Construction and validation of a tracheostomy prediction model in mechanically ventilated stroke patients and the impact of early versus late tracheostomy on clinical outcomes: an IPTW-based analysis

Abstract

Background This study aimed to develop and validate a predictive model for tracheostomy in mechanically ventilated stroke patients and to investigate the impact of early and late tracheostomy on in-hospital outcomes. Methods A total of 508 mechanically ventilated stroke patients who were admitted to a tertiary hospital between January 2022 and January 2025 were retrospectively enrolled and divided into the tracheostomy and non-tracheostomy groups. Patients were randomly split into a training set (n = 356) and a validation set (n = 152) at a ratio of 7:3. Least absolute shrinkage and selection operator (LASSO) regression combined with the importance of Random Forest feature was used to identify key variables. Independent predictors were identified using the multivariable logistic regression, and a nomogram was constructed. Model performance was evaluated using receiver operating characteristic (ROC) curves, calibration curves, and decision curve analysis (DCA). Inverse probability of treatment weighting (IPTW) based on propensity scores was applied to assess the effects of early (≤7 days) versus late (>7 days) tracheostomy on clinical outcomes. Results A multivariable logistic regression identified midline shift, hypoalbuminemia, admission Glasgow Coma Scale (GCS) score, C-reactive protein (CRP), and prognostic nutritional index (PNI) as independent predictors (p < 0.05). The area under the curve (AUC) was 0.855 in the training set (sensitivity: 75.72%, specificity: 79.78%) and 0.849 in the validation set (sensitivity: 70.04%, specificity: 84.30%), indicating good discriminative ability. The Hosmer–Lemeshow test demonstrated good calibration (training set: χ2 = 6.943, p = 0.543; validation set: χ2 = 13.547, p = 0.094). DCA showed that the model provided a favorable net clinical benefit within a certain threshold range. IPTW analysis indicated that early tracheostomy significantly reduced ICU length of stay but had no significant effect on post-tracheostomy ventilation duration, antibiotic use duration, total hospital stay, or hospitalization costs. Conclusion The nomogram developed in this study demonstrated good performance in predicting the risk of tracheostomy in mechanically ventilated stroke patients and enabled individualized real-time risk assessment via a web-based tool. Early tracheostomy may help shorten ICU stay and could inform clinical decision-making.