Post-stroke cognition with the Oxford Cognitive Screen vs Montreal Cognitive Assessment: a multi-site randomized controlled study (OCS-CARE) [version 1; peer review: 1 approved, 1 approved with reservations]

Background: Cognitive impairment is common following stroke. The Oxford Cognitive Screen (OCS) was designed to assess focal poststroke cognitive deficits in five domains. Here, we investigated whether results generated by the OCS vs the domain-general Montreal Cognitive Assessment (MoCA) at baseline impacted patient outcomes at 6 months follow-up. Methods: Patients <2 months post-stroke were randomized to receive either the OCS and corresponding information leaflet or standard care with the MoCA at baseline. After 6 months, patients received both the OCS and MoCA. The primary registered outcome measures were the Stroke Impact Scale (SIS) and change in stroke severity (National Institutes of Health Stroke Scale; NIHSS) at 6 months. The secondary outcome was change in cognitive performance from baseline to 6month follow-up. The relationship between scores from the two cognitive screens at follow-up was also explored. Results: A total of 821 patients from 37 different hospital or rehabilitation sites (England, UK) were recruited to the OCS-CARE study, with 467 completing 6-month follow-up. Patient outcomes defined by overall SIS scores and changes in NIHSS did not differ between the OCS or MoCA groups. There were high accordance rates between the OCS and MoCA at 6 months, with severity of cognitive impairment reflected in both screening tools. Cognitive performance in both groups over the 6-month follow-up declined in 22% of patients. A larger proportion of OCS group patients demonstrated improvements in cognitive scores (49% vs 40% in MoCA). Conclusions: The type of cognitive screening test did not impact broad stroke outcome measures, and the two screening tools showed a high overall accordance. The results suggest that more of the domain-specific deficits in OCS recover subacutely, providing a more Open Peer Review


Lay summary
Stroke survivors commonly experience difficulties with various aspects of thinking, planning and remembering, as well as seeing and speaking. These are often less obvious than physical problems, but are important to identify early after stroke in order to help guide rehabilitation and other treatment services.
In this study we randomly allocated 821 stroke survivors to one of two groups shortly after their stroke: one group received a cognitive assessment (Oxford Cognitive Screen) that assesses specific problems a patient may have after stroke, or another group that received a broader measure that assesses overall thinking performance (Montreal Cognitive Assessment). We then followed-up with them 6 months later to find out how they were recovering from their stroke. We collected data using a broad stroke severity measure that assesses a variety of abilities, such as the use of both arms and legs, etc. We also asked the patient to fill out a questionnaire regarding their own physical abilities, mood, communication, activities of daily life, participation in daily living, etc.
The aim of the study was to investigate whether the information provided by a cognitive screen regarding the specific cognitive problems experienced by the patient after stroke would lead to improved outcomes 6 months later. We found that the use of a cognitive screen right after stroke did not affect overall stroke recovery. We also found the cognitive screens to be in accordance with one another, in that people who struggle with one are also likely to struggle with the other. Despite not finding differences between patient outcomes between the two groups at follow-up, we did find that in the group that received the more specific cognitive screen after stroke, a higher proportion of people showed improvements than in the broad screening group. We believe this is because some of the stroke specific problems are recovering well over time. Further research into understanding what makes a person more likely to regain particular abilities is needed.

Introduction
Cognitive impairment is a common consequence of stroke. Studies reporting the prevalence of early post-stroke cognitive impairments suggest most patients experience at least one cognitive domain deficit (Demeyere et al., 2015;Jaillard et al., 2009). However, exact prevalence estimates of post-stroke cognitive impairment vary substantially in the literature depending on the nature and timing of the assessments being performed, as well as the methods of patient selection (Hurford et al., 2013;Nys et al., 2007). Still, post-stroke cognitive impairments are known to be a major determinant of poor long-term outcome and high societal costs in regards to increased mortality and disability rates (Hakkennes et al., 2011;Patel et al., 2002), functional outcomes and developing mood-related disorders (Nys et al., 2005;Nys et al., 2006).
Despite agreement that overall cognitive impairment is common after stroke, the co-occurrences and patterns of domainspecific deficits are not well understood. Cognition is often referred to as a single entity, with a binary description of "impaired" or "non-impaired". However, post-stroke cognitive impairment is complex, and does not constitute a unitary syndrome, with significant variability in stroke lesion location and size impacting on the nature and extent of the cognitive impairments such as executive functions, memory, language and visuo-spatial abilities (Jokinen et al., 2015). Domain-specific cognitive screening targeting common post-stroke impairments, such as aphasia, apraxia, and hemi-spatial neglect, allows for a more comprehensive assessment of the specific problems and the relationships between them. This domain-specific approach is recommended in several clinical guidelines, including the UK National Institute for Clinical Excellence (NICE) guideline for stroke care (NICE, 2013), which specifically stated the need to assess performance across different domains of cognition after stroke ("attention, memory, spatial awareness, apraxia, perception"), and the RCP Clinical Guidelines for stroke (Rudd et al., 2016), which suggested that "Each cognitive domain (e.g. perception, attention, memory) should not be considered in isolation because most everyday activities draw on a range of abilities." Due to the time limitations and resource constraints in acute stroke units, lengthy neuropsychological batteries for cognitive screening in the acute setting are impractical and unfeasible. Additionally, a large proportion of stroke patients experience quite severe impairments acutely and therefore are unable to participate in extensive cognitive assessments. Currently, most cognitive screening post-stroke mirrors that of an initial dementia test, with the use of broad measures such as the Mini-Mental State Exam (MMSE) (Folstein et al., 1975)  was designed specifically for an acute stroke population with a neuropsychological foundation that provides a domain-specific cognitive profile. The OCS is structured around five domains: attention and executive function, language, memory, number processing and praxis. It takes approximately 15 minutes to complete, which is a similar testing length to the MoCA. The OCS was designed to be maximally inclusive as it is suitable for patients with aphasia and hemi-spatial neglect, and can be delivered at the bedside. Rather than an overall pass/fail score, the OCS provides a "visual snapshot" (Figure 1) of the cognitive profile to communicate, at a glance, areas of strengths and weaknesses for the specific cognitive domains. The OCS has recently been shown to be more sensitive than the MoCA Utilizing domain-specific cognitive screening and subsequently reporting results to the multidisciplinary team and patients/ carers early in the rehabilitation pathway may impact outcomes. Detecting specific domain impairments may lead to improved rehabilitation, and more appropriate therapies that consider the particular limitations in cognition. Additionally, providing domain-specific information on possible coping strategies via information leaflets selected based on patients' cognitive screen results may lead to higher levels of adjustment. Therefore, the primary aim of this study was to investigate whether domain-specific screening early in the care pathway impacted long-term outcome.
The secondary aim of the study was to determine the trajectories for domain-specific post-stroke cognitive impairments over 6 months. Many studies have reported long-term post-stroke cognitive impairment may be attributable to the emergence of post-stroke and vascular dementia (Pendlebury et al., 2015;Pendlebury et al., 2019). This may be true for global impairment, but it is less clear whether stroke-related domainspecific impairments can improve, remain stable or decline over a long period of time (Del Ser et al., 2005). Individuals may experience several different domain impairments and collectively be classed as having 'post-stroke dementia' clinically, but when considering each domain impairment independently, there may be varying recovery trajectories. For example, many stroke survivors may present with a stable cognitive impairment over time, or indeed recover from various early focal deficits, such as hemispatial neglect (Demeyere & Gillebert, 2019). This would suggest that not all long-term post-stroke impairments are eligible for dementia consideration and equally, declining impairment in chronic stroke cannot exclude the contribution of vascular and age-related pathology. Thus, we examined the changes in cognitive performance from baseline to 6-month follow-up for the two screening groups with the aim of better characterizing differing cognitive trajectories.
Here, we report data from OCS-CARE, a multi-centre, randomized controlled study that investigated whether the results generated by the OCS and corresponding patient information leaflets versus results from the MoCA impacted long-term post-stroke outcomes. At 6 months post-stroke, participants completed both the OCS and MoCA, allowing us for the first time to compare domain-specific (OCS) and domain-general (MoCA) impairment levels for stroke survivors beyond the acute phase, and in relation to the baseline cognitive profiles. Here we compare proportions of patients whose cognitive performance declined, remained stable or improved over the first 6 months post-stroke.

Trial design
In this noninferiority randomized controlled study, subacute stroke patients were assigned at a 1:1 allocation ratio to either

Participants
Between July 2014 and July 2016, we recruited a consecutive sample of 821 acute stroke patients from 37 sites in England, UK (Table 1 and Figure 2). Patients were included if they met the following criteria: (i) acute stroke patient (within 10 weeks  of confirmed stroke); (ii) adult (≤90 years of age); (iii) able to concentrate sufficiently for one hour as judged by the multidisciplinary care team in the hospital; (iv) have sufficient language comprehension to pass the first orienting tests (the OCS Picture Naming and Semantics tasks); and (v) willing and able to give informed consent themselves. Patients were excluded if they were too unwell to take part, more than 10 weeks poststroke, or had insufficient language comprehension to pass the first orienting tests in the OCS. All included participants provided written informed consent. All procedures were in accordance with the Declaration of Helsinki and approved by the West Midlands -Coventry and Warwickshire National Research Ethical Committee (REC reference 12/WM/00335).

Randomization
Individual acute stroke patients were allocated to either the OCS or MoCA screening by a computer randomization service at all 37 sites. We obtained a balanced randomization over all sites by using block randomization with a lock size of 4. The randomization sequence was concealed from all researchers until assignment. A participant was enrolled in the study by the UK Clinical Research Network research facilitators who logged on to the project website and obtain the next available randomization code and assigned the participant to either the OCS or MoCA screening group.

Blinding
Participants, test administrators, care providers and dataanalysists could not be blinded during the study.

Data analysis
Pre-specified analysis included descriptive statistics for demographics, reasons for loss to follow-up and clinical characteristics of each group (frequencies, mean and standard deviations). We quantified severity of cognitive impairment by the number of subtests on which a patient scored below the cut-off score for the OCS and by the total score for the MoCA. Patients were classified as impaired if they failed ≥1 subtest on the OCS or had a MoCA score of <26. When comparing MoCA scores over time (baseline to follow-up), a difference of ≥2 points must be observed in order to be considered clinically significant. This operationalized criterion was previously demonstrated by Tan et al. (2017).
For our primary analyses, we report the distributions of scores on the cognitive tests and outcome measures. From the initial assessment, we compared severity between both groups, as reflected in the NIHSS and Barthel Index, with Wilcoxon rank sum tests to account for skewness in the data. Linear regression was performed to evaluate the association between severity and cognitive impairment at the acute stage. At follow-up, we evaluated the mean difference between each group's NIHSS and SIS scores with a Wilcoxon Mann-Whitney Rank-Sum test with continuity correction due to the scores for both groups being not normally distributed. Assessing the change in NIHSS scores from the acute assessment to 6 months was carried out by a Welch two-sample t test, as the scores were approximately normally distributed.
Our secondary analyses compared improvement in cognition on either the OCS or MoCA in both groups by calculating the difference in scores at the acute and follow-up assessment in a Bland-Altman analysis. We also calculated the percentage of patients who improved, declined or remained stable in each group and then compared proportional distributions between both groups with a χ 2 test. Lastly, we explored the relationship between the OCS and MoCA results at follow-up through a linear regression. All statistical testing was performed at a twosided 5% significance level. Analyses were completed using R software version 1.1.383 (2017). The full analysis code is available on Figshare (Demeyere et al., 2019a), along with the Underlying data (Demeyere et al., 2019b).

Results
A total of 821 acute stroke patients were randomly assigned to either the OCS screening group (n=411) or the standard care screening group with the MoCA (n=410). Upon further investigation, 6 patients that were initially randomized did not meet inclusion criteria and were therefore excluded. Additionally, 19 patients were withdrawn from the study before completing baseline assessment (deteriorating condition or voluntary withdrawal). The randomized sample size for analysis at baseline assessment was n=399 (OCS group) and n=397 (MoCA group).
A total of 467 patients completed a follow-up visit, with lossto-follow-up being attributed to study withdrawal (n=95), unable to be contacted (n=98), incomplete assessment (n=71), death (n=29) and 27 patients were undocumented loss-tofollow-up. The CONSORT flow-diagram summarizes the patient pathway in Figure 3 and patient demographics at baseline are shown in Table 2.
Cognitive screening at baseline At baseline, 75% of patients who received the OCS were impaired in ≥1 cognitive domain and 58.37% of patients in the MoCA group were cognitively impaired based on a score <26. The distributions of the scores in both groups was skewed ( Figure 4), with more patients in the sample who had fewer cognitive impairments, represented by a low score on the OCS and a high score on MoCA.  Figure 7C). This indicates that patients had a similar outcome regardless of which cognitive screen they received.
In addition to examining the average improvement in cognitive function, we compared the proportion of patients who showed an improvement, decline or no change in cognition at 6 months.  (Figure 9A versus 9B). Figure 10 explores the relationship between MoCA scores and impairments on each of the five domains in the OCS. Qualitatively, the data indicates that the most common impairments in our sample at 6-8 months post stroke, are in the attention and memory domains. Furthermore, the data suggests there is no association between overall MoCA score and any specific OCS domain, confirming the MoCA as a domain-general measure.

Discussion
The OCS-CARE study set out to investigate cognitive outcomes at 6 months based on domain-specific versus domain general cognitive screening within 10 weeks of stroke. By 'domain-specific' we refer to problems in a particular process that contributes only to the domain in question -for example, a deficit in spatial orienting affecting spatial attention but not (say) language comprehension. By 'domain-general' we refer to a process that supports a variety of cognitive domains, such as working memory and sustained attention, which support language, memory, number processing etc. Collectively, these impairments may impact recovery differently, e.g. regarding focal cognitive impairments, hemi-spatial neglect may recover while overall cognition may still decline, leading to post-stroke dementia.   A total of 821 patients were randomised to either cognitive screening with the OCS or MoCA, and followed-up with functional outcome and cognitive measures after 6 months at 37 sites in England, UK. The initial sample consisted of 54% males, with an overall average age of 69 years and 11.8 years of education. Given the national average of 72 years for first-ever stroke (Public Health England, 2018) and this inclusive sample also recruiting recurrent stroke survivors, this cohort would appear to be slightly younger than would have been expected.  (3-17)). In addition, the skewed distributions suggest that there was a slight bias in the multi-site recruitment to this research towards younger and less severely impaired stroke survivors. Nevertheless, even in this perhaps overall milder stroke sample, cognitive impairment was highly prevalent both at baseline and follow-up. At baseline, 75% of patients demonstrated at least one cognitive domain impairment in the OCS group and 58% of patients were impaired in the MoCA group (below standard MoCA cut-off). A reliable relationship between severity of cognitive impairment and stroke severity was found in both screening groups of the study, where patients who scored higher on the NIHSS demonstrated more extensive cognitive impairments. Similarly, a relationship was found between the severity of cognitive scores and lower Barthel scores.
In regards to comparing stroke severity and impact outcomes at 6 months follow-up, no difference in the core outcome measure of Stroke Impact Scale scores or NIHSS was found, demonstrating no difference in global outcome for domainspecific vs domain-general cognitive screening.
There are a few significant limitations regarding the study design. First, no information was recorded regarding potentially differing care pathways that may have been followed given a particular cognitive profile. Though the researchers were asked to provide leaflets to patients and carers, as well as report findings in the medical notes, it is not clear whether and to which extent these cognitive reports were considered. It is likely that the 33 different settings differed in their approaches here. These differences could be due to differences in established care pathways, routine treatments, familiarity with the screening tools etc. Finally, the approaches would likely also differ depending on how many patients were recruited at the sites (with some sites only recruiting a small number of patients). Second, differences related to more specific psychological outcomes could be missed and not included in such broad functional outcome measures such as the stroke impact scale and NIHSS. A more tailored and extended measure on cognitive coping might provide a more sensitive outcome measure here. Third, we suggest that a stronger understanding of cognitive care pathways might be better achieved with randomising settings where one pathway is consistently followed, and documented in depth, rather than the present design of randomising at the individual patient level.
With regards to the secondary outcome measures that were confined to cognitive changes, we found no systematic overall improvement or decline over the 6 months in either the MoCA or OCS group. Instead, the findings demonstrated that within this heterogenous cohort, a proportion of patients improved (49% OCS; 40% MoCA group), a proportion remained stable (29% OCS; 38% MoCA) and a proportion of patients declined (22% in both groups), cancelling out a systematic effect. We found no relationship between the severity of initial cognitive impairment and the likelihood of improvement vs decline in overall cognition in either group. However, when assessing the differing proportions, it appeared that a larger proportion of patients who were assessed with OCS demonstrated improvements (49%) compared to the proportion of patients improving on MoCA (40%). These findings are reminiscent of the heterogeneous cognitive trajectories found in a longer term follow-up study by Del Ser et al. (2005). At 2 years post stroke, they found cognitive status to be stable in most cases (78%), some patients demonstrating cognitive decline and (14%) and 8% improved (total N=49). Here, we found much larger proportions of recovery, both with OCS and with MoCA. However, the timelines were different (earlier follow-up), and the cognitive status measure in Del Ser et al. (2005) is likely to be less sensitive to small levels of change.
In addition to differences in sensitivity of cognitive measures, in our study, the difference in proportions of recovery is likely to also reflect the difference in the aspects of cognition that are being measured. Specifically, the improvements may reflect recovery of stroke-specific impairments, such as reading impairments, hemi-spatial neglect and apraxia. These will likely have led to improved scores on both OCS and MoCA measures, though we suggest that the increased rates of improvement on the OCS may be driven by a more explicit measure of these focal cognitive deficits.
It is important to note that the aetiology of long-term cognitive deficits is not necessarily related to a progressive neural degeneration; following stroke, cognitive evolution beyond the acute phase of recovery is indeed heterogeneous, as many patients may remain stable or even improve over time (Del Ser et al., 2005). However, many large-scale studies consider cognitive impairments on a single score, and thereby reduce the complexity of neuropsychological understanding. When using measures borrowed from the field of dementia, it is important to note that this can lead to confounded results, where people with aphasia are excluded or fail on tasks due to language impairments (Demeyere et al., 2016; Mancuso et al., 2018). Large-scale monitoring studies have highlighted the importance of cognition in assessing outcomes, but improved measures to understand the contributing factors are needed. When time permits, cognition should be assessed in more detail, particularly long after stroke in order to determine changes over time.
Finally, when directly comparing performance on the MoCA and OCS at follow-up where both tests were completed, we found a clear agreement between both tests, with highly related levels of scores on both and no difference for the two screening groups in the study. Further exploration of the data revealed that there was no association between MoCA scores and any one specific OCS domain, indicating that the MoCA appears to be measuring a global level of functioning in the chronic stage following stroke.
In conclusion, this randomised study found no difference in global outcome measures whether OCS or MoCA was completed as the first step of a cognitive care pathway post stroke.
The study demonstrated high agreement rates between the cognitive screening measures, a clear relation between severity of stroke and severity of cognitive impairments and importantly highlighted longer-term cognitive outcomes to be highly heterogenous, with cognitive trajectories 6 months post stroke demonstrating large proportions of stability and improvements along with approximately a fifth of patients following a declining cognitive trajectory. The authors present data from a noninferiority randomized controlled study examining the relationship between performance on the OCS/MoCA and stroke outcome measures (SIS and NIHSS) at baseline and at 6 month follow-up. The large multi-centre trial found no difference in stroke outcome (SIS or NIHSS score) based on the cognitive screening tool administered at baseline. Also, they found no systematic overall improvement or decline over 6 months in either the OCS or MoCA.

Comment:
The broad topic is interesting and important. However, it remains uncertain why the authors think using a particular type of screen might alter the overall functional impact of a stroke at 6 months.They should specifically explain why they hypothesise this could happen. For example, do they think that an improved screen might lead to uncovering deficits to guide rehabilitation input after stroke?
Introduction: As above to explain the motivation of the study, more discussion is needed as to why the authors might expect the administration of a cognitive screen to change stroke outcome scores on functional/neurological measures such as the NIHSS and Barthel. Is there any previous research to support the author's hypothesis? On the face of it, the result seems to be fully expected and unsurprising, unless the assessment battery is linked to a specific intervention. Can the authors please clarify and comment?
More clarity is needed to differentiate the role of domain-specific assessment versus the giving of domain-specific feedback and strategies to the patient. Perhaps the authors believe both aspects are important, but this needs to be made clearer and consistent throughout the manuscript. The study of domain-specific impairment after stroke, it's relevance to long-term outcome and the trajectory of recovery over time is not new. The authors should make greater effort to include findings from past studies in both the introduction and discussion for completeness and transparency. Alternatively can they see how their cohort compares to an unselected UK database, e.g. SSNAP?
Specifically some measure of stroke severity at baseline (e.g. NIHSS) would be helpful. This is crucial to understand the generalisability of the findings. Can the authors please provide these data?
Please report the mean timing and range of the assessments at baseline and 6 months for the two groups. Primary outcome -For completeness, it would be interesting to see whether the results differ for the different SIS/NIHSS subscales. At present, the authors only present on overall score.
Secondary outcome -Given that the authors argue for the importance of understanding domainspecific impairment after stroke, it is surprising that the analyses does not take into account domain-specific change over time. It would be interesting to see whether there are different trajectories for the different domains as suggested by previous studies (e.g. Hurford, R., Charidimou, A., Fox, Z., Cipolotti, L., & Werring, D. J. (2013). Domain-specific trends in cognitive impairment after acute ischaemic stroke 2 .).

Discussion:
The authors' definition of Domain-general in the first paragraph is somewhat confusing. Although the MoCA produces an overall score, it would not be accurate to suggest that this score reflects a 'domain-general' cognitive process as described by the authors. As the authors pointed out, the MoCA was designed to detect dementia. The issue here is that the MoCA treats cognition as a unitary concept with a one score outcome. I would avoid suggesting that the MoCA assesses a fluid intelligence-like processes as there is no basis for this.
administers the screening assessment on the stroke ward. It would be of benefit to readers outside of the UK to explain a bit more about whom these research facilitators are, their role in research, training etc.
Methods -Participants: p6-7 -Did the UK Clinical Research Network research facilitators receive specific training on how to administer and interpret the OCS/MoCA?
Methods -Participants: p6-7 -Did the UK Clinical Research Network research facilitators/ OTs go through the tailored management advice sheet with each patient or give it to them to read in their own time? -ties in to fidelity of intervention delivery.
Methods -Outcomes: p7 -How long did the follow up testing session take?
Methods -Outcomes: p7 -Did follow up occur 6 months post-stroke or 6 months post initial baseline assessment? It would be useful to include Mean (SD) weeks since stroke based on timing of the follow-up assessment also.
Results: p8 -for context, it would be useful to the reader (particularly those outside the UK) to have a brief description of what measures are usually taken post-stroke in the acute setting as part of standard care, e.g. are all patients screened for cognitive impairment post-stroke or is this done on an ad hoc basis/ varies across sites?
Results -baseline demographics: p9 -suggest including another column describing the baseline characteristics of those lost to follow up. It would be interesting to see if they were older, more severely impaired (cognitively), had recurrent stroke etc.
Results -baseline demographics: p9 -how many participants across groups had recurrent stroke?
Results: p12 -suggest changing wording in second paragraph from average "improvement" to average "change" in cognitive function (to encompass improvement, decline, and no difference).
Results: p12 -it would be really interesting if the authors reported on the domain-specific cognitive impairments identified at 6-months by the OCS and if specific domains showed change from baseline to follow-up as measured by the OCS.

If applicable, is the statistical analysis and its interpretation appropriate? Yes
Are all the source data underlying the results available to ensure full reproducibility?