How long have I got? Estimating survival for women starting first-line chemotherapy for metastatic breast cancer (MBC): A systematic review of recent randomized trials.

Abstract

9106 Background: Most patients with advanced cancer want an idea of typical, best case, and worst case scenarios for overall survival (OS). Specific percentiles from OS curves of comparable patients are a reasonable starting point for estimating these scenarios. We summarized recent randomized trials to determine if simple multiples of the median OS from each curve provided reasonable estimates of these scenarios, and to determine the ratio of median OS to median PFS from corresponding curves. Methods: We searched the Cochrane register and major journals from 1999 to 2009 for randomized trials of first-line chemotherapy for MBC with ≥90 patients/arm. We recorded the median and following percentiles (represented scenario) from each OS curve: 90th (worst case), 75th (lower typical), 25th (upper typical) and 10th (best case). We estimated these percentiles for each OS curve by multiplying its median by the following simple multiples: 0.25 for the 90th, 0.5 for the 75th, 2 for the 25th, and 3 for the 10th. Each estimate was deemed accurate if within 0.75 to 1.33 times the actual value. Associations between trial characteristics and median OS were assessed with linear regression. Results: We analyzed 74 OS curves from 34 trials including 12,714 women followed for a median of 3 years. The average value (interquartile range, IQR) of the median OS was 22 months (18-24), of the median PFS was 7 months (6-9), and of the ratio of median OS to PFS was 3.0 (2.4-3.5). The average value (IQR) in months from all OS curves for each scenario was: worst case 6 (5-7), lower typical 12 (10-13), upper typical 36 (31-41) and best case 56 (47-60). Simple multiples of the median OS provided accurate estimates of the worst case in 75% of curves, lower typical in 97%, upper typical in 96% and best case in 96%. OS was longer in trials with a higher proportion of ER+ tumors (p=0.0018) and in trials of HER2+ women treated with trastuzumab (p=0.004). Conclusions: Worst, typical, and best case scenarios were remarkably consistent in a wide range of contemporary trials. Simple multiples of the median provided accurate estimates of these scenarios for most OS curves and are a good basis for discussing prognosis. No significant financial relationships to disclose.