J. Parreco1, R. Kozol1, R. Rattan1 1University Of Miami,Miami, FL, USA
Introduction:
Early identification of critically ill patients who will require prolonged mechanical ventilation has proven to be difficult and there are no established guidelines. The purpose of this study was to use artificial intelligence and machine learning techniques to identify patients at risk for prolonged mechanical ventilation (PMV) and tracheostomy placement.
Methods:
The Multiparameter Intelligent Monitoring in Intensive Care III database was queried for all intensive care unit (ICU) stays with mechanical ventilation and surviving hospitalization. PMV was defined as mechanical ventilation for more than 7 days. Machine learning classifiers with a gradient boosted decision trees algorithm were created for the outcomes of PMV and tracheostomy placement. The classifiers were trained using 10-fold cross validation. The variables used by the classifiers were six different severity of illness scores calculated on the first day of ICU admission including their components as well as thirty comorbidities. Mean receiver operating characteristic (ROC) curves were calculated for the outcomes and variable importance was quantified.
Results:
There were 20,262 ICU stays identified and PMV was required in 13.6% and tracheostomy was performed in 6.6% of all patients. The figure shows the mean ROC curves for the outcomes with shaded portion representing the range of cross validation folds. The classifier for predicting PMV was able to achieve a mean area under the curve (AUC) of 0.82 with an accuracy of 83% and specificity of 88%. The mean AUC for tracheostomy was 0.83 with an accuracy of 91% and specificity of 96%. The variable with the highest importance for predicting PMV was the Sequential Organ Failure Assessment (SOFA) Score (13%) and the most important comorbidity in predicting tracheostomy was cardiac arrhythmia (12%).
Conclusion:
Machine learning classifiers can be easily incorporated into existing electronic medical record systems. This study demonstrates their utility for the early identification of patients at risk for prolonged mechanical ventilation and tracheostomy. Application of these identification techniques could lead to improved outcomes by allowing for early intervention.
