The predictive value of pressure recording analytical method for the duration of mechanical ventilation in children undergoing cardiac surgery with an XGBoost-based machine learning model.

Abstract

Prolonged mechanical ventilation in children undergoing cardiac surgery is related to the decrease in cardiac output. The pressure recording analytical method (PRAM) is a minimally invasive system for continuous hemodynamic monitoring. To evaluate the postoperative prognosis, our study explored the predictive value of hemodynamic management for the duration of mechanical ventilation (DMV). This retrospective study included 60 infants who underwent cardiac surgery. Cardiac index (CI), the maximal slope of systolic upstroke (dp/dtmax), and cardiac cycle efficiency (CCE) derived from PRAM were documented in each patient 0, 4, 8, and 12 h (T0, T1, T2, T3, and T4, respectively) after their admission to the intensive care unit (ICU). A linear mixed model was used to deal with the hemodynamic data. Correlation analysis, receiver operating characteristic (ROC), and a XGBoost machine learning model were used to find the key factors for prediction. Linear mixed model revealed time and group effect in CI and dp/dtmax. Prolonged DMV also have negative correlations with age, weight, CI at and dp/dtmax at T2. dp/dtmax outweighing CI was the strongest predictor (AUC of ROC: 0.978 vs. 0.811, p < 0.01). The machine learning model suggested that dp/dtmax at T2 ≤ 1.049 or < 1.049 in combination with CI at T0 ≤ 2.0 or >2.0 can predict whether prolonged DMV (AUC of ROC = 0.856). Cardiac dysfunction is associated with a prolonged DMV with hemodynamic evidence. CI measured by PRAM immediately after ICU admission and dp/dtmax 8h later are two key factors in predicting prolonged DMV.