Recursive partitioning analysis (RPA) of prognostic factors in three radiation therapy oncology group (RTOG) brain metastases trials

TLDR

Recent reports of promising outcomes with radiosurgery or stereotactic surgery for brain metastases highlight the need to determine whether these benefits arise from the therapy itself or from patient selection. The study aims to assess whether improved outcomes in brain metastases are attributable to therapy alone or partly to patient selection, and to develop prognostic classes for testing new treatments in homogeneous groups. The authors analyzed 1,200 patients from three RTOG brain metastases trials using recursive partitioning analysis to evaluate 18 pretreatment characteristics and 3 treatment variables, yielding a prognostic tree that defines three patient classes. The RPA identified three prognostic classes, with class 1 (KPS ≥ 70, age < 65, controlled primary, no extracranial metastases) showing median survival of 7.1 months, class 3 (KPS < 70) 2.3 months, and other patients 4.2 months.

Abstract

Promising results from new approaches such as radiosurgery or stereotactic surgery of brain metastases have recently been reported. Are these results due to the therapy alone or can the results be attributed in part to patient selection? An analysis of tumor/patient characteristics and treatment variables in previous Radiation Therapy Oncology Group (RTOG) brain metastases studies was considered necessary to fully evaluate the benefit of these new interventions.The database included 1200 patients from three consecutive RTOG trials conducted between 1979 and 1993, which tested several different dose fractionation schemes and radiation sensitizers. Using recursive partitioning analysis (RPA), a statistical methodology which creates a regression tree according to prognostic significance, eighteen pretreatment characteristics and three treatment-related variables were analyzed.According to the RPA tree the best survival (median: 7.1 months) was observed in patients < 65 years of age with a Karnofsky Performance Status (KPS) of at least 70, and a controlled primary tumor with the brain the only site of metastases. The worst survival (median: 2.3 months) was seen in patients with a KPS less than 70. All other patients had relatively minor differences in observed survival, with a median of 4.2 months.Based on this analysis, we suggest the following three classes: Class 1: patients with KPS > or = 70, < 65 years of age with controlled primary and no extracranial metastases; Class 3: KPS < 70; Class 2- all others. Using these classes or stages, new treatment techniques can be tested on homogeneous patient groups.