Predicting the risk of in-hospital mortality in traumatic brain injury patients on invasive mechanical ventilation in the intensive care unit: Construction and validation of an online nomogram

Abstract

To explore mortality risk factors and to construct an online nomogram for predicting in-hospital mortality in traumatic brain injury (TBI) patients receiving invasive mechanical ventilation (IMV) in intensive care unit (ICU). We retrospectively analysed TBI patients on IMV in ICU from MIMIC-Ⅳ database and two hospitals. Least Absolute Shrinkage and Selection Operation (LASSO) regression and multiple logistic regression were used to detect predictors of in-hospital mortality and to construct an online nomogram. The predictive performance of nomogram was evaluated using area under the receiver operating characteristic curves (AUC), calibration curves, decision curve analysis (DCA), and clinical impact curves (CIC). 510 from MIMIC-Ⅳ database were enrolled for nomogram construction (80%, n=408) and internal validation (20%, n=102). 185 from two hospitals were enrolled for external validation. LASSO-Logistic regression revealed predictors of in-hospital mortality among TBI patients on IMV in ICU included Glasgow Coma Scale (GCS) after ICU admission, Acute Physiology Score Ⅲ (APS Ⅲ) after ICU admission, neutrophil and lymphocyte ratio after IMV, blood urea nitrogen after IMV, arterial serum lactate after IMV, and in-hospital tracheotomy. The AUC, calibration curves, DCA, and CIC indicated the nomogram had good discrimination, calibration, clinical benefit, and applicability. The multi-model comparisons revealed the nomogram had higher AUC than GCS, APS Ⅲ, and Simplified Acute Physiology Score Ⅱ. We constructed and validated an online nomogram based on routinely recorded factors at admission to ICU and at the beginning of IMV to target prediction of in-hospital mortality among TBI patients on IMV in ICU.