Statistical analysis

Questionnaires were completed online and there was no question item missing data. Descriptive statistics were used to summarise the number of patients approached, screened, eligible, consented, and randomised. Reasons for non-consent, exclusion, and drop-out, at each stage of the study, were recorded. Similarly, descriptive statistics were computed to report adherence to the intervention.

Internal consistency of the measures was assessed using the Cronbach alpha coefficient at both T0 and T1. The EQ5D health state utility score was calculated from individual health profiles using the value set for England [9]. Mean and standard deviations (SD) are provided for all self-report outcomes by visit and by treatment. Estimates of treatment effect at T1 were based on an analysis of covariance (ANCOVA) to estimate the postintervention mean difference. The analysis adjusted for the baseline level of the outcome variable, baseline DHI, age and sex. Group allocation was included as an indicator variable following the intention-to-treat principle. Given the feasibility nature of the trial, with a small sample size not powered to detect between group differences, the statistical significance of any post-randomisation group differences was not assessed; instead, effect sizes were calculated as standardised mean differences using Hedge’s g (SMDg) applying the small sample bias correction factor [15].