Rehabilitation of Cognitive Deficits Poststroke: Systematic Review and Meta-analysis of Randomized Controlled Trials

Systematic Review and Meta-Analysis of Randomized Controlled Trials

Mairead O'Donoghue, BSc; Siobhan Leahy, PhD; Pauline Boland, PhD; Rose Galvin, PhD; John McManus, MD; Sara Hayes, PhD


Stroke. 2022;53(5):1700-1710. 

In This Article


Study Design

This systematic review and meta-analysis were conducted in accordance with the Preferred Items for Systematic Reviews Meta-Analyses statement.[28] The review protocol was registered with the International Prospective Register of Systematic Reviews (; Unique identifier: CRD42019125289) and has been published previously.[29] Data available on request from the authors.

Eligibility Criteria

Types of Study. Randomized controlled trials and quasi-randomized control trials, as defined by the Cochrane Handbook for Systematic Reviews of Interventions were included.[30] Feasibility studies and pilot studies were also included. Studies published in the English language with full text available were included.

Participants. Studies that included adults aged 18 years or older with a clinical diagnosis of ischemic or hemorrhagic stroke, in the acute, subacute, or chronic stage poststroke were included. As outlined in the review protocol, people poststroke with a confirmed cognitive impairment were to be included. However, the current review includes individuals poststroke with or without a confirmed cognitive impairment. Mixed cause studies (eg, traumatic brain injury and stroke mix) were included only if separate data for people poststroke was extractable.

Interventions. Interventions of which the primary or secondary aim is to improve cognitive function after stroke were included. Interventions of any duration or conducted at any time since stroke were included. Pharmacological interventions (including over-the-counter medications) were excluded.

Controls. Studies comparing an intervention that affects cognitive function poststroke with standard care, no treatment control, waitlist control, or active control were included.

Outcomes. Change in cognitive function postintervention in individuals poststroke was the primary outcome of interest. Outcome measures that focused on general cognitive status, or ≥1 domain-specific aspects of cognition such as executive function, attention, memory, perception, limb apraxia, and neglect in line with Australian Clinical Guidelines for Stroke (2020) were included. Cognitive outcome measures related to cognitive domains other than those listed were excluded. Cognitive screening tools were also included. Secondary outcome measures included quality of life, using stroke-specific or generic quality of life measures, functional abilities, mobility, balance, and depression.

Search. The following databases were searched: PubMed, Embase, CENTRAL, PsycInfo, and CinAHL. The Vista and databases were also searched for potentially eligible ongoing clinical trials. The search was carried out from inception to October 2019. For details of the search strategy, see the Supplemental Material.

Screening. The searches were exported and saved in a master reference management library (Clarivate Analytics Endnote X7) and duplicates were removed. Titles, abstracts, and full texts were screened independently by 2 review authors (M.O.D. and S.H., P.B., or R.G.). Any disagreements regarding full-text eligibility were resolved by discussion among all authors.

Data Extraction. Data, including author, study design, sample size, population characteristics (age, type of stroke, time since stroke), intervention characteristics (intervention type, intervention content, duration of intervention, setting of intervention), control group (standard care, waitlist control or active control), primary and secondary outcomes postintervention were extracted and entered into a standardized recording data extraction form. Study authors were contacted for missing data or further information. Data were recorded as mean and SD where possible for studies at postintervention and follow-up.

Risk of Bias. Two authors (M.O.D. and either S.H., S.L., or R.G.) independently assessed the validity of studies using the Cochrane Risk of Bias Tool. Bias was assessed as low, unclear, or high for individual components from 5 domains: selection bias, performance bias, attrition bias, reporting bias, and other bias. Disagreements were resolved by group consensus.

Data Synthesis. Separate analyses were performed for trials comparing cognitive interventions with a standard care control, or with a waitlist control intervention, and trials comparing 2 active interventions. Interventions were classified into 6 categories for analysis as follows: multiple component interventions, cognitive rehabilitation interventions, physical activity interventions, noninvasive brain stimulation protocols (NIBS), occupational-based interventions, and other interventions. For details on the categorization of these interventions, please see the Supplemental Material.

Meta-analyses were conducted using the Cochrane Review Manager software (RevMan 5.3). For continuous data, the treatment effect was calculated using standardized mean differences (SMD) and 95% CI where different studies used different scales to assess the same outcome and using mean differences (MD) and 95% CI where studies used the same scales to measure relevant outcomes. Analyses are presented according to the type of intervention compared with a control group (standard control, active control, or waitlist control). Where studies presented data from multiple intervention/control groups, all relevant interventions/control groups were combined into a single group and analyzed accordingly.[30]

Subgroup analysis was performed where pooled studies included participants that were <3 months poststroke, between 3 and 6 months poststroke, and >6 months poststroke. We previously outlined that a narrative synthesis would be provided on studies not included in the meta-analysis.[29] However, due to the number of studies identified and range of outcome measures used, the characteristics of all studies (n=64) were reported but a narrative synthesis of studies that did not provide data for the meta-analysis was not performed.

Outcomes are listed according to general cognitive functioning and the following cognitive domains: attention, memory, executive function, perception, apraxia, neglect. The impact of heterogeneity on results was assessed using the I2 statistic. When the I2 was <50% there was little concern about statistical heterogeneity.[30] Random-effects models were used where there was statistical heterogeneity≥50%.[30]