In September 2012, the decision was taken to close the trial to recruitment owing to the difficulties described in detail inChapter 4. At this point we were 15 months into the trial and we had recruited 170 individuals. A closedown plan was devised to consider ethical issues and to maximise the scientific value of the trial. It was agreed with the funder that recruitment would cease immediately, but that included participants would still receive the intervention that they were randomised to and be followed up to 1 year. This meant that the whole study would finish in January 2014 and would now be considered a feasibility study. The methods for the feasibility study are similar to those described for the original trial with some key alterations. These are summarised inTable 3followed by more detail on key components.
TABLE 3 Summary of design changes
Study component Changes to design Sample size, recruitment
and retention
Sample size reduced from 950 (90% power at 5% significance level: mean BMI difference of 1.7 kg/m2) to 166a(80% power at 5% significance level: mean BMI difference of 3.5 kg/m2) Incentives offered to participants for completing assessments increased from £10 to £20. Incentives (£10 vouchers) also offered to interview participants
Length of follow-up Reduced from 3 years to 1 year post randomisation Intervention Group sessions discontinued
Analysis of primary and secondary outcomes
Planned subgroup analyses (age, amount of weight lost, method of weight loss, satisfaction with weight loss) not undertaken; some exploratory analyses carried out (gender, binge eating, source of recruitment), but others (smoking, weight-affecting medication) dropped. Implementation intentions not included as a mediator. CACE estimated using two-stage least squares instrumental variable regression and not multilevel mixture analysis
Process evaluation Group component of intervention dropped therefore GF focus group not run. Participants who withdrew from the intervention not willing to be interviewed
Economic evaluation Estimation of long-term cost–utility (economic Markov modelling) not undertaken a Actual number recruited minus exclusions.
Objectives
The main objectives were to assess the feasibility, acceptability, compliance and delivery of a 12-month multicomponent intervention, as well as recruitment and retention. To give an indication of effect sizes for a larger trial, we evaluated the impact of the intensive or less intensive intervention on participants’BMI (primary effectiveness outcome) at 12 months from randomisation.
Design
This was a feasibility study of a three-arm individually randomised controlled trial. The population recruited was as described inParticipants. As described in theScientific summary, the intervention changed as we stopped delivering the group element. Outcome measures used were as described inTable 2with the exceptions noted inTable 3. After completing baseline assessments, participants were followed up at 6 months, during the intervention, by post and at the end of the intervention (approximately 12 months post randomisation, with a 3-month window to allow for any delay in receiving the intervention). A 1-year follow-up was conducted face to face in most cases, although a few were conducted by telephone with self-reported weight.
Sample size
The original sample size of 950 was based on detecting an effect size of 0.309 with 90% power at a significance level of 5%. Given that the study was now a feasibility study, the sample size was not derived statistically but based on actual recruitment at the time of closure (n=166). This allows us to estimate a percentage of 50 (the percentage with the greatest associated variability) to within 7.6 percentage points either side for the whole sample, or to within 15.9 percentage points for percentages within each study arm (n=38). These are the achievable precisions for estimation of the retention rate (or any other proportion). Although not primarily concerned with effectiveness, we still feel that it is informative to present power to detect differences. With our current sample size of 166 (four participants from the original 170 recruited were ineligible and randomised in error and subsequently withdrawn) and assuming 30% attrition, we have 39 in each of the three arms. We are therefore able to detect a difference in BMI of 3.537 kg/m2at the 5%
significance level with 80% power between any two arms (primary comparison between the control and intensive arms). This is a difference of about 9.75 kg between groups based on average height.
Analysis of primary and secondary outcomes
The main feasibility outcome was the proportion of participants remaining in the study for a year. A secondary feasibility outcome was the proportion of participants who adhered to the intervention in each treatment arm. Comparisons were made between those who completed the study and those who dropped out, to highlight any biases. The main effectiveness analysis was intention to treat and complete case comparing the intensive intervention arm with the control on average BMI using a three-level linear regression model to account for clustering within therapists and groups (reduced to a one-level model where there is no evidence of clustering). The main analysis examined the latest end point (12 months), controlling for individual patient characteristics (baseline BMI, age, gender, ethnicity, source of recruitment, current BMI and percentage weight loss). Both intervention groups were compared with the control. We completed augmented analyses for BMI and weight, for which self-reported follow-up weights were included with verified follow-up weights. The statistical analysis plan specified that if the primary outcome was missing for more than 10% of cases then multiple imputation would be used.
Secondary outcomes included waist circumference, waist-to-hip ratio, self-reported physical activity, proportion maintaining weight loss (defined as successful when the participant’s weight at the end of the trial is less than or equal to his or her weight at baseline), self-reported dietary intake, health-related quality of life, health service and weight control resource use, binge eating, alcohol consumption, smoking status, psychological well-being, duration of participation and dropout from the intervention. All analyses controlled for individual patient characteristics. Binary outcomes were analysed using hierarchical logistic
regression and continuous outcomes (appropriately transformed if necessary). Mediation analyses were conducted on self-efficacy, social support, intrinsic motivation, habits and self-monitoring as measured at 6 months using a hierarchical model and controlling for baseline randomisation variables in accordance with Baron and Kenny guidelines.118Consolidated Standards of Reporting Trials–Patient Reported
Outcomes guidelines are referenced for reporting patient-reported outcomes.131Exploratory subgroup
analyses investigated WLM in binge eaters, gender and source of recruitment.
Sensitivity analyses
A sensitivity analysis was conducted assuming that non-responders returned to weight levels prior to weight loss (i.e. not baseline but previous BMI). A CACE was estimated using two-stage least squares intrumental variables regression.119This modelling focuses on estimating the intervention effects in the presence of
non-compliance, but also incorporates adjustments for loss to follow-up. Compliance was defined as follows: for the intensive arm, attending five of the six face-to-face MI sessions; for the less intensive arm, attending both face-to-face MI sessions. Potential differences on basic demographics and the primary outcome between full responders and those providing a minimum data set were described and all models were tested for adequate fit using appropriate diagnostics.
Process evaluation methods
Minor changes to the process evaluation are listed inTable 3.
Economic evaluation
There are limitations to undertaking economic evaluations alongside a feasibility study. In particular,
the change in design meant (1) the within-trial analyses would only reflect short-term (12-month) differences in costs and effects, (2) the study was less likely to have sufficient power to detect statistically significant differences in costs or effects and (3) the absence of data on longer-term WLM or costs made estimation of lifetime cost–utility through economic modelling unfeasible. Although it would not be possible to establish definitive cost–utility/effectiveness results, we nevertheless applied as far as possible the methods described in
Cost-effectiveness analysisto estimate parameters and unknowns to inform future research. Specifically, (1) methods to determine costs of training and intervention delivery were unchanged; (2) methods for collecting resource use and EQ-5D data were unchanged apart from the final data collection being at 12 not 36 months; and (3) the planned estimation of long-term cost–utility was dropped.