Validation of the CALGB and EORTC Prognostic Models for Mesothelioma based on multiple CALGB Trials (Alliance)

Abstract

BACKGROUND: Prognostic models play an important role in the design and analysis of mesothelioma treatment trials. The European Organisation for Research and Treatment of Cancer (EORTC) and the Cancer and Leukemia Group B (CALGB) prognostic models are two well-known tools to predict survival in patients with malignant mesothelioma. In this retrospective validation study, we aim to assess the performance of these two mesothelioma prognostic models for overall survival (OS) with multiple clinical trials data from CALGB. METHODS: Using 204 patients with malignant pleural mesothelioma, the EORTC model (Curran et al 1998) was developed using a Cox regression with white blood cell (WBC) count, ECOG performance status (PS), diagnosis, histological type, and gender as prognostic variables. Using 337 patients with malignant mesothelioma, the CALGB model (Herndon et al 1998) was developed using a crossvalidated exponential regression tree with PS, age, haemoglobin (Hgb) level, WBC count, chest pain indicator, and weight loss indicator as prognostic variables. In this validation study, 602 mesothelioma patients from fifteen completed CALGB treatment trials accrued between June 1984 and August 2009 were included. As the CALGB model was developed using the seven earlier studies, 266 patients from eight recent studies were included in the validation. For the EORTC model, we analysed all studies as well as just those eight recent studies. The concordance of predicted survival times and risk scores was estimated by c-index (Harrell et al 1996). Secondary endpoint of interest includes progression-free survival (PFS). Sensitivity analysis and multiple imputations were used to handle missing data. We also compared our results with PS alone. RESULTS: (1) For OS, the EORTC model produced c-indices equal to 0.592 and 0.610 for the fifteen and eight studies respectively. For the eight recent studies, the CALGB model produced c-indices equal to 0.618 and 0.593 without and with imputation respectively. PS alone produced c-indices equal to 0.591 and 0.564 for the fifteen and eight studies respectively. (2) For PFS, the EORTC model produced c-indices equal to 0.569 and 0.598 for the fifteen and eight studies respectively. For the eight recent studies, the CALGB model produced c-indices equal to 0.585 and 0.560 without and with imputation respectively. PS alone produced c-indices equal to 0.568 and 0.553 for the fifteen and eight studies respectively. See Table 1. CONCLUSION: The EORTC and CALGB models perform similarly, with little improvement in prognostic ability from either compared to using PS alone. Further improvement on these existing prognostic models is warranted.