Data analysis

The correlations between the scores from the test and videotape were examined using the Spearman rank-order correlations. Correlation coefficients of 0.80 or higher indicated excellent correlation. Those of 0.50 to 0.79 indicated moderate correlation and those 0.00 to 0.49 indicated poor correlation (26). Intra-rater and inter-rater reliability were calculated using interclass correlation coefficient (ICC) model 3, k and 2, k, respectively, for the S-BESTest, Brief-BESTest and Mini-BESTest (27). The ICC values were interpreted using the criteria: 0.8 indicates good reliability, 0.8 to 0.6 indicates moderate reliability and 0.6 to 0.4 indicates weak reliability (20, 27).

The concurrent validity of the S-BESTest, Brief-BESTest and Mini-BESTest was assessed with the BBS using the Spearman rank-order correlations. Floor and ceiling effect of S-BESTest, Brief-BESTest, Mini-BESTest, and BESTest were calculated as the percentage for minimum or maximum possible scores of the sample scoring, respectively. Floor and ceiling effects greater than or equal to 20% were interpreted as significant (28). Comparisons of balance scores between baseline and two weeks post-rehabilitation and between two weeks and four weeks post-rehabilitation were analyzed by using paired t test with significance level at p< 0.05.

Internal and external responsiveness of the S-BESTest, Brief-BESTest, Mini-BESTest, and BESTest were assessed. Internal responsiveness refers to the possibility of detecting any statistical change before and after a known treatment. The internal responsiveness was examined using the standardized response mean (SRM) and minimal detectable change (MDC) (29, 30). SRM of 0.8 or greater represented a large change, values from 0.5 to 0.8 represented moderate change, and values of 0.2 to 0.5 represented small change. MDC was calculated as the SEM multiplied by 1.96(√2) (29, 30). SEM was calculated as SD multiplied by √(1- reliability). The limitation of internal responsiveness is that it lacks information on the quality of changes, such as worsening or improvement (31).

In contrast, external responsiveness is associated with the concept of clinical relevance, which depends on the choice of external standard (32). In this study, 2 external standard scales; the BBS and the GRC, were selected to compare between the change in performance and patient’s own perception. External responsiveness was assessed by using Receiver Operator Curve (ROC) analysis to establish which version of the BESTest could best identify patients whose balance had improved using a change in BBS score of 7 points as the milestone value for deciding if change had occurred (33, 34). The Area Under the Curve (AUC) value was used to reflect this. The AUC values were compared across test versions using a t-test and significance level of 0.05. The ROC analysis was repeated using the change in GRC (5 points) (25). The optimal cutoff score was also chosen from the sensitivity and specificity (27). The area under the curve (AUC) was used to interpret the probability of correctly discriminating between patients with and without balance improvement (29). An AUC of 0.8 or greater indicated excellent discrimination (27). Paired t test was used to compare the AUC between 2 testing scales with significance level at p< 0.01. A likelihood ratio demonstrates accuracy of post-test probabilities; values of LR+ above 5 and values of LR- below 0.2 were considered meaningful (27).