Authors: Christine Hazelton, Katie Thomson, Alex Todhunter-Brown, Pauline Campbell, Charlie SY Chung, Liam Dorris, David C Gillespie, Susan M Hunter, Kris McGill, Donald J Nicolson, Linda J Williams, Marian C Brady
Categories: Activities of Daily Living, Adult, Humans, Perceptual Disorders, Perceptual Disorders/etiology, Perceptual Disorders/rehabilitation, Randomized Controlled Trials as Topic, Stroke, Stroke Rehabilitation, Stroke/complications, Vision Disorders, Vision Disorders/rehabilitation, Medicine General & Introductory Medical Sciences
Source: The Cochrane Database of Systematic Reviews
Interventions for perceptual disorders following stroke
Authors: Christine Hazelton, Katie Thomson, Alex Todhunter-Brown, Pauline Campbell, Charlie SY Chung, Liam Dorris, David C Gillespie, Susan M Hunter, Kris McGill, Donald J Nicolson, Linda J Williams, Marian C Brady
Abstract
Background
Perception is the ability to understand information from our senses. It allows us to experience and meaningfully interact with our environment. A stroke may impair perception in up to 70% of stroke survivors, leading to distress, increased dependence on others, and poorer quality of life. Interventions to address perceptual disorders may include assessment and screening, rehabilitation, non‐invasive brain stimulation, pharmacological and surgical approaches.
Objectives
To assess the effectiveness of interventions aimed at perceptual disorders after stroke compared to no intervention or control (placebo, standard care, attention control), on measures of performance in activities of daily living.
Search methods
We searched the trials registers of the Cochrane Stroke Group, CENTRAL, MEDLINE, Embase, and three other databases to August 2021. We also searched trials and research registers, reference lists of studies, handsearched journals, and contacted authors.
Selection criteria
We included randomised controlled trials (RCTs) of adult stroke survivors with perceptual disorders. We defined perception as the specific mental functions of recognising and interpreting sensory stimuli and included hearing, taste, touch, smell, somatosensation, and vision. Our definition of perception excluded visual field deficits, neglect/inattention, and pain.
Data collection and analysis
One review author assessed titles, with two review authors independently screening abstracts and full‐text articles for eligibility. One review author extracted, appraised, and entered data, which were checked by a second author. We assessed risk of bias (ROB) using the ROB‐1 tool, and quality of evidence using GRADE.
A stakeholder group, comprising stroke survivors, carers, and healthcare professionals, was involved in this review update.
Main results
We identified 18 eligible RCTs involving 541 participants. The trials addressed touch (three trials, 70 participants), somatosensory (seven trials, 196 participants) and visual perception disorders (seven trials, 225 participants), with one (50 participants) exploring mixed touch‐somatosensory disorders. None addressed stroke‐related hearing, taste, or smell perception disorders. All but one examined the effectiveness of rehabilitation interventions; the exception evaluated non‐invasive brain stimulation. For our main comparison of active intervention versus no treatment or control, one trial reported our primary outcome of performance in activities of daily living (ADL):
Somatosensory disorders: one trial (24 participants) compared an intervention with a control intervention and reported an ADL measure.
**Touch perception ** no trials measuring ADL compared an intervention with no treatment or with a control intervention.
Visual perception disorders: no trials measuring ADL compared an intervention with no treatment or control.
In addition, six trials reported ADL outcomes in a comparison of active intervention versus active intervention, relating to somatosensation (three trials), touch (one trial) and vision (two trials).
Authors' conclusions
Following a detailed, systematic search, we identified limited RCT evidence of the effectiveness of interventions for perceptual disorders following stroke. There is insufficient evidence to support or refute the suggestion that perceptual interventions are effective. More high‐quality trials of interventions for perceptual disorders in stroke are needed. They should recruit sufficient participant numbers, include a 'usual care' comparison, and measure longer‐term functional outcomes, at time points beyond the initial intervention period. People with impaired perception following a stroke should continue to receive neurorehabilitation according to clinical guidelines.
Summary of findings
Background
Description of the condition
Stroke is the second most frequent cause of mortality (Katan 2018), and the second leading cause of disability‐adjusted life‐years worldwide (Lanas 2021). In 2019, over 12 million people experienced a stroke (GBD 2019 Stroke Collaborators), with over 100,000 occurring in the United Kingdom (UK) each year (Royal College of Physicians 2021; Public Health Scotland 2021). Stroke can impact a person's ability to process sensory information (Mercier 2001). It is estimated that one in five stroke survivors experience a perceptual disorder (Rowe 2009).
Perception is the ability of the brain to interpret and integrate information received by the senses. It can involve multiple steps in the processing of sensory information, including understanding the information received, organising it, and assigning meaning (Lezak 2012). This includes hearing (auditory), smell (olfactory), somatosensation, touch (tactile), taste (gustatory), and visual systems. We present our definition of perception and details of the senses and disorders included in this review in the Types of participants section.
Perceptual disorders reduce an individual’s ability to understand their relationship to the environment and, in turn, to respond appropriately. Hearing perception disorders can affect listening and communication ability, impacting on participation in rehabilitation (Koohi 2017a). Inability to smell can result in safety concerns (e.g. detecting smoke, gas, or stale food (Wehling 2015)), whilst altered perception of touch can place affected limbs at risk of injury (such as burns) and negatively impact independence in everyday activities (Doyle 2010). Disordered perceptions of touch and somatosensation impair the control of posture and movement; lead to poorer performance of motor tasks; contribute to slower and/or poorer recovery of motor function; and negatively impact independence in everyday activities (Abe 2012; Carey 2011b; Doyle 2010; Tyson 2008). Taste dysfunction can lead to impaired appetite due to unpleasantness of eating, potentially contributing to malnutrition, weight loss, and increased risk of depressive symptoms (Dutta 2013). Studies have shown that visual perceptual disorders are associated with reduced ability in activities of daily living (Prince 2017), greater disability, reduced quality of life (Ali 2013), and are a predictor of poor self‐care (Bernsprang 1987).
Despite the adverse impact of perceptual disorders on long‐term quality of life, they are frequently undetected by healthcare professionals (Dutta 2013). Current guidelines include limited information on perception, with little guidance on how to select or implement interventions (NICE 2013; Royal College of Physicians 2016).
Description of the intervention
A range of interventions to address perceptual disorders currently exists. Interventions often involve a rehabilitation focus, aiming to restore, compensate, or substitute for a loss of function. They vary, depending on the sense affected and the specific nature of the dysfunction. Rehabilitation interventions can include those led by healthcare professionals (including physiotherapists and occupational therapists), such as functional training in everyday tasks for visual perceptual disorders (Edmans 2000), or exercises for vertical training and postural control for somatosensory disorders (An 2020). Technology‐based interventions can include robotic devices for gait training (Yun 2018), computer training with neurofeedback to test changes in visual perception (Cho 2015), or specialist hearing devices (Koohi 2017b). A range of specialist equipment is available and includes touch discrimination blocks for touch perceptual disorders (Carey 1993), and a suspension device for posture correction (Jahn 2017). In addition, non‐invasive brain stimulation (Koo 2018), such as transcranial direct current stimulation, has been described in the treatment of somatosensory or touch disorders. Pharmacological interventions for auditory or visual hallucinations have also been reported (Chen 2011; Fifer 1993).
How the intervention might work
Rehabilitation interventions
Rehabilitation interventions are "designed to optimize functioning and reduce disability in individuals with health conditions in interaction with their environment" (World Health Organization 2011). They can be categorised as restorative, compensatory, or substitutive in approach (Kerkhoff 2000). Restorative interventions involve direct training of the impaired function with the aim of recovering this function. Strategies can include repetition of exercises or games to achieve postural control (Choi 2018), training with different materials to improve texture discrimination (Carey 1993), or practice of (visual) perceptual tasks (e.g. card sequencing or figure copying (Edmans 1991)). In each case, the aim is that skills gained in therapy contexts will transfer to everyday activities.
Compensatory approaches to rehabilitation focus on using a spared function to overcome limitations. Strategies might include route training to compensate for impaired visual perception (Brunsdon 2007), and learning emotional strategies to assist in the development of coping skills for perceptual impairments (McDowell 2019). Substitution approaches may involve the use of external devices or equipment, such as a specialist hearing aid as a substitute for an auditory perceptual disorder (Fifer 1993).
Non‐invasive brain stimulation
Transcranial direct current stimulation (tDCS) involves non‐invasive brain stimulation that modulates cortical excitability through use of weak electrical currents. Stimulation is delivered via electrodes placed on the stroke survivor's scalp (Koo 2018), with the aim of modulating the primary somatosensory cortex, whereas galvanic vestibular stimulation (GVS) uses current applied to the left and right mastoids to stimulate the vestibular system at the perceptual level (Nakamura 2014).
Pharmacological interventions
A range of pharmacological interventions, including aripiprazole and haloperidol (Chen 2011; Nguyen 2011), have been described as interventions for auditory or visual hallucinations.
Why it is important to do this review
Perceptual disorders following stroke can affect all senses (hearing, smell, somatosensation, taste, touch and vision). Currently, information on interventions available for perceptual disorders is limited. A range of healthcare professionals care for stroke survivors, yet some may not be fully aware of the range of perceptual impairments that can present following stroke (Dutta 2013). As a result, some impairments may go undiagnosed (Jones 2006), and therefore, untreated. Stroke survivors may be unaware of their own perceptual disorder(s) and lack understanding of the impact they have on everyday activities, which may in turn reduce their self‐confidence (Hazelton 2019a). Visual perceptual disorders are arguably the most well known, yet care provision for these disorders varies, with significant inequity in access even within the UK (Rowe 2015). Barriers to service provision include healthcare providers' limited knowledge, unclear treatment protocols, and a lack of evidence to inform effective rehabilitation (Pollock 2011a).
The first version of this review was published in 2011 (Bowen 2011). This updated and revised version reflects the recognition that an updated synthesis of new evidence in this field is needed.
Stakeholder involvement
In order to ensure that this review update was relevant and meaningful to key stakeholders, a group of five people with lived experience of stroke and four with expert clinical knowledge were involved throughout all stages of the review. This stakeholder group was formed as part of a wider project relating to evidence for perceptual problems after stroke (Hazelton 2019b). We adopted a 'top and tail' approach to the stakeholder group's input into the review (Pollock 2019a); that is, the stakeholder group contributed to decision making at the planning and methods stage of the review update, and during the interpretation of results during the final stages of the review update. We facilitated involvement through the use of structured methods, such as discussion and independent voting via a series of meetings and online communications. The stakeholder group informed and agreed on the definitions of perception and the senses used in this version of the review. The group also established priorities for outcome measures, which informed the primary and secondary outcomes of interest for this review update. The group members considered the results of the review and contributed ideas which have been incorporated into the organisation, interpretation of data analyses, discussion, and conclusions of this review.
Objectives
To assess the effectiveness of interventions aimed at perceptual disorders after stroke compared to no intervention or control (placebo, standard care, attention control), on measures of performance in activities of daily living.
Methods
Criteria for considering studies for this review
Types of studies
We included randomised controlled trials (RCTs) of interventions that targeted perceptual disorders as a consequence of stroke. In the case of cross‐over trials, we analysed data from baseline to the point of cross‐over.
Types of participants
Participants were adults (18 years and older) with impaired perception as a result of a stroke. We used the definition of perception set out in the World Health Organization's (WHO) International Classification of Functioning, Disability and Health (ICF) (b156): “specific mental functions of recognizing and interpreting sensory stimuli” (World Health Organization 2001). We used this definition because it is internationally recognised, can be applied across all senses, and distinguishes perception from closely related functions, such as sensation and cognition. Following detailed discussion with our stakeholder group, we refined and clarified the scope of the definition to include disorders affecting six hearing (auditory), smell (olfactory), somatosensation (relating to information from the skin, mucous membranes, limbs, and joints; includes proprioception, pressure, and temperature), touch (tactile), taste (gustatory), and vision (visual). Although touch could be considered a component of somatosensation, we chose to present these separately. We considered touch and somatosensation to be separate and distinguishable functions and believed that this would be more accessible to stroke survivors and families, for whom touch is well‐recognised as one of the five “traditional” senses.
We excluded trials of interventions for disorders of sensation (informed by the WHO ICF categories of ‘Sensory functions and pain’ (b210‐b229)). Thus, for example, we excluded interventions for visual field loss, or for disorders of touch that involved only detection of tactile stimuli, but no recognition or interpretation of that sensory information. We excluded disorders of other specific mental functions, such as memory (b144), thought (b160), and higher‐level cognitive functions (b164). We refer to these collectively as cognition or cognitive function throughout the rest of this review.
Clinically, neglect is widely accepted to be an attentional disorder (Corbetta 2011), and we therefore excluded interventions for unilateral neglect (which are considered in another Cochrane Review (Longley 2021). We excluded apraxia as it relates to the knowledge and production of movement rather than perception (Heilman 2003); it is also the focus of another Cochrane Review (West 2008). We excluded trials relating to the perception of other stroke‐related problems or the consequences of such problems, such as perception of pain, numbness, or weakness. We excluded trials that used other definitions of perception, such as those relating to the way in which an object or concept is regarded, understood, or interpreted.
Given the complex nature of perception and perceptual disorders and the associated language, we anticipated challenges in the application of our perception definition and eligibility criteria (above), and in distinguishing between perceptual disorders and those of other functions. We therefore planned to use relevant stroke and disorder details (lesion location, classification systems used, or reported theories of neural function) to support the application of our definition. We included trials that combined perceptual disorders with disorders that affected other functions; we also planned to include trials where the precise nature of the perceptual disorder could not be determined.
In discussion with our stakeholder group, we decided how to address more complex, stroke‐related disorders, as follows.
We excluded balance it frequently incorporates a broad range of non‐perceptual input; it is a stage ‘after’ perception (analogous to reading and vision); it has an evidence base that often takes a physical function approach; and its inclusion could make the results less meaningful. We included Pusher Syndrome, "a clinical disorder following left or right brain damage in which patients actively push away from the nonhemiparetic side" (Karnath 2003). Whilst the aetiology and mechanism of the disorder are not fully understood, we considered it a disorder of the perception of body posture.
Pusher Syndrome and balance disorders present differently, but on occasion may not be well differentiated in clinical we made our decision to include Pusher Syndrome but exclude balance disorders based on detailed discussion with our stroke‐specialist clinicians within our research and stakeholder group. Our main consideration was to focus as much as possible on perception, rather than the many, closely‐related disorders seen after stroke (as detailed above). In order to do this, clear boundaries were required; however, we are aware that other approaches to this problem could have been used.
Types of interventions
We included any intervention that addressed a perceptual disorder. We grouped trials according to the sense affected (i.e. hearing (auditory), smell (olfactory), somatosensation (including proprioception, pressure, and temperature), touch (tactile), taste (gustatory), or vision (visual)). We planned to include and code interventions that addressed perceptual disorders across more than one sense as a 'mixed' group.
We categorised interventions by the perceptual disorder targeted, followed by the therapeutic approach rehabilitation, pharmacological, non‐invasive brain stimulation, surgical, or assessment and screening intervention. We further subcategorised rehabilitation interventions by the mode of restitution, compensation, substitution, or a combination of these (Table 5).
Two review authors (CH/KMc/KT) independently classified included intervention approaches, with input from a third author (DG/SH) where differences could not be resolved through discussion. We included interventions delivered in any stroke care setting (hospital, community, and outpatient settings) and any geographical location.
We included trials in which a comparison was made between an active treatment group that received an intervention for a perceptual disorder, versus a group that received no treatment or control intervention (placebo, standard care, attention control – see Table 6) or an alternative perceptual intervention.
In addition, we aimed to examine whether one active intervention was more effective than any other active intervention for people with perceptual deficits after stroke in relation to the outcomes listed above. We also considered if some interventions were more effective for stroke survivors with specific demographic variables (including age, stroke severity, time since stroke), as measured by the primary outcome (ADL).
Types of outcome measures
Stakeholder Involvement
Our stakeholder group informed the selection of included outcome measures. This process involved identifying relevant outcome categories, then using a consensus process to rank them by importance in the context of this review.
Primary outcomes
Our primary outcome was performance in activities of daily living (ADL) immediately post‐intervention (‘immediate’ time point). We also planned to analyse ADL when measured at a follow‐up time point of three months after the intervention. We recognise that other follow‐up time points – for example, a longer (6‐ to 12‐month) follow‐up – are also important, but chose to focus on a short‐term (3‐month) follow‐up (1) based on experience in other reviews, we considered that trials were most likely to provide data for this time point (Hazelton 2022; Longley 2021; Pollock 2019b); and (2) synthesising data for this time point would enable us to explore whether any treatment effects observed immediately after treatment were maintained in the short term. We included any validated, standardised measure of ADL, such as the Barthel Index (Mahoney 1965), Functional Independence Measure (Keith 1987), Modified Rankin Scale (Rankin 1957), Katz Index of Activities of Daily Living (Katz 1983), Assessment of Motor and Process Skills (AMPS) (Fisher 1994), and Rehabilitation Activities Profile (Lankhorst 1996). Where trials provided data on more than one ADL outcome measurement instrument, we extracted and analysed the measure occurring earliest in the above list.
Secondary outcomes
Additional outcomes of relevance to this review
extended activities of daily living (EADL), measured by scales such as the Frenchay Activities Index, Nottingham Extended Activities of Daily Living scale, Lawton Instrumental Activities of Daily Living, and Rivermead Activities of Daily Living score;quality of life and participation. This included three groups of quality of life (QoL), measured by scales including the EQ5D, Health‐related quality of life scale, Quality of Well Being Scale, SF36, Stroke Impact Scale; social activities and participation, measured by scales including the Social Problem‐Solving Video Measure, Australian Community Participation Questionnaire, and the ICEpop CAPability measure for Older people;mobility and navigation, measured by scales such as the Rivermead Mobility Index, 6‐Minute Walk Test, functional ambulation, and Timed Up and Go Test;mental health and psychological well‐being stroke survivors, measured by scales such as the Hospital Anxiety and Depression Scale, Beck Depressive Inventory, General Health Questionnaire, and Geriatric Depression Scale discharge destination;family, friends, and carers, measured by scales such as the Carer Strain Index, Brief Family Distress Scale, and Perceived Caregiver Burden;perceptual function, measured by scales such as the Rivermead Perceptual Assessment Battery, Motor‐Free Visual Perception Test, Birmingham Object Recognition Battery, and Chessington Occupational Therapy Neurological Assessment Battery; adverse events; for example, falls, death, fatigue, and accident rates.
Other outcomes noted, but not analysed
sensation; for example, visual acuity, visual fields;motor ability (including balance); for example, grip strength, Berg Balance Scale, Postural Assessment Scale for Stroke, Motor Assessment Scale, Fugl‐Meyer Assessment of Motor Recovery after Stroke, and Motricity Index; cognition (including attention); for example, the Mini–Mental State Examination, Wechsler Adult Intelligence Scale (WAIS), line cancellation, and star cancellation.
Search methods for identification of studies
We did not apply any language limitations. We arranged for translations of publications if required.
Electronic searches
We searched the following electronic bibliographic databases and clinical trial registers (last searched on 9 August 2021):
Cochrane Stroke Group’s Register, the Cochrane Central Register of Controlled Trials (CENTRAL), and the Cochrane Database of Systematic Reviews (CDSR) in the Cochrane Library (2021, Issue 8) (Appendix 1);MEDLINE (Ovid; from 1946) (Appendix 2);Embase (Ovid; from 1974) (Appendix 3);ERIC (EBSCO; from 1966) (Appendix 4);CINAHL (EBSCO) (Cumulative Index to Nursing and Allied Health Literature; from 1982) (Appendix 5);AMED (Ovid; from 1885) (Appendix 6);PsycINFO (Ovid; from 1806) (Appendix 7);Epistemonikos Database (www.epistemonikos.org/) (Appendix 8);Web of Science – Core Collection (Appendix 9);Centre for Reviews and Dissemination (Database of Abstracts of Reviews of Effects (DARE) and the NHS Economic Evaluation Database (NHS EED), and Health Technology Assessment (HTA) (www.crd.york.ac.uk/CRDWeb/) (Appendix 10);ProQuest Dissertations and Theses Global (from 1997) (Appendix 11).
We developed all the search strategies with the help of the Cochrane Stroke Group Information Specialist (MEDLINE Ovid; Appendix 2). Since the previously published version of this review (Bowen 2011), the search strategies have been extensively updated to include any newly identified, relevant controlled vocabulary terms and keywords. The MEDLINE Ovid search strategy included the Cochrane Highly Sensitive Search Strategy for identifying randomised trials in MEDLINE: sensitivity‐maximizing version (2008 revision) as described in the Technical Supplement to Chapter 4, 'Searching for and selecting studies', in the Cochrane Handbook for Systematic Reviews of Interventions (Lefebvre 2021).
Searching other resources
In an effort to identify further published, ongoing, and unpublished trials, we conducted supplementary searches of the following
registers of ongoing US National Institutes of Health Ongoing Trials Register ClinicalTrials.gov (www.clinicaltrials.gov) (Appendix 12);World Health Organization International Clinical Trials Registry Platform (trialsearch.who.int) (Appendix 12);OpenGrey (www.greynet.org/opengreyrepository.html);Grey Matters: a practical tool for searching health‐related grey literature, (www.cadth.ca/grey-matters-practical-tool-searching-health-related-grey-literature) (last accessed 24 November 2020);Google Scholar (scholar.google.com/) (top 250 most relevant entries);NIHR Clinical Research Network (www.nihr.ac.uk/);Physiotherapy Evidence Database (PEDro) (pedro.org.au);OTseeker (www.otseeker.com);PROSPERO International prospective register of systematic reviews (www.crd.york.ac.uk/prospero/).
In addition, we searched national and international guidelines, government websites, relevant healthcare professional (HCP) websites, and the websites of relevant charities and patient support organisations. We also contacted research, professional associations or foundations, and experts in the field.
We completed forward citation tracking using Science Citation index and Google Scholar (scholar.google.com/) and searched the reference lists of included trials.
Data collection and analysis
Selection of studies
The Cochrane Stroke Information Specialist (JC) conducted the database searches, while three review authors (PC, CH, KMc) searched other supplementary resources. One review author (KMc/KT) excluded duplicates and any titles which were obviously not related to stroke and perception. We obtained the abstracts for the remaining references and two review authors (CH, KMc, KT) independently considered each of these abstracts, excluding any trials that clearly did not meet the inclusion criteria. The review authors resolved any disagreements through discussion, involving a third review author or clinical specialist from the stakeholder group with expertise in the relevant field for expert opinion where needed. Two review authors (CH/KMc/KT) obtained and independently assessed the full texts of all potentially relevant trials. We resolved any disagreements following discussion with an additional member of the research team or Clinical Expert Group (see Acknowledgements). We recorded the reason(s) for each exclusion at the full‐text screening stage.
Data extraction and management
One review author (KMc/KT) extracted data into drafted, piloted, Excel‐based extraction forms, and a second review author (CH/KT) checked data entry in full. We resolved disagreements through discussion or by involving a third review author if necessary.
Where possible we collected the following data.
Study: country, setting, year, design, number of centres, number of trial arms.Methods: randomisation method, prospective power calculation, recruitment details, dropout details.Participant: inclusion criteria, exclusion criteria, number, age, sex, stroke details (type, time since stroke, hemisphere affected, severity), perceptual disorder, sense(s) affected and method of diagnosis, severity, presence of other stroke‐related impairment.Intervention: classification (active intervention/no treatment/control), intervention approach (rehabilitation/non‐invasive brain stimulation/pharmacological/surgery/assessment and screening). Rehabilitation interventions were further classified as restitution, compensation, substitution, or a combination of these. Description using TIDieR headings (Hoffmann 2014), of materials, procedures, who delivered, mode, where, session and duration details, tailoring and modification. Outcomes: instrument name and aspect recorded, time point of data collection and results. For dichotomous data, we extracted the numbers who specifically did, or did not, experience the outcome in each group; that is, the 2x2 table. For continuous data, we extracted means and standard deviations for each intervention group. Where these were unavailable, we contacted authors and requested the data or calculated them using Cochrane methods (Higgins 2020). For all outcomes, we recorded any significance test, t, f, P values, and directions of findings. If a trial provided data from more than one outcome measure for the primary outcome (ADL), we extracted them in the order set out in Primary outcomes. Where a trial reported an outcome measure not listed above, we sought data on its validity and reliability in comparison to other outcome measures and ordered them from most to least appropriate. Clinical experts in the relevant topic area then considered and approved these lists.
Assessment of risk of bias in included studies
We used the Cochrane risk of bias (ROB‐1) tool to judge risk of bias (Higgins 2011). Two review authors (CH, KT) independently performed assessment, grading the risk due to selection bias (randomisation, allocation concealment), performance bias (differences in the interventions), detection bias (masking of outcome assessment), attrition bias (trial withdrawals), reporting bias, and any other sources of bias.
Two review authors independently categorised the risk of bias as high, low, or unclear, with any disagreements between authors discussed with a third review author to reach consensus. Reasons for judgements were transparently reported. We looked for an intention‐to‐treat analysis when participants had dropped out of trials and sought evidence of additional publications if outcome measures were not fully reported.
If judgements for performance bias and detection bias differed (e.g. low and unclear), we labelled the overall category as unclear. Where we judged a trial as having a high risk of performance or detection bias, then we considered blinding as an overall category as having a high risk of bias.
Measures of treatment effect
We used Review Manager software (version 5.4; Review Manager 2020) to carry out statistical analyses to determine treatment effects. We used a random‐effects model for meta‐analysis throughout. For dichotomous variables, we calculated a Peto odds ratios with 95% confidence intervals (CI). For continuous data, we calculated the mean difference (for measurements in the same scale) or standardised mean differences (for measurements on different scales) and 95% CIs.
We treated ADL and other ordinal scales for secondary outcomes as continuous outcomes (as an accepted meta‐analytic technique for ordinal outcome data is not yet available).
Where reported outcomes used a measurement scale where a lower value indicates a better outcome, we multiplied the reported values by ‐1, so that in all analyses a higher value was indicative of a better outcome.
We used final‐value scores for analysis. If trials reported change‐from‐baseline values and the baseline value was available, we calculated the final‐value scores. If trials reported change values and the baseline value was not available, we used these data in meta‐analyses but planned sensitivity analyses to investigate the effect of including the data.
We used data from trials involving stroke survivors. For trials with a mixed population, we planned to extract the stroke‐specific data where these were available. Where these were unavailable, we used mixed population data when more than 80% of participants were stroke survivors. We planned to conduct sensitivity analyses to investigate the effect of including these data.
Unit of analysis issues
We did not anticipate any specific unit of analysis issues. Where trials had more than one eligible active intervention group within the same comparison (against a control, placebo, standard care, or no treatment group), we divided the control group data between the multiple pair‐wise comparisons to ensure there was no double counting of participants within any one analysis.
Dealing with missing data
Where we identified missing outcome data, we requested these data from the trialists. If data remained unavailable, we calculated the value where possible; for example, by estimating a standard deviation (SD) based on a reported standard error, using Cochrane Handbook for Systematic Reviews of Interventions methods (Higgins 2020). We planned to conduct sensitivity analyses to investigate the effect of entering assumed values.
Assessment of heterogeneity
We calculated statistical heterogeneity using the I^2^ statistic, and discussed the results in light of Cochrane Handbook guidance on heterogeneity (Deeks 2020). We used the I^2^ statistic to categorise the level of heterogeneity as
I^2^ of 0% represents no heterogeneity;0% < I^2^ < 30% may represent some heterogeneity;30% ≤ I^2^ < 50% may represent moderate heterogeneity;50% ≤ I^2^ < 75% may represent substantial heterogeneity;I^2^ ≥ 75% may represent considerable heterogeneity.
We conducted pre‐specified subgroup analyses (Subgroup analysis and investigation of heterogeneity), and considered further sensitivity analyses based on characteristics arising during data extraction.
Assessment of reporting biases
We compared the availability of our planned outcomes (i.e. as listed in study protocols or methods) with those reported in the included trials. We noted where study authors described an outcome as measured but did not report it or where data were unavailable for analysis. We planned to examine a funnel plot for possible publication bias if we identified 10 or more trials reporting a single outcome (Egger 1997).
Data synthesis
Our analysis pooled trials that (1) active intervention to no treatment; and (2) active intervention to a control intervention (placebo, standard care, or attention control). We stratified these analyses according to intervention approach category (rehabilitation/non‐invasive brain stimulation/surgery/pharmacology/assessment and screening).
The pre‐specified comparisons
active intervention for hearing perception impairment versus no treatment;active intervention for hearing perception impairment versus control (attention control/standard care/placebo);active intervention for smell perception impairment versus no treatment;active intervention for smell perception impairment versus control (attention control/standard care/placebo);active intervention for somatosensory perception impairment versus no treatment;active intervention for somatosensory perception impairment versus control (attention control/standard care/placebo);active intervention for touch perception impairment versus no treatment;active intervention for touch perception impairment versus control (attention control/standard care/placebo);active intervention for taste perception impairment versus no treatment;active intervention for taste perception impairment versus control (attention control/standard care/placebo);active intervention for visual perceptual impairment versus no treatment;active intervention for visual perceptual impairment versus control (attention control/standard care/placebo).
The planned outcome measures and time points for each comparison above
ADL: immediate post‐intervention time point;ADL: follow‐up time point (three months);EADL: immediate post‐intervention time point;QoL: immediate post‐intervention time point;Mental health and psychological well‐being: immediate post‐intervention time point;Perceptual immediate post‐intervention time point;Adverse immediate post‐intervention time point.
We directly compared one active intervention with another, where we considered it meaningful to do so. The review author team took these decisions by considering the perceptual disorder addressed, the nature of the intervention, and the outcome measures assessed in the relevant trials.
We tabulated outcome measures available at follow‐up time points.
For trials reporting other outcomes of interest (sensation, cognition, motor function), we tabulated outcome measures at the immediate and follow‐up time points, but did not conduct a meta‐analysis of these data.
Subgroup analysis and investigation of heterogeneity
Following stakeholder input, we planned to conduct the following subgroup analyses, to explore different treatment approaches and participant/stroke
treatment rehabilitation, non‐invasive brain stimulation, surgery, pharmacology, assessment and screening;participants: age (adult 18 to 65 years, older adult > 65 years), sex;stroke stroke severity, time since stroke, type of stroke, laterality of stroke.
When 10 or more trials were included in a single analysis (with sufficient information to determine the subgroups), we planned to use established subgroup analysis methods (Deeks 2020).
Sensitivity analysis
We planned to carry out sensitivity analyses on the primary outcome to explore the risk of bias categories (selection bias, performance bias, detection bias, attrition bias reporting bias and any other sources of bias) and publication type.
Where possible, we explored the effect of including trials which were at ‘high’ or ‘unclear’ risk of bias.
Summary of findings and assessment of the quality of the evidence
We presented our findings for (1) the intervention versus no treatment comparison, and (2) the intervention versus control comparison, containing intervention subgroups where relevant. We presented data for six different activities of daily living, extended activities of daily living, quality of life, mental health and psychological well‐being (of stroke survivors and family, friends, and carers), perceptual function, and adverse events.
Two review authors independently rated our confidence in the cumulative evidence for each synthesis using GRADE methodology and considering design, inconsistency, indirectness, imprecision, and publication bias. Where necessary, we reached agreement through discussion. We downgraded the level of evidence accordingly, with one downgrade for each concern, and a maximum of two downgrades for any one criterion. Beginning with a default grade of high quality for each comparison, one downgrade reduced the level of evidence to moderate quality, two downgrades reduced it to low quality, and three or more downgrades reduced it to very low quality. We used the following definitions of evidence quality (Guyatt 2008):
high when further research is very unlikely to change our confidence in the estimate of effect;moderate when further research is likely to have an important impact on our confidence in the estimate of effect and may change the estimate; low when further research is very likely to have an important impact on our confidence in the estimate of effect and is likely to change the estimate;very low when we are very uncertain about the estimate.
We considered other factors that may have affected the quality of evidence and transparently recorded the reasons for downgrading evidence quality.
Results
Description of studies
See: Characteristics of included studies; Characteristics of excluded studies; Characteristics of ongoing studies; and Characteristics of studies awaiting classification.
Results of the search
Our database and additional searches identified 94,434 records. In addition, we considered 10 trials identified in the earlier version of the review (Bowen 2011). We retrieved 318 full‐text reports to screen for eligibility. We excluded 252 of these as they were obviously irrelevant to the review. Of the remaining 66 reports, we included 18 studies (19 reports), excluded 25 studies (25 reports), identified 11 ongoing studies, and categorised 11 as 'awaiting classification'. We present the flow of literature and results of our search and screening process in the PRISMA flow diagram (Figure 1).
Included studies
We included 18 trials (541 participants, of whom 535 (98.9%) were stroke survivors) (An 2019; An 2020; Bergmann 2018; Carey 2011a; Chen 2012; Cho 2015; Choi 2018; De Bruyn 2018; Edmans 2000; Kang 2009; Kim 2015; Koo 2018; Lee 2021; Lincoln 1985; Park 2015; Seim 2021; Yang 2015; Yun 2018). Seventeen of the trials recruited only stroke survivors. Of 33 participants included in Lincoln 1985, six had a head injury.
The Bowen 2011 version of this review included six trials and 339 participants (Dirette 1999; Edmans 2000; Hajek 1993; Lincoln 1985; Mazer 2003; Taylor 1971). Following consideration of their populations in relation to the revised inclusion criteria for this update, we excluded four of these trials (Dirette 1999; Hajek 1993; Mazer 2003; Taylor 1971) (see Excluded studies below).
Most included trials were two‐arm RCTs, with Kim 2015 randomising participants across two active interventions groups and a no‐treatment group. One trial used a cross‐over design (Lee 2021).
Trials were conducted across seven countries in Asia, Australia, Europe, and North America (Table 7). Recruitment details were often absent or limited. Most trials (14/18; 77.8%) recruited from a single site. Three recruited across two to six sites (Carey 2011a; Chen 2012; De Bruyn 2018). Trials took place in a hospital or medical facility; none were community based.
Between 11 and 80 (mean 29.9, SD 15.9) participants were randomised to the trials. We present the details of dropouts, including the number of participants lost during intervention delivery and during any follow‐up, in Table 8.
Included comparisons
The 18 included trials featured 20 comparisons relevant to our review.
Intervention versus no treatment
There were three comparisons of an intervention with no treatment (Kim 2015: a three‐arm RCT, two relevant comparisons; Cho 2015). The former trial explored pressure sense training for touch perception disorder; the latter explored neurofeedback training for visual perceptual deficit.
Intervention versus control (placebo/sham/control)
There were three comparisons of an intervention with a Koo 2018 (somatosensory), Seim 2021 (touch), and Lincoln 1985 (visual). The nature of the control group varied. Two trials compared treatment to a placebo sham transcranial direct current stimulation (tDCS), where electrodes were placed on the scalp, but no current/stimulation applied (Koo 2018); and a sham glove which did not vibrate like its active counterpart (Seim 2021). The control group in Lincoln 1985 used conventional occupational therapy, with participants completing activities that were “not specifically designed to improve perceptual abilities”.
Intervention 1 versus intervention 2
Fourteen studies compared two different interventions for perceptual disorders (An 2019; An 2020; Bergmann 2018; Carey 2011a; Chen 2012; Choi 2018; De Bruyn 2018; Edmans 2000; Kang 2009; Kim 2015; Lee 2021; Park 2015; Yang 2015; Yun 2018). The nature of the interventions varied, depending on the nature of the disorders and treatment approach taken (detailed below).
Population
We present details of the participants in included trials in Table 9.
The time since stroke onset the shortest was approximately 19 days (Koo 2018), and longest 4.3 years (Seim 2021). Where reported, trials included haemorrhagic and ischaemic stroke. Two trials described stroke severity using the National Institutes of Health Stroke Scale (NIHSS) measure (Carey 2011a; Yun 2018). Participant ages (mean) ranged from 48.8 years in Lincoln 1985 to 75.5 years in De Bruyn 2018.
We identified no eligible RCTs that addressed interventions for disorders of hearing, taste, or smell perception. Three trials (five randomised comparisons) addressed touch perception dysfunction (Kim 2015; Lee 2021; Seim 2021), seven somatosensory (An 2019; An 2020; Bergmann 2018; De Bruyn 2018; Koo 2018; Yang 2015; Yun 2018), seven visual (Chen 2012; Cho 2015; Choi 2018; Edmans 2000; Kang 2009; Lincoln 1985; Park 2015), and one had mixed touch and somatosensory disorder (Carey 2011a). Mixed‐sensory categories were not included in our analysis plan; we excluded these data from analysis.
Touch perception disorders were diagnosed using tools that included the revised Nottingham Sensory Assessment (Lee 2021) and Semmes Weinstein monofilament (Kim 2015; Seim 2021).
Somatosensory perception disorders included Pusher Syndrome in five trials, diagnosed using specialised rating scales, such as the Burke Lateropulsion Scale (An 2019; Yun 2018), and Scale of Contraversive Pushing (Bergmann 2018; Yang 2015), or less specific disorders of somatosensation via practical assessments, such as pinprick and light touch tests (De Bruyn 2018; Koo 2018).
Visual perception disorders addressed generalised visual perceptual disorders (Cho 2015; Choi 2018; Edmans 2000; Kang 2009; Lincoln 1985; Park 2015), diagnosed using specialised tests, such as the Motor‐free Visual Perception Test (MVPT) (Choi 2018; Kang 2009), Rivermead Perceptual Assessment Battery (Edmans 2000; Lincoln 1985), or cognitive tests that include visual perception subsections (such as the Modified Mini‐Mental State Examination (MMSE); Cho 2015; Park 2015), as well as visual memory deficit (Chen 2012), diagnosed using a complex figure‐drawing test.
Interventions
Across 20 comparisons in this review, we evaluated 32 interventions (excluding no‐treatment and control groups). We categorised these by one intervention involved non‐invasive brain stimulation (Koo 2018); all others used a rehabilitative approach, either compensation (one Edmans 2000), restitution (25 An 2019; An 2020; Bergmann 2018; Chen 2012; Cho 2015; Choi 2018; De Bruyn 2018; Edmans 2000; Kang 2009; Kim 2015; Lee 2021; Lincoln 1985; Park 2015; Seim 2021; Yang 2015; Yun 2018), restitution combined with another rehabilitative approach (four Bergmann 2018; Carey 2011a; Lee 2021; Yun 2018), and one where the nature of the rehabilitation approach was unclear (Carey 2011a).
Details of the interventions, covering the materials, procedures, provider, therapy modality, location, intervention session details and overall duration, and tailoring or modification of the intervention are available in the Characteristics of included studies table. We provide short summaries below.
Interventions for somatosensation perception disorders
There were seven trials (196 participants) exploring 13 interventions (An 2019; An 2020; Bergmann 2018; De Bruyn 2018; Koo 2018; Yang 2015; Yun 2018).
####### Interventions
Rehabilitation (restitution) (10 interventions) game‐based vertical posture training (two interventions) (An 2019; An 2020)standard posture training (two interventions) (An 2019; An 2020)conventional physiotherapy (for Pusher Syndrome) (two interventions) (Bergmann 2018; De Bruyn 2018)physiotherapy plus sensorimotor training (De Bruyn 2018)physiotherapy plus motor training (De Bruyn 2018)computerised interactive visual feedback training (Wii Fit) (Yang 2015)Mirror feedback training (Yang 2015)Rehabilitation (restitution and substitution) (two interventions)robot‐assisted gait training (Bergmann 2018; Yun 2018)Non‐invasive brain stimulation (one intervention)tDCS (Koo 2018)
####### Materials and procedures
One non‐invasive brain stimulation intervention used relevant equipment to deliver tDCS stimulation to the appropriate hemisphere and region (Koo 2018). All other interventions used rehabilitative approaches. For Pusher Syndrome therapy, these could be grouped into either game‐based postural training with supporting equipment or conventional physiotherapy for Pusher Syndrome. For non‐Pusher disorders, one trial explored sensory‐motor training focusing on sense discrimination, and table‐top ‘motor therapy’.
Game‐based postural training used one of three (named) Wii Fit (computer‐based exercises and balance board) (Yang 2015), Lokomat (computer‐based exercises, supportive harness, and treadmill) (Bergmann 2018; Yun 2018), and Spine Balance 3D (computer‐based exercises and whole‐body tilt apparatus) (An 2019; An 2020). Each was used to provide physical therapy, with participants asked to achieve a set body position or movement in response to the computerised exercises, and in relation to any positional change caused by the associated equipment. Conventional physiotherapy for Pusher Syndrome often involved postural training and weight shifting, using visual cues in the room to regulate posture, alongside verbal feedback from the therapist. Materials used were often unclear but included a chair for seated exercises and a mirror for feedback (An 2019; An 2020; Bergmann 2018; Yang 2015; Yun 2018). Sensory‐motor training used the Study of the Effectiveness of Neurorehabilitation on Sensation (SENSe) approach with three sensory discrimination texture discrimination, limb position sense, and object recognition (De Bruyn 2018). Materials in these tasks included different textures (fabric, wallpaper, plastic, and sandpaper), different objects of varying shape, size and materials. A range of exercises were used, such as smoothing out fabric whilst appreciating the texture, moving the limb to a specific position, and arranging bottles in order of weight. ‘Motor Therapy’ used tabletop games such as chess (with clear cognitive and attentional demand), with a set programme of upper limb exercises used to improve gross movement and dexterity (De Bruyn 2018).
####### Delivery
All interventions were delivered one‐to‐one, in a hospital setting. A physiotherapist delivered five of 13 interventions (which for one trial specified that they had “more than five years’ experience” (An 2020; Yang 2015; Yun 2018), while the remaining providers were not reported, or unclear.
####### Sessions and duration
There was a great deal of similarity in the timing and duration of the rehabilitation interventions. Sessions typically lasted 30 to 60 minutes per day, were conducted three to five times per week, for a duration of two to four weeks in total. In contrast,** **non‐invasive brain stimulation was delivered for 20 minutes per day for 10 days (Koo 2018).
####### Tailoring and modification
No modification was noted in any intervention or trial. The tailoring of interventions varied hugely. For seven interventions, it was not mentioned (An 2019; An 2020; Bergmann 2018; De Bruyn 2018; Koo 2018; Yang 2015); for others, this was clearly fundamental to the intervention delivery, where the exercises were tailored to participant ability before training began (An 2019; An 2020; De Bruyn 2018; Yun 2018). In others, the difficulty level was altered relative to performance; for example, the “the speed and range of trunk movement” was increased or exercises were “changing from sitting to standing” (An 2020).
Interventions for touch perception disorders
Three trials (70 participants) explored five interventions (Kim 2015; Lee 2021; Seim 2021).
####### Interventions
Rehabilitation (restitution) (four interventions) pressure sense perception training on stable surface (Kim 2015)pressure sense perception training on unstable surface (Kim 2015) hand exercises (without glove) (Lee 2021)a vibrating glove “VTS Glove” (Seim 2021)Rehabilitation (restitution and substitution) (one intervention) robot glove‐based hand exercises (Lee 2021)
####### Materials and procedures
Interventions were of two main pressure sense training involving exercises on a stable or unstable surface (Kim 2015), and hand exercises, either with or without a glove to assist (Lee 2021; Seim 2021). In pressure sense training, participants stood on either a stable foam block, or on an unstable balance pad. They were asked to shift weight to their affected side, and pressure in the heel was measured to ensure a desired level was reached (Kim 2015). The hand exercises included a range of passive range of motion tasks, and task‐based activities and games. The addition of a robotic glove was used to detect movement and provide a simultaneous display of performance on a computer screen, as well as providing sensory stimulation (Lee 2021). A different, vibrating, glove was used to provide stimulation to skin on the palm and fingers; it did not require any exercises (Seim 2021).
####### Delivery
Interventions were delivered by physiotherapists (two interventions; Kim 2015), or occupational therapists (OTs) (two Lee 2021), in one‐to‐one sessions in a hospital setting (four Kim 2015; Lee 2021), with the vibrating glove used by participants themselves in their own home (one Seim 2021).
####### Sessions and duration
These varied from 30‐minute sessions three days per week (Kim 2015), to three‐hour sessions seven days per week (Seim 2021). Total duration ranged from four to eight weeks.** **
####### Tailoring and modification
None of the trials reported modification of the interventions. Tailoring to participants' ability was not reported for two interventions (Lee 2021; Seim 2021). The interventions with tailoring comprised the pressure sense training, which tailored training to participants' heel pressure, and allowed progression to a harder stage when a suitable pressure level was achieved (Kim 2015), and robot‐assisted hand exercises, whose settings were adjusted to participants' ability (Lee 2021).
Interventions for visual perception disorders
Seven trials (225 participants) explored 12 interventions (Chen 2012; Cho 2015; Choi 2018; Edmans 2000; Kang 2009; Lincoln 1985; Park 2015).
####### Interventions
Rehabilitation (restitution) (11 interventions) image global processing training (Chen 2012) image rote repetition training (Chen 2012)neurofeedback (NFB) training (Cho 2015)Wii Fit virtual reality training (WVRT) (Choi 2018)general balance training (Choi 2018)transfer of training perceptual treatment (Edmans 2000)functional perceptual treatment (Edmans 2000)computerised visual perception rehabilitation with motion tracking (Kang 2009)computer‐based cognitive rehabilitation programme (Kang 2009)OT‐led perceptual training (Lincoln 1985)computer‐based cognitive rehabilitation training (Park 2015)conventional cognitive rehabilitation (Park 2015)Rehabilitation (restitution and compensation) (one intervention) functional perceptual treatment (Edmans 2000)
####### Materials and procedures
All interventions used a rehabilitation approach and could be grouped into five main paper‐based tasks, occupational therapist‐led task‐based training, physical interventions, cognitive and perceptual exercises, and neurofeedback training**.**
Three interventions used paper‐based in two, participants traced then reproduced a complex figure (Rey–Osterrieth Complex Figure) either as a whole figure, or broken down into its component parts (Chen 2012); the third used “conventional cognitive rehabilitation” the exact nature of which was not clearly stated (Park 2015). Three interventions used an occupational therapy approach, training visual perceptual skills using functional and task based training (Edmans 2000; Lincoln 1985). Although not well described, this training included simple perceptual activities, such as stick length sorting, colour matching, and parquetry mosaic tasks. One trial explored physical interventions for visual perceptual disorders alongside balance disturbance (Choi 2018). Wii Fit with balance board training used a range of activities to stimulate interest and motivation, such as simulated tightrope walking and slalom, which encouraged multidirectional weight shifting. Balance training used a balance board, with the participant asked to shift weight, using a mirror for feedback. One approach used computerised exercises – these frequently were called “cognitive” in nature, but had a clear focus on improving visual perceptual skills, including object recognition, object constancy, figure‐ground organisation, visual discrimination, and visual organisation (Kang 2009; Park 2015).
####### Delivery
Delivery was poorly reported for vision interventions. It was explicitly stated that a physiotherapist (one intervention; Choi 2018) and an occupational therapist (three interventions; Kang 2009; Lee 2021) delivered interventions. All were delivered on a one‐to‐one basis, and in a hospital setting.
####### Sessions and duration
Two interventions (involving paper‐based repetition training) were delivered in a one‐off single session of 90 minutes (Chen 2012).** **For the other interventions, sessions typically lasted 30 minutes, repeated three to five times per week, for four to six weeks.
####### Tailoring and modification
Again, no intervention modification was reported. Tailoring was either not reported or was unclear for nine interventions; in three others, the intervention exercises and difficulty were based on participants' perceptual ability (Lincoln 1985; Park 2015).
Interventions for hearing, taste, or smell perception disorders
No trials assessed interventions for hearing, taste, or smell perception disorders.
Outcome measures
The outcome measures used in each included trial are detailed in Table 10, which gives the outcome measure category, details of the specific tests used in each category, and timing of assessment.
Our primary outcome of performance in activities of daily living was measured by seven trials. The tools used were the Modified Barthel Index (Edmans 2000; Kang 2009; Lee 2021), Korean modified Barthel Index (An 2019; An 2020; Koo 2018; Yun 2018), and Edmans' ADL index (Edmans 2000).
The order of frequency of reporting other outcomes measures perception (11 trials), adverse events (six trials), quality of life and participation ‐ mobility and navigation (four trials) and eADL (one trial). No trials included measures of quality of life and participation ‐ social and participation ability, quality of life and participation ‐ QoL, or mental health and psychological well‐being.
In all six trials reporting adverse events, this was not a pre‐specified measure, but instead was reported in an ad hoc manner within the results. It was not possible to establish whether the data reflected the number of participants experiencing an adverse event during the trial or the number of adverse event occurrences, irrespective of the number of participants affected. As such, we chose not to include these data in any analyses; instead we presented the available information in a narrative format.
We identified that all five of the trials addressing the somatosensory disorder of Pusher Syndrome used measures of Pusher Syndrome severity. These included the Burke Lateropulsion Scale and the Scale for Contraversive Pushing. We extracted, analysed, and presented these data, as we considered they provided useful treatment effectiveness information.
Excluded studies
We have given details for 25 excluded reports, where inclusion and exclusion decisions were more difficult. Reasons for exclusion were primarily due to the absence of a perceptual disorder diagnosis (see Characteristics of excluded studies).
Four of the excluded trials were included in the Bowen 2011 version of this review (Dirette 1999; Hajek 1993; Mazer 2003; Taylor 1971). The adjusted criteria for our update required trial participants to have a perceptual disorder diagnosis. These four trials did not meet this criterion, and were excluded.
Ongoing studies
We identified 11 ongoing trials (CTRI201804013372; DRKS00021654; Mazer 2009; NCT02524015; NCT03154138; NCT03888326; NCT03991390; NCT04490655; NCT04703218; NCT04818073; NCT04911738) (see Characteristics of ongoing studies). Mazer 2009 was listed as an ongoing study in the first version of this review (Bowen 2011), and details of this completed trial remain unavailable. The remaining ongoing trials address disorders of smell perception (one RCT), somatosensation (seven RCTs), and vision (two RCTs).
Studies awaiting classification
Eleven trials identified are awaiting classification (Chiu 2020; Kim 2016; Kim 2020; Kitisomprayoonkul 2012; Koval'chuk 2011; Leer 1984; NCT04446273; Matz 2007; Morioka 2003; Muffel 2020; Weinberg 1982) (see Characteristics of studies awaiting classification). One of these, Weinberg 1982, was considered in the 2011 version of the review (Bowen 2011); it remains unclear whether the participants had a confirmed perceptual disorder.
Risk of bias in included studies
See Figure 2 and Figure 3 for an overview of our assessments.
Allocation
All included trials reported randomisation, but the method used was unclear for five trials (An 2019; Cho 2015; Choi 2018; Lincoln 1985; Seim 2021). The concealment of allocation was clearly stated by six trials (An 2020; Bergmann 2018; Carey 2011a; Chen 2012; De Bruyn 2018; Edmans 2000), including the use of sealed, opaque envelopes or a sealed box; however, this was unclear for the remaining trials.
Blinding
In complex rehabilitation trials, adequate blinding of clinicians delivering and participants receiving the intervention is a significant challenge. Blinding of outcome assessors is more achievable and was reported by many trials (13 RCTs) but was unclear for five trials (An 2019; Cho 2015; Kim 2015; Park 2015; Yun 2018). We judged six trials as having a high risk of bias (Carey 2011a; Chen 2012; Choi 2018; De Bruyn 2018; Edmans 2000; Kang 2009).
Incomplete outcome data
We judged three trials as having an unclear risk of bias in this domain (Chen 2012; Cho 2015; Lincoln 1985), with the remaining trials either including all participants in their analyses or accounting for all participants (with reasons given for dropouts) and an intention‐to‐treat analysis conducted. De Bruyn 2018 had a high risk of bias as three participants were excluded from both primary and follow‐up analyses.
Selective reporting
For 16 of the 18 included studies, outcome measures were well reported. One trial provided only partial data for three outcomes, as determined by comparing the published protocol and results article (De Bruyn 2018). The Carey 2011a trial failed to provide data for a secondary outcome, as determined by comparing the methods and results within one published article. We assessed both of these as having a high risk of bias in this domain.
Other potential sources of bias
Two trials had a high risk of bias. In Bergmann 2018, several participants had severe cognitive deficits and eight out of 38 participants were unable to complete a cognitive examination because of the severity of their cognitive or language impairments. Whilst no statistically significant baseline differences were found between the groups and no correlation was found between the cognitive examination and the outcome measure scores, the authors noted that these deficits might have influenced the participants' intervention response. The authors felt this was of particular relevance to the control group intervention because the intervention involved more explicit learning processes, plus the control group had more pronounced cognitive deficits.
In De Bruyn 2018, the mean age of the experimental group was significantly older than the control group, and with more right hemispheric lesions.
Comparability of groups at baseline was adequate for An 2020, Carey 2011a, Chen 2012, Choi 2018, Edmans 2000, Kang 2009, Kim 2015, Koo 2018, Lee 2021, Yang 2015, and Yun 2018. We assessed four trials as having an unclear risk of bias in this respect, either due to a lack of reporting of the baseline characteristics of participants (Cho 2015; Park 2015; Seim 2021), or due to difficulties "securing homogeneity" of participants, the nature of which was not fully explained (An 2019).
For Lincoln 1985, there was a change to eligibility criteria part‐way through the recruitment phase of the trial to include participants with left hemisphere strokes, subarachnoid haemorrhage, and head injury, due to slow recruitment. It is unclear what interim analyses were undertaken and what the decision‐making process was for continuation, adaptation, and eventual stopping of the trial. We assessed this trial as having an unclear risk of bias.
Effects of interventions
See: Table 1; Table 2; Table 3; Table 4
We present below the results for each sense (1) intervention versus no treatment or control comparisons; and (2) active intervention 1 versus active intervention 2 comparisons. Outcomes are only listed under each comparison if data were available in the included trials (i.e. there were no data available for any outcomes not listed below).
Hearing perception
No trials. See Table 1.
Smell perception
No trials.
Somatosensation perception
See Table 2. We identified seven trials. Two of these addressed general somatosensory disorders, Koo 2018 and De Bruyn 2018, while five addressed interventions for Pusher Syndrome: An 2019, An 2020, Bergmann 2018, Yang 2015, and Yun 2018. Following discussion amongst the members of the review team, the two groups have been presented separately, as either Pusher Syndrome or not Pusher Syndrome, in order to enable clinically useful synthesis of the data.
Intervention versus no treatment or control
Pusher Syndrome
No trials.
Not Pusher Syndrome
One trial with 24 participants compared transcranial direct current stimulation (tDCS) to sham treatment (Koo 2018).
####### Activities of daily living
Koo 2018 used the Korean modified Barthel Index. Analysis showed no difference between active intervention and control (mean difference (MD) 10.08, 95% confidence interval (CI) ‐2.47 to 22.63; P = 0.12; very low‐quality evidence; Analysis 1.1).
####### Quality of life and participation ‐ mobility and navigation
Koo 2018 used the Functional Ambulation Category measure. Analysis showed no difference between active intervention and control (MD 0.50, 95% CI ‐0.38 to 1.38; P = 0.27; very low‐quality evidence; Analysis 1.2).
####### Perception
Koo 2018 measured somatosensory perception (modified Nottingham Sensory Assessment) in a comparison of intervention with control, but the summary data were unreported (subscales only) and could not be included in this analysis.
####### Adverse events
Koo 2018 recorded adverse events, stating that “all the participants completed the stimulation sessions successfully without complaining about any discomfort during the procedure”.
Active intervention 1 versus active intervention 2
Pusher Syndrome
Five trials addressed this An 2019, An 2020, Bergmann 2018, Yang 2015, and Yun 2018. Trials compared an intervention training posture and/or movement using computerised, game‐based exercises, supported by assistive equipment such as balance boards, treadmills or harnesses, to conventional physiotherapy for Pusher Syndrome. Interventions in both groups were considered similar, and we combined the data in meta‐analyses.
####### Activities of daily living
Three trials (80 participants) used the Korean modified Barthel Index (An 2019; An 2020; Yun 2018). Intervention 1 (computerised balance and movement training) was more beneficial than intervention 2 ("standard" Pusher Syndrome physiotherapy) (MD 10.19, 95% CI 4.94 to 15.44; P < 0.001, I^2^ = 0%; very low‐quality evidence; Analysis 2.1).
####### Quality of life and participation ‐ mobility and navigation
One trial (30 participants) was included, which used the Performance‐Oriented Mobility Assessment‐B (POMA‐B) outcome measure (Bergmann 2018). Analysis showed no evidence of a difference between interventions 1 and 2 (MD 1.00, 95% CI ‐1.51 to 3.51; very low‐quality evidence; Analysis 2.2).
####### Perception
One trial (30 participants) was included, which assessed perception of subjective visual vertical (Bergmann 2018). There was no evidence of a difference between interventions 1 and 2 (standardised mean difference (SMD) 0.52, 95% CI ‐0.21 to 1.25; very low‐quality evidence; Analysis 2.3).
####### Adverse events
Three studies reported on adverse events; in each case, none were reported (An 2020; Yang 2015; Yun 2018).
####### Pusher Syndrome outcomes
Four trials (86 participants) were included; they used the Burke Lateropulsion Scale (An 2019; An 2020), or the Scale for Contraversive Pushing (Bergmann 2018; Yang 2015). Analysis showed a tendency for intervention 1 (computerised balance and movement training) to be more beneficial than intervention 2 (standard Pusher Syndrome physiotherapy) (SMD 1.03, 95% CI 0.33 to 1.73, P = 0.004, I² = 50%; very low‐quality evidence; Analysis 2.4).
Not Pusher Syndrome
One trial with 36 participants addressed this comparison, comparing the SENSe intervention, consisting of sensorimotor therapy and sensory discrimination tasks, with cognitive table‐top games and motor exercises (De Bruyn 2018).
####### Perception
De Bruyn 2018 used the Nottingham Sensory Assessment. Analysis showed there was no evidence of a difference between intervention 1 and intervention 2 (SMD ‐0.38, 95% CI ‐1.04 to 0.28; very low‐quality evidence; Analysis 2.3).
Touch perception
See Table 3. We identified three Kim 2015, Lee 2021, and Seim 2021. We were able to analyse data from Kim 2015 and Lee 2021; Seim 2021 did not report any outcomes of interest.
Intervention versus no treatment or control
Kim 2015 compared two active interventions with no treatment.
Quality of life and participation ‐ mobility and navigation
For this outcome, we included one trial with two relevant comparisons (30 participants), both using the Timed Up and Go Test (Kim 2015). Analysis showed there was no difference between active intervention and no treatment (MD 6.50, 95% CI ‐4.81 to 17.81; P = 0.26, I^2^ = 50%; Analysis 3.1). Due to a number of methodological concerns, we judged there was insufficient evidence to support a conclusion based on these data.
Perception
We included one trial with two relevant comparisons (30 participants), both using a dynamometer‐based measure of proprioception (Kim 2015). Analysis showed a tendency for the active intervention to be more beneficial than no treatment (MD 4.64, 95% CI 3.06 to 6.21; P < 0.001, I^2^ = 0%; very low‐quality evidence; Analysis 3.2).
Active intervention 1 versus active intervention 2
Two trials addressed this Lee 2021 compared robot glove‐based hand exercises to hand exercises alone, and Kim 2015 compared pressure sense perception training on a stable surface to an unstable one (balance board). As the interventions were quite different in nature, where more than one trial measured the same category of outcome measure, we did not combine data in a meta‐analysis.
Activities of daily living
We included one trial (24 participants), which used the Modified Barthel Index (Lee 2021). This trial demonstrated that there was no difference between the interventions (MD ‐0.41, 95% CI ‐12.31 to 11.49; very low‐quality evidence; Analysis 4.1).
Quality of life and participation ‐ mobility and navigation
We included one trial (20 participants), using the Timed Up and Go Test (Kim 2015). This trial provided evidence that intervention 2 (training on the unstable balance board) was more beneficial than intervention 1 (training on the stable balance board) (MD ‐11.60, 95% CI ‐19.50 to ‐3.70; very low‐quality evidence; Analysis 4.2).
Perception
We included two trials (44 participants): one assessed proprioception using a dynamometer (Kim 2015), and the other used the kinaesthetic subtest of the revised Nottingham Sensory Assessment (rNSA) (Lee 2021). We did not combine data from these two trials for individually, neither trial provided evidence of a difference between interventions 1 and 2 (Analysis 4.3).
Adverse Events
One trial collected data on adverse events, reporting “no safety concerns or adverse events" (Lee 2021).
Taste perception
No trials.
Visual perception
See Table 4. We identified seven relevant Chen 2012, Cho 2015, Choi 2018, Edmans 2000, Kang 2009, Lincoln 1985, and Park 2015.
Intervention versus no treatment or control
One trial addressed a comparison of intervention with no treatment, comparing neurofeedback training to no treatment (Cho 2015), and one trial addressed a comparison of intervention with control, comparing perceptual training to a conventional (not perceptual) therapy (Lincoln 1985).
Extended activities of daily living (eADL)
We included Lincoln 1985 (33 participants), which used the Rivermead ADL scale. This trial showed there was no evidence of a difference between active intervention and control (MD 0.94, 95% CI –1.60 to 3.48; P = 0.47; very low‐quality evidence; Analysis 5.1).
Perception
Cho 2015 (27 participants) assessed perceptual outcomes using the Motor Free Visual Perception Test (MVPT), and Lincoln 1985 measured perception using the Rivermead Perceptual Assessment battery. As Lincoln 1985 only reported subscale scores for the Rivermead Perceptual Assessment battery, we did not include these data in the analysis. Data from Cho 2015 showed no difference between active intervention and no treatment (MD ‐1.75, 95% CI ‐5.39 to 1.89; P = 0.35; very low‐quality evidence; Analysis 5.2).
Active intervention 1 versus active intervention 2
Five trials addressed this Chen 2012, Choi 2018, Edmans 2000, Kang 2009, and Park 2015. The interventions were very dissimilar, including OT‐led training in practical tasks, paper‐based repetition exercises, and computer‐based games; thus, statistical pooling of data was inappropriate.
Activities of daily living
Two trials (96 participants) were included; both used the Modified Barthel Index (Edmans 2000; Kang 2009). We did not combine data due to differences in the interventions. We judged the evidence from each included trial to be of very low quality (Analysis 6.1).
Quality of life and participation ‐ mobility and navigation
One trial (28 participants) used the 10‐Metre Walk Test to assess mobility (Choi 2018). It compared Wii Fit virtual reality training to general balance training for visual perceptual disorders. This trial showed no evidence of a difference between the two interventions (MD ‐0.12, 95% CI ‐13.62 to 13.38; very low‐quality evidence; Analysis 6.2).
Perception
We included five trials (163 participants) (Chen 2012; Choi 2018; Edmans 2000; Kang 2009; Park 2015). They used a range of perception outcome the Modified Taylor Complex Figure (Chen 2012), the MVPT (Choi 2018; Kang 2009; Park 2015), and the Rivermead Perceptual Assessment Battery (Edmans 2000). We did not pool data due to differences in the interventions, but displayed them as standardised mean differences (Analysis 6.3). We judged the evidence from each included trial to be of very low quality.
Discussion
Summary of main results
Our review update explored evidence of the effectiveness of interventions for perceptual disorders after stroke. It included disorders of hearing, smell, somatosensation, taste, touch, and visual perception. The main comparisons of interest were active interventions versus no treatment, and active interventions versus control (placebo, standard care, attention control). Our primary outcome measure was participants' performance in activities of daily living.
We included 18 trials (541 participants; 535 with stroke). They addressed somatosensation (seven trials, 196 participants), touch (three trials, 70 participants), vision (seven trials, 225 participants), and mixed perceptual disorders (one trial, 50 participants). There is insufficient evidence to determine the effectiveness of any one intervention for any sensory modality, nor the effect of one intervention relative to another. Only adequately‐sized trials, using relevant outcome measures and with appropriate pretrial development studies, will provide answers to these clinically important questions.
Interventions for hearing perception disorders
We found no RCTs exploring the effectiveness of any intervention for hearing perception disorders in people after stroke.
Interventions for smell perception disorders
We found no RCTs exploring the effectiveness of any intervention for smell perception disorders in people after stroke.
Interventions for somatosensation perception disorders
We found no evidence of a difference between intervention and control for ADL and navigation and mobility outcomes; this evidence is of very low quality. We found no difference between two active interventions for somatosensory perception dysfunction (not Pusher Syndrome) on perception outcomes; this evidence is of very low quality.
For Pusher Syndrome, we found no difference between game‐based posture training and standard physiotherapy for measures of mobility and navigation and perception. Game‐based posture training may be more beneficial than standard physiotherapy for improving ADL and measures of Pusher Syndrome severity. All evidence was of very low quality.
Interventions for taste perception disorders
We found no RCTs exploring the effectiveness of any intervention for taste perception disorders in people after stroke.
Interventions for touch perception disorders
We found no difference between intervention and no treatment for navigation and mobility outcomes, but there may be a beneficial effect of active intervention on perceptual function; this evidence is of very low quality.
Evidence relating to one intervention versus another was varied, and insufficient to draw generalisable conclusions.
Interventions for visual perception disorders
We found no difference between intervention and no treatment on measures of perception; this evidence was of very low quality. There was no difference between active intervention and control for measures of eADL.
We identified some data for outcomes of ADL, navigation and mobility, and perception from trials comparing one intervention to another. Due to differences in the interventions and comparisons conducted, we did not pool data for statistical analysis, and it was not possible to draw generalisable conclusions.
Key findings of this review
Perceptual disorders have a significant impact on the daily lives of many stroke survivors, but there is an absence of RCT evidence to inform clinical practice for hearing, taste, and smell disorders after stroke. For vision, touch, and somatosensation, the data provide very low‐quality evidence, and are insufficient to support generalisable conclusions about the effectiveness of any one intervention. To date, there is insufficient evidence from adequately powered, appropriately designed, and well reported RCTs to reach clear conclusions.Interventions identified primarily used a rehabilitative approach, aiming to restore participants' impaired perceptual function. The nature of the interventions varied, which limited our ability to pool data and reach generalisable conclusions.The quality of the evidence relating to all comparisons and outcomes was very low.Trialists did not report the involvement of people with perceptual disorders in the design, delivery, or interpretation of the included trials.No trials addressing the main comparisons of intervention versus no treatment or control measured our primary outcome (performance in activities of daily living) at a follow‐up time point.
Overall completeness and applicability of evidence
Participants
Most participants had stroke, with mean age ranging from 48.8 to 75.5 years, and results are therefore generalisable to this population. The time point of intervention delivery varied from 19 days to 4.3 years after stroke. As the appropriate timing, treatment, and intervention dose may vary by stroke chronicity (Bernhardt 2017), this limits generalisability and confidence in the findings.
Our systematic search did not identify any trials that addressed stroke‐related perceptual disorders of hearing, taste, or smell. Whilst each sense provides information about the environment, the processing and understanding of this information varies hugely across the senses. Consequently, findings in one sense are not transferable, and cannot be used to inform treatment of disorders in another.
The perceptual disorders addressed in the trials of interventions for touch and vision varied. Participants recruited to trials of interventions for touch disorders experienced dysfunction of both the upper and lower limbs. The tests used to determine and describe participants' visual perception did not provide detailed information about the specific visual skills impaired (e.g. Rivermead Perceptual Assessment Battery, MVPT) and thus the wider applicability of results is unclear. Within somatosensory disorders, there was a subset of included trials that recruited stroke survivors with Pusher Syndrome, and the findings therefore appear generalisable to this population.
Methods
The included trials recruited between 11 and 80 participants, with limited reporting of the use of a priori power calculations. As smaller trials provide less precise estimates of effect, this is a consideration for the trials included in this review.
Interventions
We considered to what extent the interventions included in this review reflect current practice in the regions where the studies were conducted.
Included trials originated from Asia, Australia, Europe, and North America. Many interventions such as OT‐led perceptual training for visual disorders, Pusher Syndrome physiotherapy, and computerised visuo‐cognitive training appear to reflect known treatment for perceptual impairments. Other approaches may be less available. The use of tDCS is more limited, often only available through clinical trials or limited (fee‐paying) service providers (Schjetnan 2013). According to our clinical specialists, the highly technical, equipment‐dependent training for Pusher Syndrome, such as the Lokomat or Spineboard 3D tools, are not part of routine clinical care. The analyses relating to these trials may therefore have limited clinical relevance and applicability.
Comparisons
Of the 20 comparisons included in this review, 14 compared two active interventions. Six included a comparison to either no treatment or a placebo, standard care, or attention control intervention. The included interventions varied; in many cases, we did not consider it appropriate to pool data.
Outcomes
There was limited reporting of performance in activities of daily living, our primary outcome measure. This measure was utilised by seven trials, often using the modified Barthel Index (standard or Korean version). The most frequently reported outcome of interest was perceptual function, with a huge variety of tools both within and across senses. Few trials included the outcomes that would show the transfer of any intervention effect beyond change in perceptual ability to impact on everyday life, such as mental health and psychological well‐being, or participation and quality of life. Given that our stakeholder group members identified and prioritised these outcomes, this suggests that researchers should consider their choice of outcomes, so that trials can provide more meaningful data to stroke survivors and healthcare professionals. It was also clear that outcomes were most frequently measured immediately after the intervention; there was no evidence of the effects of an intervention on our primary outcome at any other time point, resulting in an absence of evidence on the longevity of any effect.
Adverse event data were inconsistently reported, and we could not include them in any statistical analyses.
It is unclear why there are few RCTs exploring interventions for perceptual disorders following stroke. Our scoping review of interventions for perceptual disorders after stroke included all study designs and highlighted that there is a paucity of all study designs on the topic, not just RCTs (Hazelton 2022). The causes of a lack of intervention research for perceptual disorders are likely multifactorial, potentially linked to two fundamental issues.
First, there is limited awareness and understanding of perceptual disorders in stroke, and thus there is a need for further research. This covers lack of awareness of the presence, nature, frequency, and impact of perceptual disorders, in both stroke survivors, families, and healthcare professionals delivering care (Bamiou 2015; Dutta 2013; Falkenberg 2020). This is compounded by a lack of validated tools to identify perceptual disorders and poor epidemiological data to demonstrate that these are lasting problems worthy of research attention (Colwell 2021; Koohi 2019; Pollock 2011a; Rowe 2015). There is also limited guidance and training in this field (Pollock 2011a). It is also possible that in those with severe stroke, perceptual impairments may be masked by other conditions (e.g. cognitive disorder/aphasia) which may preclude formal assessments and limit self‐report.
Second, there are difficulties in conducting research in this field. As we can see from the review, most interventions are rehabilitative in nature. Clinical rehabilitation services may not be well set up to do research, often involving disparate groups of professionals working in different services (NICE 2013). Clinical research needs critical mass and leadership, which is potentially lacking in this field. The specific interventions are also ‘complex’ in nature. There are a range of known challenges to conducting trials of complex interventions, which include fully understanding the mechanisms of action, standardising the interventions, population included and context of delivery, achieving required recruitment, and retention rates and choice of appropriate outcome measures (Datta 2013; Richards 2015; Skivington 2021; Tarquinio 2015). All of these are directly applicable to research into perceptual disorders in stroke, given the complex interventions, population heterogeneity, and variation in delivery setting across acute and rehabilitation services, forming clear barriers to the conduct of trials. There is additionally potential difficulty in achieving the funding needed to support appropriately powered, complex rehabilitation trials.
Quality of the evidence
We judged the quality of evidence using the GRADE approach and found the evidence included within the meta‐analyses to be of very low quality. Key factors contributing to downgrading of the evidence within these comparisons included risk of bias, imprecision, and indirectness.
Risk of bias
We identified methodological concerns in the majority of included trials, with insufficient details available from published reports. We judged six of the 18 included trials to be at low risk of bias for allocation concealment, and none to be at low risk of bias for blinding of outcome assessment. Other potential sources of bias included poor reporting of group differences at baseline.
Imprecision
The number of participants within the included trials was small, ranging from 11 to 80 participants, with only two of the 18 trials including 50 participants or more. Across our review, we included 541 randomised participants, but trial variations limited the pooling of data which, in turn, limited the conclusions that could be drawn based on the evidence.
Indirectness
Interventions: our review synthesised the available evidence relating to a wide range of different interventions for perceptual disorders after stroke. We found variations in the interventions evaluated, both between senses, and within individual senses.Outcomes: across included trials, there was inconsistency in the outcomes assessed by individual trials, and the outcome measurement instruments used. Notably, some of the outcomes that our stakeholder group identified as priorities, including extended activities of daily living and mental health and psychological well‐being, were not reported by any of the included trials. The variations in outcome measure instruments limited the ability to pool data from individual trials in a meaningful way, and where measures were pooled, limited our certainty in the result.
In summary, we judged the quality of the evidence synthesised within this review to be very low, and this limits our confidence in the results. Currently, there is insufficient high‐quality evidence to support any generalised conclusions about the effectiveness of any specific interventions for perceptual disorders after stroke. Future research needs to be designed to enhance trial quality and limit risk of bias, in order to produce results which are clinically useful and meaningful to people with perceptual disorders after stroke.
Potential biases in the review process
Publication bias
We conducted a thorough systematic search, which included multiple literature databases, trials databases, and grey literature sources, in an effort to identify all relevant published trials. Despite our efforts, there remains a risk that a relevant trial may have been missed in our screening process (especially at title screening stage, which was the responsibility of one review author). Where uncertainties arose in the screening process, we contacted the trialists for further information; we thank all those who provided valuable information at this stage.
Categorisation of interventions
We categorised interventions using two main first, the sense targeted by the intervention, and second, the intervention approach (rehabilitation, non‐invasive brain stimulation, pharmacological, surgery, screening and assessment). We further categorised rehabilitation interventions by whether they aimed to restore lost function, compensate for lost function, or use an external device or modification to substitute for that lost function. Two independent review authors classified all interventions, involving a third review author where discrepancies could not be resolved through discussion. While alternative approaches to intervention classifications exist, our approach was derived from perceptual disorder research by Kerkhoff 2000 and has been used in other Cochrane Reviews in associated disorders (Chung 2013; Pollock 2019b). The consistent use of a pre‐planned, transparent, clinically‐relevant approach to categorisation, applied by independent review authors, is a strength of this review.
Categorisation of comparisons
Comparisons in this review related to one of three active intervention, no treatment, or control (attention control/standard care/placebo). In some cases, our categorisation of an intervention differed from that reported by the trialists; for example, a physiotherapy intervention specific to Pusher Syndrome functioned as a control intervention in one trial (compared to robot‐assisted gait training) (An 2020). For our purposes, however, we classified this as an active intervention. Two review authors independently performed categorisation, and consulted a third author where agreement could not be reached by discussion. We have clearly stated the definitions we used in this process (Types of interventions).
Inclusion participants
Defining perception is a challenge; application of that definition consistently across senses and trials is also challenging. Our review team, in collaboration with our stakeholder group, discussed, agreed, and voted on key issues relating to definitions in advance of undertaking this review update. These issues
the definition of we used the WHO ICF definition of “the specific mental functions of recognizing and interpreting sensory information” (World Health Organization 2001);the senses hearing, taste, touch, smell, somatosensation, and vision;the disorders included and excluded. We excluded balance and pain. We included one specific disorder, Pusher Syndrome, as we considered this a disorder of somatosensory perception.
At times, the lack of standardised terminology and reporting limitations resulted in challenging judgements relating to perceptual disorders. On occasion, it was difficult to reach consensus on whether disorders were perceptual, cognitive, sensory, or motor in nature, or a combination of these. In each case, we returned to our definition to provide clarity and support our decision making. We recognise other review authors may have made different decisions, and that some inadvertent exclusion of potentially relevant trials may have occurred. We are confident that our overall review conclusions are unlikely to have been impacted by the exclusion of those trials. A strength of our approach was the expertise and experience in conducting systematic reviews of complex interventions delivered to highly heterogeneous participant groups, working in partnership with clinical experts.
Our review included participants with a perceptual disorder. Thus, we excluded from this review trials that evaluated an intervention addressing perceptual function in a generalised stroke population (who may or may not have had a relevant disorder). The previous version of this review took an alternative approach (Bowen 2011).
Outcomes
Our review's primary outcome was participants' performance in activities of daily living. We identified seven trials that measured ADL. From our secondary outcomes of interest, eADL, quality of life and participation ‐ mobility and navigation, perception, and adverse events were reported, while quality of life and participation ‐ social activities and participation, QoL, and mental health and psychological well‐being outcomes were absent. The outcome measures sought in this review update were identified and prioritised by the research team alongside our stakeholder group members, to ensure their relevance to stroke survivors and to understand the impact of perceptual disorders on their daily lives. We also noted measures of sensation, motor function, and cognition, but excluded these from our analysis. Classification of outcome measurement instruments rarely presented challenges. Where uncertainties arose, we consulted topic experts, specifically in relation to navigation and mobility outcomes versus those of motor function, and sensation versus perception in relation to touch or somatosensory function.
Given the breadth of senses included and the international trial activities we identified, we encountered several outcome measurement instruments not included in our predefined lists. In order to determine which of these should be included in analysis, we sought data on the tool development and prioritised those with the highest validity and reliability measures. We confirmed these decisions with a topic expert in the relevant field.
Only six out of 18 studies mentioned the occurrence of adverse events; in each case, none were reported. It was unclear whether collection of adverse event data was planned as part of the studies' methods and therefore how thorough the reporting process was. The exact nature of adverse event data collection was also unclear, meaning these data were difficult to interpret. Thus, we did not meta‐analyse them. Adverse event data are known to often be handled with less rigour than other outcome measures (Peryer 2022), but they are important to consider within a review, to determine potential harms arising from an intervention and guide clinical practice. RCT data can provide important safety data, and standardised terminology is in place to support clarity of reporting relating to the nature and possible aetiology of adverse events.
Agreements and disagreements with other studies or reviews
Agreements and disagreements with the previous version of this review
The 2011 version of this review identified and included six trials (338 participants), each using sensory stimulation to address visual perceptual disorders. The authors concluded that "there is insufficient evidence to support or refute the view that perceptual interventions are effective" (Bowen 2011).
Our update changed the scope of this review
limiting it to a stroke population, excluding those with other neurological impairments;including all interventions, not only those classified as “non‐pharmacological”;broadening the scope of the searches, to make sure that disorders across all six senses were adequately addressed;extending the dates of the search to August 2021.
Our review included a larger number of trials and participants, across three senses rather than just vision, and with a larger range of intervention approaches. However, our conclusions remain there remains insufficient evidence to reach generalised conclusions about the effectiveness of interventions for perceptual disorders.
Agreements and disagreements with other reviews
Several systematic reviews (Cochrane and non‐Cochrane) are of relevance to our review and relate to vision, touch, and few reviews of the literature consider hearing, taste, and smell perceptual disorders due to stroke.
For vision, several Cochrane Reviews address related (but not overlapping) disorders, focused on sensory/sensorimotor function (Pollock 2011b; Pollock 2019b), or visual attention (Longley 2021). These may form a natural complement for clinicians who manage a range of disorders of the visual system. Reviews of interventions for visual perception also exist, typically focusing on interventions for visual neglect. Jutai 2003 conducted a "critical review and synthesis" of eight studies (six RCTs) of visual perceptual disorders following stroke, including both spatial neglect and apraxia. In contrast to our findings, they concluded there was "strong evidence" that specific treatment of perceptual disorders improved perceptual functioning based on summarising the original studies’ findings as three positive, one negative, and one mixed. Our systematic review differed in methods and conclusions from the Jutai 2003 review and had just two studies in common (Edmans 2000; Lincoln 1985). While we share the Jutai 2003 conclusion that no one intervention approach has proven efficacy over any other, based on our data, we concluded that there is insufficient evidence to support decisions relating to the effectiveness of specific treatments for visual perceptual disorders.
Cicerone 2005 updated a systematic review of the effectiveness of cognitive rehabilitation for people with traumatic brain injury or stroke to 2002. All studies identified considered neglect rehabilitation or interventions for visual field loss, making no recommendations relevant to the topic of this review. Similarly, Hanna 2017 included visual perception, but with a focus on visual neglect, with no studies in that review meeting our inclusion criteria.
In relation to touch and somatosensation perception disorders, a number of systematic reviews have addressed this topic. Doyle 2010 reviewed interventions for sensory impairment of the upper limb after stroke, including RCTs that addressed “any impairment that impacted on sensory registration, perception, or discrimination,” with a focus on the effect on measures of sensation. Doyle 2010 included a total of 13 studies, with no overlap in those included here; they found insufficient evidence to determine intervention effectiveness. Schaburn 2009 reviewed retraining of sensation after stroke, and this was expanded in 2019 to include 38 RCTs (Serrada 2019), conducting meta‐analysis involving 13 comparisons. The review explored passive, active, or combined sensory training, aiming to quantify the effect on impairment and function. Whilst they do not explicitly state the definition of sensation they applied, the background section describes it as “the ability to accurately perceive and discriminate sensations of pain, temperature, pressure and vibration, as well as the ability to accurately locate body parts in space (proprioception)”. They found limited data, with some evidence to support the use of passive sensory training. There were clear differences in the inclusion criteria for the Schaburn 2009 and Serrada 2019 reviews compared to our Cochrane Review, including (1) nature of the intervention, (2) body part considered, and (3) nature of the disorder addressed. This review required a clear diagnosis of a post‐stroke perceptual disorder, resulting in limited overlap of included studies and comparability of findings.
Only one review, with a very limited search strategy, unclear method of synthesis, and lack of quality appraisal, addressed Pusher Syndrome interventions (Thanaya 2019). The researchers identified 10 studies, exploring robot‐assisted gait training, visual feedback, galvanic vestibular stimulation, and physiotherapy interventions. It identified two RCTs, both of which were included here (Bergmann 2018; Yun 2018), and also concluded that further RCTs in this field were required.
Authors' conclusions
Implications for practiceCurrently, there is insufficient evidence to support or refute the view that any specific intervention is effective at reducing the impact of impaired perception after stroke. More evidence is required before decisions are made on the provision of these services. Absence of evidence is not evidence of absence; but at present it is not possible to make recommendations regarding specific interventions for specific senses nor specific disorders.
Implications for researchFuture randomised controlled trials provide a sufficiently detailed theoretical rationale for, and description of, the interventions, including type and dose, to allow implementation into clinical practice and research replication;provide a standard care control group, carefully documenting the content and amount of standard care, which can be highly variable;include detailed diagnostic information on individuals' perceptual problems, given the heterogeneity in perceptual problems in terms of type, severity, and likely impact on everyday function;endeavour to conduct trials with a low risk of bias through rigorous methodological development and reporting (e.g. ensure allocation concealment; attempt to blind outcome assessors and report the success or failure of blinding; report all loss to follow‐up; report results from all outcome measures; control for other possible sources of bias);be of sufficient size to have adequate statistical power to answer clinically important questions about long‐term functional outcomes;ensure that key outcomes, such as activities of daily living, psychosocial benefits, and quality of life, are used;adopt an intention‐to‐treat approach to measurement of outcomes in all individuals as well as to analysis of measured outcomes by treatment group.Guidance supporting stroke rehabilitation intervention development and trials of their effectiveness should be used (Bernhardt 2019), to ensure the appropriate research questions and study designs are used in any pretrial as well as randomised controlled trial work, and in turn to reduce research waste and address the clinical uncertainties relating to perceptual disorders after stroke. We further suggest the development of standardised terminology for perceptual disorders, to aid clarity of reporting and understanding for researchers, clinicians, and stroke survivors, across all the senses.
What's new
DateEventDescription20 December 2021New citation required and conclusions have changedConclusions changed.20 December 2021New search has been performedThe previous 2011 version of this systematic review included participants who experienced perceptual disorders as a result of stroke or other non‐progressive brain injuries. It included six trials (339 participants). This 2022 review focuses more specifically on stroke survivors while expanding the inclusion criteria to potentially include all interventions to address perceptual disorders (see Differences between protocol and review for full details of changes). The review now includes 18 trials and 541 participants, of whom 535 (98.9%) were stroke survivors. We excluded four trials from the 2011 review as a consequence of the narrower participant inclusion criteria.
History
Protocol first Issue 2, 2008 Review first Issue 4, 2011
DateEventDescription9 May 2008AmendedConverted to new review format.