Modelling interval over-dispersed censored survival data

Abstract

Introduction. Unknown exact timing of events and quite possible heterogeneity of counts distributions across time intervals together with profile insufficiencies pose challenges to data modelling. it calls either for mixture of latent variables distributions or generalised distributions that accommodate heterogeneity. Our goal was to examine capacity of Negative Binomial Type 2 (NB2) regression to deal with presented challenges. Methods and data. We suggested application of NB2 regression with support of power analysis implemented in R package «ltable». We examined efficacy by simulated example to demonstrate the capacity to unveil the data generation mechanism. Results confirmed that counts strongly over-dispersed which is usual situation. Covariance matrix of model parameters was not stable enough and sensitive due to the paucity of profiles that is also typical. Fit to the data was good as chi-square test is less than 1 per degree of freedom with actual value of 0.09. That was supported by residuals report. Regression effects were of expected directions and magnitudes and revealed data generation mechanism. Power analysis validated the output and substantiated true data generation mechanism of interval censored survival data. We suggest that tentative example supports the effectiveness of modelling. In its turn power analysis helps to validate the output and to reveal and substantiate true data generation mechanism of interval censored survival data. Conclusions. Modelling interval over-dispersed censored survival data is still a challenge due to heterogeneity and profile shortage that underpowers hypotheses tests. We suggest NB2 based regression to process such data with support of power analysis. Simulated data analysis supports the effectiveness of modelling.