Use of motivational interviewing in behavioural interventions among adults with obesity: A systematic review and meta-analysis

Summary This review aimed to identify whether motivational interviewing (MI) (a counselling approach for supporting behaviour change [BC]) helps to reduce bodyweight and BMI in an adult obesity context. This included evaluating effectiveness of MI interventions within this population and reporting the methodology used, including theoretical underpinnings and identification of BC and MI techniques. Eight databases were searched using controlled vocabulary. Eligible studies included adults with obesity (BMI ≥ 30 kg/m 2 ), author-reported interventions using MI aiming to reduce body weight or BMI, and comparator groups not receiving an MI intervention. Data extraction and quality appraisal tools were used to identify study characteristics, intervention content was coded for techniques, and random-effects meta-analysis were conducted to investigate effects. Meta-analysis of 12 studies indicated no overall pooled effect on bodyweight and BMI outcomes between intervention and control groups (SMD = (cid:1) 0.01 [95%CI (cid:1) 0.13 to 0.12, P = .93]). Findings were limited by multiple sources accounting for risk of bias, and poor reporting of intervention fidelity and content. Intervention and control content descriptions indicated similar techniques, with social support, goal setting (behaviour) and self-monitoring of behaviour occurring most frequently across both. Findings do not contribute additional evidence for MI use in this context, however methodological limitations were identified which must be resolved to better identify the intervention effects on obesity-related outcomes.

and United States as a Body Mass Index (BMI) of 30 and higher. 3 It holds serious implications for health including an increased risk of illnesses such as Type 2 diabetes, hypertension and cardiovascular disease when compared to individuals without obesity. 4 Research has identified a multifactorial basis for the condition, stemming from genetic, behavioural, environmental and social aspects 3 which may be impacted by the obesogenic environment. 5 These prevalence rates indicate need for development of evidence-based, effective interventions that reduce bodyweight and the risk of adverse health outcomes.
Particularly important is identification of interventions that can be implemented in a range of settings to boost accessibility of effective treatment through methods such as telehealth. 6,7 Motivational interviewing (MI) is a communication approach designed to assist an individual in reducing ambivalence about behaviour change, via four core processes: engaging with an individual, focusing on specific behaviours to change, evoking change talk, and planning to enact change. 8 It is a patient-centred, non-judgemental, directive set of skills utilized by practitioners to discuss changing patient behaviours for improved health outcomes. Skills that align with the core spirit of MI (partnership, compassion, evocation, and acceptance) and aim to respect the individual's autonomy and build upon motivation for change are used. These include open-ended questions, affirmations, reflective statements and expressions of empathy among other recently defined specific skills. 9 MI is one approach used to support preparation for behaviour change and maintenance of progress within weight-loss settings through raising motivation, self-efficacy and improving adherence to other weight-related interventions. 10 Whilst there is evidence across health behaviour settings such as alcohol and substance use, 11,12 there are less conclusive findings as to the appropriate quantity and delivery of MI within obesity care. In 2019, Patel and colleagues 13 identified 15 trials utilizing MI within telehealth settings for weight loss in adults living with overweight and obesity and found it performed better than no treatment in around 54% of 11 occasions, but in the majority of cases using an active comparator, MI did not perform better. Armstrong and colleagues 14 conducted in 2011 a meta-analysis of randomized controlled trials recruiting adults with overweight and obesity, and identified an overall significant, moderate (SMD = 0.51) effect of MI to improve weight loss over comparator interventions such as treatment as usual and advice from non-MI trained practitioners. Barnes and colleagues' recent review of papers utilizing MI with adults with overweight and obesity (2015) 6 drew similar narrative conclusions to the earlier findings with 54.2% of included studies reporting clinically significant weight loss of at least 5% baseline body weight, although this did not quantitatively synthesize findings using meta-analytic techniques.
Providing greater detail about specific intervention components, such as which skills are utilized, would assist with ensuring methodological replication. This would also allow identification of intervention components necessary for effective weight loss and maintenance.
One way to provide greater specificity within intervention reporting is to code descriptions for behaviour change techniques (BCTs). BCTs are formal descriptors for the active components of behaviour change interventions that are "observable, replicable, and irreducible … designed to alter or redirect causal processes that regulate behaviour." 15 and can be utilized to provide more specific detail about what was delivered within interventions. Understanding of relevant BCTs may assist in replication and therefore, development of effective, targeted interventions. A taxonomy of 93 techniques clustered into 16 groups has been created by Michie and colleagues 15 for use within intervention design and reporting. Within the obesity management context, health authorities have recommended use of techniques such as goal setting (1.1), selfmonitoring of behaviour (2.3), review of behavioural goals (1.5), feedback on performance (2.2) and action planning (1.4) as effective for weight loss outcomes. 16 Accounting for the fact that MI includes techniques additional to those defined within existing taxonomies, researchers have recently developed a taxonomy of techniques specific to MI 9 which can be used to clarify what is occurring within MI implementation. There is currently no clear understanding of the influence of techniques within MI interventions for obesity-related outcomes. Previous reviews have not explicitly identified the BCTs reported in published trials of MI for weight outcomes through coding the intervention design specifically within MI and weight loss contexts, but recent research has begun to identify techniques present within in-person physical activity counselling sessions. 17 MI is a popular approach and further investigation to clarify the efficacy of the intervention and its individual components is required. The use of Hardcastle and colleague's technique framework will support identification of present components in MI interventions.
The primary objective of this review was to investigate the effectiveness of MI for adiposity outcomes in populations with obesity aiming to lose bodyweight, and to evaluate its effectiveness utilizing metaanalytic methods. Additionally, the review aimed to report the presence of BCT and MI specific techniques, and theoretical underpinnings in MI research to inform understanding of how MI is utilized in research. Specifically, this review aimed to synthesize studies of participants with obesity that included at least one group receiving an intervention of MI and a non-MI comparator group, and reported outcomes of bodyweight including BMI and kilograms as change scores or final measurements.

Methods are reported in accordance with the Preferred Reporting
Items for a Systematic Review and Meta-Analysis guidelines (PRISMA 18 ; see supplementary files [Appendix S1] for checklist). The protocol can be accessed from PROSPERO (https://www.crd.york.ac. uk/prospero/display_record.php?ID=CRD42018114697).

| Study eligibility
For inclusion, trials were required to recruit adults (18 years and above) with a BMI of 30 kg/m 2 and higher examining the use of MI for the reduction of bodyweight. Trials needed include at least one arm providing MI and one arm offering a comparator without MI. Studies required at least one follow-up point and the full-text needed to be available in English. Both published and unpublished work was eligible for inclusion. Cross-sectional or single-group studies were not eligible, nor were studies recruiting participants with obesity as a result of a pre-existing condition or as a secondary effect of medication.

| Search strategy
Original searches took place 13 After removal of duplicates, researchers screened the remaining articles at title, abstract and full-text level. Papers were assessed with responses of "yes," "no," or "unclear" for eligibility characteristics.
Articles assessed as 'yes' or 'unclear' were included for the next stage.
Screening was carried out by one researcher with 25% secondscreened by an external researcher.

| Risk of bias
Papers were assessed for risk of bias following data extraction. Randomized controlled trials were assessed using the Cochrane Collaboration risk of bias tool. This determines a separate value for each type of bias, and determining an overall risk level for each paper is not recommended ( 21 ) but permits comparison. This assessed each paper for bias related to; random sequence generation and allocation concealment, blinding, blinding of outcome assessment, attrition and selective reporting as high, unclear, or low risk. 50% of studies were assessed for by a second researcher. Disagreements were resolved through discussions. Nonrandomized articles were assessed for risk of bias using the Newcastle-Ottawa quality assessment scale 22 which utilizes a starring system to report on selection, comparability, and outcome.

| Extraction of data
Where multiple intervention arms were in place, the most passive intervention with data reported was selected as the comparator. This was defined as the arm closest to providing treatment as usual/no treatment. Outcome measurements from baseline and the latest follow-up point were extracted as reported. Additionally, behavioural measures such as outcomes of physical activity or dietary intake were extracted. Where multiple papers were published on the same dataset, the latest possible follow-up point was used for extraction and earlier papers reviewed for methodological information.
Data extraction forms were designed using the template for intervention description and replication checklist (TIDieR 23 ). This included a description of the intervention delivery methods and content, interventionist backgrounds, setting, and frequency. Information regarding the use of fidelity tools was extracted if reported. The primary outcome was measures of adiposity, such as BMI and weight in kilograms.
Secondary outcomes of interest were measures of motivation or adherence to other treatments, such as behavioural weight loss programmes (BWLP). Where reported, both BMI and bodyweight information was extracted. Also extracted was information about attrition, reasons for drop-out and how analyses handled this, such as use of intent-to-treat analysis or data transformation. 25% of papers were extracted a second time by a researcher external to the research team to confirm accuracy of extraction.

| Coding of BC techniques and MI techniques
Intervention content was coded for techniques in line with the Behaviour Change Technique Taxonomy v1 (BCTTv1 15 ; and MI-specific techniques, including content-based (eg, agenda mapping) and relational techniques (eg, offer emotional support) as defined by Hardcastle and colleagues. 9 Intervention techniques were coded only when clearly present and applied to a target behaviour related to the outcome behaviour (eg, weight loss through dietary or nutritional changes, activity level changes, adherence to other intervention designed for weight loss outcomes such as BWLP). Intervention content was examined through the published article, any supplementary materials and protocols. Intervention technique coding was conducted by one researcher (HM) and 50% were second coded (AC). Intervention and comparator descriptions were reviewed and each technique noted as present or absent. Both researchers completed the BCTTv1 online training module and coded using agreed definitions. Any discrepancies were resolved through discussion. To quantify agreement between coders, prevalence-adjusted bias-adjusted kappa statistics 24 were calculated for each set of coding. Use of prevalence and bias adjusted kappa allows for a high prevalence of absent or present responses from the coders and is reported in addition to Cohen's kappa. 25

| Data analysis plan
Quantitative data was synthesized using Review Manager 5.3. 20 Reported outcome measurements differed between studies. Both BMI and bodyweight in kilograms were reported as change scores or final measurement values; to examine the pooled effect, only final measurement values for these were inputted as means and SD to Review Manager as more papers reported this outcome rather than change scores. As BMI and bodyweight in kilograms represent different scales it was not possible to combine both change and final value measurements from each scale within a SMD meta-analysis. The final measurement of both outcomes were pooled within an SMD metaanalysis. If SDs were not reported, they were calculated from 95% confidence intervals or SE using procedures from the Cochrane Handbook . 26 Where papers reported both BMI and bodyweight, BMI was selected for inclusion within pooled effects analysis. Additional analyses using mean difference were not pre-specified within the review protocol. Papers were grouped by the outcome measure of weight in kilograms and BMI and each outcome type examined in a separate analysis. This allowed for the examination of pooled change from baseline and final measurement scores within each scale individually, as it is assumed that the SD represents the same thing, 26 increasing the number of studies that could be included in analyses.
Review Manager weighted the effects according to sample size and provides a mean difference (MD) or standardized mean difference (SMD) score along with a significance value and 95% confidence intervals for continuous data. Heterogeneity was assessed by checking I 2 values. This was examined using a significance cut-off of P < .10 with percentages of 0 to 40% for potentially not important, 30 to 60% as moderate, 50 to 90% as substantial and 75% + for significant heterogeneity. 23 It was expected due to differences in MI delivery and study populations that significant heterogeneity would be present and therefore random effects meta-analyses were conducted.

| RESULTS
The number of papers assessed for inclusion is summarized in Figure 1. Initial database searches yielded 1588 records. Following title and abstract screening, 157 full texts were screened for inclusion.
The most frequent reasons for exclusion were a non-Motivational Interviewing intervention or single group design. Thirty-one papers were deemed appropriate for inclusion in the review with high reliability between researchers for the screening process (k = 0.75). Figure 1 reports the identification of studies.

| Study characteristics
Publication dates ranged from 2008 to 2020. Sample sizes ranged from 19 to 864 with 6249 participants recruited across all studies.
MI participants experienced significant improvements in HEI (P = .02), and a nonsignificant trend of improvements was seen for non-MI (P = .15). Increases in general self-efficacy for the intervention group (P = .08), not for the comparator group (P = .84). No significant difference between groups (P = . 16).No significant changes for either MI (P = .40) nor non-MI group (P = .32) and no significant differences between groups (P = .63).
Significantly higher fruit and vegetables intake in the MI group at follow-up (P = .005).
Condition did not significantly predict moderate exercise (P = .56), vigorous exercise (P = .80), fast food consumption (P = .07), sweetened drink intake (P = .12), fruit intake (P = .40) or vegetable intake (P = .20 MI group energy intake significantly reduced overall (P < .001) and activity duration significantly increased (P < .001). Intervention group scored significantly lower on the disinhibition scale of EDE-Q (P = .02), but no significant effect of time. No significant effects of treatment group on any other variables. Significant decreases in EDE-Q restraint over time (P = .01), shape concern (P = .01), increases in flexible control (P = .01), and rigid control (P < .001).
Greaves ( Intervention group significantly increased physical activity (walking) between baseline and 6 months (P = .006, d = .24) and baseline and 18 months (P = .032, d = .20) but no significant differences between groups over time. Stages of change showed significant increases between baseline and 6 months (P < .001, d = .29) and significant decreases from 6 to 18 months (P < .001, d = .29) for MI group. Decrease in fat intake for MI group (P < .001, d = .43) between baseline and 6 months, which was maintained to 18 months (P < .001, d = .38).
No significant differences in physical activity, dietary intake, or theory-based measures between groups. Significantly greater change from baseline in the intervention group for self-efficacy (P = .07), and re-structuring plans (P = .006) Karlsen (2013 Modest increases in autonomous self-regulation observed in both groups over time, but no significant differences by condition (P = .83). Reductions in controlled motivation observed, no significant differences between groups (P = .56). Perception of autonomy support at post-treatment higher in MI group (P = .08, d = .77).
A significant interaction between time and condition (P = .002) for dietary intake with a mean change of 0.06 for intervention. Significant interaction for time and condition (P = .02) for physical activity step counts, indicating significant increases in steps for the intervention group at 4 months which was not maintained to 8 months. Control-MI group did not significantly change step counts in the first 4 months but significantly increased between 4 and 8 months. Significant interaction effects for weight and exercise self-efficacy (P < .001, P < .001 Changes over time between groups were not significantly different for "get up and go" tasks (P = .23). No significant difference in step counts over time between groups (P = .39). No significant difference in changes over time for sedentary behaviour (P = .63), nor decisional balance of pros (pP = .39) and cons (p = .39), nor of diet quality changes over time (P = .45).
Depression scores and quality of life NR. Participants in the intervention group significantly improved their self-efficacy from pre to post intervention. Intervention group had significantly greater scores for negative emotions, social pressures, physical discomfort and positive activities than the comparator. Significant effect of condition (P = .001) overall and on negative emotion (P = .024), social pressure (P = .040) physical discomfort (P = .006) and positive activities (P = .017), indicating higher scores in the intervention group.
MI group scored significantly higher on selfefficacy, social pressures, food availability, physical discomfort, negative emotions, and positive activities than the comparator.

| Quality assessment
Quality assessment is reported in Supplementary materials A (Appendix S1). Risk of bias varied, but for the domains of random sequence generation (58.6%) and attrition (37.9%) a majority of studies were graded as low risk of bias. This was due to clearly described use of randomization techniques or software (eg, 34 ) and participant drop-out was explained (eg, 39   The primary reason for exclusion of randomized trials from the SMD meta-analysis was that only change scores were reported. To check publication bias, the funnel plot (see Figure 3)  Meta-regression of techniques may provide further contextualized information regarding predictive elements of effective behaviour change interventions but was outside of the current scope. 62 In comparison to earlier work, meta-analysis did not find evidence of significant difference between final physiological measurements of MI and non-MI group participants. Armstrong and colleagues 14 identified studies with a BMI ≥25 kg/m 2 and without additional intervention components, and completed a meta-analysis identifying significant effects. This was in terms of an increased bodyweight reduction compared to a control group of À1.47KG. We identified 6 papers which overlapped with Armstrong's review 28,32,50,52,53,63 and 21 papers published since 2011. Similarly to current findings, no statistically significant effect of MI on bodyweight outcomes within a pooled analysis of BMI specific outcomes was found. One consideration when conducting obesity-specific research is that BMI fails to determine adiposity from muscle and bone. 64 However, BMI is generally considered an adequate indicator and it is frequently used to determine obesity cut-offs. 65 In the current review BMI was an appropriate outcome measure due to the frequency of use which permitted data synthesis in the meta-analysis, however future research should consider reporting other measurements such as waist circumference. 4 MI centres upon resolving ambivalence for behaviour change, while promoting autonomy and building motivation. 8 As these targets map more closely to behavioural than physiological outcomes, it is plausible that whilst effects on bodyweight outcomes were inconclusive, interventions may have influenced behavioural factors such as physical activity (PA) and diet. Half of the trials measuring diet-related outcomes reported beneficial effects of MI, and around a quarter reported beneficial effects for physical activity related outcomes in comparison to the non-MI group. There may be beneficial effects of MI interventions for behavioural outcomes such as eating behaviours 31,35,[37][38][39][40]43,49 and physical activity, 31,37,39,49,52 and future investigations should assess the impact of MI for such outcomes. In a prior systematic review 6 there was found to be no significant change in behavioural outcomes for the intervention group in comparison to the control group. Conversely, Knight and colleagues 66 found that MI appeared to have beneficial effects for a range of health outcomes including psychological (eg, readiness to change), physical (eg, bodyweight and metabolic control) and lifestyle changes (eg, exercise and alcohol intake), but issues in study quality prevented meta-analysis. Mixed findings point towards a need for further study into the effects of MI on behavioural outcomes and motivation measurements.
Whilst use of methods such as online and telephone contact can improve accessibility, heterogeneity in methods can limit comparability. Miller & Rollnick 8 have highlighted the importance of the spirit of MI, which is its client-centred focus, which may not be reported within intervention descriptions or technique taxonomies. 9 Use of fidelity measures can assist in confirming the practitioner is adhering to the spirit. 6 23 Similarly, this may support avoidance of content overlap between the control and intervention. However, within the current review, there was no stated use of these of reporting templates. Multiple reviews over the past decade have highlighted the lack of reporting of fidelity and training as an issue within the field. 6,13,14 There are important limitations to consider. Searches were limited to material accessible in English language; this may exclude relevant papers and lead to location bias. However, 12 studies 10,30,31,33,35,38,41,43,[45][46][47]54 were identified that were based outside of the US and UK. Use of final measurement score over time is less ideal for inclusion within meta-analysis than use of change scores, but this was dependent on outcomes published in the articles. Inaccessibility To conclude, this review identified that a range of methodologies of MI have been researched for effectiveness of weight loss outcomes, and a meta-analysis found no significant overall effect. However, no significant increases in weight comparable to the comparator groups were observed. Frequently reported BCTs included social support, self-monitoring of behaviour, problem solving, goal setting and information about health consequences, which is consistent with current recommendations from government bodies such as Public Health England, 16 and demonstrates integration of MI with these techniques.
The novel use of recently itemized MI techniques 9 was a feasible method for coding descriptions and identified overlap with an existing BCT taxonomy. It is possible, but unsupported by the current evidence-base, that MI is a beneficial approach within overweight and obesity settings, and thus further research should address the methodological problems identified by this review and seek to confirm the most likely mechanisms of action related to this complex intervention.