Overcoming the diagnostic gap in mild cognitive impairment in Parkinson's disease: a pilot study employing a machine...

Cognitive impairment is a clinically significant, non-motor symptom of Parkinson's disease (PD) commonly associated with reduced quality of life, increased caregiver burden, and higher risk of progression to dementia. Mild cognitive impairment in PD (PD-MCI) is expressed heterogeneously, with likely prognostic implications. This pilot study evaluated the feasibility and preliminary diagnostic performance of a Machine Learning/Augmented Reality (ML/AR)-based digital assessment for identifying PD-MCI to compare with clinician-led classification.

The Altoida NeuroMarker (hereafter, "NeuroMarker") is a 10 min, self-administered digital cognitive and functional assessment performed by tablet, comprised of thirteen task challenges. The NeuroMarker was administered to 21 patients with PD. NeuroMarker-based MCI classification was compared to clinician-led classification using a confusion matrix to compute sensitivity, specificity, PPV, NPV, accuracy, and Cohen's κ. Clinical assessments included the MMSE, ACE-III, Hoehn and Yahr stage, and BDI.

The NeuroMarker identified all six clinician-classified PD-MCI cases and classified an additional 11 patients with likely MCI. Sensitivity was 100% (95% CI: 54.1-100), specificity was 26.7% (95% CI: 7.8-55.1), PPV was 35.3% (95% CI: 14.2-61.7), NPV was 100% (95% CI: 39.8-100), accuracy was 47.6% (95% CI: 25.7-70.2), and κ = 0.17. Group differences were observed for age, ACE-III, sex, and education.

These preliminary findings suggest that the NeuroMarker may identify clinician-recognized PD-MCI cases, with the potential to also flag patients with early or subthreshold cognitive impairment. However, the study's wide confidence intervals, low agreement, smaller sample size, and absence of longitudinal confirmation limit interpretation. Larger studies utilizing comprehensive neuropsychological assessment and longitudinal follow-up are required.