547 Creation of a Machine Learning Model to Predict Cognitive Outcomes Post Deep Brain Stimulation Surgery for Parkinson’s Disease

Abstract

Abstract Background Parkinson’s disease (PD) is the second most common neurodegenerative disorder and continues to affect an increasingly large portion of the world’s population. Limitations of the medical treatment of PD with levodopa have led to deep brain stimulation (DBS) therapy being used to alleviate symptoms in advanced PD patients. Despite its efficacy with motor symptoms of PD, DBS has been noted to lead to worsening of cognitive outcomes in a percentage of cases. There is currently no predictive model for individual patient cognitive outcomes post DBS. Aim The aim of this study is the creation of a machine learning (ML) model which can help facilitate decision making for clinicians regarding offering DBS to patients. ML has the potential to analyse multiple variables and produce a predictive model for neurocognitive outcomes. The volume of variables analysed by ML is beyond the remit of calculations that can be performed by individual medical professionals for each patient. Method This study used mild cognitive impairment (MCI) as the outcome, at six months post DBS surgery to measure neurocognitive impact. Clinical and radiomics variables were used to train the models. Results Two predictive models were successfully generated – random forest and logistic regression, which showed an accuracy of 0.87 and 0.74 respectively. Conclusions The study concluded that it is possible to use ML to generate a prediction model for outcomes post DBS. The model can help optimise the decision-making process behind offering DBS surgery to patients and lend insight into the possible causes of poor outcomes.