The Zero-truncated Poisson with Right Censoring: an Application to Translational Breast Cancer Research.

We propose to analyze positive count data with right censoring from Behbod et al. (2009) using the censored zero-truncated Poisson model (CZTP). The comparison in truncated means across subgroups in each cell line is carried out through a log-linear model that links the un-truncated Poisson parameter and regression covariates. We also perform simulation to evaluate the performance of the CZTP model in finite and large sample sizes. In general, the CZTP model provides accurate and precise estimates. However, for data with small means and small sample sizes, it may be more proper to make inference based on the mean counts rather than on the regression coefficients. For small sample sizes and moderate means, the likelihood ratio test is more reliable than the Wald test. We also demonstrate how power analysis can be used to justify and/or guide the choice of censoring thresholds in study design. A SAS macro is provided in Appendix for readers' reference.