A statistical comparison of prognostic index systems for brain metastases after stereotactic radiosurgery or fractionated stereotactic radiation therapy

Abstract

Aims: Prognostic indices are commonly used in the context of brain metastases radiotherapy to guide patient decision-making and clinical trial stratification. The purpose of this investigation was to compare nine published brain metastases prognostic indices using traditional and novel statistical comparison metrics.

Materials and methods: A retrospective review was carried out on two institutional databases of 501 patients diagnosed with brain metastatic disease, who received either stereotactic radiosurgery (n = 381) or fractionated stereotactic radiation therapy (n = 120) between 2002 and 2011. Descriptive statistics were generated for patient, tumour and treatment factors, as well as prognostic indices distribution. To identify predictors of overall survival, Kaplan-Meier estimates and multivariable Cox proportional hazard analyses were carried out. Prognostic indices were compared with each other using novel metrics, including: net reclassification improvement (NRI) index, integrated discrimination improvement (IDI) index and decision curve analysis (DCA).

Results: Multivariable Cox modelling confirmed the importance of all individual prognostic indices component factors except for 'active primary cancer' tumour status. When traditional and novel comparative metrics were incorporated, the available published prognostic indices were found to have important general classification benefits as follows: Radiation Therapy Oncology Group recursive partitioning analysis (RTOG RPA; NRI and DCA), Rades et al. first index (RADES I; IDI and DCA), Golden grading system (GGS; IDI and DCA) and Rotterdam index (RDAM; major misclassification rate and NRI). The graded prognostic assessment system had the smallest misclassification rate (5%) in terms of high-risk (i.e. poor prognosis) classification.

Conclusions: Summarising the various comparative approaches used in this report, we found that the RTOG RPA, GGS, RADES I and RDAM systems were superior in more than one metric studied. Of these, only the RTOG RPA has been extensively validated using large datasets and clinically utilised both at the patient level and in clinical trials.