This study developed a balance exercise programme for PwMS, based on an expert round of 13 experienced physiotherapists and a study group of 66 PwMS. The three main findings of the study are as follows: first, an expert round successfully established the content validity of the proposed balance dimensions and the allocation of 19 key balance exercises to these dimensions. Secondly, fit to the Rasch model was adequate if dimension 3 “stepping” and dimension 4 “walking” were combined, since, together, these two dimensions formed an overarching unidimensional construct. Thirdly, the difficulties of the balance exercises were adequate to cover the ability spectrum of the PwMS (i.e. adequate targeting). In summary, it was possible to create a balance programme for the proposed balance dimensions “stable BOS”, “sway” and “step and walk”.
Regarding the content validity of the proposed set of balance exercises, the exercises were explored from the perspective of health professionals and of PwMS, as proposed by the Cosmin group [49, 50]. The study included professionals with considerable experience in prescribing exercises for PwMS. Only physiotherapists were included in the expert round, as this group was considered to be particularly relevant for prescribing balance exercises for PwMS. Evidence for adequate content validity was analysed regarding relevance (i.e. the included exercises were relevant for the respective balance dimensions and the population of interest), comprehensiveness (i.e. key exercises for each balance dimension were integrated), and comprehensibility (i.e. PwMS and health professionals rated instructions and response options as understandable). The perspective of PwMS regarding the exercises was evaluated after each exercise performance, and participants provided feedback on the clarity of the exercise description and instructions, safety aspects, balance exercise, image quality and clarity, and the possibility of performing the exercise at home and adapting the exercise to their home surroundings. The perspective of PwMS will be used to adapt the balance programme within the categories reported above.
Rasch analysis found that the item fit of all 19 balance exercises to the rating scale model was adequate; i.e. no balance exercise was excluded based on the item fit statistics. However, this was only seen when the balance dimensions were analysed separately. Therefore, every balance exercise provides information about the respective latent dimension. Similarly, the assumption of unidimensionality was fulfilled only when the balance exercises were analysed in three separate latent dimensions. When all exercises were analysed together the PCA of residuals exceeded the threshold of 2 eigenvalues, which was used as an indicator for unidimensionality. From a clinical point of view, the separation of balance exercises into distinct dimensions is supported by evidence; e.g. Shumway-Cook and Woollacott  proposed a systems model of postural control, which emphasizes that postural control for stability and orientation requires a multitude of neural and musculoskeletal systems. Therefore, the dimensions of balance exercises in the current study require a similar interaction of postural control systems within each of the balance dimensions. Furthermore, PwMS with balance problems can increase their balance abilities within each dimension separately, making targeted training possible.
In addition, the proposed latent traits of the balance exercises in the current study are very similar to the three categories of balance problems proposed by Cameron and Nilsagard . The first two traits are almost identical. Only the category “delayed responses to postural displacements and perturbations” differs slightly from the current proposed latent trait “step and walk”, because, in order to design a programme that could be carried out at home or with minimal external help, external perturbations were not included in the balance programme in the current study. However, all the exercises in the third category require participants to control their balance during postural displacements.
Other authors have analysed unidimensionality of balance tests and have confirmed unidimensionality over a wide range of balance tests, which are similar to the balance exercises used in the current study. For example, La Porta et al.  reported that 12 of the Berg Balance Scale items showed unidimensionality when evaluated in samples of different aetiologies in neurological rehabilitation. A further example is a study by Franchignoni et al. , which reported that the 14 items of the Mini-BESTest showed unidimensionality in a heterogeneous neurological sample. The different findings regarding unidimensionality might be the result of a difference in study samples (e.g. heterogeneous clinical samples versus only PwMS), a difference in item characteristics (e.g. focus on ability testing versus focus on exercise performance), and different methodological criteria to confirm unidimensionality.
Analysis showed that targeting of the balance exercises was adequate; i.e. the range of balance exercise difficulty covered the ability estimates of most participants. However, there was a lack of very difficult exercises for dimension 2 “sway” and of easy balance exercises for dimension 3 “step and walk”.
To our knowledge, this is first study to evaluate the targeting of balance exercises in PwMS. These data might help to improve the effectiveness of balance exercises in PwMS by enabling better targeting of exercises to participants’ abilities. Several RCTs have analysed the effectiveness of balance exercises in PwMS . Some studies aimed to tailor the difficulty of balance exercises to the participants’ ability, with the selection of exercises based on clinical reasoning. For example, Cattaneo et al.  chose motor and sensory training modalities based on the individual’s abilities. To support health professionals in the selection of exercises the difficulty of balance exercises should be presented on a clear progression line. In addition to an unambiguous progression in exercise difficulty, the distance in difficulty between the exercises should be stated clearly. This could facilitate a more objective allocation of exercises.
The physiotherapists of the expert round ranked the exercises regarding the difficulty similar to the ranking based on the Rasch measures (logits), however there were some discrepancies. For example, in dimension 2 there was a larger discrepancy in the exercise e11 “rolling ball forwards”. This may be because the difficulty of the exercise varies greatly depending on how it is performed, i.e. whether the ball is used as a base of support or rolled forward without pressure. Therefore, the standardisation and instructions for the exercises need to be improved.
This study has several limitations. First, during the process of establishing the content validity of the balance programme quantitative methods were used, such as surveys. No qualitative research methods were used to establish content validity. In addition, the sample size of the expert round (13 physiotherapists) was relatively small, and did not include other professions, such as sports scientists. In contrast, 65 PwMS provided data for the content validity of the balance exercises.
A further limitation was the relatively low number of balance exercises integrated into the Rasch analysis. This was based on pragmatic reasoning; the PwMS should be able to complete the set of balance exercises within a single session. Adequate structural validity was reported for these 19 exercises, which can be used as “key exercises” in the balance programme, while clinicians can also integrate modifications to increase the difficulty of each exercise. For example, by incorporating additional head, eye, arm, leg or trunk movements, a change in surface conditions, dual tasks, or reduced visual information.
A further potential limitation is the focus of the balance exercises. Cattaneo et al.  reported on two different balance exercises programmes (training of motor or sensory strategies). Within the current study, the exercises can be classified into the category “training of motor strategies”. Therefore, this study could not report on the difficulty of exercises that alter the sensory environment. We propose to integrate such modifications into the balance exercises with the aim of increasing their difficulty. For example, exercise e6 “tandem stance” has a difficulty of 1.99 logits. If PwMS are asked to perform this exercise with eyes closed or on an unstable surface, the difficulty will increase and the exercise will be more challenging. However, further research is needed to determine by precisely how much the difficulty will increase.
An additional limitation was identified in analysing the threshold values of the balance exercises. The difference in logits between thresholds 1, 2 and 3 were very small. This was also observed during the measurement sessions. For some participants it was challenging to score the difficulty of the exercises. In particular, the categories “very easy” and “easy” were difficult to separate. The scoring system of the subjective difficulty should be investigated further and adapted in future studies. A possible solution would be to combine these options, although this method is controversial. This was not done in the current study because some authors suggest that data should be re-measured after changes to the scoring system .
Furthermore, the local dependency (i.e. high residual correlations) of some items (especially in dimension 2) needs to be addressed in the future development of this balance exercise programme. We were not able to precisely estimate the local dependency within our data set, because less than the required 20 exercises were analysed together within one dimension .
Dimension 2 showed a low person separation index and reliability. However, this is probably due to the low number of items (n = 4)  and in line with the standardised reliability (i.e. reliability standardised on 50 items), which was 0.92. More items are needed to increase the ability to distinguish between low and high performers in dimension 2.
There were disordered categories in two dimensions. Possible reasons for disordering are: a) the low counts in the categories could lead to random errors or idiosyncratic findings, b) there could be a problem with the rating scale, e.g. difficulty to understand the meaning of the response options. Since the response options were the same for all items, but the disorder was not consistent, we believe that the low number of observations, which increase the standard error of the estimates, was responsible for the problem . Furthermore, because only three categories showed disordered categories and because we would like to keep the response options the same over all exercises, we did not perform a recoding of these categories.
A limitation in the current study was that differential item functioning was not analysed. This analysis was not done because the sample size was not large enough to conduct an adequate analysis. Further research is needed to explore whether exercise progression might be different in specific subgroups. For example, Sosnoff et al.  reported that fall risk differed between groups of PwMS. Factors such as cerebellar or brain stem lesions increased the fall risk. Similarly, PwMS with impaired visual function showed greater balance impairments. Therefore, a larger well-powered study should investigate whether the exercise progression is comparable between these subgroups, or if the balance programme should be modified for each subgroup.
Implications for practice
The proposed balance programme is one aspect of a multicomponent rehabilitation programme aiming to decrease fall risk in PwMS . However, as balance impairments are reported to precede mobility impairments in PwMS , it can be assumed that a targeted balance exercise programme is especially valuable for this population. The 19 key exercises described here can be used as the basis for an extensive home-based exercise programme. The established difficulty estimates for each balance exercise can be used by health professionals to identify the optimal challenge point for training of balance abilities in PwMS. To increase the challenge, each exercise can be modified as follows: arm movements (slow and fast in different planes), trunk and head movements, eyes closed, or addition of secondary motor or cognitive tasks (dual task). Furthermore, different surface conditions can be used (from stable to unstable). The exercises, including adaptations to the difficulty levels, could be implemented via web or tablet applications, including videos and more detailed instructions, and with the option of gathering feedback on difficulty from patients.
Implications for research
The findings of this study should be investigated further in larger studies. In addition, the response options for the difficulty estimates should be modified, due to the limited information about difficulty on several threshold values (i.e. the difference between thresholds 1–3 was very small). A possible solution would be to combine these response options.
An interesting approach for further research would be to assess the difficulty levels of all exercises (including the adaptations) and to develop recommender systems analogous to computer adaptive testing. After each rating of exercise difficulty, the computer would suggest the most appropriate exercise to the PwMS.
In addition, differential item function should be evaluated in potential subgroups, such as PwMS with or without cerebellar lesions, spasticity and different forms of MS (e.g. primary progressive MS or relapsing-remitting MS).