Introduction: Acute respiratory failure (ARF) is a leading cause of intensive care unit admission and is associated with high morbidity and mortality. We aim to develop and validate multivariate logistic regression models to predict in-hospital mortality in critically ill patients of ARF and develop a nomogram. Methods: We analyzed ARF patients from the MIMIC-IV v3.1 database. ARF patients were identified by using ICD codes 9 and 10. Demographics, vital signs, clinical, laboratory, and treatment-related variables were included in our analysis. Missing data was handled by multiple imputations using mice with 5 datasets. Lasso regression was used for selection of variables. A multivariate logistic model was created by using selected variables, and model performance was assessed by C-index and calibration plot. Internal validation was performed by bootstrapping using 500 iterations. A nomogram was developed from the final model. Results: Out of 9833 patients (mean age: 66.8±16.2; 53.9% were male), 2525 (24.7%) died in the hospital. By using LASSO regression, age, race, anion gap, total bilirubin, SAPSII score, respiratory rate, history of diabetes, sepsis, cardiogenic shock, liver disease, AKI, pneumonia, obesity, neurological disorder, and malignancy were identified as significant predictors of in-hospital mortality. The model demonstrated very good discrimination (C index: 0.826). Internal validation using 500-bootstrap calibration demonstrated good calibration (mean absolute calibration error = 0.022), with predicted probabilities closely matching observed outcomes. Conclusions: We develop and validate logistic regression model and nomogram for predicting in-hospital mortality in critically ill patients with acute respiratory failure. Our model demonstrated good discrimination ability and can serve as a potential tool for early identification of high risk ARF patients. Future studies should focus on external validation of model and its clinical application.
Basit et al. (Sun,) studied this question.