Abstract
BACKGROUND: Hypoxemia is a common symptom in critically ill patients, with over half of patients in intensive care unit (ICU) experiencing varying degrees of hypoxemia, and its severity is an independent risk factor for patient mortality. Invasive mechanical ventilation is the main treatment for hypoxemia, but prolonged mechanical ventilation may lead to complications such as alveolar damage and accumulation of inflammatory factors. Therefore, clinical consensus emphasizes early extubation to reduce risk. However, some patients are prone to respiratory failure or re-intubation failure after extubation, resulting in a poor prognosis. Oxygen therapy, as a commonly used supportive measure after extubation, can improve oxygenation. However, there is currently limited research on the survival of patients with hypoxemia who receive post-extubation oxygen therapy after invasive ventilation, and there is a lack of effective prognostic tools. This study aimed to develop and validate a predictive model that integrates clinical features and laboratory indicators to identify the survival risk of hypoxemic patients receiving post-extubation oxygen therapy after mechanical ventilation, providing a basis for clinical treatment decisions. METHODS: A total of 7,340 patients from the Medical Information Mart for Intensive Care-IV (MIMIC-IV) database were randomly assigned to a training set and a validation set in an 8:2 ratio. Logistic regression, least absolute shrinkage and selection operator (LASSO), and extreme gradient boosting (XGBoost) models were introduced to establish prediction models. The SHapley Additive exPlanations (SHAP) tool was applied for visualization. The performance of the models was evaluated by using the area under the receiver operating characteristic (AUC of ROC) curves as well as the decision curve analysis (DCA). RESULTS: The predictive factors incorporated in the nomogram included minimum blood urea nitrogen (BUN), maximum partial pressure of oxygen (pO(2)), Charlson comorbidity index, age, duration of mechanical ventilation, weight at admission, Oxford Acute Severity of Illness Score (OASIS), minimum international normalized ratio (INR), severe liver disease, and minimum lactate level. The model exhibited great discrimination ability in the training set and test set, with AUC values of 0.887 [95% confidence interval (CI): 0.863-0.911] and 0.915 (95% CI: 0.884-0.947), respectively. Hosmer-Lemeshow test demonstrated that the model possessed a good fit. DCA analysis demonstrated that the nomogram had a higher clinical net benefit. CONCLUSIONS: The nomogram constructed in this study can accurately predict the risk and prognostic factors of hypoxemic patients who undergo invasive ventilation and receive oxygen therapy after extubation, which will aid in guiding clinical treatment in the future.