Factors Affecting Hospital Length of Stay Using Mixed Poisson Regression Models

Modeling of Hospital Length of Stay (LOS) is of great importance in healthcare systems, but there is paucity of information on this issue in Iran. The aim of this study was to identify the optimal model among different mixed poisson distributions in modeling the LOS and effective factors.

In this cross-sectional study, we studied 1256 records, including 15 variables associated with LOS in Sari Imam Khomeini Hospital (2016). Discrete Uniform-Poisson (UP) and Generalized Poisson-Lindley (GPL) distributions were fitted on LHS and modeling was performed.

Mean LOS was 4.95 days. According to the Z-test, data were overdispersed (P<0.001). GPL distribution was the best model (Akaike value=5994.61). GPL regression model showed significant relationships between LOS and age, sex, marital status, occupation, death, inpatient ward, and diagnosis (P<0.05). Longer LOS were seen in patients of lower ages and those who were employed. Mean LOS in women was 1.40 times higher than men. The LOS in internal ward, surgery, emergency, and maternity wards were (2.68, 1.57, 1.62, 0.78 times, respectively) higher than those in oncology ward. Mean LOS was considerably higher in patients with musculoskeletal disorders (8.51 days).

Hospital length of stay was different in all wards, so any Mixed Poisson Distribution that better fits such data could be used.