Designing predictive models for appraisal of outcome of neurosurgery patients using machine learning-based techniques

Abstract

Recently, machine learning (ML) methods have been progressively applied in medical researches. Predicting ICU outcome helps hospitals to better manage their resources. In addition, by recognition of the most important factors in predicting ICU outcome, some new protocols may be developed to consider these factors in the treatment. The present study is a descriptive applied study in the field of ML and intensive care unit (ICU) to determine the predicting factors associated with the outcome of neurosurgery ICU. Data of 1200 patients admitted to the neurosurgery ICU, [BLINDED FOR REVIEW] during the period of 2017 to 2019 was studied. Demographic features (age, sex, weight, height), comorbidities, ICU admission time, ICU re-admission, smoking and related complications were collected. Neighborhood Component Analysis (NCA) was used as a non-parametric strategy for choosing features with the aim of enhancing prediction accuracy of regression and classification algorithms. Using this approach, we found that the most important predictor factors for the outcome of neurosurgery patients were age, weight, hypertension, diabetes insipidus, diabetes mellitus, prior operation, smoking and alcohol, sex, blood glucose, ischemic heart disorder, meningitis, acute kidney injury, tracheostomy, prior ICU admission, seizure, and fever. The application of ML enables the prediction of outcomes of neurosurgery patients.

Ähnliche Arbeiten