Table 2:

GEE regression models for the effect of PPT score on cognitive impairment

Independent variable type	Outcome variable (≤ 10th percentile on indicated test)	Odds ratio	95% CI (low)	95% CI (high)	Adjusted p
Baseline PPT score	SDMT	0.81	0.69	0.94	0.043
	Phonemic fluency	0.85	0.71	1.02	0.439
	Delayed recognition	0.94	0.86	1.02	0.535
	Delayed recall	0.97	0.90	1.04	1.000
	New dot	0.91	0.83	1.01	0.436
	Digit-span	0.99	0.93	1.05	1.000
	Odd-man-out	1.00	0.89	1.13	1.000
PPT score (time-varying)	SDMT	0.79	0.68	0.93	0.021
	Phonemic fluency	0.78	0.67	0.92	0.020
	Delayed recognition	1.04	0.96	1.13	0.891
	Delayed recall	0.98	0.91	1.06	1.000
	New dot	0.93	0.84	1.02	0.521
	Digit-span	0.98	0.89	1.08	1.000
	Odd-man-out	0.88	0.79	0.98	0.103

Two GEE logistic regression models were constructed for each cognitive test (dependent variable). For models measuring the effect of baseline PPT (score at enrollment) on cognitive test score over time, baseline values were also used for all other adjusting covariables. For the non-baseline models, the values at each follow-up assessment were used for all covariables. Baseline cognitive test score was used to adjust all models. Odds ratios (ORs) and 95% confidence intervals (95% CI) correspond to the relationship between bimanual PPT score and impaired cognitive test score (≤ 10^th percentile). P-values for baseline and longitudinal models were separately adjusted using the Holm method, using the number of tests (7) as the number of multiple comparisons. Covariables adjusted for in all models include sex, education (years), age, disease duration, UPDRS III subsections (rigidity, postural tremor, postural instability, hand movements), and Hoehn & Yahr stage.