Objective: To explore the feasibility of collecting voice samples using digital voicemail in People with Parkinson’s disease (PwPD).

Background: For many PwPD, the increased adoption of telehealth services that occurred during the pandemic could improve their access to clinical care and research opportunities. In a rural state like Arkansas, where 3 of the 4 movement disorders trained neurologists practice at one location, determining the feasibility of remote monitoring is particularly important.

Method: Voice samples were collected using digital voicemail from 50 PwPD and 49 control adults (CA) who reported no history of neurologic, psychiatric, or speech disorders. Participants used their cellular or land-line phones to leave a voicemail that included two tasks: sustaining the sound “Ah” for at least 3s, and reading the Rainbow Passage (RP). The voicemail system digitized the analog recording at 8kHz. Acoustic features derived using Praat software included fundamental frequency, harmonic-to-noise ratio, local jitter, and local shimmer, as these have previously been shown to be affected in PwPD. Mel-spectrograms were generated and converted to image files, and a convolutional neural network (CNN) was then trained to classify these images as coming from PwPD or CA. PwPD also underwent additional evaluations, including Unified Parkinson’s disease Rating Scale (UPDRS) and Montreal Cognitive Assessment (MoCA).

Results: The CNN classifiers were found to be 94% accurate in distinguishing PwPD from CA for the Ah samples and 76% accurate for the RP samples. Disease duration for PwPD was positively correlated with the standard deviation of fundamental frequency on both the Ah sample (r=0.41; p<0.004) and the RP (r=0.49; p<0.001), as well as duration of the RP (r=0.41; p<0.004) and jitter in the RP (r=0.50; p<0.001). MoCA scores were inversely correlated with durations of both the Ah (r=-0.41; p<0.004) and the RP samples (r=-0.52; p<0.001), and the total UPDRS score was positively correlated with duration (r=0.37; p<0.01) and jitter of RP samples (r=0.40; p<0.005).

Conclusion: These findings support the feasibility of using telephonic voice samples to remotely collect features that distinguish PwPD from CA in a rural population who might otherwise find it difficult to access care or participate in research protocols.

« Back to 2022 International Congress

MDS Abstracts - https://www.mdsabstracts.org/abstract/feasibility-of-telephonic-voice-samples-to-remotely-monitor-people-with-parkinsons-disease-residing-in-medically-underserved-rural-regions/