RELEASE: REdressing Long-tErm Antidepressant uSE in adults in general practice (RELEASE): A two-arm parallel-group cluster RCT effectiveness-implementation hybrid type-1 study

Study identifiers

• Wallis, Katharine A., Donald, Maria, Horowitz, Mark, Moncrieff, Joanna, Ware, Robert S., Byrnes, Joshua, Thrift, Karen, Cleetus, MaryAnne, Panahi, Idin, Zwar, Nicholas, Morgan, Mark, Freeman, Chris, and Scott, Ian (2023). RELEASE (REdressing Long-tErm Antidepressant uSE): protocol for a three-arm pragmatic cluster randomised controlled trial effectiveness-implementation hybrid type-1 in general practice. Trials 24 (1) 615 1-14. https://doi.org/10.1186/s13063-023-07646-w ¹

• Australian New Zealand Clinical Trials Registry Number: ACTRN12622001379707p

• Ethical approval: The University of Queensland Human Research Ethics Committee (2022/HE001667)

Protocol variation

The study was initially designed as a two-arm cRCT (effectiveness-implementation hybrid type 1) testing RELEASE vs Usual Care in 24 general practices. Due to additional funding received after the start of the trial and after two practices had been recruited, in mid 2023 the trial transitioned from a two-arm cRCT to a three-arm cRCT to test both RELEASE and RELEASE+ (a more intensive version of RELEASE) versus Usual Care in 28 general practices.

However, due to delays and hurdles in getting signed agreements for development of the RELEASE+ App, we had to pare-back the App from that originally planned to the extent that we now expect the App to add little to the effectiveness of the RELEASE intervention. Therefore, in mid 2024 we changed the analysis plan from RELEASE v Usual Care and RELEASE+ versus Usual Care to RELEASE and RELEASE+ combined versus Usual Care, reflecting that the primary outcomes of interest are the between-group comparisons of RELEASE and RELEASE+ combined versus Usual Care. Thus, while the hypothesis for three-arm trial was that the RELEASE and RELEASE+ interventions were sufficiently different to justify a three-arm trial, the hypothesis was changed to be that the RELEASE and RELEASE+ interventions are essentially the same and will have the same, or similar, effect. This change occurred after participant enrolment began and has no impact on current participants in the trial.

1.1 Study objectives

1.2 Outcomes

1.3 Intervention/s

• Usual Care: practices may provide medication review as usual. This is permitted as in both arms of the trial all prescribing decisions are made as usual by GP and patient together.

• RELEASE: The RELEASE interventions are user-informed multi-modal interventions based in general practice targeting both GPs and patients. The nub of the RELEASE intervention/s is the drug-specific hyperbolic tapering plans which have the following implementation strategies:

1.4 Setting

Twenty-eight GP practices will be recruited through The University of Queensland’s Practice-based Research Network (UQGP Research) and other networks using email, telephone calls and outreach visits. These practices will be in south-east Queensland.

1.5 Randomisation

Randomisation is clustered at the practice level to avoid contamination between intervention and Usual Care arms. Practice randomisation occurred after baseline data collection to avoid bias in GP de-selection of potentially eligible patients. Randomisation was designed assuming a three-arm trial in a 3:2:2 Usual Care:RELEASE:RELEASE+ ratio via a central web-based randomisation service (Griffith University). Practices were stratified by size (>5 / <5 FTE GPs). Randomisation occurred in blocks of size 7 or 14, with block size randomly assigned in a 1:1 ratio. The allocation ratios were chosen with regard to optimising sample size for the primary outcome, as well as the secondary comparison within the RELEASE groups.

History of Study: Due to the additional funding received after the start of the trial and subsequent transition from a single RELEASE intervention to the standard RELEASE and RELEASE+ interventions, the first two practices recruited were randomised and allocated into one of two arms (Usual Care or (standard) RELEASE) and thereafter practices were randomised and allocated into Usual Care, RELEASE or RELEASE+.

1.6 Data collection

Participant e-survey data at baseline, 6- and 12-months is collected via the secure research data collection tool REDCap (Research Electronic Data Capture, Vanderbilt, USA). For the baseline data, REDCap generates an automated email to participants with a ‘closed’ link to the baseline e-survey.

For the 6- and 12-months data, practices send an SMS to participants three days before the e-survey due date to foreshadow an email coming from REDCap inviting participants to complete the 6- and 12-month e-surveys. REDCap is programmed to email all participants on the survey due date to invite survey completion. For participants who do not respond to this email invitation, REDCap sends a reminder email on days 7, 14, and 21 following the due date; and an SMS reminder on day 4 and if no response again on day 9. For participants who have not responded, the research team phone to invite participation at two weeks after the due date. REDCap is programmed to only send email and SMS reminders to participants who have not yet completed their survey.

1.7 Sample size

To calculate sample size, we assumed 12% of participants in the Usual Care group and 30% of participants in each of the intervention groups will achieve the primary outcome (stopped antidepressants by 12 months). Because the study was designed to involve two comparisons, Usual Care vs RELEASE and Usual Care vs RELEASE+, we designed the study to allocate more participants to the Usual Care arm as this increases the overall efficiency of the trial (that is, for fixed type I and type II errors we require fewer total participants). The ratio chosen for allocation was 3:2:2. Using a two-sided alpha significance level of 0.035 (chosen to facilitate comparison of the two intervention arms against a single control arm) and 90% power then, without accounting for clustering, we would require 142 participants in the Usual Care arm and 95 in each of the intervention arms (332 total).

As this is a cluster RCT, we need to account for the non-independence of observations from individual participants recruited from the same practice. We assume there will be a mean of 20 participants recruited from each practice who will provide data on the primary outcome, and that the within-practice intra-cluster correlation = 0.03, giving a design effect of 1.57. Consequently, we require 522 participants to provide data. To achieve this number, and to account for the possible drop-out of practices once enrolled, we aimed to recruit 28 practices. Assuming 20% attrition, we required 653 participants to enroll and provide baseline data. This requires the 28 practices to recruit an average of 23.3 patients.

If the study is considered as a two-arm trial, and continuing to assume 12% of participants in the Usual Care group achieve the primary outcome, these recruitment figures give 90% power to detect a statistically significant difference (two-tailed alpha=0.05) between the RELEASE combined and Usual Care groups of when at least 26% of the RELEASE combined groups have ceased anti-depressants at 12-months.

2.1 General principles

Outcomes will be analysed on an intention-to-treat basis. The unit of analysis is the individual patient.

All models that investigate patient-level effects will include the stratification factor practice size as a fixed effect, and GP practice as a random effect. Longitudinal associations will be investigated using analyses that account for multiple observations per participant. Individual analyses will be determined by data structure, for example, three-level hierarchical mixed-effects models with ‘patient’ nested within ‘GP’ nested within ‘GP practice’ where appropriate. With regard to missing data, the structure and pattern of missing data will be assessed and if necessary, a sensitivity analysis undertaken based on imputed data using a multiple imputation model (see below). Data will be stored de-identified.

2.2 Evaluation of demographic and baseline characteristics

2.3 Planned analysis of the primary outcome

The number and proportion of participants who experience the outcome antidepressant cessation (yes/no) at 12-months. Between-group differences will be assessed using mixed-effects logistic regression, with effect estimates presented as odds ratio (95% CI); P value. Treatment group (RELEASE/Usual Care) will be included as the main fixed effect, time will be included as a fixed effect (6-/12-months) along with a group-by-time interaction term, practice size will be included as a fixed effect, and GP practice will be included as a random effect to account for probable non-independence of results from participants who attend the same GP practice.

For the main analysis: participants who actively drop out before assessment will be classified as ‘missing’; participants who are lost to follow-up will be classified as ‘missing’; participants who die during the study will be excluded from analysis.

A supplementary analysis will use a generalised linear model with binomial family and identify link to calculate the absolute risk difference in the primary outcome between the RELEASE and Usual Care groups. This will allow the calculation of the ‘number needed to treat to benefit’ and 95% CI.

2.4 Planned analysis of secondary outcomes

Secondary outcomes will also be analysed using mixed-effects models, using linear regression for interval outcomes, logistic regression for binary outcomes, and Poisson regression for count outcomes. When data is recorded at 6- and 12-months mixed-effects models will be used with time included as a main effect and a group-by-time interaction.

For secondary outcomes: participants who actively drop out before assessment will be classified as ‘missing’; participants who are lost to follow-up will be classified as ‘missing’; participants who die during the study will be excluded from analysis.

2.5 Sensitivity Analyses

A sensitivity analysis will be conducted to assess any effect of the first two practices randomised prior to the transition to the three-arm trial.

Sensitivity analyses will be conducted for all outcomes to assess the effect of loss to follow up and withdrawal. Multiple imputation by chained equations will be used for all primary and secondary outcomes. Imputation variables will be determined by considering the baseline demographic and clinical characteristics of participants who do, and do not, provide primary outcome data. Variables that are imbalanced will be included as imputation variables. Each outcome will be imputed 50-times.