Local relapse rates are falling after breast conserving surgery and systemic therapy for early breast cancer: Can radiotherapy ever be safely withheld?

Surgical management of the primary tumour is more standardised 

The emergence of surgical oncology, recognition of breast surgery as a sub-speciality in many countries, multidisciplinary working and written guidelines have done much to improve the management of women with early breast cancer over the last 30 years. The American College of Surgeons, American College of Radiology, College of American Pathologists and Society of Surgical Oncology published standards in 1992 [14]. This document addressed technical aspects of surgery, the importance of specimen orientation using sutures and clips, frozen section evaluation of margins with immediate re-excision if positive, radiographic evaluation of removal of non-palpable lesions and pathologic reporting. Following publication of national guidelines in several European countries, the European Society of Surgical Oncology issued a document in 1997 [15]. The recommendations, based mainly on British Association of Surgical Oncology and Danish Breast Cancer Co-Operative Group guidelines, focused on the importance of multidisciplinary teams with specialist expertise in breast surgery, imaging and pathology.

A United States (US) Pattern of Care Study survey in 1998–1999 confirmed better compliance with breast conservation standards, compared to previous reports [16], but although guidelines have had a major influence on practice, there is no direct way of evaluating impact on clinical outcome. A retrospective study conducted between 1986 and 1991 in Scotland investigated the impact of treatment by specialist breast surgeons on disease outcome in 2776 patients [17]. The actuarial LR rates at 8 years were 7% and 14% in women treated by specialists and non-specialists, respectively, with a hazard rate of 1.54 (95% CI 1.1–2.2) for local relapse after adjustment for nodal status, tumour size and histological prognostic grade. The risk of receiving inadequate treatment was 24% and 47% (p<0.001) in patients referred to specialists and non-specialists, respectively. Adequacy of treatment was defined according to the United Kingdom (UK) King’s Fund Consensus Statement (1986) [18]. Treatment was judged inadequate according to three parameters: breast conservation surgery performed on tumours >30mm, positive resection margins and omission of radiotherapy. If these findings are representative, one of the key explanatory variables for treatment by a specialist surgeon is likely to be microscopic resection margins.

Relating surgical margin status to local tumour relapse risk is not straightforward 

Many retrospective studies have analysed the correlation of margin status with local control after breast conservation surgery and radiotherapy [4], [19], [20], [21], [22], [23], [24], [25], [26]. There is general agreement that local relapse risk is higher in the presence of positive margins, although the relevance of focally positive margins is disputed. According to a recent review [24], interpretation of the literature is hampered by the lack of consensus on the definition of negative and close margins, with >1, 2, 3 or 5mm being used to define negative margins in different case series. The Milan II trial reported a higher LR rate in patients treated with lumpectomy and radiotherapy (TART) compared to those treated with quadrantectomy and radiotherapy (QUART) [27]. However, the interpretation is limited by differences in RT volume and dose between randomised arms.

In the 1980s, an association was reported between an extensive intraductal component (EIC) in the tumours of young women and an increased risk of LR and further, often bulky, residual tumour at re-excision [28]. Although subsequent series were initially confirmatory, recent studies have failed to confirm the association in the presence of negative microscopic margins [29], [30], [31]. It is likely that better tumour localisation, more systematic margin assessment, readiness to re-excise in the presence of positive or narrow margins and more effective non-surgical therapies (see below) now control for the excess risk of local relapse associated with EIC. Perhaps for the same reasons, it is difficult to identify a difference in LR between narrow margins and wide margins in the current literature. It is particularly significant that large randomised clinical trials have failed to confirm the importance of margin width. This has been investigated in a combined analysis of the EORTC 10801 and DBCG-82TM trials, which tested breast conservation surgery plus whole breast radiotherapy versus radical mastectomy in 1772 patients [32]. It has also been explored in the EORTC 22881-10882 boost trial testing a tumour bed boost dose in 5318 women following whole breast radiotherapy [33]. In univariate and multivariate analyses, microscopic involvement of the excision margins (EORTC 10801 and DBCG-82TM) and width of microscopic margin around invasive carcinoma (central pathology review of the EORTC 22881-10882) were not associated with increased LR after breast conservation surgery and breast radiotherapy. All trials specified a tumour bed boost dose of at least 15Gy in 2.0Gy fractions following 50Gy in 25 fractions to the whole breast, and it is likely that this higher dose compensates, at least partially, for the presence of narrowly complete or focally incomplete margins.

If higher doses and more accurately directed radiotherapy are important factors in reducing LR rates (see below), narrow or incomplete resection margins might be highly significant risk factors in patients in whom radiotherapy is withheld. The prognostic value of margin status after local excision without radiotherapy was studied in only 2 of the 10 randomised trials in the Early Breast Cancer Trialists’ Collaborative Group (EBCTCG) overview [34], [35]. The Uppsala-Őrebro trial reported a HR of 1.84 (95% CI 0.89–3.77) for a resection margin <5mm versus ⩾5mm in univariate analysis of 381 women entered in both treatment arms [34]. The St. George’s trial (UK) reported on margins in 210 women randomised to receive no radiotherapy, in whom retrospective review of excision margins in 191 specimens was performed [35]. If the surgical excision biopsy at frozen section was judged clinically clear (as sole criterion), the LR rate was 32/73 (44%), compared to 24/69 (35%) if microscopic margins were <5mm and 10/49 (22%) if margins were ⩾5mm. On the basis of this limited evidence, a minimum requirement would be to implement prospective margin width analysis in protocols testing the selective avoidance of radiotherapy.

Adjuvant systemic therapies are recommended to a higher proportion of patients than ever before 

Systematic overviews of adjuvant systemic therapy by the EBCTCG and European St. Galen conferences in recent decades have been highly influential in defining treatment recommendations [36], [37]. Following publication of the 1992 EBCTCG overview, the Fourth International Conference on Adjuvant Therapy of Primary Breast Cancer held at St. Galen recommended that “almost all patients with clinical and pathologic stage I and II breast cancer, regardless of age, menopausal, nodal, or receptor status, will benefit from some form of adjuvant chemotherapy and/or hormonal therapy both in terms of improved disease-free survival, as well as overall survival” [37]. The 1998–1999 Patterns of Care Study reported an increased use of systemic therapy compared to previous surveys: tamoxifen and chemotherapy were prescribed in 55.8% and 36% of patients, respectively, compared to 43% and 25.5% in the 1989 study [16]. The following paragraphs address the effects of systemic agents on LR risk.

Endocrine therapy reduces the risk of local relapse by around 50% 

The EBCTCG overview provides high level evidence that 20mg tamoxifen daily for 5 years reduces LR risk by about 50% in patients with early stage ER-positive disease (relative risk (RR)=0.47, SE 0.08) [38]. In recent years, the benefits of adjuvant endocrine therapy in post-menopausal women have increased with the introduction of aromatase inhibitors (Table 3). This additional relative benefit ranges from around 50% to 20–30% [39], [40], [41], [42], [43], [44], [45]. Results of longer follow-up are awaited to evaluate the long-term effectiveness of different strategies adopted with aromatase inhibitors (initial adjuvant treatment, switch after 2–3 years of tamoxifen or extended therapy after 5 years of tamoxifen).

Table 3. Relative risk (RR) for ipsilateral local relapse (LR) reported in randomised trials testing systemic therapies in women with early breast cancer

	Systemic therapy	RR for local relapse
	Tamoxifen for 5 years vs no tamoxifen (EBCTCG overview) [38]	0.47
	Anastrozole switch after 2 years of tamoxifen vs 5 years of tamoxifen (ABCSG 8, ARNO 95, ITA) [39]	0.50a
	Exemestane switch after 2–3 years of tamoxifen vs 5 years of tamoxifen (IES) [40]	0.72b
	Anastrozole vs tamoxifen (ATAC trial) [41]	0.83c
	Letrozole vs Tamoxifen (BIG I-98) [41]	0.70d
	Extended HT after 5 years of tamoxifen with
	Anastrozole for 3 years (ABCSG 6a) [44]	0.81e
	Letrozole (5 years) vs placebo (MA-17) [45]	0.54f
	CMF-based or anthracycline-based polychemotherapy vs no chemotherapy (EBCTCG overview) [38]	0.63g
		0.70h
	Docetaxel-containing chemotherapy [46]
	DAC 6 vs FAC 6 (BCIRG 001)	0.75i
	FEC 3-D 3 vs FEC 6 (PACS 01)	0.66j
	Paclitaxel-containing chemotherapy [46]
	AC 4-P4 vs AC 4(NSABP B28)	0.82k
	Trastuzumab vs no trastuzumab
	NASBP B-31+NCCTG N9831 trials [47]	0.47
	HERA trial [48]	0.75

aMedian follow-up of 30 months; bMedian follow-up of 55.7 months; cMedian follow-up of 47 months; dMedian follow-up of 26 months; eMedian follow-up of 60 months; fMedian follow-up of 30 months; gWomen aged <50; hWomen aged 50–69; iMedian follow-up of 55 months; jMedian follow-up of 59.7 months; kMedian follow-up of 64.4 months.

Cytotoxic chemotherapy reduces the risk of local relapse independent of endocrine therapy and in an age-dependent manner 

Polychemotherapy reduces the risk of local relapse by ⩾30% (Table 3) [38]. The benefit is age-dependent, being greater in women under 50 years of age (RR=0.63 compared to 0.70 for patients aged 50–69 years). A direct comparison between anthracycline chemotherapy and CMF (cyclophosphamide, methotrexate and fluorouracil) reported a moderate but highly statistically significant advantage for the former regimen in the recurrence rate ratio (0.89, SE 0.03). The relative reductions in local and distant relapses were concordant when comparing polychemotherapy versus no chemotherapy, or tamoxifen versus no tamoxifen, so it is assumed that a recurrence rate ratio of 0.89 is indicative of a benefit in local control when anthracyclines are used. Taxanes in the adjuvant setting appear to be beneficial compared to anthracycline-based therapies [46]. Of the randomised trials testing taxane-containing regimens against other chemotherapy regimens, the few that reported data on the type of first event suggest a positive effect on local control (Table 3).

Trastuzumab confers independent benefits in terms of local relapse in the HER2+ subgroup 

Interim reports of three randomised trials have shown that the addition of trastuzumab reduces LR rates in women with HER2+ tumours. The combined analysis of the National Surgical Adjuvant Breast and Bowel Project (NSABP) trial B-31 and of the North Central Cancer Treatment Group trial N9831 reported a RR of 0.47 for LR in the trastuzumab group [47]. A positive impact on LR was also found in the Herceptin Adjuvant (HERA) study, although its magnitude was reduced after a median follow-up of 23.5 months (RR=0.75, compared to 0.45 after 1 year of follow-up) [48].

Significant technical advances in radiotherapy have been achieved, but it is difficult to judge their impact on local control 

The EBCTCG overview of radiotherapy effects confirms that the benefits are independent and additive with those of adjuvant tamoxifen and cytotoxic chemotherapy [38]. The overview of the breast conservation trials conducted between 1976 and 1998 reported a RR of 0.3 (SE 0.03) for LR by radiotherapy. In the last 20 years, technical advances, including immobilization devices (breast board, vacuum bag), CT-based treatment planning and electronic portal imaging devices, have improved accuracy and reproducibility of patient set-up, definition and localisation of target volume, homogeneity of dose distribution and precision of set-up verification [49], [50], [51], [52], [53]. Due to the lack of data from prospective trials or population studies, it is not possible to quantify the impact of these changes on local control. A change in practice, particularly relevant to localising the tumour bed boost, is more widespread use of surgical clips to identify the excision cavity. Boost planning based on clinical evaluation (physical examination, location of scar, clinical and surgical notes) has been shown to be inadequate in a high percentage of cases: studies testing accuracy of clinical boost fields in relation to the position of surgical clips reported inaccuracies in 42% [54], 54% [55] and 68% [56] of plans. The percentages were even higher when clinically derived plans were compared to CT-based plans [57], [58]. However, a retrospective analysis of LR rates in 1364 women (556 with clips, 808 without) failed to show any effect of this procedure on local control [59], whereas a smaller series of 188 patients (155 with clips, 26 without) reported lower LR rates when the boost was planned using surgical clips identified with a CT-simulator [60].

There is a radiation dose–response for local control 

The use of high tumour bed boost dose levels has already been referred to as a possible confounding factor in studying the importance of surgical excision margins. Reliable evidence of a dose–response has been generated by the EORTC boost trial, which randomised 5318 patients after complete tumour resection and 50Gy in 25 fractions of whole breast radiotherapy to a fractionated tumour bed boost of 15 or 16Gy versus no boost. Local relapse risk was approximately halved in the arm allocated boost (HR 0.55 95% CI 0.42–0.73) [33]. It is possible that more systematic administration of a boost dose and/or dose escalation from 10Gy to 16Gy in 2.0Gy fractions may be currently contributing to lower rates of local relapse, particularly if the accuracy of boost localisation has improved as well.

The breast cancer population is aging, and age is a powerful independent prognostic factor for LR with or without radiotherapy 

The US Surveillance Epidemiology and End Results (SEER) population based data confirm an increase in the proportion of elderly breast cancer patients in recent decades [61]. Women 65 years or older represented 37% of breast cancer patients diagnosed in 1973 compared to 46.7% in 1995. This trend will continue: in the US, 35 million people were aged ⩾65 years and 4.2 million ⩾85 years in 2000; in 2020, the projected numbers are 54.6 million and 7.3 million; in 2050, they reach 86.7 million and 20.9 million, respectively [62]. These trends are important, since age is associated with LR risk for several reasons. Before analysing the reasons, evidence for an age effect will be briefly reviewed.

In univariate analysis of the EBCTCG systematic overview of radiotherapy effects, the 5-year LR risks in women treated with breast conservation surgery were inversely related to age, regardless of whether radiotherapy was given (Table 5). In women randomised to no radiotherapy after tumour excision, LR rates were 33%, 23%, 16% and 13% in node-negative women aged <50, 50–59, 60–69, and ⩾70 years, respectively [38]. The contributing trial from Sweden noted a 3% risk reduction per year of increasing age when analysing the whole population of the study [34]. In the EORTC boost trial, a strong correlation between LR rates and age was confirmed in both the univariate and multivariate analyses (Table 5, Table 6) [33]. In the EORTC 10801 and DBCG-82TM trials, patients aged >60 years reported a 9-fold lower risk of LR compared to those aged ⩽35 years [32].

The CALGB C9343 trial recently tested the effects of radiotherapy in women aged ⩾70 years treated by lumpectomy and 5 years of tamoxifen for pT1 pN0 ER-positive invasive carcinomas [8]. The 5-year LR rates were 1% (95% CI 0–2) and 4% (95% CI 2–7) after radiotherapy and no radiotherapy, respectively. Comparable outcomes have been recorded in the US community setting by an analysis of the SEER-Medicare database [63]. Rates of ‘second breast cancer events’ (derived from the second ipsilateral breast cancer rates reported by SEER and/or the subsequent mastectomy rates reported by Medicare) in a cohort of 8724 patients selected with the same characteristics as those eligible for the CALGB trial, and treated between 1992 and 1999 were 1.1% (95% CI 0.79–1.4) and 5.1% (95% CI 4.1–6.2) following radiotherapy and no radiotherapy, respectively. In conclusion, the association between age and LR risk is real, and explanations are now explored.

Factors explaining the age effect are not well defined 

The favourable local outcome observed in the elderly, even when no radiotherapy is given, is not fully understood. A lower probability of residual disease is a strong possibility, although the EORTC 22881-10882 boost trial failed to find a correlation between age and margin involvement [33]. An association between increasing age and the presence of more favourable biologic features, including higher content of ER and PR, lower proliferative rates, lower expression of c-erbB2 and higher frequency of diploidy, has been reported in an analysis of pathological data of women aged ⩾55 years derived from the San Antonio database and SEER registry [64]. Similar associations were found by two Italian studies [65], [66], one of which also reported a higher proportion of well and moderately differentiated tumours with increasing age (Table 4). Although the available data do not allow a definitive explanation for lower LR rates observed in the elderly, it seems likely that they are explained by a lower probability of residual disease after surgery and a more indolent natural history of residual disease, each contributing to lower annual hazards of local relapse.

Table 4. Frequencies (percentages in table) of tumour characteristics according to patient age group derived from the San Antonio database [64], the Istituto Nazionale Tumori of Milan patient series [66] and the Breast Cancer Registry in Verona [65]

		Tumour characteristics (%)	Age groups (years)
			<35	35–44	45–54	55–64	65–74	75–84	⩾85
		ER status:
	San Antonio	Positive				83	87	90	91
	INT Milan	Negative	42	31	27	24	19	18
	Verona Registry	Negative		27a	19	16	13	12b	17
		PgR status
	San Antonio	Positive				57	63	64	66
	INT Milan	Negative	51	40	39	48	42	37	36
	Verona Registry	Negative		33a	30	41	42	36b
		Ploidy:
	San Antonio	Diploid				46	50	52	52
	INT Milan	Diploid	19	32	27	23	27	33	34
		Proliferation:
	San Antonio	High S-phase #				29	24	20	19
	INT Milan	TLI⁎ >3%	63	58	54	53	46	43	38
	Verona Registry	Ki 67 >30		16a	10	9	10	8b
		p53:
	San Antonio	Negative				51	55	59	61
	INT Milan	Positive	30	25	24	23	17	18	18
		c-erbB2:
	San Antonio	Negative				79	85	86	90
	Verona Registry	Negative		53a	52	57	61	64b
		Differentiation:
	Verona Registry	G3		40a	32	28	25	28b

a Age group ⩽44 years. b Age group ⩾75 years. ⁎ [3H]Thymidine labelling index. # Fraction.

Tumour characteristics as predictors of local relapse risk 

Many studies have tested for associations between LR and tumour pathology, but the data conflict and are difficult to interpret. The results of well-conducted univariate and multivariate analyses for LR in three radiotherapy trials (EORTC 10801, DBCG-82TM, EORTC boost trial) and the EBCTCG overview of radiotherapy effects after breast conservation surgery are summarised in Table 5, Table 6 [32], [33], [38]. Tumour size was correlated to LR risk in both the EORTC boost trial (univariate and multivariate analyses) and the EBCTCG overview, but not in the EORTC 10801 or DBCG-82TM trials. Axillary node status was associated with LR only in the EBCTCG overview and only in univariate analysis, making it difficult to compare to multivariate analyses in the other three trials. All datasets showed a similar correlation between tumour grade and LR in univariate analysis. In the multivariate model, grade was significant only in the EORTC boost trial. ER status influenced LR rates in the univariate analysis, but had no effect in the multivariate model. Lymphovascular invasion was the only tumour-related factor that had an independent prognostic value for LR in the EORTC 10801 and DBCG-82TM trials. In the attempt to investigate these issues further, multivariate analyses of retrospective data from five single institution series were taken into account [19], [67], [68], [69], [70]. The only tumour-related factor that had a prognostic value in all series that analysed it (four out of five) was the presence of lymphovascular invasion. Tumour grade was correlated with LR only in one study [68]. Tumour size and nodal status did not reach statistical significance in any of the series. Several retrospective studies have also analysed the relationship between lymphovascular invasion and other tumour characteristics, showing a correlation of the presence of invasion with larger size, higher grade, positive nodal status [71], [72], [73]. A more systematic reporting and evaluation of lymphovascular invasion in future prospective trials should be considered in order to understand further the mechanisms underlying the prognostic significance of this pathologic feature.

Table 5. Univariate analysis of age and tumour factors for LR in the EORTC 10801 and DBCG-82TM trials [32], EORTC 22881-10882 trial [33] and EBCTCG overview [38]

		EORTC 10801 and DBCG-82TM		EORTC 22881-10882		EBCTCG overview
			10-year LR rate (95% CI)		5-year LR rate (95% CI)		5-year LR risk
							RT	no RT
	Age	⩽35 years	35% (17–53)	⩽40 years	14.5% (11.1–18.0)	<50 years	10.9%a	33.1%a
		36–40 years	9% (2–17)	41–50 years	7.24% (5.80–8.72)	50–59 years	7.0%a	22.7%a
		41–50 years	9% (7–13)	51–60 years	3.75% (2.84–4.66)	60–69 years	4.2%a	16.3%a
		51–60 years	11% (7–16)	>60 years	3.22% (2.35–4.09)	⩾70 years	2.6%a	13.1%a
		>60 years	7% (4–11)
	Tumour size	⩽1.0cm	4% (0–8)	Log diameter dominant lesion in mm	HR: 1.19 (1.02–1.37)	1–20mm	5.3%a	20.4%a
		1.1–2.0cm	13% (9–16)			21–50mm	14.2%a	35.2%a
		⩾2.1cm	11% (7–15)
	Nodal status	N−	10% (7–13)	N0	5.41% (4.70–6.12)	N−	6.7%	22.9%
		N+	11% (7–15)	pN1 (1–3)	5.23% (3.73–6.86)	N+	11.0%	41.1%
				pN2 (>4)	4.92% (1.75–8.17)
	Grade	G1–G2	7% (5–10)	G1	3.1% (1.9–4.4)	G1	4.3%a	13.5%a
		G3	15% (10–19)	G2	5.3% (3.1–7.7)	G2	9.2%a	26.1%a
				G3	10% (7.0–14)	G3	12.1%a	33.5%a
	ER status	Not analysed		ER−	7.31% (5.68–9.03)	ER−	12.3%a	30.4%a
				ER+	4.53% (3.73–5.34)	ER+	6.0%a	25.2%a
	VI	VI−	8% (6–11)	Not analysed		Not analysed
		VI+	15% (10–20)

VI: Vascular invasion.

a Subgroup analysis in node-negative patients.

Table 6. Multivariate analysis of tumour characteristics associated with local relapse (LR) risk in the EORTC 10801 and DBCG-82TM trials [32] and EORTC 22881-10882 trial [33]

		EORTC 10801 and DBCG-82TM		EORTC 22881-10882
			HR for LR (95% CI)		HR for LR (95% CI)
	Age	⩽35 vs >60	9.24 (3.74–22.81)	⩽40 years	1.00
		36–40 vs >60	1.53 (0.54–4.33)	41–50 years	0.52 (0.36–0.76)
		41–50 vs >60	1.32 (0.60–2.86)	51–60 years	0.28 (0.19–0.42)
		51–60 vs >60	1.61 (0.75–3.46)	>60 years	0.26 (0.17–0.40)
	Tumour size	1.1–2.0 vs ⩽1.0cm	3.23 (0.97–10.72)		1.27 (1.11–1.45)
		⩾2.1cm vs ⩽1.0cm	2.88 (0.80–10.19)
	Nodal status	pN+ vs pN−	0.77 (0.44–1.36)
	Grade	G3 vs G1–G2	1.62 (0.97–2.71)	G1	1.00
				G2	1.59 (0.95–2.66)
				G3	1.73 (1.04–2.86)
	ER status			ER+ vs ER−	0.91 (0.65–1.26)
	VI	VI+ vs VI−	2.31 (1.34–4.00)

VI: Vascular invasion.

A prognostic algorithm to predict local relapse risk has been developed based on the literature 

An estimate of relative risk of LR for each prognostic factor on the basis of published data has been made during the development of a computer-based model (designated IBTR!) constructed to predict individual LR risk after breast conservation surgery with and without radiotherapy [74]. Among the tumour characteristics, lymphovascular invasion was attributed the highest relative risk (RR): absent=1.00, present=1.76. The RRs for tumour size and grade were smaller: T ⩽1cm=0.80, T 1.1–2cm=1.00, T >2cm=1.36; G1=0.79, G2=1.00, G3=1.29. Nodal and ER status were not included in the model, since their impact on LR could not be estimated independently of systemic therapy. It was assumed that all patients with node positive disease would receive endocrine therapy and/or chemotherapy, which were included among the variables of the model. Use of endocrine therapy and chemotherapy was attributed RR of 0.41 and 0.72, respectively (no use=1.00). Factors attributed the greatest prognostic value in IBTR! were margin status (negative margin RR=1.00, close [0–2mm]=1.82 and positive=3.23) and age at diagnosis (e.g. >70 years=0.63, 51–55 years=1.00, <40 years=2.07).

In IBTR!, each variable (age, margin status, lymphovascular invasion, tumour size, tumour grade, tamoxifen and chemotherapy use) is treated as independent of the others. This is not consistent with the results of the multivariate analyses mentioned above, and it may introduce inaccuracies at the two extremes of the prognostic spectrum. To evaluate the magnitude of these uncertainties, and the utility of this tool in assisting with decision making, validation on independent datasets is needed.

Gene-expression profiles do not yet identify a low-risk group 

Gene-expression profiling has improved the characterization and prognostic stratification of breast cancer [75], [76]. Several models have been used to identify and classify subtypes with distinct gene-expression patterns. Three alternative gene-expression profiles (70-gene model, wound-response signature, hypoxia-induced profile) have been tested for their ability to predict LR after breast conserving treatment in individual patients [77]. Only the wound-response signature, combined with a standard supervised analysis (an optimization method that uses clinical data on LR), showed evidence of identifying patients with different LR risks: the 10-year recurrence rate in the validation set was 5% for the low-risk group and 29% for the high-risk group. The sensitivity and specificity of this classifier were 88% and 74%, respectively. The prognostic value of the supervised wound-response model was also shown in the multivariable analysis that included known clinicopathologic factors for LR; the estimated HR for LR was 16 (95% CI 1.9–125). Before gene-expression profiles can be used in clinical practice to predict the individual risk of LR after breast conservation surgery, further studies on larger population series that include patients of all age groups (in the above-mentioned study all patients were less than 53 years old) and patients not receiving radiotherapy are needed.

The observation (no radiotherapy) arms of breast conservation trials suggest which patients might safely avoid radiotherapy 

Of 16 randomised trials [6], [7], [8], [9], [10], [34], [35], [78], [79], [80], [81], [82], [83], [84], [85] (Table 2) testing the effects of radiotherapy after breast conservation surgery, three identified a low-risk group of patients that potentially might be spared radiotherapy. In the Uppsala-Őrebro trial, women aged >55 years with pT1 pN0 carcinomas with no comedo or lobular component had a LR rate without radiotherapy of approximately 1% per year (11% at 10 years, 95% CI 4.0–18.0). The authors concluded that routine use of radiotherapy in this subgroup of patients has a low cost-effectiveness [34]. The Milan III trial proposed broader criteria to select women that could be spared post-operative radiotherapy: age >65 years and tumour <25mm; after local extensive resection and axillary dissection this subgroup of patients had a 10-year LR rate of 4.4% in the no-radiotherapy arm [82]. The CALGB trial C9343 tested the effect of radiotherapy after lumpectomy and tamoxifen in women aged 70 years or older with pT1 pN0 ER-positive tumours [8]. The actuarial 5-year LR rate was 4% in the no-radiotherapy arm, and the authors concluded that withholding radiotherapy is a realistic option in this subgroup. The 15-year reduction in breast cancer mortality due to radiotherapy would be around 1% in the subgroup identified in the Uppsala-Őrebro trial and <1% in the Milan III and CALGB trial C9343 trials. Based on the one-to-four ratio between breast cancer mortality and prevented LR reported by the EBCTCG overview [38], a further two out of 16 randomised trials identified a subgroup that would have a 15-year survival benefit of around 1% following radiotherapy [7], [9]. In a planned subgroup analysis of 611 patients with pT1 and ER-positive tumours enrolled in the Canadian study testing the effect of radiotherapy with tamoxifen, the 5-year LR rate was 5.9% in women treated without radiotherapy [7]. Similarly, the ABCSG study 8 reported an estimated 5-year actuarial LR rate of 5.1% in post-menopausal women not allocated radiotherapy with tumour size <3cm, G1-2, node-negative status and positive oestrogen and/or progesterone receptor status, [9]. In conclusion, the criteria that helped identify low-risk groups in five randomised trials were age, tumour size and ER status. Non-tumour factors influencing life expectancy will now be considered, since they also have an important impact on the magnitude of benefit to be expected from radiotherapy.

Comorbidities and age are independent and additive factors predicting lifetime freedom from LR 

The SEER-Medicare database of women treated by breast conservation surgery with or without radiotherapy (N=8724) between 1992 and 1999 has been analysed to take account of competing risk of death from any cause [63]. The adjusted number needed to treat (NNT) in order to prevent 1 LR was calculated using the 8-year survival probability point estimate, after stratification for age and comorbidities. At two ends of a spectrum were (i) a subgroup of women aged 70–74 years with no comorbidities, who had an adjusted NNT of 21 patients (95% CI 16–31) and (ii) a subgroup of women aged ⩾85 years with moderate to severe comorbidities (scores 2–9, see below), with an adjusted NNT of 125 patients (95% CI 94–185). These results translate in an absolute risk reduction for local relapse at 8 years of 4.8% in the former group and 0.8% in the latter.

The comorbidity score used to calculate the adjusted NNT from the SEER-Medicare dataset was the Charlson Weighted Comorbidity Index, which was previously tested for its ability to predict risk of death from comorbidities over a 10-year follow-up period in 685 breast cancer patients [86]. The relative risk for each increasing level of comorbidity index (Table 7) was 2.3 (95% CI 1.9–2.8). At 10 years, the survival rates of the testing population, stratified according to an index score of 0, 1, 2 and 3, were 93%, 73%, 52% and 45%, respectively. During the validation process, the impact of increasing age was also analysed; the relative risk for each additional decade of age was 2.4 (95% CI 2.0–2.9), with 40 years taken as the zero rank (age 40 years score=0, 50=1, 60=2, etc.). Similar results were found when the data were analysed using a comorbidity score developed by Kaplan and Feinstein, with a relative risk of 2.0 (95% CI 1.6–2.4) for each increased comorbidity rank and 2.3 (5% CI 1.8–2.8) for each additional decade of age. The results suggest that an increase of 1 in comorbidity score has the same detrimental effect on life expectancy as an additional decade of age. Overall, the studies confirm the importance of taking into account distinct age and comorbidity factors when constructing an algorithm for patients that might be spared radiotherapy after breast conservation surgery.

Table 7. Charlson Weighted Comorbidity index [86] (example: a patient with peripheral vascular disease and moderate liver disease would have a comorbidity score of 4)

	Assigned weights	Comorbidities
	1	Myocardial infarct
		Congestive heart failure
		Peripheral vascular disease
		Cerebrovascular disease
		Dementia
		Chronic pulmonary disease
		Connective tissue disease
		Ulcer disease
		Mild liver disease
		Diabetes
	2	Hemiplegia
		Moderate or sever renal disease
		Diabetes with end organ damage
		Any tumour
		Leukaemia
		Lymphoma
	3	Moderate or severe liver disease
	6	Metastatic solid tumour
		AIDS

It is not possible to integrate these variables and to propose a change in standard practice without prospective evaluation 

Much progress has been made in recent years in identifying factors associated with falling LR risk after breast conservation surgery. Cancer-related and treatment-related factors are undoubtedly important, but patient demographics and comorbidities are equally powerful influences on lifetime risk of LR. These factors are likely to vary considerably between countries and cultures, making it hazardous to propose standard criteria for evaluation. Proposals for selective avoidance of breast radiotherapy must also take into account the priorities of patients and health resource providers [87], [88]. Before changes in practice are introduced, impact on overall survival, quality of life and health economic consequences need to be evaluated in well-designed clinical studies. These need not be randomised, but do need to be every bit as rigorous as randomised trials, including central pathology review.