P.203Towards high-resolution clinical digital biomarkers for Duchenne muscular dystrophy

Clinical trials for Duchenne muscular dystrophy (DMD) assess functional performance in artificial set-ups, relying on subjective and motivation-dependant tasks. Furthermore, current outcome measures are subject to intra-rater & inter-rater variability in the patient performance itself. As a consequence, a larger sample size and longer duration of clinical trials are required, often delaying access to novel therapies. In the KineDMD study, we pursue a patient-centric approach to obtain a complete data-driven picture of full-body behavioural capacity in real-life over a 12 month period. Our Ethomics approach (Ethology & Omics) has 2 sides: "At home" & "In clinic". "In clinic" we focus on full-body motion analysis using high-resolution wearable sensors that can track full-skeleton movements of patients during standardised clinical assessments, therefore reducing the subjective assessment of clinical endpoints by data-driven objective analytics. DMD boys and age-matched healthy boys (HC) are assessed in the clinic during 6-monthly visits in conjunction with physio assessments whilst wearing a sensor suit of 17 sensors velcro-attached to normal clothing, capturing full body kinematics of 22 joints. To this end we are developing a number of machine learning algorithms that analyse the behaviour. Our interim analysis of the 15 DMD subjects and 10 HC reveal that motor coordination in HC and DMD show algorithmically identifiable static and dynamic differences. Our Gaussian Process regression approach combining features of individual gait-cycles of the 6-minute walk test was able to estimate from full-body kinematics the North Star Ambulatory Assessment (NSAA) total score of the boys with good accuracy. It is also possible to predict the NSAA scores (both cross-sectional and longitudinal) from the suit data of the NSAA assessment. The predicted and actual scores were highly correlated, supporting the notion that machine learning can accurately predict the results of functional assessments. The ultimate goal of this study is to derive new digital biomarkers, which are independent of artificial set-up and subjective biases and can be assessed in real life/home environment, to capture individual variations in disease progression and benefit to therapies.