Risk of falls for hospitalized patients: A predictive model based on routinely available data

Abstract

The incidence rate of falls is often used as an indicator of nursing care outcome. Comparing outcome between different settings should, however, make allowance for case mix. To measure the incidence of falls, describe their circumstances and develop a prediction model based on routinely collected data to allow comparison between hospital settings with different case mix. A dynamic population of patients hospitalized over a year in which a case was defined as any accidental fall systematically reported on an ad hoc form. A Swiss university hospital of 800 beds; 634 falls were reported for 26,643 hospitalizations over 236,307 hospitalization days. First fall rates were analyzed using a Poisson regression model with routinely computerized discharge data as independent variables. The incidence rate of first falls was 2.2 per 1000 patient-days. For subsequent falls the rates of incidence increased with the number of falls. Five independent variables played a significant role: age, gender, morbidity predisposition, surgical procedure and length of stay. Two of the interactions between these variables were significant and remained in the model (length of stay with age, morbidity with age). The model offers good medical plausibility and satisfactory predictive performance. The proposed model can be used by national health agencies to compute expected first fall rates accounting for case mix. Hospitals can use these rates for evaluation. Recommendations for measuring, monitoring and assessing fall rates are also given.