What are the functional outcomes of right hemisphere stroke patients with or without hemi-inattention complications? A critical narrative review and suggestions for further research

Abstract Purpose: There is widespread acceptance that patients demonstrating neglect/hemi-inattention (HI) following right hemisphere stroke (RHS) underachieve functionally compared to their counterparts without neglect. However, empirical evidence for this view needs examination. The purpose of this review is to critically appraise relevant studies that compared outcomes from RHS patients with/without hemi-attention and suggest more robust follow-up research. Method: Twelve studies published in 1995–2013 were critically reviewed. Two independent reviewers appraised design features including sample representation, assessment and data analysis methods. Strengths and limitations were highlighted. Results: Results were largely inconsistent. Considerable heterogeneity within patient groups and across studies complicated interpretation. Evidence suggested average group disparity in scores between patients with and without HI at discharge but the cause of functional disparity could not be attributed specifically to HI from the data and modelling results available. Conclusion: The relationship between HI status and functional recovery warrants further investigation in studies with stronger methodology to ensure rigour and robustness in the results. Pending further research, HI status should not be regarded as a key predictor of functional recovery or rehabilitation potential in patients with RHSs. This group should continue to receive appropriate therapeutic intervention aimed at maximising their functional recovery post-stroke. Implications for Rehabilitation Findings from this review demonstrate a paucity of evidence to support the presence of hemi-inattention as a key predictor of functional recovery in patients with right hemisphere stroke; as such, practitioners should take this into consideration when planning rehabilitation programmes of their patients. In the initial months following right hemisphere stroke, there are wide-ranging differences in the rate and amount of functional recovery in patients, with and without hemi-inattention. Practitioners should not limit the aspirations of their patients based on the presence or absence of hemi-inattention. This review has identified a number of measurement limitations in commonly employed assessment tools for hemi-inattention and overall functional recovery. As such, practitioners should take the limitations of specific measures into account when interpreting the results contextually and with respect to their patients’ situation.


Introduction
Hemi-inattention (HI), commonly referred to as ''neglect'', is a complex, heterogeneous and disabling condition which acutely affects up to 80% of patients with right hemisphere stroke (RHS) dysfunction [1,2]. Despite considerable research and advances in the field, HI remains poorly defined as a condition per se. This is supported by the use of multiple descriptors (e.g. unilateral neglect, unilateral inattention) and taxonomies in the literature [3][4][5].
Clinically, HI is characterised by reduced attention and/or spatial awareness to details in the environment (commonly towards the left side of the body). HI can affect one or more functional domains (e.g. sensory-motor, visual-spatial) [6,7] and often coexists with anosognosia and depression [8,9]. HI has been regarded ß 2015 The Author(s). Published by Taylor & Francis. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. as responsible for delayed and challenging rehabilitation, reduced safety awareness, poor functional outcomes, increase in dependency levels and risk of institution care [10][11][12].
Historically, findings from published studies have reported disparity in functional ability scores; with patient groups affected by HI (HI+) underachieving compared to those without (HIÀ) [11,13,14]. Traditionally the cause of this disparity has been largely attributed to the presence of HI, although findings from predictive models have been conflicting and inconclusive [10,13,[15][16][17]. This has led to considerable confusion and uncertainty about the clinical importance and significance of differences thought to be associated with HI [9,[18][19][20]. The paucity of relevant evidenced-based reviews has not helped to clarify the predictive role of HI or to promote good rehabilitation practice.
The last systematic review was undertaken by Jehkonen et al. [21]. The authors focused on the methodological quality of 26 studies published in 1996-2005, which evaluated the impact of Neglect on functional ability in predominantly generic stroke patient samples with mixed lesion sites. Jehkonen et al. [21] highlighted as an issue considerable differences in patient samples and inconsistencies in results but nonetheless concluded that HI had a significant negative impact on functional outcome, either as an independent predictive factor or in the presence of other variables. Their findings corroborated those of earlier reviews [2,22] which were not specifically focused on the relationship between HI and functional ability. Jehkonen et al. [21] recommended further research on homogeneous patient groups with respect to right/left hemispheric lesions to improve consistency in the results.
Given the paucity of research in this area, an in-depth evidence-based critical review of relevant studies is offered here with a different approach to that taken by Jehkonen et al. [21]. However, considering the extent of methodological differences between studies [21], a narrative review was appropriate. This enabled the inclusion of sufficient, relevant contemporary studies which would have otherwise been excluded by the more stringent selection criteria of a pure systematic review. Narrative reviews ''lay out the most recent and best knowledge of various aspects of a problem'' [23, p. 427], and are considered appropriate when a diversity of research methods are used in the studies considered as relevant (rather than focusing only on randomised controlled trials), where studies have used different outcome measures and/ or non-equivalent samples [24] and when studies are of relatively poor methodological quality [25].
The current review examined traditional claims made by previous studies and reviews [21,22] about the negative impact of HI on function; more specifically the strength of the relationship between HI status and functional recovery following RHS. Another aim was to estimate the magnitude of functional differences between HI± patient groups. The current review extended the work carried out by Jehkonen et al. and used a more rigorous and systematic approach to the selection of studies and the review process. Consequently it included fewer (n ¼ 12) but more homogenous studies with RHS patient samples. Theoretically similarly designed studies tend to be more comparable than heterogeneous stroke studies.
Both the discrepancy in HI± patient scores and the relationship between HI and functional recovery are of interest to rehabilitation professionals. Together with other indicators (e.g. stroke severity) they may be used to predict likely change in function with time since stroke. This knowledge can guide rehabilitation decisions, e.g. as to which patients are suitable for early supported home discharge schemes. Currently there is an urgent need for reliable predictors and indicators to support the transfer of in-patient rehabilitation services to appropriate stroke survivors in the community. The final aim was to formulate more robust research strategies based on the limitations of studies to date.

Method
A literature search was conducted from 1995 to February 2015 of the databases MEDLINE, AMED, CINAHL, PsycINFO and COCHRANE using several descriptors of neglect subtypes in the literature including HI, spatial, visual, unilateral, personal, extrapersonal, motor, sensory, hemi and representational. The words; stroke, CVA, functional* and activities of daily living (ADL) were added to the final search so that studies focused on specific functional activities were included. Children or young adults ( 18 years) and non-human samples were excluded.
The search yielded three Cochrane reviews and 195 publications; AMED (70), CINAHL (86), MEDLINE (102) and PsycINFO (57). In line with the aims of the review, and supported by recommendations from Jehkonen et al. [21], only studies that compared the homogeneous patient groups with respect to hemispheric lesion site (RHS) and presenting comparisons of patients with or without HI were selected (including intervention designs); all other studies with the heterogeneous patient samples and no HI group comparison were excluded. In addition, functional ability had to be quantifiably measured so that the HI± group differences in scores could be calculated. Two reviewers read the abstracts and, when in doubt, the publication to determine relevance. This process led to the selection of 12 international studies.
The following information (source, aims, design type, demographic data, assessment tools, data analysis method, results and findings) were extracted from each study and are summarised in Table 1. The critical evaluation process was guided by a checklist described in Appendix 1. Importantly, it focused on the extent of representation of the RHS patient sample with respect to stroke and HI severity levels, time to baseline assessment and follow-up observations, the type of data collected and appropriateness of assessment tools, extent of statistical data analysis undertaken including modelling specifications and, where appropriate, the extent of adjustment undertaken for established confounding factors (e.g. age, stroke severity, time since stroke) and handling of missing data. Each study's strengths and limitations were identified as part of the review process, these are summarised in Table 1; also included is the authors' assessment of the methodological quality of each study. This was graded (A-D) according to the Grading of Recommendations Assessment, Development and Evaluation (GRADE) [26].
On the GRADE scale; A is high and assigned to well-performed Randomised controlled trials (RCTs) and observational studies with consistent results and/or strong effects. B is moderateserious flaws in the design in which the estimated effect is likely to be considerably different than the true effect. C is low -studies with serious limitations in which the true effect is likely to be very different than the estimated, e.g. failure to include relevant confounding factors. D is very low -as in C but any estimated effect is very uncertain and highly unlikely to reflect the true effect.
Morbidity was documented in studies [16,29]; stroke was associated with hypertension, diabetes and heart disease. Stroke severity was not always made clear. It was reported as moderate in two studies [15,16], whilst study [30] indicated that patients with severe stroke were excluded. Two studies [11,14] recruited only patients with (perceived) good rehabilitation potential. Stroke severity was unreported in seven studies [10,12,27,28,29,31,32].

Definition of function and ''Neglect/HI'' syndrome
Conceptually, functional ability/outcome was rarely defined but inferred from ADL measurement scales, mainly the Barthel Index (BI) [33] and Functional Instrumental Measure (FIM) [34]. ''Neglect/HI'' tended to be traditionally defined as a failure to orient, report or respond to stimuli located on the opposite side to the site of the brain lesion which cannot be explained by either primary sensory or motor deficits [35]. Different studies referred to HI sub-types interchangeably, e.g. visual neglect [15], unilateral spatial neglect [12] but effectively measured the same condition because the measurements used cannot differentiate between sub-types of neglect (e.g. visual, spatial and unilateral) [3,4,36] but provide an overall measure of the degree or profundity of the condition.

Research settings
Research settings were insufficiently described to allow clear comparison between countries, e.g. termed as a rehabilitation facility or hospital in Israel and a stroke unit in England. They tended to be either acute in-patient hospital and/or community rehabilitation facilities which would suggest research on samples assessed at varying intervals after stroke onset.

Confirmation of stroke
Stroke was reportedly confirmed by a neurologist in all the studies and by radiological means in 42% [10,14,16,28,31]. Stroke severity was measured in three studies [15,16,27] but the scale score was only reported once in [16] (using the Canadian Neurological Scale -CNS ¼ 7). Aetiologically, infarct was predominant but the majority of studies also included haemorrhage.

Time to first (1st) observation
Baseline measures were taken at variable (non-comparable) times across the studies, making direct comparison difficult. Time to 1st observation was not reported in study [29] and unclear in [32]. In hospital settings, initial measurement varied from 7 to 15 days since stroke in studies [12,15], up to 30 days [10,28,30,31], up to 40 days [14,16] but occurred after 2-6 months in community rehabilitation facilities [11,27].

Other assessments
Functional ability was assessed by the FIM in all but four studies [15,16,27,31] which used different versions of the BI. Most studies measured or recorded additional factors which ranged from length of stay (LOS), discharge destination outcome, continence status, aspects of cognitive-motor function including perception, muscle strength, balance and tactile sensation. Validated measurement scales were generally used for these purposes.

Statistical data analysis
Data tended to be summarised by group (HI±) scores. Median or mean statistics were frequently reported with standard deviation (SD), and inter-quartiles to a lesser extent. Data distribution was rarely described but inferred from the summary statistic used. Rasch data transformation was undertaken by two studies [29,30] in an attempt to ''normalise'' a skewed data distribution. Estimation and management of missing data were not specifically reported, with one exception [30].

Type of data analysis
The type and extent of data analysis varied substantially. For clarity, only general tendencies are described in this section; for specific details refer to Table 1.
The majority of studies carried out preliminary tests for uni/ bivariate associations (e.g. neglect Â functional ability) and/or group (HI±) score comparison, e.g. [11,12,30]. The correlation coefficients used were Spearman's rho or Pearson's r, whereas t-test, Mann-Whitney U and Chi square test were frequently used for group comparisons. In order to minimise type I error (i.e. a false statistically significant result), one study [31] adjusted for multiple testing of the same participants over time by means of Bonferroni correction. Adjustment for small sample size was reported in two studies [15,32] but the adjustment method (Pillai's trace) was only described once in study [32].
Eight studies [10][11][12]15,16,[28][29][30] used regression methods to evaluate various relationships between predictor or explanatory variables (IV's) with one or more dependent variables (DVs), including functional ability. However, the type of model used was not clearly identified (predictive versus associative model). Therefore, it was difficult to assess the suitability of the models employed for the purpose of answering the question posed. For example, Paolucci et al's study [16] modelled the impact on later function of a large number and combination of IV's (admission stroke severity score, gender, type of lesion, hypertension, diabetes, heart disease, unilateral spatial neglect, depression, epileptic seizures post-stroke, family support, education level, discharge destination in various combinations) but the extent of adjustment for confounding factors was variable -stroke severity was inconsistently adjusted for and no adjustment for differences in age was undertaken. Therefore, it was difficult to tease out specific relationships and infer cause from complex regression models. Differences in age, gender and duration of in-patient stay were adjusted in some models evaluated by studies [11,16,30]. Furthermore, the rationale behind the choice, order of entry and measurement level (continuous/categorical) of IV's was rarely stated (e.g. study [10] used stepwise methods, whereas study [16] used forward stepwise), which complicated understanding and interpretation of the results.
Three out of four studies with more than one follow-up point evaluated change in functional ability over time by means of ANOVA's and/or multiple regression analysis [10,14,27]; study [10] specified repeated measures ANOVA. However, both methods have considerable limitations which potentially impact on study power and accuracy of results, especially in serial models with more than one follow-up and several repeated measures on the same individual. To this end, ANOVA requires complete data sets, which is problematic in stroke research due to the likelihood of missing data in the long-term. Ordinary single or multivariate regression analysis does not take into account correlation generated by multiple responses from the same individual on the same assessment measure/s (this violates the statistical assumption of independent observations in regression analysis [38,39]). Consequently, both the validity and accuracy of ordinary regression results are threatened including any inferences based on the results.

Main results and findings
A substantial number of findings were reported across studies, only those pertaining to functional outcome and HI are summarised in this section (refer to Table 1 for details by study).

Disparity between the HI± group scores
All the studies found statistically significant disparities in average HI± group scores wherein the HI+ patients scored less that the HIÀ in overall functional ability and sensory-motor components on the BI, FIM and RMI (Rivermead Mobility Index) [40], at discharge and up to one year post-stroke onset. Relatively less (statistically non-significant) disparity was found on cognitive FIM sub-scale scores in three studies [14,29,30]. Nevertheless, the magnitude of differences reported across studies was considerably variable even when the same measurement scales were used at discharge. Study [15] reported a difference between the HI± groups of 2 BI units (10%); whilst other studies reported differences of 7 BI units (35%) [31], 10 FIM units (8%) [14] and 430 FIM units (24%) [30]. However, it must be pointed out that time to 1st observation and discharge point were also considerably variable across the studies (see later section), therefore it is difficult to extrapolate further from the findings to isolate specific influences of HI.

Progress rates
In general, similar rates of progress between the HI± groups were found prior to discharge but again these rates varied across studies even though the samples were homogenous with respect to lesion side (RHS). Four studies [10,14,27,31] followed up patients beyond discharge; study [27] found that specific HI training improved functional ability of the HI+ group but gains were not maintained by the end of the study (estimated from highly variable published data; recorded about 6-10 months after stroke).

Modelling results
In regard to multiple-regression modelling results, six out of eight studies reported that HI significantly and negatively predicted outcome in a variety of models of differing complexity and specification [10,16]. However, without a clear approach to modelling, such as identifying the strategy (e.g. step-up/down) used to build the models, the technique used to implement that strategy (forward/backward step-wise) and the decision criteria (e.g. theory driven) used within the technique, it was difficult to follow the modelling process and know the true effect of individual variables (inclusive of HI) as evident from the following examples.
Whilst four studies [10][11][12]16] reported independent prediction of HI for the measured functional outcomes, they did not clarify what ''independent'' means (e.g. explains more than 10% of the DV). In contrast, two studies [15,29] did not find a significant relationship between HI status and functional ability in any of the models evaluated -using either FIM or BI (0-20 scale) scores as the DV and HI as a categorical or continuous predictor variable, respectively. Notably the measurement of HI was itself a variable, since different studies used different measurement levels and scales.
In regard to covariates, pre-stroke function was unrelated to functional ability at admission (T0) according to three studies [15,16,30]. Baseline function was significantly, positively related to functional outcome at discharge according to three studies [12,28,29]. Age was significantly, negatively related to change in functional ability according to studies [28,29] but unrelated to final functional ability in models evaluated by studies [15,30]. Cognitive ability was unrelated to ADL as measured by BI and FIM in studies [15,30], respectively, but positively related in study [10].
Overall model fit HI generally explained little of the final variance in functional ability (DV). Study [30] reported the largest amount of variance explained in the DV (discharge FIM scores) as 49%; 44% of which were explained by admission FIM scores and a further 5% by BIT scores (HI levels). A key limitation of these analyses was that none of the studies undertook basic sensitivity analysis, e.g. how well the models met regression assumptions, such as normal distribution of residuals. In addition, confidence intervals (CIs) around regression coefficients or standard error (SE) sizes were rarely reported which complicated the interpretation of regression coefficients.

Overall quality
Taking everything into account, individual studies were graded as C/D on the GRADE scale. This reflected (but was not limited to) serious limitations in the design and data analysis methods described in previous sections. In particular, the exclusion of RHS patients with higher levels of stroke severity and HI, effectively limited representativeness of the samples and generalisation to the wider clinical population. There was a lack of adjustment of potential confounding factors (such as stroke severity, age and time since stroke) in statistical models estimating specific effects of HI on outcomes; and insufficient data/details to enable the reader to make informed decisions (e.g. omission of confidence intervals around regression coefficient estimates which enable accurate interpretation of the result and lack of sensitivity analysis to support validity and conclusions made from the findings).

Discussion
The purpose of the review was threefold -to test traditional claims that neglect/HI has a deleterious impact on functional ability after stroke, to assess the extent of differences between RHS patient groups (with and without HI) and clarify the relationship between HI and functional recovery. In addition, suggestions for more rigorous research were formulated based on a critique of these studies.
Based on findings from the 12 reviewed studies, it is apparent that the presence of HI+ is linked to poor functional outcomes in RHS patients when compared to their counterparts without HI impairments (HIÀ). However, it was not possible to assess the extent of differences between the HI± groups at specific points in time (elapsed after stroke) because of considerable variation and inconsistency in design within and across studies -not just in assessment measures but also in sample mix (age, stroke severity); time to 1st observation as well as follow-up assessments; and methods of data analysis. Neither was it possible to draw firm conclusions on the relationship between HI status and change in basic functional ability over time. In other words, the specific contribution of HI to the functional ability and outcome of RHS patients cannot be known with certainty from the results of the reviewed studies. It is possible that other influential factors, such as stroke severity, age and time since stroke substantially, explain the discrepancy in the HI± patient group scores reported in the literature. To this end, the implications of poor methodology in the field of stroke and HI were also highlighted in previous reports [2,21,22]. There is also recent discussion [38] of the importance of optimising data analysis methods in the field of neglect/HI to enhance the accuracy of predictive models.
Aside from clear methodological limitations, such as exclusion of patients with severe stroke and marked HI+ at baseline, it is likely that unmeasured patient factors also influenced the results and are responsible for a proportion of unexplained variance in functional ability post-stroke. These include pre-stroke educational levels and personality characteristics which would be expected to vary across patients. Furthermore, seven studies [12,15,16,[28][29][30]32] used discharge as the only follow-up point. Although discharge is a key point of change in the stroke recovery pathway, it is subject to natural variation in stroke recovery rate and initial severity. Furthermore, the optimal discharge point may be influenced by differing cultural expectations as to when the recovering patient is ready to be discharged to community care and where this will be undertaken, i.e. home or institution. From the description available in past reviewed studies, it would appear that contextual features, such as stroke unit versus acute general rehabilitation hospital, were not sufficiently comparable in care and rehabilitation provision across different countries especially at the time that they were undertaken [10,12,[14][15][16]. This lack of comparability in context of care is in itself a source of important variation when evaluating functional outcomes from different countries. In relation to the issues associated with discharge as an informative follow-up point, these can be minimised by including at least one standardised fixed follow-up point for all the patients, e.g. baseline -discharge -follow-up 6 months after stroke.

Sample representation and size
Although all patients in the studies selected had RHS, there were considerable sample differences in age and recruitment settings which were not directly comparable. Furthermore, time to 1st observation and patient selection criteria were substantially different; and patients with severe cognitive impairment, and probably severe neglect/HI+ levels, tended to be automatically excluded at the recruitment phase. This strongly suggests that the samples were not sufficiently representative of the RHS population which limits the generalisation of any findings. It is imperative that future studies are conducted with representative (severity-inclusive) samples especially in predictive/explanatory relationship studies. This can be achieved by augmenting the design to suit specific requirements of severely cognitively impaired patients and then applying advanced methods of data analysis (e.g. multi-level modelling -MLM). Specific advantages of MLM over traditional methods of regression analysis and ANOVA are highlighted later on.
The importance of a large enough sample size in relation to study power and reducing type I error cannot be overemphasised. For instance, two studies [10,27] followed up patients for longer periods (up to one year after stroke) but with relatively small starting samples (40 and 59, respectively). Although attrition rate was not reported, this can be substantial at one year post-stroke which further reduces study power to detect differences between the HI± groups. Therefore, future studies should ensure sufficient initial sample size, particularly in longer term follow-up designs (lasting more than 3 months after stroke) to account for possible attrition; an allowance of 15-20% is cited in the literature [41,42].

Measurement
The accuracy of results depended heavily on the precision of instruments and consistency in measurement. Only 50% of studies followed evidence-based recommendations in favour of diagnosing neglect/HI with validated test batteries rather than single cancellation task/s; the rationale [35,43,44] being that several tests are more likely to detect neglect than a single test. To this end, the proportion of the HI+ patients detected in studies that assessed with the BIT is considerably higher than in those that used single tests. Like other test batteries, the BIT has measurement limitations in that it is restricted to assessment of HI in activities conducted within close proximity (arm's length) of surrounding body space and possibly mental representation of objects. In addition, similar to other test batteries (e.g. Chaterine Bergago scale [45]), the BIT cannot distinguish between different sub-types of the disorder [3,36] because all the assessed tasks inherently draw on visual, sensory, motor, spatial, perceptual and cognitive function to varying degrees. These components are intricately inter-twined and dependent on one another which make it virtually impossible to determine the relative contribution of individual components to the results or behaviour observed. Consequently (and of relevance to clinicians and researchers), it is not possible to accurately explain test results in terms of specific sub-types of HI in routine rehabilitation settings. That being said, assessment batteries provide an overall score of HI severity and are indicative of the type of perceptual, cognitive and/or motor impairments present.
Pending the development of more comprehensive and practical assessment tools, future researchers and clinicians should follow evidence-based assessment guidelines, i.e. use of validated assessment batteries for HI. Interpretation of results will also be enhanced by reporting the proportion of patients with mild compared to severe neglect/HI in the sample by categorising the HI+ patients at baseline. This is possible with the BIT [36,46].
Functional ability in the reviewed studies was assessed with at least two versions of the BI (using scales of 0-20 and 0-100) with differing sensitivity and precision [47,48], making comparison difficult. Furthermore, neither version assesses social/communicative components. They are therefore possibly more limited than the FIM in this respect [46]. In comparison, the Extended Barthel Index version has been used in several stroke studies [49,50]. It is a validated measure and overcomes some of the limitations of other BI versions in that it includes assessment of social and communicative components and accounts for time taken to complete ADL tasks.
Ideally future studies should measure functional ability with comprehensive evidence-based measures; however it must be acknowledged that current choices are limited not only by content in terms of validity but also reliability over time and applicability across acute and community settings.

The position, interval and frequency of observations
In relation to design, the position, interval and frequency of observations is of critical importance in recording important changes in ability at the points when they are likely to occur. Furthermore, the amount and quality of information collected partly determines the type and extent of data analysis that is possible in order to answer specific research questions [39,51,52]. This ability was greatly compromised within and across the reviewed studies by the variability in time around the initial measurement point and follow-up, the number and frequency of observations made over time, and the duration of individual studies.
In some studies, the extent of variation around the mean LOS blurred the beginning and end of the study (e.g. study [16] reported mean LOS (HI+) 117 ± 61 days versus (HIÀ) 81 ± 38, and study [29] reported (HI+) 29 versus (HIÀ) 22, range 3-75 days). This would suggest considerable differences in LOS and patient exposure to care within and across studies which have implications for assessing recovery.
At baseline, the total variation in initial measurement points across studies was 7 days to 6 months post-stroke in studies [27,32], respectively. During this period, both spontaneous and rehabilitation-driven recovery processes are known to actively contribute to outcome [53][54][55]. Based on known recovery patterns, the impact of spontaneous recovery processes is expected to confound rehabilitative processes in at least 11/12 studies. Future studies should aim to collect initial data as early as is pragmatically possible, ideally within the 1st week of stroke. This would potentially streamline data collection procedures, enhance comparison of results across studies and minimise the effect of spontaneous recovery processes on outcome so that cause (e.g. influence of HI) can be more confidently inferred from regression results.
Future studies could also consider fixing the last follow-up point whilst still recording HI and functional change at discharge. Better still would be to statistically adjust for the confounding effect of time elapsed post-stroke. An optional growth curve modelling (MLM) approach could be used to model change over time in time-variant factors -this approach has been applied in a predictive stroke patient study [56]. The advantages afforded by MLM in relation to analysis of stroke data are next summarised in this review.
Multi-level modelling approaches to data analysis The advantages of multi-level modelling over traditional regression methods of analysis should not be overlooked when modelling dependency in the data due to multiple measurements from the same patient over time and interdependency between potential explanatory factors associated with functional ability (e.g. cognitive-motor processes). MLM can also handle data missing at random and unequal interval observations -all of which were potentially problematic features of the reviewed studies. As a result, MLM regression estimates are highly accurate and relatively stable compared to estimates obtained by ordinary multivariate regression analysis [51,52]. MLM has other useful features, such as estimation of unexplained variance within and across patients over time (in serial studies). It is imperative that future studies consider the advantages afforded by novel advanced statistical techniques in serial designs, which provide rich information but need to be appropriately statistically analysed for optimal results [38,57,58].
From the discussion so far it can be deduced that regression estimates in 8/12 studies which modelled outcome are likely to be overinflated with under-estimated standard errors. In turn, this increases the risk of type 1 error which has adverse implications for interpretation of results, not least the predictive importance of HI on functional ability.

Confounding factors in stroke functional outcome studies
Another issue complicating the inference of causality from regression results is the lack of statistical adjustment for the established confounding effect of stroke severity and to a lesser extent chronological age. Besides the natural variation of both factors in different samples, stroke severity and age are both associated with HI [21,59,60]. Therefore, it is important that these particular variables are modelled in studies exploring multiple predictor variables and functional outcome [39,61]. As already stated, the use of MLM can greatly help with teasing out complicated relationship dynamics. By the same token, the need for an adequately powered study is emphasised. Considering that only six studies started with approximately 100 patients and some cases [10,27] also analysed predictive multivariate models, it is likely that respective studies were underpowered to detect true effects. This increases the uncertainty of the modelled results and detracts from the validity of their findings.

Strengths and limitations of the review
The current review extended the work carried out by Jehkonen et al. by reviewing a homogeneous (versus mixed) patient sample with respect to hemispheric lesion (RHS). Further, a relatively more rigorous and systematic approach to study selection and the overall review process was undertaken.
The current review focused on group comparisons of functional outcomes of RHS patients with and without HI and the relationship of HI status with functional change with time since stroke, whereas Jehkonen et al. undertook a generalised review of the methodological issues from a wider range of studies which did not always include patient comparisons in the design (HI±). Group comparisons allow for the calculation of mean differences between patient groups and estimation of the modelled relationship between HI (group) status and functional outcome. This resulted in new insights into the data, such as the lack of adjustment for established confounding factors in past studies (e.g. stroke severity, time since stroke and age), leading to a different conclusion, i.e. no inferences could be made from the data available on either the relationship between neglect and functional outcome or the magnitude of difference in measureable scores between the patients with and without HI due to considerable heterogeneity in design and methods used to statistically analyse the data. Currently there is an urgent need for valid predictors and indicators to support the transfer of inpatient stroke rehabilitation services to appropriate stroke survivors living in the community.
Furthermore, due to the structure and layout of the review by Jehkonen et al., it is not possible to tell which studies found what and when in terms of results and time since stroke. The current review is more detailed in this respect. It is also more systematic and transparent both in the selection of studies and their evaluation (a checklist was used to ensure parity and consistency during the review process). The methodological quality of each study was separately graded. All these factors contribute to the rigour of the review and the validity of the findings.
The main limitation of the review is the possibility that relevant studies may have been missed due to indexing and multiple terms used to describe the ''neglect'' syndrome although a thorough search of the literature has been conducted several times. In addition, only English language publications or translations were considered.
The review does not address HI conditions arising from specific cortical or sub-cortical structures or anterior/posterior circulation division or left hemisphere lesions. Although HI is associated with different brain areas, HI emanating from the right hemisphere is commonly encountered in rehabilitation settings and challenging to treat [4][5][6][7]9]. In addition, the interpretation of results from assessments of HI in the left hemisphere lesions is confounded by speech and language impairments which frequently accompany LHS conditions. Non-language based clinical assessment for HI is urgently needed for this purpose. There is a lack of comparative group studies specifically on HI/neglect emanating from sub-cortical structures [63,64] probably because this is harder to identify in routine clinical settings and possibly less commonly encountered.

Conclusions
To conclude, based on findings from this review, the evidence for an important relationship between HI status and functional outcome in RHS patients (with or without consideration of other influential factors) cannot be substantiated from the data available. This is due to significant methodological limitations of published studies highlighted in this review and the relatively few studies published in this field in the last decade [29][30][31][32]. The evidence that differences exist between the HI± patient groups on discharge and in the early months that follow tends to be better substantiated and is of interest to rehabilitation practitioners. However, on its own, this knowledge does not advance practice in the field.
In light of findings from this review, it is recommended that rehabilitation practitioners not only take into account the severity of HI in decision making but also recognise that evidence is weak for it being an independent marker of future functional recovery.
Given the extent of differential progress possible in patients with RHS with and without HI, it is recommended that rehabilitation practitioners consider both the rate and amount of functional recovery over time. This is likely to be more informative overall than isolated measurements of HI soon after stroke. In relation to measurement, it is recommended that rehabilitation professionals sufficiently consider the psychometric limitations of commonly employed assessments of HI and overall functional ability (e.g. the BIT, BI and FIM) when interpreting the results.
Further research is warranted to identify the magnitude of the differences within and between the patient HI± groups and importantly the precise relationship between HI status and functional recovery with time since stroke. For this reason, well-designed, longitudinal, serial studies on large RHS patient samples are required which can withstand the scrutiny of future reviews on the subject.
It is recommended that future rehabilitation research in the field takes into account important methodological features to improve the quality and robustness of their findings. These include but are not limited to adequate sample size to detect differences in groups of patients with and without HI, with a full spectrum of stroke severity and HI, sufficient amounts of data by means of standardised methods repeated over time rather than isolated data points, and with sufficient follow-up to allow functional recovery to occur over time since stroke.

Declaration of interest
This work received no external funding. The authors report no conflicts of interest.