Biomarkers predictive of a response to extended endocrine therapy in breast cancer: a systematic review and meta-analysis

Biomarkers are critical tools for predicting prognosis and guiding treatment in breast cancer. Common breast cancer biomarkers include hormone receptors (e.g., estrogen receptors), which contribute to tumor subtyping [1, 2]. Women with tumors overexpressing the estrogen receptor (i.e., ER +) are recommended to take at least five years of adjuvant endocrine therapy (ET) [3]. Tamoxifen is guideline treatment for premenopausal women and an alternative to aromatase inhibitors for postmenopausal women. Aromatase inhibitors (i.e., anastrozole, letrozole, exemestane) are indicated only in postmenopausal women [3]. In a meta-analysis of clinical trials of five years of adjuvant ET in early-stage breast cancer, tamoxifen approximately halved recurrence rates [4], as did aromatase inhibitors in postmenopausal women [5]. Numerous clinical trials have confirmed these findings, providing a basis for the current guideline recommendation of at least five years of treatment [6].

Though the benefits of ET are pronounced, 20–40% of treated patients recur 5–20 years after diagnosis [7]. Recurrences have been documented even 39 years after primary diagnosis [8‐10]. This hazard of late recurrence suggests a benefit of extending ET beyond the traditional five-year course. Several trials have shown a modest survival benefit and reduction of recurrence risk with extended ET depending on the duration, type, and sequence of the drugs [11, 12]. Other trials have found no improvement in overall survival with extended treatment [13‐15]. The inconsistency in findings is further complicated as ET has long-term side effects. In the Adjuvant Tamoxifen, Longer Against Shorter (ATLAS) and the Adjuvant Tamoxifen-To Offer More (aTTom) trials—which evaluated extended tamoxifen use—there were increased risks of endometrial cancer and pulmonary embolism among women assigned to extended tamoxifen compared with placebo [11, 12]. Toxicities are also seen with long-term use of aromatase inhibitors, including increased risk of hypercholesterolemia, osteoporosis, fracture, and musculoskeletal syndrome [13‐18].

Although clinical trials show that continuing ET beyond five years reduces late recurrence risk, it is essential to balance benefits with the risks of overtreatment using predictive and prognostic markers [19]. A prognostic biomarker informs the likelihood of a clinical outcome independent of any treatment received. In contrast, a predictive biomarker provides information on individuals most likely to respond to a specific treatment, differentiating patients likely to benefit from patients unlikely to benefit. To determine whether a biomarker is predictive, the study must include individuals who were treated (i.e., with extended ET), to compare them with untreated patients (i.e., those who stopped treatment after five years) [20]. There are several predictive and prognostic tests recommended by the American Society of Clinical Oncology (ASCO) and the US National Comprehensive Cancer Network (NCCN), such as OncotypeDx [21]. These tests characterize women by their risk of recurrence and have been pivotal in identifying low-risk patients who can forego chemotherapy [22‐25].

The Breast Cancer Index (BCI) assay was developed in 2011 and is the only NCCN- and ASCO-approved test to predict benefit from extended ET [26]. The assay involves two parts: (1) the molecular grade index (MGI), a 5-gene predictor that measures tumor grade and proliferation and (2) the predictive panel, based on the expression ratio of HOXB13 and IL17BR (i.e., the H/I ratio or H/I) [27]. The BCI predictive panel stratifies patients into two groups: BCI (H/I) High, which indicates potential benefit from extended ET, and BCI (H/I) Low, which indicates low likelihood of benefit [21, 26‐28]. The 2022 ASCO guideline update recommended that the BCI test be used to guide decisions about extended ET among ER + patients with node-negative disease or 1–3 positive nodes [26]. However, the evidence in premenopausal and perimenopausal women, and in those with > 3 positive lymph nodes, is limited. Predictors of early and late recurrences may differ according to menopausal status, generating a potential evidence gap in this setting [31]. Such information–perhaps provided by tumor biomarkers–could help patients and providers decide whether extending ET is worthwhile.

In this systematic review, we aimed to evaluate studies investigating biomarkers predictive of response to extended ET. Rather than focusing on late recurrence risk prediction, this review only involved populations treated with extended ET or standard duration treatment and in whom a predictive biomarker was assayed to predict response to the extended treatment.

In this review, four studies examined the utility of a biomarker in predicting clinical benefit from extended ET using the BCI Predictive assay and one study examined the predictive ability of Ki67 and PgR status. Of these, predictive capacity was seen for Ki67 in one study and for BCI Predictive assay in three distinct study populations. Though Villasco et al. concluded that no predictive response was seen with Ki67 risk stratification, there did appear to be a distinction in late distant recurrence risk comparing low versus high Ki67 level [38]. Remaining included studies consistently showed that a BCI (H/I) High score predicted benefit from extended ET, while a BCI (H/I) Low score did not. The BCI Predictive assay measures estrogenic and other proliferative signaling pathways in the progression of breast cancer, providing a risk-based score on the predicted benefit of additional therapy after completing the standard five years of ET [41]. However, the low number of eligible studies in this systematic review highlights the need for further research in this setting.

Our search identified many studies that investigated the utility of biomarkers in predicting overall late recurrence risk. Late recurrences occur when dormant cells remain inactive for some time, before reactivating to cause relapse. The underlying biology of dormancy remains poorly understood but is an active area of research. The Early Breast Cancer Trialists’ Collaborative Group (EBCTCG) periodically reviews the continued follow-up in trials such as ATLAS and aTTom to evaluate strategies for reducing late recurrence [7]. In EBCTCG’s latest study on late recurrence risk, clinicopathological features such as original tumor/lymph node status and Ki-67 status were predictive of recurrence from 5 to 20 years (level of evidence: 1B) [7, 42]. In this systematic review, late recurrence risk was often deemed a proxy for individuals who may benefit from treatment extension. These studies use evidence from biomarkers that stratify the risk for late recurrence, but this addresses a different question than that of predicting extended ET benefit. Though these studies may hint at vulnerable patients, they do not evaluate the predictive ability of the biomarker itself. Other molecular tests (e.g., OncotypeDX) have shown some prognostic value in the setting of late recurrence, but are not recommended for decision-making due to the lack of predictive studies [26].

Recently published ASCO guidelines recommend BCI testing to assess potential benefit of extended ET in disease with negative nodes or 1–3 positive nodes. However, the recommendation is only supported by intermediate evidence quality and a moderate strength [26]. In the guideline, Andre et al. note that the collective evidence from five studies—three of which were identified in this systematic review (Bartlett et. al, 2019; Sgroi et. al, 2013; and Noordhoek et. al, 2021)—demonstrated a consistent predictive benefit of extended ET. Of the two studies not eligible for this review, one was published as an abstract only, so did not meet our a priori eligibility criteria [43]. Importantly, this study found that BCI (H/I) score was not predictive of an improvement in recurrence-free interval after extended letrozole therapy [43]. The second study that was ineligible for our review and cited in the ASCO guidelines investigated BCI in node-positive patients. Though the study included patients on extended therapy, it did not directly compare them to individuals who completed the guideline five years of therapy and thus did not evaluate predictive ability of the BCI (H/I) score [41]. The fourth study that was eligible in this review was published after release of the ASCO guidelines. In a changing landscape of treatments and biomarker testing availability, it is essential to generate more evidence to support these guidelines. For example, since the 2015 approval of the first cyclin-dependent kinase (CDK) 4/6 inhibitors, no study has investigated the combined role of these drugs in addition to extended ET [44, 45]. As treatments change and improve, we must continue to generate both trial-nested and real-world evidence to understand the dynamics of extended ET.

Our review also calls attention to the lack of generalizability resulting from features of the populations of included studies. One of the three studies—the Trans-aTTom trial—included premenopausal and perimenopausal women. However, these women only comprised about 8% of the population (n = 25 premenopausal; n = 28 perimenopausal) [32, 33]. ASCO guidelines state that their recommendations for the predictive ability of BCI ‘cannot definitively be made’ for premenopausal and perimenopausal women [26]. Additionally, cancer clinical trial participation has historically been predominately composed of non-Hispanic white and higher-income individuals, limiting the generalizability of findings [46]. For example, in the overall MA.17 trial, 91.9% of participants were non-Hispanic white women [13]. Not only are minority populations less likely to partake in clinical trials, but they are also less likely to receive biomarker testing [47]. In a meta-analysis of testing inequalities, lower socio-economic position was associated with a decrease in predictive biomarker test utilization (OR = 0.86, 95% CI 0.71, 1.05, 10 studies) [48]. The mean cost of the BCI Predictive assay is $3,450, and in the US is only covered by Medicare under certain criteria [49]. Given this high cost and lack of evidence in socio-economically disadvantaged populations, the recommendation of routine BCI testing is introducing what will inevitably be a disparity both in receiving these tests and in understanding their clinical utility in underrepresented populations.

Another theme in these studies was making inferences based on hypothesis tests of treatment/biomarker interaction terms. All BCI-related papers reported this statistic and used it as evidence supporting their conclusion of a predictive effect of BCI (H/I) score on extended ET response. In this test, a full model including an interaction term between BCI (H/I) score and extended ET is compared with a reduced model without an interaction term. A likelihood ratio test is used to determine whether the interaction term coefficient is statistically significantly different from zero based on a p-value threshold of 0.05. Significant interaction terms were interpreted as supporting the predictiveness of the BCI (H/I) test. In the context of log-linear models, this tests whether there is a departure from multiplicativity of effects, which is often difficult to interpret, particularly when evaluating the predictive ability of a biomarker [50]. If available, future studies should consider investigating departure from additive effects or using stratified estimates of effect to measure of interdependence [51]. Regardless, caution should be taken when interpreting results from likelihood ratio tests in this setting.

This review outlines the limited research on biomarkers that predict a benefit from extended ET, including by use of commercially available tests, such as BCI. It is important to include premenopausal and perimenopausal women in future studies, as current studies in this area have nearly no representation of these important subpopulations of breast cancer patients who face the longest time at risk for recurrence. Additionally, diverse trial populations are essential, both because biomarker testing is differentially offered to many minority populations, but also because of lacking diverse trial. As breast cancer survival improves, the need to personalize treatment decisions will become increasingly important. Without sufficient evidence, healthcare teams and patients will face a difficult decision in balancing the benefits and risks of ET extension.