This is a randomised, parallel, two-group trial. Participants were randomised to either the SWiM-C intervention or to a standard advice wait-list control group. Participants completed outcome assessments online at baseline, and at four months and twelve months post-baseline. A detailed description of trial methods is provided elsewhere [15]. Ethical approval was obtained from the Cambridge Psychology Research Ethics Committee (Application No: PRE.2020.049) on 24/04/2020. All participants gave written, informed consent. Clinical trial registration: ISRCTN 12107048.

Individuals were eligible to participate if they were adults with overweight or obesity (age ≥ 18 years; BMI ≥ 25 kg/m2), living in the UK, had a good understanding of written English, were willing to be randomised to either intervention or a standard advice group and to complete outcome assessments online, owned a set of bodyweight scales, and had not received bariatric surgery in the two years prior to the study.

Participants were recruited using mailing lists and social media advertisements from local weight management services and obesity organisations (e.g. Obesity UK), and through a volunteer database from a previous study [18].

A detailed description of the intervention is provided elsewhere [15, 19]. Briefly, the intervention is an ACT-based, web-based, guided self-help intervention. It includes access to an online web platform with 12 weekly modules (SWiM sessions) consisting of psychoeducational content, reflective exercises, and behavioural experiments. After participants had completed session 4, they received a telephone call from their SWiM coach (a 20-minute call focusing on reviewing exercises, troubleshooting, and ensuring participants understood the content). The coach also sent a tailored email at week 10.

The wait-list control group received standard advice in the form of a leaflet from the European Association for the Study of Obesity (EASO) on diet, physical activity, and mood for people living with obesity during the COVID-19 pandemic [15]. They received access to the SWiM-C web platform after completing the 4 month outcome assessment but did not have telephone or email support from a coach.

Participants were randomised to either the intervention (SWiM-C) or the standard advice group using a 1:1 allocation ratio with block randomisation (block size 6) stratified by BMI classification (25 to <30, 30 to <40, 40+ kg/m2) and sex (male, female). The randomisation sequence was computer-generated by the trial statistician and programmed into the database by the data manager. The sequence was unknown to research staff and participants.

Following informed consent via webform, participants completed baseline assessments using online questionnaires. Participants were then randomised to SWiM-C (participants received access to the SWiM-C intervention) or standard advice (participants were emailed the EASO leaflet). At four months and twelve months post-baseline, participants completed follow-up measures using online questionnaires. Participants were given honoraria for completing questionnaires (£10 for baseline and £20 respectively for the 4-month and 12-month follow-up). Honoraria for assessment completion were not dependent on intervention engagement.

To check the quality of data entry in the online surveys, we tabulated categorical variables and produced histograms of continuous variables to check whether distributions looked plausible. Additionally, webforms had built-in validation checks (e.g., for weight participants received a warning if the value was below 40 or above 200 but were able to continue if they confirmed). We assessed the data for biologically implausible weight change using the definition by Chen et al. [20].

The target sample for the trial was 360 participants to allow detection of a 1 kg difference in weight change between intervention and standard advice (assuming a standard deviation of 6 kg and a correlation between baseline and follow-up measures of 0.9) with 90% power and 95% confidence.

The primary outcome was change in self-reported weight from baseline to twelve-month follow-up (kg). Participants entered their weight into an online questionnaire. Participants were asked to weigh themselves on the day of questionnaire completion and were given detailed instructions on how to measure their weight at home (e.g. placing the scales on firm flooring, wearing light clothing).

All outcomes were assessed using validated self-report questionnaires. Eating behaviour was measured using the Three-Factor Eating Questionnaire (TFEQ-R21) [21], which provides one score of 0-100 per subscale (cognitive restraint of food intake, uncontrolled eating, and emotional eating). Higher scores indicate higher restraint and more uncontrolled and emotional eating, respectively. Experiential avoidance/psychological flexibility was assessed using the Acceptance and Action Questionnaire Weight Related-Revised (AAQW-R) [22]. Scores on the AAQW-R range from 10 to 70, with higher scores indicating higher experiential avoidance and lower psychological flexibility. Volume of total physical activity in MET-min per week was measured using the International Physical Activity Questionnaire (IPAQ) [23]. We assessed three domains of mental health: depressive symptom severity using the Patient Health Questionnaire (PHQ-8 [24]; scores range from 0–24), anxiety symptom severity using the Generalized Anxiety Disorder 7-item scale (GAD-7 [25]; scores: 0–21), perceived stress using the Perceived Stress Scale (PSS-4 [26, 27]; scores: 0–16). Higher scores indicate higher symptom severity for depression, anxiety, and stress, respectively. Wellbeing/capability was assessed using the ICEpop CAPability measure for Adults (ICECAP-A [28], scores: 0–1, higher scores indicate higher wellbeing). Cronbach’s alpha values for all questionnaires are shown in Table S1 (supplement). Secondary outcomes also included change in self-reported weight from four months to twelve months.

At baseline, we also measured age (years), sex, ethnicity, educational qualifications, marital status, and height.

Analyses were pre-specified in a statistical analysis plan (available at www.isrctn.com/ISRCTN12107048). To summarise baseline characteristics by randomised group, we present means and standard deviations (SDs) for continuous measures and the number and percentage of individuals within each category for categorical variables. Percentages were calculated using the number of non-missing values as the denominator.

The intervention effect on weight at twelve months (and 95% confidence interval [CI]) was estimated from a random intercepts linear regression model, using measures of change in weight from baseline to four months and twelve months as outcomes. The model included intervention group, assessment timepoint, the interaction term for group by assessment timepoint, the randomisation stratifiers (sex, BMI classification), and baseline value of weight as fixed effects, and random intercepts to allow for the repeated measures on each individual. Individuals were included in the analysis in the group to which they were randomised. Secondary outcomes were analysed using the same approach as described for the primary outcome.

Random intercept models use all available data and assume missing data are missing at random. To explore this assumption, we described baseline characteristics of participants with and without missing data on any outcome at twelve months. Because there were more missing values of weight in the intervention than in the standard advice group, we conducted a sensitivity analysis assuming data were missing not at random (MNAR) using pattern mixture models. Specifically, we imputed missing weight data using multiple imputation by chained equations (MICE). We then multiplied imputed values of weight by a varying factor (0% [MAR], or increasing or decreasing the values by 10%, 20%, 30%, [MNAR]) [29]. Subsequently we examined the estimates of the intervention effect under these 7 scenarios. If the influence under these scenarios is small, the analysis is considered to be robust against departures from the MAR assumption.

Participants in the wait-list standard advice group were given access to the SWiM-C web platform after the four-month outcome assessment. We therefore conducted per-protocol analyses excluding standard advice participants who engaged with the intervention. We compared intervention participants who engaged with the SWiM-C intervention with standard advice participants who did not engage with the SWiM-C intervention to assess whether the findings were influenced by the degree of engagement with the intervention. We defined non-engagement with the intervention as accessing less than the first four sessions (participants were only able to proceed through sessions sequentially), since ACT-based content was introduced from session 4. We performed two per-protocol analyses of the primary outcome with two different levels of engagement: 1) comparing intervention participants who completed at least 4 SWiM-C sessions with standard advice participants who accessed less than 4 SWiM-C sessions, and 2) comparing intervention participants who completed at least 8 SWiM-C sessions with standard advice participants who accessed less than 4 SWiM-C sessions.

To estimate baseline-adjusted differences between the study groups in change in weight from four months to twelve months, we used a linear regression model with change in weight from four months to twelve months as the outcome, and including baseline weight, four-month weight, the randomisation stratifiers (sex, BMI classification) and intervention group as covariates.

Data were analysed using R version 4.0.0 and R Studio version 1.0.153.