Hypofractionated and single-fraction radiosurgery for brain metastases with sex as a key predictor of overall survival

Overall survival (OS) of patients with brain metastases treated with hypofractionated (HFSRT) or single-fraction (SRS) radiosurgery depends on several prognostic factors. The aim of this study was to investigate the potential of sex as an independent predictor of OS and evaluate the predictive accuracy of common prognostic scores. Retrospective analysis of 281 consecutive patients receiving radiosurgery of brain metastases was performed. Kaplan–Meier survival curves and Cox proportional hazards models were used to compare OS between SRS and HFSRT and by sex, before and after propensity-score matching (PSM) on key baseline prognostic covariates. Prognostic scores were evaluated using Harrell’s concordance index. Median OS was 11 months after both SRS and HFSRT. After PSM, median OS was 12 months after SRS (95% CI: 7.5–16.5) and 9 months after HFSRT (95% CI: 5.0–13.0; p = 0.77). Independent prognostic factors were sex, primary tumor, KPI, and systemic disease status. Median OS was 16 months for women and 7 months for male patients (p < 0.001). After excluding sex specific tumors, PSM revealed a median OS of 16 months for women and 8 months for male patients (p < 0.01). Evaluation of prognostic indices showed BSBM to be the most accurate (Harrell’s C = 0.68), followed by SIR (0.61), GPA (0.60), RPA (0.58), and Rades et al. (0.57). OS after HFSRT and SRS did not differ, although PSM revealed a non-significant advantage for SRS. Female sex was found to be a major independent positive prognostic factor for survival, and thus should be considered in the personalized decision-making of brain metastases treatment.

Statistical analysis. Propensity scores were calculated using multivariable logistic regression with the covariates age, sex, primary tumor, number of CNS metastases, PTV volume, KPI, presence of extracerebral metastases, and systemic disease status to determine the conditional probability of the patient receiving SRS vs. HFSRT, using 1:1 matching without replacement. A Kaplan-Meier analysis was conducted to estimate differences in survival by mode for all fully documented cases in the cohort (complete case analysis, n = 274) and among the matched pairs population (n = 168). Univariate Cox proportional hazards models were performed for OS for all presumed prognostic factors. Multivariable Cox proportional hazard models were performed adjusted for prognostic factors determined by backwards stepwise selection.
To assess gender related differences in survival, Kaplan-Meier curves were fitted and the log-rank test performed on all patients (n = 274), as well as among all patients excluding those with breast cancer as a primary tumor type (n = 240). Propensity-score matching for sex was also performed considering age, primary tumor, number of CNS metastases, PTV volume, KPI, presence of extracerebral metastases, and systemic disease status using 1:1 matching without replacement and excluding patients with breast cancer as the primary tumor type. A Kaplan-Meier survival curve and log-rank test was conducted on the matched pair sample (n = 164).
The prognostic indices (PI) RPA (Recursive Partitioning Analysis) 5 , GPA (Graded Prognostic Assessment) 6 , SIR (Score Index For Radiosurgery) 7 , BSBM (Basic Score for Brain Metastases) 8 , and Rades et al. 9 were calculated (see supplementary Table 1) and their predictive accuracy of survival was evaluated using Harrell's concordance index (Harell's C). Data (all available cases, n = 265) were split randomly into training and test subsets in Stata/ MP 15.1. Separate Cox models were run for each prognostic score using the training dataset with Harrell's C estimates for the test subset. Harrell's C can take a value between 0 and 1; a value of 1 corresponds to perfect concordance whilst a value of 0.5 indicates completely random concordance. Pairwise comparisons of Harrell's C were conducted using the RPA index as the reference.
Ethics approval. This study has been approved by the research ethics committee of the Medical University Innsbruck (Reference Number: EK 1160/2018).
Smoking status was available for 251 (92%) of 274 patients (Table 1). After propensity score matching, the distribution presented as follows: 35.6% of male and 32.8% of female patients were active smokers, 22

Time to CNS metastasis (months)
Mean (SD) 29 Comparison of prognostic scores among patients with brain metastases. For all prognostic scores, Harrell's Cs were significantly larger than 0.5, although none surpass a value of 0.7 and all were accompanied by wide confidence intervals (Fig. 4). Pairwise comparisons indicate that BSBM predicts survival better than the most commonly used RPA (Table 2); training dataset models are available in Supplementary Table 4.   www.nature.com/scientificreports/
As frequently reported in literature, patient cohorts receiving HFSRT and SRS differ substantially in number and size of metastases, location and histology, as well as in other prognostic factors. Conversely to SRS, HFSRT is often favored in patients with large metastases or lesions located in eloquent brain areas 19,25 . Therefore, comparative analysis suffers from differing dose concepts and inhomogeneous patient characteristics because of unequal allocation to the two treatment modalities. In order to minimize this effect we performed a propensity score matching to pair each SRS patient with a best matching HFSRT patient, including all relevant prognostic factors for pairing. As a result, homogeneity in the distribution of prognostic factors between patients treated with SRS and HFSRT improved significantly, thereby enabling unbiased direct comparison. Survival analysis after   www.nature.com/scientificreports/ PSM showed a numerical advantage for SRS over HFSRT (12 vs. 9 months), which does not, however, translate to significance (p = 0.77). The median OS observed for SRS after PSM was 3 months longer than for HFSRT, but confidence intervals of both cohorts were strongly overlapping (HFSRT: 5.0-13.0 months, SRS: 7.5-16.5 months). Considering also the sizable number of 84 matching pairs, the absence of statistical significance regarding the numerical survival advantage of SRS reflects the clinical evidence of HFSRT being on par with SRS 22,26 . Prognostic factors and scores. In the era of personalized medicine, individualized estimation of survival time becomes more and more important, since therapy decision hinges on survival expectation. In order to detect patients who are either at risk for early death or who have a chance for long-term survival, inclusion and specific weighting of prognostic factors are decisive. Thus, independent factors have been identified in the past, including age, KPI, primary tumor, presence of extracranial metastases, number and size of intracranial metastases, and systemic disease status. These prognostic factors have been condensed into various prognostic scores of differing clinical applicability, reliability and prognostic value. The major weakness of all established scores to predict OS are the facts that they do not deliver individualized survival probabilities and that they tend to suffer from disproportionally sized prognostic groups 27 . Collaterally, and despite continuous validation of prognostic scores in use for decades, ongoing improvement in cancer therapy may lead to outdated survival estimations.
Our results confirm higher age, male sex, lower KPI, primary tumor type other than lung adenocarcinoma or breast cancer, existence of extracerebral metastases, and progressive systemic disease status as to represent hazardous factors for survival. However, multivariate analysis of our data revealed that only sex, primary tumor, KPI, and systemic disease status are independently influencing OS. Male sex and progressive systemic disease status revealed to equally contribute to the hazard of poor survival (HR: 1.5 vs. 1.6), both identified as the strongest predictors of unfavorable outcome in our cohort of 274 patients. Gender-related differences in OS of patients with brain metastases occasionally have been reported in the past also for large patient cohorts [27][28][29] . However, there exist also competing results derived from multicenter studies, which did not identify gender related differences in OS. For instance, Sperduto et al. established the prognostic GPA score by analyzing five cohorts from randomized clinical trials (1,960 patients in total) in 2011. A significant gender effect on patient outcome was not observed. Consequently, sex has not been considered in the meanwhile widely used GPA score or in other prognostic indices. However, the GPA and similar scores 5,9 are based on patient cohorts of which only a minority was treated by radiosurgery. In addition, median overall survival of 75% of the investigated patients by Sperduto et al. did not exceed 3.8 months 6 . By contrast, in our study including only patients treated by radiosurgery, median survival of the entire cohort was 11 months, and we clearly identified gender as an independent prognostic factor for our cohort. A potential gender related difference might only become evident in patients of better prognosis and increased overall survival. This is supported by the fact that female survival advantages were primarily observed in newer patient cohorts benefitting from increased utilization of targeted therapies and radiosurgery.
In order to prove the independency and validity of sex as an underestimated factor of OS prognosis, exclusion of patients afflicted by sex specific primaries and balancing the distribution of prognostic factors among male and female patients by propensity matching were performed in statistical evaluation. Even after unbiased comparison, median OS of female patients (16 months) was still twice the median OS of male patients (8 months). We therefore conclude that a favorable distribution of other decisive hazard parameters, such as the smoking status, among the female cohort can, if at all, only partially explain the pronounced survival advantage. Thus, it is very likely that sex itself or other unknown gender related characteristics of female patients might be responsible for the observed effect. Female survival advantages have been reported for a range of oncologic diseases. Sex differences have been observed in the genetic and molecular basis of cancer, and the efficacy and toxicity of anticancer therapy 30 . Especially in the treatment of lung cancer, the main origin of brain metastases, several molecular markers such as EGFR, ALK, or PD1 can be exploited for targeted therapy approaches. An increased benefit of women has been observed for EGFR inhibitors 31 . Conversely, men profit from treatment with immune checkpoint inhibitors 31,32 . The contribution of such medication on the observed gender differences in the outcome of brain metastases treatment by stereotactic radiosurgery however has still to be investigated. Irrespective of benefits for female patients that derive from pharmacogenomic differences between the sexes, the observed survival advantage of female patients with brain metastases in our cohort is valid for both applied treatment options, HSFRT and SRS.
Validation of common prognostic scores applied to our patient cohort (Fig. 4) evidenced BSBM as to best perform regarding subgroup discrimination and OS prediction. After further evaluation of the observed gender effect, sex could be included as a parameter in existing scores such as the BSBM, thereby potentially increasing their prediction accuracy. Our work reinforces the necessity to continuously adjust and reevaluate prognostic scores in use in clinical practice, especially if applied to patients with brain metastases treated with rapidly evolving radiotherapy methodologies and systemic therapies.

Conclusion
After long-term follow-up, no difference in overall survival between single-fraction and hypofractionated radiosurgery of brain metastases could be observed in our cohort of 274 consecutive patients. Female sex turned out to be a major independent prognostic factor for overall survival, only in part explainable by a more favorable distribution of other hazard parameters, if compared to the male cohort. Population-based studies or metaanalyses should be conducted to validate our findings and to encourage discussion for the inclusion of sex and gender characteristics in the decision-making for the personalized treatment of patients with brain metastases.