A nomogram for predicting distant brain failure in patients treated with gamma knife stereotactic radiosurgery without whole brain radiotherapy

Abstract

Background

We review our single institution experience to determine predictive factors for early and delayed distant brain failure (DBF) after radiosurgery without whole brain radiotherapy (WBRT) for brain metastases.

Materials and methods

Between January 2000 and December 2010, a total of 464 patients were treated with Gamma Knife stereotactic radiosurgery (SRS) without WBRT for primary management of newly diagnosed brain metastases. Histology, systemic disease, RPA class, and number of metastases were evaluated as possible predictors of DBF rate. DBF rates were determined by serial MRI. Kaplan–Meier method was used to estimate rate of DBF. Multivariate analysis was performed using Cox Proportional Hazard regression.

Results

Median number of lesions treated was 1 (range 1–13). Median time to DBF was 4.9 months. Twenty-seven percent of patients ultimately required WBRT with median time to WBRT of 5.6 months. Progressive systemic disease (χ²= 16.748, P < .001), number of metastases at SRS (χ² = 27.216, P < .001), discovery of new metastases at time of SRS (χ² = 9.197, P < .01), and histology (χ² = 12.819, P < .07) were factors that predicted for earlier time to distant failure. High risk histologic subtypes (melanoma, her2 negative breast, χ² = 11.020, P < .001) and low risk subtypes (her2 + breast, χ² = 11.343, P < .001) were identified. Progressive systemic disease (χ² = 9.549, P < .01), number of brain metastases (χ² = 16.953, P < .001), minimum SRS dose (χ² = 21.609, P < .001), and widespread metastatic disease (χ² = 29.396, P < .001) were predictive of shorter time to WBRT.

Conclusion

Systemic disease, number of metastases, and histology are factors that predict distant failure rate after primary radiosurgical management of brain metastases.

The median survival for patients with brain metastases has improved significantly over time due to earlier detection,¹ better brain-directed² and systemic therapies.³ With the advent of improved salvage therapies for intracranial metastases including radiosurgery and stereotactic radiotherapy, many patients with intracranial metastases will ultimately die from their systemic disease burden. As patients live longer, there has been an incentive to delay or avoid the neurocognitive toxicities associated with whole brain radiotherapy, as these toxicities continue to worsen with longer survival time.⁴

With increased use of stereotactic radiosurgery (SRS) in the management of brain metastases, the ability to determine the optimal patient population for this technique has become paramount. Identifying suitable patients for upfront SRS versus whole brain radiotherapy (WBRT) will not only contribute to the quality of life of the patient, but also help ease the societal tensions in determining meaningful use of expensive medical resources. There are certainly populations that may benefit from upfront whole brain radiotherapy for their brain metastases. These populations likely include patients with numerous lesions [Yamamoto et al], those with rapid intracranial progression after initial SRS, and those with poor overall prognosis. Elucidation of those patient and disease characteristics that can predict for the latter two factors will become vital in the triage of patients to the proper treatment modality.

While several publications have described factors that predict for survival of patients with brain metastases,^5,6 relatively little evidence exists to describe factors that predict for early distant brain failure (DBF) after Gamma Knife radiosurgery. A single series published out of the University of Alabama Birmingham identified factors such as <4 metastases, absence of extracranial disease and non-melanoma histology as factors that predict for DBF.⁷ More recent evidence has suggested that even within populations such as lung cancer and breast cancer, differences in histological and molecular subtypes of lung and breast cancer can yield differences in DBF rates.^8,9 Vern-Gross et al recently published data showing that patients with triple negative breast cancers exhibited increased rates of DBF relative to other subtypes of breast cancer.⁹

Materials and Methods

Data Acquisition

This study was approved by the Institutional Review Board at our institution. Data was reviewed and collected at our institution on 464 patients with brain metastases treated at our institution between January 2000 and December 2010. Electronic medical records were reviewed to determine clinicopathologic characteristics including histology, age, sex, race, date of diagnosis, RPA class, and burden of intracranial and systemic disease at the time of presentation of brain metastasis. RPA class was determined as described by Gaspar et al.⁶ Patient characteristics are summarized in Table 1. The date of SRS, number of lesions treated with radiosurgery, and marginal dose delivered were determined via the Leksell GammaPlan Treating Planning system (Elekta AB, Stockholm, Sweden).

Table 1.

Patient Characteristics

	Number (%/range)
Patients	464
Median Age	62 (43–89)
Sex
Female	217 (47%)
Male	247 (53%)
Burden of Extracranial Disease
None	112 (24.1%)
Oligometastatic	150 (32.3%)
Widespread	158 (34.1%)
Unknown	44 (9.5%)
Status of Extracranial Disease
Stable	238 (51.3%)
Progressive	165 (35.6%)
Unknown	61 (13.1%)
RPA Class
I	30 (6.5%)
II	400 (86.2%
III	34 (7.3%)
Primary Tumor
Breast
Her 2 negative	36 (7.8%)
Her 2 positive	38 (8.2%)
Unknown	5 (1.1%)
Lung
Adenocarcinoma	140 (30.1%)
Squamous cell	43 (9.3%)
Other	56 (12.1%)
Melanoma	91 (19.6%)
Renal Cell	53 (11.4%)
Median Number of Lesions	1 (1–13)
Minimal Marginal Dose	20 Gy (10–24)

Patient Follow-up and Salvage Therapy

Patients were seen in follow-up ∼1 month after their SRS procedure and every 3 months following that initial post-procedure visit. The patient surveillance schedule was standardized as previously described by Lester et al. MRIs were obtained at each scheduled follow-up visit.¹⁰ DBFs were determined based on serial imaging evidence of intracranial recurrence. DBFs were defined as any new metastases that developed outside of the previous radiosurgical target volume. Patients who developed further brain metastases after SRS were generally treated with further SRS, while WBRT was generally reserved for salvage of four or more total brain metastases over time or in the setting of short-interval DBF. With regards to local control, the local brain failure was defined as either a pathologically-proven recurrence within the GKRS treatment field, or a combination of imaging and clinical characteristics of local treatment failure. Imaging characteristics of treatment failure included an increase in area of enhancement by 25% on axial slice and/or serial increases in size of enhancement with corresponding increased perfusion on perfusion-weighted imaging.

Radiosurgery

Prior to radiosurgery, each patient underwent a high-resolution contrast-enhanced stereotactic magnetic resonance imaging (MRI) study of the brain. Treatment planning was performed using the GammaPlan Treatment Planning System. SRS was performed using either the Leksell Model B (years 2000–2004), Model C (years 2004–2009) or Perfexion (years 2009–2010) Gamma Knife units (Elekta AB, Stockholm, Sweden). The median dose delivered to the tumor margin was 20 Gy (range: 10–24 Gy). Dose was generally prescribed to the 50% isodose line, and the prescription was determined based upon the guidelines published by Shaw et al for single fraction radiosurgical treatment of brain metastases.¹¹

Statistics

Time to DBF was calculated from time of SRS to DBF or censored at time of last imaging if there was no distant failure. Time to WBRT was calculated from time of SRS to whole brain radiotherapy or censored at date of last follow-up if the patient did not receive WBRT. Time to event curves were generated using the Kaplan–Meier method and significance was tested using the log-rank test. Cox proportional hazard regression in a forward and backward stepwise method was used to determine covariates that predicted for time to DBF and time to WBRT. Differences among levels of factors were assessed with Chi-Square test.

From the univariate and multivariate Cox regression results, a nomogram was constructed. The original observed data used to create the database for nomogram construction included 464 observations. Bootstrap sampling was used to create 1,000 new datasets, each with observations. Time to DBF was analyzed in each of these bootstrap samples using two Cox proportional hazards models. The composite model included all variables identified by the above methods, but all continuous variables except for age were converted to categorical covariates for intended ease of use. The impact of age was limited to discrete estimates every 5 years. Predicted probabilities of DBF were output for each combination of covariates in both models at three-month intervals. There were 2800 unique combinations of covariates. For each combination, a median and a 95 percent confidence interval was calculated for distant brain failure, using the 1,000 bootstrap samples. The R macro “RMS” was used to generate the calibration curves for each 3 month time interval.¹² The model was assessed for accuracy at each time point and it was determined that the 6 and 9 month intervals displayed the highest accuracy. The resulting nomogram figure was generated using the “RMS” package.

Results

Distant Brain Failure and Local Failure

Kaplan–Meier analysis was used to calculate freedom from DBF and freedom from Local Failure (LF). Results of Kaplan–Meier analysis are depicted in Fig. 1. The median time to DBF for the entire cohort was 4.9 months. Freedom from DBF at 6 months, 1 year, and 2 years was 59%, 30%, and 14% respectively. Sixty-five of the 464 cases experienced a LF with a median time to LF of 10 (+/−1.1) months. For the entire cohort, the freedom from LF at 6 months, 1 year, and 2 years was 94.6%, 90%, and 87.5% respectively. The local failure rates at 1 year for breast, lung, renal cell and melanoma histologies were 91%, 92%, 85% and 92%.

Fig. 1.

Kaplan–Meier plot of freedom from DBF (Fig. 1a) and freedom from LF (Fig. 1b) for 464 patients treated with upfront radiosurgery without whole brain radiotherapy.

High Risk and Low Risk Histologies for Distant Brain Failure

Kaplan–Meier analysis was performed for time to DBF for distinct histologies including melanoma, squamous cell and adenocarcinoma of the lung, her2+ and her2 negative breast cancer, and renal cell carcinoma. The histologies with the shortest time to distant failure included melanoma (median 3.3 months) and Her2 negative breast cancers (median 3.7 months). The histology with the greatest delay to DBF was HER2+ breast cancer (median 9.5 months). Kaplan–Meier analysis of DBF as stratified by histology subtypes is depicted in Fig. 2.

Fig. 2.

Kaplan–Meier plot of freedom from DBF stratified by risk stratification groupings. Her2+ breast cancer represented the lowest risk group. Melanoma and Her2 negative breast cancer represented the highest risk group. All other histologies were included in the intermediate risk group.

Log rank tests were performed to determine if any of the histologies had significantly better or worse time to DBF. High risk histologic subtypes (melanoma, her2 negative breast, χ² = 11.020, P < .001) and low risk subtype (her2+ breast, χ² = 11.343, P < .001) were identified.

Time to WBRT

Kaplan–Meier analysis was used to calculate time to WBRT. Results of Kaplan Meier analysis are depicted in Fig. 3. The median time to WBRT for the entire cohort was 5.6 months. Two-hundred and eighty six of 410 patients (70%) who had died by last follow-up never required WBRT until time of death. Freedom from WBRT at 6 months, 1 year, and 2 years was 48%, 22%, and 4% respectively.

Fig. 3.

Kaplan–Meier plot of freedom from whole brain radiotherapy.

The histologies with the shortest time to Salvage WBRT included melanoma (median 3.3 months) and poorly differentiated lung cancers (median 3.0 months). The histology with the greatest delay to DBF was HER2+ breast cancer (median 9.5 months). Kaplan–Meier analysis of time to WBRT as stratified by histology is depicted in Fig. 4.

Fig. 4.

Kaplan–Meier plot of freedom from whole brain radiotherapy by risk stratification groupings.

Predictive Factors

A forward and backward conditional entry Cox proportional hazard regression was performed to identify factors that predicted for early DBF and for early use of WBRT. Progressive systemic disease (χ² = 16.748, P < .001), number of metastases at SRS (χ² = 27.216, P < .001), discovery of new metastases at time of SRS (χ² = 9.197, P < .01), and high-risk histology (χ² = 12.819, P < .07) were factors that predicted for earlier time to distant failure.

Progressive systemic disease (χ² = 9.549, P < .01), number of brain metastases (χ² = 16.953, P < .001), minimum SRS dose (χ² = 21.609, P < .001), and widespread metastatic disease (χ² = 29.396, P < .001) were predictive of earlier requirement of WBRT.

Nomogram Creation

Based on the factors identified above, we constructed a nomogram by bootstrap sampling to create 1000 databases, which could then be used to generated confidence intervals around the estimates for every possible combination of the identified covariates in the Cox proportional hazards model. The calibration curves showed agreement with the observed data, with the confidence intervals narrowing in the patients with the lowest and highest risk for DBF, due to the increased number of events for these populations at the 6 and 9 month assessment. The estimates were then used to generate the nomogram in Fig. 5. The resultant calibrated Cox proportional hazards models are displayed in Fig. 6.

Fig. 5.

Nomogram for distant brain failure at 6 and 9 months. Systemic Disease Status 3 = Unknown, 2 = Stable, 1 = Progressive. Abbreviations: Met num, Metastasis Number; Systemic Disease GK0, Disease status at the time of Gamma Knife; L Squamous, Lung Squamous Cell Carcinoma; L Adeno, Lung Adenocarcinoma; Her2, Human Epidermal Growth Factor Receptor 2; RCC, Renal Cell Carcinoma; L Other, Lung Other histology; NSCLC, Non-Small Cell Lung Cancer.

Fig. 6.

Calibration Curves for (a) three, (b) six, and (c) nine months.

Discussion

DBF is among the major risks of treatment of primary SRS for the treatment of brain metastases. The rate of DBF in the current series was 70% at 1 year, which is similar to DBF rates that have been reported in large prospective trials of patients receiving SRS while withholding WBRT for salvage therapy (Table 2). Prior randomized studies have demonstrated that patients who receive WBRT with radiosurgery have a reduced risk of both DBF and of neurologic death.¹³ However, the use of WBRT has been associated with an acute decrease in performance status¹⁴ and with chronic and progressive worsening of cognitive function.^4,15 Thus, providers must balance the benefits of SRS with its associated risk for potential DBF and the negative impact progressive intracranial disease can have on overall cognitive functioning,¹⁶ with the potential consequences of WBRT for those patients who survive several months after therapy.

Table 2.

Randomized trials assessing treatment outcomes when withholding WBRT

	Treatment Arm	1 year Distant Brain Failure	1 year Local Failure	1 year Overall Survival
JROSG 99–1	SRS	63%	24%	28%
	SRS + WBRT	42%	10%	39%
MDACC	SRS	55%	33%	60%
	SRS + WBRT	27%	0%	21%
EORTC 22952	SRS	44%	30%	47%
	SRS + WBRT	28%	13%	46%

A predictive model for DBF would be a clinically useful tool in identifying patients who may have early DBF and thus, would benefit from earlier WBRT. Clinical prediction tools have contributed to a variety of treatment algorithms for different disease sites and have helped identify appropriate therapies for different patient populations. In addition, such tools can have a beneficial impact on health economics as radiosurgical management would not be a wise allocation of resources for patients with a strong probability of experiencing early DBF. Furthermore, because the incidence and severity of late toxicities associated with WBRT do not plateau with time, there is an incentive to identify patients who are likely to have a more durable freedom from DBF in order to maintain quality of life for those living with metastatic disease.

Histology as a predictor of DBF has been reported by several series in the past. Sawrie et al identified melanoma histology as one of the dominant factors predicting DBF following primary SRS for brain metastases.⁷ Vern-Gross et al found that patients with triple negative breast histology had a higher rate of DBF and reduced overall survival as compared to other breast cancer populations.⁹ Kuremsky et al recently reported that patients with small cell and squamous cell lung cancers have a higher rate of DBF and a greater need for WBRT than other lung cancer histologies.⁸ Our data agrees with prior publications in that melanoma and her2 negative breast cancers were identified as high-risk histologies at risk for early DBF.

Correlations between intracranial disease burden and progression free survival have also been observed. Patel et al demonstrated that the discovery of metastases at time of radiosurgical planning was a significant factor predicting early DBF.¹⁷ In this series, the identification of further metastases at the time of radiosurgery was highly correlated to patients having progressive systemic disease. A recent study from Georgetown looking at single brain metastases from non-small cell lung cancer suggests that progression of extracranial disease also predicts for distant intracranial failure.¹⁸ In the current series, both discovery of new metastases at time of radiosurgery, and the presence of progressive systemic disease were dominant factors predictive of earlier DBF.

The limitation of this study is its retrospective nature and the fact that patients were treated at a single institution which predisposes the study to patient selection bias. The data from this series largely confirms findings from previously published series, but also suggests that certain low risk and high risk populations can be identified based on histology. The generated nomogram may ultimately prove to be a useful resource when making management decisions for individual patients. A multi-institutional independent data set is currently being generated in order to validate the developed prediction tool.

Conclusion

DBF remains problematic after single modality radiosurgery in the treatment of brain metastases. The histology of the primary tumor may affect the DBF rate.

Conflict of interest statement. None declared.