The Zero-Truncated Poisson Model with Right Censoring: An Application to Translational Breast Cancer Research

• Activity Number: 127
• Type: Contributed
• Date/Time: Monday, August 2, 2010 : 8:30 AM to 10:20 AM
• Sponsor: Biometrics Section
• Author(s): Hung-Wen Yeh*+ and Purna Mukhopadhyay and Fariba Behbod and Byron J. Gajewski
• Companies: The University of Kansas Medical Center and The University of Kansas Medical Center and The University of Kansas Medical Center and The University of Kansas
• Keywords: positive counts ; zero truncation ; right censoring ; sample size ; ductal carcinoma in situ ; zero-truncated Poisson

Abstract

Positive count data are common in medical research and are often fitted by the zero-truncated Poisson (ZTP) model. Observations with large count values are not always recorded and can be censored, creating data with simultaneous left-truncation and right-censoring. We propose to analyze such data using the censored ZTP (CZTP) model. The empirical study suggests the CZTP model provides accurate and precise estimates whereas ignoring censoring can lead to greater bias and underestimate variability. Type I error rate and power are well preserved in the CZTP model. To show an example, we applied the proposed method to a human-in-mouse transplantation study that mimics the progression of ductal carcinoma in situ (DCIS) in human. We also demonstrated how power analysis can be used to guide the choice of censoring thresholds in experimental design.