Abstract
Purpose: An estimated 24%–45% of patients with cancer develop brain metastases. Individualized estimation of survival for patients with brain metastasis could be useful for counseling patients on clinical outcomes and prognosis. Methods: De-identified data for 2367 patients with brain metastasis from 7 Radiation Therapy Oncology Group randomized trials were used to develop and internally validate a prognostic nomogram for estimation of survival among patients with brain metastasis. The prognostic accuracy for survival from 3 statistical approaches (Cox proportional hazards regression, recursive partitioning analysis [RPA], and random survival forests) was calculated using the concordance index. A nomogram for 12-month, 6-month, and median survival was generated using the most parsimonious model. Results: The majority of patients had lung cancer, controlled primary disease, no surgery, Karnofsky performance score (KPS) ≥ 70, and multiple brain metastases and were in RPA class II or had a Diagnosis-Specific Graded Prognostic Assessment (DS-GPA) score of 1.25–2.5. The overall median survival was 136 days (95% confidence interval, 126–144 days). We built the nomogram using the model that included primary site and histology, status of primary disease, metastatic spread, age, KPS, and number of brain lesions. The potential use of individualized survival estimation is demonstrated by showing the heterogeneous distribution of the individual 12-month survival in each RPA class or DS-GPA score group. Conclusion: Our nomogram provides individualized estimates of survival, compared with current RPA and DS-GPA group estimates. This tool could be useful for counseling patients with respect to clinical outcomes and prognosis.
Keywords: brain metastases, nomogram, prediction, prognosis, survival
An estimated 24%–45% of all patients with cancer develop brain metastases every year.1,2 The most common primary tumors that metastasize to the brain are melanoma, lung, breast, and renal cancer.2,3 The current standard of care for patients with brain metastasis consists of whole-brain radiotherapy (WBRT) and/or stereotactic radiosurgery and/or surgery.1,2 However, controversy still exists regarding selection of the optimal treatment regimen.
Gaspar et al. evaluated prognostic factors associated with survival among patients with brain metastasis using recursive partitioning analysis (RPA)4,5 and found that age, Karnofsky performance score (KPS), status of primary disease, and metastatic spread formed 3 classes with distinct survival times. Although these classes have been validated in successive studies, Sperduto et al. developed a new prognostic scoring system called the Graded Prognostic Assessment (GPA)6–8 and found that age, KPS, number of brain metastases, and presence/absence of extracranial metastasis accurately estimated prognosis. The GPA was then extended to include diagnosis-specific information (diagnosis-specific GPA [DS-GPA]), which improved its accuracy for estimating prognosis.9 Although these classification systems have been validated in successive studies,10–14 RPA and DS-GPA provide group rather than individualized estimates; all patients in a class have the same clinical characteristics and, therefore, the same survival.
Additional prognostic scores have been developed for patients with brain metastasis. For those undergoing radiosurgery, 2 progonsotic systems have been developed: the score index for radiosurgery (SIR) and the basic score for brain metastasis (BS-BM). Lorenzoni et al. showed that these were more accurate than RPA in predicting survival in this patient population.15,16 The interval from tumor diagnosis to WBRT was the main predictor in the intracerebral control and survival rate scoring index of Rades et al.17,18
Individualized estimation of survival among patients with brain metastasis could be useful for counseling patients on treatment decisions and optimizing therapeutic approaches. These estimates are most useful at the time of diagnosis. Nomograms are often used for estimation of individualized cancer prognosis,19,20 but this approach has not been applied to brain metastasis. The purpose of this study was to develop and internally validate a nomogram for estimation of individualized survival probabilities for patients with brain metastasis with use of data from the Radiation Therapy Oncology Group (RTOG) that could be applied at the time of initial diagnosis.
Methods
Study Population
De-identified data were provided by RTOG for 7 brain metastasis randomized clinical trials spanning over 20 years (as approved by the Case Cancer Institutional Review Board): 0118,21 0119,22 79–16,23 85–28,24 89–05,25 91–04,26 and 95–08.27 These data included information on a total of 2367 patients. After removing individuals with missing information (n = 17), 2350 patients were available for analysis.
Variables of Interest
For each patient, the following variables were obtained: primary site (breast, lung, skin-melanoma, gastrointestinal, renal, other), histology (adenocarcinoma, large cell, small cell, non–small cell, squamous cell, other), status of primary disease (controlled, uncontrolled), metastatic spread (brain only, brain and other sites), surgery for brain metastasis (partial or complete/gross total resection, biopsy, none), age, KPS (≥70, <70), RTOG RPA class (I, II, III), DS-GPA score group (0–1, 1.25–2.5, 2.75–3, 3.25–4) and number of brain lesions (single, multiple, actual number of lesions). Patient self-reported history of headaches, history of seizures, and history of visual disturbance were also assessed.
Statistical Analysis
Missing variable values were imputed by the method of chained equations (MICE28), which uses a regression model for each variable with missing data, using all other variables as predictors, in an iterative fashion until all missing values are predicted.
Kaplan-Meier was used to estimate overall survival, defined as time to death from any cause. Survival was further assessed with adjustment for variables of interest using Cox proportional hazards regression,29 RPA,30 and random survival forests (RSF).31,32 In the Cox model, continuous and ordinal variables were modeled with restricted cubic splines to relax the linearity assumption. In RPA, a full recursive classification tree was created and the least important splits were eliminated recursively to obtain the final subclassification tree, which minimized the model deviance. In RSF, 1000 trees were grown, terminal nodes were set to 20, and all possible split points for each variable were examined using RSF to find the optimal split solution. All other RSF arguments were left at default values.
For each statistical approach, 2 models were fit: the full model, which included all variables of interest, and the reduced model, which included all variables of interest except symptoms. Models were internally validated via 10-fold cross-validation, and individual predicted 12-month survival probabilities were generated to quantify predictive accuracy and calibration. The predictive accuracy for overall survival of the 3 statistical approaches for each of the 2 models (full and reduced) was calculated using the concordance index (CI), which is equivalent to the area under the receiver-operating characteristic curve for censored data ranging from 0.5 to 1. The significance levels for the differences in CIs of competing statistical approaches were produced from bootstrap. A final nomogram was built using the statistical model with the highest predictive accuracy for individualized estimation of survival based on the CI. Calibration of the final model was visually assessed as follows: all patients were assigned into quintiles of the nomogram-predicted 12-month survival probabilities, and within each quintile, the mean nomogram predicted 12-month survival probability was plotted against the Kaplan-Meier estimated 12-month survival. The accuracy for estimating survival of the current RPA and DS-GPA classes was directly compared with the nomogram. For some patients (n = 753), the number of lesions was recorded as >1 only; to calculate the DS-GPA score in this group, the score was calculated assuming 2–3 lesions and then assuming >3 lesions. These 2 scores were averaged, providing a final DS-GPA score for these patients. All analyses were performed using R, version 2.10.0 (http://www.r-project.org/).
Results
Of the 2350 patients available for study, 2273 were deceased (96.7%). Characteristics of these patients are shown in Table 1 along with the number of patients with missing information by variable of interest. The mean age across studies is similar. The majority of patients had primary lung cancer (66%). Most patients had controlled primary disease, no surgery, KPS ≥ 70, and multiple brain metastases. Thus, most were classified as RPA class II and had a DS-GPA score of 1.25–2.5 (Table 1). In addition, the vast majority of patients did not undergo surgery (data not shown); thus, this variable was not tested in the models. The overall median survival was 136 days (∼4.5 months) after randomization (95% confidence interval [CI], 126–144 days) (Fig. 1).
Table 1.
Overall characteristics of RTOG patients with brain metastasis
| RTOG Study Number and sample size |
|||||||||
|---|---|---|---|---|---|---|---|---|---|
| Variables of Interest (column %) | Overall total n = 2350 | 118 n = 183 | 119 n = 129 | 7916 n = 848 | 8528 n = 343 | 8905 n = 72 | 9104 n = 444 | 9508 n = 331 | |
| Average age (standard deviation) (N missing = 18) | 58.5 (11.1) | 57.9 (10.8) | 61.7 (12.2) | 56.8 (10.6) | 58.8 (10.8) | 60.0 (10.4) | 59.9 (11.0) | 59.4 (12.3) | |
| Primary Site and Histology (N missing = 147) | Breast-Adenocarcinoma | 207 (8.8%) | 20 (10.9%) | 7 (5.4%) | 90 (10.6%) | 32 (9.3%) | 2 (2.8%) | 37 (8.3%) | 19 (5.8%) |
| Breast-Other | 84 (3.6%) | 14 (7.7%) | 6 (4.7%) | 24 (2.8%) | 15 (4.4%) | 1 (1.4%) | 11 (2.5%) | 13 (3.9%) | |
| Lung-Adenocarcinoma | 674 (28.7%) | 50 (27.3%) | 41 (31.8%) | 212 (25.0%) | 99 (28.9%) | 28 (38.9%) | 136 (30.6%) | 108 (32.7%) | |
| Lung-Large cell | 176 (7.5%) | 10 (5.5%) | 5 (3.9%) | 65 (7.7%) | 35 (10.2%) | 5 (6.9%) | 38 (8.6%) | 18 (5.5%) | |
| Lung-Other | 206 (8.8%) | 23 (12.6%) | 23 (17.8%) | 50 (5.9%) | 19 (5.5%) | 12 (16.7%) | 40 (9.0%) | 39 (11.8%) | |
| Lung-Small cell | 150 (6.4%) | 18 (9.8%) | 17 (13.2%) | 33 (3.9%) | 15 (4.4%) | 3 (4.2%) | 40 (9.0%) | 24 (7.3%) | |
| Lung-Squamous cell | 349 (14.9%) | 13 (7.1%) | 6 (4.7%) | 189 (22.3%) | 41 (12.0%) | 8 (11.1%) | 59 (13.3%) | 33 (10.0%) | |
| Other-Adenocarcinoma | 79 (3.4%) | 4 (2.2%) | 11 (8.5%) | 27 (3.2%) | 8 (2.3%) | 0 (0.0%) | 11 (2.5%) | 18 (5.5%) | |
| Other-GI | 94 (4.0%) | 2 (1.1%) | 0 (0.0%) | 40 (4.7%) | 18 (5.2%) | 3 (4.2%) | 15 (3.4%) | 16 (4.8%) | |
| Other-Renal | 43 (1.8%) | 4 (2.2%) | 0 (0.0%) | 13 (1.5%) | 10 (2.9%) | 1 (1.4%) | 8 (1.8%) | 7 (2.1%) | |
| Other-Squamous cell | 27 (1.1%) | 1 (0.5%) | 1 (0.8%) | 13 (1.5%) | 5 (1.5%) | 0 (0.0%) | 5 (1.1%) | 2 (0.6%) | |
| Skin-Melanoma | 136 (5.8%) | 18 (9.8%) | 4 (3.1%) | 34 (4.0%) | 32 (9.3%) | 6 (8.3%) | 28 (6.3%) | 14 (4.2%) | |
| Other | 124 (5.3%) | 6 (3.3%) | 8 (6.2%) | 58 (6.8%) | 14 (4.1%) | 3 (4.2%) | 16 (3.6%) | 19 (5.8%) | |
| Status of Primary Disease (N missing = 131) | Tumor controlled | 1514 (64.4%) | 125 (68.3%) | 43 (33.3%) | 434 (51.2%) | 254 (74.1%) | 68 (94.4%) | 326 (73.4%) | 264 (79.8%) |
| Tumor uncontrolled | 836 (35.6%) | 58 (31.7%) | 86 (66.7%) | 414 (48.8%) | 89 (25.9%) | 4 (5.6%) | 118 (26.6%) | 67 (20.2%) | |
| Metastatic Spread (N missing = 18) | Brain alone | 1411 (60.0%) | 63 (34.4%) | 46 (35.7%) | 460 (54.2%) | 250 (72.9%) | 72 (100.0%) | 320 (72.1%) | 200 (60.4%) |
| Brain & other sites | 939 (40.0%) | 120 (65.6%) | 83 (64.3%) | 388 (45.8%) | 93 (27.1%) | 0 (0.0%) | 124 (27.9%) | 131 (39.6%) | |
| History of Headache (N missing = 333) | Not present | 1356 (57.7%) | 91 (49.7%) | 79 (61.2%) | 467 (55.1%) | 211 (61.5%) | 34 (47.2%) | 264 (59.5%) | 210 (63.4%) |
| Mild | 519 (22.1%) | 55 (30.1%) | 26 (20.2%) | 204 (24.1%) | 62 (18.1%) | 10 (13.9%) | 97 (21.8%) | 65 (19.6%) | |
| Moderate | 358 (15.2%) | 23 (12.6%) | 21 (16.3%) | 137 (16.2%) | 51 (14.9%) | 22 (30.6%) | 53 (11.9%) | 51 (15.4%) | |
| Severe | 117 (5.0%) | 14 (7.7%) | 3 (2.3%) | 40 (4.7%) | 19 (5.5%) | 6 (8.3%) | 30 (6.8%) | 5 (1.5%) | |
| History of Seizures | |||||||||
| (N missing = 1169) | None | 1825 (77.7%) | 144 (78.7%) | 93 (72.1%) | 543 (64.0%) | 294 (85.7%) | 64 (88.9%) | 397 (89.4%) | 290 (87.6%) |
| <1 per week | 238 (10.1%) | 8 (4.4%) | 16 (12.4%) | 96 (11.3%) | 38 (11.1%) | 6 (8.3%) | 33 (7.4%) | 41 (12.4%) | |
| ≥1 per week | 287 (12.2%) | 31 (16.9%) | 20 (15.5%) | 209 (24.6%) | 11 (3.2%) | 2 (2.8%) | 14 (3.2%) | 0 (0.0%) | |
| History of Visual Disturbance (N missing = 1171) | None | 1763 (75.0%) | 138 (75.4%) | 95 (73.6%) | 533 (62.9%) | 280 (81.6%) | 61 (84.7%) | 367 (82.7%) | 289 (87.3%) |
| Acuity decreases | 262 (11.1%) | 33 (18.0%) | 11 (8.5%) | 121 (14.3%) | 25 (7.3%) | 7 (9.7%) | 38 (8.6%) | 27 (8.2%) | |
| Field defect | 126 (5.4%) | 5 (2.7%) | 12 (9.3%) | 49 (5.8%) | 25 (7.3%) | 2 (2.8%) | 23 (5.2%) | 10 (3.0%) | |
| Other | 199 (8.5%) | 7 (3.8%) | 11 (8.5%) | 145 (17.1%) | 13 (3.8%) | 2 (2.8%) | 16 (3.6%) | 5 (1.5%) | |
| Karnofsky Performance Score (KPS) (N missing = 0) | ≥70 | 1947 (82.9%) | 180 (98.4%) | 129 (100.0%) | 450 (53.1%) | 342 (99.7%) | 72 (100.0%) | 443 (99.8%) | 331 (100.0%) |
| <70 | 403 (17.1%) | 3 (1.6%) | 0 (0.0%) | 398 (46.9%) | 1 (0.3%) | 0 (0.0%) | 1 (0.2%) | 0 (0.0%) | |
| Number of Brain Lesions (N missing = 978) | Single | 888 (37.8%) | 7 (3.8%) | 0 (0.0%) | 374 (44.1%) | 158 (46.1%) | 34 (47.2%) | 128 (28.8%) | 187 (56.5%) |
| Multiple | 1462 (62.2%) | 176 (96.2%) | 129 (100.0%) | 474 (55.9%) | 185 (53.9%) | 38 (52.8%) | 316 (71.2%) | 144 (43.5%) | |
| 1 only | 888 (37.8%) | 7 (3.8%) | 0 (0.0%) | 374 (44.1%) | 158 (46.1%) | 34 (47.2%) | 128 (28.8%) | 187 (56.5%) | |
| >1 | 753 (32.0%) | 146 (79.8%) | 101 (78.3%) | 1 (0.1%) | 185 (53.9%) | 4 (5.6%) | 316 (71.2%) | 0 (0.0%) | |
| 2 or 3 | 515 (21.9%) | 30 (16.4%) | 27 (20.9%) | 295 (34.8%) | 0 (0.0%) | 19 (26.4%) | 0 (0.0%) | 144 (43.5%) | |
| >3 | 194 (8.3%) | 0 (0.0%) | 1 (0.8%) | 178 (21.0%) | 0 (0.0%) | 15 (20.8%) | 0 (0.0%) | 0 (0.0%) | |
| RPA Class | Class I | 485 (20.6%) | 39 (21.3%) | 0 (0.0%) | 85 (10.0%) | 104 (30.3%) | 39 (54.2%) | 124 (27.9%) | 94 (28.4%) |
| Class II | 1462 (62.2%) | 141 (77.0%) | 129 (100.0%) | 365 (43.0%) | 238 (69.4%) | 33 (45.8%) | 319 (71.8%) | 237 (71.6%) | |
| Class III | 403 (17.1%) | 3 (1.6%) | 0 (0.0%) | 398 (46.9%) | 1 (0.3%) | 0 (0.0%) | 1 (0.2%) | 0 (0.0%) | |
| DS-GPA Score Group | 0–1 | 240 (11.3%) | 22 (12.8%) | 19 (17.4%) | 175 (23.2%) | 6 (1.9%) | 1 (1.4%) | 11 (2.7%) | 6 (2.0%) |
| 1.25–2.5 | 1174 (55.2%) | 115 (66.9%) | 80 (73.4%) | 342 (45.4%) | 191 (60.1%) | 38 (55.1%) | 264 (64.1%) | 144 (49.1%) | |
| 2.75–3.0 | 355 (16.7%) | 15 (8.7%) | 5 (4.6%) | 113 (15.0%) | 66 (20.8%) | 19 (27.5%) | 72 (17.5%) | 65 (22.2%) | |
| 3.25–4 | 357 (16.8%) | 20 (11.6%) | 5 (4.6%) | 123 (16.3%) | 55 (17.3%) | 11 (15.9%) | 65 (15.8%) | 78 (26.6%) | |
Fig. 1.
Overall Kaplan-Meier observed survival for all RTOG patients with brain metastasis with 95% CI lines. The dotted lines above and below solid line are the 95% CI lines. Numbers above the x axis are number of patients at risk at each time point.
For both the full and reduced models, the Cox analysis (CIs: 0.594 full and 0.604 reduced) outperformed RPA (CIs: 0.574 full and 0.578 reduced) and RSF (RSF CIs: 0.587 full and 0.587 reduced). Compared with either the current RPA classes (CI, 0.553) or the DS-GPA score groups (CI, 0.579), the full or reduced Cox models have higher CIs (P = .001 for Cox vs RPA and Cox vs DS-GPA).
Because the reduced Cox proportional hazards model had a better CI than the full Cox model, we built the nomogram to estimate 6- and 12-month survival probabilities and median survival time for the reduced model. Therefore, the nomogram included the following: primary site and histology, status of primary disease, metastatic spread, age, KPS, and number of brain lesions (Fig. 2). The incremental accuracy for estimating survival for each of the variables is shown by the decrease in the CI when the predictor variable is removed from the existing model: age (0.018), status of primary disease (0.013), metastatic spread (0.013), primary site & histology (0.010), number of brain lesions (0.0031), and KPS (<0.001). The impact of study registration year was also tested and did not have any effect on the CI. The final Cox model was well calibrated (Fig. 3); for each quintile group, the estimated versus observed 12-month survival probabilities intersected the 45-degree line, demonstrating that the predicted value approximated the observed value within a 95% confidence interval.
Fig. 2.
Nomogram for 6- and 12-month survival and median survival for RTOG brain metastases patients. BA, Breast and Adenocarcinoma; BO, Breast and Other; LA, Lung and Adenocarcinoma; LL, Lung and Large cell; LO, Lung and Other; LSM, Lung and Small cell; LSQ, Lung and Squamous cell; OA, Other and Adenocarcinoma; OG, Other and GI; OR, Other and Renal; OSQ, Other and Squamous cell; SMM, Skin-Melanoma; OO, Other and Other; PR, Partial Resection; CGTR, Complete/Gross total resection.
Fig. 3.
Calibration curve based on 12-month survival prediction. Patients were grouped into quintiles of the predicted 12-month survival probabilities. The vertical bars represent 95% confidence intervals.
We also compared the estimations from the nomogram with those from the RPA classes (Fig. 4)4,5 or the DS-GPA score groups (Fig. 5).9 We found that the nomogram-estimated 12-month survival probabilities are heterogeneously distributed within each RPA class or DS-GPA score group. These results indicate the potential use of individualized prognostic estimation obtained from the nomogram versus the group-wise prognostic estimations obtained from either the RPA or the DS-GPA.
Fig. 4.
Distribution of nomogram 12-month survival probabilities within each RPA class. Values in parentheses indicate the number of patients. Class I: KPS ≥ 70, primary tumor controlled, age <65 years and brain metastasis only; class III: KPS < 70; class II: all others, including KPS ≥ 70 and primary uncontrolled, KPS ≥ 70 primary controlled and age ≥ 65, and KPS ≥ 70 primary controlled, age ≥ 65 and brain and other metastases.
Fig. 5.
Distribution of nomogram 12-month survival probabilities within each DS-GPA class. Values in parentheses indicate the number of patients in each DS-GPA class.
Discussion
The objective of this analysis was to develop and internally validate an individualized prognostic nomogram for patients with brain metastasis. We found that the best fitting model was the Cox proportional hazards survival model and included primary site and histology, status of primary disease, metastatic spread, age, KPS, and number of brain lesions. Thus, the nomogram was built using this model (Fig. 2).
The RPA brain metastasis article5 used data from 3 RTOG trials (79–16, 85–28, and 89–05) and was based on KPS, status of primary disease, age, and metastatic spread. These were then validated using data from 91–04 and compared with the RTOG historical database.4 We showed that, although the original 4 variables included in the Gaspar RPA classes were important predictors, primary site and histology and number of brain lesions were also important for accuracy of survival estimation. In comparison with other variables, KPS provided the least amount of incremental accuracy for estimating survival, which might be accounted for by bias and interobserver variability.
The original RPA class system has been externally validated using other groups of patients with brain metastasis. Agboola et al.11 applied these classes to patients with brain metastases who had undergone resection and WBRT. They found that survival was significantly associated with RPA class assignment. Two additional studies in patients with brain metastases from renal cell carcinoma12 and small cell lung cancer13 have also validated this finding. In addition, Sanghavi et al.14 showed improvement in survival with radiosurgery for all RPA classes, but each RPA class had separate survival from the other classes. However, in most of these studies, the majority of patients tended to fall primarily into class II, which is defined by lack of characteristics of class I or class III. Thus, approaches to studying subclassification of patients in these RPA classes are warranted.
The original GPA external validation analysis by Sperduto et al. used data from 5 RTOG trials: 79–16, 85–28, 89–05, and 95–08. The GPA scores were based on age, KPS, number of brain metastases, and presence/absence of extracranial metastases.6 Recently, these scores have been extended to include information about primary site and histology (DS-GPA)9 that used data from 4259 patients. In that study, the significant prognostic factors for non–small cell and small cell lung cancer were age, KPS, presence/absence of extracranial metastases, and the number of brain metastases. The significant prognostic factors for melanoma and renal cell carcinoma were KPS and the number of brain metastases, whereas for gastrointestinal cancers, only KPS was significant. For breast cancer, KPS, breast tumor subtype, and age were significant.9,33 We also show that primary site and histology are important factors. We compared our nomogram estimations with DS-GPA9 and showed that, within each DS-GPA class, there was a wide range of individualized survival probability. However, because 224 patients had sites/histologies that were not included in the DS-GPA calculation, they were not used for the comparison of nomogram with DS-GPA. In addition, because the level of detail about number of brain lesions varied between the clinical trials used, our nomogram results differed from the DS-GPA, because the DS-GPA showed that different prognostic factors were significant for different diagnoses and that this nomogram did not take that possibility fully into account.
The SIR and BS-BM scoring systems have been shown to predict survival among patients with brain metastasis undergoing radiosurgery. Of the 7 RTOG trials, only 1 had a study arm in which patients received radiosurgery. All patients received whole brain radiation. Therefore, a comparison of our nomogram to those systems would be inappropriate. The scoring system developed by Rades et al. is based on the time from diagnosis to initiation of WBRT. This interval was uniform in patients who participated in the RTOG trials because this was one of the elements of the study design for these protocols. Applying this score to this study population would not be informative.
Our nomogram, which included additional variables, allows for individualized estimation of 6- and 12-month survival probabilities and median survival. Our analysis included one of the largest number of patients with brain metastases published to date, with 2350 patients, thereby allowing us the statistical power to examine more variables in our models than could previous models with smaller sample sizes. Our comparison of estimations from the Cox model with those from either RPA or RSF highlighted the potential to tailor predictions to individual patients rather than groups of patients. The RSF estimations were quite different from the Cox model, suggesting that the improvement in accuracy of the Cox model is relevant. Although the Cox model was less than perfect, it remained the most accurate when tested in our data.
This nomogram (Fig. 2) can be readily used in clinical practice to provide the patient and their health care team an individualized survival estimate. For each patient, the health care team member records the aforementioned variables. One then draws a line straight upward to the “Points” axis to determine how many points the patient is allocated for each variable. The sum of the individual points for these variables would then be located on the “Total Points axis.” A line is drawn straight down to the relevant estimated survival information. This information could then be shared with the patient and could be used to help personalize that patient's care by informing specific treatment decisions over others. For example, if we had a patient with squamous cell lung cancer (57 points), controlled primary disease (0 points), metastatic spread to the brain alone (0 points), aged 80 years (74 points), KPS ≥ 70 (0 points), and only one brain lesion (0 points), their total would be 131 points. Their estimated 6- and 12-month survival probability would be 25% and 8%, respectively, and their projected median survival would be 95 days. To facilitate clinical use of this nomogram, we plan to provide free software for its implementation at http://rcalc.ccf.org, under the category “Brain Cancer”.
This study has several limitations. Our data represent a large group of patients with brain metastasis from 7 clinical trials conducted at many sites, and therefore, additional data on the patients’ other therapies were not available. Many of the studies incorporated WBRT but not radiosurgery and recent advances in systemic therapies for primary cancers; thus, these data may underestimate survival. Because all these patients developed brain metastases, information from these therapies perhaps might not be significantly deterministic for survival estimation. However, recent improvements in systemic chemotherapy for primary breast, lung, and renal cancers and melanoma would not be reflected in this analysis, because the most recent trial used in this analysis began in 1995. The vast majority of patients in this study had lung cancer (66%), which may limit its applicability to other tumors. Although lung cancer is the most common source of brain metastasis, different primary tumors may behave differently. However, other types of primaries were included in this analysis, although at smaller proportions. The information available on the exact number of brain lesions was clinical trial dependent; thus, for the final nomogram model, the number of brain lesions is categorized as single or multiple to make full use of the available data (79–16 and 89–05 the exact number of brain lesions was collected, 85–28 and 91–04 only single or multiple was collected, 95–08 1, 2 or 3 brain lesions was collected, 0118 1, 2, 3 or other brain lesions was collected and 0119 this information was not collected). We readily acknowledge that patients with a varying number of total brain metastases have differing prognoses.9,33 Advances in neuroimaging have improved the ability to detect smaller metastatic lesions, which may impact the final nomogram score. To use all available data, missing values in variables of interest were imputed using a widely used multiple imputation procedure,24,34 which has been shown to be an effective method for robustly estimating missing values, such that study inference is not biased.35–38 Multiple imputation uses the best available information to impute multiple plausible values, then averages these values and adjusts the variance accordingly; thus, study power is increased because all patient data can be used. Lastly, our nomogram is built and validated internally, although we did 10-fold cross validation to control the over-fitting bias. Therefore, this model should ideally be externally validated by additional independent data. However, the methods of nomogram construction that we used have been validated for other diseases using single institutional data.39–44
In conclusion, we have developed a tool for assessing individualized survival estimates for patients with brain metastasis. The nomogram provides an individualized estimate of survival, rather than a group estimate. This tool should be useful to patients and health care teams for counseling patients with respect to prognosis. Future directions include external validation in a new dataset and expanding its development in patient cohorts with much greater rates of resection and/or radiosurgery.
Funding
This work was supported by the Radiation Therapy Oncology Group (U10 CA21661) and CCOP (U10 CA37422) from the National Cancer Institute, the Case Comprehensive Cancer Center Core (NCI P30-CA043703 to J.S.B-S., A.E.S., M.M., and M.W.K.), and and the Peter D. Cristal Chair in Neurological Surgery, Case Western Reserve University School of Medicine to A.E.S. No author received an honorarium or other form of financial support related to the development of this manuscript.
Acknowledgments
We thank the RTOG Brain and Publications Committees for graciously providing the data needed for this work. This manuscript's contents are solely the responsibility of the authors and do not necessarily represent the official views of the National Cancer Institute.
Conflict of interest statement. All authors: No reported conflicts.
References
- 1.Eichler AF, Loeffler JS. Multidisciplinary management of brain metastases. Oncologist. 2007;12:884–898. doi: 10.1634/theoncologist.12-7-884. doi:10.1634/theoncologist.12-7-884. [DOI] [PubMed] [Google Scholar]
- 2.Sawaya R, Bindal RK, Lang FF, Abi-Said D. Metastatic brain tumors. In: Kaye AH, Laws ER, editors. Brain Tumors An Encyclopedic Approach. 2nd ed. London: Churchill Livingstone; 2001. pp. 999–1026. [Google Scholar]
- 3.Barnholtz-Sloan JS, Sloan AE, Davis FG, Vigneau FD, Lai P, Sawaya RE. Incidence proportions of brain metastases in patients diagnosed (1973 to 2001) in the Metropolitan Detroit Cancer Surveillance System. J Clin Oncol. 2004;22:2865–2872. doi: 10.1200/JCO.2004.12.149. doi:10.1200/JCO.2004.12.149. [DOI] [PubMed] [Google Scholar]
- 4.Gaspar LE, Scott C, Murray K, Curran W. Validation of the RTOG recursive partitioning analysis (RPA) classification for brain metastases. Int J Radiat Oncol Biol Phys. 2000;47:1001–1006. doi: 10.1016/s0360-3016(00)00547-2. doi:10.1016/S0360-3016(00)00547-2. [DOI] [PubMed] [Google Scholar]
- 5.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;37:745–751. doi: 10.1016/s0360-3016(96)00619-0. doi:10.1016/S0360-3016(96)00619-0. [DOI] [PubMed] [Google Scholar]
- 6.Sperduto CM, Watanabe Y, Mullan J, et al. A validation study of a new prognostic index for patients with brain metastases: the Graded Prognostic Assessment. J Neurosurg. 2008;109(Suppl.):87–89. doi: 10.3171/JNS/2008/109/12/S14. [DOI] [PubMed] [Google Scholar]
- 7.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;70:510–514. doi: 10.1016/j.ijrobp.2007.06.074. doi:10.1016/j.ijrobp.2007.06.074. [DOI] [PubMed] [Google Scholar]
- 8.Sperduto PW. What is your patient's GPA and why does it matter? Managing brain metastases and the cost of hope. Int J Radiat Oncol Biol Phys. 2010;77:643–644. doi: 10.1016/j.ijrobp.2010.02.038. doi:10.1016/j.ijrobp.2010.02.038. [DOI] [PubMed] [Google Scholar]
- 9.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;77:655–661. doi: 10.1016/j.ijrobp.2009.08.025. doi:10.1016/j.ijrobp.2009.08.025. [DOI] [PubMed] [Google Scholar]
- 10.Bernad DM, Sperduto PW, Souhami L, Jensen AW, Roberge D. Stereotactic radiosurgery in the management of brain metastases from primary thyroid cancers. J Neurooncol. 2010;98:249–252. doi: 10.1007/s11060-010-0175-z. doi:10.1007/s11060-010-0175-z. [DOI] [PubMed] [Google Scholar]
- 11.Agboola O, Benoit B, Cross P, et al. Prognostic factors derived from recursive partition analysis (RPA) of Radiation Therapy Oncology Group (RTOG) brain metastases trials applied to surgically resected and irradiated brain metastatic cases. Int J Radiat Oncol Biol Phys. 1998;42:155–159. doi: 10.1016/s0360-3016(98)00198-9. [DOI] [PubMed] [Google Scholar]
- 12.Cannady SB, Cavanaugh KA, Lee SY, et al. Results of whole brain radiotherapy and recursive partitioning analysis in patients with brain metastases from renal cell carcinoma: a retrospective study. Int J Radiat Oncol Biol Phys. 2004;58:253–258. doi: 10.1016/s0360-3016(03)00818-6. doi:10.1016/S0360-3016(03)00818-6. [DOI] [PubMed] [Google Scholar]
- 13.Videtic GM, Adelstein DJ, Mekhail TM, et al. Validation of the RTOG recursive partitioning analysis (RPA) classification for small-cell lung cancer-only brain metastases. Int J Radiat Oncol Biol Phys. 2007;67:240–243. doi: 10.1016/j.ijrobp.2006.08.019. doi:10.1016/j.ijrobp.2006.08.019. [DOI] [PubMed] [Google Scholar]
- 14.Sanghavi SN, Miranpuri SS, Chappell R, et al. Radiosurgery for patients with brain metastases: a multi-institutional analysis, stratified by the RTOG recursive partitioning analysis method. Int J Radiat Oncol Biol Phys. 2001;51:426–434. doi: 10.1016/s0360-3016(01)01622-4. doi:10.1016/S0360-3016(01)01622-4. [DOI] [PubMed] [Google Scholar]
- 15.Lorenzoni J, Devriendt D, Massager N, et al. Radiosurgery for treatment of brain metastases: estimation of patient eligibility using three stratification systems. Int J Radiat Oncol Biol Phys. 2004;60:218–224. doi: 10.1016/j.ijrobp.2004.02.017. doi:10.1016/j.ijrobp.2004.02.017. [DOI] [PubMed] [Google Scholar]
- 16.Weltman E, Salvajoli JV, Brandt RA, et al. Radiosurgery for brain metastases: a score index for predicting prognosis. Int J Radiat Oncol Biol Phys. 2000;46:1155–1161. doi: 10.1016/s0360-3016(99)00549-0. doi:10.1016/S0360-3016(99)00549-0. [DOI] [PubMed] [Google Scholar]
- 17.Rades D, Dziggel L, Haatanen T, et al. Scoring systems to estimate intracerebral control and survival rates of patients irradiated for brain metastases. Int J Radiat Oncol Biol Phys. 2011;80:1122–1127. doi: 10.1016/j.ijrobp.2010.03.031. doi:10.1016/j.ijrobp.2010.03.031. [DOI] [PubMed] [Google Scholar]
- 18.Rades D, Dunst J, Schild S. A new scoring system to predicting the survival of patients treated with whole-brain radiotherapy for brain metasatases. Strahlenther Onkol. 2008;184:251–255. doi: 10.1007/s00066-008-1831-5. doi:10.1007/s00066-008-1831-5. [DOI] [PubMed] [Google Scholar]
- 19.Kattan MW, Eastham JA, Stapleton AM, Wheeler TM, Scardino PT. A preoperative nomogram for disease recurrence following radical prostatectomy for prostate cancer. J Natl Cancer Inst. 1998;90:766–771. doi: 10.1093/jnci/90.10.766. doi:10.1093/jnci/90.10.766. [DOI] [PubMed] [Google Scholar]
- 20.Kattan MW, Reuter V, Motzer RJ, Katz J, Russo P. A postoperative prognostic nomogram for renal cell carcinoma. J Urol. 2001;166:63–67. doi:10.1016/S0022-5347(05)66077-6. [PubMed] [Google Scholar]
- 21.Knisely JP, Berkey B, Chakravarti A, et al. A phase III study of conventional radiation therapy plus thalidomide versus conventional radiation therapy for multiple brain metastases (RTOG 0118) Int J Radiat Oncol Biol Phys. 2008;71:79–86. doi: 10.1016/j.ijrobp.2007.09.016. doi:10.1016/j.ijrobp.2007.09.016. [DOI] [PubMed] [Google Scholar]
- 22.Berk L, Berkey B, Rich T, et al. Randomized phase II trial of high-dose melatonin and radiation therapy for RPA class 2 patients with brain metastases (RTOG 0119) Int J Radiat Oncol Biol Phys. 2007;68:852–857. doi: 10.1016/j.ijrobp.2007.01.012. doi:10.1016/j.ijrobp.2007.01.012. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Komarnicky LT, Phillips TL, Martz K, Asbell S, Isaacson S, Urtasun R. A randomized phase III protocol for the evaluation of misonidazole combined with radiation in the treatment of patients with brain metastases (RTOG-7916) Int J Radiat Oncol Biol Phys. 1991;20:53–58. doi: 10.1016/0360-3016(91)90137-s. doi:10.1016/0360-3016(91)90137-S. [DOI] [PubMed] [Google Scholar]
- 24.Sause WT, Scott C, Krisch R, et al. Phase I/II trial of accelerated fractionation in brain metastases RTOG 85–28. Int J Radiat Oncol Biol Phys. 1993;26:653–657. doi: 10.1016/0360-3016(93)90284-3. doi:10.1016/0360-3016(93)90284-3. [DOI] [PubMed] [Google Scholar]
- 25.Phillips TL, Scott CB, Leibel SA, Rotman M, Weigensberg IJ. Results of a randomized comparison of radiotherapy and bromodeoxyuridine with radiotherapy alone for brain metastases: report of RTOG trial 89–05. Int J Radiat Oncol Biol Phys. 1995;33:339–348. doi: 10.1016/0360-3016(95)00168-X. doi:10.1016/0360-3016(95)00168-X. [DOI] [PubMed] [Google Scholar]
- 26.Murray KJ, Scott C, Greenberg HM, et al. A randomized phase III study of accelerated hyperfractionation versus standard in patients with unresected brain metastases: a report of the Radiation Therapy Oncology Group (RTOG) 9104. Int J Radiat Oncol Biol Phys. 1997;39:571–574. doi: 10.1016/s0360-3016(97)00341-6. doi:10.1016/S0360-3016(97)00341-6. [DOI] [PubMed] [Google Scholar]
- 27.Andrews DW, Scott CB, Sperduto PW, et al. Whole brain radiation therapy with or without stereotactic radiosurgery boost for patients with one to three brain metastases: phase III results of the RTOG 9508 randomised trial. Lancet. 2004;363:1665–1672. doi: 10.1016/S0140-6736(04)16250-8. doi:10.1016/S0140-6736(04)16250-8. [DOI] [PubMed] [Google Scholar]
- 28.Stuart EA, Azur M, Frangakis C, Leaf P. Multiple imputation with large data sets: a case study of the Children's Mental Health Initiative. Am J Epidemiol. 2009;169:1133–1139. doi: 10.1093/aje/kwp026. doi:10.1093/aje/kwp026. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29.Cox D. Regression models and life-tables. Journal of the Royal Statistical Society. 1972;30:248–75. [Google Scholar]
- 30.Breiman L, Friedman JH, Olshen RA, Stone CJ. Classification and Regression Trees. Boca Raton, FL: Chapman & Hall/CRC Press; 1984. [Google Scholar]
- 31.Ishwaran H, Kogalur UB, Blackstone EH, Lauer MS. Random survival forests. Ann App Statist. 2008;2:841–860. doi:10.1214/08-AOAS169. [Google Scholar]
- 32.Segal MR, Bloch DA. A comparison of estimated proportional hazards models and regression trees. Stat Med. 1989;8:539–550. doi: 10.1002/sim.4780080503. doi:10.1002/sim.4780080503. [DOI] [PubMed] [Google Scholar]
- 33.Sperduto PW, Kased N, Roberge D, et al. The Effect of Tumor Subtype on Survival and the Graded Prognostic Assessment (GPA) for Patients with Breast Cancer and Brain Metastases. Int J Radiat Onc Biol Phys. 2011;82:2111–2117. doi: 10.1016/j.ijrobp.2011.02.027. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 34.Rubin DB, Schenker N. Multiple imputation in health-care databases: an overview and some applications. Stat Med. 1991;10:585–598. doi: 10.1002/sim.4780100410. doi:10.1002/sim.4780100410. [DOI] [PubMed] [Google Scholar]
- 35.Rubin DB. Multiple Imputation for Nonresponse in Surveys. New York: J. Wiley & Sons; 1987. [Google Scholar]
- 36.Rubin DB. Multiple imputation after 18+ years (with discussion) J Amer Stat Assoc. 1996;91:473–489. doi:10.2307/2291635. [Google Scholar]
- 37.Barnard J, Meng XL. Applications of multiple imputation in medical studies: from AIDS to NHANES. Stat Methods Med Res. 1999;8:17–36. doi: 10.1177/096228029900800103. doi:10.1191/096228099666230705. [DOI] [PubMed] [Google Scholar]
- 38.van Buuren S, Boshuizen HC, Knook DL. Multiple imputation of missing blood pressure covariates in survival analysis. Stat Med. 1999;18:681–694. doi: 10.1002/(sici)1097-0258(19990330)18:6<681::aid-sim71>3.0.co;2-r. doi:10.1002/(SICI)1097-0258(19990330)18:6<681::AID-SIM71>3.0.CO;2-R. [DOI] [PubMed] [Google Scholar]
- 39.Gallina A, Chun FK, Briganti A, et al. Development and split-sample validation of a nomogram predicting the probability of seminal vesicle invasion at radical prostatectomy. Eur Urol. 2007;52:98–105. doi: 10.1016/j.eururo.2007.01.060. doi:10.1016/j.eururo.2007.01.060. [DOI] [PubMed] [Google Scholar]
- 40.Steuber T, Chun FK, Erbersdobler A, et al. Development and internal validation of preoperative transition zone prostate cancer nomogram. Urology. 2006;68:1295–300. doi: 10.1016/j.urology.2006.08.1066. doi:10.1016/j.urology.2006.08.1066. [DOI] [PubMed] [Google Scholar]
- 41.Chun FK, Steuber T, Erbersdobler A, et al. Development and internal validation of a nomogram predicting the probability of prostate cancer Gleason sum upgrading between biopsy and radical prostatectomy pathology. Eur Urol. 2006;49:820–826. doi: 10.1016/j.eururo.2005.11.007. doi:10.1016/j.eururo.2005.11.007. [DOI] [PubMed] [Google Scholar]
- 42.Karakiewicz PI, Benayoun S, Kattan MW, et al. Development and validation of a nomogram predicting the outcome of prostate biopsy based on patient age, digital rectal examination and serum prostate specific antigen. J Urol. 2005;173:1930–1934. doi: 10.1097/01.ju.0000158039.94467.5d. doi:10.1097/01.ju.0000158039.94467.5d. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 43.Graefen M, Karakiewicz PI, Cagiannos I, et al. International validation of a preoperative nomogram for prostate cancer recurrence after radical prostatectomy. J Clin Oncol. 2002;20:3206–3212. doi: 10.1200/JCO.2002.12.019. doi:10.1200/JCO.2002.12.019. [DOI] [PubMed] [Google Scholar]
- 44.Graefen M, Karakiewicz PI, Cagiannos I, et al. Validation study of the accuracy of a postoperative nomogram for recurrence after radical prostatectomy for localized prostate cancer. J Clin Oncol. 2002;20:951–956. doi: 10.1200/JCO.2002.20.4.951. doi:10.1200/JCO.20.4.951. [DOI] [PubMed] [Google Scholar]