Investigating the effect of an online enhanced care program on the emotional and physical wellbeing of patients discharged from hospital with acute decompensated heart failure: Enhanced care program for heart failure

Background: The Enhanced HF Care study is a prospective randomised open blinded endpoint (PROBE) design trial comparing the effectiveness of a digital health intervention --Enhanced HF Care -- to usual care, in increasing health related quality of life for patients hospitalised with acute decompensated heart failure (ADHF). The study aims to evaluate effectiveness in (i) improving emotional and physical wellbeing, and (ii) decreasing healthcare utilisation. This statistical analysis plan outlines the pre-specified statistical principles and procedures for analysing data from the trial, which has been updated from as described in the trial protocol to reflect lower than anticipated recruitment rates. Methods: The co-primary outcomes are emotional and physical wellbeing measured using the Minnesota Living with Heart Failure Questionnaire (MLHFQ) domains at 6-months post-recruitment. Secondary outcomes include MLHFQ emotional and physical well-being at 1-month post-recruitment, unplanned hospital readmissions, and emergency department presentations. Statisticians conducting analyses are blind to treatment allocation. Analysis: Bayesian hierarchical mixed effects models will be used to estimate treatment effects for all outcomes, with uninformative prior distributions specified for effect parameter and half-Cauchy priors on the random effect standard deviation and the within person standard deviation. For each outcome, the analysis will present mean differences between treatment groups with 95% credible intervals (highest posterior density) and posterior probabilities of treatment effect. All randomised participants will be analysed according to their assigned treatment group following an intention-to-treat framework, excluding only those who withdraw consent for data use. Intercurrent events will be handled using the estimands framework. For MLHFQ outcomes, a composite strategy will be employed where participants who die will be assigned the maximum value reflecting worst possible quality of life. Healthcare utilisation outcomes (unplanned hospital readmissions and emergency department presentations) will be analysed using a 'while alive' strategy, with follow-up extending until either death or 6 months post-randomisation, whichever occurs first. For handling missing data, we will use multiple imputation by chained equations (MICE) under a missing at random assumption when participants drop out. For completed MLHFQ subscales specifically, if only one item is missing within a subscale, we will use mean imputation for that item. However, if more than one item is missing form a subscale, the total score will be set to missing and MICE will be used to impute the total subscale score. Additional analyses will estimate the posterior predictive probability of trial success had the target sample size been achieved. Conclusion: Publication of this statistical analysis plan prior to data analysis ensures transparency in the analytical approach and minimises potential bias in the interpretation and reporting of trial results, particularly given the adaptation to Bayesian methods necessitated by lower-than-expected recruitment.