A Randomized Controlled Trial of Culturally Tailored Dance and Reducing Screen Time to Prevent Weight Gain in Low-Income African American Girls: Title and subTitle BreakStanford GEMSPreventing Weight Gain in African American Girls FREE

Child and adolescent obesity have more than tripled among African American girls since the 1960s, with the greatest increases since 1980.^{1 - 3} Body mass index (BMI) differences between white and black girls are present before age 6 years, significantly widen in older age groups, and are independent of socioeconomic status.⁴ Reviews of cardiovascular disease and cancer risk reduction studies in children and adolescents suggest that body weight may be more difficult to change than other risk factors.^{5 - 7} Few studies have tested intervention strategies specifically designed for African American girls and their families.^{8 - 13}

Girls' health Enrichment Multi-site Studies (GEMS) was a National Heart, Lung, and Blood Institute (NHLBI)–sponsored collaborative effort to develop and test interventions to reduce weight gain in African American preadolescent girls.^{8 - 13} This article describes the outcomes of the Stanford GEMS trial, a 2-arm, parallel-group, randomized controlled trial to test the efficacy of a culturally tailored after-school dance program and a family-based intervention to reduce television, videotape, and video game and computer use to prevent BMI gain among lower–socioeconomic status African American preadolescent girls. Participants were randomized to either the dance and screen time reduction intervention or an information-based health education intervention for 2 years.

The design, methods, and baseline sample characteristics of this study were previously described in detail.¹⁴ Community members were extensively involved in designing the study. The study and protocols were approved by the Stanford University Administrative Panel on Human Subjects in Medical Research and the NHLBI. A 6-member independent data and safety monitoring board was selected by the NHLBI and met 1 to 2 times per year to approve the protocol and review trial progress and safety.

To enroll a representative sample of lower–socioeconomic status African American girls, we recruited from schools, community centers, churches, and community events in low-income, predominantly African American neighborhoods in Oakland, California. We performed all assessments in the participants' homes, eliminating the need for families to come to a clinical research center.

Eligible girls were identified as “African American or black” by their parent or guardian and were aged 8, 9, or 10 years on the date of randomization. To select a community-based group at higher risk, girls were required to have a BMI (calculated as weight in kilograms divided by height in meters squared) at or higher than the 25th percentile for age and/or at least 1 overweight parent/guardian (BMI ≥25).¹⁵ Girls were excluded if they had a BMI higher than 35, had been diagnosed as having a medical condition or were taking medications affecting their growth, had a condition limiting their participation in the interventions or assessments, were unable to understand or complete the informed consent document, planned to move from the area, were homeless, or had no television.¹⁴

After completing the baseline measures, families and households were randomized by computer using the biased coin randomization procedure developed by Efron¹⁶ to produce similar sample sizes in each group.

The background and conceptual models for the treatment and active-placebo comparison interventions were described previously^{12 ,14} and were shaped by extensive formative research and the Stanford GEMS pilot study.^{12 ,17}

The dance and screen time reduction treatment intervention was created using the social cognitive model developed by Bandura.^{12 ,14 ,18 - 19} To incorporate African American culture into our intervention, we emphasized elements to address both surface structure (eg, culturally matched models, music, and language) and deep structure (eg, values as well as social and historical influences).²⁰

The GEMS Jewels after-school dance intervention was offered 5 days per week, 12 months per year (excluding school holidays), at community centers in selected neighborhoods. Daily sessions lasted up to 2½ hours and started with a 1-hour homework period and small snack followed by 45 to 60 minutes of learning and practicing dance routines. Three styles of dance were taught: traditional African dance, hip-hop, and step. Additional activities to maintain motivation included GEMS Jamboree dance performances approximately every 8 weeks for families and friends, including awards for each girl based on Kwanzaa principles; videotaped feedback allowing girls to teach each other and choreograph routines; opportunities for participant choice and control; and performances at public events. Dance classes were led by female African American college students and/or recent graduates from the local community when possible to serve as role models for dance, to maintain cultural identity, and to inspire educational achievement. Although individual girls were the unit of study, the dance intervention was considered an environmental intervention, with girls attending as often as they wished.

Sisters Taking Action to Reduce Television (START) is a home-based screen time reduction intervention designed to incorporate African or African American history and culture,¹² including up to 24 lessons during 2 years. Young adult, female, African American START mentors met with families in their homes to deliver each lesson, following the screen time reduction model developed in several prior studies.^{12 ,21 - 22}

The active-placebo^{23 - 24} health education comparison intervention was selected to address the possibility of resentful demoralization and/or compensatory rivalry.^{14 ,25} It consisted of state-of-the-art, culturally tailored, authoritative, information-based health education on nutrition, physical activity, and reducing cardiovascular and cancer risk. It included 24 monthly newsletters for the girls (“Felicia's Healthy News Flash”) and their parents/guardians (“Stanford GEMS Health Report”) and quarterly community center health lectures (“Family Fun Nights”). We used the same monitoring and incentive schedules included in our experimental treatment condition.¹⁴

Data collection was scheduled every 6 months in the participants' homes by trained, female, African American research assistants masked to experimental assignment.¹⁴ For test-retest reliability, a duplicate set of physical measures was prompted randomly for approximately 10% of participants throughout the study.

We chose BMI as the primary outcome measure for its accessibility, reliability, measurement validity, and clinical validity.^{26 - 27} Weight and height were measured indoors in lightweight clothing without shoes.¹⁴ Test-retest reliability was 1.00 for weight and 0.99 for height.

Waist circumference was measured to the nearest millimeter using a nonelastic metric tape at end expiration, with the umbilicus as a landmark.¹⁴ Test-retest reliability was 0.99. Triceps skinfold thickness was measured on the right arm.^{14 ,28} Test-retest reliability was 0.99. Resting blood pressure and resting heart rate were measured according to established protocols.¹⁴

Fasting serum insulin, glucose, and lipid levels were measured at baseline and at the final follow-up visit after an overnight fast of at least 8 hours.¹⁴ Assays were conducted by the Stanford University Hospital Clinical Laboratory, a participant in the Centers for Disease Control – NHLBI Lipid Standardization Program (with ≤5% intra-assay precision and ≤6% interassay precision).

We also evaluated additional secondary outcome measures and potential moderators and/or mediators of intervention effects. Physical activity was assessed annually on 3 consecutive days, including a weekend day, using an accelerometer (Actigraph; Manufacturing Technologies Inc [formerly, Computer Sciences and Applications], Pensacola, Florida) secured at the hip with an elastic belt.¹⁴ We used previously validated methods to clean the data and estimate mean counts per minute and minutes of moderate to vigorous physical activity,²⁹ using published count thresholds in adolescent girls (≥3000 counts per minute).³⁰ Physical activity preferences were assessed annually.^{12 ,31}

Television viewing, videotape viewing, video game use, and computer use were measured with a self-report instrument proved sensitive to change.^{12 ,21} Parents/guardians reported overall household television viewing.^{21 ,32} Frequency of eating meals with the television on was reported by the girls using items previously validated³³ and sensitive to change.^{12 ,21} Dietary intake was assessed annually by 24-hour recall on 3 nonconsecutive days, including 1 weekend day and 2 weekdays, using the University of Minnesota Nutrition Coordinating Center Nutrition Data System for Research.¹⁴ Data collectors were trained and certified in these protocols.³⁴

Self-reported psychosocial measures were assessed annually, including overconcern with weight and shape, using the McKnight Risk Factor Survey^{35 - 36} ; self-perceived body shape and body shape dissatisfaction, using female African American preadolescent body figure silhouettes³⁶ ; depressive symptoms, using the 10-item short form of the Children's Depression Inventory³⁷ ; self-esteem, using the 10-item Rosenberg Self-Esteem scale^{38 - 39} ; and school performance.¹²

To assess African American ethnic/cultural identity,⁴⁰ parents/guardians completed measures of African American family practices and values, preferences for things African American, racial segregation,⁴¹ and participation in African American activities,⁴² at baseline and at the final follow-up visit. Sexual maturation was self-reported at every visit using drawings and descriptions of the pubertal stages⁴³ and age at menarche. At baseline, parents/guardians reported the girl's race, ethnicity, date of birth, and household membership; parent/guardian educational level; and total household income.

Baseline treatment and comparison group differences and baseline differences between those with and without follow-up data were assessed using Wilcoxon rank sum tests–Mann-Whitney tests for scaled variables and χ² tests for categorical variables. To use all the prospective data collected, BMI measures (up to 5 per participant) defined individual trajectories of change during the entire 2-year trial.¹⁴ Using random regression models,⁴⁴ trajectories of change in BMI were regressed on intervention group assignment (centered), with the baseline value of BMI (centered at its mean) and the intervention × baseline BMI interaction as covariates.^{45 - 49} Random regression models have advantages compared with other repeated-measures analysis methods, accommodating differential lengths of follow-up, irregular measurement intervals, and missing data.^{44 ,50} Analyses followed intention-to-treat principles. For 18 girls with no follow-up data, BMI trajectories were imputed using a multivariate linear regression model predicting BMI change from baseline measures for each group. The same random regression models and imputation approaches were used for secondary outcome variables. The results did not change materially when the analysis was repeated including only girls with measures for follow-up visit 4 (completers analysis), so only the primary intention-to-treat analysis results are reported. We explored potential moderators of treatment effects defined a priori using the methods of Kraemer et al.^{51 - 52}

For a 2-sided test at α= .05, the planned sample size of 130 girls per group provided approximately 90% power to detect an intervention effect (Cohen d ≥ 0.4),^{53 - 54} equivalent to a difference of about 1 kg/m² at 24 months for a 5.5 SD in BMI and an estimated 0.90 correlation between the BMI at baseline and the 24-month follow-up visit.¹⁴

Participant recruitment, enrollment, and flow are displayed in Figure 1. Baseline characteristics and comparisons are reported in Table 1 and Table 2. Only 18 girls were unavailable for follow-up after baseline; 126 girls (94.0%) in the dance and screen time reduction treatment group and 117 girls (92.1%) in the health education comparison group completed at least 1 follow-up assessment (P = .54). Of all baseline measures, the 18 girls who dropped out of the study had lower waist circumference (mean [SD], 63.3 [10.0] cm vs 69.5 [12.8] cm; P = .03), triceps skinfold thickness (12.8 [5.1] mm vs 17.8 [8.3] mm; P = .01), and triglyceride levels (13 vs 195; 48.2 [17.0] mg/dL vs 65.5 [32.5] mg/dL [to convert to millimoles per liter, multiply by 0.0113]; P = .02) than the 243 girls with follow-up data, but there were no other statistically significant differences. The mean (SD) length of follow-up was 25.7 (8.5) months (median, 27.0 months; maximum, 40.6 months) for girls in the treatment group and 25.2 (9.6) months (median, 27.0 months; maximum, 38.9 months) for girls in the comparison group (P = .82).

The a priori goal for mean attendance rate at dance classes was 0.6 or higher (an average of 3 d/wk). However, the mean (SD) attendance rate for girls in the treatment group was only 0.21 (0.22) (median, 0.12; interquartile range, 0.02-0.34; minimum, 0; maximum, 0.81) for possible dance classes, from randomization to their last assessment. In addition, 9 (6.7%), 73 (54.5%), 25 (18.7%), 17 (12.7%), and 10 (7.5%) girls attended classes a mean of 0, greater than 0 to less than 1, 1, 2, and 3 or more days per week, respectively. Attendance rates decreased during the study (Figure 2). Two main challenges affected dance class attendance. First, changes in community center leadership or episodes of violent crime at or near the community centers where dance classes were held necessitated changing intervention sites 6 times. Second, the local transportation vendor abruptly ended service early in the study. We eventually provided our own vans and drivers, but attendance rates never fully recovered. At follow-up visit 4, girls reported practicing dance outside of class a mean (SD) of 2.7 (2.6) d/wk (45.1% on ≥3 d/wk) for a mean (SD) of 0.83 (0.50) hours (37.2% for ≥1 hour), confirming the motivating aspect of the intervention.

We were able to deliver a mean (SD) of 12.4 (6.3) START lessons (median, 13 lessons; interquartile range, 7-18 lessons) of 25 possible lessons. Of the families, 70.1% received at least the first 7 lessons, defined as the basic skills portion of the intervention; 29.1% received 7 to 14 lessons; 34.3% received 15 to 20 lessons; and 6.7% received 21 or more. In addition, 76.9% hooked up at least 1 electronic television time manager (TV Allowance; Mindmaster Inc, Miami, Florida) and 11.9% hooked up 2 or more. The mean (SD) reported weekly screen time budget goal was 10.0 (2.4) hours (median, 10.0; interquartile range, 7.5-12.0).

All 24 educational newsletters were sent to valid addresses for 93.7% of girls in the active-placebo health education group and their parents/guardians, and 86.9% of the girls reported reading at least half of the “Felicia's Healthy News Flash” newsletters (66.4% read all or almost all the newsletters). Families attended a mean (SD) of 1.1 (1.4) (median, 1; interquartile range, 0-2) of 8 possible evening health education events. Additional Saturday summer health education fairs were attended by 29 of 94 families (30.9%) enrolled by the summer of the first year and 18 of 127 families (14.2%) in the second summer of the study. Also, 80.4% of parents/guardians reported reading at least half of the “Stanford GEMS Health Report” newsletters (54.2% read all or almost all the newsletters). All elements of the treatment and comparison interventions were rated highly for fun and helpfulness by girls and their parents/guardians.

Outcomes are reported in Table 2. There was no statistically significant difference between groups for change in BMI (the primary outcome measure). Among prespecified secondary outcome measures, fasting total cholesterol level, low-density lipoprotein (LDL) cholesterol level, and depressive symptoms decreased statistically significantly more among girls in the treatment group. Parents/guardians in the treatment group also reported significantly increased preferences for things African American compared with parents in the comparison group. Differences between groups were in the expected direction for most other secondary outcome measures but were not statistically significant. Incidence based on high-risk thresholds of categorical outcomes are reported in Table 3. These results suggest a significant treatment group benefit in the incidence of marked hyperinsulinemia (≥30 μIU/mL [to convert to picomoles per liter, multiply by 6.945]) (Figure 3).

We explored a priori–defined baseline (prerandomization) measures as potential moderators of treatment effects on the primary outcome measure (ie, BMI) to identify subgroups that may have responded more or less to the intervention.^{51 - 52} Parent/guardian marital status (P = .01) and baseline hours of television viewing (P = .02) were statistically significant moderators of intervention effects (Figure 4).

Systematic monitoring of all injuries and other medical problems requiring a visit to a medical professional, height growth velocity, and excessive BMI or weight loss¹⁴ suggested no increased risk associated with participation in the study as a whole or between intervention groups (P ≥ .20 for all). No injuries or illnesses were judged to be probably or definitely related to study participation.

A culturally tailored, after-school, ethnic dance program and a home/family-based intervention to reduce screen time among low-income African American preadolescent girls did not reduce BMI gain compared with health education. However, girls randomized to ethnic dance and screen time reduction significantly lowered their fasting total cholesterol and LDL cholesterol levels, depressive symptoms, and incidence of elevated fasting insulin levels compared with girls randomized to health education. Changes in most other prespecified secondary outcome measures also favored the ethnic dance and screen time reduction intervention, with small to medium effect sizes,⁵³ but were not statistically significant. Consistent with the cultural orientation of the intervention, treatment group parents/guardians significantly increased their preferences for African American culture and activities. Exploratory moderator analysis indicated that the experimental intervention reduced BMI gain more than health education among the subgroups of girls who had unmarried parents/guardians and/or watched more television at the start of the study.

The effects on total cholesterol and LDL cholesterol levels, depressive symptoms, and incidence of elevated fasting insulin level may represent reduced risks of future cardiovascular disease, diabetes mellitus, and psychosocial problems. Changes in these important clinical measures have rarely been achieved through population-based interventions for children.^{55 - 56} Total cholesterol and LDL cholesterol levels decreased a mean of about 3.5 and 3.0 mg/dL more per year, respectively (or about 7.0 and 6.0 mg/dL during the 2-year study [to convert to millimoles per liter, multiply by 0.555]), among girls randomized to ethnic dance and screen time reduction than to health education. The incidence of a fasting insulin level of 30 μIU/mL or higher was about 65% lower in the dance and screen time reduction group during the study, representing a relatively small number needed to treat of 9. We believe that these differences would be of clinical and policy-making significance when applied across the population. In addition, the results of the exploratory moderator analysis^{51 ,57} suggest that the intervention might more effectively reduce BMI gain if targeted toward particular high-risk samples. We found no evidence of increased weight concerns or body dissatisfaction, consistent with prior findings¹² that interventions designed to prevent obesity have not put girls at higher risk for disordered eating problems.

It is possible that the overall BMI results indicate a failure of this intervention approach for obesity prevention. We do not believe that is necessarily the case. Previous clinical trials demonstrating the effects of reducing screen time on BMI^{12 ,21 ,58 - 59} and favorable trends in many of the measures of screen time, physical activity, and energy intake behaviors argue against a failure of the intervention model. Instead, we believe that difficulties experienced in implementation, resulting in lower-than-projected intervention doses, represent the most likely explanation. Median attendance rates at dance classes were only 11.6%, one-fifth of the goal rate, and the mean difference between groups in change in screen time was 22 minutes per year, substantially less than prior studies^{12 ,21 ,58 - 59} ; age- and sex-standardized BMI change during the study was weakly inversely associated with dance class attendance (r = −0.04) and number of START lessons completed (r = −0.12) among girls in the treatment group. Another potential explanation is that the active-placebo health education comparison intervention produced greater effects than anticipated. The active-placebo intervention has many important practical and conceptual benefits, but if it produces behavior change it could make it more difficult to detect treatment intervention effects. It is impossible to determine whether this occurred, but the rate of increase in BMI in our health education group was substantially less than that observed in some other contemporary cohorts of African American girls, including the control cohorts in the Memphis GEMS trial⁶⁰ and the GEMS pilot studies,^{10 - 12} among others.⁶¹ We still believe that the advantages of the active-placebo comparison outweigh the potential risks in this study setting, but it is important to balance these possibilities carefully when designing comparison conditions for future studies.

Other findings also deserve attention. We observed particularly high rates of cardiovascular disease and diabetes risk factors. At study entry, when girls were only 8 through 10 years of age, nearly 1 in 5 girls in the sample had elevated levels of fasting cholesterol (≥200 mg/dL) and LDL cholesterol (≥130 mg/dL) and 4.8% had markedly elevated fasting insulin levels (≥30 μIU/mL). During the subsequent 2 years, we observed an additional new onset of elevated total cholesterol, elevated LDL cholesterol, low high-density lipoprotein cholesterol, and elevated insulin levels of about 5.2%, 5.1%, 6.7%, and 11.1%, respectively, and about a 1.2% incidence of diabetes mellitus (fasting glucose level ≥126 mg/dL [to convert to millimoles per liter, multiply by 0.0555]). Accelerometer measures indicated only about one-half hour of moderate to vigorous physical activity per day at baseline, which decreased rapidly at a mean rate of about 5 minutes per day per year during the 2 years of the study. Nearly 4 of 5 girls reported having a television in her bedroom, exceeding the national average for girls this age.⁶² Girls ate about 2 breakfasts and 3 dinners per week in front of the television and consumed about 35% of their energy intake from fat. These findings indicate a particularly high-risk population and reinforce the critical importance of additional research to find effective interventions to meet their needs.

Although we experienced several implementation barriers, it should not deter others from performing research, particularly randomized controlled trials, in this or similarly low–socioeconomic status populations and settings. By involving community members in designing the study and formulating methods to address anticipated barriers, we still successfully recruited and retained a particularly high-risk sample. Less than 7% of the sample was unavailable for follow-up, and more than 86% completed their final follow-up visit after 2 years of participation.

Few previous diet and physical activity interventions have reduced body fat and weight gain,^{5 ,63} and few studies have tested obesity prevention programs specifically designed to meet the needs of preadolescent African American girls and their families, despite their increased rates of obesity and associated morbidities. Stanford GEMS was designed with particular attention to applying theory and experience to help address the paucity of research in this area. Notwithstanding implementation difficulties, an intervention applying the social cognitive model^{12 ,14 ,18 - 19} to a community-based urban setting with low-income African American girls and their families produced potentially clinically important changes in lipid levels, insulin levels, and depressive symptoms compared with health education. Although there was no significant difference between groups in BMI (the primary outcome measure), there was evidence of effects on BMI in subsets of higher-risk girls with unmarried parents/guardians and/or high baseline television viewing. These results are promising and suggest a need for continued solution-oriented research⁶⁴ to identify approaches and methods that produce greater effects in population-based samples of children.

Correspondence: Thomas N. Robinson, MD, MPH, Division of General Pediatrics and Stanford Prevention Research Center, Stanford University School of Medicine, 1070 Arastradero Rd, Ste 300, Palo Alto, CA 94304 (tom.robinson@stanford.edu).

Accepted for Publication: April 26, 2010.

Author Contributions: Dr Robinson (principal investigator) had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. Study concept and design: Robinson, Matheson, Kraemer, Wilson, Obarzanek, and Killen. Acquisition of data: Matheson, Wilson, Thompson, Alhassan, Spencer, and Fujimoto. Analysis and interpretation of data: Robinson, Matheson, Kraemer, Obarzanek, Haydel, Fujimoto, and Varady. Drafting of the manuscript: Robinson. Critical revision of the manuscript for important intellectual content: Robinson, Matheson, Kraemer, Wilson, Obarzanek, Thompson, Alhassan, Spencer, Haydel, Fujimoto, Varady, and Killen. Statistical analysis: Kraemer, Haydel, and Varady. Obtained funding: Robinson and Killen. Administrative, technical, and material support: Obarzanek, Thompson, Spencer, Fujimoto, and Killen. Study supervision: Robinson, Matheson, Wilson, and Killen. Project coordination: Thompson.

Financial Disclosure: None reported.

Funding/Support: This research was funded by cooperative agreement UO1 HL62663 from the NHLBI, National Institutes of Health.

Role of the Sponsor: An NHLBI program officer (Dr Obarzanek) was a member of the cooperative agreement steering committee and, as a coauthor on the manuscript, participated in interpretation of the data and preparation of the manuscript. The NHLBI program officer and other NHLBI scientific staff provided input on design and conduct of the study but were not involved in collection, management, or analysis of the data. The manuscript was reviewed and approved by the NHLBI before submission.

Additional Contributions: We thank the Stanford GEMS participants and their families, our Stanford GEMS field staff, and the many community members and community centers that made this study possible. We also thank Charlotte Pratt, MS, PhD, RD, from the NHLBI, the members of the data and safety monitoring board, and the Memphis GEMS investigators for their valuable input and collaboration.