Objective: To capitalise on the ubiquity of smartphones and to develop tools to objectively assess symptoms associated with PD.

Background: Accurate and reproducible outcome measures resistant to the inherent inter- and intra-rater variability associated with clinician derived measures of disease change are critically needed to inform PD research.

Methods: We obtained smartphone recordings from deeply phenotyped participants enrolled in a large longitudinal cohort study involving participants with early PD and healthy controls. Participants performed tasks assessing voice, balance, gait, dexterity, reaction time, rest and postural tremor. 2674 time-synchronised recordings of all 7 tasks were analysed from 329 participants with PD (63% male, mean age 68.1 years, standard deviation 9.3 years, mean Movement Disorder Society Unified Parkinson’s Disease Rating Scale (MDS-UPDRS) III score 28.7, standard deviation 12.5). In total, 998 features were extracted. Using the smartphone-based features, machine learning algorithms were employed to predict scores derived from semi-quantitative tests of motor function, namely the Purdue pegboard test, Timed up and go and the Flamingo test as well as the MDS-UPDRS part III, Montreal Cognitive Assessment score and Beck Depression Inventory. Model accuracy was evaluated using a 10-fold cross validation scheme, whereby the data was randomly split into training and test sets comprising 90% and 10% of the data respectively.

Results: Having demonstrated around 85% sensitivity and specificity in distinguishing PD from healthy controls using smartphone motor testing, we also predict semi-quantitative tests of motor function and cognition with relatively high levels of accuracy. This includes the prediction of the motor MDS-UPDRS score with a mean absolute error of 4.9 points, within previously observed limits of inter-rater variability of between 1.7 and 5.4 points.[1]

Conclusions: Objective smartphone assessments of voice and movement accurately predict clinical scores in early PD. Advantages include low cost, high-frequency, data capture across the clinic and home environment, with the potential for individual stratification and treatment monitoring.

References: 1. Post B, Merkus MP, de Bie RM, et al. Unified Parkinson’s disease rating scale motor examination: are ratings of nurses, residents in neurology, and movement disorders specialists interchangeable? Movement disorders: official journal of the Movement Disorder Society 2005;20(12):1577-84. doi: 10.1002/mds.20640 [published Online First: 2005/08/24].

« Back to 2018 International Congress

MDS Abstracts - https://www.mdsabstracts.org/abstract/the-use-of-smartphone-task-derived-features-to-predict-clinical-scores-in-parkinsons-disease-pd/