Predicting hospital admissions with Poisson regression analysis

Abstract

In this thesis, Poisson regression is used to predict and analyze inpatient hospital admissions for two inpatient units (Four East and Four West) at Naval Medical Center San Diego. Data that include age group, gender, beneficiary category, enrollment site and fiscal month are collected for the patient population. This information is used along with additional details about past admissions such as the location and source of admission. These data are next fit to four different models that correspond to Four East (enrolled and un-enrolled beneficiaries) and Four West (enrolled and un-enrolled beneficiaries). Stepwise selection techniques are used to arrive at final models. The final models are used to observe trends in predicted hospital admissions based on trends in current population sizes.