A structural equation model of adverse events and length of stay in hospitals

Adverse events in hospitals cause significant morbidity and mortality, and considerable effort has been invested into analysing their incidence and preventability. An unresolved issue in models of medical adverse events is potential endogeneity of length of stay (LOS): whilst the probability of suffering a medical adverse event during the episode is likely to increase as a patient stays longer, there are a range of unobservable patient and hospital factors affecting both the occurrence of adverse events and LOS, such as unobserved patient complexity and hospital management. Therefore, statistical models of adverse events which do not account for the potential endogeneity of LOS may generate biased estimates. Our objective is to examine the effects of risk factors on the incidence of adverse events using structural equation models and accounting for endogeneity of LOS. We estimate separate models for three of the most common and serious types of medical adverse events: adverse drug reactions, hospital acquired infections, and pressure ulcers. We use episode level administrative hospital data from public hospitals in the state of Victoria, Australia, for the years 2004/05 and 2005/06 with detailed information on patients, in particular medical complexity and adverse events suffered during admission. We use days and months of discharge as instruments for LOS. Our research helps assessing the costs and benefits of additional days spent in hospital. For example, it can contribute to identifying the ideal time of discharge of patients, or inform whether 'hospital at home' programs reduce rates of hospital acquired infections.