Abstract
Background/Objectives Predicting unfavourable outcomes with machine learning (ML) such as discharge morbidity and length of stay has significant benefits in all aspects of patient care, from admission to surgical planning, and discharge. Aneurysmal subarachnoid haemorrhage (aSAH) is a condition associated with significant morbidity and mortality, hence identifying risk factors for poor outcomes and discharge planning with ML may yield significant benefit to patient outcomes and hospital efficiency.
Methodology Data was extracted from a neurovascular database and electronic medical records at the Royal Adelaide Hospital, SA. Patients admitted for aSAH over two-years were included. Patient and aneurysm characteristics including radiological measures were extracted. Discharge Modified Rankin scale (mRS), length of stay and discharge location were extracted. The data was randomly split into a 75%/25% train-test ratio to train ML models including logistic regression, XGBoost, random forest, and decision-tree models. The primary outcome was the area under the curve (AUC) to determine the model's predictive ability.
Results 128 patients were included, with a mean age of 58.5 years (SD 13.1). ML models demonstrated excellent performance in predicting discharge mRS (AUC 0.9), survival to discharge (0.94) and discharge destination (AUC 0.89). World Federation of Neurosurgical Societies grade, Fisher grade and antithrombotic use were strong predictors of poor outcomes.
Conclusion ML models were shown to provide great predictive value for discharge planning from several clinical and radiological variables. Implementation of these models may yield significant benefits to patient outcomes and hospital efficiency. Larger multi-center studies are needed to develop more robust ML models.
