|Friday, February 16|
|PS2 Poster Session 2 and Refreshments||
Fri, Feb 16, 5:15 PM - 6:30 PM
A Practical Guide for Modeling Length of Stay with Focus on Right Skewness and Zero Inflation (303624)
Song Wu, Stony Brook University
Jie Yang, Stony Brook University
Keywords: length of stay, right skewness, zero inflation, linear regression models, generalized linear regression models, zero inflated models, Hurdle models
Length of stay (LOS) is a key metric in evaluating health service expenditures and assessing in-hospital caring quality. Therefore, it is frequently studied in clinical research studies. Often the distribution of LOS is right skewed and/or zero inflated. Because of this, popular models for analyzing LOS include linear regression models with and without log transformation of length of stay, generalized linear regression models assuming normal, Poisson and negative binomial distributions, and zero inflated and Hurdle regression models assuming Poisson and negative binomial distributions. However, none of these 9 statistical models is consensually recommended to model LOS. A practical guide is presented here to choose the optimal model for length of stay using its distribution characteristics, based on both simulation studies and real data analysis. Implementation of these models using SAS and R will also be introduced.