The efficacy of molecular subtyping in predicting postoperative recurrence in breast-conserving therapy: a 15-study meta-analysis

Background Recent research displays that breast cancer (BC) is a heterogeneous disease and distinct molecular subtypes yield different prognostic outcomes. Methods We conducted a meta-analysis to clarify the role of molecular subtypes in recurrence risk after breast-conserving therapy (BCT). Eligible studies of single- (ER, PR, Her-2, and p53) and triple-molecular (Luminal A, Luminal B, Her-2, triple-negative) subtypes were identified through multiple search strategies. Pooled hazard ratios with 95% confidence intervals were calculated to assess this research topic. Results Fifteen studies involving 21,645 participants were included in the meta-analysis. Her-2 positive patients had a significantly higher recurrence risk in both overall merge (HR = 1.97, 95% CI: 1.41-2.75) and subtotal merge of local recurrence (LR) (HR = 1.93, 95% CI: 1.34-2.78). Significantly higher risk of recurrence was also observed in p53 positive patients by overall merge (HR = 1.78, 95% CI: 1.49 -2.12) and subtotal merge of LR (HR = 1.73, 95% CI: 1.44-2.07). When setting Luminal A as a baseline, Luminal B, Her-2, and triple-negative all showed significantly increased risk for both LR and distant recurrence (DR). Comparing triple-negative and non-triple-negative subtypes showed the biggest risk for overall recurrence (HR = 3.19, 95% CI: 1.91-5.31) and LR (HR = 3.31, 95% CI: 1.69-6.45). Conclusions Our meta-analysis showed significant differences in recurrence risk among various molecular subtypes after BCT. Although Her-2 and p53 positive subtypes can be considered independent prognostic biomarkers for indicating high LR risk, triple-molecular biomarkers showed higher clinical value. Triple-negative subtype showed the highest recurrence risk among all subtypes, and adjuvant chemotherapy should be considered for it.

The factors affecting recurrence are complex, including clinical and histological characteristics, with or without postoperative adjuvant radiotherapy and systemic therapy (chemotherapy and/or hormone therapy) [11]. Recent researches display that BC is a heterogeneous disease [12] and distinct molecular subtypes yield different prognostic outcomes [13][14][15][16][17][18][19]. These molecular markers mainly include estrogen receptor (ER), progesterone receptor (PR), human epidermal growth factor receptor 2 (HER-2), p53, and Ki67 [14,20]; these have immensely contributed to the selection of the optimal strategy for BCT [21][22][23]. However, the impact of molecular subtypes on LR or distant recurrence (DR) has not been systemically evaluated. Therefore, we conducted a meta-analysis to clarify the role of molecular subtypes in BC recurrence after BCT.

Search strategy
Original articles analyzing the hazard ratio (HR) of recurrence after BCT in different BC molecular subtypes were searched by online databases PubMed, Embase, and Web of Science. We selected studies carefully by the following sets of key words variably combined: 'breast cancer', 'breast-conserving surgery', 'breast-conserving therapy', 'recurrence', 'hazard ratio', 'molecular marker', and 'molecular subtype'; the last search update was performed on 10 January 2014. All eligible studies published in English were reviewed, and their bibliographies were also examined for other relevant publications. Relevant review articles were manually searched to find additional eligible studies. If more than one article was published using the same series of study subjects, we only chose the latest or most complete study for this meta-analysis. All the studies enrolled in this meta-analysis have been performed with the approval of an appropriate ethics committee. Researches carried out on humans are all in compliance with the Helsinki Declaration.

Inclusion and exclusion criteria
We followed the guidelines of the critical checklist proposed by the Dutch Cochrane Centre Meta-analysis of Observational Studies in Epidemiology (MOOSE) [24]. Articles were identified as eligible when they fit the following criteria: (1) they performed BCS on BC patients; (2) they focused on postoperative recurrence; (3) they investigated the association between recurrence and different BC molecular subtypes. Altogether, 953 studies were excluded by exclusion criteria and further quality evaluation that are presented in Figure 1.

Data extraction
All data were carefully extracted from eligible publications in duplicate by two co-authors (PJ and JZ). Any disagreements were resolved by discussion between the two authors. The extracted data elements in Table 1 include the followings: first author's last name, publication year, case nationality, dominant ethnicity, study design, number of cases, median time to follow-up, percentage of patients who received adjuvant radiotherapy (RT), percentage of patients who received adjuvant systemic therapy (AST), types of recurrence, source of HR, and characteristics of enrolled cases.
The extracted data elements in Table 2 show the HRs and 95% confidence intervals (CIs) of different recurrences among molecular subtypes. If HR was not reported directly, data were extracted from Kaplan-Meier curves of survival outcomes to extrapolate required data using the previously described methods [25][26][27]. We also wrote emails to the corresponding authors of enrolled studies to obtain additional information and original data needed for the meta-analysis.

Statistical analysis
All statistical analyses were conducted using Stata (version 11.0; StataCorp LP, College Station, TX, USA) and Excel (version 2007; Microsoft Corp., WA, USA). The aggregation of HRs and 95% CIs were calculated following Tierney's method [27]. Forrest plots were used to estimate the effect of different molecular subtypes on recurrence after BCT. The heterogeneity assumption of pooled HRs was verified by Cochran's Q-test, and the percentage of Higgins' I-squared statistic (I 2 ) was used to quantify the extent of heterogeneity explained by these characteristics of enrolled studies. If significant heterogeneity was observed (P <0.1 or I 2 >50%), a random-effects model (Der Simonian-Laird method) was applied; otherwise, the fixed-effects model (Mantel-Haenszel method) was adopted [28]. Potential publication bias was determined by Egger's linear regression test with a funnel plot [29].
To avoid the influence of heterogeneity among these studies, we also conducted a subgroup analysis stratified by different recurrence categories (LR and DR). We defined LR to include true recurrence, local failure, local recurrence or relapse, ipsilateral breast tumor recurrence, locoregional recurrence, and second ipsilateral BC. DR was defined as distant metastasis or failure, and recurrence was defined as any of LR and/or DR, the merger of LR and DR in the same study or not definitely mentioned. All P values were two-sided and a P value less than 0.05 was considered to be statistically significant.

Single-molecular subtypes and postoperative recurrence
Single-molecular subtypes are defined as the dichotomous status of a single receptor or protein, which will be excluded from the meta-analysis if less than three studies are found involved in. Altogether, four proteins (ER, PR, Her-2, and p53) are included in analyses.

Publication bias
Unexpectedly, the comparison of TN vs. LA (Figure 3c) shows obvious publication bias by Egger's test (P = 0.021). However in the other seven molecular typing comparisons, the shapes of funnel plots seem symmetrical and the results of Egger's test do not suggest any publication bias (all P >0.05) (Figure 2a, b, c, d and Figure 3a, b, d).

Discussion
BC has become a major cause of morbidity and mortality in women [46]; there are an estimated 5.2 million BC survivors worldwide [47]. A recent study reported that the incidence of BC in the United States from 2000 to 2009 showed a recent increase, especially for early-stage disease (in situ and localized) in non-Hispanic blacks and Asian/Pacific Islanders [48]. In addition, from 1976 to 2009, the incidence of advanced BC rose significantly among young women in the US (from 1.53/100,000 to 2.9/100,000); this trend may be accelerating and seems confined to women aged 25 to 39 years [49].
Currently, local excision combined with adjuvant RT has been well-established as an optimal treatment strategy for early-stage BC [2,11,50,51]. This may be an overtreatment if RT is imposed on any patient after BCS. Recurrence risk in a considerable proportion of patients is sufficiently low, especially DCIS diagnosed through screening of healthy women [52]. In contrast, for patients with high recurrence risk where only RT may be insufficient, AST (HT and/or CT) should also be administered after BCS [30]. In our meta-analysis, 16 enrolled studies display extreme difference in rates for RT (0% to 100%), HT (0% to 77%), and CT (0% to 66.4%) based on the diversity of patient age, pathological types, clinical

02) -
Local recurrence is defined to include true recurrence (TR), local failure (LF), local recurrence or relapse (LR), ipsilateral breast tumor recurrence (IBTR), locoregional recurrence (LRR), or second ipsilateral breast cancer (SIBC). Distant recurrence is defined to include distant metastasis (DM) or distant failure (DF). Recurrence is defined to include any local and/or distant recurrence, which is the merge of LR and DR in the same study or not definitely mentioned.  stages, and surgical margin (Table 1). Therefore, definite evaluation criteria are needed to identify patients at high recurrence risk from those at low risk, and to help choose a more precise postoperative treatment strategy for patients undergoing BCS.
Towards this goal, consideration must be given to the combination of underlying pathological and clinical characteristics. Recent studies have shown that molecular subtypes are prognostic for LR and DR after BCS, and immunohistochemical staining is often used to approximately identify these subtypes [18,19,53]. In this review, we take both single-molecular and triple-molecular typing into evaluation (Table 2). To our knowledge, this is the first meta-analysis study to comprehensively assess the effect of molecular subtypes on recurrence after BCT.
Our meta-analysis shows that there are significant differences in recurrence risk among various BC subtypes after BCT. First, we analyzed the efficacy of singlemolecular subtypes. Although a slightly increased risk has been found in both ER negative and PR negative patients, there is no significant effect on LR after BCT. Only Her-2 positive patients display a significantly higher risk for local and overall recurrence when compared to Her-2 negative individuals (Table 3). It reveals that Her-2 positive status could be used clinically as an independent prognostic factor of high recurrence risk, and the status of Her-2 is the most important in the combined efficacy of triple-molecular subtypes. In addition, overexpression of p53 (p53 positive) is presumed to be a surrogate for TP53 mutations, which are associated with higher tumor grade [54]. Our results display that p53 positive patients have a significantly increased risk for overall recurrence and LR when compared to p53 negative (low expression) individuals (Table 3). Thus, p53 positive subtype should also be considered an important, independent prognostic biomarker for indicating high recurrence risk. P53 positive individuals are more necessary to accept adjuvant RT combined with or without AST and will benefit more than those with p53 negative BC for preventing postoperative recurrence because p53 positive subtype also predicts better responsiveness to both RT and AST [54]. Apart from the above mentioned four biomarkers, we did not find any other receptor or protein that met our inclusion criteria. Although Ki67 is generally accepted as one of the most important molecules for BC typing and has been studied over a long period of time [55], only one study was found focusing on our subject [33].
Recently, more studies have focused on the combined efficacy of ER, PR, and Her-2 receptors since the 12th St Gallen International Breast Cancer Conference (2011) Expert Panel adopted a new approach to BC classification for therapeutic purposes based on the recognition of intrinsic biological subtypes [55]. Therefore, we also put emphasis on the relationship between triple-molecular subtypes and recurrence risk after BCT. When compared with Luminal A, Luminal B, Her-2, and triple-negative subtypes all show significantly increased risk for both LR and DR. All HR values for overall recurrence based on triplemolecular typing are greater than 2.0, which can identify patients with higher recurrence risk and shows more clinical prediction value than analyses based on singlemolecular typing (Table 4). For example, the HR value of LR (HR = 2.33) derived from triple-molecular comparison between Her-2 and Luminal A subtypes is larger than the HR value of LR (HR = 1.93) derived from single-molecular comparison between Her-2 positive and Her-2 negative subtypes. Therefore, we concluded that the clinical application of triple-molecular typing as a biomarker can better distinguish high-risk individuals compared to single-molecular typing.
Moreover, the comparison between triple-negative and non-triple-negative subtypes shows the biggest risk difference for overall recurrence (HR = 3.19) and LR (HR = 3.31) among all molecular typing comparisons. When setting Luminal A as a baseline, the HR value of triplenegative subtype (HR = 2.90) remains larger than that of Luminal B (HR = 2.23) or Her-2 subtypes (HR = 2.26) for overall recurrence (Table 4). Previous studies have shown that triple-negative receptor status is strongly associated with poor clinical outcomes [56], and that young women more frequently suffer from triple-negative tumors [57]. Thus, triple-negative subtype should be considered the biggest risk factor for recurrence and adjuvant CT should be administered in BCT.
Some limitations of this meta-analysis should be acknowledged. First, there are only three studies focused on Asians [35,36,42] and none on Africans in this metaanalysis, hindering comprehensive investigation of the association between BC molecular typing and recurrence risk after BCT. Second, the sample sizes of enrolled researches (from 39 to 8,724) vary widely, and therefore the statistical power or weight of each study is greatly different, inevitably causing bias to varying degrees. Third, the number of original studies focusing on this topic is insufficient, especially studies related to DR risk.

Conclusions
Our meta-analysis represents a quantified synthesis of all published studies and shows significant differences in recurrence risk among various molecular subtypes after BCT. Her-2 positive and p53 positive subtypes can be considered independent prognostic biomarkers for indicating high LR risk, but triple-molecular biomarkers exhibit higher clinical value than single-molecular biomarkers. Moreover, triple-negative subtype shows the biggest risk for overall recurrence and LR among all molecular subtypes and adjuvant CT should be considered in BCT. Considering the insufficient number of original studies, further research with different ethnicities is needed on this topic, especially the intensity of association between molecular typing and DR risk after BCT.