https://orcid.org/0000-0002-9589-9906
Abstract
The long-term risk of necrosis after radiosurgery for brain metastases is uncertain. We aimed to investigate incidence and predictors of radiation necrosis for individuals with more than 1 year of survival after radiosurgery for brain metastases.
Patients who had a diagnosis of brain metastases treated between December 2006 and December 2014, who had at least 1 year of survival after first radiosurgery were retrospectively reviewed. Survival was analyzed using the Kaplan-Meier estimator, and the incidence of radiation necrosis was estimated with death or surgical resection as competing risks. Patient and treatment factors associated with radiation necrosis were also analyzed.
A total of 198 patients with 732 lesions were analyzed. Thirty-four lesions required salvage radiosurgery and 10 required salvage surgical resection. Median follow-up was 24 months. The estimated median survival for this population was 25.4 months. The estimated per-lesion incidence of radiation necrosis at 4 years was 6.8%. Medical or surgical therapy was required for 60% of necrosis events. Tumor volume and male sex were significant factors associated with radiation necrosis. The per-lesions incidence of necrosis for patients undergoing repeat radiosurgery was 33.3% at 4 years.
In this large series of patients undergoing radiosurgery for brain metastases, patients continued to be at risk for radiation necrosis throughout their first 4 years of survival. Repeat radiosurgery of recurrent lesions greatly exacerbates the risk of radiation necrosis, whereas treatment of larger target volumes increases the risk modestly.
Brain metastases are a relatively common occurrence in cancer patients, with incidence estimates ranging as high as 17%. The historical management of these lesions included steroids, whole-brain radiotherapy (WBRT), and resection, the latter of which was shown to provide a survival benefit in patients with a single metastasis.1 Subsequently, stereotactic radiosurgery (SRS) has established a long track record as a safe and effective treatment in a variety of clinical scenarios, including when used in conjunction with surgery,2,3 whole-brain radiation,4 or alone.5–8 Critically, evidence suggests that using SRS in lieu of WBRT does not result in a survival detriment and reduces toxicity.9 Although the early experience with radiosurgery focused on patients with a small number of metastases (fewer than 4 metastases), recent data have suggested that patients with an intermediate disease burden also do well with a radiosurgery-alone approach.9,10 These data have led to the increased adoption of SRS in the treatment of patients with brain metastases.
The longer survival of patients with brain metastases11 also exposes them to late toxicity from cranially directed therapy. Whereas neurocognitive decline is a dreaded late complication of WBRT, SRS generally preserves cognition.10,12 The most relevant late complication in patients after SRS is radiation necrosis. Radiation necrosis poses challenges on 2 fronts. First, it presents a diagnostic dilemma for the treating physician, who must decide whether an enlarging mass or increasing edema on surveillance imaging represents treatment effect or tumor progression. More onerously, radiation necrosis may be symptomatic because of a disruption of nearby cortical function and may be life-threatening. Whereas radiation necrosis can often be managed with conservative treatment, symptomatic necrosis may require surgical intervention.
Prior reports of radiation necrosis have focused on all-comers and include patients with short survival following radiosurgery. With the validation of the diagnosis-specific Graded Prognostic Assessment (ds-GPA) in patients with brain metastases, we can now predict a priori the median survival in different cohorts of patients; patients with a ds-GPA greater than 3 frequently have survivals exceeding 1 year.13 We hypothesize that in these long-term survivors, the incidence of delayed radiation necrosis is underestimated. The present study aims to study the incidence and risk factors for radiation necrosis in patients with prolonged survival after radiosurgery for brain metastases.
Methods
Patient Selection
Our prospective institution outcomes database for patients treated with Gamma Knife radiosurgery at William Beaumont Hospital treated between December 2006 and December 2014 was reviewed retrospectively for patients who survived at least 1 year after radiosurgery for brain metastases. All patients were treated on Leksell Gamma Knife 4C (Elekta) with our previously reported technique.14,15 The project was approved by the Beaumont Institutional Review Board.
Patient Data
Using the prospective institutional database and the patient’s electronic and paper chart for cross-referencing, data were gathered on patient age, sex, date of diagnosis, date of intracranial metastasis, and date and dose of any WBRT given. Gamma Knife radiosurgery parameters including the target volume, tumor location, prescription volume, prescription dose, prescription isodose line, and whether the target was a cavity or intact lesion were also collected from a prospectively collected data sheet. Data on concurrent systemic therapy (either overlapping or initiated within 90 days of radiosurgery) were retrospectively collected for the purpose of this study.
Radiation necrosis was recorded prospectively in our cohort of patients, and additional data regarding diagnosis and treatment were confirmed retrospectively. The diagnosis of necrosis was based on clinical and diagnostic imaging findings after excluding disease progression or from histopathological findings when surgical resection was performed. The method of diagnosis was classified into 5 categories. The first was based on standard contrast/fluid-attenuated inversion recovery MRI showing an enlarging lesion that subsequently responded to steroids with no further therapy. The second was MR perfusion added to the prior criteria. The third was spectroscopy added to MRI. The fourth added both perfusion and spectroscopy. The perfusion and spectroscopy criteria for confirmation of necrosis at our institution have been previously described16 and include measurements of relative cerebral blood volume, percentage of signal-intensity recovery, choline:creatine ratio, choline:N-acetylaspartate ratio, and relative ratios of metabolites compared to normal brain. The fifth diagnostic modality was surgical biopsy (confirming predominance of necrosis with no or insignificant viable cells in a lesion).
Toxicity grading was performed as follows: grade 1 was imaging changes alone, grade 2 was requiring any medical therapy (dexamethasone, vitamin E, pentoxifylline, bevacizumab, or hyperbaric oxygen), grade 3 was requiring surgery, grade 4 was requiring intensive care unit–level care for symptoms, and grade 5 was death. Of note, this grading was not purely dependent on symptoms because some therapies were initiated because of presence of imaging findings, and we felt that capturing any use of therapy for radiation necrosis better captured the burden of this event.
Statistical Methods
Survival of patients from the time of radiosurgery was estimated using the Kaplan-Meier estimator, censoring patients at the last date of follow-up (end date of study data collection was February 1, 2016). The incidence of radiation necrosis was estimated using the cumulative incidence competing risk method and performed on a per-lesion basis. Competing risks included patient death, salvage surgery, or salvage radiosurgery for recurrence (all patients in our practice who were well enough to undergo diagnostic MRI to confirm recurrence were offered salvage therapy). We also estimated the risk in retreated lesions, with time to necrosis measured from the second radiosurgery date.
Fine and Gray models were used to estimate relevant covariates predicting the risk of necrosis. The number of shots and prescription isodose line were colinear to the remaining radiosurgery parameters (dose, conformality index, target volume). The radiosurgery parameters used in the retreatment analysis were from the second treatment, and we also included time to retreatment as a continuous factor within our model. Analyses were performed in R using the survival and cmprsk packages with additional scripts in the CRR add-on package as described by Scrucca et al.17 The optimal tumor volume cut point for predicting necrosis was analyzed using SPSS with a receiver operating characteristic curve. A Cox proportional hazards model was used to estimate the relative hazard of necrosis greater than and less than the cut point.
Results
Patient and Lesion Characteristics
Our database yielded 198 patients with 732 metastatic lesions with 1 year of follow-up following first Gamma Knife radiosurgery treatment. The median per-patient follow-up time was 23.8 months and the median lesion follow-up was 20.6 months, with 694 (94.5%) lesions having follow-up of at least 12 months. Patient and tumor characteristics are reported in Table 1. The most common primary tumor sites were lung, breast, and melanoma. WBRT was given to 78 (39%) patients. For the 72 patients with dose information available, the median dose was 35 Gy. WBRT timing information was available for all but 5 patients. In these patients, 39 received WBRT prior to first radiosurgery, with a median time to radiosurgery of 10.3 months (range, 0.2-23.7 months) and 34 received WBRT after initial radiosurgery with a median time to WBRT of 11.7 months (range, 0.3-52.2 months). There were 68 resection cavities out of the 732 lesions.
Patient and Tumor Characteristics
| Characteristic | Per-Patient Statistic (Total = 198) | Lesion Characteristics | Per-Lesion Statistic (Total = 732) |
|---|---|---|---|
| Age at SRS (median; range), y | 61; 30-89 | Resection cavities | 68 (9%) |
| Sex (N, %) Male | 70 (35%) | Retreated lesions | 34 (5%) |
| Primary site (N, %) | Tumor location (N) | ||
| Lung | 122 (62%) | Frontal | 203 |
| Breast | 42 (21%) | Parietal | 109 |
| Melanoma | 8 (4%) | Occipital | 99 |
| Other | 26 (13%) | Temporal | 74 |
| WBRT | 78 (39%) | Cerebellum | 161 |
| Median WBRT dose, Gy | 35 | Brainstem | 12 |
| Median time, WBRT to first SRSa, mo | 10.7 | Parameningeal | 30 |
| Median time, first SRS to WBRTa, mo | 11.7 | Nucleib | 14 |
| Ventricular | 26 | ||
| Tract | 3 |
| Characteristic | Per-Patient Statistic (Total = 198) | Lesion Characteristics | Per-Lesion Statistic (Total = 732) |
|---|---|---|---|
| Age at SRS (median; range), y | 61; 30-89 | Resection cavities | 68 (9%) |
| Sex (N, %) Male | 70 (35%) | Retreated lesions | 34 (5%) |
| Primary site (N, %) | Tumor location (N) | ||
| Lung | 122 (62%) | Frontal | 203 |
| Breast | 42 (21%) | Parietal | 109 |
| Melanoma | 8 (4%) | Occipital | 99 |
| Other | 26 (13%) | Temporal | 74 |
| WBRT | 78 (39%) | Cerebellum | 161 |
| Median WBRT dose, Gy | 35 | Brainstem | 12 |
| Median time, WBRT to first SRSa, mo | 10.7 | Parameningeal | 30 |
| Median time, first SRS to WBRTa, mo | 11.7 | Nucleib | 14 |
| Ventricular | 26 | ||
| Tract | 3 |
Abbreviations: SRS, stereotactic radiosurgery; WBRT, whole-brain radiotherapy.
aPatients who received WBRT prior to or after SRS.
bThalamus/Basal Ganglia/Caudate.
Patient and Tumor Characteristics
| Characteristic | Per-Patient Statistic (Total = 198) | Lesion Characteristics | Per-Lesion Statistic (Total = 732) |
|---|---|---|---|
| Age at SRS (median; range), y | 61; 30-89 | Resection cavities | 68 (9%) |
| Sex (N, %) Male | 70 (35%) | Retreated lesions | 34 (5%) |
| Primary site (N, %) | Tumor location (N) | ||
| Lung | 122 (62%) | Frontal | 203 |
| Breast | 42 (21%) | Parietal | 109 |
| Melanoma | 8 (4%) | Occipital | 99 |
| Other | 26 (13%) | Temporal | 74 |
| WBRT | 78 (39%) | Cerebellum | 161 |
| Median WBRT dose, Gy | 35 | Brainstem | 12 |
| Median time, WBRT to first SRSa, mo | 10.7 | Parameningeal | 30 |
| Median time, first SRS to WBRTa, mo | 11.7 | Nucleib | 14 |
| Ventricular | 26 | ||
| Tract | 3 |
| Characteristic | Per-Patient Statistic (Total = 198) | Lesion Characteristics | Per-Lesion Statistic (Total = 732) |
|---|---|---|---|
| Age at SRS (median; range), y | 61; 30-89 | Resection cavities | 68 (9%) |
| Sex (N, %) Male | 70 (35%) | Retreated lesions | 34 (5%) |
| Primary site (N, %) | Tumor location (N) | ||
| Lung | 122 (62%) | Frontal | 203 |
| Breast | 42 (21%) | Parietal | 109 |
| Melanoma | 8 (4%) | Occipital | 99 |
| Other | 26 (13%) | Temporal | 74 |
| WBRT | 78 (39%) | Cerebellum | 161 |
| Median WBRT dose, Gy | 35 | Brainstem | 12 |
| Median time, WBRT to first SRSa, mo | 10.7 | Parameningeal | 30 |
| Median time, first SRS to WBRTa, mo | 11.7 | Nucleib | 14 |
| Ventricular | 26 | ||
| Tract | 3 |
Abbreviations: SRS, stereotactic radiosurgery; WBRT, whole-brain radiotherapy.
aPatients who received WBRT prior to or after SRS.
bThalamus/Basal Ganglia/Caudate.
Repeat radiosurgery as salvage was performed in 31 lesions. Median time to repeat radiosurgery was 12.6 months (range, 1.4-29.7 months).
Radiosurgery Treatment Parameters
Table 2 outlines the radiosurgery parameters for each lesion stratified by intact vs resection cavity and upfront vs retreatment. In general, retreated lesions tended to be larger, as did resection cavities. These lesions were also generally treated at a lower dose.
Radiosurgery Treatment Parameters
| Parameter | Upfront Treatment (N = 732) | Retreatment (N = 31) |
|---|---|---|
| Intact lesions (N = 698) | N = 664 | N = 31 |
| Dose, median; range, Gy | 20; 11.5-24 | 16; 11-22 |
| No. of shots, median; range | 1; 1-10 | 2; 1-13 |
| Conformality index, median; range | 3.87; 0.09-26.0 | 2.20; 1.18-13.5 |
| Prescription isodose line, median; range | 50; 40-60 | 50; 33-70 |
| Tumor volume, median; range, cc | 0.12; 0.01-16.10 | 2.00; 0.03-24.50 |
| Resection cavities (N = 34) | (N = 68) | |
| Dose, median; range, Gy | 16; 12.5-22 | |
| No. of shots, median; range | 2; 1-13 | |
| Conformality index, median; range | 2.00; 1.30-3.80 | |
| Prescription isodose line, median; range | 50; 50-50 | |
| Tumor volume, median; range, cc | 7.10; 2.50-39.10 |
| Parameter | Upfront Treatment (N = 732) | Retreatment (N = 31) |
|---|---|---|
| Intact lesions (N = 698) | N = 664 | N = 31 |
| Dose, median; range, Gy | 20; 11.5-24 | 16; 11-22 |
| No. of shots, median; range | 1; 1-10 | 2; 1-13 |
| Conformality index, median; range | 3.87; 0.09-26.0 | 2.20; 1.18-13.5 |
| Prescription isodose line, median; range | 50; 40-60 | 50; 33-70 |
| Tumor volume, median; range, cc | 0.12; 0.01-16.10 | 2.00; 0.03-24.50 |
| Resection cavities (N = 34) | (N = 68) | |
| Dose, median; range, Gy | 16; 12.5-22 | |
| No. of shots, median; range | 2; 1-13 | |
| Conformality index, median; range | 2.00; 1.30-3.80 | |
| Prescription isodose line, median; range | 50; 50-50 | |
| Tumor volume, median; range, cc | 7.10; 2.50-39.10 |
Radiosurgery Treatment Parameters
| Parameter | Upfront Treatment (N = 732) | Retreatment (N = 31) |
|---|---|---|
| Intact lesions (N = 698) | N = 664 | N = 31 |
| Dose, median; range, Gy | 20; 11.5-24 | 16; 11-22 |
| No. of shots, median; range | 1; 1-10 | 2; 1-13 |
| Conformality index, median; range | 3.87; 0.09-26.0 | 2.20; 1.18-13.5 |
| Prescription isodose line, median; range | 50; 40-60 | 50; 33-70 |
| Tumor volume, median; range, cc | 0.12; 0.01-16.10 | 2.00; 0.03-24.50 |
| Resection cavities (N = 34) | (N = 68) | |
| Dose, median; range, Gy | 16; 12.5-22 | |
| No. of shots, median; range | 2; 1-13 | |
| Conformality index, median; range | 2.00; 1.30-3.80 | |
| Prescription isodose line, median; range | 50; 50-50 | |
| Tumor volume, median; range, cc | 7.10; 2.50-39.10 |
| Parameter | Upfront Treatment (N = 732) | Retreatment (N = 31) |
|---|---|---|
| Intact lesions (N = 698) | N = 664 | N = 31 |
| Dose, median; range, Gy | 20; 11.5-24 | 16; 11-22 |
| No. of shots, median; range | 1; 1-10 | 2; 1-13 |
| Conformality index, median; range | 3.87; 0.09-26.0 | 2.20; 1.18-13.5 |
| Prescription isodose line, median; range | 50; 40-60 | 50; 33-70 |
| Tumor volume, median; range, cc | 0.12; 0.01-16.10 | 2.00; 0.03-24.50 |
| Resection cavities (N = 34) | (N = 68) | |
| Dose, median; range, Gy | 16; 12.5-22 | |
| No. of shots, median; range | 2; 1-13 | |
| Conformality index, median; range | 2.00; 1.30-3.80 | |
| Prescription isodose line, median; range | 50; 50-50 | |
| Tumor volume, median; range, cc | 7.10; 2.50-39.10 |
Concurrent systemic therapy was anti–epidermal growth factor receptor (EGFR) therapy in 9 patients (23 lesions) patients, anti–human epidermal growth factor receptor 2 (HER2) therapy in 10 patients (56 lesions), and conventional cytotoxic therapy in the remaining patients. Immunotherapy (ipilimumab) was given within 90 days to only 2 patients (5 lesions).
Radiation Necrosis
Of 732 lesions treated with initial SRS, 44 patients experienced radiation necrosis events, 413 were censored because of competing risk of patient death, 10 were censored because of competing risk of surgical salvage, 31 were censored because of competing risk of repeat radiosurgery, and 234 were censored because of end of study period follow-up. On a per-patient basis, 49 out of 198 patients experienced a radiation necrosis event in our population.
The diagnosis of radiation necrosis was made by MRI and clinical judgment in 5 patients, MRI and perfusion in 7 patients, MR and spectroscopy in 2 patients, MRI, perfusion and spectroscopy in 31 patients, and surgical resection in 10 lesions. Toxicity grade was grade 1 in 22 events, grade 2 in 23 events, and grade 3 in 10 events. There were no grade 4 or 5 events Table 3.
Radiation Necrosis Events (Per-Lesion Analysis)
| Diagnostic Modality | N = 55 Necrosis Events | ||
|---|---|---|---|
| n, % | Toxicity Grade | n, % | |
| MRI + clinical judgment | 7, 12.7% | Grade 1 (radiographic only) | 22, 40.0% |
| MRI perfusion/ spectroscopy | 31, 56.4% | Grade 2 (medical therapy required) | 23, 41.8% |
| MR perfusion | 7, 12.7% | Grade 3 (surgical therapy required) | 10, 18.2% |
| MR spectroscopy | 2, 3.6% | ||
| Surgical resection | 10, 18.2% |
| Diagnostic Modality | N = 55 Necrosis Events | ||
|---|---|---|---|
| n, % | Toxicity Grade | n, % | |
| MRI + clinical judgment | 7, 12.7% | Grade 1 (radiographic only) | 22, 40.0% |
| MRI perfusion/ spectroscopy | 31, 56.4% | Grade 2 (medical therapy required) | 23, 41.8% |
| MR perfusion | 7, 12.7% | Grade 3 (surgical therapy required) | 10, 18.2% |
| MR spectroscopy | 2, 3.6% | ||
| Surgical resection | 10, 18.2% |
Radiation Necrosis Events (Per-Lesion Analysis)
| Diagnostic Modality | N = 55 Necrosis Events | ||
|---|---|---|---|
| n, % | Toxicity Grade | n, % | |
| MRI + clinical judgment | 7, 12.7% | Grade 1 (radiographic only) | 22, 40.0% |
| MRI perfusion/ spectroscopy | 31, 56.4% | Grade 2 (medical therapy required) | 23, 41.8% |
| MR perfusion | 7, 12.7% | Grade 3 (surgical therapy required) | 10, 18.2% |
| MR spectroscopy | 2, 3.6% | ||
| Surgical resection | 10, 18.2% |
| Diagnostic Modality | N = 55 Necrosis Events | ||
|---|---|---|---|
| n, % | Toxicity Grade | n, % | |
| MRI + clinical judgment | 7, 12.7% | Grade 1 (radiographic only) | 22, 40.0% |
| MRI perfusion/ spectroscopy | 31, 56.4% | Grade 2 (medical therapy required) | 23, 41.8% |
| MR perfusion | 7, 12.7% | Grade 3 (surgical therapy required) | 10, 18.2% |
| MR spectroscopy | 2, 3.6% | ||
| Surgical resection | 10, 18.2% |
Survival Analysis
The Kaplan-Meier estimate for median overall survival was 25.2 months (95% CI: 22.9-28.4 months). The survival curve is plotted in Fig. 1A. The point estimates for 2-year, 3-year, 4-year, and 5-year survival were 52.8%, 33.4%, 22.1% and 16.6%. Fig. 1B shows that the cohort of patients with radiation necrosis survived longer than those without (P = .02).
Overall survival probability as a function of time. Time is measured from radiosurgery; however, graph is truncated because no deaths occurred before 12 months per study design. A, Survival for entire cohort. B, Survival per strata (patients having radiation necrosis or not). In our cohort, patients with radiation necrosis had better overall survival (P = .03 by log-rank test).
Cumulative Incidence of Radiation Necrosis
Estimates for the cumulative incidence of necrosis for the first 5 years were 2.9%, 4.9%, 6.2%, 6.8%, and 6.8% (Fig. 2).
Cumulative Incidence of Radiation Necrosis Incidence of radiation necrosis vs competing risks of death, salvage surgery, or salvage radiosurgery.
Of the 31 retreated lesions, 11 patients experienced radiation necrosis, 15 were censored because of competing risk of patient death, 2 were censored because of competing risk of surgical salvage, and 3 were censored because of end of study period follow-up. None of the lesions included for analysis had a second radiosurgery retreatment event within the follow-up of this study. Those receiving retreatment had estimated incidences of 6.5%, 16.5%, 31.6%, and 36.3% for the first 4 years after initial radiosurgery, respectively (the retreatment group had no 5-year survivors).
Predictors of Radiation Necrosis
To find predictors of radiation necrosis in the initial SRS cohort, 3 Fine and Gray Subdistribution Proportional Hazards models were run (Table 4): Model 1 was run with the prospective Gamma Knife parameters and model 2 with systemic therapy information added. The systemic therapies considered were anti-EGFR or anti-HER2 therapy only—because only 2 patients (5 lesions) were treated with subsequent immunotherapy within the 3-month window following radiosurgery. A model was also run for repeat radiosurgery lesions.
Fine and Gray Subdistribution Proportional Hazards Model
| Model 1 | Model 2 | Model Retreat | ||||
|---|---|---|---|---|---|---|
| Covariate | Hazard ratio (95% CI) | P | Hazard ratio (95% CI) | P | Hazard ratio (95% CI) | P |
| Age | 1.003 (0.972-1.035) | .840 | 1.001 (0.970-1.033) | .960 | 0.942 (0.793-1.120) | .50 |
| Histologya—lung | 1.098 (0.384-3.143 | .860 | 0.855 (0.308-2.372 | .760 | 0.67 (0.009-49.787) | .85 |
| Histology a—melanoma | 0.385 (0.045-3.296) | .380 | 0.292 (0.035-2.447) | .260 | 7.95 (0.476-131.859) | .15 |
| Histology a—other | 1.313 (0.389-4.432) | .660 | 0.991 (0.305-3.227) | .990 | 0.57 (0.004-80.954) | .82 |
| Cavity | 0.937 (0.288-3.045) | .910 | 0.889 (0.256-3.093) | .850 | ||
| Sex (female) | 0.483 (0.235-0.993) | .048 | 0.487 (0.238-0.994) | .048 | 0.09 (0.009-0.895) | .04 |
| Rx dose, Gy | 1.141 (0.958-1.360) | .240 | 1.164 (0.976-1.387) | .092 | 0.98 (0.745-1.301) | .91 |
| Tumor volume, cc | 1.082 (1.014-1.155) | .017 | 1.096 (1.008-1.192) | .032 | 1.03 (0.934-1.146) | .51 |
| Conformality index | 0.962 (0.849-1.090) | .540 | 0.962 (0.851-1.088) | .530 | 0.74 (0.522-1.061) | .10 |
| WBRT | 1.119 (0.546-2.294) | .760 | 1.160 (0.571-2.359) | .680 | 2.51 (0.321-19.635) | .38 |
| Anti-EGFR therapy | 0.634 (0.054-7.386) | .720 | ||||
| Anti-HER2 therapy | 0.272 (0.035-2.109) | .210 | ||||
| Time to retreat, mo | 1.08 (0.911-1.282) | .37 |
| Model 1 | Model 2 | Model Retreat | ||||
|---|---|---|---|---|---|---|
| Covariate | Hazard ratio (95% CI) | P | Hazard ratio (95% CI) | P | Hazard ratio (95% CI) | P |
| Age | 1.003 (0.972-1.035) | .840 | 1.001 (0.970-1.033) | .960 | 0.942 (0.793-1.120) | .50 |
| Histologya—lung | 1.098 (0.384-3.143 | .860 | 0.855 (0.308-2.372 | .760 | 0.67 (0.009-49.787) | .85 |
| Histology a—melanoma | 0.385 (0.045-3.296) | .380 | 0.292 (0.035-2.447) | .260 | 7.95 (0.476-131.859) | .15 |
| Histology a—other | 1.313 (0.389-4.432) | .660 | 0.991 (0.305-3.227) | .990 | 0.57 (0.004-80.954) | .82 |
| Cavity | 0.937 (0.288-3.045) | .910 | 0.889 (0.256-3.093) | .850 | ||
| Sex (female) | 0.483 (0.235-0.993) | .048 | 0.487 (0.238-0.994) | .048 | 0.09 (0.009-0.895) | .04 |
| Rx dose, Gy | 1.141 (0.958-1.360) | .240 | 1.164 (0.976-1.387) | .092 | 0.98 (0.745-1.301) | .91 |
| Tumor volume, cc | 1.082 (1.014-1.155) | .017 | 1.096 (1.008-1.192) | .032 | 1.03 (0.934-1.146) | .51 |
| Conformality index | 0.962 (0.849-1.090) | .540 | 0.962 (0.851-1.088) | .530 | 0.74 (0.522-1.061) | .10 |
| WBRT | 1.119 (0.546-2.294) | .760 | 1.160 (0.571-2.359) | .680 | 2.51 (0.321-19.635) | .38 |
| Anti-EGFR therapy | 0.634 (0.054-7.386) | .720 | ||||
| Anti-HER2 therapy | 0.272 (0.035-2.109) | .210 | ||||
| Time to retreat, mo | 1.08 (0.911-1.282) | .37 |
Abbreviations: EGFR, epidermal growth factor receptor; HER2, human epidermal growth factor receptor 2; Rx, prescription; SRS, stereotactic radiosurgery; WBRT, whole-brain radiotherapy.
Five cases omitted because of incomplete data in 1 or more covariates in Model 1 and Model 2. Model Retreat considers all 31 cases that had repeat treatment. In this model the Gamma knife parameters are for the second treatment. Bold values are statistically significant.
aHistology hazard ratio is compared to breast histology as baseline.
Fine and Gray Subdistribution Proportional Hazards Model
| Model 1 | Model 2 | Model Retreat | ||||
|---|---|---|---|---|---|---|
| Covariate | Hazard ratio (95% CI) | P | Hazard ratio (95% CI) | P | Hazard ratio (95% CI) | P |
| Age | 1.003 (0.972-1.035) | .840 | 1.001 (0.970-1.033) | .960 | 0.942 (0.793-1.120) | .50 |
| Histologya—lung | 1.098 (0.384-3.143 | .860 | 0.855 (0.308-2.372 | .760 | 0.67 (0.009-49.787) | .85 |
| Histology a—melanoma | 0.385 (0.045-3.296) | .380 | 0.292 (0.035-2.447) | .260 | 7.95 (0.476-131.859) | .15 |
| Histology a—other | 1.313 (0.389-4.432) | .660 | 0.991 (0.305-3.227) | .990 | 0.57 (0.004-80.954) | .82 |
| Cavity | 0.937 (0.288-3.045) | .910 | 0.889 (0.256-3.093) | .850 | ||
| Sex (female) | 0.483 (0.235-0.993) | .048 | 0.487 (0.238-0.994) | .048 | 0.09 (0.009-0.895) | .04 |
| Rx dose, Gy | 1.141 (0.958-1.360) | .240 | 1.164 (0.976-1.387) | .092 | 0.98 (0.745-1.301) | .91 |
| Tumor volume, cc | 1.082 (1.014-1.155) | .017 | 1.096 (1.008-1.192) | .032 | 1.03 (0.934-1.146) | .51 |
| Conformality index | 0.962 (0.849-1.090) | .540 | 0.962 (0.851-1.088) | .530 | 0.74 (0.522-1.061) | .10 |
| WBRT | 1.119 (0.546-2.294) | .760 | 1.160 (0.571-2.359) | .680 | 2.51 (0.321-19.635) | .38 |
| Anti-EGFR therapy | 0.634 (0.054-7.386) | .720 | ||||
| Anti-HER2 therapy | 0.272 (0.035-2.109) | .210 | ||||
| Time to retreat, mo | 1.08 (0.911-1.282) | .37 |
| Model 1 | Model 2 | Model Retreat | ||||
|---|---|---|---|---|---|---|
| Covariate | Hazard ratio (95% CI) | P | Hazard ratio (95% CI) | P | Hazard ratio (95% CI) | P |
| Age | 1.003 (0.972-1.035) | .840 | 1.001 (0.970-1.033) | .960 | 0.942 (0.793-1.120) | .50 |
| Histologya—lung | 1.098 (0.384-3.143 | .860 | 0.855 (0.308-2.372 | .760 | 0.67 (0.009-49.787) | .85 |
| Histology a—melanoma | 0.385 (0.045-3.296) | .380 | 0.292 (0.035-2.447) | .260 | 7.95 (0.476-131.859) | .15 |
| Histology a—other | 1.313 (0.389-4.432) | .660 | 0.991 (0.305-3.227) | .990 | 0.57 (0.004-80.954) | .82 |
| Cavity | 0.937 (0.288-3.045) | .910 | 0.889 (0.256-3.093) | .850 | ||
| Sex (female) | 0.483 (0.235-0.993) | .048 | 0.487 (0.238-0.994) | .048 | 0.09 (0.009-0.895) | .04 |
| Rx dose, Gy | 1.141 (0.958-1.360) | .240 | 1.164 (0.976-1.387) | .092 | 0.98 (0.745-1.301) | .91 |
| Tumor volume, cc | 1.082 (1.014-1.155) | .017 | 1.096 (1.008-1.192) | .032 | 1.03 (0.934-1.146) | .51 |
| Conformality index | 0.962 (0.849-1.090) | .540 | 0.962 (0.851-1.088) | .530 | 0.74 (0.522-1.061) | .10 |
| WBRT | 1.119 (0.546-2.294) | .760 | 1.160 (0.571-2.359) | .680 | 2.51 (0.321-19.635) | .38 |
| Anti-EGFR therapy | 0.634 (0.054-7.386) | .720 | ||||
| Anti-HER2 therapy | 0.272 (0.035-2.109) | .210 | ||||
| Time to retreat, mo | 1.08 (0.911-1.282) | .37 |
Abbreviations: EGFR, epidermal growth factor receptor; HER2, human epidermal growth factor receptor 2; Rx, prescription; SRS, stereotactic radiosurgery; WBRT, whole-brain radiotherapy.
Five cases omitted because of incomplete data in 1 or more covariates in Model 1 and Model 2. Model Retreat considers all 31 cases that had repeat treatment. In this model the Gamma knife parameters are for the second treatment. Bold values are statistically significant.
aHistology hazard ratio is compared to breast histology as baseline.
Tumor volume was an effective continuous predictor of necrosis risk in the upfront setting. Female sex was protective against necrosis in all models. Because of the strong protective effect of female sex on radiation necrosis, a separate model was run omitting sex to assess whether sex was masking the effect of breast cancer histology; however, the regression coefficient for histology was still statistically nonsignificant (data not shown). Additionally, the model for retreated lesions was also run using cumulative dose instead of second radiosurgery dose, showing that cumulative dose was still not statistically significant (data not shown). Histology, delivery of whole-brain irradiation, and treatment of surgical cavity vs intact lesion did not predict for necrosis. Concurrent systemic therapy type was not associated with risk of necrosis; however, concurrent immunotherapy was not modeled because there were too few at-risk lesions.
Cut point analysis suggested 0.305 cc as the tumor volume threshold most predictive for necrosis. The hazard ratio for necrosis for tumor volumes greater than and less than this threshold was 4.66.
Discussion
Our large cohort of long-term survivors reveals the considerable burden of radiation necrosis after radiosurgery for brain metastases. We demonstrate that the cumulative incidence of radiation necrosis continues to increase in the first 4 years of follow-up, with a 4-year incidence of 6.8% (on a per-lesion basis) when death, salvage radiosurgery, or surgical salvage was analyzed as a competing risk. Additionally, retreatment poses an especially large risk—the 4-year incidence rises to 33.3% in these lesions. Our analysis is pragmatic on several fronts. We meticulously reviewed clinical and imaging data to classify radiation necrosis using commonly accepted criteria as described in Methods. We also graded radiation necrosis toxicity based on level of intervention necessary (1: none, 2: medical therapy, 3: surgical therapy, 4: intensive care unit–level care, 5: death), better reflecting the cost to the patient. The majority of patients had grade 1 or 2 toxicity, but 18% of lesions experiencing necrosis required surgical intervention either for diagnostic confirmation or therapeutic intervention.
The incidence of radiation necrosis after radiosurgery has been reported to range from 4% to 30%,5,18–27 with the large variance in rates arising from differing intensity of follow-up and definitions of necrosis. Specifically, radiation necrosis may be recorded only when clearly symptomatic, which ignores the diagnostic burden of necrosis and the burden of steroid use in minimally symptomatic or asymptomatic patients. A large proportion of the data supporting these estimates comes from radiosurgery outcomes for arteriovenous malformations, given their long natural history, large target sizes, and ease of distinguishing residual nidus from necrosis (as opposed to tumor progression vs necrosis). These reports may not accurately estimate the risk of necrosis in brain metastases targets, which have a different local tumor environment and disease course. Studies including brain metastases agree that adverse radiation events occur in approximately 10% of patients, with events occurring as late as 5 years after radiosurgery.24 These studies on the whole, however, estimate long-term risks from short patient follow-up, owing to the historically short survival of patients with brain metastases. This concern motivated an earlier study similar to ours in which Varlotto et al reported a late sequelae rate of 11.4% at 5 years in patients living at least 1 year after radiosurgery.28 However, our study has several important differences from this seminal study. For instance, we consider both symptomatic and asymptomatic radiation necrosis. Second, we explicitly model competing risks in our study. For instance, using a simple Kaplan-Meier estimate for necrosis incidence would have more than doubled our 4-year incidence estimate and as such, the rate of symptomatic necrosis may have been overestimated using simpler actuarial methods. Finally, our patient population is broader than most prior reports on brain metastases patients, including that by Varlotto and colleagues, given our long-standing institutional preference to treat patients with radiosurgery rather than WBRT even in patients with upward of 10 metastases.14 This difference is illustrated in our highernumber of average lesions per patient (3.7 lesions vs 1.5 lesions) and smaller median tumor volume (0.12 cc vs 1.9 cc). Similar differences are also seen when compared to the series reported by Nakamura et al.22
Similar to our study, Varlotto et al28 found that target volume was the only statistically significant treatment factor predicting radiation necrosis. Our study included tumor volume rather than target volume in multivariable analysis because tumor volume is known prior to radiosurgery. These results, however, should be considered analogous because of the inclusion of the conformity index (defined as the ration of target volume: tumor volume) in our model. We also found that tumor volumes greater than 0.3 cc (or more than ~0.8 cm in diameter) result in a more than 4 times higher hazard for necrosis compared to those below this volume cutoff. Several groups have assessed volumetric effects and shown similar increased risk of necrosis when treating larger volumes, including prospective data from the Radiation Therapy Oncology Group 90-05 dose-escalation study.29 Similar volume effects were reported by Nakamura et al.22 Some groups have also suggested the dosimetric volume V12Gy, originally reported as strongly prognostic for necrosis in radiosurgery for arteriovenous malformations,30 may possibly have significance in predicting the risk of necrosis after radiosurgery for brain metastases.31 To mitigate risks of necrosis with large volumes, a hypofractionated radiotherapy approach may be pursued. Minniti et al reported a lower risk of radiation necrosis for lesions larger than 2 cm in diameter using a hypofractionated approach,32 and similarly, Kim and colleagues reported lower toxicity with a hypofractionated approach for tumors greater than 5 mL in volume.33 A hypofractionated strategy may be especially beneficial in the treatment of resection cavities, in which volume expansions are often used along meningeal surfaces.
Korytko et al31 reported male sex to be a risk factor for radiation necrosis, but this finding has not been reported by other investigators to the best of our knowledge. Our findings agree with this earlier report and seem to not be confounded by histology. Given the strong effect seen in our cohort (hazard ratio ~0.5), we feel that exploring sex propensity differences for radiation necrosis and the potential vascular mechanisms for these differences are avenues for further study.
The substantial risk of radiation necrosis after retreatment with radiosurgery34 (36% 4-year risk of necrosis in our series) suggests that one must consider alternative methods of achieving intracranial disease control, especially in eloquent locations in patients with long projected survival. Several new strategies are emerging for the treatment of brain metastases with improved toxicity. Modern systemic agents including targeted therapy for lung and breast cancer and immunotherapy for melanoma and lung cancer have shown promising activity against brain metastases both in the setting of progressive disease after prior therapy and as an upfront treatment strategy in highly selected asymptomatic patients undergoing close surveillance.35–40 We are also encouraged by the recent maturation of trials supporting the use of memantine41 and hippocampal avoidance42 with WBRT to reduce cognitive decline in those patients requiring whole-brain therapy. All of these techniques can help reduce the need for repeat radiosurgery.
Several limitations exist in our analysis. We consider only patients with longer than 1 year of survival in our database, which may bias our results. However, this population selection allowed us to better estimate the long-term risk of necrosis by limiting the analysis to those who had a long survival, a population that is underreported in the literature. The risk of bias with this approach was minimized by considering competing risk, to the extent that such biases can be accounted for in a cohort study. We were also unable to analyze dose-volume parameters (such as V10Gy or V12Gy) because our database does not currently record these parameters. Additionally, MRI perfusion and spectroscopy have limited accuracy in diagnosing radiation necrosis prospectively; 43 however, we retrospectively reviewed each case including follow-up status to help define whether patients truly had necrosis or recurrence. Finally, this cohort of patients included only 2 patients who received immunotherapy within the 3-month time window of radiosurgery because most of these patients were treated in the “preimmunotherapy” era. Thus, we could not recapitulate recent findings that immunotherapy may increase the risk of radiation necrosis.44,45
In summary, our analysis shows that patients continue to be at risk for radiation necrosis for at least the first 4 years after treatment and that the risk of necrosis after retreatment may be unacceptably high, especially in the presence of effective systematic treatment options or less-toxic WBRT strategies. Further work is required to characterize the interaction of sex with radiation necrosis risk. In addition, a consensus-based, pragmatic definition of radiation necrosis would help in comparing results from different series and providing uniform reporting of adverse events from radiosurgery in trials.
Funding
This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors.
Acknowledgments
We would like to acknowledge Hong Ye, PhD, in the Department of Radiation Oncology at Beaumont Health, for assistance in verifying the statistics used in this paper.
Portions of this work were presented in poster form at the 2016 meeting of the American Society of Radiation Oncology, September 25 to 28, 2016, in Boston, Massachusetts.
Conflict of interest statement
PYC, DJK, and REO own stock in Greater Michigan Gamma Knife. ISG has received a research grant from Elekta, owns stock in and is on the board of Greater Michigan Gamma Knife, and is residency program director of Beaumont Health. The other authors have nothing to declare.
