Table 4.

Discriminative ability of clinical rating scales evaluated cross-sectionally.

	Baseline		Year-1 follow-up
Rating scale	Control vs. PROD AUC (P-value)	Control vs. PD AUC (P-value)	Control vs. PROD AUC (P-value)	Control vs. PD AUC (P-value)
UPDRS-III	0.70 (<0.05)∗	1.00 (p < 0.05)∗	0.70 (0.06)	0.99 (<0.01)∗∗
H&Y†	0.53 (0.19)	1.00 (<0.001)∗∗∗	0.58 (0.07)	1.00 (<0.001)∗∗∗
MoCA	0.62 (0.1)	0.72 (0.01)∗∗	0.51 (0.9)	0.58 (0.1)
SDM	0.79 (<0.01)∗∗	0.65 (0.2)	0.80 (<0.01)∗∗	0.76 (<0.01)∗∗

AUC, Area and the ROC curve (perfect discrimination = 1.0, random discrimination = 0.5). P-value represents the statistical significance of the metric computed with logistic regression models corrected for age and gender. ^†H&Y could not be corrected for age and gender given the sample distributions, a Mann–Whitney U-test was used instead. As expected, UPDRS-III and H&Y achieved a near perfect discrimination between Control and Parkinson’s disease (PD) groups. SDM had the best performance in distinguishing PROD from Controls, but failed to reliably measure a difference between Controls and Parkinson’s disease. Bolded values indicate the best metrics among the different ML models (^∗P-value < 0.05, ^∗∗P-value < 0.01, and ^∗∗∗P-value < 0.001).