Two serial MoCA assessments may support biomarker-sparing triage between Parkinson’s disease/Lewy body dementia and frontotemporal dementia: a progressive inclusion analysis of 1,129 participants
Abstract
Reliable differentiation of Parkinson’s disease/Lewy body dementia (PD/LBD) from frontotemporal dementia (FTD) affects treatment strategy and clinical trial eligibility, yet confirmatory biomarker testing remains costly and unevenly available. Whether routine cognitive trajectories can support testing prioritization has not been systematically quantified. We performed a progressive inclusion analysis of 1,129 National Alzheimer’s Coordinating Center participants with PD/LBD ( n = 385) or FTD ( n = 744) to determine the minimum number of serial Montreal Cognitive Assessment (MoCA) administrations required for diagnostic separation. Random forest classifiers used seven MoCA subdomain slopes from the first k chronological assessments ( k = 2 through k = 8), with five-fold stratified cross-validation and 500-iteration bootstrap confidence intervals. In the primary full-cohort analysis, two assessments yielded AUC = 0.785 (95% CI 0.757–0.814; sensitivity = 0.922; specificity = 0.525). The interval-restricted 6–12 month subset yielded AUC = 0.837 (95% CI 0.787–0.885), near the lower edge of published biomarker-panel ranges in an indirect comparison, although its lower confidence bound remained below 0.85. Discrimination persisted after age matching (AUC = 0.794, 95% CI 0.761–0.826; residual age gap = −0.3 years) and age restriction to 55–75 years (AUC = 0.777, 95% CI 0.741–0.814). Age alone yielded lower discrimination (AUC = 0.721, 95% CI 0.691–0.751) than MoCA slopes, while slopes plus age yielded AUC = 0.867 (95% CI 0.845–0.888). Four-assessment performance was AUC = 0.831 (95% CI 0.791–0.873), consistent with an apparent sample-size-limited plateau as eligible N contracted from 391 at k = 4 to 98 at k = 6. These findings support serial MoCA trajectory analysis as an exploratory tool for prioritizing confirmatory diagnostic testing, with age sensitivity, subtype sensitivity, and operating-point performance quantified.