Fair and Accessible Parkinson’s Disease Screening using a Machine Learning-Powered Web Platform: Research Protocol and Preliminary Results

Digital technologies offer unprecedented opportunities to screen for conditions like Parkinson’s Disease (PD) in a scalable and accessible manner. With the widespread adoption of smartphones and computers, the general public is constantly interacting with digital interfaces, leaving behind a wealth of data that can be harnessed for health screening. Keystroke dynamics, touchscreen interactions, and other digital footprints have emerged as potential indicators of PD. By analyzing patterns in keyboard typing, touchscreen gestures, and other digital indicators, it is now possible to detect subtle motor impairments associated with PD. We propose to further develop, refine, and validate a baseline predictive model for Parkinson’s disease (PD) based on keystroke and touchscreen measurements which we have developed and tested on participants in Hawaii. Through extensive experimentation, the project aims to determine the optimal combination of features that yield the highest sensitivity and specificity in distinguishing participants with and without PD while algorithmically reducing disparities in performance across race and socioeconomic status. A central challenge of this research will be ensuring fairness by mitigating biases caused by differences in laptop and desktop screen dimensions, mouse responsiveness, and other configurations. These differences are likely to vary by socioeconomic status, requiring a thorough analysis of these disparities and employment of algorithmic fairness techniques to mitigate the underlying problem. Additionally, we will conduct human-centered design sessions to understand how to create such screening tools in a manner that is sensitive to Indigenous data sovereignty. Our findings will underscore the potential of leveraging technology-measured limb movement data as a reliable and accessible method for early detection of PD. This research holds promise for screening individuals who may potentially be affected by PD earlier in an accessible and scalable manner, thus reducing socioeconomic health disparities related to early screening and diagnosis.