Dissociable language and executive control deficits and recovery in post-stroke aphasia: An exploratory observational and case series study

Abstract

A growing body of evidence indicates many, but not all, individuals with post-stroke aphasia experience executive dysfunction. Relationships between language and executive function skills are often reported in the literature, but the degree of interdependence between these abilities remains largely unanswered. Therefore, in this study, we investigated the extent to which language and executive control deficits dissociated in 1) acute stroke and 2) longitudinal aphasia recovery. Twenty-three individuals admitted to Johns Hopkins Hospital with a new left hemisphere stroke completed the Western Aphasia Battery-Revised (WAB-R), several additional language measures (of naming, semantics, spontaneous speech, and oral reading), and three non-linguistic cognitive tasks from the NIH Toolbox (i.e., Pattern Comparison Processing Speed Test, Flanker Inhibitory Control and Attention Test, and Dimensional Change Card Sorting Test). Two participants with aphasia (PWA) with temporoparietal lesions, one of whom (PWA1) had greater temporal but less frontal and superior parietal damage than the other (PWA2), also completed testing at subacute (three months post-onset) and early chronic (six months post-onset) time points. In aim 1, principal component analysis on the acute test data (excluding the WAB-R) revealed language and non-linguistic executive control tasks largely loaded onto separate components. Both components were significant predictors of acute aphasia severity per the WAB-R Aphasia Quotient (AQ). Crucially, executive dysfunction explained an additional 17% of the variance in AQ beyond the explanatory power of language impairments alone. In aim 2, both case patients exhibited language and executive control deficits at the acute post-stroke stage. A dissociation was observed in longitudinal recovery of these patients. By the early chronic time point, PWA1 exhibited improved (but persistent) deficits in several language domains and recovered executive control. In contrast, PWA2 demonstrated mostly recovered language but persistent executive dysfunction. Greater damage to language and attention networks in these respective patients may explain the observed behavioral patterns. These results demonstrate that language and executive control can dissociate (at least to a degree), but both contribute to early post-stroke presentation of aphasia and likely influence longitudinal aphasia recovery.

1.0. Introduction

The extent to which language and other cognitive processes are inter-related is a long-debated issue, one that had a foothold in disputes among early aphasiologists regarding the overlap of language and intelligence in stroke patients (see e.g., Broca, 1861; Head, 1915, 1921, 1926; Wernicke, 1970). Even today, the degree of interdependence between these abilities remains largely unanswered (Barker et al., 2020). While a lack of clearly defined and consistent terminology has plagued this line of inquiry (Head, 1921), in this paper we use the term “language” holistically and to mean an abstract, symbolic system of sounds or written symbols used to convey thoughts. Although not universally accepted (Crinella & Yu, 1999), some scholars (e.g., Butterfield & Belmont, 1977; Sternberg, 1985) propose the closest proxy to general intelligence is executive function (i.e., higher-order operations that govern the way we process information for future goals). Perhaps for this reason, executive function has been studied extensively in people with aphasia (PWA).

1.1. Executive dysfunction in aphasia

When considering executive dysfunction in PWA, an important consideration is that “executive function” is an umbrella term that comprises many cognitive subdomains. Indeed, when summarizing different assessments, Keil and Kaszniak (2002) divided executive function into four main categories: 1) concept formation and abstract reasoning, 2) set shifting and suppression, 3) concept generation, fluency and initiation, and 4) planning, strategy use and rule adherence. As an example, with regards to the first category, a plethora of evidence demonstrates that some PWA exhibit impairments in nonverbal problem-solving, decision-making, and reasoning tasks (e.g., Baldo et al., 2005; Baldo et al., 2010, 2015; Basso et al., 1973; Borod et al., 1982; Gainotti et al., 1986; Kertesz & McCabe, 1975; Lee & Pyun, 2014; Purdy, 2002). In these studies, the severity of reasoning difficulties is often significantly correlated with the degree of overall aphasia severity or deficits in specific language domains (e.g., Bailey et al., 1981; Baldo et al., 2005, 2010, 2015; Basso et al., 1981; Borod et al., 1982; Kalbe et al., 2005; Kim et al., 2020; Wall et al., 2017 but see Basso et al., 1973; Fonseca et al., 2019; Helm-Estabrooks, 2002; Kertesz & McCabe, 1975). One interpretation of these findings is that solutions to novel reasoning tasks are facilitated through inner language (reviewed in Perrone-Bertolotti et al., 2014), a conclusion that is supported by studies regarding thought and language relationships in children (e.g., Hermer-Vazquez et al., 1999; Lidstone et al., 2012; Loewenstein & Gentner, 2005).

In recent years, the construct of executive control—which most closely maps onto Keil and Kaszniak’s (2002) second category—has received increasing attention in the aphasia literature. Executive control—often used interchangeably with the cognitive control—encompasses the core cognitive skills of attention shifting, updating working memory, and inhibition (Miyake, Emerson, et al., 2000; Miyake, Friedman, et al., 2000). Related to the question of interactivity between language and cognitive systems, researchers have compared impairments on tasks taxing linguistic executive control (such as the Stroop task; Stroop, 1935) and non-linguistic, domain-general executive control tasks (such as the nonlinguistic Flanker task; Eriksen & Eriksen, 1974) in aphasia. Some studies (Gray & Kiran, 2016; Kuzmina & Weekes, 2017) have reported PWA exhibited impairments only in linguistic executive control, whereas others (Hamilton & Martin, 2005) reported impairments on both task types. Moreover, findings from studies focused solely on characterizing linguistic inhibition are mixed, with some researchers (e.g., Hunting-Pompon et al., 2015) finding PWA experience greater interference effects for such tasks than controls and others (e.g., Faroqi-Shah & Gehman, 2021) finding no between-group differences. Nonetheless, many studies report that language abilities are significantly related to linguistic and/or non-linguistic executive control skills in PWA (Allen et al., 2012; Baldo et al., 2005; Bonini & Radanovic, 2015; Faroqi-Shah & Gehman, 2021; Hunting-Pompon et al., 2015; Mohapatra & Marshall, 2020; Obermeyer et al., 2020; Olsson et al., 2019; Wall et al., 2017; Wong & Law, 2022). In general, these studies illustrate that even when groups of PWA do not exhibit deficits on non-linguistic executive tasks, relationships between language abilities and performance on both linguistic and non-linguistic executive control abilities still can exist.

These collective findings imply a tight coupling between language and various skills within the executive function umbrella. Indeed, some theoretical accounts of aphasia propose that language impairments in PWA are caused by a misallocation of attentional resources (Hula & McNeil, 2008; Luria, 1973; Shallice, 1982), a problem that is arguably intrinsically linked to executive dysfunction. Notably, however, many studies (e.g., Baldo et al., 2005; Faroqi-Shah & Gehman, 2021; Helm-Estabrooks, 2002; Kertesz & McCabe, 1975) have reported a high degree of inter-individual variability in executive dysfunction among people with similar degrees of language impairment. There is also evidence that some patients with severe aphasia with minimal to no receptive or expressive language have intact non-linguistic cognition, including executive function abilities (reviewed in Fedorenko & Varley, 2016; Varley, 2014). Along this vein, van Mourik et al. (1992) identified a group of individuals with nearly entirely spared cognition among 17 patients with global aphasia. Similarly, Marinelli et al. (2017) found that about 30 of 189 patients with global aphasia exhibited no evidence of executive dysfunction. In a group of 47 individuals with severe aphasia, Olsson et al. (2019) found that 79% of participants exhibited executive dysfunction but that significant relationships between language and executive function skills were found only in a subgroup of nonverbal individuals and not the verbal patient subgroup. Such findings suggest language and executive functions are modular constructs to some extent and imply the parts of the brain that mediate these skills may be at least partially distinct.

1.2. The neural bases of executive dysfunction in aphasia

Historically, executive dysfunction has been linked to frontal damage (Keil & Kaszniak, 2002; Luria, 1973; Shallice, 1982), and even updated models of the neural architecture of the frontal lobes demonstrate their importance in executive functioning (e.g., Badre & Nee, 2018; Lambon Ralph et al., 2017; Miller & Cohen, 2001). However, recent lesion symptom mapping (LSM) studies of executive dysfunction in PWA present mixed results. In acute stroke, Varjačić et al. (2018) implicated lesions in the left insula and adjacent white matter in poor set-switching abilities. In an acute left hemisphere stroke sample (that partially overlaps with the present study sample for research question 1), Meier et al. (2021) found that deficits in inhibitory control were related to lesions in the left intraparietal sulcus and that dorsolateral prefrontal cortex (DLPFC) damage was also implicated in a task requiring set-shifting, regions not implicated in naming impairments in the sample. In a series of studies with partially overlapping samples of patients with chronic aphasia, Lambon Ralph and colleagues found no significant lesion correlates of executive dysfunction in their first set of studies (Butler et al., 2014; Halai et al., 2017, 2018). In subsequent studies by this group (Alyahya et al., 2020; Schumacher et al., 2019; Zhao et al., 2020) and others (Baldo et al., 2005; Lacey et al., 2017), investigators reported relationships between various executive function deficits and abnormal or damaged voxels within frontal regions, but many other areas through the brain (e.g., middle, posterior inferior, and fusiform temporal gyri, temporooccipital cortex and occipital pole, precuneus, inferior parietal cortex) were also implicated.

Potential explanations for these disparate lesion findings come from the functional imaging literature. Functional imaging has demonstrated that complex cognitive functions are not mediated by isolated brain regions (e.g., frontal lobes alone) but instead are processed within intrinsic brain networks that span lobes and hemispheres. Several networks have been implicated in executive control processes, including the dorsal attention network (DAN; Fox et al., 2006), frontoparietal network (FPN; Cole et al., 2013), and cingulo-opercular network (Dosenbach et al., 2007). Prior fMRI studies (Balaev et al., 2016; Baldassarre et al., 2019; Geranmayeh et al., 2016; Sandberg, 2017) have shown PWA exhibit reduced or abnormal functional connectivity of these cognitive networks compared to healthy controls. Deficits in the skills these networks mediate would logically follow.

Furthermore, network neuroscientists are actively investigating the functional specificity of intrinsic brain networks as a means for determining which cognitive and neural resources do or do not overlap. Duncan, Fedorenko, and colleagues (Duncan, 2010, 2013; Fedorenko et al., 2011, 2012, 2013; Fedorenko & Thompson-Schill, 2014) propose that a multiple demand (MD) network comes online to exert top-down, domain-general cognitive control during various tasks as needed, particularly when tasks are difficult. Through the use of language and domain-general cognitive task functional localizers in healthy adults, these researchers have demonstrated that the language network is spatially distinct from the MD system, which encompasses the aforementioned networks and includes brain regions in both hemispheres (i.e., bilateral DLPFC, inferior frontal sulcus, frontal operculum, anterior insula, dorsal anterior cingulate cortex, and intraparietal sulcus [IPS]). However, Duncan, Fedorenko, and colleagues have also found that voxels activated during language processing are adjacent to voxels activated during domain-general cognitive tasks within the frontal lobe and other parts of the brain (e.g., parietal lobe) (Diachek et al., 2020; Duncan, 2010; Fedorenko et al., 2011, 2012, 2013; Mineroff et al., 2018; Wehbe et al., 2021). These findings imply that tissue important for either language or domain-general processing would be susceptible to infarct and subsequent damage in the event of stroke.

Moreover, it is important to note that cerebral infarcts occur along vascular territories, the boundaries of which do not perfectly align with functional networks (Charidimou et al., 2014). Specifically, aphasia most often arises from a left middle cerebral artery (MCA) stroke. While the left MCA supplies perisylvian frontal, temporal and parietal regions that are thought to comprise the canonical language network, it also supplies certain areas within other cognitive networks. Many regions within the MD network are supplied by the MCA (as visualized in Supplemental Figure 1). As such, it is logical to conclude that the degree to which PWA suffer from language impairments versus executive dysfunction is directly related to the relative damage to these established, spatially-distributed brain networks.

1.3. The current study

In this study, our goal was to investigate the extent to which language and executive control deficits dissociate in acute stroke and longitudinal aphasia recovery. Notably, much of the literature regarding executive function in aphasia is cross-sectional and uses group-level correlation analyses to demonstrate relationships between a specific type of language deficit (e.g., naming, auditory comprehension) and a certain subdomain of executive function (e.g., abstract reasoning). This approach does not provide information regarding the separability of language and cognitive constructs. As such, in our study, we first recruited a group of acute left hemisphere stroke survivors who completed not one but several linguistic and nonlinguistic cognitive assessments at a single time point. This approach ensured that the constructs under investigation were not unduly biased by the qualities of a single test and allowed us to measure different aspects of language (e.g., single-word naming, single-word writing, picture description abilities) and executive function (e.g., inhibition, set-shifting, rule maintenance). As described in greater detail in the section 2.2., our cognitive battery also included a measure of processing speed and basic visual attention with minimal burden on executive function, allowing us to frame interpretations of patients’ executive function performance against performance on a simpler attention task. The inclusion of multiple tests in itself does not solve the fact that neuropsychological assessments are rarely pure tests of a given underlying cognitive construct (Friedman & Miyake, 2017). Therefore, we then used a data-driven approach (i.e., principal component analysis) to determine the nature and number of components (i.e., cognitive constructs) present in the patient group dataset and to identify which linguistic and executive control assessments contributed to each component. This method provided information regarding the underlying structure of the data, including the degree of separation of various language and executive control measures within the acute stroke sample.

It can be argued that dissociations in single cases provide more robust evidence for the potential modularity of cognitive domains (McCloskey, 1993). In the 19th and 20th centuries, the study of single unique case patients (e.g., Phineas Gage, LeBorgne) or case series led to double dissociations that demonstrated how certain processes can be independently impaired, which launched the field of cognitive neuropsychology forward (Cubelli & Della Sala, 2017). Yet, the utility of single-case versus group studies was fiercely debated in the late 20th century (e.g., Caramazza & Badecker, 1989, 1991; Caramazza & McCloskey, 1988; Van Orden et al., 2001; Zurif et al., 1991), and since then, case study approaches have become less common in the neuroimaging age and the push towards big data. Nonetheless, single-case approaches, particularly when combined with modern neuroimaging techniques, can be critical for addressing gaps left by group studies (Deifelt Streese & Tranel, 2021; Medina & Fischer-Baum, 2017). Therefore, to address the main study aim, we also followed two individuals with acute post-stroke aphasia at subsequent subacute and early chronic post-stroke stages and report divergent recovery patterns in the context of damage to functional networks.

Using this combined cross-sectional exploratory and longitudinal case study approach, we addressed two research questions (RQs):

1. What are the contributions of language and executive control impairments on overall aphasia severity in an acute left hemisphere stroke sample?

   Based on the prior literature suggesting that language and executive function are related but potentially separable domains, we hypothesized that language assessment scores would load primarily onto one component and that executive control measures would load onto another component in the principal component analysis performed on data obtained from the group of acute left hemisphere stroke survivors. We predicted that linguistic impairments and non-linguistic executive control deficits (as captured by principal components) would independently contribute to acute overall aphasia severity.

2. Does acute to chronic recovery of language mirror longitudinal recovery of executive control in two PWA with different lesion distributions?

   The greatest lesioned area in both case patients was the left posterior temporoparietal cortex, but the extent to which other areas of the brain (e.g., anterior to mid temporal, frontal, and superior parietal regions) were affected differed between individuals. As such, we postulated that differences would be observed between patients in the extent of damage to language versus executive control functional networks. In terms of behavior, we predicted that both cases would demonstrate deficits in both language and executive function domains that would improve over time. Additionally, we hypothesized that the extent of recovery of executive control deficits in the patient with primarily temporal damage (PWA1) would exceed recovery observed in an individual with less temporal damage and greater parietal and DLPFC damage (PWA2) and that the inverse would be true of language impairments.

2.0. Material and methods

2.1. Participants

2.1.1. Patients

Patients who were admitted to Johns Hopkins Hospital (JHH) between February 2019 and November 2020¹ with a diagnosis of an acute left hemisphere stroke were recruited to participate in this study. Inclusion criteria included premorbid proficiency in English, normal or corrected-to-normal vision and hearing, and the capacity to complete testing protocols (e.g., alertness, medical stability). The exclusion criterion was a history of neurological disease affecting the brain other than stroke. Twenty-three individuals (10 women, mean age = 63.22 ± 9.78 years, mean education = 15.13 ± 2.70 years, all right-handed) completed the testing battery described in section 2.2 at the acute phase of recovery (mean time post-stroke onset = 5.09 ± 3.58 days). Two individuals with aphasia were followed at subsequent subacute (patient 1 [PWA1]: 98 days and patient 2 [PWA2]: 77 days after onset) and early chronic (PWA1: 238 days and PWA2: 214 days after onset) post-stroke time points. Additional information regarding these two individuals is provided below.

2.1.2. Case series of participants with aphasia

The two cases were selected based on their language profiles (i.e., both presented with aphasia), their lesion profiles (i.e., presence of large, left cortical lesions), and the availability of longitudinal data. The difference between participants in temporal versus frontoparietal damage was based on initial visual inspection of acute neuroimages.

PWA1 was a right-handed, 62-year-old man who was admitted to JHH in May 2019, presenting with mild right-sided facial droop and aphasia characterized by poor comprehension, repetition, and naming. PWA1’s past medical history included type II diabetes, hypertension, atrial fibrillation, cardiomyopathy, complete heart block status post pacemaker, and alcohol and tobacco misuse. Prior to his stroke, PWA1 was retired and living independently. JHH clinical radiology notes for PWA1 indicated restricted diffusion within the left temporal and parietal lobes, consistent with an acute ischemic infarct of likely cardioembolic etiology.

PWA2 was a right-handed, 60-year-old man who was admitted to JHH in June 2019, presenting with reduced speech output, word retrieval difficulties, and right upper limb weakness. PWA2’s past medical history included type II diabetes, hypertension, hepatitis C, and tobacco and substance abuse. Prior to his stroke, PWA2 was retired and living independently. PWA2’s clinical radiology notes indicated areas of restricted diffusion in the left middle frontal and precentral gyri as well as left temporal and parietal lobes, consistent with an acute ischemic stroke of a probable atheroembolic etiology.

Both individuals received speech-language therapy consistent with usual care during their hospitalization at JHH. Following the acute and into the subacute post-stroke phases, both PWA participated in a 15-session combined language and noninvasive brain stimulation treatment study aimed at improving lexical access. This treatment was not a focus of the present study, but more details about the treatment methods can be found in the Supplemental Material. Neither participant reported completing additional speech-language therapy in the time between completing the aforementioned treatment study and the 6-month follow-up time point.

2.1.3. Neurotypical controls

To evaluate executive control task performance in the two cases, a control group was identified from a publicly available database of data from 5,396 community-dwelling children and adults (ages 3–85 years old) who completed the “NIH Toolbox Norming Study” (Gershon, 2016). We selected participants with complete datasets of the NIH Toolbox (NIH-TB) tests that were included in this study (see section 2.2.2.) who reported English as a primary language and were similar in age and years of education (i.e., highest grade level of school completed) as the two cases. The final control sample included 132 individuals (84 women) with a mean age of 60.50 ± 3.11 years (range: 55–65) and mean education of 15.22 ± 1.87 years (range: 13–20).

2.1.4. Informed consent and study approval

Each left hemisphere stroke participant or their healthcare proxy (in the case of individuals with impaired comprehension) provided written informed consent upon study entry. Study protocols were approved by the Johns Hopkins University Institutional Review Board and adhered to the Declaration of Helsinki.

2.2. Neuropsychological assessments

2.2.1. Language battery

Only the left hemisphere stroke survivors completed language assessments. At the acute time point for all patients and at subsequent time points for the two cases, we used Part 1 of the Western Aphasia Battery-Revised (WAB-R; Kertesz, 2007) to characterize deficits in auditory comprehension, repetition, and word finding/naming domains as well as determine each patient’s aphasia classification and overall aphasia severity per the Aphasia Quotient (AQ). We assessed more fine-grained language and linguistic skills in PWA through several additional measures.

Naming was further characterized using a 30-item shortened version of the Boston Naming Test (BNT; Fisher et al., 1999) for object naming and the 30-item Hopkins Action Naming Assessment (HANA; Breining et al., 2021) for action naming. In both assessments, participants were shown line drawings of objects (BNT) or actions (HANA) and asked to name them aloud. Final, spontaneously given correct responses or correct answers following a stimulus cue (BNT only) were counted as accurate. In addition to the total number of correct responses, a speech-language pathologist (CRK) classified BNT naming errors according to the 10-error classification schema described by Goodglass et al. (2001) for the two case study participants. We then condensed errors into four main types: semantic (i.e., real-word errors related to the target in meaning), phonological (i.e., nonword and real-word phonemic paraphasias with > 50% phonological overlap with the target word), mixed (i.e., real-word errors semantically and phonologically related to the target word), and unrelated (i.e., single or multi-word responses unrelated to the targeted in semantics or phonology). For the HANA, we used a similar classification schema with the additional category of morphological errors (e.g., named object rather than action).

We used a published, 14-item version of the Pyramids and Palm Trees Test (PPT; Breining et al., 2015) and an in-house shortened, 15-item version of the Kissing and Dancing Test (KDT; Bak & Hodges, 2003) to capture nonverbal associative semantic skills for objects and actions, respectively. For these assessments, participants pointed to one of two line drawings that was associated in meaning with a target item. While the picture versions of the PPT and KDT do not explicitly require language, participants were not forbidden from naming or describing the pictures when making their decision. Performance was captured by the total number of correct trials for all individuals.

To index discourse abilities, patients completed the original (Goodglass et al., 2001) and new (Berube et al., 2019) Cookie Theft picture description tasks. In each task, participants were asked to describe the pictured scene, and their responses were recorded and subsequently transcribed. From each transcribed Cookie Theft sample, we obtained measures of total content units (CU), i.e., objects and actions pictured in the scene, reflecting lexical-semantic skills, and number of syllables/CU, reflecting communication efficiency of content. Scoring was done according to procedures described in Yorkston & Beukelman (1980).

Finally, we used the Psycholinguistic Assessments of Language Processing in Aphasia (PALPA; Kay et al., 1992) subtest 35: Spelling-Sound Regularity Reading to capture single-word reading abilities. During this test, patients were asked to orally read a list of 30 regularly-spelled and 30 irregularly-spelled single words. Performance was measured in terms of total correct items within each word type category. To address RQ1, we collapsed across word types to yield a single single-word oral reading score for each patient.

2.2.2. Cognitive battery

All participants completed the Pattern Comparison Processing Speed (PCPS) Test, the Flanker Inhibitory Control and Attention (FICA) Test, and the Dimensional Change Card Sort (DCCS) Test from the NIH Toolbox (NIH-TB) Cognition Battery (Gershon et al., 2013). The latter two tasks were selected to measure distractor inhibition (both tasks) and set-shifting (DCCS Test) abilities. The PCPS Test was selected based on findings from prior work demonstrating that PWA often exhibit slowed domain-general processing speed and a variety of attention deficits that can impact other executive function skills (Faroqi-Shah & Gehman, 2021; Korda & Douglas, 1997; Murray, 2012; Spaccavento et al., 2019; Villard & Kiran, 2015, 2018), and that the PCPS Test requires less engagement of executive function skills. The NIH-TB tasks have been validated in healthy adults and individuals with acquired brain injury (Heaton et al., 2014; Tulsky et al., 2017; Weintraub et al., 2014) and were used over other standardized assessments because they are quick and simple to administer (≤ four minutes per test). In the current study, patient participants completed the tasks on an iPad through the NIH-TB application.

For each trial of the PCPS Test, participants saw two images and decided as quickly as possible whether the images matched by selecting a YES button for a match or NO for mismatch. Match and mis-match trials were randomly presented. In mismatch trials, the images differed either in color, the addition or subtraction of a component (e.g., flower stem), or the number of items within the image (e.g., one versus many clouds). Participants completed as many trials as possible within 85 seconds. Task performance was indexed by accuracy per trial as well as total number of trials completed within the allotted time. In the present study, we used the total number of completed trials as the proxy metric of visual processing speed given that most participants performed at near-ceiling accuracy.

The FICA Test measures inhibition of visual distractors, similar to the classic Flanker task (Eriksen & Eriksen, 1974). For each trial of the FICA Test, participants saw a row of arrows and indicated which direction (left or right) the middle arrow faced. The arrows flanking the middle arrow either faced the same (i.e., congruent) or different (i.e., incongruent) direction as the target arrow. Between trials, participants placed their index finger on “home base”, a colored circle placed three inches from the edge of the iPad. The task included 20 trials, and trial-by-trial accuracy and reaction time (RT) data were captured by the application. We indexed performance by accuracy and average RT measures for each participant.

For each task trial of the DCCS Test, participants decided which one of two pictures matched a target image according to a target dimension (i.e., shape or color). During the task, the target dimension for “repeat” trials matched the preceding trial’s dimension, whereas the match dimension of “switch” trials differed from the preceding trial’s dimension. Similar to the FICA Test, participants placed their index finger on “home base” between trials. The task included 30 trials, and trial-by-trial accuracy and reaction time (RT) data were recorded by the application. Like the FICA Test, we indexed performance by accuracy and average RT measures for each participant.

2.3. Neuroimaging methods

Upon admission to JHH, both case patients with aphasia underwent a clinical imaging protocol that included a diffusion-weighted imaging (DWI) sequence². Trained technicians manually delineated the acute stroke volume visible on DWI in a slice-by-slice fashion using MRIcron (Rorden & Brett, 2000). DWI b0 maps were warped to an older adult template in standard space (Rorden et al., 2012) using SPM12 routines (https://www.fil.ion.ucl.ac.uk/spm/software/spm12/), and the normalization parameters were subsequently applied to the lesion maps.

To obtain an objective measure of the integrity of brain tissue critical for the language and executive control processes under investigation, we quantified the extent of damage to functional networks of interest. We used Neurosynth (Yarkoni et al., 2011; https://neurosynth.org/analyses/terms/), an online repository of fMRI meta-analytic data, to search the terms “attention” (1831 contributing studies), “executive” (786 contributing studies), and “language” (1101 contributing studies) and downloaded the corresponding association test activation map for each network (corrected for multiple comparisons at a false discovery rate [FDR] of 0.01). We subsequently binarized the images, resampled and warped the network maps to the older adult template space, and isolated the left hemisphere (since patients had left hemisphere-only strokes). We then created subject-specific network masks by intersecting the normalized lesion maps with the final left hemisphere only-network masks. We determined total lesion volume and the amount of damaged tissue within each left hemisphere functional network for each patient using MarsBaR routines (Brett et al., 2002).

Notably, studies included in each search term in Neurosynth are based on a procedure by which the full text of each article is tagged for frequently occurring terms, and for each term of interest, the entire database is divided into two sets of coordinates (i.e., studies including the term versus studies without the term) (Yarkoni et al., 2011). Therefore, the spatial distribution of each mask relies on consistent use of terminology, agreed-upon theoretical underpinnings, and gold standard, “pure” neuropsychological assessments used across researchers regarding constructs of interest. As highlighted in the Introduction, this is an issue regarding neuropsychological assessment of cognition, particularly of executive function; as such, while we present the amount of damage incurred to each mask in both cases, we also report and discuss the percent damage to a combined mask that includes voxels from both the attention and executive masks.

2.4. Statistical approach

Statistical analyses were performed using R 4.1.0 (R Core Team, 2021) and the psych (Revelle, 2021), RVAideMemoire (Hervé, 2021), and singcar (Rittmo & McIntosh, 2021) packages. To address RQ1, we first conducted a principal component analysis (PCA) with varimax rotation on the language and cognitive data from the acute stroke group. Language measures included total correct items on the BNT, HANA, PPT, KDT, and PALPA 35 as well as original Cookie Theft total CU. Cognitive measures included the total number of trials on the PCPS test and accuracy and average RTs from the FICA and DCCS Tests. WAB-R domain scores were not entered into the PCA because WAB-R AQ served as our proxy for overall aphasia severity for this aim. The rationale for implementing PCA was to ascertain the underlying structure of the data and determine the amount of variance in overall aphasia severity each orthogonal component explained. We retained components with eigenvalues ≥ 1.0. Assessments with loadings of ≤ −0.50 or ≥ 0.50 onto a given component were assigned to that component in analysis interpretations. We extracted single-subject factor loadings onto each retained component which were used as independent predictors of acute WAB-R AQ scores via hierarchical regression³. We performed checks on model assumptions using the car (Fox & Weisburg, 2011), gvlma (Pena & Slate, 2014), and MASS (Venables & Ripley, 2002) R packages.

To address RQ2, we used Cochran’s Q tests to determine changes over time (at acute, subacute and early chronic stages) in the two case patients’ performance on language (i.e., BNT, HANA, PPT, KDT, new Cookie Theft CU, and PALPA35) and NIH-TB accuracy measures. Cochran’s Q test is an extension of the McNemar Chi-squared test that compares paired dichotomous response data at more than two time points. In the event of a significant Cochran’s Q test result, we conducted pairwise comparisons using McNemar Chi-squared tests and corrected p-values for multiple comparisons at a FDR of q < 0.05. We also assessed each patient’s degree of impairment at each time point according to published norms for the language measures (excluding the KDT) or compared to neurotypical control performance for NIH-TB metrics. For the language measures, we considered impaired performance to be scores greater than 1.5 standard deviations below the healthy control means reported in Breining et al. (2021) for the BNT and HANA, Breining et al. (2015) for the PPT, Berube et al. (2019) for new Cookie Theft Total CU and syllables/CU, and Kay et al. (1992) for PALPA 35. For the NIH-TB comparisons, we used the Bayesian Test of Deficit described by Crawford and colleagues (Crawford et al., 2011; Crawford & Garthwaite, 2007). This test determines the posterior probability that a case patient is more impaired than a control group sample and outputs a standardized case score (z_cc), 95% confidence intervals, and the percentage of the control sample likely to score below the case patient. Given potential influences of demographics (age, education, sex) on executive functions, we determined relationships between NIH-TB measures (accuracy, RTs) and demographics in the healthy control group prior to conducting the case-control comparisons. Shapiro-Wilk tests of normality revealed most control sample NIH-TB measures were not normally distributed, so nonparametric tests (Spearman correlations, Wilcoxon rank sum tests) were used. In subsequent case-control comparisons, we controlled for any demographic variable significantly related to NIH-TB performance in the control sample.

3.0. Results

3.1. RQ1: What are the contributions of language and executive control deficits on acute overall aphasia severity?

The PCA of the acute behavioral scores yielded two components that explained approximately 71.5% of the variance in test performance. As shown in Figure 1, all language measures plus DCCS accuracy scores loaded onto component 1. The remaining NIH-TB measures loaded onto component 2. Given these loadings, we coined component 1 “language” and component 2 “executive control”.

Figure 1.

Acute sample results. Language and executive control assessments loaded primarily onto separate factors (left); the opacity of the shading reflects loading strength (i.e., more opaque, higher loading), and the hue reflects loading direction (i.e., red = negative loading and blue = positive loading). Language and executive control component variables were both significantly related to overall aphasia severity in the acute stroke group (right). Of note, the patients included in the hierarchical regression (n = 22, excluded PWA2) had WAB-R AQ scores that ranged from 7.5 to 100. The single-subject factor loadings are weighted scores that reflected each patient’s scaled performance on tests that loaded highly onto each component. Lower single-subject factor loadings reflect more impaired performance within a given domain; the opposite is true of higher scores.

Next, we use hierarchical regression to determine the independent contributions of each factor to acute aphasia severity. The base linear regression model included just the language component. The overall model was significant (F(1,20) = 50.33, p < 0.001, R² = 0.716), and the language component was significantly related to AQ (β = 19.730, SE = 2.781, t = 7.094, p < 0.001). The addition of the executive control component to the base model also resulted in a significant overall model (F(2,19) = 77.71, p < 0.001, R² = 0.891) and both the language (β = 19.730, SE = 1.766, t = 11.172, p < 0.001) and executive control (β = 9.769, SE = 1.766, t = 5.532, p < 0.001) components were significant predictors of AQ scores (Figure 1). To ensure demographic characteristics did not drive the latter set of results, we conducted a third model adding age (in years), sex (dummy-coded with female as the referent), and education (in years) as nuisance regressors to the second model. The overall model was significant (F(5,16) = 29.82, p < 0.001, R² = 0.903, adjusted R² = 0.873). Holding the other variables constant in this model, the language (β = 20.436, SE = 2.024, t = 10.098, p < 0.001) and executive control (β = 10.546, SE = 2.244, t = 4.700, p < 0.001) components remained significant predictors, but age (β = 0.051, SE = 0.233, t = 0.220, p = 0.829), sex (β = 2.794, SE = 3.644, t = 0.767, p = 0.454), and education (β = −0.917, SE = 0.743, t = −1.233, p = 0.235) were not significant.

Comparison of the base model (with only the language component) and the second model (with both components) via ANOVA indicated that the second model was a significantly better predictive model than the base (p < 0.001) and that executive control deficits added an additional 17% explained variance in AQ beyond the explanatory power of language impairments. Comparison of the second and third models indicated that the inclusion of demographics did not improve the prediction of WAB-R AQ (p = 0.360). Checks on linear regression assumptions indicated significant link function violations for both the language-only base (p < 0.002) and language plus executive control (p < 0.003) models, driven by outliers (unadjusted p < 0.05) on both ends of the WAB-R AQ range that influenced model residuals. Therefore, we re-ran these two models using robust regression using the Huber method, which de-weights the influence of outliers (Rousseeuw & Leroy, 2005) (i.e., n = 4 patients in the base model, n = 2 patients in the second model). While the strength of the relationships weakened, both components remained significant predictors of WAB-R AQ in the respective models (see Supplemental Table 2 for complete robust regression results). These collective findings indicate that while much of the variance in acute overall aphasia severity was explained by language impairments, executive dysfunction additionally contributed to overall aphasia severity at the acute post-stroke stage.

3.2. RQ2: Does longitudinal recovery of language mirror longitudinal executive control recovery in two PWA with different lesion distributions?

3.2.1. Network lesion characteristics

Figure 2 depicts each patient’s lesion map overlaid onto the left hemisphere attention, executive, and language network masks. PWA1’s total lesion volume was 90.99 cc whereas PWA2’s lesion volume was nearly half the size, at 46.58 cc. PWA1 had greater damage to the language network mask (33.4%) than PWA2 (8.2%). Only 3.2% of the attention mask was damaged in PWA1 compared to greater damage (45.1%) in PWA2. Both patients had minimal damage to the executive mask (PWA1: 3.6%, PWA2: 2.3% damage). Despite the lack of consistent use of terminology in the field, there was minimal spatial overlap between the attention and executive masks with only 200 voxels overlapping in the masks (corresponding to 2.8% of the attention mask and 2.5% of the executive mask). As such, the pattern of damage to the combined attention+executive mask largely aligned with the attention mask-only results: only 3.4% of the combined mask was damaged in PWA1 compared to 23.0% of the combined mask in PWA2.

Figure 2.

Overlap of case study patients’ lesions and attention (ATTN, top), executive (EXEC, middle), and language (LANG, bottom) network masks. In each section, the masks are shown in green, lesions in red, and the overlap is in yellow.

3.2.2. Language deficits and recovery

Table 1 includes language scores across all three time points for both case study patients.

Table 1.

Language assessment scores across time for case patients

	PWA1			PWA2
	Acute	Subacute	Chronic	Acute	Subacute	Chronic
Days Post Stroke	3	98	238	4	77	214
WAB-R Spontaneous Speech (/20)	2	14	18	13	18	16
WAB-R Auditory Verbal Comp (/10)	0.45	8.8	8.55	4.3	8.45	8.95
WAB-R Repetition (/10)	1	4.5	4.10	5.7	9.8	9.2
WAB-R Naming & Word Finding (/10)	0.3	9.1	8.60	6.4	8.8	9.3
WAB-R Aphasia Quotient (/100)	7.5	72.8	78.5	58.8	90.1	86.9
WAB-R Aphasia Subtype	Global	Conduction	Conduction	Wernicke’s	Anomic	Anomic
BNT (/30)	1*	14*	14*	22	23	26
HANA (/30)	0*	6*	14*	18	22	25
PPT (/14)	13	14	14	14	12*	14
KDT (/15)	7	13	15	14	13	14
New Cookie Theft Total CU	2*	37	24	15	24	18
New Cookie Theft Syllables/CU	10.00*	6.90	6.13	14.00*	8.92*	12.06*
PALPA 35 Regular Words (/30)	2*	25*	27*	18*	25*	24*
PALPA 35 Irregular Words (/30)	2*	18*	16*	17*	25*	28*

Notes: In the latter half of the table, bolded scores followed by an asterisk indicate scores that fall below the cutoffs for impairment based on published norms. Impairment cutoffs based on published norms are reported in Supplemental Table 1.

Abbreviations: WAB-R = Western Aphasia Battery-Revised, BNT = Boston Naming Test, HANA = Hopkins Action Naming Assessment, PPT = Pyramids and Palm Trees Test, KDT = Kissing and Dancing Test, CU = content units, PALPA = Psycholinguistic Assessments of Language Processing in Aphasia.

3.2.2.1. PWA1

At the acute stage, PWA1 was classified as having global aphasia and demonstrated severe impairments in auditory comprehension, repetition, and word retrieval. Compared to published norms, he scored below normal limits on all standardized assessments (see Table 1). Consistent with these impairments, he exhibited difficulty understanding and responding to basic social questions (e.g., “what’s that? I’m sure too much I try” to “how are you today?”). In structured picture description tasks, he produced little meaningful content (e.g., complete new Cookie Theft picture description: “That’s a door. Two t- two /gɔr/ a /gɔr/ um. Man two there there over there yes cooking”). He exhibited severe anomia, and the majority of his errors on naming tests were unrelated to the target items (52% of BNT errors, 60% of HANA errors) with semantic and morphological errors being the second most common error types on the BNT (40% of errors) and HANA (30% of errors), respectively. While his oral naming skills for objects and actions was comparable, he exhibited worse nonverbal semantic association skills for actions compared to mostly intact object knowledge. On PALPA 35, PWA1 attempted oral reading via phoneme-to-grapheme conversion but was unsuccessful in most attempts, resulting in primarily phonemic paraphasias and real-word formal errors for both regularly (e.g., “envoy” for envy, “plunk” for plank) and irregularly (e.g., “boil” for bowl, /wɔrf/ for wolf) spelled words.

By the subacute testing time point, PWA1 exhibited improved abilities in all language domains on the WAB-R but demonstrated the greatest persistent impairments in repetition, consistent with his subacute diagnosis of conduction aphasia. With a couple notable exceptions, PWA1’s language skills at six months post-stroke onset were comparable to his 3-month abilities. At the 6-month time point, he was still classified with conduction aphasia, and most domains remained stable (or slightly decreased) compared to the subacute time point with the exception of his spontaneous speech score.

PWA1 still exhibited impaired object and action naming at both the subacute and early chronic time points. However, he displayed an overall significant increase in BNT object naming over time (χ²(2) = 22.53, p < 0.001); pairwise comparisons revealed significant change from the acute to early chronic time points (Z = −3.61, q < 0.001) and acute to subacute stages (Z = −3.61, q < 0.001) but not subacute to early chronic stages (Z = 0, q = 1). He also exhibited significant change over time in HANA action naming (χ²(2) = 19.73, p < 0.001) with significant gains from acute to early chronic time points (Z = −3.74, q < 0.001), acute to subacute stages (Z = −2.45, q = 0.014) as well as subacute to early chronic stages (Z = −2.53, q = 0.014). At the subacute time point, PWA1’s predominant error types on the BNT and HANA were semantic, phonological, and mixed errors (as well as morphological errors during verb naming) with relatively few unrelated errors compared to the acute time point. By the early chronic stage, PWA1 produced equal proportions of semantic, phonological, and unrelated/morphological errors on the BNT/HANA.

In terms of nonverbal semantics, PWA1 had perfect scores on the PPT at the subacute and early chronic stages. Given his near perfect score on the PPT at the acute stage, no significant changes were observed over time (χ²(2) = 2.00, p = 0.368). In contrast, PWA1 demonstrated significantly improved semantic knowledge for actions at subsequent time points (χ²(2) = 11.56, p = 0.003). Pairwise comparisons indicate his KDT score significantly improved from acute to early chronic time points (Z = −2.83, q = 0.014) and trended towards significance between acute and subacute phases (Z = −2.12, q = 0.051) but that PWA1’s KDT scores were comparable between subacute and early chronic time points (Z = −1.41, q = 0.157).

At follow-up time points, PWA1’s output for picture description tasks was characterized by hyperverbosity (particularly at the subacute stage), but he produced more meaningful content in a more efficient manner compared to his acute time point performance. Indeed, PWA1’s total CU and syllables/CU were within normal limits at both follow-up time points. Consistent with this information, the Q-test revealed PWA1’s total CU on the new Cookie Theft picture description task changed over time (χ²(2) = 45.80, p < 0.001). He produced significantly more CU from the acute to early chronic (Z = −4.49, q < 0.001) and acute to subacute phases (Z = −5.75, q < 0.001). He produced fewer CU at the early chronic time point compared to the subacute phase (Z = 2.84, q = 0.005), but both scores were comparable to healthy control performance. Of note, despite having normal CU and syllables/CU scores at the two follow-up time points, his picture description samples still lacked core content as most utterances focused on the people in the scene but not the key actions (e.g., stealing cookies, stool tipping over).

While still impaired compared to published norms at both follow-up time points, PWA1’s single-word oral reading also greatly improved for both regular (χ²(2) = 44.54, p < 0.001) and irregular words (χ²(2) = 19.83, p < 0.001) on PALPA 35. Pairwise comparisons revealed similar patterns for both word types: his performance increased from acute to early chronic stages (regular: Z = −5.00, q < 0.001, irregular: Z = −3.50, q < 0.001), a finding driven by the change from acute to subacute time points (regular: Z = −4.80, q < 0.001, irregular: Z = −3.77, q < 0.001) given minimal change from the subacute to early chronic phases (regular: Z = −1.00, q = 0.317, irregular: Z = 0.58, q = 0.564). At both the subacute and early chronic time points, PWA1 often regularized irregularly-spelled words (e.g., /yæk/ for yacht, /bukwɪt/ for bouquet), consistent with surface dyslexia.

3.2.2.2. PWA2

At the acute time point, PWA2 was classified with Wernicke’s aphasia. He exhibited impairments in all language domains but to a lesser extent than PWA1’s acute language deficits. PWA2 understood and responded appropriately to most personal and basic conversational questions but exhibited difficulty with comprehending more complex questions, recognizing single auditory words, and following directions with more than one step. His picture description included many core concepts (e.g., reference to “stealing some cookies”) but also a high number of mazes and some paraphasias (e.g., “They they the the um the child the dog is doing the dishes for the child”). PWA2 scored on the low end of the normal range for both the BNT and HANA; many of his final accurate responses were self-corrections. For incorrect responses, he produced predominantly semantic errors on both tests (71% of BNT errors, 44% of HANA errors). He scored at or near ceiling on the PPT and KDT, indicating relatively intact semantic knowledge for objects and actions. He exhibited comparably impaired oral reading for regularly and irregularly spelled words.

At the subacute and early chronic stages, PWA2 exhibited improved overall language abilities and was classified with anomic aphasia at both time points. PWA2 made slight gains in oral naming over time, but given his already relatively high acute scores, no significant changes were observed for either the BNT (χ²(2) = 3.25, p = 0.197) or the HANA (χ²(2) = 5.69, p = 0.058). On both tests and at both time points, PWA2’s naming errors were mainly semantic in nature (circumlocutions, category coordinate or category superordinate responses) in addition to some morphological errors (e.g., “prescription” for prescribe) on the HANA. Excluding a slightly lower 3-month PPT score, PWA2 exhibited comparable performance on the PPT (χ²(2) = 4.00, p = 0.135) and KDT (χ²(2) = 0.67, p = 0.717) across time points. Unlike the acute time point, PWA2’s total new Cookie Theft CU count was comparable to controls’ at the early subacute and chronic time points, but no significant changes in the CU totals were seen (χ²(2) = 3.15, p = 0.207). Of note, the number of syllables/CU PWA2 produced was greater than 1.5 SDs above the control mean at each time point, indicating he was less efficient than controls in communicating salient picture content.

Similar to PWA1, PWA2 exhibited improvements over time in single-word oral reading on PALPA 35 for both regular (χ²(2) = 7.17, p = 0.028) and irregular (χ²(2) = 13.86, p < 0.001) words, yet he remained impaired at each time point. Pairwise comparisons revealed significant or near-significant changes in both word types from acute to early chronic (regular: Z = −2.12, q = 0.051, irregular: Z = −3.32, q = 0.003) and acute to subacute (regular: Z = −2.33, q = 0.051, irregular: Z = −2.31, q = 0.031) time points; changes from the subacute to early chronic time point were not significant (regular: Z = 0.38, q = 0.705, irregular: Z = −1.34, q = 0.180). PWA2’s reading errors were mostly phonological in nature (e.g., “blue” for blood, “content” for context) with some errors for irregular words made due to adherence to typical letter-to-sound correspondence (e.g., /sʌwɔrd/ for sword).

3.2.3. Executive control deficits and recovery

We found no significant relationships between demographics (age, education, and sex) and NIH-TB measures (i.e., PCPS number of trials and FICA and DCCS proportion correct and average RT) in healthy controls (p > 0.216). Therefore, we did not control for demographics in the case-control comparisons. NIH-TB performance at a single time point for controls and across time points for both case patients is shown in Figure 3.

Figure 3.

NIH-TB performance in controls (left) versus case patients (right) for the A) Pattern Comparison Processing Speed (PCPS) Test total number of items, B) Flanker Inhibitory Control and Attention (FICA) Test proportion correct (left) and average reaction time (RT, right), and C) Dimensional Change Card Sort (DCCS) Test proportion correct (left) and average RT (right). Note that PWA2 refused to complete the DCCS Test beyond the practice trials at the acute stroke phase. Annotations within each bar in the patient graphs denote the results of statistical comparisons between each patient and the control group at that time; asterisks indicate significantly impaired performance in PWA compared to controls. The annotations above the brackets in the FICA and DCCS accuracy plots reflect significance testing for changes over time per the Cochran Q-Tests. P < 0.001***, < 0.01**, < 0.05*, n.s. = not significant.

3.2.3.1. PWA1

At the acute time point, PWA1 exhibited impaired performance on all NIH-TB measures. PWA1 completed significantly fewer PCPS Test trials than the control group (z_cc = −3.831, [CI: −4.34, −3.35], p < 0.001), and only 60% of responses were correct on these few trials. On the FICA Test, PWA1 was significantly less accurate (z_cc = −3.968, [CI: −4.48, −3.44], p < 0.001) and slower (z_cc = −7.355, [CI: −8.27, −6.43], p < 0.001) than the control group. Similarly, PWA1 had significantly poorer accuracy (z_cc = −13.124, [CI: −14.78, −11.54], p < 0.001) and longer RTs (z_cc = −2.710, [CI: −3.08, −2.34], p = 0.004) than controls on the DCCS Test. Less than 1% of the control sample exhibited poorer performance than PWA1’s acute scores across measures.

PWA1 demonstrated gains across NIH-TB metrics over time to the extent no differences were found for several measures in the case-control comparisons. PWA1 completed the same number of PCPS trials at the subacute and early chronic time points with 100% accuracy. His total trial score did not significantly differ from controls (z_cc = −0.234, [CI: −0.41, −0.06], p = 0.408); of note, approximately 41% (CI: 34.09, 47.61) of controls completed fewer PCPS trials than he did. PWA1 did not exhibit significantly increased FICA accuracy over time (χ²(2) = 2.00, p = 0.368), yet the slight increase to 100% accuracy that he attained at the subacute time point and retained by the chronic stage resulted in no differences between his scores and controls (z_cc = 0.236, [CI: 0.06, 0.41], p = 0.593). PWA1’s FICA average RT also decreased over time to the point his RTs did not significantly differ from controls at either the subacute (z_cc = 0.468, [CI: 0.29, 0.64], p = 0.679) or early chronic (z_cc = 0.387, [CI: 0.21, 0.56], p = 0.650) time points.

PWA1 also demonstrated significantly increased accuracy on the DCCS Test over time (χ²(2) = 29.368, p < 0.001). Pairwise comparisons revealed PWA1 had significant increases in DCCS Test accuracy from the acute to chronic (Z = −4.24, q < 0.001) and the subacute to chronic (Z = −3.87, q < 0.001) time points but no significant change from the acute to subacute stages (Z = −1.34, q = 0.180). Despite these gains, PWA1 still had significantly poorer DCCS accuracy than controls at the subacute (z_cc = −11.325, [CI: −12.67, −9.92], p < 0.001) and early chronic (z_cc = 0.387, [CI: 0.21, 0.56], p = 0.650) time points, with approximately 1% or fewer controls exhibiting poorer accuracy than PWA1. PWA1’s subacute (z_cc = 1.163, [CI: 0.94, 1.380], p = 0.876) and chronic (z_cc = −1.081, [CI: −1.29, −0.86], p = 0.142) stage average RT did not differ from controls, meaning that he was as quick to respond as the majority of controls but with a higher error rate.

3.2.3.2. PWA2

Similar to PWA1, PWA2 exhibited impaired performance on all measures of completed NIH-TB tasks at the acute time point. PWA2 completed more PCPS Test trials than PWA1 but still significantly fewer items than the control group (z_cc = −2.906, [CI: −3.29, −2.52], p = 0.002) with less than 1% of controls completing fewer trials. His accuracy on completed PCPS Test trials was 86%. PWA2 also exhibited poorer accuracy (z_cc = −30.883, [CI: −34.61, −27.13], p < 0.001) and longer RTs (z_cc = −5.723, [CI: −6.45, −5.00], p < 0.001) than the control group on the FICA Test. PWA2 attempted the DCCS Test but requested to terminate the task after four practice trials.

At subsequent time points, PWA2 completed more PCPS trials than he did at the acute stage but still significantly fewer than controls (z_cc = −2.598, [CI: −2.96, −2.24], p = 0.005). Of the trials he completed, he had some incorrect responses (82% and 94% accuracy at subacute and early chronic time points, respectively). While his FICA Test accuracy increased over time, those increases were not significant ((χ²(2) = 5.25, p = 0.072). His FICA accuracy scores also remained significantly lower than controls at both the subacute (z_cc = −26.437, [CI: −29.73, −23.23], p < 0.001) and early chronic (z_cc = −8.655, [CI: −9.72, −7.59], p < 0.001) phases; 0% of controls had poorer FICA Test accuracy than PWA2. PWA2’s average RTs were also longer than controls at both subacute (z_cc = −4.951, [CI: −5.57, −4.32], p < 0.001) and early chronic (z_cc = −9.396, [CI: −10.55, −8.22], p < 0.001) follow-ups.

Unlike the acute time point, PWA2 completed the DCCS Test at both follow-up visits. Unexpectedly, his DCCS accuracy significantly decreased from the 3- to 6-month time points (χ²(1) = 5.556, p = 0.018). His DCCS accuracy was significantly lower than controls at both the subacute (z_cc = −2.927, [CI: −3.31, −2.54], p = 0.002) and chronic stages (z_cc = −8.925, [CI: −10.03, −7.83], p < 0.001). PWA2’s average RTs were also longer than controls at both subacute (z_cc = −3.321, [CI: −3.76, −2.88], p < 0.001) and early chronic (z_cc = −4.031, [CI: −4.54, −3.51], p < 0.001) phases.

4.0. Discussion

In this study, we used a combined cross-sectional observational and longitudinal case series design to investigate the relationship between language and non-linguistic cognition in acute left hemisphere stroke survivors and acute to early chronic stage aphasia recovery. In the acute stroke group, we found that language and non-linguistic cognitive tests primarily loaded onto separate components and that these two components were independently associated with overall acute aphasia severity. The case study participants both presented with aphasia at the acute phase, but the individual with greater temporal damage (PWA1) than the patient with additional superior parietal and frontal damage (PWA2) exhibited more severe aphasia in all language domains. Both patients exhibited impaired attention and executive control skills at the acute stage and demonstrated recovery of language and executive control abilities over time, but a dissociation was observed. That is, by the early chronic stage, PWA1 still exhibited more severe aphasia and associated language deficits but relatively intact executive control abilities, whereas PWA2 had greater language recovery but still showed executive dysfunction. We discuss each of these findings in greater detail below.

4.1. Language and executive control deficits both contribute to acute aphasia severity

For RQ1, we elected to use PCA to derive our predictive variables because it is a data-driven method that is agnostic to preconceived notions regarding the underlying structure of the neuropsychological assessments. As such, PCA is well suited to address the question regarding the degree of inter-dependence between scores on different assessments within a sample. Similar to previous studies that implemented PCA on assessments given to stroke patients (e.g., Alyahya et al., 2020; Butler et al., 2014; Halai et al., 2017; Lacey et al., 2017; Mohapatra & Marshall, 2020; Schumacher et al., 2019; Stefaniak et al., 2022), the cognitive and language measures loaded onto separate components consistent with our expectations with one key exception: accuracy on the DCCS Test. These findings indicate DCCS accuracy scores correlated more strongly (and thus were more similar) to language measures than other executive control tests, suggesting that language processing may have been necessary to successfully complete this task.

In addition to terminology confusion, a main issue in neuropsychological assessment is that most standardized tests do not isolate executive dysfunction since they also exert demands on other (e.g., visual, motor, language) systems (Friedman & Miyake, 2017; Keil & Kaszniak, 2002; Miyake, Emerson, et al., 2000; Mohapatra & Marshall, 2020). For example, the Wisconsin Card Sorting Test (WCST; Grant & Berg, 1993) is considered a gold standard assessment of executive functions, yet it requires color and number processing, and task success relies on the ability to generate category rules based on verbal feedback and maintain rules within (most likely) verbal working memory. The degree of language demands in the DCCS Test differs substantially from the WCST (i.e., no numeric processing, abstract categorization, or rule generation), yet the target dimension in the DCCS Test was a written word (i.e., “shape” or “color”) each trial, which likely taxed the language system, particularly for participants with aphasia. This is important to keep in mind when considering the case series results, given that the only NIH-TB measure for which PWA1 still exhibited impaired performance by the subacute and chronic phases was DCCS Test accuracy.

Another central problem of measuring non-linguistic cognition in individuals with aphasia is that language is the vehicle through which task instructions are provided for assessments (de Koning, 2009; Gorelick et al., 2011; Keil & Kaszniak, 2002). While this is also true of the NIH-TB tasks, the written instructions are supplemented by visual aids (e.g., a star symbol directing the participant’s attention to the middle arrow in the FICA Test and dimension target in the DCCS Test) and clear examples, and participants complete a series of practice trials before completing each task. In the present study, lab members also read aloud written instructions and allowed patients to repeat practice trials with additional instructions as needed. On average, accuracy on the practice trials was high across the group (PCPS Test: 100%, FICA Test: 96.74%, and DCCS Test: 96.65%). As such, we can conclude that performance on the NIH-TB tasks was not an artifact of difficulty comprehending task instructions. Importantly, we can also safely assume that the NIH-TB measures that loaded onto the executive component (i.e., PCCS Test Total Items, FICA Test accuracy and average RT, DCCS Test average RT) reflect deficits distinct from impairments captured by the language component.

The key finding from this aim is that both language and executive function components were significantly related to acute aphasia severity in the left hemisphere acute stroke sample. This result accords with other studies in chronic aphasia that have reported a relationship between aphasia severity and patient performance on tests of executive control and problem-solving (Bailey et al., 1981; Baldo et al., 2005; Basso et al., 1981; Bonini & Radanovic, 2015; Lee & Pyun, 2014; Mohapatra & Marshall, 2020). Unlike previous studies, we also demonstrated that executive dysfunction explains a significant degree of aphasia severity not accounted for by language deficits alone. The tests that comprised the language principal component included measures of naming and spontaneous speech abilities, domains also measured by the WAB-R that are captured in the AQ score. Unsurprisingly, we found that the language component had a stronger relationship with AQ than the executive control component did. Yet, executive dysfunction still explained an additional nontrivial amount of variance (i.e., 17%) in acute aphasia severity, meaning that patients with poor language skills and executive dysfunction had more severe aphasia than patients with language impairments alone. Overall, the group results from RQ1 indicate that in acute left hemisphere stroke, language and executive control deficits were separable via PCA and that each component was related to overall acute aphasia severity. The case series followed up on these results by investigating whether dissociations between language and executive control were observed during stroke recovery.

4.2. Dissociable recovery of language deficits and executive control impairments is possible in post-stroke aphasia

At each time point, PWA1 presented with more impaired language than PWA2. For example, at the acute stage, PWA1 and PWA2 both demonstrated impairments in communication of salient content during picture description, word- and sentence-level auditory comprehension and repetition, and single-word oral reading, but these deficits were more pronounced in PWA1 compared to PWA2. Because of this fact, PWA1 had greater room for improvement in language skills over time than PWA2, and PWA1’s longitudinal language gains indeed exceeded PWA2’s. For both PWA, the greatest increase in language skills occurred between the acute and subacute phases with very little significant change between the 3- and 6-month post-stroke marks. This is consistent with findings from prior studies (Lazar et al., 2010; Pashek & Holland, 1988; Pedersen et al., 1995) showing the rate of spontaneous recovery is quickest within the first three months following stroke (but see Nicholas et al., 1993 regarding maximal recovery up to six months in severe aphasia and El Hachioui et al., 2011 for variable recovery timelines by language domain). Despite significant longitudinal change, by the early chronic time point, PWA1 still presented with a more severe aphasia type than PWA2 (conduction compared to PWA2’s anomic aphasia) that was characterized by anomia, frequent production of semantic and phonemic paraphasias in conversation and structured tasks (e.g., picture description, confrontation naming), repetition impairments, and impaired oral reading. In contrast, PWA2’s deficits by six months post-stroke were limited mainly to decreased communication efficiency and minor word retrieval difficulties.

Both patients’ language deficit profiles and aphasia classifications largely matched expectations given their respective lesion sizes and locations. Both PWA had temporoparietal damage, which is consistent with the fluent aphasia subtypes diagnosed for each individual and timepoint (excluding the acute phase for PWA1). PWA1 had a larger lesion volume and greater damage to the language network than PWA2. Unlike PWA2, whose lesion was restricted to posterior temporal cortex, PWA1’s lesion spanned large swathes of tissue within the anterior, mid, and posterior middle and superior temporal gyri. Many studies have demonstrated that damage to these temporal regions can result in speech recognition and auditory discrimination deficits (e.g., Fridriksson et al., 2018; Mirman, Chen, et al., 2015; Mirman, Zhang, et al., 2015; Pillay et al., 2017; Schwartz et al., 2012) and naming impairments (e.g., Alyahya et al., 2018; Fridriksson et al., 2018; Yourganov et al., 2016) as well as deficits in underlying linguistic processes such lexical-semantics (e.g., Alyahya et al., 2020; Cloutman et al., 2009; Halai et al., 2018; McKinnon et al., 2018; Meier et al., 2019; Schumacher et al., 2019; Schwartz et al., 2009; Stark et al., 2019; Walker et al., 2011), phonology (e.g., Alyahya et al., 2020; Halai et al., 2017, 2018; Pilkington et al., 2017; Schumacher et al., 2019), and syntax (e.g., den Ouden et al., 2019; Fridriksson et al., 2018; Kristinsson et al., 2020; Magnusdottir et al., 2013). Superior and middle temporal damage and dysfunction has also been associated with poorer acute to chronic stroke language recovery (Hillis et al., 2006, 2018; Meier et al., 2020). Therefore, the difference in language abilities between these two patients may be attributed to differences in the extent and location of temporal lobe damage.

In terms of executive control abilities, both PWA exhibited significantly worse performance than controls on all NIH-TB measures at the acute time point. By the subacute time point, PWA1 achieved scores that were commensurate with control performance and maintained or improved those scores by the early chronic time point. As the one exception, PWA1’s DCCS Test accuracy was still significantly lower than controls at both follow-up time points. Given that language skills may be required to successfully complete the DCCS Test (see the RQ1 results), PWA1’s persistent aphasia may have prevented him from achieving control-like accuracy on the DCCS Test even as his other NIH-TB scores improved.

PWA1’s recovery of executive control skills is in direct contrast to PWA2’s NIH-TB longitudinal scores. PWA2 demonstrated only a very slight increase in the total number of items completed on the PCPS Test from the acute to early chronic time points. There was an upward (but nonsignificant trend) in PWA2’s FICA Test accuracy scores over time, but his largest increase in accuracy from the 3- to 6-month post-stroke time points coincided with a speed tradeoff. PWA2’s DCCS Test scores are more difficult to interpret because the task was not completed at the acute stage per his request and the fact his scores worsened rather than improved from the subacute to early chronic time points. Nevertheless, unlike PWA1, PWA2 had significantly worse scores than controls on all NIH-TB measures at each time point, demonstrating that his attention and executive control deficits persisted by six months following his stroke.

Notably, only a little over 50% of the tissue within the attention mask remained for PWA2. In contrast, PWA1 had only 3% of the attention mask damaged. PWA2’s lesion overlapped only slightly with frontal lobe voxels within the attention mask but almost entirely encompassed the mask cluster within the superior parietal lobule and intraparietal sulcus. These regions are functionally connected to and work in conjunction with visual and prefrontal areas within the dorsal attention network to orient attention to relevant external stimuli (Fox et al., 2006; Humphreys & Lambon Ralph, 2015, 2017). To reiterate, PWA2 demonstrated persistent impairments not only in the more challenging NIH-TB tasks that taxed inhibition and attention shifting but also in the PCPS Test, which captures basic visual attention and proceeding speed. Processing speed is believed to underpin other cognitive skills, including executive functions, especially in older adults (Lindenberger et al., 1993; Salthouse, 1996; Sliwinski & Buschke, 1999). In Faroqi-Shah and Gehman (2021), patients with chronic aphasia demonstrated domain-general slowed processing speed (also as measured by the PCPS Test) which they found underscored slower reaction times on linguistic tests. Interestingly, although many of PWA2’s language deficits had resolved by the early chronic phase, he still exhibited reduced communication efficiency on the new Cookie Theft picture description task, but PWA1 did not. Thus, between-patient differences in parietal integrity and basic processing speed skills may explain the difference in executive control recovery observed between PWA1 and PWA2. Note that causal relationships cannot be determined from these data, so this conclusion is only speculative.

4.3. Study limitations and future directions

This study has several limitations, including aspects of our participant sample. First, it is important to acknowledge that the patient group for RQ1 constituted acute left hemisphere stroke survivors, but not all individuals had aphasia according to the WAB-R. Relatedly, while the robust regression follow-up analyses for RQ1 still revealed significant relationships between WAB-R AQ and both the language and executive components, the collective regression findings should be treated with caution given that very few patients in the sample had moderate to severe aphasia. It is possible that the RQ1 results would be different with a more severely language-impaired group, particularly given that the language component would likely have even greater explanatory power of aphasia severity in a group of all PWA and a greater number of PWA with moderate-severe aphasia specifically. One limitation of the case series was that the two case patients both had multiple comorbidities and a history of substance misuse/abuse. While both patients were living independently and had no history of pre-stroke cognitive issues per family, we are not able to definitively determine that their cognition was within normal limits prior to stroke.

Notably, the generation of orthogonal factors corresponding to language and executive control domains in the acute stroke group in RQ1 provides evidence for construct separability. However, it should be noted that loadings of the language tests onto the executive control component (and vice versa) were not zero, meaning that there was still a relationship—albeit weaker—between the language and executive control measures. Furthermore, the statistical approach used in RQ1 provides no information about the degree of relative impairment of language and executive control in individual patients nor does it answer the question of the potential interactivity of these deficits on aphasia severity. Proponents of the use of case studies in cognitive neuropsychology point out that group-level statistical tests cannot reliably demonstrate that patients within a group do or do not share specific deficits (Caramazza, 1986; McCloskey, 1993). This is particularly true for assessing longitudinal outcomes, as our patient case results demonstrate the importance of studying individual trajectories of performance across tasks.

Often, a central aim of case series studies in cognitive psychology is to demonstrate proof of a double dissociation in deficit profiles between patients, i.e., presence of impairment A but not B in one individual/patient group and the opposite in another individual/patient group (Van Orden et al., 2001). Stronger evidence of a dissociation comes from RQ2. Specifically, PWA1’s persistent language deficits and good recovery of executive control skills dissociates executive control from language, whereas PWA2’s persistent executive dysfunction and good recovery of language skills dissociates language from executive control. While these results suggest language and executive control are dissociable to a certain degree, there are two main limitations of our study (and possibly the scope of the question we aimed to answer) that preclude the claim of a true double dissociation in our case series.

First, while we assessed language and executive control using a variety of measures, there are linguistic and cognitive domains that we did not assess. Specifically, the language assessment battery included several measures of naming and semantic abilities but few to no syntactic or phonological processing tests. Similarly, the included NIH-TB tasks were limited to assessments of visual attention, inhibitory control, and set-shifting and thus did not capture other non-linguistic cognitive abilities (e.g., working memory) that contribute to executive control. For RQ1, it is possible that the factor loadings and/or predictive utility of the principle component variables would be altered with the inclusion of other assessments. For RQ2, the addition of phonological tests in particular would have formed a more complete picture of each patient’s language recovery. Future studies that aim to test acute to chronic stroke relationships between language and cognition more broadly should include additional assessments that cover the breadth of linguistic and cognitive processes that underlie the constructs of language and executive control (e.g., see Stefaniak et al., 2022 employ such an approach to study subacute to chronic aphasia recovery).

The second limitation lies in drawbacks of the lesion approach. We elected to use Neurosynth to create our masks instead of publicly available atlases because Neurosynth allows users to generate functional activation maps based on meta-analytic (rather than single study) data from specific search terms of interest. In Neurosynth, users have the option of selecting two different types of functional map outputs: one based on an “association test” that includes voxels preferentially related to a given search term and another based on a “uniformity test” that includes voxels consistently activated across included studies. There is greater certainty with an association test that the voxels retained in the map are specific to a given search term (e.g., attention, language), but the final maps have fewer voxels than the uniformity test maps. Consequently, the executive mask used in this study was small (i.e., less than 10% the size of the language mask), which may be one reason why both case patients had minimal executive network damage. However, it is important to note that the executive mask was slightly larger than the attention mask, and the size of the masks do not discount the possibility that a small, focal area of damage (e.g., PWA2’s lesion within the middle frontal gyrus that overlapped with the executive mask) could have a large impact on network dysfunction and corresponding cognitive abilities. Indeed, with extensive damage to a large mask, it is impossible to determine within an individual patient the specific area of damage that causes specific deficits. Therefore, future studies should incorporate lesion network localization methods (e.g., lesion seeding of resting state data, see Boes et al., 2015) that more clearly demonstrate the functional impact of an individual patient’s brain damage. It is also important to note that while the two case patients differed in the degree of damage to the language and attention networks, both individuals had some degree of damage to each mask. As such, a complete dissociation in terms of structural damage and behavioral performance cannot be made.

Moreover, an arguably greater disadvantage of lesioning pre-existing masks (either derived from atlases or repositories like Neurosynth) is that it is impossible to determine which damaged regions are key hubs within individual-specific functional networks. The use of functional localizer fMRI experiments in healthy adults shows that there is spatial similarity between individuals in terms of regions activated by language and domain-general cognitive tasks but that the specific voxels that are recruited within such regions vary between people (see e.g., Fedorenko et al., 2012). Without the use of fMRI, we cannot say for certain whether the damaged voxels within our network masks represent tissue necessary for mediating language and executive control specifically for our two case patients. Future studies that combine lesion methods, fMRI functional localizers, and extensive neuropsychological assessment stand the best chance of providing a true double dissociation between language and executive control in PWA.

The strongest evidence from such an approach would come from longitudinal study of the same individuals, from acute to chronic post-stroke stages. Recently, Stefaniak et al. (2022) investigated relationships between functional brain activation and speech/fluency, semantic/executive, and phonology principal components (derived from a neuropsychological battery) in a group of 26 patients with mild-moderate aphasia at approximately two weeks and again at four months post-stroke. They reported that functional brain-behavior correlates and relationships were mostly distinct for each component, suggesting a dissociation between these behavioral constructs. Stefaniak et al. (2022) provide an elegant example of how functional imaging and in-depth neuropsychological data can be combined to study longitudinal aphasia recovery, but it should be noted that individual recovery patterns were not delineated, and localizer tasks and an acute stroke time point were not incorporated into this study. Inclusion of acute stroke data is crucial to fully describe aphasia recovery trajectories. Acute stroke lesion and functional correlates may most closely align with the spatial distribution of undamaged language and cognitive networks in healthy people because this time window occurs before the onset of neural reorganization of structure-function relationships (Cramer, 2008; Hartwigsen & Saur, 2019; Hillis & Heidler, 2002; Kiran, 2012; Turkeltaub, 2019). In contrast, the triangulation of lesion, fMRI, and neuropsychological data in chronic stroke provides insight into neural reorganization of these networks for recovery. This is important from the standpoint of understanding neural redundancy in the face of network damage and determining the potential interactivity of functional networks (e.g., recruitment of domain-general regions to bootstrap language processing) in chronic aphasia.

5.0. Conclusions

This study demonstrates that language and non-linguistic executive control dissociate in their contributions to acute aphasia severity and in longitudinal aphasia recovery during the first six months after left hemisphere stroke. The challenges and barriers in assessing non-linguistic cognition in aphasia can be overcome, even in acute stroke. Certain assessments (e.g., PCPS and FICA tests) require little to no language processing and likely provide purer measures of non-linguistic attention and executive control skills than others (e.g., DCCS Test). Anterior to posterior temporal lobe damage may have resulted in PWA1’s greater language impairments (compared to PWA2), and superior parietal/intraparietal sulcus damage may have resulted in PWA2’s greater impairments executive control (compared to PWA1). Despite these findings, we cannot conclude that language and executive control are or are not distinct cognitive modules (Van Orden et al., 2001). Instead, it is likely impairments in each domain can co-exist to varying degrees within an individual patient given the spatial proximity of language and domain-general network voxels within the brain and the likelihood of damage to both types of networks. As such, the inter-dependence on language and other cognitive processes remains unknown, but future work that incorporates extensive neuropsychological assessment, lesioned network localization, and functional activation localizers in other aphasia case series is well-suited to addressing this outstanding question.