Parkinson's typing study: post-error recovery time reveals disease severity with p<10⁻⁹

A preprint from arXiv (q-bio.NC) by Navin Bondade analyzed natural typing errors (backspace events) from 57 participants in the MIT-CSXPD dataset, including 27 individuals with Parkinson's disease (PD) and UPDRS-III motor scores. The study tested whether passively collected keystroke dynamics could separate two cognitive-motor stages: noticing an error (error monitoring) versus resuming normal typing rhythm (post-error motor recovery). Using continuous accelerated failure time (AFT) survival models—chosen because discrete time-to-event methods lost the signal—the authors found that post-error recovery time varies strongly with disease severity (p<10⁻⁹ in two independent sub-cohorts), while pre-error keystroke instability does not (r=0.164, p=0.413). The two measures are uncorrelated (r=-0.065), confirming a genuine dissociation.

The result held after controlling for raw typing speed and was cross‐validated with an independent clinical finger‐tapping test (p=0.011). The authors relate this to known electrophysiology: error detection is largely spared in PD, but post-error motor adjustment depends on subthalamic nucleus circuits. This work suggests that everyday typing—requiring no wearables or clinic visits—could serve as a passive digital biomarker for Parkinson's disease severity, potentially enabling low‐cost, large‐scale screening or remote monitoring.