All Times ET
Virtual
Comparison of Statistical Regression Models in Analysis of Hospital Length of Stay (305322)
*Gustavo Fernandez, The University of Texas, Rio Grande Valley
Kristina Vatcheva, The University of Texas, Rio Grande Valley
Keywords: Hospital length of stay, Count data, Poisson regression, Negative Binomial regression, Zero Inflated regression
In Healthcare, the hospital length of stay (LOS), defined as the duration of a single episode of hospitalization, is a key indicator of hospital activities including hospital care management, cost of care, quality control, appropriate use of hospital services, hospital planning, and use of hospital resources. LOS is a metric used for measuring the efficiency with which hospitals provide care to their patient population. Many studies have been conducted to predict patients’ LOS and to identify important socio-demographic predictors of LOS. Some of those studies used linear, logarithmic transformation, and logistic regression approaches which has been criticized for being inadequate in modeling LOS. We conducted simulation studies to compare different regression methods: Poisson, Negative Binomial, Zero Inflated regression models for analysis of count data. Specifically, we generated data to demonstrate the potential differences between regression methods using different simulation scenarios for sample size, proportions of zeros, and over-dispersion. In addition, analyses to illustrate the aforementioned methods were conducted using empirical count data for hospital LOS.