Data analysis plan

Analysis of effectiveness on the primary and secondary outcomes will use intention to treat, and will be blinded for intervention allocation by having a researcher not involved in the analyses recode the groups using two random numbers (eg, groups ‘1’ and ‘2’, instead of ‘intervention’ and ‘control’). Our primary outcome will be compared between groups as the number of falls per person-year using Poisson or negative binomial regression (depending on the data distribution). In secondary analyses, complier averaged causal effect analysis will be used to correct for imperfect adherence, and the proportion of fallers will be compared between groups using modified Poisson or logistic regression (depending on the data distribution). The effect of group allocation on secondary outcomes will be analysed using generalised linear models. The effect of in-person versus telehealth assessment will be explored using a covariate accounting for method of assessment.

Economic analysis will be conducted from a health and community care provider perspective and comprise a cost-effectiveness analysis and a cost–utility analysis.⁵⁶ The analysis will include costs associated with intervention delivery and healthcare utilisation, and the outcomes of interest will be the number of falls prevented and health-related quality of life. Bootstrapping will be used to estimate a distribution around costs and health outcomes, and to calculate the confidence intervals around the incremental cost-effectiveness ratios. A cost-effectiveness acceptability curve will be plotted to provide information about the probability that the intervention is cost-effective at different willingness to pay thresholds.

Data will be regularly monitored for omissions and errors. All analyses will follow a detailed statistical analysis plan, which will be publicly registered prior to analysis of the data. We will not have to account for missing data in the primary outcome as a correction for follow-up duration is part of the analysis. Missing data in other outcomes will be imputed using estimated means or multiple imputation depending on the mechanism of missingness. In case of a large proportion of missing data, which is anticipated for some outcomes due to our switch to telehealth, we will report both imputed and complete cases results where appropriate.