Motivational Interviewing to Treat Adolescents With Obesity: A Meta-analysis

We used the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement and checklist to guide the conduct and reporting of this review. Before data extraction was complete, we developed and registered a protocol on PROSPERO (#CRD42017072342), available in full on the program Web site (http://www.crd.york.ac.uk/PROSPERO/display_record.php?ID=CRD42017072342).

The protocol predefined the objectives, methods, principal focus (concept) and context, research question, and inclusion and exclusion criteria for this SR and meta-analysis, and described the search, data extraction, and data synthesis strategies. We conducted searches on September 26, 2016 and April 16, 2018 and identified studies published from 1997 to 2017 in the following 7 databases: Medline, Embase, Cumulative Index to Nursing and Allied Health Literature, PsychINFO, Web of Science, Cochrane Library, and Google Scholar. Search terms were “adolescent obesity” and “motivational interviewing” (see Supplemental Information for terms used and PubMed search).

For inclusion, studies were required to be published in the English language; to include adolescent participants (ages 12–19 at study enrollment) with overweight or obesity (BMI percentile ≥85%); to focus on an MI intervention targeting weight management; and to report at least 1 predetermined primary outcome (change in pounds, kilograms, BMI, BMI z score or percentile from baseline to last available follow-up) or secondary outcome (change in nutrition, physical activity, sleep behaviors, cardiometabolic outcomes, or quality of life). We excluded case studies, qualitative studies, editorials, and MI-based interventions focused on behavior change not directly related to weight management (eg, alcohol, substance, and condom use).

We used an SR citation–screening Web application, abstrackr,²⁴ to manage the abstract screening process. Four authors independently participated in screening of titles, abstracts, and full-text articles identified through the searches against the protocol. The first author resolved conflicts between the screeners. Five of the articles included in the Cushing et al¹⁷ meta-analysis in which MI for adolescent health behaviors was evaluated met our inclusion criteria and were included in this SR.

Nine additional articles met the inclusion criteria during title and abstract screening. However, on full-text review, 2 of these articles had participants that were mostly outside of our target age range, and 7 contained examinations of outcomes other than our predefined primary or secondary outcomes of interest (see Fig 1).

Two independent observers extracted information from each study and 2 authors checked data extraction for completeness and accuracy (see Table 1). When possible, we reported results only for adolescents 12 to 19 years of age when the study sample also included younger or older participants. If the mean participant age was less than age 12 years, we contacted authors for adolescent participant specific data.

We used Review Manager (RevMan),⁴¹ the Cochrane Collaboration’s software for preparing SRs and meta-analyses, to organize, manage, and analyze the data using an inverse-variance statistical method. We used the Grades of Recommendation, Assessment, Development, and Evaluation (GRADE)⁴² software (GRADEpro) to rate the quality of the evidence for outcomes as recommended by the Cochrane Handbook for Systematic Reviews.⁴³ For each outcome, 2 authors independently extracted data and cross-checked against the data that were entered in RevMan. Throughout the article selection process, data abstraction, computation, calculation, evaluation, and synthesis process, 2 authors resolved disagreements through joint examination of the articles and discussion until consensus was reached.

We used The Cochrane Collaboration’s Tool⁴³ for assessing risk of bias in RevMan⁴¹ to assess included studies across 7 domains. We included randomized and nonrandomized studies in the risk of bias assessment and extracted data regarding each domain. Two authors rated each domain as being high, low, or unclear risk of bias using criteria indicated by the Cochrane Handbook for Systematic Reviews.⁴³ We used the following rules for judging risk of bias for incomplete outcome data for each individual study: the final sample dipped below the sample size calculation, imbalance in numbers or reasons for missing data between groups, loss to follow-up >20%,^44,45 or substantially different rates in attrition between groups.⁴³ 

We assessed the quality of evidence using the GRADEpro tool,⁴² which considers within-study risk of bias, directness of evidence, heterogeneity, precision of effect estimates, and risk of publication bias. We imported data from RevMan⁴¹ into GRADEpro.⁴² Two authors independently rated the quality of evidence for each comparison and outcome across the included studies and then produced a “Summary of Findings” table (see Table 2) using the GRADE Handbook⁴² criteria. When CIs included or crossed 0, we conducted calculations for comparison groups for each outcome using *GPower Sample Size Calculator⁴⁶ to determine optimal information size⁴⁷ using a 1-sided α of .05 and power of .80. The actual means and SDs from the meta-analysis of each outcome were used to calculate effect sizes, which ranged from 0.01 to 0.27. Two authors conducted and cross-checked calculations.

When there was >1 follow-up period reported, we selected the point with the greatest improvement in outcome measurements. When >1 arm in the intervention using MI existed, we selected the intervention arm that had the greatest improvement in outcome measurements.

For consistency in measurement outcomes, 2 authors converted and cross-checked measurement units to the American Medical Association preferred units of measurements where needed.⁴⁸ For studies missing required data elements, we e-mailed authors a request for the missing data, sent a second e-mail, and e-mailed a coauthor when needed. When possible, for studies where data were not available or authors did not respond to requests, we computed SDs from the available data using formulas and methods recommended by the Agency for Healthcare Research and Quality for handling missing continuous data instead of omitting the study.⁴⁹ Two authors conducted the computations and cross-checked for consistency.

To investigate statistical heterogeneity, we used a fixed-effects model in RevMan⁴¹ and produced Forest plots with the I² statistic. Forest plots provide visual variability in point estimates of the effect size and CIs; I² quantifies the percentage of the variability in effect estimates due to heterogeneity rather than to sampling error (chance).⁵⁰ A significant Q (Cochran Q = χ²) with P < .05 or I² value >50% suggests substantial heterogeneity.⁴³ If heterogeneity was present, we performed a random effects analysis, which equally weighs all included studies to account for between study variance due to sample size differences.⁵¹ 

To investigate reporting bias, we used Funnel plots produced by RevMan⁴¹ software, which provide visual scatter plots of the effect estimates against the study’s size. In the absence of bias and between study heterogeneity, the scatter will be due to sampling variation and the plot will resemble a symmetrical inverted funnel.⁵¹ Heterogeneity, reporting bias, and chance may lead to asymmetry.⁵¹ 

We reported outcomes from all 17 studies as an SR to synthesize the data and only included RCTs in the meta-analysis per the Cochrane Handbook, section 13.1.2.⁴³ We excluded 6 studies from the meta-analysis. Four were nonrandomized studies; 1 RCT did not have follow-up mean or SD values that are necessary for meta-analysis, and we were unable to get these values or additional information from the authors to compute mean and SD values; and 1 RCT did not have a non-MI control group. Sufficient data were available for meta-analysis across 11 studies; we conducted meta-analyses using RevMan⁴¹ to produce overall estimated pooled treatment effects as relative effect estimates and mean differences (MDs) with 95% CIs for each outcome. We included the following outcomes in the meta-analysis: BMI, BMI percentile, BMI z score, waist circumference, fasting glucose, triglycerides, total cholesterol, and insulin. An MD was appropriate for this review because RCTs contained reports of outcomes as continuous data from standard measurement scales. Health behavior and quality of life outcomes are reported only as a qualitative synthesis.

We performed a sensitivity analysis as set forth by the Cochrane Handbook, section 9.7.⁴³ We requested, received, and included our eligibility age range (ages 12–19 at study enrollment) data in the analysis. Only 1 study that met inclusion criteria was included in SR; this study was excluded from the meta-analysis because of missing data that could not be computed or imputed, but the study did not contain reports of effects on weight-related outcomes. We undertook the entire meta-analysis twice for all outcomes using a fixed-effect model followed by a random effects model, and the overall results were not affected.

Our electronic search yielded 1545 records through database searching and an additional 169 records through review of citations and hand searching. After we removed duplicates, there were 1336 abstracts. The first round of double screening excluded 1310 records on the basis of title and abstract. We identified 26 full-text articles for additional review and determined that 17 studies met the inclusion criteria for this SR, including 13 RCTs and 4 other types of studies (1 quasi-experimental, 1 cohort, and 2 pre and post). Eleven RCTs were included in the meta-analysis (see Fig 1). Full details of the included studies are provided in Table 1.

All 17 studies examined anthropometrics, and only 3 reported significant effects on BMI,^31,37 BMI percentile,³² BMI z score,³⁷ and waist circumference³²; 1 of these studies did not have lasting effects by the final follow-up period at 6 months.³¹ Seven studies examined cardiometabolic outcome measures, and only 1 reported significant decreases for total cholesterol and triglycerides that were not clinically relevant.³⁷ Fourteen studies examined physical activity. Three studies reported significant effects on self-reported sedentary behaviors^32,36,39; 2 reported contained reports of significant effects on physical activity duration,^31,37 energy expenditure,^31,37 and self-reported activity measures^37,39; and 1 reported significant effects on fitness.³⁰ Out of the 11 studies that examined self-reported dietary habits, 4 studies reported significant effects,^34,36,37,39 and 1 study reported overall success in meeting diet and physical activity behavior goals.²⁹ Three studies evaluated quality of life outcomes, and 1 reported significant effects on self-reported school functioning, emotional functioning, physical health, and psychosocial health.^27,28,37 All 17 studies were focused on lifestyle changes and incorporated general education on nutrition and/or physical activity into the core MI-based intervention sessions. Fifteen studies included both didactic nutrition and physical activity education, with 2 of those also adding an exercise class component. Two studies were focused on didactic physical activity education, with 1 of those also adding an exercise class component. Three of the 16 studies augmented MI with cognitive behavioral therapy (CBT), and 11 studies involved parents. Nine of the 16 studies reported significant improvements in nutrition and/or physical activity habits.

Overall, there was high variability in the number of MI sessions included in the interventions. The majority of the studies had relatively short-term follow-up periods, and the biggest outcome improvements tended to occur in studies with follow-up periods that were 6 months or less. Reported improvements were primarily in nutrition and physical activity behaviors versus anthropometric or cardiometabolic outcomes.

Risk of bias assessments are presented for each domain as percentages across all 17 studies in the SR (see Fig 2) and for each study (see Fig 3). Overall, risk of biases common to the majority of the included studies were related to lack of blinding of participants, personnel, and those assessing outcomes. In addition, approximately half of the studies were assessed as being at high risk for bias related to allocation concealment and incomplete outcome data. Overall, the risk of publication bias was deemed low; funnel plots produced for each relevant outcome from the 11 RCTs included in the meta-analysis (see Supplemental Information for Fig 4) appear to be fairly symmetrical, although smaller studies tend to have larger effect sizes.

We included 11 total RCTs in the meta-analysis, with a total of 1245 participants, follow-up duration of 3 to 13 months, 1 to 16 intervention sessions, and study sample sizes of 32 to 357 participants (see Table 1). Participants were predominantly female sex. Overall, there was evidence of heterogeneity for only 1 cardiometabolic outcome out of 8 outcomes that were examined (triglycerides, I² = 77%, P = .01).

The results of the meta-analyses are presented in Table 2 (see Supplemental Information for Figs 5–12). We found nonsignificant average reductions in BMI and BMI percentile. There were no discernible effects on BMI z score, waist circumference, glucose, triglycerides, total cholesterol, or insulin. The optimal information size necessary for detecting a statistically significant MD was not met for any outcome. Dose response gradients or plausible confounders were not detected on the basis of criteria set forth by the GRADE Handbook.⁴² 

Although most of the studies reported MI training for providers, training efforts were highly variable, ranging from online learning modules to 3 full days of direct training. The majority of the studies did not discuss ongoing coaching or supervision. Ten studies specified that Motivational Interviewing Treatment Integrity coding was done; coding indicated that providers delivering the MI intervention inconsistently met levels of MI proficiency across studies.

The main finding of this SR and meta-analysis on the use of MI to treat adolescents with overweight and obesity are nonsignificant reductions in some anthropometric outcomes, no discernable effects on cardiometabolic outcomes, and some qualitative evidence of positive effects on nutrition and physical activity behaviors and quality of life. Even pooling participants across studies, there was an issue with achieving adequate power for any of our primary outcomes, which must be considered as a viable explanation for the predominantly negative findings in the meta-analysis. Overall, the quality of evidence from the studies was rated predominantly moderate, indicating moderate confidence that the outcome estimate effects are near the true value across studies.⁴³ In addition, sensitivity analysis indicated the results of the analysis can be regarded with a relatively high degree of certainty.

Our findings are somewhat in contrast with those of Cushing et al,¹⁷ in part because of different approaches to examining the data. Cushing et al¹⁷ found a small, significant positive effect size on health behaviors overall in adolescents (g = 0.16, 95% CI 0.05 to 0.27)¹⁷ using Hedges’ g calculation to determine an overall effect size for each study, and then aggregating those overall effect sizes across studies. Thus, their overall finding encompasses the effect of MI on several outcomes, including anthropometry, cardiometabolic outcomes, risky sexual behavior, repeat birth, sleep, dietary and physical activity behaviors, and asthma symptoms. Their findings for specific outcomes examined in this review, such as anthropometry, were fairly consistent with ours, with the weighted mean effect sizes for studies containing examinations of anthropometric outcomes hovering at ∼0 (g = −0.10 to 0.07).¹⁷ We felt that it was important to specifically examine the impact of MI on clinically relevant outcomes like weight status and cardiometabolic indicators because these factors are generally most closely associated with poor long-term health outcomes.

Our findings also contradict another recent SR, whose authors indicated that multifaceted interventions, including family support and guided behavior modifications, seem effective for reducing BMI in adolescents with overweight and obesity.⁵² However, the authors included all weight loss interventions and did not exclusively examine MI interventions. Similar to our SR, there was considerable variability in effectiveness between interventions.

Bean et al⁵³ argued that examining the effects of MI on outcomes beyond weight and cardiometabolic outcomes can increase understanding of the mechanisms of treatment effects. We found that there was some evidence that MI may help to improve diet and physical activity behaviors and quality of life in adolescents. However, we could not conduct separate meta-analyses for these outcomes because of variation in outcome measures across studies and limited quality of life outcome data. Many of the studies reported significant improvements in nutrition and physical activity behaviors, and of the 3 studies that evaluated quality of life, 1 reported even greater effects when parent involvement was added to MI compared with MI alone. MI may help adolescents engage effectively with other treatments that more directly affect nutrition and physical activity behaviors and quality of life.^54,55 This fits with the basic philosophy that MI primarily improves the collaborative relationship between the provider and client to build motivation to change.²⁰ 

Most of the studies in the current SR included ≤6 MI sessions and managed patients for <1 year. It is likely that more ongoing contact may be necessary to impact anthropometric and cardiometabolic outcomes. The USPSTF² found that at least 26 contact hours per year seemed to be the threshold necessary to promote weight loss in the context of behavioral interventions for pediatric patients; none of the studies included in this review met this threshold. It is likely that intensive, ongoing support is necessary to address fluctuations in motivation and adherence and metabolic and physiologic energy-balance adaptations that often frustrate long-term weight loss and maintenance efforts.⁵⁶ 

Finally, many of the studies in the current SR did not assess treatment fidelity and used provider training models that may not provide adequate support for implementing MI. Concerns regarding MI treatment fidelity are salient, given evidence that training workshops alone do not typically result in enduring changes in practice⁵⁷ and that MI skill fluctuates between providers and over time.^21,58 According to an SR by Hall et al,⁵⁹ in the absence of supervision and ongoing training after initial training, the majority of clinicians are unlikely to achieve beginning efficiency in MI. Moreover, comfort with MI may not be achieved until at least 3 months, even with ongoing use and coaching,⁵⁸ and proficiency and skill may not be achieved until 6 to 12 months.^60,61 Given the relatively short duration of many of the studies in the current SR, it is possible that many of the providers delivering the MI intervention may not have achieved proficiency and skill. This was reflected in variable findings related to provider MI proficiency in studies in which MITI coding was conducted.

Results of this SR and meta-analysis should be interpreted in the context of the limitations. Overall, there was a range of evidence quality, with fairly small sample sizes and risks of bias related to lack of blinding of participants, personnel, and those assessing outcomes, allocation concealment, and incomplete outcome data. Other reviewers might reach different conclusions about the risks of bias and strength of the evidence on the basis of their own judgements. However, we applied stringent criteria in grading the evidence and have aimed for transparency regarding the judgements that we reached. In addition, women were overrepresented in the studies that were included, potentially limiting the generalizability of the results. Finally, none of the included studies met the USPSTF-recommended 26 contact hour threshold for behavioral interventions for weight management in pediatric patients.

There is little indication in this SR and meta-analysis that MI impacts anthropometric and cardiometabolic outcomes in adolescents with overweight and obesity. This finding may reflect a true lack of effect, or it may be related to issues with inadequate power or treatment dose, delivery, or duration. Future studies should attempt to address these shortcomings. There is some evidence that MI, especially in conjunction with other supportive interventions, may positively impact nutrition and physical activity behaviors and quality of life outcomes. Standardization of nutrition and physical activity measures across interventions, as well as more routine measurement of quality of life, would facilitate a future meta-analysis on these outcomes. The full applicability of MI for weight management in adolescents is yet to be determined. However, the results of this SR and meta-analysis are applicable in clinical practice in that MI may effectively promote adolescent engagement and positive behavior changes, especially when used with complementary interventions.

CBT

cognitive behavioral therapy

CI

confidence interval

GRADE

Grades of Recommendation, Assessment, Development, and Evaluation

MD

mean difference

MI

motivational interviewing

RCT

randomized controlled trial

RevMan

Review Manager

SR

systematic review

USPSTF

US Preventive Services Task Force

The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health. We thank Daveer Menchaca, medical student, Alyssa Mirabel, medical student, Nadine Montoya, dietetic intern, Christina Fallows, dietetic intern, and Jessica Hammond, dietetic intern for their assistance in the data extraction process.