The value of machine learning models in predicting factors associated with the need for permanent shunting in patients with intracerebral hemorrhage requiring emergency cerebrospinal fluid diversion.

Abstract

Objective: To assess the efficacy of machine learning models (MLMs) in identifying factors associated with the need for permanent ventricular shunt placement in patients experiencing intracerebral hemorrhage (ICH) who require emergency cerebrospinal fluid (CSF) diversion.

Methods: A retrospective review was performed on patients with ICH requiring urgent CSF diversion who were admitted to our facility between July 2009 and May 2023. A binary logistic regression analysis was carried out to determine independent predictors linked to the development of shunt-dependent hydrocephalus following ICH. Five different machine learning models-random forest (RF), support vector machine (SVM), k-nearest neighbor (k-NN), logistic regression (LR), and Adaptive Boosting (AdaBoost)-were utilized to predict the need for permanent shunting in those with spontaneous ICH necessitating emergency CSF diversion. Additionally, RF techniques were applied to identify the factors affecting the need for permanent ventricular shunt placement in these patients.

Results: A total of 578 patients were included in the analysis. Shunt-dependent hydrocephalus occurred in 121 individuals (20.9%). In the multivariate analysis, the Graeb Score, the length of time the external ventricular drain (EVD) was in place, and an elevated intracranial pressure (ICP) greater than 30 mm Hg were significant predictors for the need for permanent CSF diversion (p<0.05). All predictive models showed commendable performance, with RF achieving the highest accuracy (0.921), followed by SVM (0.906), k-NN (0.889), LR (0.881), and AdaBoost (0.823). RF also excelled over the other models in terms of sensitivity and specificity, with a sensitivity of 0.912 and specificity of 0.892. The area under the curve (AUC) values for RF, SVM, k-NN, LR, and AdaBoost were recorded at 0.903, 0.820, 0.804, 0.801, and 0.798, respectively.

Conclusion: This research demonstrates that machine learning models can effectively predict the need for permanent CSF diversion in patients with ICH who underwent EVD placement for urgent CSF diversion, offering important prognostic insights that could facilitate early intervention and lead to potential cost reductions.