Machine Learning-Based Nomogram with Boruta and LASSO for 28-Day Mortality Prediction in Critical COPD Patients: Role...

Despite the known increased mortality of chronic obstructive pulmonary disease (COPD) patients in the intensive care unit (ICU), robust tools for early risk stratification are limited.

A total of 1817 patients with critical care COPD were retrospectively identified from the Medical Information Mart for Intensive Care IV (MIMIC-IV) database and randomly split into training and validation cohorts. Least absolute shrinkage and selection operator (LASSO) and Boruta algorithms were used for variable selection. Cox regression, restricted cubic splines (RCS), and Kaplan-Meier survival analyses with Log rank tests were performed, followed by construction of a nomogram. The nomogram was validated using the C-index, receiver operating characteristic (ROC) analysis, and calibration curves, and its clinical utility was assessed using decision curve analysis (DCA).

RCS and Kaplan-Meier analysis substantiated that patients with high Albumin-Corrected Anion Gap (ACAG) levels had significantly higher 28-day mortality. A nomogram was constructed using 12 predictors identified by combining a statistical learning method (LASSO) and a machine learning algorithm (Boruta). The model demonstrated strong discrimination, with C-indices of 0.747 and 0.766, and area under the curve (AUC) values of 0.772 (95% CI: 0.742-0.801) and 0.795 (95% CI: 0.752-0.838) in the training and validation cohorts, respectively. Multivariable Cox regression confirmed the nomogram as a useful predictor of mortality both in the training cohort and the validation cohort.

This study established ACAG as an independent predictor of 28-day mortality in COPD patients from ICU. The presented nomogram demonstrated robust performance upon internal validation, offering a practical tool for risk stratification.