Preconditions and multilevel models in studying post-surgical adverse outcomes

Abstract

A variety of adverse outcomes, such as kidney injury, death, cardiac injury, and respiratory failure affect a significant number of patients after surgery. Previous research has investigated possible predictors for these outcomes, including features extracted from physiologic time series, change points in the time series, and prior conditions. This study builds upon this previous work by further exploring time series statistics, such as entropy, long-term memory, and change point analysis, as possibly predictive measures of volatility. These statistics are further examined by conditioning on prior conditions using multilevel modeling. For prediction, we use both random forest models and the robust method of L 1 regularized logistic regression. Predictive results from these models are evaluated using receiver operating characteristic curves and their area under the curve values. While the developed models did not show improvements in predictive accuracy, they did show that change point analysis and measures of entropy and long-term memory can be useful tools in predicting post-surgical adverse outcomes. The multilevel models show the important interactions between prior conditions and time series statistics for predicting adverse outcomes.