Expected Length of Stay with a Multistate Model

Abstract

A time-inhomogeneous Markov process can be used to model the effect of a hospital infection on the expected length of stay. It also allows for modelling the time-dependency of the hospital infection. The Markov assumption provides us with the Nelson-Aalen estimators for cumulative hazards, a powerful tool to quantify the effect of hospital infection on the expected length of stay. Using a summary measure, we have found that over patients in the ICU, CMV reactivation does not prolong the stay in ICU, but instead shortents the length of stay due to a high death hazard. By assuming the state transitions to follow a Cox proportional hazards model, we have also found that CMV reactivation does have a prolonging effect on length of stay for patients with a low APACHE IV score but not for patients with a high APACHE IV score.