In his seminal paper, Sperduto et al. compared the newly published GPA with other prognostic indexes, using retrospectively the RTOG database to group BM patients in multiple levels with similar outcome. The authors conclude that the GPA index is as prognostic as the RPA. To our best of our knowledge, this is the first prospective comparison after Nieder's et al. retrospective validation, of three prognostic indexes in a multinational setting, showing that the GPA, BSBM and RPA are valid tools to prognosticate BM patients.
Our multivariate analysis has shown that three factors, namely, KPS, prognostic scores and center, were significant independent predictors for OS (Table 5). Although the former two parameters were foreseen survivorship predictors, the latter was somewhat unexpected. One center included patients with a significant better KPS (KPS ≥ 70, 81.9% vs. 65.3%; P = 0.008) and overall prognosis (RPA 3, 18.1% vs. 35.2%; P = 0.024). Although the goal of prognostication modeling, using a multivariable model, is to provide quantitative knowledge about the probability of outcomes in patients with different characteristics, the present analysis may have been influenced by the recruitment of these patients in this study. One center entered prospectively patients seen routinely in the practice of a busy radiotherapy department, whereas the other Spanish centers entered only a small part of patients in routine clinical practice (n = 58 cases). These centers accrued a majority of patients (n = 155 cases; 72.8%) in two consecutive prospective trials stemming from the GEGB group. One phase II trial, randomized BM patients to WBRT and temozolomide chemotherapy vs. WBRT alone, excluding specifically good prognosis patients who underwent surgery or radiosurgery, with or without RT, and including patients with KPS of 50 to 60. The other trial included patients treated with WBRT to prospectively assess the neurological outcome, excluding specifically patients with good prognosis who underwent surgery or radiosurgery. It is also possible that active palliative care was more readily available in the non-Spanish center, which could have a prognostic impact for these patients.
It was assumed that the RPA, when compared to the GPA scoring system, would be more easy to use. We did observe, however, that minor discrepancies (10 - 15%) between the reported and re-computed scores were substantial, irrespective of the three scoring systems. Interestingly, the 4-tier scoring system, i.e. GPA, had the highest number (n = 7) of major discrepancies, illustrating the difficulty to evaluate prognosis using a multi level scoring system and non-integer values. Importantly, the measure of agreement between the scored and re-calculated prognostic values was fairly good for the BSBM (κ value of 0.67) and good for the RPA (κ value of 0.81) and GPA (κ value of 0.81), respectively. These observed κ values, assessing the reliability of prospective scoring and retrospective computation for these categorical scales, legitimate the use of these scoring system in daily practice.
Our data suggests that the prediction of these indexes may be for short term (< 2 - 3 months) prognostication only (Table 6). We performed an alternative Cox computation in which the effect of the score on the survival was allowed to vary on time. This approach indicated that the value of the prognostic score at the time of the diagnosis was poorly associated with the survival after 3 months. This observation has not been observed in previous analyses [5, 6, 10, 12], but differential survivorship as a function of classes/groups in these series was assessed using Cox proportional hazards models only. We chose to perform a non-dependant Cox analysis, as the assumption of proportional hazards was not verified in our data, suggesting thus that the classes/groups' prognostic ability changed over time. After visual analysis of the plots, we elected to assess prognostication in three time-intervals, relevant to the clinical outcome of BM patients. This finding was unexpected and should be confirmed by further research in the framework of future prospective trials. It may well be that too few good prognostic patients (in the range of 3% to 15% using the GPA and RPA indexes in our study; Table 3) were included in this analysis with a consequential time-dependence prognostication unreliability of the studied models.
Notwithstanding the data published by Sperduto et al. initially in 2008  and updated in 2010 with the incorporation of recent data stemming from prospective randomized trials , we cannot state that one prognostication system was superior to another (Figure 1). Small patient numbers and differences in patient populations between these two set of data complicate the interpretation of these findings. The limit of the RPA index have been detailed in a pivotal editorial published by this author and is exemplified by the following clinical case of an asymptomatic young renal cell cancer patient with one brain metastasis, good performance status and two bone metastasis. The predicted survivorship of this very patient would vary more than 4 fold depending on the used prognostic (i.e. RPA or GPA) index. We must be aware however of developing a zealotry about these indexes, in which we, as physician, rely too heavily on them, be it RPA, GPA, BSBM or any other future prognostic scales, to tailor our patient's therapy. Median survival of 9.3 years was estimated in 32 BM patients treated in two leading US institutions ; among these patients, 9.4% and 28% were older than 65 years and had multiple BMs or systemic disease at brain metastasis, respectively. The majority (60%) of long-term advanced lung cancer patients were older than 65 years in another series. Having said that, there are definite arguments to use a simple prognostic scoring system, using objective (i.e. not using subjective assessment of control of primary tumor) and patient-related parameters to guide the therapeutic management of these challenging patients. More often than not, biases that influence therapeutic decision, made by physicians or family alike, could be diminished by applying selectively prognostic scores to these patients.
In summary, all studied indexes were prognostically relevant in BM patients in this prospective study. Our data did not suggest a greatest prognostic power of one scoring system compared to others. In our study, the significant OS difference observed within 3 months of diagnosis between the BSBM, RPA and GPA classes/groups was however not observed after this cut-off time point. GPA may be more difficult to use for daily prognostication of BM patients. The authors recommend that, regardless of the scoring index used, caution should be exercised by the treating physicians to use discretely these prognostic models and to comprehensively integrate other health, familial and socio-economical related parameters to this very heterogeneous population of patients with BMs.