Exact Nonparametric Confidence Bands for the Survivor Function

• Published in: The international journal of biostatistics Vol. 9; no. 2; pp. 185 - 204
• Main Author: Matthews, David
• Format: Journal Article
• Language: English
• Published: Germany De Gruyter 01.11.2013; Walter de Gruyter GmbH
• Subjects: Algorithms; Biometry - methods; Computer Simulation; Confidence Intervals; exact simultaneous confidence bands; Humans; Kaplan-Meier Estimate; Kaplan–Meier estimator; Likelihood Functions; Lung Neoplasms - drug therapy; Lymphoma, Non-Hodgkin - drug therapy; Monte Carlo Method; Noe recursions; right censoring; survivor function
• ISSN: 2194-573X; 1557-4679; 1557-4679
• DOI: 10.1515/ijb-2012-0046

A method to produce exact simultaneous confidence bands for the empirical cumulative distribution function that was first described by Owen, and subsequently corrected by Jager and Wellner, is the starting point for deriving exact nonparametric confidence bands for the survivor function of any positive random variable. We invert a nonparametric likelihood test of uniformity, constructed from the Kaplan–Meier estimator of the survivor function, to obtain simultaneous lower and upper bands for the function of interest with specified global confidence level. The method involves calculating a null distribution and associated critical value for each observed sample configuration. However, Noe recursions and the Van Wijngaarden–Decker–Brent root-finding algorithm provide the necessary tools for efficient computation of these exact bounds. Various aspects of the effect of right censoring on these exact bands are investigated, using as illustrations two observational studies of survival experience among non-Hodgkin’s lymphoma patients and a much larger group of subjects with advanced lung cancer enrolled in trials within the North Central Cancer Treatment Group. Monte Carlo simulations confirm the merits of the proposed method of deriving simultaneous interval estimates of the survivor function across the entire range of the observed sample. This research was supported by the Natural Sciences and Engineering Research Council (NSERC) of Canada. It was begun while the author was visiting the Department of Statistics, University of Auckland, and completed during a subsequent sojourn at the Medical Research Council Biostatistics Unit in Cambridge. The support of both institutions, in addition to that of NSERC and the University of Waterloo, is greatly appreciated.