Modelling Length of Stay in Hospitals using Multinomial Regression

Abstract

Hospital management is generally focused on studying the length of stay of patients since the measure has an impact on hospital resources. It is a challenging task for the hospital management to model the length of stay as they are asymmetric and heterogeneous in nature. Diabetes is a major health problem prevalent worldwide which leads to hospitalization over a time period. The present study deals with stay of diabetes patients classified as very short, short, medium and long duration of stay based on quantile classification rather than arbitrary approach. In this study, we have attempted to include an important covariate known as medical record since it assist in reducing the stay of a patient and can thereby accommodate more patients deserving treatment as inpatients. Based on the multiple levels of the response variable, we have considered fitting multinomial regression model for length of stay on diabetes. Further, this study has considered the validation of variable selection procedure for model fitting using subsampling approach. In conclusion, it has been identified that medical records is one of the important factor affecting the stay of patients and subsampling approach has been helpful in building the final model.