Optimal MMSE and MoCA cutoffs for cognitive diagnoses in Parkinson's disease: A data-driven decision tree model

Detecting cognitive impairment in Parkinson's disease (PD) is challenging due to diverse manifestations and outdated diagnostic criteria. Cognitive screening tools, as Mini-Mental State Examination (MMSE) and Montreal Cognitive Assessment (MoCA), are adopted worldwide, but despite several cutoffs has been proposed for PD, no consensus has been reached, hindered by limited sample sizes, lack of validation, and inconsistent age- and education-adjustments.

Determine the optimal MMSE and MoCA cutoffs in a large PD cohort, spanning from normal cognition (PD-NC), mild cognitive impairment (PD-MCI) to dementia (PDD), and develop a decision tree model to assist physicians in cognitive workups.

Our retrospective Italian multicenter study involves 1780 PD, cognitively diagnosed with a level-II assessment: PD-NC(n = 700), PD-MCI(n = 706), and PDD(n = 374). Optimal cutoffs (for raw scores) were determined through ROC analysis. Then, a machine learning approach-decision trees-was adopted to validate and analyze the possible inclusion of other relevant clinical features.

The decision tree model selected as primary feature a MMSE cutoff ≤24 to predict dementia, and a score ≤ 27 for PD-MCI. To enhance PD-MCIvs.PD-NC accuracy, it also recommends including a MoCA score ≤ 22 for PD-MCI, and > 22 for PD-NC. Age and education were not selected as relevant features for the cognitive workup. Both MoCA and MMSE cutoffs exhibited high sensitivity and specificity in detecting PD cognitive statues.

For the first time, a clinical decision tree model based on robust MMSE and MoCA cutoffs has been developed, allowing to diagnose PD-MCI and/or PDD with a high accuracy and short administration time.