Stereotactic Radiosurgery for Recursive Partitioning Analysis Class II / III Lung Cancer Patients with Brain Metastases in the Modern Treatment Era

Stereotactic radiosurgery (SRS) is considered the initial treatment for lung cancer patients with small-sized and limited number of brain metastases. The objective of this study was to assess clinical outcomes of SRS treatment using CyberKnife (CK) for recursive partitioning analysis (RPA) class II/III patients with one to three brain metastases from lung cancer and identify which patients in the high RPA class could benefit from SRS. A total of 48 lung cancer patients who received CK-based SRS for their metastatic brain lesions from 2010 to 2017 were retrospectively analyzed. Radiographic response was evaluated during follow-up period. Overall survival (OS) and intracranial progression-free survival (IPFS) were calculated and prognostic variables associated with OS and IPFS were evaluated. Median follow-up time was 6.6 months. Local control rates at 6 months and 1-year following SRS were 98% and 92%, respectively. The median OS of all patients was 8 months. One-year and 2-year OS rates were 40.8% and 20.9%, respectively. In multivariate analysis, uncontrolled primary disease (p = 0.008) and ECOG performance status of 2 or 3 (p = 0.001) were independent prognostic factors for inferior OS. These two factors were also significantly associated with inferior IPFS. In subgroup analysis according to RPA class, primary disease status was the only prognostic factor, showing statistically significant OS differences in both RPA class II and III (controlled vs. uncontrolled: 41.1 vs. 12.3 months in RPA class II, p = 0.031; 26.9 vs. 4.1 months in RPA class III, p = 0.011). Our results indicated that SRS could be an effective treatment option for RPA class II/III patients with brain metastases from lung cancer in the modern treatment era. SRS might be particularly considered for patients with controlled primary disease.


Introduction
Lung cancer has currently the highest incidence in the world.It is the leading cause of cancer related deaths [1].Brain metastases from lung cancer also have a high incidence rate of 20% [2].As this disease progresses, brain metastases occur more often.Among all cases of brain metastases, those from lung cancer account for 40-50% [3].Therefore, appropriate management of brain metastases is a very important issue in treatment of lung cancer patients.
In the last several decades, whole brain radiation therapy (WBRT) is a standard treatment for patients with multiple brain metastases.However, many clinical studies have shown that the quality of life of patients is deteriorated by the neurotoxicity caused by WBRT [3,4].Although median survival of patients with brain metastases treated with WBRT has been reported to be less than 6 months, the number of patients with long term survival as well as median survival of patients has increased recently due to the development of systemic therapies and choice of appropriate local treatment strategies, e.g., SRS alone, SRS plus WBRT and surgical resection.[5,6].Treatment for four or fewer oliogometastatic lesions has been gradually replaced by SRS or surgical resection, instead of WBRT which reduces neurotoxicity while not affect survival.[7,8].In addition, some studies have reported that SRS alone is an effective treatment for patients with five or more brain metastases [9][10][11].
When determining whether patients with brain metastases should receive radical local treatment, recursive partitioning analysis (RPA) classification has been widely used [5].Although it is a readily available prognostic tool to select candidates, it lacks other important prognostic factors such as histology of tumor, number of brain metastases, and molecular features of tumor.[3,12,13].To perform treatment for brain metastases based on individual assessment of prognostic factors, it is necessary to consider all significant prognostic factors for survival.Considering median survival of patients with brain metastases, it is desirable to try aggressive treatment strategy for RPA class I patients.However, the number of such patients is very limited because most patients belong to class II and III in clinical practice [14].Although some studies have reported SRS results in lung cancer patients with limited number of brain metastases, few studies have reported the treatment outcomes of SRS for patients with only RPA class II/III.Thus, the aim of this study was to analyze clinical outcomes of SRS using CyberKnife (Accuray Inc., Snnyvale, CA, USA) for RPA class II/III patients with one to three brain metastases from lung cancer and identify which patients in the high RPA class could benefit from SRS.

Treatment outcomes
LC rates at 6 months and 1-year following SRS were 98% and 92%, respectively.A representative case of good response after CK-based SRS for brain lesion is shown in Figure 1.RC rates at 6 months and 1-year following SRS were 88% and 78%, respectively.SRS site failure occurred in five patients while distant brain failure occurred in 12 patients during the follow-up period after SRS.
The median IPFS of all patients was 5.3 months.One-year and 2-year IPFS rates were 23.9% and 15.2%, respectively (Fig. 2B).Similar to the analysis for OS, primary disease status and ECOG performance status were significant prognostic factors for IPFS (Table 3).
We also analyzed OS according to primary disease status and the presence of extracranial metastases in each RPA class II and III group.In RPA class II group, median OS time of patients with controlled primary disease was significantly higher than that of patients with uncontrolled primary disease (41.1 vs. 12.3 months; p = 0.031; Fig. 3A).Patients without extracranial metastases had significantly higher median OS than those with extracranial metastases (29.1 vs. 8.5 months; p = 0.032; Fig. 3B).In the RPA class III group, only OS according to primary disease status showed statistically significant difference (controlled vs. uncontrolled: 26.9 vs. 4.1 months; p = 0.011; Fig. 4A).OS according to the presence of extracranial metastases also showed a large difference of 5.1 months (absent vs. present: 9.3 vs. 4.2 months; p = 0.072; Fig. 4B), although the statistical significance was marginal.
No symptoms or signs caused by radiation induced necrosis were observed during follow-up period after SRS.In addition, there was no case showing significant neurocognitive dysfunction.Other complications included seizure in two patients and IICP signs in three patients due to cerebral edema.

Discussion
Recently, with the advent of novel targeted or immunotherapeutic agents, rapid development of systemic therapy has improved survival and clinical outcomes of patients with metastatic lung cancer [15][16][17][18][19][20].Previously, SRS for brain metastases is known to be an effective alternative treatment modality for WBRT in patients with RPA class I predicted to have good prognosis [5,21].There are not so many clinical studies on patients with RPA class II / III who are predicted to have poor prognosis.Particularly RPA class III patients were excluded from most randomized studies because of their extremely poor prognosis [22,23].However, in the modern treatment era, when chemotherapeutic agents with acceptable toxicity and advanced supportive management are introduced, lung cancer patients with brain metastases not necessarily have lower survival rates due to old age or poor performance [24,25].
We have conducted CK-based SRS since 2010, which is the time period after application of targeted therapies such as epidermal growth factor receptor tyrosine kinase inhibitors to stage IV lung cancer in our institution.Our results showed that, with the help of advanced systemic therapies, survival benefit can be achieved through aggressive local treatment for brain lesions if primary disease is adequately controlled.
Since the publication of RPA classification system by Gaspar et al., it has been questionable whether this could be generally applied to determine the treatment strategies of patients with brain metastases [5].A few studies have been conducted to verify this system [14,26,27].Nieder et al. have confirmed that patients with RPA class I have no disagreement with aggressive local treatment such as SRS [14].
However, considering time to non-CNS death, primary disease controlled subgroups of RPA class II patients suggest that aggressive local control of brain metastases may provide survival benefit [14].
Yamamoto et al. have divided RPA class II into three subclasses by scoring four prognostic factors (KPS, tumor numbers, primary disease status, and non-brain metastases) and reported that prognostic factors that can determine RPA class II are diverse and very heterogeneous, showing significant differences among subclasses for OS (p < 0.001 for all subclasses) [27].In the above study, they included tumor numbers that could be associated with target volume rather than age, one of the prognostic factors that determine original RPA class, suggesting the importance of local control in the era of advanced systemic therapies [27].
One retrospective study has shown that RPA class III patients with brain metastases have a reasonable median OS of 7.2 months when they are treated with SRS alone [25].That study also argues that poor performance patients may be ideal candidates for SRS because low incidence of distant CNS failure during life expectancy and single faction treatment can be helpful for patients with debility and their caregivers [25].We also believe that the convenience of short fractionation is a clear benefit for poor performance patients and SRS is a more effective treatment than WBRT in terms of the possibility of combination with chemotherapy or in the absence of delaying systemic therapy.
According to our study results, significant prognostic factors for both OS and IPFS in multivariate analysis were primary disease status and ECOG performance status.Primary disease status has been previously identified as the strongest prognostic factor associated with survival and intracranial progression in several studies [12,28,29].This plays a very important role in clinical decision-making, in predicting long term outcome after treatment of patients with brain metastases, and when using treatment strategies such as upfront SRS.The ECOG scale is the simplest and most commonly used performance status scale in clinical practice.It was used in our study.Our study only included patients with RPA class II and III.The factor that determines these two classes is only Karnofsky performance status (KPS).When assessing the clinical performance status of patients, patient's ability to perform certain activities of daily living without help of others is regarded as up to 1 according to the ECOG scale and up to 70 according to KPS.We excluded RPA class from the analysis because groups categorized by ECOG (0-1 vs 2) were perfectly matched with groups categorized by RPA class (2 vs 3).This shows that the RPA class also serves as a statistically significant prognostic factor for survival.The presence of extracranial metastases is known to be a significant poor prognostic factor for survival in several clinical studies, including factors that should be considered in diagnosis-specific graded prognostic assessment (DS-GPA) by stratifying based on primary cancer histology as well as RPA classification for brain metastases [5,13,21].The presence of extracranial metastases in our study was a significant prognostic factor in univariate analysis of OS and IPFS.However, it did not show any statistically significant difference in multivariate analysis.Thus, primary disease status, performance status, and extracranial metastases are important factors that have a significant effect on OS and IPFS.
In each group of RPA class II and III, we further analyzed the OS for these two factors (primary disease status, extracranial metastases) showing significantly survival difference in univariate analysis except for age among the three factors determining RPA class II.As a result, controlled primary disease was associated with significantly superior OS in both RPA class II and III.The absence of extracranial metastases was also associated with significantly superior OS in RPA class II.Particularly, RPA class III patients with poor performance had a median survival time longer than 2 years if the primary disease was controlled.These results clearly suggest that some RPA class II/III patients can have good prognosis depending on the control of primary tumor and extracranial metastatic lesions in the modern treatment era.
When we identified adverse effects, there were no patients with significant neurocognitive dysfunction or radionecrosis.Only 10% of patients suffered from neurologic toxicities after SRS.Side effects might have been underestimated due to evaluations without a quantitative neurocognitive test.Individual setting of dose schedules at clinician's discretion might have also affected the severity of side effects.
We acknowledge that this retrospective study has several limitations including selection bias and confounding factors.In particular, patients who previously had WBRT were included.Heterogeneous lung cancer group was formed by including both SCLC and NSCLC patients.In addition, the relatively small sample size limits the statistical power.Thus, further studies with a large number of patients are needed to validate our findings.In particular, a well-designed prospective study will help determine which patients will need active local treatment or just supportive care.

Patient selection
This retrospective study was approved by Institutional Review Board (IRB) of Gyengsang National University Hospital (IRB number: 2018-05-015).
A total of 106 patients with brain metastases were treated with SRS using CyberKnife (CK) at Gyengsang National University Hospital between February 2010 and May 2017.Among them, patients with primary lung cancer who had completed planned schedule of SRS were included while patients with RPA class I or loss of follow-up were excluded.Finally, 48 lung cancer patients eligible for RPA class II / III were retrospectively analyzed.SRS was performed in patients with less than three metastatic brain lesions.The maximum diameter of each lesion was less than 5 cm according to our institutional protocol.

SRS
All patients underwent SRS using CK.During treatments, all patients were immobilized with a thermoplastic head mask in supine position.Contrast-enhanced computed tomography (CT) and contrast-enhanced magnetic resonance imaging (MRI) for brain were taken with slice thickness of 1.0 mm and 2.5 mm, respectively.CT and MRI images were then rigidly fused with respect to each other using CK planning system (Multiplan ver.3.5.4).Gross tumor volume (GTV) was defined as enhanced lesion observed by any of both images.First, GTV was delineated on a CT image.It was then modified to include clearly contrast-enhancing regions seen in MRI images.Clinical target volume (CTV) was set to the same as GTV, assuming no microscopic invasion outside the GTV.Planning target volume (PTV) was generated by adding 1-mm margin from the GTV.Organs at risk (OAR) including lenses, optic nerves, optic chasm, brainstem, and spinal cord were also contoured.If two lesions were close 1 cm less, they were planned as one target.For this reason, although the number of brain metastatic in all patients treated was 70, the actual number of planning targets was 63.Inverse treatment planning was performed for all patients using Multiplan version 3.5.4(Accuray Inc., Snnyvale, CA, USA).SRS treatment plans were basically designed so that the entire GTV and at least 95% PTV were covered by the prescription dose surface.The prescription dose was normalized median 80% isodose line (range, 75% to 85%) relative to the maximal dose.The prescription dose was basically determined based on volume-dependent dose regimen suggested in the Radiation Therapy Oncology Group (RTOG) 95-05 trial [5].It was partly revised according to the judgment of clinicians considering the tumor size and location, the timing of SRS, and prior RT dose.Details in the prescription dose actually delivered are summarized in the Table 4 along with dose schedules and biologic equivalence dose for α/β=10 (BED 10 ).

Response and outcome assessment
After SRS, follow-up MRI (or CT if ineligible for MRI) was performed every 2 to 3 months, or when clinically indicated.Tumor response was evaluated based on Response Evaluation Criteria in Solid Tumors (RECIST) criteria (version 1.1) [30].In the case of multiple lesions, the sum of the two largest lesions was used to evaluate the response.Complete response, partial response, and stable disease were classified as local control (LC) while progressive disease was classified as local failure.
Overall survival (OS), intracranial progression-free survival (IPFS), LC, and regional control (RC) were evaluated.OS was defined as the length of time from the date of diagnosis of brain metastases to death or the last follow-up.IPFS was defined as the length of time from the date of diagnosis of brain metastases to intracranial progression or death from any cause or the last follow-up.However, in the case of salvage SRS, the survival time was calculated from the time of diagnosis of recurrent brain metastases in MRI. was defined as freedom from development of new lesions within the field treated with SRS or progression in preexisting metastases.RC was defined as freedom from development of new distant brain metastases.
Adverse effects such as increased intracranial pressure (IICP) signs, neurocognitive defect, and radiation necrosis were also evaluated.Quantitative evaluation such as questionnaire on neurocognitive function was not made.However, a clear clinical record of patient's symptoms was obtained through a simple clinical interview with the clinician.

Statistical analysis
All statistical analyses were performed using SPSS software Version 21.0 for Windows.Actuarial OS, IPFS, LC rates, and RC rates were calculated using the Kaplan-Meier method.Log-rank test was used for univariate analysis to assess prognostic factors associated with OS and IPFS.Cox proportional hazard models were performed for multivariate analysis.A two-sided p-value < 0.05 was considered statistically significant.

Conclusions
In conclusion, our results showed that SRS could be a useful treatment modality for RPA class II/III patients with one to three brain metastases from lung cancer in the modern treatment era.Our results suggest that patients with older age or poor performance should not be unconditionally excluded from aggressive treatment for brain metastases and SRS might be considered as an initial treatment for RPA class II/III patients with well-controlled primary disease.

Figure 1 .
Figure 1.An example of the magnetic resonance imaging (MRI) response after CyberKnife (Accuray

Figure 2 .
Figure 2. Overall survival (A) and intracranial progression free survival (B) of recursive partitioning

Figure 3 .
Figure 3. Overall survival difference according to primary disease status (A) extracranial

Figure 4 .
Figure 4. Overall survival difference according to primary disease status and (A) extracranial

Table 1 .
Patient and tumor characteristics

Table 2 .
Results of univariate analysis of overall survival and brain failure free survival

Table 3 .
Results of multivariate analysis of overall survival and brain failure free survival OS: overall survival; BFFS: brain failure free survival; ECOG: Eastern Cooperative Oncology group; NSCLC: Non-small cell lung cancer; SCLC: Small cell lung cancer.