To develop a clinically relevant, objectively measurable and easy to perform set of motor and non-motor assessments that can categorize People with Parkinson’s Disease (PwPD) with or without freezing of gait (FOG).

FOG is a debilitating motor feature that can be difficult to observe in the clinical and research settings, especially early on when it is predominantly levodopa responsive. Early FOG may go unrecognized for years, making it difficult to develop targeted neuroprotective interventions. Objective measures outside of the freezing episodes are needed to better assess and categorize PwPD at risk.

Sixty-one participants (35 non-freezers, 26 freezers) performed 16 objective motor (gait and balance) and non-motor (cognition, neuropsychiatric, sleep, and quality of life) assessments at a single visit, resulting in 217 variables. A binary regression model was used to iteratively combine variables from these assessments using R studio, for the task of categorizing participants as freezers or non-freezers. The categorization accuracy with a probability threshold of 0.75 was calculated for each iteration. Multicollinearity was assessed for the final model.

Single assessment categorization accuracy ranged from 56%-73%.  After 48 iterations, 14 variables from 8 assessments remained in the final model with a categorization accuracy of 87% (95% CI: 75–94%, p<0.0001).  These included dual-task gait, pivot and semi-circular turns, tandem gait, feet-together stance, and quality of life, apathy and anxiety rating scales and take about 15-20 minutes to perform.

These results suggest that individual assessments have low categorization accuracy, but that there is the potential to create an objective set of assessments that could be used to stratify PwPD as freezers or non-freezers when freezing episodes are not visualized. Using machine learning models with larger datasets in the future would allow a more unbiased approach to developing a shorter assessment panel performable in the clinical setting.