Prognostic factors in brain metastases: Should patients be selected for aggressive treatment according to recursive partitioning analysis (RPA) classes?

Purpose: To determine whether or not Radiation Therapy Oncology Group (RTOG ) recursive partitioning analysis (RPA) derived prognostic classes for pati ents with brain metastases are generally applicable and can be recommended as rational strategy for patient selection for future clinical trials. Incl usion of time to non-CNS death as additional endpoint besides death from an y cause might result in further valuable information, as survival limitatio n due to uncontrolled extracranial disease can be explored. Methods: We performed a retrospective analysis of prognostic factors for su rvival and time to non-CNS death in 528 patients treated at a single instit ution with radiotherapy or surgery plus radiotherapy for brain metastases, For this purpose, patients were divided into groups with Karnofsky performa nce status (KPS) <70% and KPS greater than or equal to 70%, as proposed by the RTOG, Results: Median overall survival was 2.9 months (2.0 months for patients wi th KPS <70% and 3.6 months for patients with KPS greater than or equal to 7 0%,p < 0,001). We did not find other variables splitting patients with KPS <70% in different prognostic groups. However, advanced age, multiple brain metastases, presence of extracranial metastases, and uncontrolled primary t umor each predicted shorter survival in patients with KPS greater than or e qual to 70%. When grouped into the original RTOG RPA classes, our data set split into three subgroups with different prognosis and median survival tim es of 10,5, 3,5, and 2 months, respectively (p < 0,05), Only 3% of patients fell into the most favorable group. Median time to non-CNS death was 4.1 m onths (12.9 months in RPA class I, 4.9 months in RPA class II, and 3.8 mont hs in RPA class III, respectively,p > 0.05 for RPA class II versus In). How ever, it was 8.5 months in RPA class II patients with controlled primary tu mor, which was found to be the only prognostic factor for time to non-CNS d eath in patients with KPS greater than or equal to 70%. In patients with KP S <70%, no statistically significant prognostic factors were identified for this endpoint, Conclusions: Despite some differences, this analysis essentially confirmed the value of RPA-derived prognostic classes, as published by the RTOG, when survival was chosen as endpoint, RPA class I patients seem to be most like ly to profit from aggressive treatment strategies and should be included in appropriate clinical trials. However, their number appears to he very limi ted. Considering time to non-CNS death, our results suggest that certain pa tients in RPA class II also might benefit from increased local control of b rain metastases, (C) 2000 Elsevier Science Inc.