The Usefulness of Prognostic Tools in Breast Cancer Patients with Brain Metastases

Simple Summary Due to the variability of an individual’s prognosis and the variety of treatment options available to breast cancer (BC) patients with brain metastases (BM), establishing the proper therapy is challenging. Since 1997, several prognostic tools for BC patients with BM have been proposed with variable precision in determining the overall survival. The majority of prognostic tools include the performance status, the age at BM diagnosis, the number of BM, the primary tumor phenotype/genotype and the extracranial metastases status as an outcome of systemic therapy efficacy. It is necessary to update the prognostic indices used by physicians as advances in local and systemic treatments develop and change the parameters of survival. Free access to prognostic tools online may increase their routine adoption in clinical practice. Clinical trials on BC patients with BM remains a broad field for the application of prognostic tools. Abstract Background: Determining the proper therapy is challenging in breast cancer (BC) patients with brain metastases (BM) due to the variability of an individual’s prognosis and the variety of treatment options available. Several prognostic tools for BC patients with BM have been proposed. Our review summarizes the current knowledge on this topic. Methods: We searched PubMed for prognostic tools concerning BC patients with BM, published from January 1997 (since the Radiation Therapy Oncology Group developed) to December 2021. Our criteria were limited to adults with newly diagnosed BM regardless of the presence or absence of any leptomeningeal metastases. Results: 31 prognostic tools were selected: 13 analyzed mixed cohorts with some BC cases and 18 exclusively analyzed BC prognostic tools. The majority of prognostic tools in BC patients with BM included: the performance status, the age at BM diagnosis, the number of BM (rarely the volume), the primary tumor phenotype/genotype and the extracranial metastasis status as a result of systemic therapy. The prognostic tools differed in their specific cut-off values. Conclusion: Prognostic tools have variable precision in determining the survival of BC patients with BM. Advances in local and systemic treatment significantly affect survival, therefore, it is necessary to update the survival indices used depending on the type and period of treatment.


Introduction
Brain metastases (BM) are a serious consequence of breast cancer (BC) progression. They occur in up to a half of patients with metastatic human epidermal growth factor Table 1. Prognostic tools developed on mixed groups of patients with BM (from year 1997 to 2021).

Author, Prognostic Tool, Year of Publication N (Total and BC) Prognostic Factors Included into Index Prognostic Class/Groups and OS (m)
Gaspar [6]

Author, Prognostic Tool, Year of Publication N (Total and BC) Prognostic Factors Included into Index Prognostic Class/Groups and OS (m)
Sperduto [11]

General Prognostic Models in BM
In 1997, the Radiation Therapy Oncology Group (RTOG) developed the first prognostic score for BM patients by using a Recursive Partitioning Analysis (RPA) strategy [6]. The study based its analysis on 1200 patients enrolled in 3 RTOG trials conducted from 1979 to 1993 that were investigating WBRT with or without radiosensitizing agents. Breast cancer patients constituted only 12% of the whole group (N = 137). Out of 13 prognostic variables analyzed, 4 were statistically significant: KPS, the age, the status of the primary lesion and evidence of other systemic metastases. Based on these 4 variables, 3 main classes were established, "class I (KPS ≥ 70, primary lesion controlled, age < 65 years and BM only), class II (not class I or III) and class III (KPS < 70) with a median overall survival (OS) of 7.1, 4.2 and 2.3 months," respectively. In 1999, Lagerwaard et al. [7] developed the Rotterdam Scoring System (RSS) analyzing 1292 patients with BM retrospectively, in whom 213 (16%) were BC patients. Out of the 12 variables, 3 were statistically significant predictors of OS: performance status ECOG scale, clinical response to steroids (good, moderate, little) and extracranial disease/metastases (ECM) status (no, limited, extensive). RSS divided patients into 3 different groups with a median OS of 6.3, 3.4 and 1.3 months.
Lorenzoni et al. [9] proposed the Basic Score for Brain Metastases (BS-BM) based on the outcome of 110 patients treated with radiosurgery from 1999 to 2003. Out of the 10 potential prognostic factors, three built BS-BM: KPS (50-70, 80-100), primary tumor control (yes vs. no) and ECM (yes vs. no). The four prognostic groups were selected with a median survival of 32, 13, 3.3 and 1.9 months.
To predict the survival of BM patients treated with WBRT, Rades et al.
[10] created a scoring system by retrospectively analyzing 1085 patients. Four significant factors were identified: KPS (≥70 vs. <70), age (≤60 vs. >60), ECM (yes vs. no) and interval between tumor diagnosis and WBRT and WBRT (≤8, >8 months). Patients were divided into four groups and for each group, survival was compared for a short (5 × 4 Gy) and a longercourse of WBRT (10 × 3 Gy/20 × 2 Gy). The 6-month survival in 4 groups were 6%, 15%, 43% and 76%, respectively. Based on this data, the authors concluded that patients with a better prognosis should receive a longer course of radiation with a lower dose per fraction to avoid neurotoxicity.
Rades et al. [12] updated his index by an analysis of a group of 1797 patients after WBRT. The rate of BC patients with BM was not reported in this group. Out of the 7 potential variables, five impacted the OS: KPS (<70 vs. 70 vs. >70), age (≤60 vs. >60 years), ECM (yes vs. no), the interval between the tumor diagnosis and WBRT (≤6 vs. >6 months) and the number of BM (1 vs. 2-3 vs. ≥4). The 6-month survival rates in the 3 groups were 9%, 41% and 78%.
Barnholtz-Sloan et al. [13] developed a nomogram based on 2367 patients from 7 clinical trials. There were 291 (12%) patients with BC in the analyzed group. The nomogram included variables as follows: the primary site and histology, the status of primary disease, ECM, age, KPS, and the number of BM. The authors concluded that the nomogram provided an individual with an estimate of his or her survival and better informed an oncologist on the patient's prognosis.
In 2012, Yamamoto et al. [14] proposed a sub-classification of RPA class II, originally developed by Gaspar [6], by dividing this class into 3 subclasses. The reason was that in the RPA classification the majority of patients were in class II and clinical factors varied widely within the category. In the group of 3753 patients after radiosurgery without WBRT, 282 (7%) were BC patients.

Breast Cancer-Specific Prognostic Models in BM
In 2005, Claude et al. [19] retrospectively reviewed 120 patients with BC and BM treated with WBRT. The prognostic factors (age, performance status, tumor characteristics and pretreatment modalities) were analyzed and two of them had the statistical power to predict the OS: ECOG 0-1 vs. >1 and lymphopenia ≤700 G/L vs. >700 G/L. The OS of patients with good and poor prognoses were 7.0 and 2.0 months.
Le Scodan et al. [20] analyzed 117 BC patients with BM treated with WBRT from 1998 to 2003. Nine prognostic variables were assessed as potential predictors of the OS. Three adverse prognostic factors were selected: RPA class III or KPS < 70, negative hormone receptor (HR) status and lymphocyte count ≤700 G/L. The OS of patients without negative prognostic factors was 15.0 months, 5 months with one adverse prognostic factor and 3 months with more than one prognostic factor.
In 2009, Park et al. [21] retrospectively analyzed 125 patients with BC and BM. Eight prognostic factors were analyzed and only three of them (poor performance status ECOG ≥2, HER2-positivity and no additional systemic therapy) were identified as risk factors for a worse prognosis.
Rades et al. [28] constructed a simple model to estimate the 6-month survival based on the data of 230 BC patients and BM treated only with WBRT. The two selected prognostic variables were KPS (<70 vs. ≥70) and ECM (yes vs. no). Three groups were defined, with the rate of 6 months survival of 10%, 55% and 78%, respectively.
In 2014, Ahluwalia et al. [29] validated the Breast-GPA (Sperduto 2012) and revealed that it was prognostic for the OS; however, separation between the groups was variable. Xiong et al. [33] established two new nomograms estimating the individual OS and BC-specific survival, based on the data of 789 BC patients and BM from the Surveillance, Epidemiology, and End Results (SEER) program. There were 6 significant variables constructed in the nomogram: age (18-49, 50-64, ≥65), tumor subtype (luminal A, luminal B, triple-negative, HER2 positive), chemotherapy (yes vs. no)), surgery (yes vs. no), the number of ECM sites (0, 1, 2, 3) and the median household income (high vs. low).
In 2020, Sperduto et al. [34] published updated the Breast Graded Prognostic Assessment Liu et al. [36] proposed and validated a new nomogram, named NCCBM using a large cohort of 975 BC patients from the SEER database diagnosed from 2011 to 2014. The following variables were selected: age, race, surgery, radiation therapy, chemotherapy, laterality, grade, molecular subtype, and extracranial metastatic sites. The c-index of the NCCBM was 0.69 (95%CI, 0.67 to 0.71) in the training set and 0.70 (95%CI, 0.68 to 0.73) in the validation set. The main limitation of this nomogram was that clinical data on the tumor subtype and distant metastases were collected only after 2010 in the SEER database and therefore limited the sample size of this study. Further, information about disease recurrence or subsequent sites of disease involvement was not collected and therefore this study was unable to investigate patients who developed BM later in their disease course. Moreover, lack of detailed treatment information for patients with BM and KPS does not allow for comparison NCCBM to other prognostic models directly.

External Validation of the Applicability of Different Survival Prognostic Scores
Three separate groups of scientists [20,37,38] confirmed the prognostic value of the RPA score (Gaspar, 1997) in patients with BC and BM. Nieder et al. [22] in 2009 validated RPA, GPA, BSBM, SIR and Rades score on the group of 85 BC patients and revealed that the scores that performed best were RPA and SIR but the c-index was not used. In 2011, Villa et al. [39] validated 3 prognostic indices: RPA, GPA and BS-BM on a group of 285 patients (17% BC patients). Harrell's c-index values were 0.58, 0.61 and 0.58 for the GPA, BSBM and RPA, respectively. The authors concluded that these indices had a limited long-term prognostication capability.
Ahn et al. [25] in 2012 validated the Breast-GPA on a group of 171 BC patients treated from 2000 to 2008. In this cohort, the Breast-GPA did not discriminate among prognostic classes and the prediction model for the 1-year survival probability (area under the curve (AUC) of 0.55). In turn, Braccini et al. [40] in 2013 compared 7 published prognostic indexes: RPA (Gaspar, 1997), GPA (Sperduto, 2008), BS-BM (Lorenzoni, 2004), Breast-RPA (Niwińska, 2012), Breast-GPA , Le Scodan's score (2012) and the clinico-biological score developed in the phase I study (The Phase 1 Prognostic Score, P1PS) [41] in 250 BC patients with BM treated from 1995 to 2010. The analysis revealed that all the indices were able to discriminate patients with statistical significance (p < 0.001) for the OS according to the prognostic category. Pairwise comparisons of each prognostic index revealed statistically significant differences in survival between prognostic classes, except for the breast-GPA classes I vs. II, BS-BM scores 1 vs. 2 and Le Scodan's scores I vs. II. There were no significant differences between all prognostic indices concerning the survival predicting ability. Only minor differences were seen using Harrell's c-index (range 0.60-0.68). The authors concluded that RPA seemed to be the most useful score. RPA performed better than new prognostic indices because it was the most accurate in identifying patients with long (>12 months) and short (<3 months) survival.
In The analyses revealed that the majority of the scores were associated with the OS, but the GPA was the most accurate at identifying patients with a survival of more than 1 year and Breast-GPA was the best in selecting patients with a survival of less than 3 months.
Huang et al. [32] in 2018 performed an external validation of the RPA, GPA and Breast-GPA on a group of 411 patients. The analysis revealed overlapping between groups I and II in RPA, and between groups II, III and IV in GPA, and unsatisfactory discrimination between groups II and III in Breast-GPA. It was concluded that in those three prognostic models, the value of differentiating a patient's survival was not satisfactory. The c-index was 0.64, 0.61 and 0.63 for the nomogram Breast-GPA, GPA and RPA, respectively.
Znidaric et al. [46] in Lee et al. [15] in 2020 proposed a prognostic index for BM (PIBM) and performed a validation of 4 indices (RPA, SIR, BSBM, GPA) on a group of 311 BC patients treated with gamma knife radiosurgery. All four indices were comparable in regards to the prognostic ability (AUC range 0.59-0.63) and PIBM had the highest prognostic power (AUC~0.66).

Discussion
Since the first Recursive Partitioning Analysis published in 1997, a series of prognostic tools have been developed in BC patients with newly diagnosed BM to facilitate clinical decision-making and appropriate stratification to local and systemic therapy. This review of prognostic tools illustrates how over time, progress in the understanding of the biology of BC and new effective systemic treatments have influenced the prognosis of patients with BM .
The majority of prognostic tools in BC patients with BM include the performance status, the age at BM diagnosis, the number of BM (rarely volume), the primary tumor phenotype/genotype and the ECM status as an outcome of the systemic therapy efficacy (Table 3) [19][20][21][22][23][24][25][26][27][28][29][30][31][32][33][34][35]. Although the results of several analyses agree that patients with better performance status achieve a longer survival, several researchers emphasize that this criterion carries a risk of bias because the clinical assessment of performance status is subjective. The most often used scale to assess the performance status was Karnofsky's scale with a cut-off point of 70 (i.e., the patient could not work, but did not require care) (Tables 2 and 3). Some studies described 3 prognostic subgroups: patients in a very good general condition (90-100 points), patients in a poor general condition (≤60 points, or <70 points) and the intermediate group (61-90 points or 70-89 points). Similar controversies concern age because the biological age often differs from the chronological age. The most often used cut-off for age in prognostic tools was 60 or 65 years, less often 50 (Tables 2 and 3). Also, the prognostic significance of the number of BM has not been clearly defined. Currently, when dealing with 1 to 10 BMs for treatment qualification, BM volume and lesion location seem to be more important than the number alone [5,48,49].
On the other hand, the BC phenotype or genotype is a main predictive factor of systemic therapy efficacy and ECM control. However, studies assessing ECM status as a prognostic factor used various definitions. Typically, it was the presence of distant metastases outside the brain or their absence [22,26,28,30,34,35]. Others analyzed the ECM site (e.g., bone metastases vs. metastases to parenchymal organs or metastases in the liver vs. metastases in other sites) as well as the number of organs outside the brain with metastases [6,33]. ECM was also analyzed in terms of its activity and classified as controlled or uncontrolled [23,25,32]. Furthermore, the disease progression was defined as the appearance of new lesions or the progression of previous lesions found in computed tomography 3 months before radiotherapy or if the examination was not performed, one month after irradiation. Still, uncontrolled ECM in BC patients was associated with a worse prognosis of patients and a shorter overall survival [22,23,[25][26][27][28]30,[32][33][34][35].
Advanced TNBC and HER2-positive BC have a higher risk of BM [2,3]. It is a natural course of advanced disease. On the other hand, an effective anti-HER2 therapy allows one to extend the time until symptomatic BM. An exploratory analysis from a phase III trial CLEOPATRA suggests that pertuzumab, trastuzumab, and docetaxel in a one line setting in HER2-positive advanced breast cancer delays the onset of CNS [50]. Similarly, in the NEfERT-T study, neratinib-paclitaxel delayed the onset and reduced the frequency of central nervous system progression [51]. Moreover, an effective anti-HER2 treatment after BM improved the OS, and new drugs that penetrate the blood-brain barrier were active in the brain and in the ECM [52]. According to the literature, the median OS in BC patients with BM is the longest in patients with HER2-positive treated with anti-HER2 regiments, then with luminal HER-2-negative and the shortest in TNBC [53,54]. Other prognostic factors included in prognostic tools, e.g., the time of diagnosis to the occurrence of BM, or disturbances in laboratory parameters (lymphopenia, increase in lactate dehydrogenase) seem to be derivative of the disease control [17,19,20,27].
Predicting the survival of BC patients with BM is difficult; therefore, prognostic tools are crucial in stratifying different patients' outcomes. However, they are limited by their retrospective nature and may underestimate survival in the modern era with the growing number of effective systemic agents. Furthermore, the prognostic tool developed for single institution cohorts might be biased by institutional practice patterns; therefore, external validation for new prognostic tools are the gold standard and should be obtained whenever possible [55,56]. Moreover, this validation should be performed in a cohort of patients with similar characteristics and demographics, and treated during a similar time period. The diversity of populations between cohorts may explain discrepancies in results and may reveal that they are not as predictive for the OS as the original one and may even underestimate the real survival [57].
International multidisciplinary recommendations, EANO-ESMO and NCCN, do not recommend prognostic tools in BC patients with BM [5,49]. The eligibility criteria in clinical trials with new systemic therapies in these patients are based mainly on the performance status, stabilization after local treatment (surgery and/or WBRT), the need for immediate local therapy (in patients with untreated BM) and the need for a daily dose of corticosteroids to control symptoms of BM. Furthermore, patients with leptomeningeal disease are usually excluded from clinical studies dedicated to BM (www.clinicaltrials.gov; accessed on 25 January 2022).
In clinical practice, the decision on the treatment sequence of BC patients with newly diagnosed BM should be made by a multidisciplinary team including a medical oncologist, radiotherapist and neurosurgeon, according to the accepted standard operating procedures (SOP). In the management algorithm, it is crucial to provide supportive care to patients with a poor prognosis and an expected survival of less than 3 months; for example, patients with uncontrolled ECM and/or more than 10 BM where systemic pharmacotherapy and WBRT are mainly used. In turn, for the minority of patients, with small and few lesions (1 to 10) depending on the volume (<15 mL), who may experience long-term survival or even cure, several approaches are used in combination (surgery followed by SRS/SRT or systemic pharmacotherapy, SRS/SRT or systemic pharmacotherapy) [5,49].
The universal, ideal prognostic tool should be simple and easily usable. Electronic access to such indices improves their usefulness in clinical practice, e.g., for the modified updated Breast GPA index (a free online calculator available at brainmetgpa.com). Furthermore, new prognostic tools in BC patients with BM should be used more in clinical trials.

Conclusions
Prognostic tools have variable precision in determining the survival of BC patients with BM. Progress in local and systemic treatment significantly affects the parameters of survival. Hence, it is necessary to update the prognostic indices used, depending on the period of treatment. Free access to prognostic tools online may increase the frequency of their use in clinical practice. Clinical trials in BC patients with BM remain a broad field for the application of prognostic tools.