| Sumario: | Background. Bedside monitors in intensive care units (ICUs) routinely measure and collect patients' physiologic data in real time in order to continuously assess the health status of critically ill patients. With the advent of increased computational power and the ability to store and rapidly process big datasets in recent years, this physiologic data show promise in identifying specific outcomes and/or events during patients' ICU hospitalization. Methods. We introduce a methodological workflow capable of automatically extracting meaningful information from continuous-in-time vital signs to predict patient outcomes. Our prediction algorithms are based on robust and scalable machine-learning techniques. We demonstrate the feasibility and applicability of this approach to a pilot study aimed at predicting whether or not a patient will experience a prolonged ICU length of stay (defined as longer than 4 days) in a cohort of 284 mechanically ventilated patients collected from a pediatric medical and surgical ICU at Boston Children's Hospital, using only information collected or available during the first 24 hours of mechanical ventilation. Results. Our methodology achieves predictive accuracies above 83% (AUC) by using only vital sign information collected from bedside physiologic monitors. In addition, we show that combining vital sign information with patients' demographic and clinical history data contained in electronic health records, improves the accuracy of our approach to accuracies of 90% (AUC). The predictive power of our methodology is assessed strictly on an unseen hold-out validation set of patients. Conclusions. Our proposed workflow may prove useful in the design of scalable approaches for real-time predictive systems in ICU environments exploiting real time vital sign information from bedside monitors.
|
|---|