Characteristics that affect score reliability in the Berg Balance Scale: a meta-analytic reliability generalization study.

Introduction: The Berg Balance Scale is the most widely used clinical scale used to assess balance performance in neurological conditions. Reliability is not an unalterable value of a scale across different applications, but is a property referring to the results obtained with the scale, not the instrument in itself. The results of a scale can vary depending on the context where it is applied and the characteristics of the participants, therefore the reliability coefficient should be reported in each study. A systematic review and meta-analysis of the reliability coefficients obtained in different applications of the test with the data at hand is the best method to examine how the reliability of a test scores varies. The objectives of this systematic revision are: To determine the mean of internal consistency, intra and interrater reliability of the Berg balance scale in the clinical, nonclinical and mixed populations, to determine the methodological and substantive characteristics and to propose a predictive model enabling researchers and clinicians to use it in the future to estimate the expected reliability based on the characteristics of the most relevant studies.
Evidence Acquisition: The MEDLINE (Pubmed), EMBASE and CINAHL databases were searched from 1989 to 2015. Two reviewers independently selected empirical studies published in English or in Spanish that applied the Berg Balance Scale and reported any reliability coefficient.
Evidence Synthesis: The 80 samples in the 65 studies reported any the Berg Balance Scale reliability estimate. Coefficient alpha ranged from 0.62 to .98, with a mean of 0.92. For intra-rater agreement, the mean intraclass correlation was ICC+=0.957, and for inter-rater agreement ICC+=0.97. The SD of the Berg Balance Scale scores presented statistically significant relationships with the coefficient alpha and with ICC (intra-rater). In addition, the clinical population and the institutionalized population presented statistically significant relationships with the coefficient alpha. The sample size and mean scale results were also statistically related to ICCs (intra-rater).
Conclusions: The alpha coefficient and intra- and inter-rater agreement for Berg Balance Scale scores was very satisfactory. Several characteristics of the studies were statistically associated to the alpha coefficient and with intra-rater reliability.