Evaluation of patient safety culture using a random forest algorithm

Abstract

Safety culture is a multidimensional concept that may be associated with medical errors and patient safety events in healthcare delivery systems. However, limited evidence is available regarding which safety culture dimensions drive overall patient safety. Moreover, the use of advanced statistical analysis has been limited in past studies of safety culture data. To address these issues, we use hospital-level aggregate survey data from U.S. hospitals to analyze the relationship between the defined safety culture dimensions and the patient safety grade. We use a tree-based machine learning algorithm, random forests, to estimate accurate and stable associations. The results of our analysis show that safety perception, management support, and supervisor/manager expectations are the leading drivers of patient safety grade. More specifically, safety problems in the work unit and work climate provided by hospital management are specific drivers of patient safety outcomes. The random forest model sheds new light on the most important cultural features relevant to patient safety.