Types of studies

We will include randomised controlled trials (RCTs) in the review. We will also include clinical trials where participants are quasi‐randomly assigned to one of two or more treatment groups. Cross‐over trials will be considered as RCTs according to The Cochrane Collaboration's guidelines (Higgins 2005).

Types of participants

Participants will be adults (aged 18 years or over) with clinically defined stroke and confirmed cognitive impairment as specified in the trial. We will exclude trials where the focus of the intervention was on improving language skills or perceptual skills, or both. We will exclude trials with mixed aetiology groups unless participants who have had (and only had) a stroke comprise more than 50% of the participants. We will include these trials only when data are either provided separately for participants with stroke in the published article or are available from the trial authors.

Types of interventions

We will include all occupational therapy interventions for cognitive impairment in people with stroke. Occupational therapy interventions in this review are defined as interventions indexed in major international occupational therapy texts (Katz 2005; Pedretti 2001; Trombly 2002). Furthermore, if papers reporting an intervention were authored by an occupational therapist or the intervention in the study was administered by occupational therapists, or both, we will also include these. These interventions may take either a remedial or a compensatory approach, or both. The remedial approach focuses on training specific cognitive deficits using media such as pencil and paper, computer tasks, and board games. In a compensatory approach, interventions may include (1) training skills for daily activities (e.g. dressing, ambulation, driving, managing a meal) and vocation using compensatory strategies; (2) advising and educating about the use of assistive devices that aid cognitive function, such as an alarm watch, a hand‐held computer, or a medication container; and (3) educating patients, families, and caregivers about strategies to overcome patients' cognitive impairment. The dynamic interactional approach (previously referred to as multicontextual) is an integrated approach, encompassing both remedial and compensatory elements to encourage generalisation of the treatment effect achieved in a clinical setting to patients' real life performance situation (Toglia 2005). We will consider the dynamic interactional approach as a third type of intervention in this review, separate from remedial and compensatory approaches. We will not include trials examining drug effects on cognitive function following stroke.

Types of outcome measures

The primary outcome measure will be assessments of basic activities of daily living. We will consider assessments of instrumental activities of daily living, community integration, resumption of life roles, and specific cognitive functions, such as attention and memory or general cognitive function, as secondary outcome measures. We will describe differences in adverse outcomes (such as death) between the treatment groups.

Study selection

One review author (CK) will review the titles identified in references and searches and eliminate obviously irrelevant studies. We will obtain the abstracts of the remaining studies. Using the titles and abstracts obtained from the searches, two review authors (CK and TH or SB) will independently complete the first study selection according to the four inclusion criteria (types of studies, participants, interventions, and outcome measures). The first study selection will result in the categories of included, excluded, or unsure. The full texts of the studies that are marked as included or unsure will be obtained and two review authors (CK and TH or SB) will independently complete the second study selection to finally decide on each trial's inclusion or exclusion. We will resolve disagreement by discussion based on the inclusion criteria. If no consensus is reached, a third review author will arbitrate.

Assessment of methodological quality

We will present the included trials in tabular form to summarise their methodological quality. There are four sources of potential bias in trials of intervention effectiveness: selection bias, performance bias, detection bias, and attrition bias. According to the definition by Juni et al, selection bias refers to biased allocation of participants to comparison groups (i.e. absence of, or inadequate, allocation concealment) (Juni 2001). We will assess the method of allocation concealment of the selected studies according to three categories using the criteria suggested in the Cochrane Handbook: A ‐ adequate; B ‐ unclear; and C ‐ inadequate (Higgins 2005). Performance bias refers to unequal provision of care apart from the treatment under evaluation (i.e. lack of blinding of therapists or participants, or both). Detection bias refers to biased assessment of outcome (i.e. lack of blinding of outcome assessors). Attrition bias refers to biased occurrence and handling of deviations from the treatment protocol (i.e. lack of analysis according to intention to treat) and loss to follow up. To evaluate the four types of bias in each eligible trial, the eight internal validity items adapted from the PEDro scale will be applied in this review (Table 1) (Moseley 2002). After reviewing a trial, each of the eight items will be assigned 'Yes' (present) or 'No' (absent or not reported) to indicate the methodological quality of the selected studies according to the criteria used in the OTseeker database (http://www.otseeker.com/scale.htm).

Data extraction

Two review authors (TH and SB) will independently record the following information using a self‐developed data extraction form.
(1) Sample characteristics such as: age, level of education, sex, first or recurrent stroke, type and severity of stroke, time since onset of stroke, type of cognitive impairment, sample size, number of drop outs.
(2) Methodological quality: according to the eight internal validity items as described in Table 01 (Table 1).
(3) Details of the interventions: type of interventions (remedial, compensatory, or dynamic interactional approach), materials used in interventions (e.g. cards, boards, paper and pencil exercises, computer games), duration and frequency of interventions and follow up, individual or group therapy.
(4) Outcome measures: the outcome measures used in the trial and when they were administered.

We will extract data from published reports or request data from the first author when necessary. For each trial, we will require the following summary statistics for each outcome that is measured as continuous data: the mean change in the outcome from baseline, the standard error of the mean change, and the number of participants in each treatment group at each assessment. Where changes from baseline are not reported, we will extract, if available, the mean, the standard deviation, and the number of participants in each treatment group at each time point. For dichotomous data, we will extract the number of participants and the number assessed on the outcome of interest in each treatment group.

The baseline assessment is defined as the latest available assessment prior to randomisation, but no longer than two months prior. In cross‐over trials, we will not analyse further any data collected after the cross over.

We will resolve differences in data extraction by discussion. If no consensus can be achieved, the third review author will be consulted. We will attempt to contact study authors to obtain missing information.

Data analysis

The outcomes measured in clinical trials of cognitive impairment often arise from an ordinal rating scale. Where the rating scales used in the trials have a reasonably large number of categories (more than 10) we will treat the data as continuous outcomes arising from a normal distribution.

If the results of trials are found to be similar, we will synthesise these using meta‐analysis. For continuous data, since trials may not use the same rating scale to assess an outcome, we will calculate two types of estimate. The measure of the treatment difference for any outcome will be the weighted mean difference when the pooled trials use the same rating scale or test, and the standardised mean difference (the absolute mean difference divided by the standard deviation) when they use different rating scales or tests. We will calculate each one, together with the corresponding 95% confidence interval (CI). For dichotomous data, we will compute the relative risk or odds ratio with 95% CI.

The results of all trials will be pooled to present the overall estimate of the treatment effect using a fixed‐effect model and viewed to assess heterogeneity. We will test heterogeneity between trial results by using I‐squared (I²) estimates (Higgins 2003). An I² value above 75% will be considered substantial, indicating heterogeneity between trial results. In this case, subgroup analysis (e.g. separating participants with different severity or separating different treatment duration) will be applied to see if homogeneous results can be generated. Otherwise, a random‐effects model will be used (in which case the confidence intervals will be broader than those of a fixed‐effect model).