Hamed Asadi, Richard Dowling, Bernard Yan and Peter Mitchell
Introduction: Subarachnoid haemorrhage due to aneurysm rupture is a major cause of death and disability. Accurately predicting the outcome for those patients who have endovascular treatment from a set of predictive variables may identify high-risk patients and guide treatment approaches, leading to decreased morbidity. Logistic regression models allow for the identification and validation of predictive variables. However, advanced machine learning algorithms offer an alternative, in particular, for large-scale multi-institutional data, with the advantage of easily incorporating newly available data to improve prediction performance. Our aim was to design and compare different machine learning methods, capable of predicting the outcome of endovascular intervention in acute subarachnoid haemorrhage and aneurysm rupture.
Method: We conducted a retrospective study on a prospectively collected database of patients with acute subarachnoid haemorrhage due to aneurysm rupture who underwent endovascular intervention. All demographic, clinical and procedural data was collated including information from follow up imaging studies. Using SPSS®, MATLAB® and RapidMiner®, classical statistics as well as machine learning algorithms were applied to design a supervised machine capable of classifying these predictors into potential good and poor outcomes. It was attempted to predict the final patients’ outcome based on modified Rankin Scale (mRS), and a dichotomised outcome, good or bad, as well as mortality, recanalization rate and need for retreatment. Subsequently, these algorithms were train
Results: We included 236 consecutive acute subarachnoid haemorrhage patients with ruptured intracerebral aneurysm treated by endovascular technique, with a mean age of 52.7 (SD=13.7). All the available demographic, procedural and clinical factors were included into the models. The overall accuracy in predicting the exact mRS was just below 50%, which increased to above 75% in prediction of the dichotomised (good or bad) outcome, and approximately 85% in prediction of mortality. Prediction of recanalization had an overall accuracy of just below 50%; however, there was an approximately 90% accuracy in prediction of those patients requiring retreatment.
Discussion: We showed promising accuracy of outcome prediction, using supervised machine learning algorithms in particular in prediction of final outcome as good or bad as well as the probability of needing retreatment in future, with potential for incorporation of larger multicenter datasets, likely further improving predictive accuracy. Finally the filtered and optimized dataset was introduced into a decision induction module and a simplified prognostication tree was designed representing a pictorial relationship between the predictors and the final outcome in a relatively easy to interpret way.
English 
Spanish 
Chinese 
Russian 
German 
Japanese 
Portuguese 
Hindi