Adapted Motivational Interviewing and Cognitive Behavioural Therapy for Food Addiction

Background: Food addiction is the concept that individuals can be "addicted" to foods, particularly highly processed foods. This concept has attracted growing research interest given rising obesity rates and engineering of food products in industrialized countries. Food addiction is assessed using the validated Yale Food Addiction Scale (YFAS), which applies substance use disorder criteria from the most recent Diagnostic and Statistical Manual of Mental Disorders to the consumption of these types of foods. Prevalence estimates of food addiction range from 8-15% in two nationally representative samples in the U.S. and Germany. Although food addiction is not a currently recognized mental disorder, individuals do identify as being addicted to foods and self-help organizations such as Overeaters Anonymous have existed since 1960 to purportedly treat it (i.e., through abstinence). However, little research has been conducted on how abstinence approaches work and such methods may even be harmful for individuals with eating concerns, given the risk of disordered eating. Currently, there are no empirically supported treatments for food addiction. However, evidence-based treatments do exist for addictions and eating disorders, such as Adapted Motivational Interviewing (AMI) and Cognitive Behavioural Therapy (CBT), which may prove beneficial for food addiction, given neural similarities between substance addiction and binge eating, and the potential for high ambivalence. AMI is designed to allow clients to voice their own motivations for change and the use of AMI skills by therapists has been shown in meta-analyses to predict this type of change talk, which then predicts positive behavioural outcomes, Given that food addiction is also associated with internalized weight bias and lower eating self-efficacy, AMI techniques in fostering acceptance, highlighting client strengths, and providing psychoeducation may help to lower self-blame and bolster confidence to change one's eating habits.

Method: The current study proposes a randomized controlled trial using a four-session AMI and CBT intervention for food addiction. Due to COVID-19 limitations, all intervention sessions will be conducted virtually over secure videoconferencing technology or by telephone, expanding access to all adult community members across the province of Ontario in Canada. The intervention combines the personalized assessment feedback and person-centred counselling of AMI with CBT skills for eating disorders, such as self-monitoring of food intake and stimulus control. The aim is to motivate participants to enact behavioural change, such as moderate consumption of processed foods in a harm reduction approach. Twenty randomly selected session tapes will be reviewed by two trained coders to assess for MI adherence using the most commonly used MI fidelity measure. The intervention condition will be compared to a wait-list control (WLC) group. Both groups will be assessed at pre- and postintervention periods, as well as over a 3-month follow-up period to assess maintenance effects.

Hypotheses

Primary Hypotheses - Food Addiction and Binge Eating Frequency (H1-H3)

• H1: Compared to WLC, AMI will lead to a significantly greater reduction in food addiction symptoms (using the YFAS 2.0) at postintervention and up to 3 months postintervention.
• H2: Compared to WLC, AMI will lead to a significantly greater reduction in self-reported consumption of highly processed foods specified in the YFAS 2.0 (using the Canadian Diet History Questionnaire II) at postintervention and up to 3 months postintervention.
• H3: Compared to WLC, AMI will lead a to significantly greater reduction in number of binge eating episodes (using select Eating Disorder Examination Questionnaire questions) at postintervention and up to 3 months postintervention.

Secondary Hypotheses - Readiness for Change, Eating Self-Efficacy, and Weight Bias Internalization (H4-H6)

• H4: Compared to WLC, AMI will lead to a greater increase in motivation for changing one's food addiction symptoms (e.g., reducing consumption of highly processed foods; using Motivational Rulers) at postintervention.
• H5: Compared to WLC, AMI will lead to a significantly greater increase in eating self-efficacy (using the Weight Efficacy Lifestyle Questionnaire) at postintervention.
• H6: Compared to WLC, AMI will lead to a greater reduction in weight bias internalization (using the Modified Weight Bias Internalization Scale) at postintervention.

Secondary Hypotheses - Other Eating-Related Constructs (H7-H14)

It is hypothesized that AMI will lead to significantly greater reductions in other eating-related constructs compared to WLC at postintervention and up to 3 months postintervention, in terms of:

• H7: self-identified food addiction,
• H8: addiction-like eating behaviour (AEBS),
• H9: binge eating symptoms (Binge Eating Scale),
• H10: loss-of-control eating (Loss of Control over Eating Scale),
• H11: emotional eating (Emotional Eating Scale),
• H12: general appetite for palatable foods or hedonic hunger (Power of Food Scale),
• H13: cravings for specific highly processed foods (Food Craving Inventory),
• H14: Body Mass Index (BMI)

Tertiary Hypothesis - Working Alliance (H15-17)

Given that a collaborative partnership is key component of MI and that there is a robust positive association between working alliance and treatment outcomes, it is hypothesized that there will be positive associations between postintervention working alliance (using the Working Alliance Scale Short Form Revised) and postintervention motivation for change (H15), eating self-efficacy (H16), and weight bias internalization (H17).

Sample Size: Based on a power analysis and previous effect sizes following AMI interventions for binge eating (Cohen's d = 0.76), a total sample size of n = 58 is needed. Accounting for previous withdrawal rates and an inclusion rate of 44.6% in a similar study, a total of 131 individuals should be recruited. Recruitment is estimated to take place over 5 months beginning in March 2021. Given the 3-month follow-up, participation is estimated to end in March 2022.

Analyses: To determine whether both AMI and WLC groups are equivalent in terms of sample characteristics as a result of randomization, independent samples t tests will be conducted on baseline variables such as age, BMI, YFAS severity, and binge eating frequency. To determine whether sample characteristics differ between treatment completers and dropouts, independent samples t tests will be conducted on the same baseline variables and working alliance. Lastly, to determine if equal proportions dropped out of the AMI and WLC groups, a chi square test will be conducted.

Primary, Secondary, and Tertiary Outcomes: For the primary outcomes (i.e., YFAS symptoms, binge eating frequency, and consumption of highly processed foods), given the between-groups and repeated-measures mixed design, a 2 (group: WLC vs. AMI) x 4 (time: baseline, postintervention, and 1- and-, 3-month follow-up) mixed analysis of variance (ANOVA) will be conducted on SPSS statistical software. For the secondary outcomes (i.e., readiness for change, eating self-efficacy, and weight bias internalization), a 2 (group: WLC vs. AMI) x 2 (time: baseline, postintervention) mixed ANOVA will be used to compare the WLC and AMI groups from pre- to postintervention. For the other secondary eating-related outcomes, the same 2 (group) x 4 (time) mixed ANOVAs described above will be used to compare WLC and AMI groups across time. For the tertiary outcomes, to determine if there is a positive association between working alliance and readiness for change, eating self-efficacy, and weight bias internalization, two-tailed, bivariate, Pearson's correlation analyses will be performed. To explore the changes in readiness for change, eating self-efficacy, weight bias internalization, and working alliance from pre- to postintervention, paired samples t tests will be used for the AMI group for these four constructs. Prior to data analyses, data will be checked for bias and corrected as necessary. Any interactions from the ANOVAs will be followed up with planned contrasts, with the control group and baseline as the base categories for the between-groups and repeated-measures variables, respectively.

Assuming that there are no valid reasons to ignore missing data and to conduct complete case analysis (e.g., if less than 5% of data are missing), and assuming that data are missing at random, multiple imputation will be conducted on SPSS for the missing values, with at least 50 imputed datasets in order to reduce sampling variability in the imputation process. Results from complete case analyses and multiple imputation analyses will be compared for differences. To reduce bias of the imputation model, the model will include any variables that predict missing data. SPSS will automatically scan the data for a monotone pattern of missing values, and if such a pattern is present, a monotonic multiple imputation will be conducted. The default number of iterations per missing variable used will be 10 but at least 50 imputed datasets will be computed. If data are not assumed to be missing at random, sensitivity analyses will be performed for missing binary data.

Satisfaction Evaluation: Descriptive statistics will be obtained for the three quantitative satisfaction questions (e.g., an average score for how satisfied participants were with the research study). Qualitative responses from the open-ended questions will be analysed as per thematic analysis methodology.

Treatment Adherence: The minimum threshold of MI adherence will be based on the Motivational Interviewing Treatment Integrity Code (MITI) basic competence and proficiency thresholds for clinicians. Specifically, summary scores must fall in at least the "fair" scores (i.e., the Relational score must be 3.5/5, the Technical score must be 3/5, 40% of the reflections must be complex reflections, and the reflection-to-question ratio must be 1:1). If all four domains meet these thresholds, then the session will be rated as 100% fairly adherent, which will be the minimum goal. An average percentage of "fair" adherence across raters and tapes will be calculated to determine if sessions met this threshold. To determine interrater reliability for each summary score, a two-way mixed-effects model, using the mean of two raters (k = 2) and absolute agreement will be used. Intraclass correlation coefficients and their 95% confidence intervals will be reported from the SPSS output.