@ARTICLE{10.21494/ISTE.OP.2019.0393, 

TITLE={Zero-inflated regression models for right-censored counts, with an application to healthcare utilization}, 
  
AUTHOR={Van Trinh Nguyen, Jean-François Dupuy, }, 
	
JOURNAL={Biostatistics and Health Sciences}, 
	
VOLUME={1}, 

NUMBER={Issue 1}, 
	
YEAR={2019}, 

URL={https://openscience.fr/Zero-inflated-regression-models-for-right-censored-counts-with-an-application}, 
	
DOI={10.21494/ISTE.OP.2019.0393}, 
	
ISSN={2632-8291}, 
	
ABSTRACT={Zero-inflated models for censored and overdispersed count data have received little attention so far, except for the zero-inflated Poisson (ZIP) model which assumes that overdispersion is entirely caused by zero-inflation. When additional overdispersion is present, useful alternatives to ZIP are given by the zero-inflated generalized Poisson (ZIGP) and zero-inflated negative binomial (ZINB) models. This paper investigates properties of the maximum likelihood estimator (MLE) in ZIGP and ZINB regression models when the count response is subject to right-censoring. Simulations are used to examine performance (bias, mean square error, coverage probabilities and standard error calculations) of the MLE. Results suggest that maximum likelihood yields accurate inference. A simple, efficient and easy-to-implement methodology for variable selection is also proposed. It is applicable even when the number of predictors is very large and yields interpretable and sound results. The proposed methods are applied to a dataset of healthcare demand.}}