Background
The Modified Version of the Postural Assessment Scale for Stroke Patients (SwePASS) [
1,
2] (see Additional file
1) is an ordinal scale aimed for use in stroke rehabilitation and research. The SwePASS is a modification of the original French version called the Postural Assessment Scale for Stroke Patients (PASS) [
3] in terms of different definitions of categories. The PASS [
3], a clinical tool that assesses lying, sitting and standing, for monitoring patient’s weekly progress, was developed by Benaim and colleagues [
3], from the Fugl-Meyer Assessment of balance and mobility [
4] and the BL Motor Assessment [
5]. In the development of the PASS, poor postural performance was related to impairment in body stabilization [
3] i.e. maintain a posture and ensure equilibrium. The construct derived from the theory for the PASS, described by Benaim et al. [
3], involves terms of both postural control and postural performance. The items of the SwePASS may be linked to the categories related to activities of maintaining (d410) and changing (d415) a body position in lying, sitting and standing, as described in the mobility chapter (d4) of the International Classification of Function (ICF). Thus, to clarify the content validity, the construct that SwePASS refers to is postural control of persons with stroke, as the ability in body stabilization during every day activities of lying, sitting and standing.
The SwePASS has been shown to be a highly reliable [
1] and valid [
2] scale in acute stroke as well as to be responsive to change over time [
6]. This suggests that the SwePASS may be a reliable and a usable clinical indicator of postural control in activities of lying, sitting and standing
. Whether the SwePASS scale can be regarded as unidimensional, is not clear. Unidimensionality of the scale will support the assumption that one construct is assessed. Unidimensionality is required in order to sum the item scores into a valid total score. For the total score to be a valid indicator of the person’s ability (i.e. does it makes sense to report a single score for a patient’s performance on the SwePASS?) there must be evidence of the internal construct validity and reliability of the SwePASS. One way to provide evidence that the total sum score of the items holds adequate internal construct validity is to use Rasch analysis [
7‐
10]. By using this method, one can also change a nonlinear transformation of ordinal raw scores into interval measures, assuming that fit to the Rasch model is supported. In addition, using Rasch analysis is also about testing hypotheses. The hypothesis is that the SwePASS is measuring one single construct, here postural control as the patient’s ability in body stabilization (that is, maintain a posture and ensure equilibrium) during every day activities of lying, sitting and standing. The hypothesis about the hierarchy of the items in the SwePASS concerns that the items in sitting and lying are the easiest ones and that the items in standing are the most difficult ones. The aim of the present study is to evaluate the construct validity of the SwePASS, more specifically to evaluate the following measurement properties: unidimensionality, targeting of persons and items, category function, local independence, invariance, and the reliability of the SwePASS using Rasch analysis [
7].
Discussion
At this initial validation stage, the Rasch analysis provided evidence that SwePASS did not work as originally intended. On the basis of these initial findings, the scale was modified. After adjustments of categories and local dependencies, this first Rasch analysis of the SwePASS shows that the SwePASS could be considered as unidimensional and that it works as a global measure of postural control as the ability to body stabilization during every day activities at an individual level in patients with stroke, regardless of gender, age and location of stroke lesion.
At the same time, the initial analysis highlighted a number of potential problems. For instance, 4 of 12 items showed disordered thresholds. From a clinical perspective this is unsatisfactory, both for the patient and for the physiotherapist, who must be able to trust the results of the measurement instrument. Particularly, it is important for a physiotherapist to know that a scale has sound measurement properties based on items that reflect the hypothesized structure and hierarchy of the construct (a logical progression of the ability to body stabilization). An improvement in ability or a good ability shall give a higher or a high score, while deterioration in ability or a low ability must give lower or a low score. Using a scale with disordered thresholds, a change in categories will be difficult or even impossible to evaluate. A misjudge of a patient’s postural control may have consequences on the patient’s self-efficacy and to safety. Disordered thresholds, which may result from several different factors, are likely caused by the item designer as a failure of the hypothesis behind the items. Counter-intuitive, this was addressed by re-scoring according to statistical criteria as well as to clinical judgement. For instance, in Item 4 (‘Sitting without support’), raters may find it difficult to discriminate between the two first categories (0 = Cannot sit; 1 = Can sit with slight support, for example by 1 hand), as the word “slight” may be too vague despite the extended explanation (‘by one hand’), thus leading to disordered thresholds. The new created observational framework of 3 categories for this item may be discriminated more easily by the physiotherapist (Cannot sit or require support for sitting / Can sit without support >10 s / Can sit without support > 5 min). In Item 7 (‘Standing without support’), the category 3 does not seem to add much information of postural control. Furthermore, in Item 8, (‘Standing on the non-paretic leg’), the raters may find it difficult to between the narrow time ‘cut-offs’ in category 1 and 2. By contrast, Item 10 (‘Standing, picking up a shoe from the floor’), the judging does not include any time-restriction, but instead includes a factor related to need of support. Clearly, picking up a shoe from a standing position is a demanding task. If either the patient or the physiotherapist has doubts about the patient’s chance to perform the task, they may avoid trying, automatically resulting in a score of 0. Standing and picking up a shoe is both a complex and dual task, why being able to implement it, whether with one or two persons support, means more ability to body stabilization than not being able to do it at all.
In contrast with the Items 4, 7, 8 and 10, the rescoring of Items 1-3 was based only on clinical implications and not disordered thresholds. Theses 3 items (rolling on the sides and postural change from supine to sitting) were rescored, as we considered that for the purpose of measuring postural control in such relatively easy activities the differences between 0 (cannot perform the activity) and 1 (can perform the activity with the help of two persons) would be minimal. Indeed, after these modifications, the difficulty hierarchy of the items changed and became more consistent with the expected progression in capability of postural control.
The local dependency revealed in the SwePASS between Item 6 ‘Standing with support’ and Item 11 ‘Sitting down from standing up’ might be explained by the fact that the two items both test the ability to stand up. To work as intended, the scale should not include one or more similar items, both in terms of affection of estimation as well as of time efficiency and effort. As a consequence of the local dependency revealed in SwePASS, the items 6 and 11 were combined into a testlet. This solution led to a fit to the Rasch model, with only a marginal reduction of the reliability value, indicating that the SwePASS has the potential to be used at the individual level.
Since stroke patients represent a very heterogeneous group, the result demonstrating nonexistence of DIF is important and positive for the clinician. In SwePASS, the responses of persons to items are determined only by the patients’ ability of postural control and are not influenced by their gender, age or stroke location. Consequently, de facto facilitates the interpretation of the scale.
According to the item location, defined in logits and presented in Table
2, it is fully expected that the items associated with supine and sitting are less demanding on postural control than the items associated with standing positions. As expected the two items involving the procedure of standing on one leg turned out to be the most difficult ones. As the clinical experience suggest, it is also expected that the ‘Supine to affected side lateral’ (Item 1) is less difficult than ‘Supine to non-affected side lateral’ (Item 2).
Recently, La Porta and colleagues performed a Rasch analysis of the Berg Balance Scale (BBS) [
25]. BBS is another clinical scale to assess balance in elderly [
26,
27], including patients with stroke, with several similar items as in the SwePASS. In that Rasch analysis, disordered thresholds were detected in 11 of 14 items [
25]. Thus, compared to the current study of SwePASS, the problem with disordered thresholds was even larger in the BBS.
There are three major limitations of the study which need to be discussed, namely the suboptimal targeting, the sample size and the inability to provide a transformation table from raw scores to interval data. The results of this study show that the SwePASS is not optimally targeted to the examined population. There is a need in both clinic and research to assess even the patients that are most severely affected. This is done in clinic. However, the most severely ill patients were not included in the study. This was partly based on the view of the responsible physician that it might be unethical to ask for informed consent in a medical more or less critical condition or at least in a chaotic situation for the relatives. If these patients had been included, it is plausible that more patients had demonstrated a lower capacity in postural control, as shown at the lower end of the scale. The suboptimal targeting is not only a limitation of the study, but also a negative finding. The small number of data (participants) in this study (n = 150) and the poor targeting means no optimal power to detect and adjust for problems in the scale, only give indications of possible problems. Relative to the sample size, even if the data satisfy the Rasch model expectations, the raw score obtained from the scale could not be converted into an interval scale whose estimated unit of measurement is the logits. The transformation of ordinal data into interval data, leading to the possibility of using parametric statistical analyses, requires a larger population than was used in the current study. As a guiding principle, Linacre [
28,
29] has proposed a sample size of 250 for accurate estimation and appropriate degree of precision using Rasch analysis, regardless of targeting of persons to the scale. From a clinical point of view, interval data would simplify the interpretation of change in postural control using SwePASS, as well as allowing a presentation of the results in terms of a sum score. In addition, there is always a risk of type I or type II errors when analysing data with Rasch analysis and this risk becomes much higher if the sample is small and not well targeted, as in this study. However, even considering that some findings of the analysis are highly significant, further studies with larger samples are needed to confirm the findings. Before then, it cannot be advocated that that categories of the SwePASS should be changed.
A review on balance scales in 2009 [
30] showed that there were around 30 instruments measuring balance and postural control, and this number is constantly increasing [
31,
32]. The SwePASS is specifically developed for persons with stroke, modified according to clinical relevance and has also been shown that be quick to perform in clinic [
1]. SwePASS has been investigated in different ways with methods that are recommended to handle ordinal data, concerning its psychometric properties regarding previously presented results of reliability [
1] and responsiveness [
6] as well as the current internal construct validity. The psychometric evidence of the predictive validity [
2] may be interpreted very cautiously. Even if the total score was used dichotomized, it can be considered bias before dealing with adjustments of scoring categories and local dependency. Overall, the new scale, the SwePASS, may be a scale of interest to both the clinician and the researcher.
Competing interests
The authors declare that they have no competing interests.
Authors’ contributions
CUP contributed to the study design, acquisition, analysis and interpretation of data, drafting and editing of the manuscript. KSS contributed the overall study design, acquisition, analysis and interpretation of data and manuscript editing. ÅLN contributed to the analysis and interpretation of data and manuscript editing. All authors’ have read and approved the final version submitted for publication.