Objective: We designed an algorithm to detect response to medication in an early PD population using at-home, unsupervised, unobtrusive typing data.

Background: Advances in technology are opening a new era to remotely evaluate people with PD. In previous studies we have shown that features of in­lab keyboard typing can be used to evaluate motor skills and to classify subjects as having PD or not [1]. More recently we have shown the same capability from typing on a touch­screen based information and from keyboard data at-home [2]. We now hypothesized that typing on an electronic device, a habitual behavior, likely controlled by the nigro­striatal dopaminergic pathway, could allow for objectively and non­-obtrusively monitoring parkinsonian features and response to medication in an at­-home setting.

Methods: We designed a naturalistic prospective validation study to evaluate whether typing patterns changed in accordance with responsiveness to medication. 31 early PD subjects, who were going to start a dopaminergic drug, and 30 matched controls were enrolled. We remotely monitored their typing pattern over a 6-­month follow­-up period while antiparkinsonian medications were being titrated (fig.1). A novel deep learning algorithm (nQRNN) was developed to detect participants’ outcome defined as the response to medication assessed by the UPDRS-­III minimal clinically important difference (MCID) at the final visit (6 months). Further, we tested if this model could predict that outcome earlier than 6 months.

Results: The nQRNN had an overall moderate kappa agreement (k=0.50) and fair 0.73 area under the ROC curve with the time­-coincident UPDRS-­III MCID-­based classification of response (fig.2). Furthermore, the nQRNN at week 3 (and beyond) could reliably predict which subjects would respond and which wouldn’t. (fig.3)

Conclusions: This preliminary study suggests that a habitual task based on remotely-­gathered unsupervised typing data at home allows for an accurate and predictive classification of drug response in PD. If confirmed by a larger prospective study, this approach could provide supplementary information to clinicians for a more continuous monitoring of motor symptoms of PD, thus helping to take informed decision on therapeutic strategies and disease management. Also, this tool could be useful as a cost­ effective and reliable outcome measure for clinical trials to test response to medication.

References: 1. Giancardo L, Sánchez-Ferro Á, Arroyo-Gallego T, et al. Computer keyboard interaction as an indicator of early Parkinson’s disease. Sci Rep 2016;6:34468. doi:10.1038/srep34468. 2. Arroyo-Gallego, T, Ledesma-Carbayo, M, Butterworth, I, et al. Detecting Motor Impairment in Early Parkinson’s Disease via Natural Typing Interaction with Keyboards: Validation of the neuroQWERTY Approach in an Uncontrolled At-Home Setting. JMIR 2018 in press. doi: 10.2196/jmir.9462.

« Back to 2018 International Congress

MDS Abstracts - https://www.mdsabstracts.org/abstract/objective-monitoring-of-drug-response-in-early-pd-patients-using-remote-at-home-typing-data-through-machine-learning-analysis/