Original Article
A modified recursive partitioning analysis for predicting overall survival in patients with non-small cell lung cancer and central nervous system metastases
Abstract
Background: Non-small cell lung cancer (NSCLC) is a major cause of brain metastases. Nonetheless, patients with central nervous system (CNS) spread are poorly represented in clinical trials. We sought to evaluate the overall survival (OS) of patients with NSCLC and CNS metastases.
Methods: Patients with NSCLC and CNS metastases treated at A. C. Camargo Cancer Center from January 2007 to December 2017 were selected. The primary endpoint was OS following the diagnosis of CNS metastasis. The Kaplan-Meier method was applied to calculate OS. Prognostic factors were assessed by the Cox Proportional Hazards model. As an exploratory analysis, a survival tree was generated based upon the two most statistically significant variables in the multivariate model and one additional clinically meaningful variable.
Results: In total, 311 patients were included. Median OS was 10.3 months (95% CI, 8.7–13.1 months). ECOG performance status 2–4 (HR 2.12; 95% CI, 1.40–3.20; P<0.01) and the absence of a driver mutation (HR 3.30; 95% CI, 1.85–5.90; P<0.01) were strongly associated with worse OS. A Modified Recursive Partitioning Analysis (mRPA) was developed based on the curves generated by the survival tree. mRPA stratified our cohort in four subgroups with significantly different OS (3.1 to 43 months) and it outperformed both RPA and GPA in predicting OS in our population.
Conclusions: OS in our cohort was better than previously reported. However, prognosis is widely variable and is mostly dictated by performance status and the presence of a driver mutation.
Methods: Patients with NSCLC and CNS metastases treated at A. C. Camargo Cancer Center from January 2007 to December 2017 were selected. The primary endpoint was OS following the diagnosis of CNS metastasis. The Kaplan-Meier method was applied to calculate OS. Prognostic factors were assessed by the Cox Proportional Hazards model. As an exploratory analysis, a survival tree was generated based upon the two most statistically significant variables in the multivariate model and one additional clinically meaningful variable.
Results: In total, 311 patients were included. Median OS was 10.3 months (95% CI, 8.7–13.1 months). ECOG performance status 2–4 (HR 2.12; 95% CI, 1.40–3.20; P<0.01) and the absence of a driver mutation (HR 3.30; 95% CI, 1.85–5.90; P<0.01) were strongly associated with worse OS. A Modified Recursive Partitioning Analysis (mRPA) was developed based on the curves generated by the survival tree. mRPA stratified our cohort in four subgroups with significantly different OS (3.1 to 43 months) and it outperformed both RPA and GPA in predicting OS in our population.
Conclusions: OS in our cohort was better than previously reported. However, prognosis is widely variable and is mostly dictated by performance status and the presence of a driver mutation.
