Outcome of BMI²⁺: Motivational Interviewing to Reduce BMI Through Primary Care AAP PROS Practices

Eighteen practices were randomized (9 to intervention and 9 to usual care [UC]) in 2017. Parents were recruited and enrolled in the intervention between November 30, 2017, and April 30, 2019, and intervention delivery to parents ended January 31, 2021. See supplement for CONSORT diagram.

Full details about the protocol were previously published²⁰ . Briefly, practices and pediatric clinicians were recruited through the AAP’s Pediatric Research in Office Settings national primary care practice-based research network and all used a common electronic health record (EHR) vendor. Institutional Review Board (IRB) approval was obtained from the AAP, the IRB of record for the trial. University of Michigan formally relied on the AAP IRB. The Children’s Hospital of Philadelphia IRB deemed this study exempt. All parents of eligible children from intervention practices were mailed study information, and interested parents provided informed consent and Health Insurance Portability and Accountability Act authorization.

Practice randomization was conducted by a statistician blinded to practice identity, stratified based on racial and ethnic composition and size of their patient population. Intervention practices were compensated.

Over the course of the study, 1 intervention practice of the 9 stopped recruiting after 33 patients for reasons unrelated to the study but allowed continuation of the RD calls and data extraction for analyses. Two UC practices of 9 discontinued using the EHR vendor, so their outcome data were truncated.

Twelve providers (from 9 intervention practices) received 2 days of in-person training in MI as well as billing and coding for weight management services and the study protocol. Continuing medical education and maintenance of certification parts II and IV were offered. MI booster training sessions were provided for pediatric clinicians (optional) and required for RDs.

The UC pediatric clinicians continued their standard care. They were trained in study procedures via phone and given access to recorded webinars describing current treatment recommendations and billing and coding for weight management services. At the end of the study, UC pediatric clinicians were offered a free MI training.

• Ages 3 to 11

• BMI value for age and sex ≥85th percentile

• ≥1 health supervision visits with a study clinician in past 2 years

• Type I or type II diabetes

• Medications known to affect growth and mood or behavior

• Limiting, severe medical disorder

• Sibling enrolled, withdrew, or refused consent

• Recommendation by the primary care clinician not to participate.

The intervention comprised 4 key components¹ : in-person and telehealth MI counseling by the pediatric clinicians² ; telephone MI counseling from a RD³ ; text messages with reminders and tailored behavioral messages; and⁴  study portal for RDs, pediatric clinicians, and parents.

Pediatric clinicians were asked to complete 4 MI-based in-person or telehealth sessions with enrolled parents over 2.0 to 2.5 years. Because of COVID-19, the study was extended about 6 months to allow parents to complete their MI sessions and final height and weight measurements (see Supplemental Information). MI sessions were conducted both as stand-alone office visits and as part of other scheduled appointments, primarily health supervision visits.

RDs scheduled up to 6 telephone counseling sessions with enrolled parents during the study period. RD counseling was available in English and Spanish. Although all RD sessions were offered free of charge, pediatric clinicians billed for their visits.

The intervention focused on discrete target behaviors that included: sweetened beverages, eating out, fruits and vegetables, sweets, screen time, and physical activity²⁰ .

Enrolled parents received tailored text messages (English or Spanish) aimed at enhancing motivation and reinforcing behavior change throughout the intervention period. Some texts contained links to additional educational content. In addition, parents received automated reminders to schedule and complete RD counseling calls.

MI fidelity was assessed using standardized patient role-playing for both the pediatric clinicians and RDs and audio recordings of actual patient encounters for RDs using the ONE-PASS system²¹ . Pediatric clinicians received feedback on 1 standardized patient encounter at the end of their 2-day training, and an additional feedback session on a second standardized patient encounter was optional during the course of the intervention. RDs received feedback for ≥10% of their sessions, with more frequent supervision provided for RDs who did not meet competence criteria (average score ≥5) on the ONE-PASS.

The main outcome was child’s BMI as a percent of the 95th percentile for children of the same age and sex, a more reliable measure of adiposity change than other measures, such as BMI z-score, particularly on the upper end of the BMI distribution^22,23 . The primary outcome was assessed using data from baseline through 24 months. As an exploratory outcome, we report results through 36 months for approximately 70% of the sample. All BMI values were used regardless of the number of values in the record.

For each child enrolled in an intervention practice, we matched them based on age and gender, using EHR data to 3 children from a control practice. For all matching, the child had to have at least 1 BMI value 10 months or more from the start of enrollment. We did an initial exact match on paired practice, birth year with a 1-year caliper, and follow-up BMI data at least 10 months from the start of enrollment, and then rematched across practices any cases without 3 matches within the same paired practice.

Mixed effects linear regression allowing for repeated measures was used to examine the change in percentage of the 95th percentile BMI in intervention children compared with UC children over the course of the study. Covariates included child sex, child age, baseline BMI (ie, last percentage of 95th percentile BMI obtained before the start of the study), ethnicity and race (Hispanic, Non-Hispanic white, Non-Hispanic Black, Non-Hispanic other, and missing), elapsed time since the start of the intervention (measured in weeks; time-varying covariate), time elapsed squared (to account for a nonlinear effect of time), and insurance type (commercial, Medicaid, and unknown). For the intervention arm only, we had parents complete a baseline questionnaire that included demographics, and this self-reported data were used to supplement any missing data.

Our linear mixed-effects models were run using SAS Software (version 9.4) Proc Mixed Procedure. Each model accounted for both the clustering of observations within the same practice by including a random effect for site and repeated visits within each patient. This linear mixed-effects approach allows for patients to have a varying number of follow-up visits and differential follow-up times.

Postestimation tests were done to estimate the effect of intervention at enrollment, 6 months, 12 months, and 24 months after the study start. To aid in interpretation, we also report results using raw BMI, using the same covariates, except age at the time of visit (time-varying) was used in place of age at the first visit on or after the study start (fixed).

We examined outcomes for the subset of children whose parents actively participated in the intervention, defined as receipt of ≥50% of the total MI counseling sessions from pediatric clinicians (≥2 or more), and RDs (≥3) versus all eligible children in UC practices. We examined outcomes for the total eligible population of participating clinicians in the intervention practices who had a follow-up visit with height and weight at least 10 months from baseline. This was considered the population effect and was intended to detect possible “spillover” effects of practice participation and of the pediatric clinicians’ counseling among families that did not directly enroll in the study. A priori, we were interested in 2 potential effect modifiers, race and ethnicity and sex as these have been associated with differential weight status as well as response to prior interventions^3,16,24,25 . We therefore tested if the main outcomes differed across race and ethnicity groups by including a 3-way interaction term (intervention, time, and categorical race and ethnicity). We similarly examined the moderating effect of child sex and age.

The study was powered to detect an effect of around 0.20. Sample size estimates accounted for practice-level clustering, assuming an intraclass correlation coefficient between 0.001 and 0.03, based on prior studies. Based on these assumptions, we required 7 practices per study group (2 additional practices were recruited in each group to account for practice attrition) and an average of 35 enrolled parents per intervention practice (target n = 316).

At baseline, children in the 2 arms were similar with regard to age and sex. However, intervention youth, using unadjusted data, had a significantly higher percent of the 95th percentile than UC youth, 106.4 compared with 102.7, as well as higher raw BMI, 21.4 versus 20.5 (Table 1). There was also a significantly higher percentage of Hispanic and Black and commercially insured youth. Mean counseling dose was 2.1 of 4 sessions for pediatric clinicians and 2.3 of 6 sessions for RDs.

The primary outcome analysis indicated a significant treatment x time interaction (b = 0.017, 95% confidence interval [CI] 0.0066 to 0.027). Youth in the intervention arm showed a greater relative increase in their percent of the 95th percentile (per month), ie, gained more relative weight for age and sex than youth in the UC arm over the course of the study (Fig 1). Results were identical when using raw BMI as the outcome. There were no significant between-group differences at enrollment, 6 months, 12 months, and 24 months for either percent of the 95th BMI percentile or raw BMI. In terms of absolute weight gain, intervention youth gained approximately 0.5 more BMI units and 1.0 additional kg at 2-year follow-up relative to UC youth (Table 2).

Although not our primary timeframe for evaluating the intervention, we continued to collect data for months 24 to 36, and there was an overall treatment x time effect favoring the UC group. The contrast at month 36 was not significant for either percent of the 95th percentile or raw BMI.

When comparing only youth whose parents (32%) were more engaged in the counseling sessions, the treatment x time interaction was not significant, nor were any of the pairwise comparisons at 6, 12, and 24 months. When including all youth of intervention clinicians, not only those enrolled in the program, the treatment x time interaction for distance above the 95th percentile (interaction b = 0.00173; CI 0.0067 to 0.017) and raw BMI (interaction b = 0.0032; CI 0.0021 to 0.0044) were statistically significant in favor of the UC group. None of the contrasts at specific time points were statistically different for either outcome. The 3-way interaction of race and ethnicity, intervention, and time was significantly associated with the main outcome (P < .0001). When stratifying by race and ethnicity, the difference in percent from the 95th percentile, between intervention and UC children over the course of the study was significant for Black (b = 0.06; CI 0.025 to 0.092) and Non-Hispanic other youth (b = 0.06; CI 0.017 to 0.01) but not for white (b = 0.002; CI −0.013 to 0.02) or Hispanic youth (b = −0.003; CI −0.02 to 0.01). See Fig 2. Looking at the contrasts at the 4 study time points, none of the between-group differences were significant within any race and ethnicity group. There was no significant 3-way interaction between treatment and age or sex, so no further stratified analyses by sex were conducted.

Adverse events were monitored by the lead study dietitian (author K.S.) and reviewed with the Data Safety Monitoring Board as needed.

This study tested the effectiveness of a motivational interviewing intervention, previously found to be efficacious²⁶ , when delivered by pediatric clinicians and RDs under unprecedented, COVID-19 pandemic conditions. There was no overall benefit of the intervention, and contrary to expectations, some children in the intervention arm appeared to gain more weight, based on percent of the 95th percentile and raw BMI, than the matched children from UC practices. Although the absolute excess weight gain among intervention youth relative to UC youth was small, around 0.5 BMI units or 1 kg over 2 years, the reverse effect was nonetheless statistically significant and unexpected. The clinical significance of the findings is unclear. There are several potential explanations for these results.

First, MI in combination with our other intervention elements may be ineffective for weight control in this age group. Although this is inconsistent with the positive results observed in the BMI pilot and the BMI²  efficacy study,^26 –28  it is consistent with several prior MI studies in overweight children^16,29  and teens¹⁸ , which reported largely null effects. None, however, reported a negative effect as we observed. Second, is insufficient dose or type III error³⁰ . Although this might explain null results, it would not necessarily explain the reverse effect. Moreover, the dose response analyses did not indicate any positive effects for families receiving a higher dose, although this may be limited by small sample size of the high dose group. Thus, insufficient dose is an unlikely explanation for our results, particularly the inverse effects. A related possible explanation is that more limited intervention skills, particularly on the part of pediatric clinicians, may have contributed to the results. On the standardized patient encounter immediately after the initial MI training, the clinicians’ average score was in the lower range of fidelity. For RDs, a randomized sample of 2 to 5 sessions from each RD was scored by an MI supervisor with 81% exceeding criteria for competence. Both clinicians (optional) and RDs (required) were offered additional coaching throughout the study. We had felt these decisions aligned with the “real world” nature of the project. Finally, a key difference between the current study and the prior 2 positive studies was that we approached practices using a particular EHR and did not require a prior level of interest in MI.

Although the overall negative intervention effect was significant for the full sample, further analyses found that the negative outcomes were only significant among Black and other youth. Given the small number of Black youth in our study who came from a small number of practices, and that the negative intervention effect was also observed for “other” youth further suggests caution in generalizing our findings or over-interpreting them. Nonetheless, there is some evidence from prior studies that some African American patients may not have the same preference for or response to MI or similar interventions^24,31 –33 . Our findings suggest that providers should try to tailor their communication style to the needs and preferences of their patients^33,34 . These preferences could be affected by racial nonconcordance of the counselor, which has been shown to affect clinical outcomes³⁵ . Whether racially concordant providers may alter the impact of MI is an important research priority which we are unable to answer in our study because of the small number of clinicians and participants who were Black.

There may have been other methodologic differences in the current study that could have biased our results. Only parents in the intervention group had to actively enroll into the study, whereas UC youth were matched after the fact, based on age and gender. This could have created several biases. If the intervention group was more motivated to participate, this would likely have predisposed them to have better, not worse outcomes. On the other hand, the intervention participants might have had greater need for help than the UC families, which might have contributed to worse outcomes. Youth in the intervention arm were more likely at baseline to have BMI values greater than or equal to the 120th percentile for age and sex. Although this is adjusted for in the main outcome analyses, it nonetheless could have influenced response to the intervention. Youth with more severe obesity might have needed a more structured and intensive intervention than was provided.

More broadly, this study suggests that the model used needs reconfiguration. The required dose to achieve weight effects may be unattainable in our current healthcare delivery system. Further, the use of dietitians and pediatricians may need to be reconsidered. It may be beneficial to include other health care professionals who are primarily trained in behavioral change counseling and motivation, perhaps providing them with nutrition counseling skills or perhaps engaging dieticians only once or twice during the intervention. Additionally, although our intervention included some limited digital components (eg, text messaging), a more intensive eHealth intervention may increase effectiveness.

Finally, this study was conducted largely during the heart of the COVID-19 pandemic, including the March through June 2020 lockdown, when US youth gained considerable extra weight^2,36,37 . One study conducted in the Kaiser Permanente Health system showed that Black and Hispanic youth appeared to gain more weight than other racial groups during lockdown³⁸ . During the COVID, pandemic clinicians may have been hesitant to ask families to tackle another stressor (ie, weight), particularly since many youth had reduced access to healthy foods and their normal physical activities. Whereas again this might contribute to the lack of intervention effects, this is an unlikely explanation for the negative effects observed.

Contrary to 2 prior studies, MI delivered by pediatric clinicians and RDs did not improve the weight status of participating children. Methodologic and cultural factors as well as the COVID-19 pandemic may have contributed to our findings.

We thank all BMI²⁺ practices (listed below), pediatricians, nurse practitioners, registered dietitians, office staff, caregivers, and families who participated in our study. BMI²⁺ practices participating in this study, named here with their permission, included: Cypress Pediatrics; Delaware Valley Pediatric Associates, PA; El Paso Pediatric Associates; Holland Pediatric Associates; KID-DOC Pediatrics; Palmetto Pediatrics and Adolescent Clinic, PA; Pediatric Associates of Kingston; Pediatric Center at Renaissance; Pediatrics in Brevard, Skagit Pediatrics, LLP; Tendercare Pediatrics; Union Pediatrics.

AAP

American Academy of Pediatrics

IRB

Institutional Review Board

MI

motivational interviewing

RDs

Registered Dietitians

UC

usual care

UM

University of Michigan