Causes and consequences of unawareness (anosognosia) of tool-action errors after left-hemisphere stroke

Abstract

Many individuals with limb apraxia after left-hemisphere stroke exhibit a lack of awareness of their tool-related action errors, i.e., unawareness of apraxia (UA; also called anosognosia of apraxia). Little is known about the prevalence of UA, the relationship between UA and apraxia severity, or its underlying mechanisms. Here, we assessed both the causes and consequences of UA. Based on a mechanistic model, we hypothesized that UA may arise because of deficits in representations signaling how tool-related movements should look and feel—a component of action knowledge—and that degradation of this knowledge impedes the detection of mismatches between planned and actual tool-related actions. We further predicted that a consequence of UA is a reduction in error-correction attempts. Fifty-six individuals with chronic LCVA gestured to show how to use tools. Immediately after the gesture production task, participants were asked if they made any errors. All participants also completed an action knowledge task to measure the integrity of tool-related movement goals. Individuals were denoted as exhibiting UA if they performed below a normative cutoff for apraxia yet reported making no errors. Our sample included 21 individuals with apraxia; of these, nearly half (48%) exhibited UA. These two groups made a comparable number of gesture errors and were of equivalent stroke severity, yet individuals with UA had significantly more impaired action knowledge. Additionally, individuals with UA were less likely to attempt to correct their errors compared to individuals who were aware of their apraxia. These data support the hypothesis that action knowledge (how tool actions look and feel) serves a key role in error detection and awareness of apraxia and may contribute to the difficulties with everyday tasks experienced by many people with apraxia.

Introduction

Limb apraxia is a neurological condition that commonly occurs after a left-hemisphere cerebral vascular accident (LCVA) and is characterized by errors in pantomimed tool-use gestures and actual tool use in the absence of elementary sensorimotor deficits^1–7. Some apraxic individuals exhibit a remarkable phenomenon characterized by unawareness of their praxis deficits (i.e., unawareness of apraxia – UA), a condition also known as anosognosia of apraxia^8–10. For example, an individual with UA might demonstrate how to use a pair of scissors with a flagrantly incorrect hand posture and movement trajectory but maintain that the gesture was produced correctly when queried about its quality. Deficient awareness of deficit (i.e., anosognosia) has been identified in several other post-stroke syndromes including aphasia, hemiplegia, and spatial neglect, and is linked to poor rehabilitation outcomes and increased caregiver dependence¹¹. However, there are very few studies of UA, and little is known about its prevalence or underlying mechanisms. Given the potential impact of UA on functional independence, the limited research on this disorder represents a critical gap.

In this study, we tested two hypotheses about the causes and consequences of UA that were based on a novel mechanistic model of error awareness (Fig. 1). In this model, an aspect of action knowledge (i.e., stored information about how an action looks and feels¹²) serves as a goal for the specification of motor commands sent to the body^2,13–15 (Fig. 1 - yellow box). Sensory feedback about the movement of the body is sent back to the brain (Fig. 1 - pink box), enabling a comparison between the actual and desired movement. This comparison provides an error signal that facilitates performance awareness; if there is a mismatch between the goal and the produced movement, motor commands are generated to correct the error^16–18 (Fig. 1 – purple box). Based on this model, our first hypothesis was that reduced action knowledge (and hence an impaired motor goal) gives rise to deficient performance awareness (UA).

Figure 1:

Mechanistic model of performance awareness. Action knowledge serves as a movement goal defining how the movement should look and feel (yellow box). This goal informs motor commands sent to the body. In return, the brain receives visual and proprioceptive feedback that details how the body moved (pink box). This sensory feedback is compared with the desired movement goal, providing an error signal used for performance awareness (purple box). We hypothesize that the movement goal is poorly specified in UA, resulting in a degraded error signal that reduces subsequent error corrections (i.e., via the controller) and increases unawareness of errors.

Our second hypothesis addresses the consequences of UA on the ability to correct one’s errors. Error correction attempts are commonly viewed as a measure of the integrity of performance monitoring and awareness in both language and motor domains^19,20. In a prior study, we examined error corrections in a tool-use gesture task in individuals with LCVA¹³. We found a positive linear relation between error correction attempts and action knowledge as assessed by a task in which people had to recognize the correct use gesture associated with a given tool^4,13,21. These data were consistent with the hypothesis that error correction attempts depend upon an intact movement goal. However, UA was not assessed in that study, leaving unclear whether the observed relationship between reduced action knowledge and reduced error correction attempts has any relevance for the clinical UA syndrome. Accordingly, our hypothesis was that we would observe a specific relationship between action knowledge, error correction attempts, and UA that would not be attributable to the overall severity of neurologic impairment or apraxia.

Materials and Methods

Participants

Sixty-six chronic left hemisphere stroke (LCVAs) survivors were recruited from the Research Registry of the Jefferson Moss Rehabilitation Research Institute²². All participants experienced a stroke more than 6 months prior to participation, and did not have a history of psychiatric diagnosis, other neurologic disorders, traumatic brain injury, or drug/alcohol abuse. All participants were right-handed prior to stroke onset. To ensure that participants did not have major language comprehension impairments, they were required to achieve a score of 4 or higher on the comprehension subtest of the Western Aphasia Battery²³ (WAB). Additionally, a measure of stroke severity was acquired for each participant via the National Institutes of Health Stroke Scale²⁴ (NIHSS). Ten of the 66 participants were excluded because they did not have a lesion limited to the left hemisphere (n = 4), were left-handed (n = 3), experienced significant memory or comprehension issues preventing task completion (n = 2) or did not complete all the relevant tasks (n = 1). A sample of 16 right-handed neurotypical adults who were matched in age, education and sex was also included (Table 1). The neurotypical participants completed a telephone version of the Montreal Cognitive Assessment^25,26 (MoCA) to confirm that their cognitive functioning was within normal limits (score equal to or above 18). Informed consent was obtained for all participants in accordance with the policies of the Institutional Review Board of Thomas Jefferson University. Participants were compensated for their participation.

Table 1:

Demographics for the tested sample. To ensure that the groups did not differ significantly in age and education, two-sample t tests were conducted. A chi-squared test was conducted to test whether the proportion of females and males was statistically different across groups. Demographics are reported in the form mean ± standard deviation.

	LCVA (N= 56)	Neurotypical Controls (N=16)	Statistics
Age (years)	63.77 ± 10.56	64.69 ± 9.95	t(70) = 0.31; p = 0.76
Education (years)	14.50 ± 2.69	15.88 ± 2.31	t(70) = 1.86; p = 0.07
Sex	24 females	8 females	χ2(1) = 0.26; p = 0.61
	32 males	8 males
Stroke onset (months)	115.08 ± 79.74
Phone MoCA		20.31 ± 1.20

Tasks

Gesture-to-Sight (GTS) of Objects Task

Thirty-nine colored photographs of common manipulable objects with a distinct use action were included in the GTS task. These objects included carpentry tools (e.g., hammer), household articles (e.g., clothes iron), office supplies (e.g., scissors), and grooming items (e.g., razor). Participants were seated approximately one meter from a 22-inch computer monitor and instructed to gesture the use of a target object with their ipsilesional left hand. Each object photograph was presented once, in random order, over the course of two blocks. After the object was shown for three seconds, a beep signaled participants to start the gesture. Gestures were recorded by digital camera and scored offline by one of two trained coders who were reliable with each other (> 85% agreement by Cohen’s Kappa²⁷). Each trial was first scored for the presence of a content error (i.e., zero for error, one otherwise) if another recognizable gesture was substituted for the target gesture (e.g., brushing teeth gesture when the object was a comb). Items receiving a content score of one were subsequently scored on three spatiotemporal dimensions: hand posture, arm posture, and amplitude/timing. Following detailed praxis scoring guidelines long in use in our laboratory²¹, each dimension was scored separately and a score of zero was assigned when an error occurred (one otherwise). For each trial the best attempt was scored. That is, if participants first produced an error but then adequately corrected the error, they received a score of one. Further coding guidelines were developed to specifically identify any trials with error correction attempts. These trials included a small percentage (8.73 %) of trials with successful error correction attempts (i.e., trials on which participants ultimately received the maximum 3-point score). Trials in which the participant produced a content error (0.32 % of trials) or stated that they did not recognize the object (1.10 % of trials) were removed from the analysis. For the remaining trials, the sum of the three spatiotemporal gesture components (hand posture, arm posture, and amplitude/timing) were averaged for each item, resulting in scores ranging from 0 to 1.

Apraxia error awareness question

We modified an existing anosognosia questionnaire for aphasia²⁸ to assess whether participants were aware of their errors. Immediately after the gesture task, they were asked: “During the gesture task, did you ever feel that you knew the purpose or reason for using the object, but found it difficult to show how to use it?” This wording focuses on any gesture production difficulties participants may have experienced despite good object recognition. Participants’ yes or no responses, along with their GTS score, were used to determine whether a participant with apraxia experienced UA (see Analyses section for further details).

Gesture Recognition (GR) Task

The GR task is a verbal action-gesture matching task long in use in our lab^21,29. Each trial began with a target verb phrase (e.g., “tightening a bolt”) presented on a computer screen along with the simultaneous playing of an audio recording of the phrase. Two short (3-5 second) videos were then played. In one video, an actor was shown gesturing the target action. In the other video, which served as a foil, a spatiotemporal error (29 trials) or semantic error (27 trials) was depicted. Spatiotemporal errors were comprised of target actions with an error in one of the same three dimensions (hand posture, arm posture, or amplitude/timing) used to score performance in the GTS task. Semantic error videos depicted a substitution of an action that was semantically related to the target action. For example, for the target verb phrase “cutting paper”, the foil video depicted a “slicing with a knife” gesture. The order of the correct and foil videos was randomized. Stimuli were presented via a program developed on PsychoPy (Open Science Tools Ltd, Nottingham, UK).

Prior to administration of the GR task, we administered a pretest to ensure that individuals could comprehend the verb phrases used in the GR task (e.g., “tightening a bolt”) and knew which tool was associated with that phrase. Participants viewed verb phrases presented singly on a 22-inch computer monitor while an audio recording of the verb phrase was played. Participants then viewed three pictures of numbered tools: the correct tool (e.g., the tool associated with tightening a bolt, i.e., a wrench) and two semantic foils (e.g., a screwdriver and a hammer). Participants were instructed to select, using the number pad on the keyboard, the tool that best matched the verb phrase. There were 34 different test items presented with Eprime 2 (Psychology Software Tools, Pittsburgh, USA). Any verb phrases that a participant failed in the pretest were removed from scoring in the GR task (5.66 % of trials).

NIH Stroke Scale

As a control for overall stroke severity, we administered the NIH Stroke Scale²⁴, consisting of 11 items assessing language, consciousness, inattention, vision, motor function, and sensation (Max. score of 42; higher scores indicate greater impairment).

Data Analyses

Identifying individuals with apraxia and UA

LCVA participants were subdivided into apraxic and non-apraxic groups based on a cutoff score calculated from the average GTS scores across all trials of the neurotypical participants. LCVA individuals scoring less than .85 (i.e., more than two standard deviations below the mean of the neurotypicals) were classified as apraxic. People with apraxia were further subdivided according to their responses to the apraxia awareness question. If participants with apraxia denied that they had issues showing how to use tools in the GTS task, they were classified as having UA. The LCVA sample was thus subdivided into three groups: non-apraxic individuals (Non-Apraxic), individuals experiencing apraxia with awareness of their deficit (Apraxic-Aware), and individuals experiencing unawareness of apraxia (UA).

Statistical analyses

To test the study predictions, we analyzed GTS and GR accuracy, as well as the proportion of trials with error correction attempts in the GTS task. We also considered whether UA was related to apraxia severity and/or overall stroke severity using GTS accuracy and the NIHSS score, respectively.

To assess whether the Apraxic-Aware and UA groups differed in GTS accuracy, we used a Linear Mixed Model with Group as a fixed effect. By definition, these groups had lower GTS accuracy than the Non-Apraxic group, so the Non-Apraxic group was not included in this analysis. Participant was included as a random effect. NIHSS data were analyzed with a one-way ANOVA with group as a fixed effect (Non-Apraxic vs. Apraxic-Aware vs. UA). GR accuracy and the proportion of trials with error corrections were analyzed in separate General Linear Mixed Model models with a binomial family. For GR accuracy, we included Group (Non-Apraxic vs. Apraxic-Aware vs. UA) as a fixed effect and Participant as a random effect. For the proportion of trials with error correction attempts, we use a similar model but only included the Apraxic-Aware and UA groups (as these groups produced sufficient errors for analysis). Where relevant, Tukey post-hoc tests were conducted to compare pairwise differences between groups. Finally, to test the prediction that action knowledge and error correction attempts would be positively related, we ran a one-sided Pearson’s correlation between total GR accuracy and the proportion of GTS trials with error correction attempts. All analyses were conducted in R (R Core Team).

Results

UA is common in individuals with apraxia

Of the 56 LCVA individuals tested, 21 had limb apraxia and 35 were Non-Apraxic. Among the individuals with apraxia, there were 10 individuals with UA and 11 Apraxic-Aware individuals. Thus, in our sample 47.62% of individuals with apraxia experienced UA. The proportion of individuals with UA is broadly consistent with a previous study in which 50% of individuals with limb and/or bucco-facial apraxia showed reduced awareness of their deficit¹⁰.

Apraxia severity and overall stroke severity are not associated with UA.

We first tested whether UA could simply be explained by the severity of apraxia or overall cognitive and sensory-motor deficits related to stroke. There was no difference in the performance of the Apraxic-Aware versus UA groups on the GTS task (χ²(1) = 0.33, p = 0.57; Fig. 2A). For the NIHSS, although we found an effect of group (F_{(2, 53)} = 7.21, p = 0.002; Fig. 2B), post-hoc tests revealed that this was not attributable to differences between UA individuals and either the Apraxic-Aware (p = 0.81) or Non-Apraxic (p = 0.08) groups. Rather, Apraxic-Aware individuals had significantly greater overall stroke impairment relative to Non-Apraxic individuals (p = 0.003). Together, these findings indicate that UA is explained neither by apraxia severity nor overall stroke severity.

Figure 2: Individuals with UA showed equivalent performance to other groups in apraxia and stroke severity.

A) Plot shows the proportion correct trials on the Gesture-to-Sight (GTS) task assessing apraxia severity. No difference in GTS accuracy was observed between Apraxic-Aware [0.76 ± 0.02] and UA [0.77 ± 0.02] groups. B) Plot shows stroke severity (NIH Stroke Scale) between the three groups [Non-apraxic = 2.94 ± 0.36 ; Apraxic-Aware = 6.27 ± 1.10; UA = 5.10 ± 1.19]. Each colored dot represents a unique participant. The group average is depicted by black dots, and error bars represent the group standard error of the mean (SEM). Significance level is indicated with a star: ★★ p < 0.01.

Individuals with UA have a selective impairment in action knowledge and are less likely to correct errors

To test the hypothesis that UA is caused by deficient action knowledge, we analyzed Gesture Recognition (GR) accuracy across all three groups. We found a group effect (χ²₍₂₎ = 35.02, p < 0.001); post-hoc tests revealed an impairment in action knowledge for people with UA when compared to both the Apraxic-Aware (p < 0.001) and Non-Apraxic groups (p < 0.001; Fig. 3A). No statistical difference was observed between the Non-Apraxic and Apraxic-Aware groups (p = 0.15). Viewed together with the results reported above, this indicates that individuals with UA have a specific deficit in action knowledge.

Figure 3: Individuals with UA had more deficient action knowledge and fewer error correction attempts than individuals in the Apraxic-Aware group.

A) Individuals with UA showed a selective impairment in gesture recognition (GR) when compared to the other two groups [Non-apraxics = 0.86 ± 0.01; Apraxic-Aware = 0.83 ± 0.02; UA = 0.69 ± 0.02]. B) Proportionally fewer error trials were associated with correction attempts in UA than Apraxic-Aware individuals. [Apraxic-Aware = 0.32 ± 0.05; UA = 0.09 ± 0.03]. The group average is depicted by black dots, and error bars represent the group standard error of the mean (SEM). Significance levels are indicated with stars: ★★★ p < 0.001. C) The correlation between GR accuracy and the proportion of trials with error corrections was significantly positive. Each colored dot represents a unique participant.

Our second prediction was that UA would impact individuals’ abilities to detect and correct their own errors. We found that the proportion of error trials that contained an error correction attempt was significantly smaller for individuals with UA compared to Apraxic-Aware individuals (χ²₍₁₎ = 13.71, p < 0.001; Fig. 3B). Moreover, we found that action knowledge (GR accuracy) was positively correlated with the proportion of error trials with error correction attempts (ρ = 0.60; p = 0.002; Fig. 3C). Thus, individuals with poor action knowledge are less likely to correct their errors.

Discussion

Limb apraxia is a common disorder, occurring in 25%-50% of individuals following a stroke in the left hemisphere^30,31 and is associated with poor post-stroke recovery³². A significant proportion of these individuals (nearly 50% in our study) are not aware of their impairments, suggesting that UA may be a major contributing factor to disability. Why UA arises and how it impacts tool-use ability has not been well-understood. This study tested two predictions derived from a mechanistic model of performance awareness (Fig. 1) in which action knowledge serves as a movement goal that specifies how an action should look and feel^2,13–15. According to this model, imprecise and/or slowly-activated action knowledge in individuals with UA disrupts the ability to compare movement goals against executed movements, with resulting imprecision or “noise” in an error “mismatch” signal that enables error detection and correction. Thus, we hypothesized that UA should be specifically associated with both deficient action knowledge and reduced error correction attempts. The current study confirmed these predictions and additionally suggested a reliable relationship between the magnitude of impairment in action knowledge (a cause of UA) and the magnitude of reduction in error correction attempts (a consequence of UA).

It is noteworthy that we found no relationship between UA and apraxia severity, consistent with the results of a prior study¹⁰. This suggests that UA does not simply reflect an overall degradation of the praxis system. Instead, we showed that UA was specifically related to a deficit in action knowledge (i.e., stored movement goals). Note that not all apraxic impairments arise from action knowledge deficits; they may also arise from a downstream disruption in transforming movement goals into a motor command to be executed^2,3,33–38. Whereas gesture recognition tasks probe the former, tests of gesture imitation are frequently used to probe the latter. This is because imitation does not necessitate accessing stored action knowledge; instead, visual input (i.e., the gesture to be imitated) can be directly transformed into a representation that guides the desired movement³⁹. Indeed, a prior study using action imitation to probe UA at this more downstream point along the praxis pathway found an association between apraxia severity and the ability to detect self-produced (but not other-produced) errors⁴⁰. Taken together with the present results, these data suggest that different factors may contribute to unawareness of tool-use production errors (where a stored movement goal is required) versus imitation errors (where one’s own actions must be matched to a recently-performed action performed by another person).

Consistent with this possibility, our model suggests several possible loci for unawareness of praxis performance errors. First, impairments in sensory feedback processing (Fig. 1 – pink box), may result in deficient awareness that generated movements deviate from movement goals. We recently demonstrated that individuals with apraxia exhibit a specific reduction in the ability to use proprioceptive information to guide movements¹⁴. Accordingly, it is possible that UA in some individuals may be attributable to deficits in using proprioceptive feedback to detect their own errors. Alternatively, performance unawareness may arise from an impaired ability to compare intact movement goals to generated movements, as has been hypothesized for anosognosia of hemiplegia^41–43. Both the proprioceptive feedback and “comparator” accounts could explain the disparity between impaired detection of one’s own errors and intact detection of other errors that was observed by Scandola and colleagues (2021) in an imitation task. In summary, error monitoring failures may plausibly arise for at least three reasons: failure to retrieve a movement goal, deficits in proprioceptive feedback integration, or deficient comparison of intact goals with feedback. Furthermore, the locus of failure may depend in part on the input and output requirements of the task. It will be of interest to pursue these distinctions in future research.

Error correction attempts by apraxic-aware individuals suggest that they were able to detect the discrepancy between the expected and actual movement after the movement was produced, although they were unable to prevent the error from occurring in the first place¹³. In other domains (e.g., speech, typing), error correction attempts are commonly viewed as a key index of the integrity of performance monitoring and a window into the processes that are used for error detection⁴⁴. In picture naming, correction attempts consisting of successive approximations of the target word (“conduite d’approche”) are a hallmark of conduction aphasia, a language disorder characterized as a failure to transmit intact knowledge of phonemes/syllables/words to an intact production planning system in a timely way to prevent errors. In this disorder, errors are detected by the auditory comprehension system as a mismatch between the way the phoneme/syllable/word is supposed to sound and auditory feedback about what was actually produced. These detected errors in turn prompt another attempt at producing the correct utterance: approximations to the target word are thus viewed as “clean-up” attempts by the intact comprehension system^45–47. Indeed, in several language models⁴⁸, intact knowledge of the movement goal (how an utterance should sound) is a prerequisite for error correction^49,50. We propose that apraxic-aware individuals leverage their intact action knowledge to employ a similar correction strategy based on the desired look and feel of tool actions.

As noted earlier, many classic and contemporary accounts of apraxia posit two major subtypes: one characterized by deficits in action knowledge, and the other by a disconnection between intact knowledge and motor planning^2,3,33–38. The former subtype, typically referred to as ideational or conceptual apraxia, is described as an inability to conceptualize a proper action plan, i.e., single object-related actions⁵¹ or a sequence of complex multistep actions⁵². Our view is that ideational apraxia is most clearly identified by poor performance on gesture recognition or other concept-focused tasks that do not have a gestural output requirement, such as those tapping manipulation knowledge⁵³. In contrast, ideomotor apraxia has been linked to underlying deficits in processes further “downstream”, such as the integration of sensory information to form a motor plan¹⁴ or selecting among action competitors^54–56. In its purest form, ideomotor apraxia is associated with spatiotemporal errors in production in the context of intact gesture recognition⁵⁷. Nevertheless, differential diagnosis is often difficult: gesture errors are produced in both subtypes for different reasons, and action recognition and production deficits often co-occur to varying degrees². Given the strong relationship we observed between gesture recognition and the ability to detect one’s own errors, it may be clinically useful when attempting differential diagnoses to consider poor error detection and UA as hallmarks of the ideational apraxia syndrome, whereas apraxia with error awareness may be more consistent with a diagnosis of ideomotor apraxia.

The present study had several limitations. Our UA assessment was very brief, based on a prior questionnaire developed for assessment of unawareness of aphasia²⁸. There are also existing anosognosia questionnaires for hemiplegia⁵⁸, spatial neglect^59,60, and everyday complex actions^8,9, but we are not aware of a specific assessment tool for gestural praxis actions. These existing assessments may serve as templates for a more thorough UA questionnaire. Another limitation of the study is that we inquired about participants’ views of their own performance only after the entire gesture task was complete, rather than after each individual trial. The temporal dynamics of perception of performance may be an interesting issue for future research. For example, it may be possible to immediately judge one’s performance on a given trial as impaired but be unable to integrate this into a long-term belief that one has problems with the task as a whole^61–63. Future research should explore the relationship of trial-level and task-level assessments. Finally, from both theoretical and clinical perspectives it will be of interest to further examine the relationship of UA to anosognosia in other domains (e.g., hemiparesis, aphasia, spatial neglect) as well as to further explore the hypothesis that UA may arise because of deficits at several different points in processing (e.g., sensory integration or comparison), as noted earlier.

These results may also be informative from a clinical perspective. Treatments for limb apraxia often focus on practicing gesture production⁶⁴. The present findings suggest that rehabilitation approaches may benefit from an additional focus on restoring action knowledge in individuals with UA. Indeed, prior research indicates that remediation of action knowledge may also benefit gesture production accuracy⁶⁵. Assessing UA in rehabilitation settings may facilitate the provisioning of interventions targeted at individuals’ reduced self-awareness and help to improve caregivers’ awareness of this challenge. Taken together, our findings of the relationship of UA to impaired action knowledge and reduced error correction highlight the importance of systematically assessing UA when testing for the presence of limb apraxia.

Data Availability

Relevant data and scripts are available athttps://osf.io/cq5vg.