Development and validation of the STeP score for predicting tracheostomy in patients with sepsis using a nationwide...

Among patients with sepsis admitted to the intensive care unit (ICU), a substantial proportion require mechanical ventilation, and a subset eventually undergo tracheostomy. Early identification of patients at high risk for tracheostomy may facilitate timely decision-making and improve clinical communication.

We conducted a nationwide, retrospective study using the Japanese Intensive care PAtient Database (JIPAD). Adult patients with sepsis (Sequential Organ Failure Assessment score of ≥ 2, excluding viral pneumonia) who required mechanical ventilation between 2018 and 2022 were included. The primary outcome was tracheostomy within 14 days of ICU admission. Seventy-five variables available within 24 h of ICU admission were collected. Using least absolute shrinkage and selection operator (LASSO) regression with tenfold cross-validation, we selected predictors to build a multivariable logistic regression model (Sepsis Tracheostomy early Prediction [STeP] model). A simplified scoring system (STeP score) was also derived. Predictive performance was assessed using the area under the curve (AUC) of the receiver operating characteristic (ROC) in a temporally independent validation cohort.

Among 7357 eligible patients (training: 5374; validation: 1983), 1013 (13.8%) underwent tracheostomy. The STeP model, based on 8 LASSO-selected variables, demonstrated good discrimination (AUC: 0.76 in training, 0.74 in validation). The simplified STeP score (range, 0-17), derived from the same predictors, achieved an AUC of 0.73 in the validation cohort. Patients were stratified into low (≤ 2 points), moderate (3-6 points), and high (≥ 7 points) risk groups for tracheostomy, with corresponding tracheostomy rates of 4.0%, 13.6%, and 27.1%, respectively.

We developed and validated a robust prediction model and simplified risk score (STeP score) for tracheostomy within 14 days in ICU patients with sepsis. Early risk stratification using variables available within 24 h may support timely tracheostomy planning. A web-based calculator is publicly available to facilitate bedside implementation.