Comparison Among Machine Learning Algorithms for Predicting Shunt-Dependent Hydrocephalus

Hydrocephalus is one of the leading causes of mortality and morbidity in neurosurgery. Extraventricular drain placement is a common first-line treatment for acute HCP patients. Approximately 6.5-25% of HCP is needed for the subsequential ventriculoperitoneal shunt. The primary objective of the present study was to compare classification performances among various supervised ML algorithms. Moreover, the secondary objective was to investigate factors associated with shunt-dependent HCP. The retrospective cohort study was conducted on 412 patients with hydrocephalus who underwent extraventricular drain placement. Clinical features and ventricular indexes were collected. Supervised ML algorithms were trained and tested for classification between the subsequential ventriculoperitoneal shunt group and the successful extraventricular drain removal group. As a result, the logistic regression algorithm had excellent performance, with an area under the receiver operating characteristic curve of 0.92. In summary, these ML algorithms are useful to predict shunt-dependent HCP and may be implemented in clinical settings as preliminary screening instruments to decrease unnecessary procedures.