Are Standardized Tests Sensitive to Early Cognitive Change in Parkinson’s Disease?

Abstract

Introduction

Cognitive deficits within the first years of Parkinson’s disease (PD) diagnosis are commonly reported, and progression to dementia greatly impacts independence. Identifying measures sensitive to early changes is critical for trials of symptomatic therapies and neuroprotection.

Methods

A sample of 253 newly diagnosed PD patients and 134 Health Controls (HC) completed a brief cognitive battery annually over a 5-year period through the Parkinson’s Progression Markers Initiative (PPMI). The battery included standardized measures of memory, visuospatial functions, processing speed, working memory, and verbal fluency. Inclusion criterion for HCs was performance above a cutoff for possible Mild Cognitive Impairment (pMCI) on cognitive screening (MoCA ⩾ 27) The PD sample was therefore divided to match HCs on baseline cognitive testing (PD-normal n = 169; PD-pMCI n = 84). The multivariate approach to repeated measures examined rates of change between groups on cognitive measures.

Results

An interaction indicating slightly greater decline over time in PD-normal relative to HCs was observed on a measure of working memory: letter-number sequencing. Differential rates of change were not observed on any other measures. Motor symptoms on the dominant right upper extremity accounted for performance differences on a test with writing demands (Symbol-Digit Modality Test). PD-pMCI performed worse than PD-normal on all cognitive measures at baseline, but did not decline faster.

Discussion

Working memory appears to decline slightly faster in early PD compared to HCs, while other domains remain similar. Within PD, faster decline was not associated with lower baseline cognition. These findings have implications for clinical trial outcome selection and study design.

Introduction

Cognitive impairment in Parkinson’s disease is common and contributes significantly to diminished quality of life and loss of independence. Mild deficits have been reported early in the course of illness, with several studies reporting about 20% meeting criteria for Mild Cognitive Impairment (MCI) at time of diagnosis, increasing to 40–50% after 5 years,¹ and progression to dementia in about half after 10 years.² A number of clinical trials using disease-modifying interventions have attempted to target cognitive decline early in the course of PD. Efficacy outcomes for these trials are often based on cognitive screening tests, such as the Montreal Cognitive Assessment (MoCA)³ or Mattis Dementia Rating Scale (DRS-2),⁴ or composite measures from neuropsychological test batteries (e.g., NCT02953665, NCT04180865, NCT03100149, NCT01560754, NCT02642393, etc.). However, the natural progression of early PD-related cognitive decline, as captured by such tests, is not well-established.

The Parkinson’s Progression Markers Initiative (PPMI)⁵ is a multicenter longitudinal investigation to comprehensively characterize PD motor and non-motor symptom progression. The primary cohort includes 423 individuals with PD enrolled early in the course of illness, prior to initiation of dopaminergic therapy, and 195 neurologically healthy controls (HC). These participants completed cognitive testing at baseline and yearly follow-up evaluations over a 5-year span. Notably, presence of MCI, as indicated by performance below a cutoff score of 27 on the MoCA, was an exclusionary criterion for HC but not PD participants. Several papers have been published on this dataset documenting emergence of cognitive deficits in PD. For instance, Weintraub and colleagues found decline over 5 years on MoCA scores was minimal in PD (2%) and no different than HC.⁶ However, no studies have examined cognitive change over time in PD patients who are similar to HC at baseline according to MoCA performance. Such information is critical for selecting measures that might indicate response to disease-modifying therapy, as many clinical trials also exclude individuals with question of MCI. The primary purpose of this study was to determine whether PD patients who were cognitively similar to HCs at baseline would evidence differential change over time on measures from the PPMI cognitive test battery. We also sought to determine whether possible MCI status at baseline in the PD group (MoCA ⩽ 26) was associated with different trajectories of change on the test battery.

Methods

Sample: The PD sample was divided based on MoCA scores at baseline to match HCs, with the matched PD sample (PD-normal) having MoCA scores ⩾ 27, and PD with possible MCI (PD-pMCI) having MoCA scores ⩽ 26. It is noted that this cutoff is slightly more conservative than the more commonly used cutoff score of ⩽ 25 for identifying PD-MCI,⁷ and does not take into account subjective reporting of decline relevant to formalized diagnostic criteria.⁸ All PPMI participants who completed cognitive assessments at all time points from baseline to 5-year follow-up were included in the analysis.

Cognitive Tests: All cognitive tests administered in PPMI were examined for change over time. This included total correct for the Judgment of Line Orientation Test (JOLO),⁹ Letter-Number Sequencing (LNS),¹⁰ and Symbol Digit Modalities Test (SDMT).¹¹ Semantic verbal fluency¹² was based on the total correct for Animals, Fruits, and Vegetables condition. From the Hopkins Verbal Learning Test (HVLT-R),¹³ Total Learning and Delayed Recall were examined. Of note, all six (6) parallel forms of the HVLT-R were used in this study to mitigate practice effect, with administration order fixed sequentially from 1–6. Given common use of cognitive screening tests as efficacy outcomes, we examined change over time on the MoCA. We also explored whether a more generalized pattern of cognitive decline, not fully captured by individual tests, was occurring at differential rates in PD. A global cognitive composite score was derived using all test outcomes except the MoCA (as this itself is a composite). Means and standard deviations from the entire HC sample at baseline (n = 195) were used to derive z-scores for each measure. The global composite thus reflected the participant’s average z-score on the test battery at each time point.

Statistical Analysis: Group differences at baseline were evaluated using independent-sample t-tests. Group differences in cognitive performance over time were examined using the multivariate approach for repeated measures. Mauchly’s test was evaluated to assess assumption of sphericity. Type-I error control was set at two-tailed alpha = .05 for all measures except for the two HVLT-R outcomes, where the Bonferroni approach was used to control for increased within-family false discovery risk (alpha = .025).

Results

Group Composition: Complete datasets for all time points were obtained from 134 Healthy Controls, 169 PD-normal and 84 PD-pMCI, representing 68.7%, 59.9%, and 59.7% of the originally enrolled groups, respectively. Retention rate was not significantly different between HC and PD, X² (2, 622) = 3.261, p = .071, or PD cognition groups X² (2, 423) = .005, p = .944. No significant groups differences were seen for education or gender composition. However, as shown in the Table, groups were found to differ by age, with PD-pMCI being older at time of entry (mean = 63.40, SD = 8.57), than PD-normal (mean = 59.15, SD = 10.18), or HC (60.56, SD = 10.82), F(2,386) = 4.192, p = .016. As such, age was included as a covariate.

PD-Normal vs. HC

Baseline: PD-Normal performed significantly worse than HC on the SDMT at baseline, t(302) = 3.950, p < .001, and had a lower global composite, t(302) = 1.988, p = .047. These corresponded to a mean difference in raw SDMT score of 4.58, 95% CI: [2.29, 6.86], and global z-score difference of .144, 95% CI: [.002, .285]. Statistically significant differences between groups at baseline were not seen on any other measure.

Change over time: Mauchly’s test showed violation of sphericity for all measures; as such Wilks’ lambda was interpreted. A main effect of time was observed on the MoCA, F(5,290) = 3.523, p < .001. Follow-up repeated contrasts revealed a statistically significant drop in both groups from baseline to year 1, F(1,307) = 62.476, p < .001, with no significant changes thereafter. No other main effects of time were observed.

Differential change: A statistically significant interaction indicating different rates of change between groups was observed on LNS, F(5,297) = 2.986, p = .012. This is illustrated in the Figure 1, with the HC group appearing stable over time and the PD-Normal showing a linear decline. This corresponded to a mean raw score decline over 5 years of .684, 95% CI: [.34, 1.03], and a median 5-year decline of 1.0. From normative data, in adults 65–69 years of age, a raw score difference of 1 corresponds to about .33 SD.¹⁴ No other group × time interactions were statistically significant, though a trend was observed on SDMT, p = .066

Figure 1.

Change Over 5 Years on Cognitive Tests

A statistically significant interaction indicating faster decline in PD participants matched to Health Controls on cognitive screening at baseline was observed on Letter-Number Sequencing, but no other measures. PD patients with lower scores at baseline did not decline faster over the first 5 years than those with normal scores.

PD-Normal vs. PD-MCI

Baseline: As shown in Table 1, PD-pMCI performed worse than PD-Normal at baseline on all cognitive test measures.

Table 1. Demographic and Test Performances by Group.

		Healthy Controls		PD Normal		PD-pMCI		HC vs PD Normal		PD Normal vs PD-pMCI
		(n = 135)		(n = 169)		(n = 84)		p-value		p-value
Age		60.34 (11.07)		59.52 (10.18)		63.40 (8.57)		0.482		0.004
Education		16.04 (2.91)		15.75 (2.86)		15.24 (2.91)		0.392		0.183
Women		53 (39%)		59 (35%)		25 (30%)		0.474		0.479
Judgment of Line Orientation		13.13 (2.02)		13.24 (1.85)		12.62 (2.32)		0.605		0.018
Semantic Fluency		52.88 (11.16)		50.96 (11.41)		46.04 (10.68)		0.142		0.001
Letter-Number Sequencing		11.1 (2.51)		11.15 (2.62)		9.94 (2.21)		0.862		< .001
Symbol-Digit Modalities Test		47.24 (11.01)		42.66 (9.21)		39.39 (10.28)		< .001		0.011
HVLT-R Total Learning		26.7 (4.24)		25.77 (4.65)		23.18 (4.5)		0.078		< .001
HVLT-R Delayed Recall		9.42 (2.33)		9.12 (2.13)		7.59 (2.49)		0.236		< .001
Global Composite		.07 (.62)		−.07 (.62)		−.53 (.63)		0.047		< .001

Change over time: Mauchly’s test showed violation of sphericity for all measures; as such Wilks’ lambda was interpreted. The main effect of time was significant on the MoCA, F(5,290) = 2.986, p = .012, and a trend for effect of time was observed on the SDMT, p = .072. No other main effects of time were observed.

Differential change: A statistically significant interaction was found between groups over time on the MoCA, F(5,241) = 12.01, p < .001. As show in the Figure, PD-Normal participants evinced a decline between baseline and year 1, Δ = 1.275, 95% CI: [.905, 1.646], while the PD-pMCI group improved, Δ = .845, 95% CI: [1.354, .336]. Follow-up repeated contrasts revealed a statistically significant decline from year 1 to year 2 in the PD-pMCI group. No other contrasts were statistically significant in either group. There were no statistically significant interactions between group and time on any of the other measures to suggest differential rates of change based on cognitive status at baseline.

Follow-up analysis: Laterality of motor symptoms to dominant hand.

The SDMT is the only task in the battery requiring graphomotor responding. To determine whether baseline differences between PD-Normal and HC were secondary to motor impairment, measures from the MDS-UPDRS Part III¹⁵ Motor Examination reflecting upper extremity functioning (i.e., rigidity, pronate/supinate, hand movements, postural tremor, kinetic tremor, and rest tremor) were calculated separately for the right (RUE) and left (LUE).

In right-hand dominant participants (HC n = 108, PD-Normal n = 147), univariate ANOVA found that the baseline difference on SDMT was no longer statistically significant after controlling for RUE functioning, F(1,253) = 2.128, p = .146. However, group differences remained controlling for only LUE function F(1,253) = 8.637, p = .004. Repeated measures multivariate ANOVA found that the previously observed trend for change in SDMT over time was no longer apparent after including RUE functioning at baseline as a covariate, p = .414.

In comparisons between right-hand dominant PD-Normal (n = 147) and PD-pMCI (n = 69), baseline differences on SDMT remained statistically significant in univariate ANOVA after controlling for RUE function, F(1,215) = 5.950, p = .016, and LUE function, F(1,215) = 6.402, p = .012. Controlling for RUE function in the multivariate analysis did not reveal a statistically significant main effect of time or group × interaction.

Discussion

The cognitive test battery in PPMI found little evidence of accelerated decline over the first 5 years of illness in PD patients who perform within normal limits on a cognitive screen (MoCA > 26) at baseline. The single exception was on a measure of working memory, LNS, where a modest, linear decline was observed in PD compared to stable performances in HC. Baseline differences between PD-normal and HC groups could be identified on a global composite from the test battery, but not on any individual tests. An apparent baseline difference on a test of processing speed, SDMT, was not significantly different after accounting for motor impairment in the dominant right hand. PD patients who performed below cutoff (MoCA ⩽ 26) at baseline also demonstrated lower baseline performances on measures of working memory, processing speed, verbal fluency, visuospatial functioning, memory, and the global composite than PD patients who performed within normal limits. However, there was no evidence of differential rate of change over the 5-year study period on any of these measures between PD cognitive status groups.

Previous studies have suggested the importance of dopamine-dependent fronto-striatal systems and working memory in PD. As such, it is not particularly surprising that working memory declined in PD-normal over the first 5 years compared to the HC group which remained stable. Dopaminergic depletion may increase vulnerability in the face of particularly cognitively demanding working memory tasks which engage multiple components of cognition including attention, executive control, and the short-term storage of verbal and visuospatial information.¹⁶ Research suggests increased demand for neural resources may be needed to compensate for nigrostriatal dopamine deficiency in PD participants performing memory tasks.¹⁷ Functional neuroimaging studies for example, demonstrate working memory tasks require a more extensive activation within the fronto-parietal network, even when performance is similar in controls.¹⁸ Furthermore, dopaminergic medications can reduce compensatory hyperactivation in PD participants during working memory tasks.¹⁹ Thus, as dopamine depletion reaches a threshold, compensatory mechanisms may no longer be sufficient leading to an accelerated decline in working memory performance. In contrast, episodic memory and visuospatial functions are typically associated with cholinergic tone, which is less disrupted in early idiopathic PD. Indeed, in vivo structural neuroimaging studies show that reductions in substantia nigra pars compacta (SNc) volume precede cholinergic basal forebrain degeneration.²⁰ Over a 12-month period in a smaller sample of early PD patients (n = 138), investigators in the NET-PD trial found a statistically non-significant decline on LNS, Δ = −0.5, 95% CI: [−1.1, 0.2]; while this trial was terminated following futility analysis, the trend toward decline is consistent with our findings.

The current findings also suggest that for PD patients who perform poorly on cognitive screening at baseline, deficient test performances can be observed across multiple domains compared to those who perform within normal limits. However, the presence of early deficits was not associated with accelerated rate of decline on any of these tests over the first 5 years. This finding may appear at odds with results from the ICICLE study, where PD patients with MoCA scores less than 26 at baseline were more likely to convert to dementia within the first three years;²¹ however, it is important to note that the focus of the current investigation was the psychometric properties of the tests, not clinical status, and it is likely that with lower scores at baseline, more individuals in the PD-pMCI group likely crossed diagnostic thresholds for dementia. It is also noted that baseline MoCA scores in ICICLE study were lower for the PD group, 25.4 (3.4) than HCs 27.0 (2.5), and from a much larger cognitive battery containing 27 measures, only phonemic fluency and two attention indices from computerized testing identified differential rate of change between groups.

Results from this study have implications for outcome measure selection in clinical trials. Investigations aimed at slowing or preventing cognitive decline in early PD would appear more likely to capture treatment effect with working memory measures similar to LNS rather than tasks from other domains. LNS has also demonstrated relevance to real-world functioning in other populations. For instance, in schizophrenia performance on the LNS was predictive of occupational attainment even after controlling for a number of other neurocognitive measures.²² That said, enthusiasm for LNS as a single measure is tempered by the rather diminutive change over time, amounting to a median 5-year decline of only about 1/3 of a standard deviation. While a “global cognitive index” is attractive as a concept and can simplify data analysis, the global index based on this battery did not detect a differential rate of change in early PD that could be targeted in neuroprotective trials. Our findings also suggest that it would not be necessary to screen-in participants based on cognitive deficit at baseline, as those with reduced scores on the MoCA did not show more precipitous decline. However, for shorter-term trials of symptomatic therapies, such inclusion criteria would be needed given the lack of baseline differences between PD-normal and HC. Across indications, use of tasks requiring graphomotor responding is cautioned, and application of appropriate controls to account for impact of motor impairment on such measures is warranted.

The current study certainly has limitations. First, the typical cutoff score for identifying PD-pMCI with the MoCA is < 26, and in this study the PD sample was split at < 27 to match inclusion criteria for HC participants in PPMI. Applying the more traditional cutoff of < 26 could diminish some of the observed differences between PD groups at baseline. Additionally, while six parallel versions of the HVLT-R were used to mitigate practice effect, their equivalence has been debated, and it is possible that changes in episodic memory were obscured by measurement error. There were also no measures of visual memory, executive functioning (e.g., set-switching, concept formation), or attention included in the PPMI battery; deficits in these important domains may manifest and progress differentially in early PD. Finally, this study was limited to changes seen in early PD, and extrapolation to trials of symptomatic therapies in more advanced stages is not recommended. Future investigations will examine sensitivity of these tests to structural changes observed on neuroimaging and correlation with clinician ratings of cognition and other non-motor PD symptoms.