USING COMPUTATIONAL ARGUMENTATION TO INTEGRATE ACTIVITY AND QUALITY OF LIFE DATA TO PREDICT DISEASE PROGRESSION IN PATIENTS WITH HIGH-GRADE GLIOMA

Abstract AIMS Many patients with high-grade glioma report weakness, balance problems and fatigue. Patient assessment tools capture a snapshot of health state. New wearable technology allows the collection of longitudinal physical ac- tivity (PA) data from the patient directly. However, PA data is fraught with issues of volume, fragmentation and heterogeneity. Computational Argumentation is a developing AI approach that has the advantage of explain- ability and transparency. We have previously published the detailed technical aspects of the work; here we focus on the clinical data. METHODS 31 HGG participants from the BrainWear clinical trial provided acceleration data and EORTC QLQ-C30/BN20 questionnaires. Patients were classified as having progressive or stable disease at multiple timepoints (deter- mined by MRI scans or mortality status). We built an Argumentation-based AI model that treats each data point as an argument in favour of its labelling. Arguments consider clinician preferences to determine which features are most important. We assessed the performance of the system by comparing the classification of unlabelled data points with existing AI techniques. RESULTS We identified 110 timepoints from 31 patients. We used 60 for model training and 50 for testing. Our novel argumentation-based approach achieves a prediction accuracy of 80% on the test set. Our approach outper- forms similar interpretable models, with a decision tree and a k-nearest neighbour achieving 75% and 70% respectively. CONCLUSION This study leverages accelerometer and questionnaire data as predictive measures of progressive disease. We use a model that reasons with patient data, using past patients as evidence for the prediction made about a new patient. This transparent, fully interpretable model considers clinician preferences in the reasoning system, leading to increased performance. with the final debate providing patient-centred reasoning for the outcome predicted.