis	[6a,20a,21°,26,27]	190	4.94
Skin retraction + Fibrosis	[25]	3	0.08
Soft tissue infection	[20a]	6	0.15
Wound infection	[6a,13,23,24a,26]	59	1.53
Wound dehiscence	[13,26]	11	0.28
Wound dehiscence + infection	[13]	6	0.15
Wound necrosis	[13]	1	0.02
Side Effect Grade		Events	% [Events/Total events (981)]
G1 - G2	7 [6,13,20,23,24,26,27]	317	32.32
G3 - G4	6 [6,9,20,23,24,26]	14	1.42
Not Reported Grade		650	66.25

° The maximum grade of any side effect was reported, independently of alleviation or resolution.

^aSome patients developed more than 1 side effect.

The CTCAE scale v3.0 or v4.0 was used to evaluate acute and/or late side effects in 6 papers [13,20,23,24,26,27]. The RTOG scale assessed acute and late side effects in 2 papers [6,21] and late side effects in 1 [20]. The LENT-SOMA scale was used to assess late side effects in 1 paper [13]. Three studies [9,25,28] did not report a scale.

3.5. Cosmetic results

Cosmetic results were evaluated independently by patients in 3 studies [[21], [22], [23]] and physicians in 5 [13,[21], [22], [23],26]. The Harvard Scale [29] was used in all studies [13,[21], [22], [23],26]. Results are reported in Table 5.

Table 5.

Cosmetic evaluation and patient satisfaction.

N° studies	Evaluated by	TOT	EXCELLENT - GOOD N° (%)	FAIR – POOR N° (%)
5 [13, 21, 22, 23, 26]	Physician	1115	981 (87.98)	134 (12.01)
3 [21, 22, 23]	Patient	154	142 (92.20)	12 (7.79)

4. Discussion

The present meta-analysis and systematic review provided evidence suggesting that IOeRT, as delivered by mobile linacs, is a feasible option for administering PBI as part of BCT in selected, low-risk early breast cancer patients. Overall, local recurrences ranged from 0 % to 11.3 % with a pooled proportion of 3.8 % and a 5-year probability of 6 %. These results appear higher than phase III randomized controlled trials (RCTs) comparing WBI with PBI, as delivered by brachytherapy or EBRT [[30], [31], [32], [33], [34], [35], [36], [37]]. Differences in techniques as far as regards planning, dosimetry and target coverage need to be taken in account when evaluating results from several reports [38]. Patient selection criteria may also account for the discrepancy in local relapse rates. Furthermore, IOeRT can only be performed on pre-pathology, the results of which might have excluded some patients from this treatment modality.

In 4 phase III RCTs which were designed as non-inferiority studies, PBI emerged as a valid option in well-selected patients. As in the present meta-analysis, follow-up length impacted on the relapse rate as the longer the follow-up, the higher the relapse rate. In the GEC-ESTRO trial (accrual 2004–2009), patients were treated with interstitial brachytherapy. At a medium follow-up of 6.6 years, the 5-year cumulative incidence of local recurrence was 1.4 % [30]. Recently updated results showed a 10-year local recurrence rate of 3.5 % [31]. Other studies used EBRT to administer PBI in moderate or ultra-hypo-fractionated schedules [[32], [33], [34], [35]]. With a schedule of 40 Gy in 15 fractions and a median follow-up of 72 months, the UK IMPORT LOW trial (accrual 2007–2010) reported a local relapse rate of 1 %, with an estimated 5-year cumulative incidence of 0.5 % [32]. Using the same fractionation, the Danish PBI trial (accrual 2009–2016) reported a 2.3 % recurrence rate at a median follow-up of 7.6 years. The risk of local recurrence was 1.2 % and 3.1 % at, respectively, 5- and 9-years [33]. The RAPID trial (accrual 2006–2011) used EBRT to administer 38.5 Gy in 10 fractions delivered twice a day over 5–8 days. At a median follow-up of 8.6 years the local recurrence rate was 2.6 %. The 8-year cumulative rate of local recurrence was 3.0 % [34]. In the equivalence NSABP B-39/RTOG 0413 trial (accrual 2005–2013), PBI was delivered with 3D-EBRT or by interstitial/endocavitary brachytherapy. A dose of 38.5 Gy was administered with EBRT and 34 Gy with brachytherapy in 10 fractions, given twice daily, over 5 treatment days within an 8-day period. At a median follow-up of 10.2 years, the local recurrence rate was 4 %; the 10-year cumulative incidence was 4.6 %. Even though PBI did not meet the equivalence criteria, the 1 % difference between PBI and WBI relapse rates suggested that PBI was an acceptable alternative for some women [35]. Finally, the Florence trial (accrual 2005–2013) used an intensity modulated technique to deliver PBI (30 Gy in 5 non-consecutive once-daily fractions). At a median follow-up of 5 years, the local recurrence rate was 1.5 % [36]; the 10-year cumulative incidence rose to 3.7 % [37]. On the basis of the results of these studies, both moderate- and ultra-ultra-hypofarctionated schedules can be used to deliver PBI with EBRT [39].

Unlike the present meta-analysis investigating IOeRT, meta-analyses of PBI trials that used different techniques suggested IORT was the least appropriate technique, as it was associated with inferior local control [[40], [41], [42], [43], [44]]. Poor local control rates may have been due to the inclusion of only the TARGIT trial [5,45,46] and/or the IEO trial [6,11] as those are the only prospective randomized phase III trials. Even though including the IEO trial in the present meta-analysis impacted negatively on the results, other factors may also have come into play such as length of follow-up, sample size, and tumour-, patient- and treatment-related features.

Patient selection needs to be taken into consideration when evaluating relapse rates as patient- and tumour-related features impact strongly upon local recurrence rates. In fact, not all patients in the present meta-analysis were carefully selected. Some had tumours over 2.5–3 cm in size (cut-off suggested by the international guidelines), involved lymph nodes (see Table 2) and other adverse risk factors for local relapse. In the IEO trial a post-hoc univariate analysis of risk factors for local relapse demonstrated that patients who had at least one of the following factors had significantly increased risk of local recurrences: grade 3 tumours; size over 2 cm; high proliferative (Ki-67) index >20 %; luminal B or triple-negative subtype, four or more positive axillary nodes. An unplanned analysis in a small subgroup of patients with G1, luminal A tumours, under 1 cm in size, with a Ki-67 < 14 % showed a 1.3 % local recurrence rate at 10-years [11]. Guenzi et al. [7] observed 3.4 % (8/235) local recurrences at a median follow-up of 6 years in patients with adverse biological risk factors such as high histological grade, ki67 > 20, invasive lobular carcinoma. Relapses did not occur in any patient with a luminal A tumour. Other risk factors negatively impacting on the risk of relapse were e.g., tumour multifocality [8,13], positive margins, age under 50 years associated with extensive intraductal component, ER negative tumours [8].

One might argue that patient selection for IOeRT, like other PBI techniques, and RT omission are quite similar. However, different age ranges are recommended for omitting WBI or receiving PBI (≥65–70 years vs ≥ 50 years). In trials comparing WBI with no WBI, RT to the breast significantly lowered the risk of local relapse, even though it did not impact on survival outcomes [47]. In the PRIME II trial [48], which enrolled well-selected luminal A breast cancer patients, the 10-year cumulative incidence of local relapse was 9.5 % after no RT vs 0.9 % after WBI, with an HR of 10.4 %. The incidence was significantly higher in patients with ER-low tumours and in patients no longer taking endocrine therapy. Consequently, the recent ESMO guidelines indicated the RT omission after BCS as an investigational approach [3].

Doubts arose as to whether relapses in the present analysis were real ipsilateral relapse or second primary tumours. Local relapse features (i.e., time lapse to recurrence, histology, relapse in the same or different quadrant as the primary tumour) were not always reported in the studies under analysis, with only 4 papers [6,12,21,27] reporting a 1.9 % pooled proportion of ipsilateral second breast cancers. On the other hand, given the well-known difficulty in distinguishing between local relapse and a new primary tumour, one can hypothesize that many papers included both as ipsilateral breast recurrence.

The pooled proportion of relapse-related mastectomies was 2.6 %, as reported in 6 papers [6,8,9,11,21,27]. One might speculate that some patients with local relapse or a new primary tumour underwent a second BCS. This may be an option, particularly when PBI was part of first-line treatment which is, theoretically, one of the advantages of using PBI in suitable candidates.

In the present study, the nodal recurrence pooled proportion was 1.1 %, which was higher than Haussmann's meta-analysis of randomized clinical trials [41]. As the latter included patients with good prognostic features at low risk of local and nodal relapse, differences in patient selection likely account for the (minor) discrepancy. In the present cohort, T2 or T3 tumours were found in 10 % of patients and pN1-N2 cases in 15 %. Consequently, one may assume that most would have needed regional node irradiation. Furthermore, not all patients with 1–2 positive SLN underwent ALND. As they received PBI, they may not have benefited from the WBI-associated incidental irradiation of the axillary lymph node area [49,50]. Regional node irradiation should be considered in this subset of patients, especially if they harbour unfavourable features such as microscopic extracapsular extension or lympho-vascular invasion [[51], [52], [53]].

Like the Oxford meta-analysis, the present study analysed “recurrences of any type” which were, however, particularly impacted by local relapse rates. The pooled proportion of 7.2 % was extrapolated from 13 papers [[6], [7], [8], [9],[11], [12], [13], [14],20,21,[25], [26], [27]], which is not dissimilar to other studies in low-risk patients.

Finally, only 4 studies [9,[11], [12], [13]] reported 5-year OS probabilities. The pooled proportion of 96 % was expected because, even though the relapse rates were high in some studies, they did not impact upon OS.

One of the main limitations of this meta-analysis, as with many others, is that it was based on published data and did not regard individual patients. Another was the high heterogeneity index across papers for several outcome parameters. Indeed the I² was >50 % for all but one. Furthermore, almost all the studies were not randomized so data were derived from heterogeneous groups, some of whom may not have been candidates for IOeRT according to current guidelines. Finally, in some papers sample sizes were small, follow-up was short, part of the data was not clearly reported and not all studies reported all outcome parameters in a standardized way. Technical data, particularly information on collimator size as a function of tumour size, were often lacking, leading to speculations that IOeRT poor target volume coverage might have been due to use of applicators that were too small. Moreover, study biases, missing data, and poorly reported data sets account for the difficulties in analysing side effects, which were not always graded on an internationally accepted scale and mostly described as minimal, mild, moderate or severe. Some patients were affected by more than 1 side effect and incidences were at times reported as percentages or on graphs. The timing of side effects was not always clearly indicated as acute or late, particularly when referring to fibrosis, and evolution over time was not always reported.

The present paper provides a piece in the jigsaw puzzle of evidence for more widespread acceptance of IOeRT as part of PBI delivery in breast cancer management, as long as patients are carefully selected. Furthermore, well-informed patients may prefer IOeRT because of its convenience.

As stated in the recent ESMO guidelines [3], any technique is suitable as long as it provides full coverage of the entire target volume [1,54]. On the other hand, according to the ASTRO guidelines [4], since more robust IOeRT data are needed, patients should be enrolled in prospective clinical trials or a multi-institutional registry.

In the future, improved target volume coverage by using real-time imaging and treatment planning during the surgical and IOeRT procedure will facilitate optimal selection and positioning of the linac's applicator and electron energy [55]. FLASH-irradiation is set to be explored in this field as well, as a possible solution for increasing irradiated volumes without increasing side effects [56].

5. Conclusions

Despite the limitations of the present meta-analysis and systematic review IOeRT, as delivered by mobile linacs, appeared to be a feasible PBI option for selected, low-risk early breast cancer patients. As only one phase III randomized trial [11] has been conducted to date one might argue that IOeRT still requires investigation in well-designed trials. On the other hand, the phase I/II or retrospective/prospective studies that were analysed in the present paper reported lower LR rates than the IEO trial [11]. Since inappropriate selection criteria may have been one of the factors underlying the local reurrence rates, careful patient recruitment seems essential for good outcomes with IOeRT.

CRediT authorship contribution statement

Cynthia Aristei: Writing – review & editing, Writing – original draft, Validation, Project administration, Methodology, Conceptualization. Federico Camilli: Validation, Investigation, Data curation. Valeria Epifani: Validation, Investigation, Data curation. Simona Borghesi: Validation, Methodology, Investigation. Isabella Palumbo: Validation, Investigation. Vittorio Bini: Validation, Formal analysis. Philip Poortmans: Writing – review & editing, Validation, Supervision, Methodology, Conceptualization.