Prediction model of central nervous system infections in patients with severe traumatic brain injury after craniotomy

Abstract

The aim of this study was to develop and evaluate a nomogram to predict CNS infections in patients with severe traumatic brain injury (sTBI) after craniotomy. This retrospective study was conducted in consecutive adult patients with sTBI who were admitted to the neurointensive care unit (NCU) between January 2014 and September 2020. We applied the least absolute shrinkage and selection operator (LASSO) and multivariate logistic regression analysis to construct the nomogram, and k-fold cross-validation (k=10) to validate it. The receiver operator characteristic area under the curve (AUC) and calibration curve were applied to evaluate the predictive effect of the nomogram. The clinical usefulness was investigated by decision curve analysis (DCA). A total of 471 patients with sTBI who underwent surgical treatment were included, of whom 75 patients (15.7%) were diagnosed with CNS infections. The serum level of albumin, cerebrospinal fluid (CSF) otorrhoea at admission, CSF leakage, CSF sampling, and postoperative re-bleeding were associated with CNS infections and incorporated into the nomogram. The results showed that our model yielded satisfactory prediction performance with an AUC value of 0.962 in the training set and 0.942 in the internal validation. The calibration curve exhibited satisfactory concordance between the predicted and actual outcomes. The model had good clinical use since the DCA covered a large threshold probability. We established a straightforward individualized nomogram for CNS infections in sTBI patients in the NCU, which could help physicians screen high-risk patients to perform early interventions to reduce the incidence of CNS infections in sTBI patients.