A Data-Driven Operational Definition of Square-Wave Jerks and Its Diagnostic Utility in Progressive Supranuclear Palsy.

Abstract

Background: The Movement Disorder Society diagnostic criteria for progressive supranuclear palsy (PSP) list frequent square-wave jerks (SWJ) as a suggestive oculomotor feature. A standardised operational definition of SWJ is however presently lacking.

Objective: To create a data-driven optimal SWJ definition and to assess its diagnostic utility in PSP.

Methods: We prospectively recruited PSP n = 25, Parkinson's disease (PD) n = 32, Alzheimer's disease (AD) n = 35, and controls n = 88. SWJ were recorded with a visual fixation task using video-oculography. Multiple candidate definitions of SWJ were generated by systematically varying morphological features, producing 32 candidate definitions. A nested cross-validation was used to obtain unbiased diagnostic performance and optimised SWJ definition. Results were then externally tested in a fully independent sample of n = 25 PSP, n = 41 PD and n = 25 controls.

Results: The optimal selected SWJ definition consisted of saccade amplitude 1°-8°, intra-SWJ intervals 20-600 ms, directional opposition within ±45°, and amplitude similarity ≤ 0.25. Only SWJ rate, and not morphological features, was distinctive for PSP. SWJ rate gave out-of-fold AUCs of 0.81 for PSP versus controls, 0.71 for PSP versus AD, 0.63 for PSP versus PD, and 0.76 for PSP versus all non-PSP groups. Post hoc analyses using this data-driven definition on the external validation cohort yielded relatively comparable results. SWJ rate > 24/min had high specificity, but low sensitivity, for PSP when compared with PD.

Conclusions: SWJ rate was significantly increased in PSP but with insufficient discriminant ability for use as a stand-alone diagnostic marker. This data-driven definition may serve as a reproducible operational definition for future research.