We assembled a single-center retrospective cohort of 7215 adult surgical patients
hospitalized between 2000 and 2010 who had intraoperative physiologic time series data
available. We developed generalized additive models applying non-linear mathematical
modeling of preoperative clinical risk factor data to calculate probabilistic risk scores for
postoperative AKI. We incorporated physiologic time series with approximately 600
observations per patient including mean arterial blood pressure (MAP), minimum alveolar
oxygen concentration and heart rate into the models and assessed AKI prediction
associated with intraoperative events. We also included available lab results, medications
administered and estimated blood loss. We extracted time series features and added them
to the preoperative model.