Re‐irradiation in patients with progressive or recurrent brain metastases from extracranial solid tumors: A novel prognostic index

Abstract Background Most studies evaluating factors associated with the survival of patients with brain metastases (BM) have focused on patients with newly diagnosed BM. This study aimed to identify prognostic factors associated with survival after brain re‐irradiation in order to develop a new prognostic index. Methods This 5‐year retrospective study included patients treated with repeat‐radiotherapy for recurrent BM at the “Instituto Nacional de Cancerología” of Mexico between 2015 and 2019. Significant variables in the multivariate Cox regression analysis were used to create the brain re‐irradiation index (BRI). Survival and group comparisons were performed using the Kaplan–Meier method and the log‐rank test. Results Fifty‐seven patients receiving brain re‐irradiation were identified. Most patients were women (75.4%) with a mean age at BM diagnosis of 51.4 years. Lung and breast cancer were the most prevalent neoplasms (43.9% each). Independent prognostic factors for shorter survival after re‐irradiation were: Age >50 years (hazard ratio [HR]:2.5 [95% confidence interval [CI], 1.1–5.8]; p = 0.026), uncontrolled primary tumor (HR:5.5 [95% CI, 2.2–13.5]; p < 0.001), lesion size >20 mm (4.6 [95% CI, 1.7–12.2]; p = 0.002), and an interval <12 months between radiation treatments (HR:4.3 [95% CI, 1.7–10.6]; p = 0.001). Median survival (MS) after re‐irradiation was 14.6 months (95% CI, 8.2–20.9).MS of patients stratified according to the BRI score was 17.38, 10.34, and 2.82 months, with significant differences between all groups. Conclusions The new BRI can be easily implemented for the prognostic classification of cancer patients with progressive or recurrent BM from extracranial solid tumors.


| INTRODUCTION
Brain metastases (BM) are one of the most detrimental factors impacting the clinical course of cancer patients, with many of them presenting neurologic deficits that negatively affect their overall performance and quality of life. 1 It is widely accepted that lung cancer, breast cancer, melanoma, and renal cell carcinoma have the highest incidence of BM. However, available estimates vary substantially across studies, ranging from 8.5% to over 50%. [2][3][4] Advances in imaging techniques and systemic therapies have led to a higher number of early-stage diagnoses, better progression-free survival, and overall survival (OS), but also to a higher lifetime risk of being diagnosed with BM. 5 The increased incidence of BM observed over the past 10 years is consistent with the low brain accumulation and activity of most anticancer drugs. 6 Thus, surgical resection, wholebrain radiation therapy (WBRT), and stereotactic radiosurgery (SRS), often used sequentially or in combination, have remained the preferred treatments for patients with symptomatic BM. 7 In recent years, evidence has accumulated supporting the upfront use of systemic therapies with proven activity in the central nervous system (CNS), such as newer-generation tyrosine kinase inhibitors and immunecheckpoint inhibitors, in molecularly selected patients. 8 In addition to the type of local and systemic therapies received, survival in patients with BM has been shown to be associated with other clinical factors including age, performance status, primary tumor control and histology, extracranial disease burden, as well as the number, size, volume, and location of the metastatic brain lesions. [9][10][11][12][13] Some of these factors have been included in composite prognostic indexes such as the Disease-Specific Graded Prognostic Assessment (DS-GPA) scale, 14 the Recursive Partitioning Analysis (RPA) score, 15 the Score Index for Radio-surgery (SIR), 16 which can be used in clinical practice to guide treatment decisions as well as in clinical research to assess patient eligibility and to perform stratified data analysis. 17 Although the efficacy of repeat radiotherapy, in terms of local control and survival, in patients with progressive or recurrent BM has been established, 18 there is little agreement regarding which factors should guide patient stratification and treatment decisions. 19 The primary objective of this study was to identify prognostic factors associated with the survival of cancer patients following a second course of radiotherapy for the treatment of progressive or recurrent BM and to develop a new prognostic index.

| Study design and patients
This single-center retrospective study aimed to identify prognostic factors associated with survival after brain re-irradiation therapy. The study included patients with brain metastases (BM) treated with a first course of local radiation therapy and who underwent a second course of radiation, due to central nervous system (CNS) recurrence or progression, at the "Instituto Nacional de Cancerología" of México (INCan) between 2015-2019. Recurrent/progressive brain metastases were defined as metastases that after initial therapy recurred/progressed anywhere in the brain (in both original and non-original sites). Additional inclusion criteria were age (≥18 years), histopathological confirmation of cancer (extracranial solid tumor), measurable CNS disease prior to re-irradiation, and availability of MRIs before and after treatment. Patients with cranial bone metastases or whose medical records failed to document basic clinical characteristics, treatment modality, or relevant follow-up data were eliminated. A cohort of 57 patients fulfilled these criteria and were included in the analysis.

| Study endpoints and assessments
The primary outcome was the median duration of survival after re-irradiation, defined as the time between the date of re-irradiation and the date of death (due to any cause) or last contact, right censoring data at a cutoff date for the analysis on June 31, 2020. Variables with potential prognostic value were extracted from electronic medical records at baseline, first radiation, and re-radiation. Local responses (evaluated by brain MRIs within 2 months after each course of radiation) were determined by an independent central imaging group according to Response Evaluation Criteria in Solid Tumors version 1.1 (RECIST v1.1). The objective response rate (ORR) was defined as the proportion of patients with complete or partial responses. Disease control rate (DCR) was defined as the proportion of patients with either complete response, partial response, or stable disease. BM-related symptoms (seizures, headaches, vertigo, nausea, as well as visual, somatic, motor, and speech deficits) were recorded at baseline and after each course of radiation, classifying symptom reduction as complete, partial, or unchanged.

K E Y W O R D S
prognostic index score, recurrent brain metastases; re-irradiation, repeat radiotherapy Recursive Partitioning Analysis (RPA) was performed as previously described, 15 with a class assigned to each patient for both the first and second course of brain radiation. Due to the inclusion of primary tumors without a diagnosis-specific Graded Prognostic Assessment (GPA) index, diagnosis-specific GPA indexes [20][21][22] were calculated at the same time points for only 52 patients.

| Statistical analysis
Data were summarized as means with standard deviations (SDs) or as medians with ranges for continuous variables, and as proportions for categorical variables. Two groups comparisons between continuous variables were performed using the Student's t-test. The Chi-square test and the Fisher's Exact Test were used for categorical data. In the very few instances where significance values differed between tests, the results from the Fisher's Exact Test were reported. Comparisons between longitudinally collected and dependent dichotomous data were performed using the McNemar's paired test, while the marginal homogeneity test was used for multi categorical dependent data. Median duration of survival after brain re-irradiation and survival curves were estimated using the Kaplan-Meier method. Differences in the estimated survival curves were assessed, overall and between group pairs, using the log-rank test. p-values <0.05 with a two-sided test were considered statistically significant. Univariate and multivariate Cox regression analyses were performed to identify prognostic factors significantly associated with time-to-event outcomes. Variables that after adjustment for potential confounders remained significant in the multivariate Cox regression analysis were used to construct a new prognostic index: The brain re-irradiation (BRI) score. For simplicity, a score of 1 was assigned to each of these variables and score classes without significantly different survival times were grouped together. Data were analyzed using the SPSS software, package version 26 (IBM Corp.).

| Population characteristics at first and second irradiation
The characteristics of patients at first radiation and re-irradiation are shown in Table 2. The median time between radiotherapy courses was 13.4 months (interquartile range [IQR], 7.9-19.7 months). The recurrence rate of BM after re-irradiation was 29% at 12 months, and 48% at 24 months (Figure 1). At the time of the first radiotherapy, 42 (73.7%) patients had an Eastern Cooperative Oncology Group performance status (ECOG-PS) between 0 and 1, and 15 (26.3%) between 2 and 3. In contrast, at re-irradiation 31 (54.4%) patients had an ECOG-PS of 0-1 and 26 (45.6%) an ECOG-PS of 2-3, a change that was statistically significant in a paired analysis (p = 0.013). The proportion of patients achieving primary tumor control increased from 38.6% (n = 22), at the first radiation, to 54.4% (n = 31) at re-irradiation (p = 0.049).
Regarding radiation techniques, while the percentage of patients receiving WBRT decreased from 87.6% to 56.2% between the first and the second course of radiotherapy, the percentage of patients receiving conformal radiotherapy increased from 3.6% to 35.1% (p < 0.001). Similarly, there was a statistically significant reduction in the radiation dose (p < 0.001) and in the number of fractions (p < 0.001), with hypo-fractionated schemes predominating at re-irradiation. In contrast, there were no significant differences between radiotherapy courses in terms of KPS, GPA, RPA, or extracranial metastatic involvement.

| Neurological symptoms and response
BM-related neurological deficits and responses after the first and second radiation are shown in Table 2. Whereas 18 patients reported being asymptomatic prior to the start of the first radiation treatment, 14 reported being asymptomatic prior to the beginning of the second cycle of brain radiation. The most frequent focal neurological deficits reported by patients reported a partial improvement. The only factor significantly associated with a favorable neurologic response was age at re-irradiation (Table 3). In contrast to patients with partial symptom improvement, patients with complete symptom resolution were more likely to be ≤50 years old (p = 0.006).

| Radiological response
Local tumor responses after the first and second course of radiation are shown in Table 2. After initial radiotherapy, four patients (7%) had a complete response, 25 (43.9%) a partial response. Stable disease was reported in 20 patients (35.1) and disease progression in eight patients (14%). After re-irradiation, 14 patients (29.2%) had a partial response and 26 patients (54.2%) had stable disease. There were no re-irradiated patients with a complete response.
Response rates according to primary tumor diagnosis are shown in Table 4. After the first radiation, complete responses were observed only among breast cancer patients (4 out of 25). Similarly, partial responses were more frequent among breast cancer patients than in patients with other malignancies, who were more likely to have stable or progressive disease (p = 0.049). The objective response rate for the entire population was higher after the first radiation than after re-irradiation (50.9% vs. 29.2%), with up to 72% of breast cancer patients achieving an objective response (p = 0.018). In contrast, there were no significant differences between malignancies with regards to the proportion of patients with specific responses after re-irradiation. Of note, the probability of achieving an objective response after re-irradiation was lower among breast cancer patients who had responded to the first course of radiation (p = 0.007). T A B L E 2 (Continued)

| Brain re-irradiation index (BRI)
Since each of the four variables that remained significant in the multivariate analysis had similar effect sizes, a score of 1 was assigned to age (>50 years), primary tumor status (uncontrolled), BM size (>20 mm), and time between radiotherapies (<12 months) to create a 3-tiered prognostic index (scores of 0-1, 2, and 3-4), with the highest score corresponding to the highest risk of death. Of 57 patients, 20 (35%) had a score of 0-1, 27 (47.4%) had a score of 2, and 10 (17.6%) had a score of 3-4. The median survival after re-irradiation of patients classified according to the BRI score were 17.38 month (95% CI, 16.2-18.4) for the score of 0-1, 10.34 months (95% CI, 4.0-16.6) for the score of 2, and 2.82 months (95% CI, 0.0-6.2) for the score 3-4, with significant survival differences across all groups (p = <0.001) and between pairs (Figure 2A). In contrast, there was no statistical difference in the duration of survival after re-irradiation according to primary diagnosis ( Figure 2B). When patients were classified according to the recursive partitioning analysis (RPA) score ( Figure 2C), significant survival differences were found across groups (p < 0.001). However, when pairwise comparisons were performed, only Classes 1 and 2 differed from Class 3 (p = 0.001), with no differences between Group 1 and group 2 (p = 0.202). Of note, at the time of re-irradiation only one patient with a KPS below 70 was assigned to the RPA Class 3 (Table 2). Similarly, no survival differences were found when patients were classified according to the DS-GPA, either in the pooled analysis ( Figure 2D), or when lung cancer ( Figure 2E) and breast cancer were analyzed separately ( Figure 2F), with p-values of 0.597, 0.302, and 0.854, respectively.
It has been reported that prognostic factors differ between patients with newly diagnosed BM and patients with recurrent BM. 9,25 However, a limited number of studies have evaluated

Cumulative recurrence of brain metastasis (%)
Time from brain radiation, mo whether prognostic indexes developed in patients with newly diagnosed BM can be used in patients with recurrent BM. 36,39,40 Thus, for comparison purposes, survival was analyzed using two commonly used indexes, the Diagnostic-Specific Graded Prognostic Assessment (DS-GPA) for melanoma, breast and lung cancer, [20][21][22] and the recursive partitioning analysis (RPA). 15 Of note, balanced group sizes could only be obtained with the new BRI score. Furthermore, in five patients with primary tumors for which there is currently no DS-GPA, prognosis was successfully assessed using the BRI score.
The DS-GPA index failed to classify patients into groups with significantly different risks of death after re-irradiation. This result may be explained by the fact that the DS-GPA was developed to estimate the survival of patients with newly diagnosed BM. Thus, excluded from the analysis were patients with recurrent BM or leptomeningeal metastases, which are the most frequently diagnosed metastases in long-term survivors. Another possible explanation is that the DS-GPA for each patient was calculated using the mutation status

T A B L E 3 (Continued)
results from diagnostic tumor biopsies, since updating this information at re-irradiation would have required obtaining repeat tissue biopsies. In light of the well-known limitations of tissue biopsies and the growing use of liquid biopsies for the detection of clinically actionable mutations, 41 it will be important to determine whether mutation status results from liquid biopsies can be integrated into existing prognostic indexes or used for the development of novel prognostic systems. It has been noted that the DS-GPA scale can be used to calculate the life expectancy of patients at any given time point after treatment initiation. 14,42 However, this approach might not be suitable for patients with recurrent BM who have different survival trajectories.
The results of this study stand in contrast with previous research showing that the presence neurological symptoms 30,33,35 and a larger number of BM 25,39 are independent predictors of reduced survival. However, they are consistent with other studies in which survival has not been found to be associated with neurological symptoms 34,36,43 or with the number of BM. 9,24,27 In contrast to other studies, 31,33-36 Kased et al., previously reported that KPS (≥70) was an independent prognostic factor of longer survival in newly diagnosed patients but not in patients treated with repeat SRS for recurrent BM. 9 In the current study, KPS was not found to be a significant predictor of survival after reirradiation in either the univariate or the multivariate analysis. A possible explanation for this might be the underrepresentation of patients with a low KPS, which could be the result of selection and information bias. For instance, patients with a low KPS are less likely to be treated with repeat radiotherapy due to death, loss-tofollow-up, or lack of consent. Similarly, patients with a low KPS score are more frequently offered best supportive care rather than repeat radiotherapy, which according to most clinical recommendations is reserved for patients with good prognosis. Finally, given the subjectivity of KPS scoring, the possibility of assessment bias cannot be excluded. This accords with a study by Caballero et al., in which the prognostic value of KPS could not be assessed because of a limited number of patients with low scores. 25 More recently, it has been noted that inclusion criteria based on KPS is too restrictive, even in studies of patients with poor prognosis. 42 Similarly, it has been noted that the selection of variables to be included in a multivariate model requires careful consideration since, for instance, the RPA and DS-GPA overlap and are not independent of age, KPS and extracranial disease status. 44 In this study, the survival analysis of patients stratified according to RPA scores did not yield meaningful results as it has in previous studies. 16,23,36 Importantly, whereas the population of this study included patients treated between 2015 and 2019, previous studies had been restricted to the analysis of patients treated up to the year 2011, 28 19 Since in this study a large proportion of patients were treated with repeat-SRS, the MS estimates herein reported are broadly consistent with other studies of repeat-SRS, in which MS ranges from 3.0 to 10.8 months. 19 Although the impact of targeted therapies was not evaluated, its contribution to improved MS or its possible association with the higher KPS scores observed cannot be excluded. The lack of association between KPS and survival precluded the possibility of performing an exhaustive evaluation of other indexes, which rely on the performance status of patients, 17 including the re-irradiation score for patients treated with repeat-WBRT proposed by Logie et al. in 2017. 36 Nevertheless, what is clear from our results is that even in a population where most patients had a good KPS (≥80), there are marked survival differences that are related to other factors, further highlighting the importance of developing composite indexes specific for patients with recurrent BM as well as the urgent need to incorporate them into routine clinical practice.

| CONCLUSIONS
The proposed BRI can be easily implemented in clinical practice for the prognostic classification of cancer patients with progressive or recurrent BM arising from extracranial solid tumors and could potentially be used to help guide therapeutic decisions after a first course of local radiotherapy. The BRI has the added advantage of using clinical variables that are less prone to subjective interpretation. Nonetheless, considering the relatively small sample size and the retrospective nature of this study, prospective and larger-scale studies are necessary to validate this novel prognostic index.  data interpretation and formal analysis, data visualization, original draft writing, draft review and editing. Federico Maldonado Magos: Project conceptualization, data curation and investigation, methodology, data interpretation and formal analysis, data visualization, project administration and supervision, original draft writing, draft review and editing. Jenny G Turcott: Project conceptualization, data curation and investigation, methodology, data interpretation and formal analysis, data visualization, project administration and supervision, original draft writing, draft review and editing. Juan-Manuel Hernandez-Martinez: data curation and investigation, methodology, data interpretation and formal analysis, data visualization, original draft writing, draft review and editing. Bernardo Cacho-Díaz: Cases collection and methodology, draft review and editing. Andrés F. Cardona: Cases collection and methodology, draft review and editing. Aida Mota-García: Cases collection and methodology, draft review and editing. Francisco Lozano Ruiz: Cases collection and methodology, draft review and editing. Maritza Ramos-Ramirez: Cases collection, draft review and editing. Oscar Arrieta: Project conceptualization, data curation, data interpretation and formal analysis, funding acquisition, methodology, project administration and supervision, and writing-review and editing. All authors read and approved the final manuscript.

FUNDING INFORMATION
This study did not receive any specific funding from agencies in the public, commercial, pharmaceutical, or from the for-non-profit sector.