Study design, setting, and population
For this retrospective cohort study, we identified youth (n = 38,776) with obesity (defined as BMI-for-age ≥ 95th percentile) [20] between the ages of 3 and 17 years who received care at the Kaiser Permanente Medical Offices in Orange County, California, between April 2014 and Dec 31, 2018 (Fig. 1). The study cohort was described in detail elsewhere [19]. The study protocol was reviewed and approved by the KPSC Institutional Review Board. Informed consent was waived for this study.
Youth were eligible for the study if they were obese (BMI-for-age ≥ 95th percentile), did not have insulin-dependent diabetes, chromosomal anomalies predisposing to obesity such as Prader-Willi syndrome or Trisomy 21, eating disorders including bulimia or binge eating, prior history of medications or surgery for the treatment of obesity, did not have a history of prolonged steroid use (> 6 months) for treatment of a chronic illness, and were not pregnant.
For the intervention group, we identified youth who were enrolled in a family-based weight management program using International Classification of Diseases, 9th and 10th Revision, Clinical Modification (ICD-9 and ICD-10) Z71.3 and V65.3 and a KPSC-specific code for weight management. We included youth participating in one or more sessions over 12 months of follow-up (n = 341) to assess a dose-response relationship (Fig. 1).
As control groups, we identified youth with obesity who were referred to the family-based weight management program but did not participate in any sessions (Ref-CG, n = 317), as well as area-matched youth who were never referred and were not enrolled in a K.P. weight management program (Area-CG, n = 801).
Family-based weight management intervention
The Orange Country pediatric weight management program started in 2014 to support families in managing their child’s body weight. Individual appointments provided the ability to identify specific needs and barriers to change. Four pediatric care providers administered the family-based weight management program (FB-WM). The program used motivational interviewing (MI) approaches [21,22,23,24,25,26]. Youth were referred to the program by their care provider. The program consisted of 30 min counseling appointments provided by a pediatric nurse practitioner or pediatrician. While many counseling models rely heavily on directive advice and information exchange, an MI counselor generally avoids direct attempts to convince or persuade. MI is a patient-centered counseling style that explores, strengthens, and guides an individual’s motivation for change [27, 28]. It relies on specific techniques, which include reflective listening and eliciting change talk [29, 30]. The visit included the child and usually one or more parent or caregivers; in some cases, the entire family. Family members accompanying the child varied from visit to visit. About 1/3 of visits with adolescent patients took place without the presence of a parent or guardian by choice. In these cases, parents were contacted by phone to ensure family involvement. These visits were offered to provide patient-centered care and to avoid exclusion of teenagers with working parents. The visits focused on one or more of the following target areas: snack foods, sweetened beverages, eating out, whole grains, fruits, vegetables, sweets and desserts, portion size, T.V./screen time, video games, and physical activity. Provider and patient determined together which behaviors were most amenable to intervention, which goal they wanted to commit to, and work into a personalized plan. This meal/exercise plan was created and provided for each patient/family. Handouts developed by Kaiser Permanente for weight management supported the goal setting. If needed, we sent letters to schools, babysitters, and other caretakers asking for their help supporting the family. The core of the intervention used the 5-4-3-2-1-GO! Tool developed by the Consortium to Lower Obesity in Chicago Children (CLOCC) [31,32,33]. This plan implemented 5 fruits/vegetables per day, 4 glasses of water daily, 3-servings of low-fat dairy daily (alternatives provided for children with lactose intolerance or those who refuse dairy), 2 h or less of screen time per day, 1 h physical activity per day, and 0 sugared drinks. The healthy plate materials were adopted [34]. These guidelines were reviewed at each visit and used to review progress, provide feedback, and support goal setting.
A written food diary was reviewed at follow-up visits, and if not available, was obtained by memory recall. Encouragement was given for positive changes, regardless of a weight change. Provider and patient identified problem areas when applicable and discussed solutions. Behavioral goals and BMI changes were reviewed if appropriate.
The program was designed to be completed in 3 to 6 thirty minute counseling sessions (1.5 to 3 h). Patient, family, and care providers decided together when the intervention was complete but had the option to continue if additional counseling sessions were needed, even beyond the original length of the 6-month program. In the present analysis, all patients were included as “intention-to-treat” even if the patient/family discontinued the program before the third visit.
Efforts to train all providers in MI techniques are ongoing and not fully implemented. Informal training has been provided to many but not all pediatric care providers, and they are currently not certified in MI with standardized quality controls and regular training. The same four providers consistently provided care. The fidelity of the intervention based on a written framework was maintained through regular team meetings but without formal fidelity assessments.
Intervention and control groups
The analytic cohort consists of three groups: 1) a group of children who received a family-based behavior-changing weight management intervention (FB-WMG), 2) a control group of children referred by their provider who did not participate in the intervention (Ref-CG), and an area control group (Area-CG) matched for sex, age, relative distance from the median BMI-for-age, zip code and time of the first visit (allowing +/− 1 year around the window of their match). We used a 2-step matching approach. In the first step, we matched 230 out of 341 children using index date (+/− 365 days), age (+/− 1 year), relative distance from the median BMI-for-age (+/− 0.5%), and exact zip code. During the second step, another 111 children were matched using index date (+/− 548 days), age (+/− 1 year), and relative distance from the median BMI-for-age (+/− 0.5%). The index date for FB-WMG was at the first appointment. For Ref-CG, the referral date (or nearest office visit date with BMI) was used as the index date. For Area-CG, the index date was the date of the BMI measure used for matching.
Height and weight were measured in light clothing without shoes and used to calculate sex-specific BMI-for-age [20]. To determine BMI at 6 months of follow-up, BMI was calculated from weight and height measured during a routine outpatient visit within 1 month before and 3 months after 6 month of follow-up. If no weight and height were available close to the 6-month follow-up, BMI was calculated by linear regression using 2 weights and heights within 3 months around the 6-month follow-up (< 10% of measures). To determine BMI at 12 months of follow-up, BMI was calculated from weight and height measured during a routine outpatient visit closest to the 12 months follow-up within 3 months before and 3 months after 1-year follow-up. If no weight and height were available close to the 12 months of follow-up, BMI was calculated by linear regression using 2 weights and heights within 6 months around the 12-month follow-up (< 5% of measures).
Study outcome
Change in body weight was calculated for each follow-up visit as the absolute and relative difference in the distance from the median BMI-for-age and sex [35]. This metric is a more reliable measure of change in adiposity, particularly for individuals in the upper end of the BMI distribution compared to other methods such as BMI z score [35]. It also does not have an upper limit (as has BMI-for-age percentile) and can be used to assess adiposity across the entire BMI spectrum.
Covariates
We obtained race and ethnicity information from health plan administrative records and birth records. We categorized race/ethnicity as non-Hispanic White, Hispanic (regardless of race), African American, Asian or Pacific Islander, and other, multiple or unknown race/ethnicity. In addition, we used median household income and education in the patient’s residential census tract as area-based measures of socioeconomic status [36, 37]. Census-tract household income was classified using the individual’s likelihood of a median household income of < $45,000, $45,001 to $80,000, $80,000 or more. Neighborhood education was categorized using an individual’s likelihood of education with some college or higher. We also used insurance through government healthcare assistance programs (yes/no), such as MediCal, as an additional proxy for socioeconomic status.
Statistical analysis
Baseline characteristics of the study cohort are presented for all three study groups: for the family-based behavior-changing weight management intervention group (FB-WMG), referred control group (Ref-CG), and area control group (Area-CG) using means with standard deviation (S.D.) or medians with interquartile range (P25 – P75) for continuous variables as appropriate, and the number of observations with percentage for categorical variables. Differences in characteristics between groups are assessed using the t-test, chi-squared test or fisher’s exact test contrasting intervention against control groups.
We conducted sensitivity analyses restricting FB-WMG youth to those with at least three visits and their matched Area-CG. The primary outcome measure was a change in the relative distance from the median BMI-for-age between baseline and 12-month follow-up. In addition, we present outcomes of the 6-month follow-up. Adjusted differences-in-difference (DID) and confidence intervals in relative distance from the median BMI-for-age between intervention and control groups are derived using multivariable linear regressions with robust standard error.
All models are adjusted for age, sex, race/ethnicity, baseline BMI, state-subsidized insurance coverage, neighborhood income and education, and length of KPSC membership. For simplicity and consistency, we included all three study groups in one model. We performed additional analyses using separate mixed linear regression to compare FB-WMG and Area-CG to account for the matching process. The results were essentially consistent and did not affect the overall conclusion. All analyses were performed using SAS statistical software version 9.4 (SAS Institute Inc., Cary, NC).