Statistical Methods
We conducted an intention-to-treat analysis for all randomized
participants. Rubin’s method of multiple imputations was used to
estimate the treatment effect by combining five imputed values to
complete missing data. Continuous parameters were compared using the
Wilcoxon Mann-Whitney test; Hodges-Lehmann estimate was used to
calculate a 95.0% confidence interval. Categorical variables were
compared using the Fisher Exact test.
We also performed ancillary analysis by comparing THI subscales, using a
multivariate regression model to adjust for the possible effect of
‘education level’ and ‘guardian help to fill the THI score’ on THI
responsiveness, as well as ‘mild hearing loss’ and ‘psychiatric illness’
on clinical performance, and performed a sensitivity analysis to test
the efficacy in those participants who completed the study, the
per-protocol population.
We used Stata software (version 16.1) for all statistical analyses.