Abstract:Background: Hospital length of stay (LOS) is a key indicator of hospital care management efficiency, cost of care, and hospital planning. Hospital LOS is often used as a measure of a post-medical procedure outcome, as a guide to the benefit of a treatment of interest, or as an important risk factor for adverse events. Therefore, understanding hospital LOS variability is always an important healthcare focus. Hospital LOS data can be treated as count data, with discrete and non-negative values, typically right skewed, and often exhibiting excessive zeros. In this study, we compared the performance of the Poisson, negative binomial (NB), zero-inflated Poisson (ZIP), and zero-inflated negative binomial (ZINB) regression models using simulated and empirical data. Methods: Data were generated under different simulation scenarios with varying sample sizes, proportions of zeros, and levels of overdispersion. Analysis of hospital LOS was conducted using empirical data from the Medical Information Mart for Intensive Care database. Results: Results showed that Poisson and ZIP models performed poorly in overdispersed data. ZIP outperformed the rest of the regression models when the overdispersion is due to zero-inflation only. NB and ZINB regression models faced substantial convergence issues when incorrectly used to model equidispersed data. NB model provided the best fit in overdispersed data and outperformed the ZINB model in many simulation scenarios with combinations of zero-inflation and overdispersion, regardless of the sample size. In the empirical data analysis, we demonstrated that fitting incorrect models to overdispersed data leaded to incorrect regression coefficients estimates and overstated significance of some of the predictors. Conclusions: Based on this study, we recommend to the researchers that they consider the ZIP models for count data with zero-inflation only and NB models for overdispersed data or data with combinations of zero-inflation and overdispersion. If the researcher believes there are two different data generating mechanisms producing zeros, then the ZINB regression model may provide greater flexibility when modeling the zero-inflation and overdispersion.

Some Theoretical Comparisons of Negative Binomial and Zero-Inflated Poisson Distributions

A Comparative Study of Several Models for Zero Inflation Counting Data

On performance of parametric and distribution‐free models for zero‐inflated and over‐dispersed count responses

A comparison of statistical methods for modeling count data with an application to hospital length of stay

The Zero-inflated Negative Binomial-Exponential Distribution and Its Application

Score Tests for Zero-Inflated Double Poisson Regression Models

Comparisons of Order Statistics from Heterogeneous Negative Binomial Variables with Applications

Zero-inflated non-central negative binomial distribution

The Positive Negative Binomial-Lindley Distribution and Its Applications

Poisson and negative binomial regression models for zero-inflated data: an experimental study

Quasi-negative binomial distribution: Properties and applications

The Zero-Inflated Poisson - Probit regression model: a new model for count data

Comparison of BINAR(1) Models with Bivariate Negative Binomial Innovations and Explanatory Variables

A new multivariate zero-adjusted Poisson model with applications to biomedicine

Zero-inflated Loss Count Models and Their Applications

The Multivariate Component Zero‐inflated Poisson Model for Correlated Count Data Analysis

Two Methods of Parameter Estimate and Their Comparison for the Negative Binomial Distribution

Zero-inflated Poisson Models with Censored Data and Its Application

ASYMPTOTIC PROPERTIES IN THE PROBIT-ZERO-INFLATED BINOMIAL REGRESSION MODEL

One mixed negative binomial distribution with application

Inference on a bivariate binomial distribution with zero-inflation applicable to baseball data