Global Coherence and Cognition in Parkinson's Disease

Abstract

Purpose:

This study compared global coherence (GC) in individuals with Parkinson's disease (PD) to a healthy older adult (HOA) group during single (sitting) and dual (stationary cycling) tasks. Additionally, it explored the relationship between GC and cognition in PD.

Method:

Thirty-seven individuals with PD and 19 HOAs participated in the prospective, cross-sectional study. Participants completed discourse monologues elicited using published prompts while seated and while pedaling a stationary bicycle. Four rating levels of GC were analyzed (GC1 = no relationship to the topic, GC2 = remote relationship, GC3 = conditional relationship, and GC4 = complete relationship) using a published protocol with good interrater reliability and test–retest stability. Participants completed a battery of cognitive tasks, from which four latent factors were extracted: processing speed, working memory, inhibition, and updating.

Results:

Linear mixed modeling identified significant effects of GC level and GC level interactions with group, processing speed, and inhibition. The Group × GC Level interaction reflected that the PD group had a higher proportion of GC2 and GC1 utterances and fewer GC4 utterances than the HOA group. No differences between single and dual task conditions were found. Faster speed of processing predicted more GC4 utterances, whereas slower speed of processing predicted more G1 utterances. Better inhibition predicted fewer GC2 utterances. Group also predicted GC4 and GC2 proportions.

Conclusions:

Individuals with PD experienced greater difficulties with GC than HOAs. Processing speed and inhibition contributed significantly to GC across groups. Analysis of GC should be considered an informative addition to assessment of communicative effectiveness in PD.

Supplemental Material:

https://doi.org/10.23641/asha.20416056

Conservative projections of Parkinson's disease (PD) prevalence in North America estimate a rise in over 1 million cases by the year 2030 (Marras et al., 2018), indicating an imperative for speech-language pathologists to better understand communication impairments beyond motor speech in this growing population. PD, a neurodegenerative movement disorder, is widely known to affect motor aspects of speech, but the effects of PD on the expression of language are less well understood, particularly for aspects of discourse production. Analysis of discourse, defined here as language that is connected by successive utterances (Grimes, 1975; Harris, 2013), has received little attention in the PD literature compared to other acquired neurogenic communication disorders, possibly because of a disproportionate focus on motor-speech evaluation and treatment as opposed to language. Also complicating the issue is that assessment of language in this population is typically performed using aphasia batteries, which are not sensitive enough to expose language deficits in individuals with PD without dementia, who tend to perform at ceiling (N. Miller, 2017). Critically, to develop therapies that address the cognitive–linguistic and motoric functionality in PD, there is a crucial need for more investigation into the interrelationships among those processes affecting communication (Smith & Caplan, 2018).

In a large study surveying people with PD about changes experienced in communication, one of the four most commonly reported symptoms was “getting off topic” (Schalling et al., 2017), a problem that could disrupt not only the coherence of their speech but also their comprehensibility. Getting off topic is an example of poor global coherence (GC). GC, theoretically derived from Agar and Hobbs (1982), is defined here as a macrolinguistic measure of topic maintenance that reflects the extent to which a discourse unit maintains an overall/overarching semantic theme (Glosser & Deser, 1990). Furthermore, Sherratt (2007) reported correlations between “non-specific elements,” empty phrases or sentences unrelated to the topic, to lower syntactic complexity and less appropriate story grammar, thus tying GC to other aspects of complex language production. Therefore, based on Sherratt's and Schalling and colleagues' findings, GC analysis has the potential to be an important measure of language change in PD and other populations.

Both cognition and language can be affected early in people with PD (Altmann & Troche, 2011; Holtgraves & Giordano, 2017). The ability to maintain GC is hypothesized to reflect higher order complex cognitive–linguistic elements of discourse expression (Rogalski et al., 2020), although there is no consensus for which specific aspects of cognition are related to GC. Therefore, this study focuses on identifying cognitive predictors of GC and whether these differ in PD and healthy older adults (HOAs). Thus, the goals of the following study are to examine GC in PD compared to HOAs and to explore the relationship between GC and cognition in PD.

PD, Discourse, and GC

Language declines in PD are evident even in the early stages of disease progression (Altmann & Troche, 2011; Holtgraves & Giordano, 2017); however, most studies focus on language processing and comprehension. Relatively few have examined language in the context of discourse expression in PD. In those that have, researchers report that, compared to control groups, individuals with PD exhibit fewer informative words and utterances (Holtgraves et al., 2013; Murray, 2000; Reddy et al., 2016; Roberts & Post, 2018); more irrelevant information, sequencing, and perseveration errors (Godbout & Doyon, 2000); more incomplete or incorrect cohesive ties (Ellis et al., 2015); and fewer main events (Roberts & Post, 2018). Although it is unclear what causes GC and other language changes in PD, many studies hypothesize underlying associations with cognition (for a review, see Auclair-Ouellet et al., 2017; Ellis, Henderson, et al., 2016; Rogalski et al., 2020).

Researchers have reported GC impairments in adult populations with neurogenic communication disorders (e.g., aphasia, traumatic brain injury, dementia) and in HOAs (for a review, see Ellis, Henderson, et al., 2016). Comparatively few have examined GC in PD, and of those that have, none have reported significant GC deficits in PD without dementia (PD). Ash et al. (2011) compared the narrative discourse of four groups: PD, PD with dementia (PDD), dementia with Lewy bodies (DLB), and controls. The individuals with PDD and the individuals with DLB were combined into one group due to shared symptoms: highly variable cognition, inappropriate behaviors, and more severe cognitive impairments relative to nondemented individuals with PD, due to Lewy bodies (protein incursions between neurons) in the frontal lobe. Those with PD and no dementia typically have most Lewy bodies in subcortical regions at least in the mild to moderate stages of the disease. Ash et al. elicited discourse samples using a children's wordless picture book and analyzed narrative organization using numerous macro- and microlinguistic parameters. One macrolinguistic analysis of narrative organization similar to GC was “search theme maintenance.” Each participant's transcript was rated with a single score of 0–4, depending on how many a priori criteria the speaker met that were related to the central events in the book. The authors found no statistical differences between the PD subgroup and controls on search theme maintenance but did find differences between controls and the DLB/PDD subgroup on this and another measure of narrative organization. Notably, difficulty with search theme maintenance correlated with impairments in executive functioning, specifically inhibitory control and verbal fluency, but only in the DLB/PDD subgroup. The authors concluded that both cognitive and linguistic processes subserve narrative discourse in DLB/PDD due to disease affecting the frontal lobes. Narrative organization did not appear to be impaired in the PD without dementia subgroup, and the researchers further acknowledged that these individuals were also indistinguishable from controls on most other linguistic and cognitive measures.

Sanchez and Spencer (2013) analyzed the discourse of a large group of individuals with PD in both the “on” and “off” states of dopaminergic medication. Participants spoke for 1 min on the topic of their choice, and individual utterances were coded using a modified 5-point GC scale (Glosser & Deser, 1990; Van Leer & Turkstra, 1999), where a score of 1 represents no relationship of the utterance to the topic and a score of 5 represents a definite relationship of the utterance to the topic. The authors computed mean GC score for each transcript and additionally calculated the percentage of utterances rated as having no relationship to the topic (GC1), probable relationship (GC3), and definite relationship (GC5). They found no significant effect of participants being “on” or “off” their Parkinson's medications on mean GC scores. Unfortunately, there was no control group for comparison.

Lastly, Ellis and colleagues elicited discourse from a control group and a PD group in the “off” medication state, using the prompt “discuss a typical day” (Ellis, Fang, et al., 2016). Similar to Sanchez and Spencer (2013), discourse samples were analyzed using a modified 5-point GC scale (Glosser & Deser, 1990; Van Leer & Turkstra, 1999), and mean GC ratings were computed, as well as the percentage of utterances with a definite relationship to the topic (GC5). The PD group and the control group did not differ in mean GC. Although the PD group produced a somewhat lower proportion of utterances with GC5 ratings, the group difference was not significant. The authors attributed the nonsignificant findings to the small sample size (11 in each group). They further asserted that using an overall mean GC rating score and only the highest GC rating might not yield a true representation of the ability to maintain GC (Ellis, Fang, et al., 2016). These findings suggest that using the percentage of utterances at each GC level as the dependent variable might yield a better assessment of GC than mean GC score in a larger group of people with PD.

There are several possible explanations for the results reported in the previous research. The sample sizes were relatively small (i.e., Ash et al., 2011; Ellis, Fang, et al., 2016), and in the study with the largest sample size of the three, there was no control group for comparison (i.e., Sanchez & Spencer, 2013). Moreover, the discourse elicitation methods were different for each study, and it could be that some methods, such as telling a story from a picture book, or prompts, like “Please describe a typical day,” were not sensitive enough to capture impairments in GC (e.g., Marini et al., 2005; Van Leer & Turkstra, 1999; Wright et al., 2013). Finally, the GC analysis methods were inconsistent across studies, although the two studies that approached significance (i.e., Ellis, Fang, et al., 2016; Sanchez & Spencer, 2013) both used counts of utterances at each GC level as opposed to overall comparisons of means.

GC and Cognition

Another aim of this study is to extend the literature on the relationship between GC and cognition. Despite the lack of a strong theoretical framework for GC (Ellis, Henderson, et al., 2016), several studies have reported relationships between aspects of cognition and GC. GC has been associated with working memory (WM) in individuals with aphasia (Cahana-Amitay & Jenkins, 2018) and traumatic brain injury (Coelho et al., 2012). Also in traumatic brain injury, connections between GC and executive function (Le et al., 2014) and delayed recall (Galetto et al., 2013) have been reported. In older adults without brain injury, GC has been linked to executive function (i.e., inhibition) and episodic memory (Wright et al., 2014). In poststroke individuals without aphasia, Barker et al. (2017) found that better performance on tasks measuring WM and executive function (i.e., inhibition) correlated with GC errors. Correlations were also found between GC and attention and processing speed in poststroke individuals without aphasia (Rogalski et al., 2010).

Oftentimes these studies have not included a comprehensive battery of cognitive tasks. One exception is Kim et al. (2019), who found that individuals with mild cognitive impairment (MCI: both amnestic and nonamnestic) had lower overall mean GC than healthy controls. They further examined correlations between GC and a wide range of cognitive tasks in the MCI groups only. They found that better mean GC was associated with better immediate and delayed memory on a test of verbal learning, as well as with better inhibition, and semantic and phonemic verbal fluency.

To our knowledge, Ash et al. (2011) is the only study that has explored GC and cognition in PD. Their measure of GC, called “search theme maintenance,” was related to increased word fluency and better inhibition in the 14-person group with dementia. Unfortunately, correlations between search theme maintenance and cognitive measures were not assessed in the PD and HOA groups. Thus, robust findings that GC is related to various aspects of cognition in other groups, with and without neurogenic disorders, has not been replicated in PD without dementia, even though people with PD share deficits in many of the same cognitive abilities that have predicted GC in other groups.

Elucidating relationships between GC levels and cognitive tasks will enhance understanding of the underlying theoretical mechanisms of GC by exploring the relationship between proportion of utterances at each level of GC and cognition in individuals with PD who have been assessed on a great variety of cognitive variables. This study presents a secondary analysis of pre-intervention discourse samples collected during a study reported by Altmann et al. (2015, 2016). The study was designed to examine the effect of exercise on a broad range of cognitive tasks and two language tasks, one of which was discourse. Pre- and post-intervention assessments were completed in both a single task condition (sitting in a quiet room) and a dual task condition (while riding an exercise bicycle), to test the hypothesis that small changes in cognition would improve single task performance, but larger changes would also improve dual task performance. The study found that the aerobic exercise intervention improved executive function and improved the ability to stay focused in a picture-constrained sentence production task (i.e., improved “completeness”; Altmann et al., 2016); however, the discourse samples have not been analyzed yet.

Dual Task Effects

Doing two things at the same time can result in a “dual task effect” where performance of one task interferes with the simultaneous performance of another task (Woollacott & Shumway-Cook, 2002) due to limitations in shared attentional resources (Kahneman, 1970; Pashler, 1994). Dual task effects on discourse have been examined to some extent in healthy populations. For example, Kemper et al. (2003) found that older adults slowed their speech in response to walking or finger tapping dual tasks, whereas young adults produced shorter sentences that were less syntactically complex. Subsequently, Kemper et al. (2010) reported that, using a hand–eye coordination task, easier tasks had little effect on language performance of either group, but when the dual task increased in difficulty, the speech of both groups became dysfluent, less syntactically complex, and ungrammatical.

There is a large literature documenting that individuals with PD typically experience greater dual task costs than HOAs in balance-based or walking tasks (for a review, see Kelly et al., 2012; Raffegeau et al., 2019). Consistent with previous findings, Salazar et al. (2017) reported that both HOA and PD groups performed worse on an executive function task in a walking dual task than a single task. Furthermore, although individuals with PD performed worse than HOAs, no interactions with group were significant. In contrast to most studies on dual task effects in PD, Altmann et al. (2015) reported dual task benefits on cycling speed during easy, fast-paced cognitive tasks and no change in cycling during harder tasks, although benefits were smaller for the PD group than the HOA group. A follow-up analysis of cognitive task performance by Hazamy et al. (2017) found dual task benefits on response times for some difficult visuospatial cognitive tasks and a dual task cost only for the 2-back task in both the PD and HOA groups, with no group interactions. Hsiu-Chen et al. (2020) replicated these findings, individuals with PD showed dual task benefits on three measures of cognition during concurrent cycling. However, they also found that performance on two of those tasks declined with concurrent walking. Overall, these findings suggest that dual task effects on cognition in both HOAs and PD depend on the type of motor and cognitive tasks employed.

Studies looking at dual task effects on communication variables in PD have been largely limited to examining the impact of motor tasks on speech measures. For instance, Ho et al. (2002) examined the effects of a concurrent visuomotor tracking task on volumetric and temporal measures of speech in individuals with PD and healthy controls. The authors reported that when simultaneously counting and completing the distractor motor task, persons with PD exhibited dual task interference in the form of significant reductions in speech intensity, higher rates of intensity decay, and reduced counting durations (i.e., counting during a breath span). Additionally, dual tasking resulted in a slower rate of speech during a conversational speech task in persons with PD. None of the dual task costs were reported for healthy adults. Bunton and Keintz (2008) reported dual task costs on intelligibility in a PD group during single word, sentence, and monologue speech tasks when turning a nut on a bolt, whereas no such costs were reported for healthy controls. Additionally, Dromey et al. (2010) reported bidirectional interference in individuals with PD when simultaneously reading target sentences and completing a postural adjustment task (i.e., rise to toe). Specifically, when dual tasking, individuals with PD demonstrated degradation of speech articulation as evidenced by reduced diphthong extents and slopes along with declines in postural stability. However, healthy young and aged-matched controls did not demonstrate significant changes to speech while dual tasking. Greater dual task interference effects in individuals with PD compared to health controls have also been reported on sequential nonword repetition accuracy and sequence duration during completion of a concurrent visuomotor tracking task (Whitfield & Goberman, 2017).

Interestingly, others have found dual task benefits on conversational speech intensity during visuomotor tracking (Adams et al., 2010) and simultaneous walking (McCaig et al., 2016) in PD. Moreover, McCaig et al. (2016) found that concurrent walking had no effect on conversational speech rate in either controls or individuals with PD. These findings support the idea that the impact of dual tasking on speech may depend upon the nature of the concurrent task

To our knowledge, this study is the first to examine the effects of dual tasking on GC in PD. Considering that other difficult tasks in this same project showed dual task benefits (Altmann et al., 2015; Hazamy et al., 2017), it is possible that GC will improve in the dual task in people with PD and, possibly, HOAs. On the other hand, in a prior study examining GC in a stroke population without aphasia, we found no changes in GC from single (sitting) to dual (walking) tasks (Rogalski et al., 2010). Thus, it is also possible that GC will be unaffected, suggesting it may be somewhat protected from dual task effects.

This study analyzes pre-intervention discourse samples from the participants with PD described in Altmann et al. (2015) and Hazamy et al. (2017), as well as a group of HOAs recruited specifically to investigate the unexpected findings of dual task benefits on cycling reported in Altmann et al. (2015). Because it is known that people with PD have cognitive deficits compared to matched peers (e.g., Dirnberger & Jahanshahi, 2013), slightly older adults were recruited to provide a better match in terms of cognitive levels for the PD group, as it was unclear whether the unexpected dual task benefits were related to cognition, disease processes, or some other variable. Both groups were tested with the same tasks but different stimuli in single and dual task conditions. Moreover, based on the findings in Ellis, Fang, et al. (2016) and Sanchez and Spencer (2013), this study analyzes the proportion of utterances falling in each of the four GC levels (Wright et al., 2014), rather than using mean GC levels. Thus, the basic design of the study is 2 Group (PD, HOA) × 2 Conditions (single, dual task) × 4 GC levels. The contributions of cognitive covariates to proportion of utterances at each GC level will also be investigated. This study addresses the following research questions.

1. Does the distribution of utterances across GC levels differ for participants with PD without dementia and HOAs? Based on prior studies where GC differences approached significance when examining the levels of GC in smaller groups of PD and control participants (Ellis, Fang, et al., 2016) and a larger group of PD participants without a control group for comparison (Sanchez & Spencer, 2013), we predicted there would be group differences in the proportion of utterances falling at both the high and low end of the GC scale.

2. To what extent do cognitive abilities predict GC at each of the four levels? Furthermore, do the cognitive predictors of GC differ between PD without dementia and HOA groups? We expect that executive function, particularly inhibition, and WM will significantly predict GC, and this will not differ between groups.

3. This exploratory question asks whether the cycling dual task impacts GC in PD without dementia or HOAs. There are too little data in the literature to make predictions about possible dual task effects, and whether these effects might be costs or benefits.

Method

Participants

Thirty-eight individuals with PD (31 males and seven females) and 21 HOA participants (six males and 15 females) were included in the current analysis. One participant with PD who screened into the study scored in the impaired range on the Dementia Rating Scale-2 (Jurica et al., 2001), suggesting dementia, and was excluded from all analyses. Two HOA participants were missing cognitive data and excluded from all analyses. All participants were part of a larger study at the Center for Movement Disorders and Neurorestoration in Gainesville, Florida, examining outcomes of exercise treatment in PD (Altmann et al., 2016). Complete demographic data for this study are reported in Hazamy et al. (2019), and descriptive information for the participants can be found in Table 1. Participants were screened with the Mini Mental State Examination at intake, and only those with scores greater than 24 completed the study. The PD and HOA groups were similar in education, depression, and apathy but differed in age. The PD group was younger than the HOA group. As discussed above, individuals with PD have reductions in overall cognitive abilities compared to age-matched peers (e.g., Dirnberger & Jahanshahi, 2013); thus, older HOAs were recruited in order to better match the PD group on overall cognition. Even so, 68% of the PD group and 73% of the HOA group were between 60 and 80 years old.

Table 1.

Descriptive information for Parkinson's disease (PD; n = 37) and healthy older adult (HOA; n = 19) groups.

Descriptive information	PD	HOA	Range	t value	p
	M (SD)	M (SD)
Age	64.89 (9.96)	72.74 (9.14)	33–89	−2.663	.010
Education (years)	17.44 (3.98)	18.42 (1.97)	10–28	−1.007	.319
MMSE a	29.22 (0.99)	30.00 (0)	28–30	−3.413	.001
Dementia Rating Scale-2	140.00 (3.69)	141.63 (1.77)	131–145	−1.815	.075
Beck Depression Inventory-2	5.47 (5.23)	3.58 (4.02)	0–17	1.376	.175
Apathy Evaluation Scale	8.59 (5.85)	8.32 (2.58)	0–19	0.229	.820
UPDRS b	33.89 (10.86)		15–53
Hoehn & Yahr Scale c	2.14 (0.58)		1–3

Note. Bolded values are statistically significantly different between groups at p < .05.

^a MMSE = Mini Mental State Examination.

^b Unified Parkinson's Disease Rating Scale: higher scores (maximum: 199) indicate greater disability from PD.

^c Measures five stages of the progression of symptoms in PD.

Participants with PD were included if they had a clinical diagnosis of idiopathic PD, a modified Hoehn and Yahr scale score between 1 and 3 in the “on” medication state (Hoehn & Yahr, 1967), a stable response to medications, and no presence of atypical parkinsonism, or history of motor fluctuations. The Hoehn and Yahr scale (1967) is a 5-point, clinical rating scale that categorizes the severity of motor symptoms in PD from 1 = no impairment to 5 = extremely impaired. Testing took place in the “on” medication state, to minimize fall risk during the dual task. Exclusion criteria for the PD and HOA groups were as follows: significant history of falls, presence of significant cognitive or psychiatric disorder, taking medications interfering with cognitive function, and taking moderate to high doses of beta-blockers. All participants had normal or corrected-to-normal vision. All signed informed consent approved by the institutional review board of the University of Florida.

Procedure

Participants completed a large battery of tasks while sitting in a quiet room (single task) and while stationary cycling (dual task; Altmann et al., 2016). The order of the single and dual tasks was counterbalanced across participants. An effort was made to keep all tasks to about 3 min each, to minimize fatigue during the dual task. Thus, discourse samples were 3 min long. All tasks were presented using DirectRT software (Jarvis, 2006). Full procedures are available in Altmann et al. (2015, 2016).

Discourse Task

Instructions and prompts were presented on a laptop screen (single task) or a projection screen (dual task). In each condition, participants responded to one of four discourse prompts. All responses were recorded on the computer via wireless headset microphone for later transcription. Four different prompts were used to prevent content repetition across the four testing sessions (pre- and post-intervention, single and dual task). Prompts were counterbalanced across single and dual tasks and test session. The discourse prompts were the following:

1. “Please tell me about a person who had an important impact on your life.”

2. “What do you think was the most important event of the last 100 years?”

3. “Please tell me about a vacation or event from your past that you remember well.”

4. “What do you think is the most important invention of the last 100 years?”

Participants were given as much time as needed to think about their responses before speaking; however, once participants began speaking, all narratives were terminated at 3 min. Transcripts were terminated at the end of the last complete clause. This study analyzes the pre-intervention discourse samples.

Discourse Transcription, Segmentation, and GC Coding

Five trained research assistants from the final author's lab orthographically transcribed and mazed discourse responses according to the Systematic Analysis of Language Transcripts (SALT; J. Miller & Iglesias, 2008). Mazed words (e.g., filler words, grammatical errors) were excluded from the transcripts before resubmitting them to SALT analysis to obtain group comparison data on number of C-units, mean length of utterance (MLU), number of words, number of words per minute, and percentage of pause time.

The transcripts were then entered into Excel spreadsheets and segmented into C-units, any independent clause plus any related clauses (Loban, 1963), by a master's level graduate student in the first author's lab, blinded to task (single or dual) and group (PD or HOA). GC ratings were completed by this graduate student and checked by the first author, also blinded to task and group. Each C-unit was scored with a number from 1 to 4 in terms of its relation to the topic of discourse, according to the published scoring criteria guidelines by Wright et al. (2013) and supplementary training materials provided by Wright. Briefly, a score of 1 signifies that the utterance is entirely unrelated to the topic; 2 signifies that the utterance is only remotely related to the topic, including noncritical tangential information; 3 signifies that the utterance is related to the topic, including relevant tangential information; and 4 signifies that the utterance is overtly related to the topic (Wright et al., 2013). Differences among the four GC levels are perceptible to clinicians with minimal training (Rogalski et al., 2020). The percentages of each participant's utterances in each of the four GC levels are the dependent variables of interest. A sample scored discourse sample from a participant with PD is included in Supplemental Material S1.

Reliability

Reliability was completed on a randomized 16% of the transcripts. As reported in Hazamy et al. (2019), transcription interrater reliability for mazed and nonmazed words was good (both Cronbach's α > .92). Point-to-point reliability for GC was calculated as the number of same ratings divided by the total number of ratings (same and different). Intrarater reliability for GC was 91.4%, and interrater reliability was 83.8%. Disagreements between raters over coding particular utterances were settled by discussion.

Cognitive Tasks

Cognitive tasks were chosen to span a broad range of cognitive abilities, with at least two measures each of processing speed, WM, and two executive function domains (Miyake et al., 2000), inhibition, and updating. There were 12 cognitive tasks in total, all taking about 3 min each, as stated above. All tasks required verbal responses to maintain safety during the dual task. Easier tasks were interspersed with more difficult tasks to avoid fatigue. Task stimuli were randomly presented for all tasks but the digit spans. This study employs data from the single task condition to best represent participants' baseline cognitive performance at the time the discourse samples were collected. As the plan was to look at cognitive predictors of GC in a linear mixed model with follow-up regressions, all cognitive tasks were entered into a principal components analysis to extract uncorrelated latent factors. The table of factor loadings is available in Supplemental Material S2.

Two tasks assessed verbal and visual speed, respectively. Articulation speed was assessed using a syllable repetition task. The dependent variable was the number of times participants repeated “pa” within 10 s. Visual processing speed was assessed with a task measuring how fast a person verbally responded “go” each time a blue star appeared on the screen at varying time intervals. Response time was the dependent variable.

There were four short-term and WM tasks. Verbal short-term memory was assessed using the Digit Span Forward task. Verbal WM was assessed using the Digit Span Backward task (Wechsler, 1987). For these tasks, participants heard and repeated back increasingly longer strings of numbers in either verbatim order (digit span forward) or reverse order (digit span backwards). The dependent variable for both tasks was the number of correctly recalled lists out of 14. Visual WM was assessed using a task in which participants saw one to four individually presented tic-tac-toe boards with two adjacent dots presented at a set interval (1,500 ms each) and then an array of tic-tac-toe boards. Participants had to verify if the sequence of stimuli in the array matched the sequence of individual stimuli they had seen previously. The score was percent correct. Finally, the Operation Span task, a measure of domain-general WM (or “working attention”; Conway et al., 2005), required participants to read aloud six letters on a screen, then verify if one to four simple arithmetic equations were correct or not (e.g., 2 + 4 = 7 = no), followed by recalling the letters, in order, that were presented initially. The dependent variable for the operation span task was the mean number of letters recalled in sequence. Updating was assessed using 0-back, 1-back, and 2-back tasks (Braver et al., 1997). The n-back tasks used the same tic-tac-toe stimuli as the visual memory task and included 40 trials with 10 critical “yes” trials. The 0-back task required participants to state “yes” or “no” if a stimulus matched or did not match a previously specified target stimulus. For the 1-back task, participants viewed a continuous series of screens and responded “yes” or “no” if the current stimulus matched or did not match the immediately preceding stimulus. The 2-back task was similar except participants were asked if current patterns matched or did not match patterns from two screens prior. Scores for all n-back tasks were the proportion of correct responses to critical “yes” stimuli.

The Digit Symbol Substitution task (Wechsler, 1958) measures multiple functions, including processing speed, complex attention, and visuoperceptual function (Jaeger, 2018). There is some support that executive functioning influences the Digit Symbol Substitution task (Thornton & Carmody, 2012) and that the task is sensitive to cognitive dysfunction and changes in cognitive functioning across a variety of clinical populations (Jaeger, 2018). In the version adapted for use in this study, an array of nine symbols paired with the numbers 1–9 were displayed at the top of a screen. Participants were shown a large symbol below and responded verbally with the corresponding number. The dependent variable for this task was response time for accurate responses.

Finally, the Stroop task (Stroop, 1935) consisted of two subtasks. In the Stroop Color task, participants named the font color (either red, blue, or green) of a string of four bolded Xs on the screen. In the Stroop Color–Word task, participants were presented with color words displayed in incongruous font colors (e.g., the word “green” was presented in red font) and were asked to name the font color. Dependent variables for these tasks were response times for Stroop Color, and Stroop Interference calculated as the response time for the Stroop Color–Word task minus the response time for Stroop Color. Stroop Color is often considered a processing speed measure, whereas Stroop Interference is considered a measure of executive function, specifically the ability to inhibit cognitive interference (Stroop, 1935).

Data Analysis Plan

Cognitive scores were submitted to a principal components analysis to extract uncorrelated latent variables for use as covariates in the primary analysis. Note that, due to its dubious status as a cognitive measure, the “Pa” task was excluded from this analysis.

Prior to data analysis, data were checked for integrity including outliers. The contributions of task (single, dual), GC level, and cognition to the proportion of utterances at each level were analyzed using linear mixed models (LMMs) with random participants and items (i.e., topics), but not random intercepts as the models did not converge. Significant interactions identified in the LMM were explored using stepwise regressions. Statistical significance was set at a critical value of α < .05. Specifically, Research Question 1 and Exploratory Question 3 are addressed by the fixed effects, that is, the effects of GC level, group, and dual task. Research Question 2 is addressed in the results for the covariates. All statistics were computed using SPSS 27.

Results

Cognitive Characteristics

Regarding cognitive measures, the PD group performed significantly worse on 1-back, 2-back, operation span, and digit symbol substitution tasks than the HOA group (p < .05). A principal components analysis with varimax rotation extracted four latent factors from the 11 cognitive tasks. The first factor, accounting for 24.65% of variance in cognitive scores, was labeled speed, because the strongest factor loadings were from the DSST, Stroop Colors, and the Star task. Higher Speed scores reflected higher (slower) response times on these tasks and, therefore, greater cognitive slowing. The second factor, accounting for 20.74% of the variance, was labeled WM, because the largest factor loadings were from digit span forward, digit span backwards, and operation span. Higher WM scores indicated better WM. The third factor, which accounted for 14.94% of the variance, was labeled Inhibition, as the largest factor loadings were from Stroop Interference and the 1-back task, which is very fast paced and requires inhibition of the prepotent “No” response. Higher Inhibition scores meant participants had little difficulty inhibiting prepotent responses in the Stroop Color Word task. The fourth factor, which accounted for 9.74% of the variance, was labeled Updating, because the strongest factor loadings were from the 2-back and 0-back tasks, which require continuous updating of WM. Higher scores in Updating meant better updating performance. The full table of factor loadings is included in Supplemental Material S2. Group scores for the four factors are shown in Table 2. Independent-samples t tests assessed factor scores for group differences. Although the PD group scored lower than the HOA group in the first three factors, there were no significant group differences.

Table 2.

Discourse and cognitive variables for Parkinson's disease (PD) and healthy older adult (HOA) groups.

Variable	Single task		Dual task
	PD M (SD)	HOA M (SD)	PD M (SD)	HOA M (SD)
No. of participants	37	19	37	19
Discourse
C-units	24.8 (9.8)	25.9 (11.3)	26.1 (8.9)	26.0 (7.8)
MLU	13.5 (3.1)	15.9 (4.9)	12.2 (3.1)	14.7 (3.3)
Number of words	318.1 (100.8)	339.3 (75.4)	312.9 (107.7)	366.3 (78.7)
Number of words/min	127.7 (31.9)	131.6 (24.2)	122.4 (34.7)	135.1 (26.1)
Percent pause time	9.8 (10.1)	2.4 (2.8)	14.4 (11.8)	2.9 (3.2)
Cognition
“Pa” syllables a	54.0 (9.9)	57.6 (8.2)
Stroop Color words a (ms)	962 (238)	902 (108)
Speed factor	0.06 (1.00)	−0.30 (0.57)
Star task (ms)	480 (204)	421 (54)
Digit symbol sub. (ms)	2903 (838)	2584 (499)*
Stroop Color Xs (ms)	656 (153)	634 (105)
Visual memory (%)	71.0 (20.1)	84.5 (18.6)*
Working memory factor	−0.19 (.894)	0.31 (1.12)
Digit span forward	8.38 (2.35)	9.1 (2.9)
Digit span backward	6.4 (2.0)	7.0 (2.3)
Operation span	2.7 (1.2)	3.9 (1.3)**
Inhibition factor	−0.08 (1.15)	0.27 (0.38)
Stroop interference (ms)	306 (234)	267 (79)
1-back (%)	93.3 (11.8)	97.9 (5.4)
Updating factor	0.01 (1.23)	−0.03 (0.31)
2-back (%)	78.6 (24.0)	89.1 (13.0)**
0-back (%)	99.0 (2.9)	99.5 (2.3)

Note. Bolded values are statistically different between groups at *p < .05 and **p < .01. C-units = communication units; MLU = mean length of utterance.

^a Excluded from principal components analysis.

Discourse Characteristics

Repeated-measures ANOVAs were used to assess the effects of task (single, dual) and group (PD, HOA) on characteristics of discourse samples, shown in Table 2. There were only two with significant findings. The main effect of group was significant for MLU. Individuals with PD produced shorter sentences than HOAs (M _PD = 12.89, SD = 3.11; M_HOA = 15.15, SD = 4.10); F(1, 57) = 9.334, p = .003, η² = .141. Additionally, individuals with PD spent more time in pauses than HOAs (M _PD = 12.14 s, SD = 10.99, M _HOA = 2.40, SD = 3.03); F(1, 56) = 16.943, p < .001, η² = .249. There was also a significant interaction between task and group in pause time, because participants with PD had longer pause times in the dual task (M = 14.430, SD = 11.843) than in the single task (M = 9.827, SD = 10.155), t(36) = 3.600, p = .001, but HOAs showed no task effect (M _Dual = 2.58, SD = 3.19; M _Single = 2.21, SD = 2.80), t(18) = 1.832, p = .607. It is not clear if the effect of the dual task on pause time in the PD group was due to exercise-related respiration problems or cognitive interference in language planning. The main effect of group on pause time may reflect longer language formulation and speech planning times in the PD group and has been previously reported in Ash et al. (2012) and Smith et al. (2018). Despite the PD group having longer pause times, groups did not differ significantly on number of C-units, length of speaking time, number of words, or number of words/minute (all p > .05). Notably, single and dual task discourse did not differ for any of the other variables nor did it interact with group.

Question 1: Group Differences on GC Levels

LMM was employed to assess group, task, and GC level effects, as well as cognitive covariates. Table 3 presents the means and standard deviations of the proportions of utterances receiving each GC rating for both groups during single, dual, and combined single/dual task conditions. Because the dependent variables were proportions of utterances that added to 1.0 for each participant and because Research Question 1 specifically asked about the distribution of responses across GC levels between Groups, only GC level and interactions with GC level were included in the LMM.

Table 3.

The means and standard deviations of the proportions of utterances receiving each global coherence (GC) rating for the Parkinson's disease (PD) and healthy older adults (HOA) groups during single, dual, and combined single/dual task conditions.

GC level	Single task		Dual task		Combined single and dual
	PD M (SD)	HOA M (SD)	PD M (SD)	HOA mean (SD)	PD M (SD)	HOA M (SD)
GC1	0.13 (0.15)	0.05 (0.10) *	0.14 (0.20)	0.08 (0.11)	0.13 (0.14)	0.07 (0.08)
GC2	0.23 (0.16)	0.12 (0.09) **	0.19 (0.14)	0.14 (0.12)	0.21 (0.10)	0.13 (0.07) **
GC3	0.27 (0.15)	0.27 (0.12)	0.29 (0.14)	0.31 (0.17)	0.29 (0.12)	0.29 (0.10)
GC4	0.37 (0.21)	0.55 (0.20) **	0.39 (0.24)	0.47 (0.16)	0.38 (0.17)	0.51 (0.12) **

^*Group comparison is < .05.

^**Group comparison is < .01.

Addressing Exploratory Question 3, the effect of the dual task and all interactions with dual task were not significant (all p > .25), so after the first set of models, task was excluded from all analyses, which improved model fit. All subsequent analyses were computed on the mean proportion of utterances combined across single and dual tasks in each GC level.

Because groups differed in age, age was tested as a potential covariate. Neither age nor any interaction with age was significant, so it was not included in the final model. This left a model that included GC level, Group × GC Level, and interactions between each cognitive factor and GC level. GC level interactions with WM and Updating were not significant (p = .385 and .624, respectively), and their removal improved the model. The final model, therefore, included GC level, F(3, 208) = 76.166, p < .001; GC Level × Group, F(4, 208) = 3.929, p = .004; GC Level × Speed, F(3, 208) = 4.843, p = .001; and GC Level × Inhibition, F(4, 208) = 2.688, p = .032. Details of the final model are included in Supplemental Material S3.

Addressing Research Question 1, the main effect of GC level was explored with planned, paired comparisons, which revealed that the proportion of utterances falling in each GC level was significantly different from all other levels, with all p values less than .001. The Group × GC Level interaction was explored with independent-samples t tests with Bonferroni corrections for multiple comparisons (i.e., p was required to be below .0125 to be significant). As shown in the right two columns of Table 3, the groups differed significantly in the proportion of utterances in GC Level 4, t(54.8) = −3.507, p = .004, and GC Level 2, t(55.6) = 3.575, p = .001, but not in GC1, t(55.6) = 2.323, p = .023, or GC3, t(56) = −0.578, p = .567.

Question 2: Effects of Cognition on GC Levels

In the LMM, the effects of Speed and Inhibition interacted with GC level. To address Research Question 2 and explore these interactions, stepwise regressions were employed to determine the amount of variance Speed and Inhibition accounted for at each GC level. Additionally, Group was added as a predictor in the second step of the regression to determine the percent of variance that it accounted for at each GC level. Details of the regressions are shown in Table 4.

Table 4.

Cognitive and group predictors of the proportion of utterances rated at each GC level.

GC level	Predictors	r 2	B (SE)	p value	t value
GC1	(Constant)		0.113 (.006)	.016	6.865
	Speed	.127	0.052 (.019)	.019	2.800
GC2	(Constant)		0.272 (.034)	.034	7.965
	Inhibition	.189	−0.038 (.012)	.012	−3.231
	Group	.100	−0.066 (.024)	.024	−2.727
	Total r 2	.289
GC3	No significant predictors
GC4	(Constant)		0.274 (.058)	.058	4.702
	Speed	.168	−0.065 (.022)	.022	−2.920
	Group	.098	0.110 (.041)	.041	2.657
	Total r 2	.266

Note. A high Speed score means slow responses. A high Inhibition score means better inhibition ability. The independent variable Group was dummy coded (PD as 1, HOA as 2) to facilitate its use in the regression.

The best predictors of the proportion of utterances categorized at GC4 were Speed, accounting for 16.8% of the variance, and Group, accounting for 9.8% of the variance, for a total 26.6% of the variance. HOAs and participants with faster speed of processing produced a higher proportion of utterances characterized as GC4. There were no significant predictors of G3 proportions, with the three predictors accounting for a total 1% of variance. In contrast, proportions of utterances categorized as G2 were predicted by Inhibition, accounting for 18.9% of the variance, and Group, accounting for an additional 10.0% of variance, for a total of 28.9% of the variance. Participants with worse inhibition and participants with PD produced more utterances categorized as GC2. Proportions of utterances categorized as GC1 were predicted by Speed, accounting for 12.7% of the variance. Group accounted for an additional, nonsignificant 5.2% of variance. Participants with slower processing speed produced more utterances characterized as GC1. In summary, slower speed of processing predicted more unrelated GC1 utterances, and faster speed of processing predicted more on-topic GC4 utterances. Inhibition predicted GC2 scores, with poor inhibition associated with more largely off topic, tangential utterances. A diagnosis of PD was associated with fewer GC4 utterances and more GC2 utterances.

Discussion

Question 1: Group Differences on GC Levels

This study's primary aim was to compare GC in a PD group to a control group of HOAs by analyzing discourse monologues elicited during single and dual tasks. As predicted, the PD group was impaired in GC as demonstrated by having a significantly smaller proportion of utterances categorized as GC4 (overtly related to the topic) and a higher proportion of utterances categorized as GC2 (only remotely related to the topic, including noncritical tangential information) compared to the HOA group. However, group differences in GC3 and GC1 were not significant, which we did not predict. To the best of our knowledge, this study is the first to report significant impairments in GC in individuals with PD without dementia. Contrary to our findings, Ash et al. (2011) found no differences between controls and the PD group without dementia. One reason might have to do with their discourse elicitation method: story retell using a picture book. Wright et al. (2014) also found no differences between younger and older adults using picture stimuli, but did find differences using personal recounts (e.g., describing a vacation). Furthermore, those personal recounts were associated with increased tangential utterances. Wright et al. additionally found that picture stimuli resulted in higher ratings of GC for both groups than personal topics and posited that picture stimuli allow for more goal-directed discourse. Picture books also provide a topic, a constrained set of events, and the order of these events to guide the discourse, which could facilitate maintaining GC. Personal recounts often contain somewhat off topic, perhaps GC3, utterances, according to Wright et al., accounting for the lower mean GC scores. This study expands on this literature by suggesting that individual variations in cognitive ability and disease state may influence the frequency of departures from completely on-topic, GC4 utterances in a discourse sample, as well as how far the speaker might veer off topic. Considering this, personal recount discourse samples may be more effective at identifying GC impairments in people with cognitive impairments and may give a better snapshot of an individual's performance in-real world situations than more constrained language production tasks.

The findings in this study are consistent with those of Ellis, Fang, et al. (2016), who employed a 5-point GC scale. They found participants with PD had fewer utterances with GC5 ratings than the control group, although this was not significant, possibly because of the small number of participants. Additionally, discourse in the latter study was elicited with a prompt to “discuss a typical day,” so the immediateness and lack of novelty of the topic may have made it easier for participants to stay on topic.

This study additionally explored whether a cycling dual task would affect GC, and especially if the dual task effect differed between groups. In fact, there were no effects of the dual task on either GC or other characteristics of discourse, like number of different words or words per minute (see Table 2). Participants did divide their words (which did not differ between single and dual task samples) into slightly more C-units and produced significantly longer sentences in the dual task. This may have been related to higher breathing rates in the dual task. Similarly, pause time showed a dual task effect, and that may also have been confounded with exercise-related breathing difficulty. Dual task effects, especially costs, have generally been attributed to limitations in shared resources (Kahneman, 1970; Pashler, 1994); the lack of dual task effect on GC suggests that shared cognitive resources may be minimal between cycling and producing coherent discourse. Cycling on a stationary bicycle is an easy, relatively automatic task that relies on procedural memory. In contrast, language generation is a highly volitional task that relies on declarative memory, especially for content. Therefore, perhaps it is not a surprise that there was no effect of cycling on GC and other discourse characteristics.

Alternatively, the dual task benefits papers from our group suggest another explanation. Altmann et al. (2015) suggest that people assess the difficulty of tasks or combinations of tasks, and cortical arousal increases with the perceived difficulty of the tasks. Increased arousal increases available cognitive resources. If someone underestimates the difficulty of the tasks or one of the tasks is extremely difficult, dual task costs arise. If someone overestimates the difficulty, dual task benefits arise. Considering that cycling speeds increased by about 7% (nonsignificant) during the dual task in participants with PD (unpublished data), and that there were no dual task effects on GC, both groups did an excellent job of estimating the difficulty of this combination of tasks. Indeed, it is common for people to talk while doing any number of tasks. Kemper et al. (2010) found that the effects of easy and moderately difficult tasks did not affect the fluency, grammatical complexity, or content in the discourse of HOAs, and only very difficult tasks disrupted discourse catastrophically. Stationary cycling is an easy motor task that requires little attention relative to walking (Lambourne & Tomporowski, 2010), and is unaffected in PD (Snijders et al., 2012). Therefore, many aspects of discourse, including GC, appear to be unaffected by concurrent, easy tasks like cycling.

Question 2: Effects of Cognition on GC Levels

The second research question explored the relationship between cognition and GC in the PD group. The original prediction was that WM and inhibition would influence GC. These predictions were partially supported. The study found significant effects of speed of processing and inhibition on GC levels, and speed of processing is related to WM in older adults (Salthouse, 1992). In fact, different levels of GC had different predictors. This provides considerable support for analyzing GC levels separately, rather than using a mean GC score, since the same mean GC score could be attained in many different ways. In particular, the findings argue against conflating GC2 and GC1, as they have completely different predictors.

Faster processing speed predicted more GC4 utterances and fewer GC2 and GC1 utterances. The processing speed factor comprised input from the digit symbol substitution task, previously found to correlate with GC (Rogalski et al., 2010), as well as the Star task and Stroop colors, all tasks that required speeded lexical output. Therefore, speed of lexical access or minimally, speed of articulation initiation, may also contribute to this factor. Considering that participants were given time to plan their discourse before recording began, those with faster speed of processing may have activated a more complete plan for their narrative and then may have been able to reactivate that plan with each utterance or through self-monitoring. In contrast, those with slower processing speed may have only partially activated a plan for their discourse and then likely spoke slower during the task, leading to longer times between reactivations of their narrative plan. Consequently, parts of their discourse plan may have decayed in memory, whereas the extra time allowed for competing unrelated ideas to become active.

The results of this study also suggest that inhibition, one of the core executive functions (Miyake et al., 2000),played a role in maintaining GC during discourse production in participants with PD. Thus, in this study, poorer inhibition predicted a higher proportion of GC2 utterances across groups. To account for this, we hypothesize that activation spreads within the discourse production system much like it has been posited to do throughout the lexical access system (Martin et al., 1994). When someone is asked a question, or given a prompt in an experiment, an entire event or series of events and concepts is likely activated that share connections among them and are linked to the eliciting question. Through existing links in declarative memory, both semantic and autobiographical, spreading activation could lead to partial or even full activation of related ideas and events. The activation of some of these linked ideas could be augmented by feedback from self-monitoring. In a healthy system, inhibition would keep the spread of activation to a manageable level, allowing full activation of only highly related ideas so that the discourse can proceed, while suppressing activation of off-topic ideas. In people with poor inhibition, activation and feedback would have a chance to spread for a longer time, increasing the probability of producing a statement that is only vaguely related to the topic. This could explain how off-topic utterances could occur that are unrelated to memory impairments. However, memory impairments leading to rapid fading of the topic or slow processing speed, allowing additional time for activation to spread and feedback to occur, could exacerbate this effect, depending on the cognitive deficits in the target population. In summary, our central hypothesis is that inhibition is likely essential to curtail spreading activation among events and concepts partially related to the discourse topic, so that the speaker can maintain GC.

The findings that inhibition contributes to poor GC are consistent with those of several studies in the literature. Ash et al. (2011), Kim et al. (2019), and Barker et al. (2017) reported relationships between GC, variously measured, and Stroop Color–Word performance in Lewy body disease with dementia, MCI, and poststroke individuals without aphasia, respectively. Moreover, the cognitive aging literature also documents links between Stroop tasks and off-topic verbosity in older adults (e.g., Pushkar et al., 2000), a phenomenon in which, during discourse, the speaker adds extensive, unnecessary detail or completely goes off topic (i.e., GC2 and GC1 utterances, respectively).

Another interesting finding in this study was that there were no cognitive predictors of the proportion of GC3 utterances. It is likely that there are numerous subtypes of GC3 utterances, especially given that relevant tangential information would be considered as GC3. Furthermore, it is common for older adults to add explanations of events or items, which might also be categorized as GC3, for younger listeners (James et al., 1998), such as the researchers who ran the current experiment.

In summary, these findings suggest that speed of processing and inhibition, a core executive function, may play significant roles in the GC of individuals with PD and HOAs. Specifically, the findings suggest that maintaining good GC may require sufficient processing speed to support the continuous generation of iterative utterances that are overtly related to the topic (GC4). Furthermore, when processing speed is too slow, GC may be compromised, leading to completely unrelated utterances, GC1. Additionally, these finding are consistent with the idea that suppressing remotely related utterances (GC2) may depend on inhibition ability in these two groups.

Clinical Implications of Findings

This study illustrates how using the 4-point GC ratings levels instead of overall GC means can be beneficial in exposing language impairments in PD. Using mean scores within a group and the group's standard deviation can reveal variation in mean GC within the group, but cannot capture the relative frequency of different types of tangentiality within the group. For example, if someone has on-topic discourse (GC4) punctuated by equivalent periods of tangentiality (GC2), their mean score (GC3) would be equivalent to someone who is generally on topic throughout their discourse sample. Furthermore, the results suggest a dissociation in the cognitive foundations of GC2 and GC1 utterances, which may reflect distinct problems. In this study, the pattern of fewer GC4 and more GC2 ratings in the PD group compared with the HOA group reflects greater instances of tangential discourse in PD. This is consistent with findings indicating that individuals with PD rate themselves as being more “off topic” (Schalling et al., 2017). Moreover, in conjunction with the finding that high GC correlated with untrained raters' levels of interest and attention (Rogalski et al., 2020), these findings suggest that listeners might be less interested in communicating with someone who produces fewer on-topic utterances and/or greater off-topic utterances. Individuals with PD face communication challenges that affect discourse production (e.g., N. Miller, 2017; Schalling et al., 2017), which, in turn, can lead to social avoidance (N. Miller et al., 2008). This is worrisome because in PD and HOAs, social connections and communication have been shown to substantially affect quality of life (Takahashi et al., 2016).

PD is a movement disorder that affects language early in the disease progression (Altmann & Troche, 2011; Holtgraves & Giordano, 2017), but oftentimes language deficits are not captured by standardized language tests (N. Miller, 2017), or they are overshadowed by more obvious motor speech impairments. Language deficits in PD are thought to stem from cognitive impairments (Auclair-Ouellet et al., 2017); however, this study and Troche and Altmann (2012) both find significant effects of PD on language production even when cognition is accounted for. GC is hypothesized to be subserved by cognition in PD (Ellis, Henderson, et al., 2016; Rogalski et al., 2020), a hypothesis supported by the current findings. On the other hand, individuals with PD still produced a smaller proportion of GC4 and larger proportion of GC2 utterances even after cognition was taken into account. Therefore, there is either some other, untested cognitive variable contributing to GC or PD itself can impact GC, apart from individual differences in cognition. Taken together, the relationships observed in this study and previous studies suggest that analysis of GC, particularly utterances that might be categorized as GC2 and GC4, which are discernable by untrained clinicians (Rogalski et al., 2020), may reflect higher order cognitive–linguistic changes in PD and other disorders.

Evaluation of GC may provide a way to discern the more subtle cognitive–linguistic characteristics that might be masked by the audible changes to voice and motor speech in PD, leading to treatments tailored to engage cognition and language in addition to voice and speech (Ash et al., 2011). The data in our study suggest that processing speed has pervasive effects on maintaining GC. What this could mean clinically is that individuals with PD may benefit from a discourse-related cognitively stimulating therapy program focusing on improving processing speed and inhibition. Research on whether inexpensive “brain games” that focus on processing speed and inhibition affect GC might be fruitful. Indeed, Payne and Stine-Morrow (2017) reported that a computerized WM training program improved memory for sentences and comprehension of ambiguous sentences. Similarly, Novick and colleagues (Hussey et al., 2017; Novick et al., 2014) found improvements in both syntactic ambiguity resolution and sentence comprehension following training with an executive function task requiring conflict resolution (i.e., a 6-back task with high conflict among trials). Unfortunately, to our knowledge, no one has looked at the effects of brain training games on a language production task. Furthermore, the extent to which cognitive slowing and inhibition deficits result from permanent changes in cortical function in PD is unknown, so these abilities may not be amenable to intervention in this group. Alternatively, a treatment that is focused directly on the macrolinguistic impairment in GC may yield more direct benefits (Rogalski et al., 2010). In other words, a therapeutic program incorporating discourse and topic maintenance goals could be one way of enhancing communication in PD.

Limitations and Future Directions

Although the PD and HOA groups were similar on a number of different descriptive and discourse variables, they did differ in age. The HOA group was significantly older than the PD group. Comparison to age-matched peers would likely increase the group differences in cognition and GC reported here. Future studies should compare a PD group with an age-matched control group. Another potential limitation was that point-to-point reliability was at the low end of what is considered “good reliability,” which may have affected these results. Future studies should take pains to ensure very high reliability in studies like this. Another possible limitation was the 3-min limitation on discourse duration. Based on the hypothesis above about how deficits in inhibition might result in increased GC2 and GC1 utterances, a longer sample might elicit an even larger proportion of those types of utterances. Furthermore, there may be gradually increasing instances of GC2 and GC1 utterances as the discourse continues. This could be a fruitful topic for future research. Additionally, the exceedingly high level of education in both groups may have masked real effects of WM and updating on GC. Research that specifically targets individuals with PD with lower education levels might significantly bolster understanding of the cognitive foundations of GC. Another limitation of this study is that cycling is a relatively automatic motor task. Future studies investigating dual task effects on GC should use a more difficult motor task like walking, which might yield very different results about how dual tasking could affect GC. Finally, future studies might investigate listener judgments of GC using discourse samples with the same mean GC but differing variability in GC levels to investigate how variability in GC within a discourse influences overall perception of its GC.

Conclusions

Our study is the first to document significant GC differences in a group of individuals with PD compared to HOAs and to link GC with specific cognitive abilities. Higher proportions of on-topic utterances were predicted by faster speed of processing, whereas off-topic utterances were predicted by slower processing speeds. In addition, somewhat off-topic tangential utterances were predicted by poorer inhibition, indicating a dissociation in the predictors for the proportions of utterances rated GC2 and GC1. Importantly, the relationships between GC and cognitive variables were identified across both groups, not just the PD group. On the other hand, there were significant group differences in GC4 and GC2 above and beyond what was predicted by cognitive factors. These findings are important clinically because they indicate specific cognitive abilities that could be trained in conjunction with an explicit macrostructure-focused intervention that might, in turn, improve GC, as well as overall communication with communication partners. Improved communication for those with PD could lead to improvements in their quality of life.

Author Contributions

Yvonne Rogalski: Conceptualization (Lead), Data curation (Lead), Formal Analysis (Lead), Methodology (Lead), Visualization (Lead), Writing – original draft (Lead). Sarah E. Key-DeLyria: Formal Analysis (Supporting), Writing – original draft (Supporting), Writing – review & editing (Lead). Audrey Hazamy: Data curation (Lead), Methodology (Supporting), Writing – review & editing (Supporting). Lori J. P. Altmann: Data curation (Lead), Funding acquisition (Supporting), Investigation (Lead), Writing – review & editing (Supporting), Writing – original draft (Supporting).

Funding Statement

Research reported in this publication was partially supported by the National Institute of Aging of the National Institutes of Health under award number R21AG033284. Participant recruitment, data collection, and initial transcriptions were funded by the National Institutes of Health (NIH), approximately 60% of the project, while the remainder of the project was funded by the National Parkinson Foundation Center of Excellence, the Inform Research Database at the University of Florida, and the home institutions of the authors, which funded the authors' time. The content is solely the responsibility of the authors and does not necessarily represent the official views of the NIH.