Zero-augmented accelerated spatial failure model for modeling hospital length of stay data

Abstract

Hospital length of stay (LOS) is often used as an indicator for hospital efficiency and resource utilization. LOS is nonnegative with presence of zeros and typically positively skewed with a long right tail, which may not be adequately modelled by traditional distributions, such as lognormal. We developed a zero-augmented accelerated frailty model for modeling the extreme skewness with the presence of zeros. Levels of utilization of health services may vary geographically, so conditional autoregressive priors were used to provide spatial smoothing across neighboring hospital health districts. The random effect terms are further linked to investigate if the capacity for longer LOS are consistently higher or lower at the health district level. Modeling and inference used the Bayesian approach via Markov Chain Monte Carlo simulation techniques. We demonstrated the proposed model for modeling the LOS of patients admitted due to chronic lower respiratory disease in Saskatchewan, Canada.