Design, Participants and Setting

This article presents a cohort analysis of a national multi-centre sample of left and right hemisphere stroke patients who were admitted to inpatient specialist rehabilitation programmes in the UK during a 3-year period between May 2010 and April 2013.

In the UK, the majority of stroke patients will make a good recovery with the support of their local (Level 3) stroke rehabilitation services. A smaller number of patients have more complex needs that require expertise, equipment and facilities of a district (Level 2) or tertiary (Level 1) specialist rehabilitation centre. Typically, these services take a selected population of mainly younger stroke patients with a mixture of physical, cognitive, communicative and/or psychosocial difficulties. Detailed criteria for admission to such services are available on the British Society of Rehabilitation Medicine website [15]. Outcome evaluation in this group must take account of the full range of disabilities, rather than just physical function.

The UK Rehabilitation Outcomes Collaborative (UKROC) provides the national clinical database for specialist rehabilitation in the UK. Established in 2010 with funding from the UK National Institute for Health Research (Programme grant RP-PG-0407-10185), the UKROC database collates information on needs, inputs and outcomes of all the case episodes of in-patient specialist rehabilitation of those admitted to specialist rehabilitation (Levels 1 and 2) services in England. Other UK centres participate on a voluntary basis. A national training programme is in place to ensure that clinical teams are trained in the use of the UKROC tools and outcome measures.

The dataset consists of demographic information and process data, together with a hierarchical system of outcome measurement that includes the Barthel Index (at the simplest level), the FIM and the UK FIM+FAM (at the most detailed level) [11], rated on admission and discharge. At the start of data collection, services could choose which of these measures to report as an outcome measure, depending on the time that clinicians were willing/able to spend collecting the data [11]. Since April 2013, however, reporting of the UK FIM+FAM has been mandatory for all level 1 and 2 specialist rehabilitation services in England [10].

For the purpose of this analysis, we extracted all the case episodes for stroke patients in whom a full set of UK FIM+FAM data was collected on both admission and discharge. Subarachnoid haemorrhage was excluded because it often causes a diffuse injury and pattern of deficit, which is atypical within the usual stroke population. The UKROC dataset includes a field for primary localisation of brain injury, including left and right hemisphere, as well as bilateral, frontal brainstem and diffuse. This localisation is recorded by the treating clinical teams. The dataset does not include information on neuroimaging, so we cannot exclude the possibility that some of the localisation data was misreported.

Extracted data were transferred to Microsoft Excel for cleaning, and then analysed using the IBM Statistical Package for Social Sciences (SPSS version 21).

Measures—The UK FIM+FAM

The UK FIM+FAM consists of 30 items [8]. Each item is rated on seven levels with a score ranging from 1—‘Total dependence’ to 7—‘Complete independence’. Nine items address basic self-care including bladder and bowel management; seven items address transfers and mobility; six items address communication, and nine items address cognitive and psychosocial function. Scores are rated by the multidisciplinary team, according to the published scoring manual, within 10 days of admission and within the last 7 days before discharge from the rehabilitation programme. Rating takes approximately 20–30 minutes depending on the complexity of the case and the experience of the team. Further detail regarding development of the UK version is detailed elsewhere [8], and specific information on scoring (including the scoring manual for the UK FAM items) may be found on our website [16].

Analysis

The UK FIM+FAM generates ordinal data and there is continued debate about the approach to statistical analysis in this context. Some authors favour techniques based on Item Response Theory such as Rasch analysis [17] whilst others support initial evaluation using traditional psychometric approaches based on Classical Test Theory, such as factor analysis [10,18]. Even though they are based on parametric assumptions, principal components and factor analysis are widely used in this context and have generally been considered appropriate for the initial stage of exploring and describing the relationships among a large set of variables, even where assumptions of normality may not strictly hold [19]. In this paper we present a traditional psychometric analysis. We are in the process of exploring Rasch analysis, which will be presented for publication separately.

We debated carefully whether to use parametric or non-parametric statistical analysis. According to Altman and Bland 2009, rank methods are sometimes useful, but parametric methods are generally preferable as they provide estimates and CIs and generalise to more complex analyses, especially where data may have many possible values (ie, long-ordinal data) and samples are large [20]. Factor analysis already uses parametric assumptions, and for our primary analysis, we therefore used parametric techniques (t tests) for subscale analyses where ‘long ordinal’ data (range 28 to 96 points) approximated to a normal distribution. (For completeness, an equivalent analysis using non-parametric methods is provided in S1 and S2 tables, confirming that both methods gave similar results.) Non-parametric techniques were used in any event for item-level analyses that involved ‘short ordinal’ data (range 7 points) that were typically skewed, and so would not fulfil the assumptions of parametric techniques. To allow for multiple tests, the threshold for significance of two-sided P values was taken as 0.05/number of tests.

Ethics

The UKROC database collates de-identified data as part of routine clinical practice and the programme registered as a Payment by Results Improvement Project. The analysis of this routinely-collected data is classed as service evaluation, which does not require research ethics permission in the UK.

Part 1: Psychometric Analysis

Exploratory factor analysis.

EFA was conducted on sample A. All items loaded reasonably strongly onto the first principal component with all 30 loadings >0.3. Inspection of the scree plot suggested a 3-factor solution with three principal components with Eigenvalues >1.5.

As a previous factor analysis in a general neurorehabilitation sample [10] suggested both a 2-factor solution and a 3-factor solution, both solutions were explored (see Table 2). Results showed:

• The 2-factor solution accounted for 64% of the variance and divided neatly into Motor (16 items) and Cognitive (14 items) subscales.
• The 3-factor solution accounted for 69% of the variance:
  • 15 out of 16 items again loading strongly on the first component reflecting the Motor function.
  • The 14 item cognitive subscale was split into two subscales:
    • Nine items loaded onto the second component (Psychosocial function).
    • Five items loaded on to the third factor (‘Communication’).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Principal component loadings after Varimax rotation.

https://doi.org/10.1371/journal.pone.0147288.t002

The 3-factor solution was considered the most promising model for stroke patients, as it accounted for 5% more of the variance and was readily interpretable. The only item that did not load onto any of these three components by >0.5 was ‘Swallowing’, which loaded weakly onto both the motor (0.433) and the communication (0.443) components. For subsequent analyses, it was included in the motor subscale on the basis of clinical relevance.

The internal consistency was high for the whole scale with Cronbach’s alpha = 0.96. Alpha coefficients for the Motor, Psychosocial and Communication subscales were 0.97, 0.93 and 0.88 respectively.

Responsiveness to Change.

All UK FIM+FAM subscales showed significant improvement between admission and discharge (p < 0.0001) as shown in Table 3.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Change in the UK FIM+FAM subscale scores from admission to discharge.

https://doi.org/10.1371/journal.pone.0147288.t003

The ‘composite radar chart’ “Fig 2” illustrates these changes at the individual item level. Although the largest changes were seen in the motor items (especially those reflecting mobility and continence), significant gains were seen for all items confirming the relevance of cognitive and psychosocial measurement in the stroke population.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Composite radar chart of median item scores on admission and discharge for the whole stroke population.

Legend: The radar chart (or “FAM splat”) provides a graphic representation of the disability profile from the data (n = 1539). Scale items are arranged as spokes of a wheel from 1 (total dependence) to 7 (total independence) run from the centre outwards. Thus a perfect score would be demonstrated as a large circle. This composite radar chart illustrates the median scores on admission and discharge. The shaded area thus represents the change in median score from admission to discharge.

https://doi.org/10.1371/journal.pone.0147288.g002

Part 2: Left versus Right Hemispheric Strokes

Table 4 shows the difference in UK FIM+FAM subscale scores between left and right strokes on admission. Overall, there was no significant difference in total UK FIM+FAM score between the left and right strokes; however the patterns of disability were different. After correcting for multiple tests, left hemisphere strokes showed significantly higher motor scores (Mean 5.7, 95%CI 2.7, 8.6 p<0.001), while left hemisphere strokes had significantly lower cognitive scores, primarily in the communication domain (-6.8 95%CI -7.7, -5.8 p<0.001). There was no significant group difference overall or in the psychosocial domain.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 4. Mean differences between left and right hemisphere strokes on admission.

https://doi.org/10.1371/journal.pone.0147288.t004

The differences at item level shown in “Fig 3” were analysed and the results are provided in Table 5. Patients with right hemisphere stroke had significantly lower scores for dressing, toileting, bed and car transfers, locomotion and stairs; whilst patients with left hemisphere strokes had lower levels of all five communication items, memory and orientation.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Composite radar chart of median item scores for Left and Right hemisphere strokes on admission.

This composite radar chart illustrates the median scores for Left and Right hemisphere strokes. Left Hemisphere scores are shown in red, while Right hemisphere strokes are shown in blue. Left-sided strokes had higher scores for some motor items but lower scores for communication and some cognitive items.

https://doi.org/10.1371/journal.pone.0147288.g003

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 5. Item level differences between left and right strokes.

https://doi.org/10.1371/journal.pone.0147288.t005

Study Limitations

1. The study was carried out in a selected stroke population of mainly younger adults with complex needs. It cannot be assumed that the findings would necessarily be reflected in a more typical older stroke population. However, the findings would have relevance for other countries that offer specialist rehabilitation services for selected groups of stroke patients with more complex needs.
2. The data were recorded in the context of routine clinical practice and 13% of episodes had incomplete FIM+FAM data. Although the included sample was not significantly different from the total population with respect to demographics or total functional scores, we cannot exclude the possibility of sample bias.
3. Within the main subgroups of left and right hemisphere stroke, there will inevitably be a range of pathologies that may impact the outcome. For example, the Bamford classification [32] separates strokes into total and partial anterior circulation, and lacunar strokes that are known to carry different outcomes [33]. In this sample, we know that the proportions of infarcts to haemorrhage strokes were similar for both sides of stroke, but the UKROC dataset does not include the Bamford classification or equivalent. Hence, we cannot be certain that the groups were well-matched for pathological severity. That said, however, patients referred to specialist inpatient rehabilitation are a selected sample of patients with more complex disabilities, whose recovery trajectory is likely to be slower. Therefore, we would anticipate a relatively low proportion of small vessel lacunar strokes in this study population.
4. The factor analysis was carried out on a sample where admission and discharge data was pooled together and randomised. The population was therefore heterogeneous, which may partly explain why the CFA indices were acceptable but did not meet the strictest criteria. Heterogeneity could potentially have been reduced by restricting the sample to admission values only, but we deliberately used a broader sampling method to ensure data representation across the whole scale range.
5. Although the sample size exceeded the usual standards for factor analysis, and by pooling and randomisation of the samples, we reduced, as far as possible, the relationship between samples used for EFA and CFA, they cannot be said to be fully independent. The results of CFA therefore require confirmation in a fully independent sample.

In summary, despite the above-recognised limitations, this study provides confirmation that the UK FIM+FAM is a valid instrument for use as a measure of functional independence in stroke patients. Its scaling properties are broadly similar in this group to those previously reported in a general neurorehabilitation population and in traumatic brain injury. It demonstrates deficits in cognitive, communicative and psychosocial function that change during rehabilitation. In this study, the FIM+FAM differentiated between patients with left and right hemisphere stroke in a manner that resonates with clinical experience, thus suggesting that it is an appropriate tool to use in this population, especially where the clinical team wishes to extend outcome measure beyond the simple recording of physical disability and independence in basic activities of daily living.