Semiparametric model for recurrent event data with excess zeros and informative censoring

• Published in: Journal of statistical planning and inference 2012, Vol.142 (1), p.289-300
• Main Authors: Zhao, XiaoBing, Zhou, Xian, Wang, JingLong
• Format: Article
• Language: English
• Subjects: Additive rates model; Applications; Exact sciences and technology; General topics; Information censoring; Insurance, economics, finance; Local likelihood; Mathematics; Nonparametric inference; Probability and statistics; Probability theory and stochastic processes; Recurrent event data; Sciences and techniques of general use; Special processes (renewal theory, markov renewal processes, semi-markov processes, statistical mechanics type models, applications); Statistics; Zero-recurrence
• ISSN: 0378-3758; 1873-1171
• DOI: 10.1016/j.jspi.2011.07.016

Recurrent event data are often encountered in longitudinal follow-up studies in many important areas such as biomedical science, econometrics, reliability, criminology and demography. Multiplicative marginal rates models have been used extensively to analyze recurrent event data, but often fail to fit the data adequately. In addition, the analysis is complicated by excess zeros in the data as well as the presence of a terminal event that precludes further recurrence. To address these problems, we propose a semiparametric model with an additive rate function and an unspecified baseline to analyze recurrent event data, which includes a parameter to accommodate excess zeros and a frailty term to account for a terminal event. Local likelihood procedure is applied to estimate the parameters, and the asymptotic properties of the estimators are established. A simulation study is conducted to evaluate the performance of the proposed methods, and an example of their application is presented on a set of tumor recurrent data for bladder cancer. ► A semiparametric marginal rate model is developed to analyze recurrent event data. ► The model can accommodate excess zeros and informative censoring in the data. ► Local likelihood procedure is applied to estimate the model parameters. ► The asymptotic properties of the estimators are established. ► Application of the model is demonstrated on a set of bladder cancer data.