8 pages

/lp/ou_press/superior-prognostic-value-of-cumulative-intracranial-tumor-volume-mlZVrn3kcG

- Publisher
- Congress of Neurological Surgeons
- Copyright
- Copyright © 2017 by the Congress of Neurological Surgeons
- ISSN
- 0148-396X
- eISSN
- 1524-4040
- D.O.I.
- 10.1093/neuros/nyx225
- Publisher site
- See Article on Publisher Site

Abstract BACKGROUND Two intracranial tumor volume variables have been shown to prognosticate survival of stereotactic-radiosurgery-treated brain metastasis patients: the largest intracranial tumor volume (LITV) and the cumulative intracranial tumor volume (CITV). OBJECTIVE To determine whether the prognostic value of the Scored Index for Radiosurgery (SIR) model can be improved by replacing one of its components—LITV—with CITV. METHODS We compared LITV and CITV in terms of their survival prognostication using a series of multivariable models that included known components of the SIR: age, Karnofsky Performance Score, status of extracranial disease, and the number of brain metastases. Models were compared using established statistical measures, including the net reclassification improvement (NRI > 0) and integrated discrimination improvement (IDI). The analysis was performed in 2 independent cohorts, each consisting of ∼3000 patients. RESULTS In both cohorts, CITV was shown to be independently predictive of patient survival. Replacement of LITV with CITV in the SIR model improved the model's ability to predict 1-yr survival. In the first cohort, the CITV model showed an NRI > 0 improvement of 0.2574 (95% confidence interval [CI] 0.1890-0.3257) and IDI of 0.0088 (95% CI 0.0057-0.0119) relative to the LITV model. In the second cohort, the CITV model showed a NRI > 0 of 0.2604 (95% CI 0.1796-0.3411) and IDI of 0.0051 (95% CI 0.0029-0.0073) relative to the LITV model. CONCLUSION After accounting for covariates within the SIR model, CITV offers superior prognostic value relative to LITV for stereotactic radiosurgery-treated brain metastasis patients. Brain metastases, Cumulative intracranial tumor volume, Net reclassification improvement, Prognostic model, Radiosurgery ABBREVIATIONS ABBREVIATIONS AIC Akaike Information Criterion BM brain metastases CI confidence interval CITV cumulative intracranial tumor volume IDI integrated discrimination improvement KPS Karnofsky Performance Score LITV largest intracranial tumor NIH National Institutes of Health ROC receiver operating characteristic SIR scored index for radiosurgery SRS stereotactic radiosurgery WBRT whole brain radiation therapy NRI net reclassification improvement Recent estimates suggest that 25% to 45% of the cancer patient population will develop brain metastases (BM).1 Because most chemotherapeutic agents are unable to penetrate the blood–brain barrier, surgery and radiation are the mainstay treatments for BM.2 For surgically inaccessible BMs or small BMs without associated neurological deficits, radiation is the preferred treatment. Radiation delivery can be achieved by use of whole brain radiation therapy (WBRT),3 hippocampal sparing WBRT,4 and stereotactic radiosurgery (SRS).5 Because of potential deleterious neurocognitive effects associated with WBRT,6 SRS is increasingly utilized for BM patients with a limited number of metastases.7,8 Based on the Surveillance Epidemiology and End Results multistate cancer registry, SRS as treatment for BM has more than tripled in the past decade.9 While recent studies show increased survival for a subset of patients with BMs after the introduction of molecularly targeted therapies10 and immunotherapies,11 the survival expectation for patients with BM as a whole remains poor.12 With an expected median survival of 2 to 7 mo for all BM patients,2,13 a major challenge in clinical management involves weighing the inherent risks of potential intervention relative to survival expectation.14 In this context, several prognostic scales have been developed for patients afflicted with BM.15-17 While the specifics of these scales differ, they generally involve the same clinical variables, including age, Karnofsky Performance Score (KPS), extracranial disease status, histology, and the number of metastases.15-17 Intracranial tumor volume is an important prognostic factor for SRS-treated BM patients.17-20 The first prognostic scale to acknowledge the importance of intracranial tumor volume is termed Scored Index for Radiosurgery (SIR). SIR included 5 prognostic factors, including age, KPS, extracranial disease status, number of brain lesions, and volume of the largest intracranial tumor (LITV).17 Our previous studies suggested that the cumulative intracranial tumor volume (CITV), or the sum volume of all intracranial metastases, may also serve as a critical prognostic factor for SRS-treated BM patients.18-20 Here, we utilized the framework established by the SIR to test the prognostic performance of CITV relative to LITV. METHODS Study Cohorts All data collection was performed under Institutional Review Board approved retrospective review, which included a waiver of informed consent. The first dataset consists of 3061 consecutive BM patients treated by TS (1998-2014). This study population is termed the “TS cohort”. A posthoc sample size analysis of the TS cohort receiver operating characteristic (ROC) curves indicated that a minimum of 2055 patients would be necessary for a power of 0.8021 to detect the effect size seen in the cohort. To ensure our validation cohort was sufficiently powered to detect a difference of similar magnitude, we combined the cohorts of patients treated by MH and CC (1994-2014) to a resulting data set of 2793 patients. This study population is termed the “MC” cohort. Details of these cohorts have been previously published.18-20,22 In brief, every patient had a primary tumor confirmed outside of the central nervous system and underwent SRS without craniotomy. All patients were referred for SRS by a radiation or medical oncologist as well as a neurosurgeon. A multidisciplinary tumor board reviewed each of the SRS candidates. Patients who were recommended and underwent surgical resection were not included in the present analysis. For the purposes of our analyses, patients who underwent multiple radiosurgery treatments (n = 273—4.7% of all analyzed patients) were labeled as new patients, based on previously published findings that survival is comparable between patients receiving a singular round of SRS vs those receiving multiple SRS treatments.20 Decisions to pursue SRS vs WBRT were personalized based on the patient's general medical status as well as the status of their oncologic disease. In general, the clinical practice pattern of the 3 treating surgeons conform to generally accepted guidelines, which include reserving WBRT for those with prohibitively large tumor volumes, those unable to follow up with surveillance MRIs, those with miliary disease, and those receiving radiation for the sole purpose of palliation.23-25 Patient baseline characteristics for each data set are available in Table 1. SRS was performed as previously described.18 TABLE 1. Demographics and Clinical Characteristic of Study Patientsa TS cohort MC cohort Number of patients 3061 2793 Socio-demographics Age (mean, SD) 64.5 ± 11.1 63.2 ± 12.0 Sex (% male) 59.2% 58.9% Patient properties KPS (mean, SD) 85.5 ± 14.4 85.0 ± 13.4 Number of metastases (mean, SD) 6.6 ± 9.0 5.7 ± 8.9 Active extracranial disease (%) 68.6% 85.6% Largest lesion (cc; mean, SD) 5.3 ± 6.8 6.4 ± 8.9 CITV (cc; mean, SD) 8.3 ± 10.2 8.7 ± 11.8 Type of cancer (%) Breast 11.0% 11.9% Gastrointestinal 12.7% 9.1% Lung 65.5% 61.6% Urogenital 6.2% 4.8% Other 4.6% 12.6% Survival Overall survival (years) (mean, SD) 1.1 ± 1.4 0.9 ± 1.4 Patients with minimum 1 yr survival 33.4% 27.6% TS cohort MC cohort Number of patients 3061 2793 Socio-demographics Age (mean, SD) 64.5 ± 11.1 63.2 ± 12.0 Sex (% male) 59.2% 58.9% Patient properties KPS (mean, SD) 85.5 ± 14.4 85.0 ± 13.4 Number of metastases (mean, SD) 6.6 ± 9.0 5.7 ± 8.9 Active extracranial disease (%) 68.6% 85.6% Largest lesion (cc; mean, SD) 5.3 ± 6.8 6.4 ± 8.9 CITV (cc; mean, SD) 8.3 ± 10.2 8.7 ± 11.8 Type of cancer (%) Breast 11.0% 11.9% Gastrointestinal 12.7% 9.1% Lung 65.5% 61.6% Urogenital 6.2% 4.8% Other 4.6% 12.6% Survival Overall survival (years) (mean, SD) 1.1 ± 1.4 0.9 ± 1.4 Patients with minimum 1 yr survival 33.4% 27.6% acc, cubic centimeters; CITV, cumulative intracranial tumor volume; KPS, Karnofsky performance score; SD, standard deviation. View Large Radiosurgery Parameters and Overall Survival In both cohorts, in-house electronic medical record data were gathered, including age, KPS, extracranial disease status, number of BM, CITV, LITV, and overall survival. Only patients for whom each of these variables were known were analyzed; patients with any missing data (515 of 6369, or 8.1%) were excluded for the purposes of our analyses. We defined overall survival as the time (in months) from SRS to time of death. LITV was defined as the largest contiguous lesion present on the pre-SRS (T1-weighted postcontrast image) as calculated by the Elekta's Gamma Plan software. As previously published,18 CITV was defined as the sum volume of all treated lesions, such that in patients with a single lesion, the CITV and LITV values were identical. To facilitate comparison of the prognostic value of LITV and CITV, we set out to compare the sensitivity and specificity of 1-yr survival prognostication by SIR relative to a model where LITV was replaced with CITV (termed CITV-modified SIR). The SIR scale (termed SIR model) used in this study is as originally proposed17 and included age, KPS, number of metastases, extracranial disease status, and LITV (see Table 2). We used the optimized cut-offs of LITV (<5, 5-13, and >13 cc) as previously reported by Weltman et al.17 For the CITV-modified SIR model, all variables of SIR were included except that LITV was replaced with CITV. The optimized CITV cut-points (<3.5, 3.5-13, and >13 cc) were derived using methods similar to those used by Weltman et al17 for LITV (see Table 2). Median survival time (in months) for each of the point groupings is shown in Table 3. TABLE 2. Scoring Components for SIR- and CITV-Modified SIR Modelsa SIR model CITV-modified SIR model 0 points 1 point 2 points 0 points 1 point 2 points Age ≥60 yr 51-59 ≤50 yr ≥60 yr 51-59 ≤50 yr KPS ≤50 60-70 ≥80 ≤50 60-70 ≥80 Number of lesions ≥3 2 1 ≥3 2 1 Extracranial disease status PD PR CR PD PR CR Largest lesion size <5 cc 5-13 cc >13 cc – – – CITV – – – <3.5 cc 3.5-13 cc >13 cc SIR model CITV-modified SIR model 0 points 1 point 2 points 0 points 1 point 2 points Age ≥60 yr 51-59 ≤50 yr ≥60 yr 51-59 ≤50 yr KPS ≤50 60-70 ≥80 ≤50 60-70 ≥80 Number of lesions ≥3 2 1 ≥3 2 1 Extracranial disease status PD PR CR PD PR CR Largest lesion size <5 cc 5-13 cc >13 cc – – – CITV – – – <3.5 cc 3.5-13 cc >13 cc aKPS, Karnofsky performance score; PD, progressive disease; PR, partial remission; CR, complete remission; cc, cubic centimeters; CITV, cumulative intracranial tumor volume; SIR, scored index for radiosurgery. View Large TABLE 3. Median Survival Time (in months) of SIR- and CITV-Modified SIR Point Groupingsa Point total 0 1 2 3 4 5 6 7 8 9 10 SIR model 2.7 2.2 3.4 4.3 7.0 7.7 9.5 13.6 21.6 23.3 42.0 CITV-SIR model 2.5 2.6 3.8 5.1 7.3 8.4 10.6 15.3 21.2 27.4 42.0 Point total 0 1 2 3 4 5 6 7 8 9 10 SIR model 2.7 2.2 3.4 4.3 7.0 7.7 9.5 13.6 21.6 23.3 42.0 CITV-SIR model 2.5 2.6 3.8 5.1 7.3 8.4 10.6 15.3 21.2 27.4 42.0 aSIR, scored index for radiosurgery; CITV, cumulative intracranial tumor volume. View Large Statistical Methods In order to justify direct comparison of CITV to LITV within the SIR model, we first needed to identify CITV as an independent predictor of survival. This was achieved by first performing a series of univariable analyses comparing the proportions and distributions of each SIR factor (ie, age, KPS, extracranial disease status, number of BM, LITV) across CITV groupings, to determine the extent of variability amongst SIR factors within distinct CITV groupings. Continuous factors were compared using ANOVA tests, while categorical factors were compared using chi-square tests. We next performed both a multivariable linear regression analysis and Cox proportional analysis to examine the predictive significance of CITV for patient survival (in months) after accounting for the remaining SIR factors (age, KPS, extracranial disease status, and number of BM). For relative comparison, the same set of multivariable analyses were performed examining the predictive capacity of LITV (ie, using the original SIR model factors). To more thoroughly compare the prognostic significance of CITV to LITV, a final set of multivariable analyses (linear regression and Cox proportional hazards) were performed including both CITV and LITV as predictors against the outcome of continuous survival; these models were examined to determine which variables retained prognostic significance upon the inclusion of both CITV and LITV. For all multivariable models, the Akaike Information Criterion (AIC) was calculated, such that the goodness of fit could be compared between models which included CITV vs LITV. In brief, the AIC allows for the quantification of a model's goodness of fit while at the same time penalizing models that are over-fit and would limit the external validity of the model to new data (ie, other patients not contained within the dataset). When comparing 2 models, a lower AIC indicates an improved goodness of fit; a reduction in AIC of at least 2 is necessary for one model to be deemed superior to the other.26 Finally, to directly compare the prognostic value of our CITV-modified SIR model to the traditional SIR model, we relied on the net reclassification improvement (NRI) and integrated discrimination improvement (IDI) indices, as well as the calculated change in area under the ROC curve (ΔAUC) to determine the extent of prognostic improvement afforded by the replacement of LITV with CITV within the SIR model. In each case, patient survival was defined according to the discrete time point of 12 mo. This time point was chosen in part due to its prevalent usage in the literature.27-29 Calculating ΔAUC allowed for comparison of true positive and false positive rates for a given test as a whole. Calculating NRI and IDI helps quantify the number of correctly reassigned cases (ie, false negative to true negative, false positive to true positive) after implementing a change in the model such as the replacement of LITV with CITV. Taken together, all 3 values provided complementary information as to the relative improvement in a prognostic model which may result from a change in one (or more) of its variables.30 Statistical comparisons were performed with R version 3.2.3 and the predictABEL package for ΔAUC, NRI, and IDI calculations. R’s plotROC package was used to plot the ROC curves. All P < .05 were considered statistically significant. RESULTS Comparisons of LITV and CITV Detailed descriptions of the 2 study cohorts (TS and MC) have been previously published.18-20,22 Pertinent demographic and clinical variables pertaining to these cohorts are shown in Table 1. Both cohorts showed comparable demographics, with similar distribution in age (mean age of 64.5 and 63.2) and sex (59.2% and 58.9% male). The distribution in cancer types was also similar, with a predominance of lung cancer patients in both cohorts (65.5% and 61.6% for TS and MC cohorts, respectively). We first characterized relative distribution of LITV and CITV for each SRS-treated patient in the TS cohort. We found that LITV and CITV differed significantly for the majority of the patients. The CITV was at least 50% greater than the LITV in ∼80% of the patients (Figure 1A). CITV was as high as 400% of LITV for 10% to 30% of the SRS-treated BM patients for each category of LITV. These results suggest that the majority of the BM patients treated harbored multiple intracranial lesions. Moreover, the sum volume of the intracranial lesions (CITV) differed significantly from that of the largest tumor volume (8.33 vs 5.33 cc, P < .001). FIGURE 1. View largeDownload slide Discrepancy between CITV and LITV in the TS cohort. A, Histogram showing the number of patients (y-axis) with a given largest lesion size (x-axis). Relative increase of CITV to largest lesion was color coded such that increasing intensity reflects greater discrepancy between CITV and LITV. CITV was as high as 400% of LITV for 10% to 30% of the SRS-treated BM patients for each category of LITV (shown in 1 cc increments on the x-axis). B, ROC curve showing increased AUC for CITV relative to LITV for prognosticating 1-yr survival. FIGURE 1. View largeDownload slide Discrepancy between CITV and LITV in the TS cohort. A, Histogram showing the number of patients (y-axis) with a given largest lesion size (x-axis). Relative increase of CITV to largest lesion was color coded such that increasing intensity reflects greater discrepancy between CITV and LITV. CITV was as high as 400% of LITV for 10% to 30% of the SRS-treated BM patients for each category of LITV (shown in 1 cc increments on the x-axis). B, ROC curve showing increased AUC for CITV relative to LITV for prognosticating 1-yr survival. Prognostic Value of CITV The results of our univariable analyses are shown in Table 4. As shown in this table, KPS, extracranial disease status, number of lesions, and LITV were all found to differ for patients according to their CITV point grouping (ie, <3.5, 3.5-13, >13 cc). Only age was found to not significantly differ according to CITV point grouping. TABLE 4. Univariable Analysis of Each SIR Factor as Distributed Across CITV Groupings (<3.5, 3-13, >13)a CITV grouping >13cc 3.5-13cc <3.5cc Statistical significance Age in years (mean, SD) 64.3 ± 12.0 64.8 ± 11.1 64.3 ± 10.8 P = .843 KPS (mean, SD) 77.2 ± 15.0 84.2 ± 14.5 90.8 ± 11.7 P < .001 Number of metastases (mean, SD) 7.9 ± 10.3 6.7 ± 9.3 5.7 ± 8.0 P < .001 Extracranial disease status (col %) • PD 65.9% 65.7% 72.4% P < .001 • PR 8.0% 12.2% 13.4% • CR 26.1% 22.1% 14.2% LITV in cc (mean, SD) 14.0 ± 9.5 5.4 ± 2.9 1.0 ± 0.8 P < .001 CITV grouping >13cc 3.5-13cc <3.5cc Statistical significance Age in years (mean, SD) 64.3 ± 12.0 64.8 ± 11.1 64.3 ± 10.8 P = .843 KPS (mean, SD) 77.2 ± 15.0 84.2 ± 14.5 90.8 ± 11.7 P < .001 Number of metastases (mean, SD) 7.9 ± 10.3 6.7 ± 9.3 5.7 ± 8.0 P < .001 Extracranial disease status (col %) • PD 65.9% 65.7% 72.4% P < .001 • PR 8.0% 12.2% 13.4% • CR 26.1% 22.1% 14.2% LITV in cc (mean, SD) 14.0 ± 9.5 5.4 ± 2.9 1.0 ± 0.8 P < .001 aCITV, cumulative intracranial tumor volume; cc, cubic centimeters; SD, standard deviation; KPS, Karnofsky performance score; PD, progressive disease; PR, partial remission; CR, complete remission; LITV, largest intracranial tumor volume. View Large Following our univariable analyses, we performed a series of multivariable analyses to determine the degree of prognostic significance of CITV for patient survival in the context of covariates from the SIR model. The results of our first multivariable analysis—a linear regression analysis, using the outcome of patient survival defined in months alive—are shown in Table 5. As shown in this table, CITV was found to be significantly predictive of patient survival (in months) after accounting for patient age, KPS, number of metastases, and extracranial disease status. The AIC calculated for this was lower than that of a similar model in which LITV is used in place of CITV (25501 vs 25511), indicating an improved goodness of fit over the original SIR model. Finally, a third model, in which CITV and LITV were both included, found that CITV retained its prognostic significance (P < .001), while LITV dropped out of the model as a prognostic factor (P = .173). TABLE 5. Multivariable Analysis of Survival (in months), Demonstrating Statistical Significance of CITV as Prognostic Factor After Correcting for SIR Variables in a Point-Based SIR Modela Outcome = survival in months CITV-modified SIR SIR model SIR model + CITV Estimate Statistical Estimate Statistical Estimate Statistical coefficient significance coefficient significance coefficient significance Age grouping (per point increase) 0.91 P = .029 0.85 P = .029 0.91 P = .030 KPS grouping (per point increase) 5.1 P < .001 5.31 P < .001 5.02 P < .001 Number of metastases grouping (per point increase) 2.34 P < .001 2.69 P < .001 2.45 P < .001 Systemic disease status grouping (per point increase) 6.98 P < .001 6.97 P < .001 6.99 P < .001 LITV grouping (per point increase) 2.375 P < .001 0.85 P = .173 CITV grouping (per point increase) 2.44 P < .001 1.89 P < .001 Akaike Information Criterion (AIC) 25501 25511 25501 Outcome = survival in months CITV-modified SIR SIR model SIR model + CITV Estimate Statistical Estimate Statistical Estimate Statistical coefficient significance coefficient significance coefficient significance Age grouping (per point increase) 0.91 P = .029 0.85 P = .029 0.91 P = .030 KPS grouping (per point increase) 5.1 P < .001 5.31 P < .001 5.02 P < .001 Number of metastases grouping (per point increase) 2.34 P < .001 2.69 P < .001 2.45 P < .001 Systemic disease status grouping (per point increase) 6.98 P < .001 6.97 P < .001 6.99 P < .001 LITV grouping (per point increase) 2.375 P < .001 0.85 P = .173 CITV grouping (per point increase) 2.44 P < .001 1.89 P < .001 Akaike Information Criterion (AIC) 25501 25511 25501 aCITV, cumulative intracranial tumor volume; SIR, scored index for radiosurgery; KPS, Karnofsky performance score; LITV, largest intracranial tumor volume. View Large We performed a second multivariable analysis using a Cox proportional hazards model. The results of this analysis again showed that CITV remained a significant prognostic factor for survival after accounting for patient age, KPS, number of metastases, and extracranial disease status (Table 6). The AIC for our CITV model was lower than that of SIR (42428 vs 42445), indicating an improved goodness of fit over the original SIR model. Finally, when examining both CITV and LITV together in the same Cox model, only CITV retained its prognostic significance (P < .001), while LITV became statistically insignificant (P = .393). TABLE 6. Cox Proportional Hazards Analysis Demonstrating Prognostic Superiority of CITV Over LITV Within the SIR Modela Outcome = death (survival time in months) CITV-modified SIR SIR model SIR model + CITV Hazard Statistical Hazard Statistical Hazard Statistical ratio significance ratio significance ratio significance Age grouping (per point increase) 0.915 P = .001 0.917 P = .001 0.914 P = .001 KPS grouping (per point increase) 0.572 P < .001 0.559 P < .001 0.574 P < .001 Number of metastases grouping (per point increase) 0.872 P < .001 0.854 P < .001 0.869 P < .001 Systemic disease status grouping (per point increase) 0.681 P < .001 0.684 P < .001 0.681 P < .001 LITV grouping (per point increase) 0.854 P < .001 0.967 P = .393 CITV grouping (per point increase) 0.842 P < .001 0.860 P < .001 Akaike Information Criterion (AIC) 42428 42445 42429 Outcome = death (survival time in months) CITV-modified SIR SIR model SIR model + CITV Hazard Statistical Hazard Statistical Hazard Statistical ratio significance ratio significance ratio significance Age grouping (per point increase) 0.915 P = .001 0.917 P = .001 0.914 P = .001 KPS grouping (per point increase) 0.572 P < .001 0.559 P < .001 0.574 P < .001 Number of metastases grouping (per point increase) 0.872 P < .001 0.854 P < .001 0.869 P < .001 Systemic disease status grouping (per point increase) 0.681 P < .001 0.684 P < .001 0.681 P < .001 LITV grouping (per point increase) 0.854 P < .001 0.967 P = .393 CITV grouping (per point increase) 0.842 P < .001 0.860 P < .001 Akaike Information Criterion (AIC) 42428 42445 42429 aCITV, cumulative intracranial tumor volume; SIR, scored index for radiosurgery; KPS, Karnofsky performance score; LITV, largest intracranial tumor volume. View Large Taken together, these results suggest that CITV allows for finer survival discrimination within each value of LITV. To confirm this hypothesis, we generated an ROC for LITV and CITV as prognostic factors for 1 yr survival after SRS treatment in a univariable analysis (Figure 1B). Consistent with our hypothesis, the AUC for CITV (red) was greater than that of LITV (black), suggesting that CITV was a superior prognostic variable. Prognostic Value of LITV Relative to CITV in the SIR Model Finally, in order to directly compare the prognostic utility of CITV to LITV within the SIR model, we determined the prognostic value of the SIR in the TS cohort and compared this value to a model where LITV was replaced with CITV using the aforementioned statistical measures: NRI > 0 and IDI. For both measures, the prognostic value of the SIR model was significantly improved when LITV was replaced with CITV. The NRI > 0 was 0.2574 (95% confidence interval [CI] 0.1890-0.3257, P < .001) when comparing the SIR model where LITV was replaced with CITV (termed CITV-modified SIR model) relative to conventional SIR model (Table 7). IDI revealed an improvement of 0.0088 (95% CI 0.0057-0.0119, P < .001) when comparing the CITV-modified SIR model relative to the conventional model. These results both support the notion that CITV is a superior prognostic factor relative to LITV. TABLE 7. Prognostic Performance of the SIR and CITV-Modified SIR Models for TS and MC Cohortsa TS cohort Value 95% CI ΔAUC 0.0098 P = .001 Continuous NRI > 0 0.2574 0.1890-0.3257 (P < .001) IDI score 0.0088 0.0057-0.0119 (P < .001) MC cohort Value 95% CI ΔAUC 0.0093 P = .007 Continuous NRI > 0 0.2604 0.1796-0.3411 (P < .001) IDI score 0.0051 0.0030-0.0073 (P < .001) TS cohort Value 95% CI ΔAUC 0.0098 P = .001 Continuous NRI > 0 0.2574 0.1890-0.3257 (P < .001) IDI score 0.0088 0.0057-0.0119 (P < .001) MC cohort Value 95% CI ΔAUC 0.0093 P = .007 Continuous NRI > 0 0.2604 0.1796-0.3411 (P < .001) IDI score 0.0051 0.0030-0.0073 (P < .001) aΔAUC, area under curve; NRI, net reclassification improvement; IDI, integrated discrimination improvement. View Large Result Validation in an Independent Cohort To validate our findings, we analyzed LITV and CITV in an independent cohort of 2793 patients treated by MH and CC (termed MC cohort). These analyses showed results highly comparable to those observed in the TS cohort. As was previously seen in the TS cohort, the CITV was at least 50% greater than the LITV in ∼80% of the patients in the MC cohort (Figure 2A). Similarly, CITV was as high as 400% of LITV for 10% to 30% of the SRS-treated BM patients for each LITV category. Univariable analyses (Table A, Supplemental Digital Content) showed that, consistent with our TS cohort, univariable analyses of our MC cohort found that KPS, number of metastases, extracranial disease status, and LITV all differed for patients according to their CITV point grouping (ie, <3.5, 3.5-13, and >13 cc). CITV was better associated with survival than LITV in our multivariable analyses (results are shown in Tables B and C, Supplementary Digital Content). Further, our CITV-based ROC curve showed a greater AUC relative to our LITV-based ROC (Figure 2B). When LITV in the SIR was replaced with CITV, NRI > 0 showed an improvement of 0.2604 (95% CI 0.1796-0.3411, P < .001), and IDI showed an improvement of 0.0051 (95% CI 0.0030-0.0073, P < .001; Table 7). These results validate the findings observed in the TS cohort and support the superior prognostic value of CITV relative to LITV for SRS-treated brain metastasis patients. FIGURE 2. View largeDownload slide Discrepancy between CITV and LITV in the MC cohort. A, Histogram showing the number of patients (y-axis) with a given largest lesion size (x-axis). Relative increase of CITV to largest lesion was color coded such that increasing intensity reflects greater discrepancy between CITV and LITV. CITV was as high as 400% of LITV for 10% to 30% of the SRS-treated BM patients for each category of LITV (shown in 1 cc increments on the x-axis). B, ROC curve showing increased AUC for CITV relative to LITV for prognosticating 1-yr survival. Results were derived from the TS cohort. FIGURE 2. View largeDownload slide Discrepancy between CITV and LITV in the MC cohort. A, Histogram showing the number of patients (y-axis) with a given largest lesion size (x-axis). Relative increase of CITV to largest lesion was color coded such that increasing intensity reflects greater discrepancy between CITV and LITV. CITV was as high as 400% of LITV for 10% to 30% of the SRS-treated BM patients for each category of LITV (shown in 1 cc increments on the x-axis). B, ROC curve showing increased AUC for CITV relative to LITV for prognosticating 1-yr survival. Results were derived from the TS cohort. DISCUSSION Relevance of Findings We previously demonstrated the importance of CITV as a prognostic factor for SRS-treated BM patients18-20. Here we compared CITV and LITV17 in terms of their prognostic significance through a variety of statistical models, all of which demonstrate a superiority of CITV over LITV for predicting patient survival within the framework of the SIR model. Admittedly, the improvement in survival prognostication when comparing CITV and LITV was modest. However, this modest effect is to be expected since CITV includes LITV. The fundamental question is to what extent the tumor volume not accounted for by the LITV contributes to prognostication. Our study suggests these tumor volumes contribute to survival prognostication in a statistically significant albeit modest manner. To the best of our knowledge, this study is the first to address this issue. Our analysis indicates that ∼80% of the patients with comparable LITV exhibited significantly different CITV, suggesting that CITV offers a more granular stratification of the SRS-treated BM population. When LITV in the SIR scale was replaced with CITV, sensitivity and specificity of predicting 1-yr survival was significantly improved in 2 independent cohorts of ∼6000 SRS-treated BM patients. Impressively, the gain in prognostication, as measured by NRI > 0 in both cohorts, exceeds the Cohen criteria of significance11,18,30 and was on par with that observed for High-Density Lipoprotein (HDL) as a prognostic factor for myocardial infarction.31 In aggregate, these results suggest that the CITV-modified SIR model offers superior prognostic value for SRS-treated BM patients relative to the SIR model with LITV, a finding further supported by the failure of prior studies to find a significant prognostic value for LITV.16,31 Potential Explanations There are several potential explanations for the superior prognostic value of CITV relative to LITV. Clinically, the differing CITV in patients with comparable LITV may influence the maximal dose of radiation dose delivered, thereby influencing the likelihood of local control.32 For instance, in 2 patients with identical LITV, the patient with significantly higher CITV may have received lowered SRS dose due to dose constraints related to single-fraction treatment.33 Alternatively, the tumors with differing CITV may reflect an underlying biology that is not captured by LITV. For example, the sizes and size distribution of the intracranial metastases may serve as imaging proxy for the biology of the primary cancer. For instance, the natural history and therapeutic responses of primary tumors that give rise to BM consisting of 1 dominant lesion and 3 smaller lesions may be different from those that give rise to 4 similarly sized lesions. These hypotheses are not mutually exclusive and may both contribute to the observed results. Limitations There are several limitations to our study. First, our study is subject to all biases inherent in retrospective analysis. To mitigate these biases, we validated our observation using an independent cohort of patients.34 Second, the SIR model was developed without consideration for tumor histology.17,35 While examining CITV and LITV in this model afford direct comparison of these variables in an established prognostic scale,17 we were unable to examine the effects of histology using an SIR-based framework for our augmented model. Future studies to examine the impact of histology on the prognostic value of CITV are warranted. Third, we rely partly on statistical measures (namely NRI and IDI) which carry inherent limitations.36,37 While NRI and IDI metrics allow for direct quantifiable comparison of the 2 different prognostic scales at hand in this study (SIR and CITV-modified SIR), both measures require a binary outcome in order to compare the SIR to our CITV-modified SIR. That said, the results of our multivariable models (using survival as a continuous variable) and AIC calculations highlight the prognostic superiority of CITV over LITV. Finally, our study fails to consider the survival impact of molecularly targeted therapy, immunotherapy, or the genetic mutation profile of the tumor. To the extent, survival expectations are increasingly modified by these agents, future studies should be devoted to study the interaction between CITV and these agents. CONCLUSION In 2 independent patient cohorts totaling ∼6000 patients, CITV consistently outperforms LITV as a prognostic variable when applied in the context of the SIR model. Disclosures The project described was partially supported by the National Institutes of Health (NIH), Grant 1TL1TR001443. The content is solely the responsibility of the authors and does not necessarily represent the official views of the NIH. There are no other financial disclosures relevant to this manuscript. The authors have no personal, financial, or institutional interest in any of the drugs, materials, or devices described in this article. REFERENCES 1. Patchell RA. The management of brain metastases. Cancer Treat Rev . 2003; 29( 6): 533- 540. Google Scholar CrossRef Search ADS PubMed 2. Eichler AF, Loeffler JS. Multidisciplinary management of brain metastases. Oncologist . 2007; 12( 7): 884- 898. Google Scholar CrossRef Search ADS PubMed 3. Gaspar LE, Mehta MP, Patchell RA et al. The role of whole brain radiation therapy in the management of newly diagnosed brain metastases: a systematic review and evidence-based clinical practice guideline. J Neurooncol . 2010; 96( 1): 17- 32. Google Scholar CrossRef Search ADS PubMed 4. Gondi V, Wolfgang TA, Mehta MP. Why avoid the hippocampus? A comprehensive review. Radiother Oncol . 2010; 97( 3): 370- 376. Google Scholar CrossRef Search ADS PubMed 5. Wang CC, Floyd SR, Chang CH et al. Cyberknife hypofractionated stereotactic radiosurgery (HSRS) of resection cavity after excision of large cerebral metastasis: Efficacy and safety of an 800 cGy x 3 daily fractions regimen. J Neurooncol . 2012; 106( 3): 601- 610. Google Scholar CrossRef Search ADS PubMed 6. McDuff SG, Taich ZJ, Lawson JD et al. Neurocognitive assessment following whole brain radiation therapy and radiosurgery for patients with cerebral metastases. J Neurol Neurosurg Psychiatry . 2013; 84( 12): 1384- 1391. Google Scholar CrossRef Search ADS PubMed 7. Linskey ME, Andrews DW, Asher AL et al. The role of stereotactic radiosurgery in the management of patients with newly diagnosed brain metastases: a systematic review and evidence-based clinical practice guideline. J Neurooncol . 2010; 96( 1): 45- 68. Google Scholar CrossRef Search ADS PubMed 8. Sneed PK, Lamborn KR, Forstner JM et al. Radiosurgery for brain metastases: Is whole brain radiotherapy necessary? Int J Radiat Oncol Biol Phys . 1999; 43( 3): 549- 558. Google Scholar CrossRef Search ADS PubMed 9. Halasz LM, Weeks JC, Neville BA, Taback N, Punglia RS. Use of stereotactic radiosurgery for brain metastases from non-small cell lung cancer in the United States. Int J Radiat Oncol Biol Phys . 2013; 85( 2): e109- e116. Google Scholar CrossRef Search ADS PubMed 10. Preusser M, Capper D, Ilhan-Mutlu A et al. Brain metastases: pathobiology and emerging targeted therapies. Acta Neuropathol . 2012; 123( 2): 205- 222. Google Scholar CrossRef Search ADS PubMed 11. Cohen JV, Kluger HM. Systemic immunotherapy for the treatment of brain metastases. Front Oncol . 2016; 6: 49. doi:10.3389/fonc.2016.00049. Google Scholar CrossRef Search ADS PubMed 12. Nussbaum ES, Djalilian HR, Cho KH, Hall WA. Brain metastases: histology, multiplicity, surgery, and survival. Cancer . 1996; 78( 8): 1781- 1788. Google Scholar CrossRef Search ADS PubMed 13. Langer CJ, Mehta MP. Current management of brain metastases, with a focus on systemic options. J Clin Oncol . 2005; 23( 25): 6207- 6219. Google Scholar CrossRef Search ADS PubMed 14. Pollock BE, Brown PD, Foote RL, Stafford SL, Schomberg PJ. Properly selected patients with multiple brain metastases may benefit from aggressive treatment of their intracranial disease. [Review] [32 refs]. J Neurooncol . 2003; 61( 1): 73- 80. Google Scholar CrossRef Search ADS PubMed 15. Gaspar L, Scott C, Rotman M et al. Recursive partitioning analysis (RPA) of prognostic factors in three Radiation Therapy Oncology Group (RTOG) brain metastases trials. Int J Radiat Oncol Biol Phys . 1997; 37( 4): 745- 751. Google Scholar CrossRef Search ADS PubMed 16. Sperduto PW, Berkey B, Gaspar LE, Mehta M, Curran W. A new prognostic index and comparison to three other indices for patients with brain metastases: an analysis of 1,960 patients in the RTOG database. Int J Radiat Oncol Biol Phys . 2008; 70( 2): 510- 514. Google Scholar CrossRef Search ADS PubMed 17. Weltman E, Salvajoli JV, De Oliveira Borges SR, Toledo R, Anselmo R, Wajsbrot DB. Score index for stereotactic radiosurgery of brain metastases. J Radiosurgery . 1998; 1( 2): 89- 97. Google Scholar CrossRef Search ADS 18. Marcus LP, Marshall D, Hirshman BR et al. Cumulative intracranial tumor volume (CITV) enhances the prognostic value of the lung-specific graded prognostic assessment (GPA) model. Neurosurgery . 2016; 79( 2): 246- 252. doi:10.1227/NEU.0000000000001123. CrossRef Search ADS PubMed 19. Gonda DD, Kim TE, Goetsch SJ et al. Prognostic factors for stereotactic radiosurgery-treated patients with cerebral metastasis: implications on randomised control trial design and inter-institutional collaboration. Eur J Cancer . 2014; 50( 6): 1148- 1158. Google Scholar CrossRef Search ADS PubMed 20. Marshall DC, Marcus LP, Kim TE et al. Management patterns of patients with cerebral metastases who underwent multiple stereotactic radiosurgeries. J Neurooncol . 2016. doi:10.1007/s11060-016-2084-2. 21. Obuchowski NA, Lieber ML, Wians FH. ROC curves in clinical chemistry: uses, misuses, and possible solutions. Clin Chem . 2004; 50( 7): 1118- 1125. Google Scholar CrossRef Search ADS PubMed 22. Serizawa T, Higuchi Y, Nagano O et al. A new grading system focusing on neurological outcomes for brain metastases treated with stereotactic radiosurgery: the modified basic score for brain metastases. J Neurosurg . 2014; 121( suppl): 35- 43. Google Scholar PubMed 23. Nabors LB, Portnow J, Ammirati M et al. Central nervous system cancers, version 2.2014: featured updates to the NCCN guidelines. J Natl Compr Canc Netw . 2014; 12( 11): 1517- 1523. Google Scholar CrossRef Search ADS PubMed 24. Pan H, Cerviño LI, Pawlicki T et al. Frameless, real-time, surface imaging-guided radiosurgery: clinical outcomes for brain metastases. Neurosurgery . 2012; 71( 4): 844- 851. Google Scholar CrossRef Search ADS PubMed 25. Kaal EC, Niël CG, Vecht CJ. Therapeutic management of brain metastasis. Lancet Neurol . 2005; 4( 5): 289- 298. doi:10.1016/S1474-4422(05)70072-7. Google Scholar CrossRef Search ADS PubMed 26. Burnham KP, Anderson DR, eds. Model Selection and Multimodel Inference . New York, NY: Springer; 2004. doi:10.1007/b97636. Google Scholar CrossRef Search ADS 27. Rava P, Leonard K, Sioshansi S et al. Survival among patients with 10 or more brain metastases treated with stereotactic radiosurgery. J Neurosurg . 2013; 119( 2): 1- 6. Google Scholar CrossRef Search ADS 28. Joshi R, Johnson MD, Maitz A, Marvin KS, Olson RE, Grills IS. Utility of graded prognostic assessment in evaluation of patients with brainstem metastases treated with radiosurgery. Clin Neurol Neurosurg . 2016; 147: 30- 33. doi:10.1016/j.clineuro.2016.05.001. Google Scholar CrossRef Search ADS PubMed 29. Trifiletti DM, Lee C-C, Kano H et al. Stereotactic radiosurgery for brainstem metastases: an international cooperative study to define response and toxicity. Int J Radiat Oncol Biol Phys . 2016; 96( 2): 280- 288. Google Scholar CrossRef Search ADS PubMed 30. Pencina MJ, D’Agostino RB, Pencina KM, Janssens AC, Greenland P. Interpreting incremental value of markers added to risk prediction models. Am J Epidemiol . 2012; 176( 6): 473- 481. Google Scholar CrossRef Search ADS PubMed 31. Cooney MT, Dudina A, De Bacquer D et al. How much does HDL cholesterol add to risk estimation? A report from the SCORE investigators. Eur J Cardiovasc Prev Rehabil . 2009; 16( 3): 304- 314. Google Scholar CrossRef Search ADS PubMed 32. Shiau CY, Sneed PK, Shu HKG et al. Radiosurgery for brain metastases: relationship of dose and pattern of enhancement to local control. Int J Radiat Oncol Biol Phys . 1997; 37( 2): 375- 383. Google Scholar CrossRef Search ADS PubMed 33. Kim YJ, Cho KH, Kim JY et al. Single-dose versus fractionated stereotactic radiotherapy for brain metastases. Int J Radiat Oncol Biol Phys . 2011; 81( 2): 483- 489. Google Scholar CrossRef Search ADS PubMed 34. Toll DB, Janssen KJM, Vergouwe Y, Moons KGM. Validation, updating and impact of clinical prediction rules: a review. J Clin Epidemiol . 2008; 61( 11): 1085- 1094. Google Scholar CrossRef Search ADS PubMed 35. Sperduto PW, Chao ST, Sneed PK et al. Diagnosis-specific prognostic factors, indexes, and treatment outcomes for patients with newly diagnosed brain metastases: a multi-institutional analysis of 4,259 patients. Int J Radiat Oncol Biol Phys . 2010; 77( 3): 655- 661. Google Scholar CrossRef Search ADS PubMed 36. Kerr KF, Wang Z, Janes H, McClelland RL, Psaty BM, Pepe MS. Net reclassification indices for evaluating risk prediction instruments: a critical review. Epidemiology . 2014; 25( 1): 114- 121. Google Scholar CrossRef Search ADS PubMed 37. Hilden J, Gerds TA. A note on the evaluation of novel biomarkers: do not rely on integrated discrimination improvement and net reclassification index. Stat Med . 2014; 33( 19): 3405- 3414. Google Scholar CrossRef Search ADS PubMed Supplemental digital content is available for this article at www.neurosurgery-online.com. Copyright © 2017 by the Congress of Neurological Surgeons

Neurosurgery – Oxford University Press

**Published: ** Apr 1, 2018

Loading...

personal research library

It’s your single place to instantly

**discover** and **read** the research

that matters to you.

Enjoy **affordable access** to

over 18 million articles from more than

**15,000 peer-reviewed journals**.

All for just $49/month

Query the DeepDyve database, plus search all of PubMed and Google Scholar seamlessly

Save any article or search result from DeepDyve, PubMed, and Google Scholar... all in one place.

Get unlimited, online access to over 18 million full-text articles from more than 15,000 scientific journals.

Read from thousands of the leading scholarly journals from *SpringerNature*, *Elsevier*, *Wiley-Blackwell*, *Oxford University Press* and more.

All the latest content is available, no embargo periods.

## “Hi guys, I cannot tell you how much I love this resource. Incredible. I really believe you've hit the nail on the head with this site in regards to solving the research-purchase issue.”

Daniel C.

## “Whoa! It’s like Spotify but for academic articles.”

@Phil_Robichaud

## “I must say, @deepdyve is a fabulous solution to the independent researcher's problem of #access to #information.”

@deepthiw

## “My last article couldn't be possible without the platform @deepdyve that makes journal papers cheaper.”

@JoseServera

DeepDyve ## Freelancer | DeepDyve ## Pro | |
---|---|---|

Price | FREE | $49/month |

Save searches from | ||

Create lists to | ||

Export lists, citations | ||

Read DeepDyve articles | Abstract access only | Unlimited access to over |

20 pages / month | ||

PDF Discount | 20% off | |

Read and print from thousands of top scholarly journals.

System error. Please try again!

or

By signing up, you agree to DeepDyve’s Terms of Service and Privacy Policy.

Already have an account? Log in

Bookmark this article. You can see your Bookmarks on your DeepDyve Library.

To save an article, **log in** first, or **sign up** for a DeepDyve account if you don’t already have one.

All DeepDyve websites use cookies to improve your online experience. They were placed on your computer when you launched this website. You can change your cookie settings through your browser.

ok to continue