Machine Learning-Based Model for Predicting Prolonged Mechanical Ventilation in Patients with Congestive Heart Failure.

Abstract

Mechanical ventilation (MV) is widely used to relieve respiratory failure in patients with congestive heart failure (CHF). Prolonged MV (PMV) is associated with a poor prognosis. We aimed to establish a prediction model based on machine learning (ML) algorithms for the early identification of patients with CHF requiring PMV. Twelve commonly used ML algorithms were used to build the prediction model. The least absolute shrinkage and selection operator (LASSO) regression was employed to select the key features. We examined the area under the curve (AUC) statistics to evaluate the prediction performance. Data from another database were used to conduct external validation. We screened out 10 key features from the initial 65 variables via LASSO regression to improve the practicability of the model. The CatBoost model showed the best performance for predicting PMV among the 12 commonly used ML algorithms, with favorable discrimination (AUC = 0.790) and calibration (Brier score = 0.154). Moreover, hospital mortality could be accurately predicted using the CatBoost model as well (AUC = 0.844). In the external validation, the CatBoost model also showed satisfactory prediction performance (AUC = 0.780), suggesting certain generalizability of the model. Finally, a nomogram with risk classification of PMV was shown in this study. The present study developed and validated a CatBoost model, which could accurately predict PMV in mechanically ventilated patients with CHF. Moreover, this model has a favorable performance in predicting hospital mortality in these patients.