When ‘good’ is not good enough: a retrospective Rasch analysis study of the Berg Balance Scale for persons with Multiple Sclerosis

BACKGROUND: The Berg Balance Scale (BBS) is one of the most used tools to quantify balance in Persons with Multiple Sclerosis, a population at high risk of falling. AIM: To evaluate the measurement characteristics of the BBS in Multiple Sclerosis through Rasch analysis. DESIGN: Retrospective study. SETTING: Outpatients in three Italian Rehabilitation centers. POPULATION: Eight hundred and fourteen persons with Multiple Sclerosis able to stand independently for more than 3 s. METHODS: The sample (N = 1,220) was split into one validating (B1) and three confirmatory subsamples. Following the Rasch analysis performed on B1, the item estimates were exported and anchored to the three confirmatory subsamples. After obtaining the same final solution across all samples, we studied the convergent and discriminant validity of the final BBS-MS using the EDSS, the ABC scale, and the number of falls. RESULTS: The base analysis on the B1 subsample failed the monotonicity, local independence, and unidimensionality requirements and did not fit the Rasch model. After grouping locally dependent items, the BBS-MS fitted the model (χ(2)(8) = 23.8; p = 0.003) and satisfied all requirements for adequate internal construct validity (ICV). However, it was mistargeted to the sample, given the striking prevalence of higher scores (targeting index 1.922) with a distribution-independent Person Separation Index sufficient for individual measurements (0.962). The B1 item estimates were anchored to the confirmatory samples with confirmation of adequate fit (χ(2) = [19.0, 22.8], value of ps = [0.015, 0.004]) and satisfaction of all ICV requirements for all subsamples. The final BBS-MS directly correlated with the ABC scale (rho = 0.523) and inversely with EDSS (rho = −0.573). The BBS-MS estimates significantly differed across groups according to the pre-specified hypotheses (between the three EDSS groups, between the ABC cut-offs, distinguishing ‘fallers’ vs. ‘non-fallers’, and between the ‘low’ vs. ‘moderate’ vs. ‘high’ levels of physical functioning; and, finally, between ‘no falls’ vs. ‘one or more falls’). CONCLUSION: This study supports the internal construct validity and reliability of the BBS-MS in an Italian multicentre sample of persons with Multiple Sclerosis. However, as the scale is slightly mistargeted to the sample, it represents a candidate tool to assess balance, mainly in more disabled people with an advanced walking disability.