Effects of a low glycemic load or a low-fat dietary intervention on body weight in obese Hispanic American children and adolescents: a randomized controlled trial^,

Abstract

Background: In Hispanic children and adolescents, the prevalence of obesity and insulin resistance is considerably greater than in non-Hispanic white children. A low–glycemic load diet (LGD) has been proposed as an effective dietary intervention for pediatric obesity, but to our knowledge, no published study has examined the effects of an LGD in obese Hispanic children.

Objective: We compared the effects of an LGD and a low-fat diet (LFD) on body composition and components of metabolic syndrome in obese Hispanic youth.

Design: Obese Hispanic children (7–15 y of age) were randomly assigned to consume an LGD or an LFD in a 2-y intervention program. Body composition and laboratory assessments were obtained at baseline and 3, 12, and 24 mo after intervention.

Results: In 113 children who were randomly assigned, 79% of both groups completed 3 mo of treatment; 58% of LGD and 55% of LFD subjects attended 24-mo follow-up. Compared with the LFD, the LGD decreased the glycemic load per kilocalories of reported food intakes in participants at 3 mo (P = 0.02). Both groups had a decreased BMI z score (P < 0.003), which was expressed as a standard z score relative to CDC age- and sex-specific norms, and improved waist circumference and systolic blood pressure (P < 0.05) at 3, 12, and 24 mo after intervention. However, there were no significant differences between groups for changes in BMI, insulin resistance, or components of metabolic syndrome (all P > 0.5).

Conclusions: We showed no evidence that an LGD and an LFD differ in efficacy for the reduction of BMI or aspects of metabolic syndrome in obese Hispanic youth. Both diets decreased the BMI z score when prescribed in the context of a culturally adapted, comprehensive weight-reduction program. This trial was registered at clinicaltrials.gov as NCT01068197.

INTRODUCTION

The prevalence of obesity [BMI ≥95th percentile for age and sex] in US Hispanic children and adolescents is considerably greater than in non-Hispanic white children 6–19 y old (1). Hispanics also have the highest prevalence of metabolic syndrome (2), which is a cluster of abnormalities that includes dysglycemia, dyslipidemia, and hypertension (3, 4) that predispose adults to type 2 diabetes (T2D)⁵ (3) and cardiovascular disease (4). Because Hispanics are the fastest growing segment of the US population (5), the increasing prevalence of overweight in this ethnic group will likely further increase the health care burden because of diabetes and its complications in the United States.

Although lifestyle modifications are known to be important in the reduction of cardiovascular disease and T2D risk factors in both adults (6) and children (7, 8), little attention has been given to the development of clinically effective and culturally competent interventions for the Hispanic community (9). With regard to dietary interventions, there has been extensive research on low–glycemic index (LGI) diets or low–glycemic load diets (LGDs), which aim to control the rise of blood glucose after food consumption (10). The glycemic index (GI) of a test food is defined as the glucose AUC from the consumption of 50 g carbohydrates from a test food relative to the consumption of 50 g carbohydrates from a standard food of either white bread or glucose (10). High–glycemic load (GL) diets are thought to increase risk of β cell exhaustion and impair glucose homeostasis by increasing glucose excursion and stimulating the release of counterregulatory hormones that augment the rise in free fatty acids after eating (11, 12).

Some (13–15) but not all (16, 17) epidemiologic studies have shown a lower risk of diabetes in individuals who consumed an LGD, but to our knowledge, there has been no research specifically on Hispanic children and adolescents, and thus, the efficacy of an LGD for this demographic segment is not known. However, an LGD might be particularly useful for weight control in obese Hispanic children for 2 reasons. First, Hispanic children tend to have high-GL diets, whereby they consume a lot of processed and refined carbohydrate foods and sweetened beverages (18, 19). Second, Hispanic children frequently manifest insulin resistance with high insulin-secretion rates (20) and have an increased prevalence of T2D (21) that might be modified by changing diet composition.

Therefore, we hypothesized that an LGD would be a more-effective dietary regimen for weight control and improvement of obesity-related comorbid conditions in obese Hispanic children than would a low-fat diet (LFD). To test this hypothesis, we compared the effects of an LGD and LFD on body composition and components of metabolic syndrome in Hispanic youth over a 2-y interval.

SUBJECTS AND METHODS

Overview of study design

From November 2003 to May 2008, we conducted a single-center, randomized clinical trial of an LGD compared with an LFD in the context of a comprehensive, culturally competent behavior-modification program in Hispanic children and adolescents. The intervention was carried out at a Children's National Medical Center (CNMC) community-based clinic, and evaluations and assessments were performed at the General Clinical Research Center (GCRC) of the CNMC (Washington, DC). The intervention consisted of 12 weekly nutrition education and dietary counseling sessions. All study participants were then followed for a total of 2 y, with monthly visits for 9 mo and 3 monthly visits for another 12 mo. Body composition and laboratory measurements to assess components of metabolic syndrome were obtained at baseline, immediately after intervention at 3 mo, and at 12 and 24 mo.

Participants

Participants were recruited through advertisements placed at community facilities such as clinics, schools, churches, and other establishments in the vicinity of the community clinic. Hispanic children aged 7–15 y with a BMI ≥95th percentile for age and sex who were otherwise healthy were eligible. Hispanic ethnicity was determined by self-identification with the Hispanic or Latino cultural group by the participant's parents and both sets of grandparents, which was established by interviewing the participant's parents. Only one child per family was eligible for enrollment as a study subject, although all family members were encouraged to participate in the program. All enrolled study subjects provided a detailed medical history and underwent a physical examination with physician assessment of pubertal development at recruitment. Exclusion criteria included any known medical condition that would interfere with study objectives or procedures, such as preexisting T2D, Cushing syndrome, untreated hypothyroidism, pervasive developmental disorder, severe asthma, untreated depression, use of medications known to promote weight gain or loss, and obesity-associated genetic syndromes.

The study was approved by the CNMC Institutional Review Board. Signed consents were obtained from parents of participating children, and assents were obtained from the children and adolescents.

Description of the intervention program

The 2 dietary intervention programs were modeled after those proposed by Ebbeling et al (22) but were adapted for the Hispanic culture. The study was conducted over a 4-y period during which a total of 6 LGD and 6 LFD groups were recruited at regular intervals, and each group was followed up for 2 y. The order in which groups occurred was determined by random assignment in blocks of 2 within strata determined by the BMI percentile, sex, and pubertal stage. Intervention sessions were purposely not held contemporaneously so as to minimize the contamination between the 2 dietary groups. Both LGD and LFD programs were offered during each season of the year, which minimized any seasonal biases. The fidelity of the intervention was promoted by developing specific workshops for each dietary group and the provision of manuals, food choice lists, and recipes specific to the dietary group. Interventionists took diligent care not to bias outcomes of any group by providing a similar intensity of the program to each dietary arm. Care was taken to keep all other components of the program the same for the 2 dietary groups including the physical activity, behavior change, and parenting components of the program. The principal investigator attended all teaching sessions to monitor program fidelity.

Two instruction manuals, one of which was specific for the LGD and one of which was specific for the LFD plan, were developed for the intervention program. The nutrition education sessions were divided into 12 modules taught over a 12-wk course. For the LGD group, participants and their parents were given instructions and specific examples to lower the GL of their diets by replacing high-GI sources of carbohydrates with LGI food sources, replacing energy from carbohydrates with energy from protein and fat, and balancing meals and snacks with LGI carbohydrates, protein, and low-fat food sources. For example, individual participants were offered LGI alternatives for typically consumed foods or favorite foods that had a high GI. The objective was to achieve macronutrient composition for the LGD of 45–50% LGI carbohydrates, 20–25% protein, and 30–35% fat each day, with an emphasis on achieving the target macronutrient distribution at each meal. For the LFD group, participants and their parents were given instructions and specific examples to limit dietary fat intake and increase the intake of grains on the basis of current low-fat dietary recommendations (23). The composition of the LFD was targeted to achieve 55–60% carbohydrates (with no discrimination by GI), 15–20% protein, and 25–30% fat.

For each dietary group, culturally appropriate dietary prescription plans for either the LGD or LFD depending on the group assignment were given to study participants and their parents. In addition to dietary manuals, participants and their parents were provided with recipe books, food-choice lists, and pantry lists to assist with their cooking and food shopping. See Figure 1, A and B, under “Supplemental data” in the online issue for illustration of the LGI and low-fat food pyramids. Cooking demonstrations and tastings were also included in the nutrition education sessions.

All subjects also participated in sessions to increase their physical activity and reduce their sedentary behaviors. The physical activity and reduction of sedentary behavior components of the program were directed at both participants and their families in individual and group sessions. These components included instructions on the role of physical activity for health and weight maintenance, activity demonstrations, participation in exercises led by a trained fitness instructor for the children, and contracting to increase physical activity and reduce total screen time.

Program delivery was through 12 weekly group sessions, which were separate for parents and children, and weekly family sessions at which the interventionist met with each child and parent individually. Groups consisted of 6–12 participants plus their parents. Postvoid weight in minimal clothing was obtained at each weekly visit by using an electronic scale (model 5002–21539; Scale-Tronix Inc). Height was measured by using a wall-mounted stadiometer (SECA 216; SECA) at baseline and the end of the intervention. Incentives were given to participants for goal achievements. The sessions were conducted in the evenings at a CNMC community clinic in the neighborhood where the families lived. All components of the intervention were accomplished by using standard tools for behavior changes that have been described by Epstein et al (24, 25). These tools included self-monitoring, social reinforcement by using praise and contracting, modification of the social environment, contingency planning, stimulus control, modeling and social skills, and the provision of feedback. In addition, parents of enrolled participants attended parenting classes. These classes were based on the establishment of parenting curricula (26–28) but targeted to dietary and activity behaviors.

Follow-up

Follow-up after the intensive weekly phase of the intervention consisted of monthly follow-ups for 9 mo and then 3-monthly follow-ups for 12 mo for a total follow-up interval of 2 y. Postvoid weights and heights were measured in the clinic at all follow-up visits as previously described.

Assessment of intervention effects

Children were admitted to the GCRC at the CNMC for their baseline and 3-, 12-, and 24-mo postintervention assessments early in the morning after an overnight fast. Weight was measured on a digital scale (Health O meter 2500 KLS; Health O meter) with the participant in underwear and a hospital gown. The electronic balance was calibrated each morning by using a known standard weight. Height was measured by using a wall-mounted stadiometer (SECA 216; SECA). Waist circumference was measured at the umbilicus by using a nonelastic tape measure to the nearest 0.1 cm. For height, weight, and waist circumferences, 3 repeated measures for each participant were obtained, and the average of these 3 measurements was used. BMI (in kg/m²) was calculated. Blood pressure measurements (Vital Signs Monitor 300 Series; Welsh Allyn) were taken after 15 min of bed rest. Total-body fat mass and fat-free mass were assessed by using air-displacement plethysmography (BodPod; Life Measurement Inc) as previously described (29). Fasting blood samples were obtained at baseline and after intervention to measure insulin, glucose, total cholesterol, LDL cholesterol, HDL cholesterol, and triglycerides. Serum insulin concentrations were measured by using a solid-phase, 2-site chemiluminescent immunometric assay (Immulite 2000 Analyzer; Diagnostic Products). Plasma glucose, which was collected in tubes that contained sodium fluoride and potassium oxalate as glycolytic inhibitors, was measured by using the hexokinase-glucose-6-phosphate dehydrogenase method (Dade Behring Inc). Serum LDL cholesterol was measured by using a homogeneous direct calorimetric method (Dade Behring Dimension RXL; Dade-Behring), and total cholesterol, HDL cholesterol, and triglycerides were measured by using an enzymatic calorimetric method on an automated analyzer (Dade Behring Dimension RXL; Dade-Behring). The insulin resistance index was determined by using the HOMA-IR and was calculated as the product of the fasting plasma glucose concentration (mmol/L) and fasting serum insulin concentration (μU/L) divided by 22.5 (30).

Dietary intake and composition were assessed with both a 24-h dietary recall and a 2-wk dietary recall by using a Block Kid Food-Frequency Questionnaire (31). The Nutritionist Pro software (version 4.2; Axxya System) was used to perform an energy and macronutrient analysis of the 24-h dietary recall. The GI of carbohydrate-containing foods was assigned on the basis of a glucose reference from published GI values (32). A weighted GI for each food item was obtained by multiplying the GI by the proportion of total carbohydrate contributed by the food item. The daily GI was calculated by summing the weighted GI values for each food item. The GL was calculated as the product of the daily GI and total carbohydrate and adjusted for energy intake (33). Dietary compliance with the assigned diet-treatment group was determined by assessing the change of meal GI, meal GL, and the percentage of energy from fat with the intervention.

Because of the nature of the dietary intervention, the study was not a double-blind randomized study. Participants were not informed of their dietary group assignment but could ascertain their group on the basis of the diets offered. The staff who obtained primary and secondary outcome measurements did not take part in the interventions and were blinded to subject group assignments.

Statistical methods

Sample-size and power analyses were carried out a priori to allow for comparison of the means of postintervention BMI between groups. Estimates were derived on the basis of a comparison of effect sizes expressed in SD units and based on the assumption that repeated measures in the same child would have a 70% correlation. The sample size was computed to detect a difference in BMI between groups of a 0.5 SD with 90% power on the basis of a 2-sided α level of 0.05. A 0.5-SD effect size is generally considered to represent a moderate difference. We calculated that 84 children (42 children per intervention group) would have 90% power to detect a 0.4-SD difference between treatment groups in the primary outcome (BMI) and time-averaged posttreatment assessments and >98% power to detect differences of ≥0.5 SD (34). To allow for a 35% dropout rate, we enrolled a total sample of 113 participants.

The effectiveness of the 2 dietary interventions was analyzed in the intention-to-treat population, which comprised all participants who were randomly assigned, irrespective of the number of intervention sessions attended or withdrawal from the study. The primary analyses used a multiple-imputation methodology (35) to estimate missing data under 2 assumptions (ie, missing at random and missing not at random) but from strata formed on the basis of predictors of BMI change including age, sex, maternal education, BMI, and waist circumference at baseline. These methods yielded similar results to the analyses of completers. Analyses were performed by using SAS software (SAS 9.1 version; SAS Institute Inc) and STATA software (release 12; StataCorp LP).

The change in BMI z score (BMI expressed as a standard z score relative to CDC age- and sex-specific norms) was the predetermined primary efficacy variable used to assess the effectiveness of treatments. The BMI z score is typically used to measure treatment outcomes in children and adolescents rather than weight change, which does not take into account normal growth (36). We assessed the difference between treatment groups by using ANCOVA models and controlled for baseline measures of the dependent variable.

Secondary efficacy variables were changes in insulin resistance and metabolic risk markers. We used the same methods as previously described to compare preintervention to postintervention changes in insulin sensitivity (by using HOMA-IR) while controlling for baseline HOMA-IR. Changes in metabolic syndrome prevalence with intervention were assessed by using metabolic syndrome criteria proposed by Cook et al (2). We compared risk of metabolic syndrome (present compared with absent) by treatment group at each of the time points of 3, 12, and 24 mo by assessing the current odds of metabolic syndrome at each time point and taking into account the status at the time just before (transition modeling). We used survival analyses to predict RR in participants in the LGD group relative to the LFD group of no longer having metabolic syndrome over time in subjects who had metabolic syndrome at baseline and of developing metabolic syndrome over time in subjects who did not have metabolic syndrome at baseline.

RESULTS

The recruitment plan and study flow are shown in Figure 1. A total of 169 Hispanic children were screened on site to identify 113 children who were randomly assigned to the LGD or the LFD (Table 1). The time lapse from random assignment to an intervention was not significantly different between the 2 dietary groups, with a mean duration of 2.26 ± 3.3 mo for the LGD group and 3.25 ± 5.8 mo for the LFD group (P = 0.26). There were no significant differences with regard to the demographic, baseline anthropometric, and clinical characteristics of the 2 study groups except that participants randomly assigned to the LFD had a significantly higher prevalence of a family history of hypertension and higher baseline serum LDL cholesterol and triglycerides. Subjects in both dietary groups had high mean fasting insulin concentrations (>16 μU/mL) and mean HOMA-IR >2.5, which indicated the presence of insulin resistance.

FIGURE 1.

Flow of participants throughout the study. GCRC, General Clinical Research Center.

TABLE 1.

Baseline characteristics of study participants by dietary group¹

	Low–glycemic load diet (n = 57)	Low-fat diet (n = 56)
Sex (percentage M)	44	59
Age (y)	11.8 ± 0.32	11.5 ± 0.3
Girls’ Tanner breast stage (%)
1	15.6	22.7
2–3	34.4	36.4
4–5	50.0	40.9
Boys’ testis volume (cc)	7.0 ± 1.4	8.8 ± 1.3
BMI (kg/m2)	31.1 ± 0.8	30.03 ± 0.6
BMI z score3	2.25 ± 0.05	2.24 ± 0.03
Body fat mass (kg)	30.8 ± 12.0	30.3 ± 9.9
Body fat (%)	42.3 ± 0.8	43.3 ± 0.7
Waist-circumference z score	1.57 ± 0.1	1.48 ± 0.07
SBP (mm Hg)	115 ± 1.2	115 ± 1.3
DBP (mm Hg)	62 ± 1.0	64 ± 1.1
Reported age of obesity onset (y)	5.7 ± 0.50	5.4 ± 0.50
Maternal age (y)	37.8 ± 1.0	40.1 ± 0.9
Maternal education (%)
Elementary plus some HS	64.9	50.0
Graduated from HS	15.8	30.4
Post-HS or college graduate	19.3	19.6
Total household income ($)	27,700 ± 2300	30,900 ± 2600
Family history of obesity (%)	71.9	69.6
Family history of T2D (%)	66.7	64.2
Family history of dyslipidemia (%)	52.8	61.8
Family history of CVD (%)	38.6	53.6
Family history of hypertension4 (%)	51.8	74.5
Fasting laboratory assessment
Insulin (μU/mL)	16.5 ± 1.7	17.2 ± 1.6
Glucose (mg/dL)	85.3 ± 7.6	82.9 ± 1.0
HOMA-IR	3.53 ± 0.4	3.53 ± 0.3
HDL cholesterol (mg/dL)	37.5 ± 1.3	39.8 ± 1.1
Triglycerides (mg/dL)4	108.1 ± 7.6	131.8 ± 7.8

¹Differences between dietary groups were assessed by using 2-factor ANOVA for continuous variables and the chi-square test for categorical variables (P < 0.05 indicated significance). CVD, cardiovascular disease; DBP, diastolic blood pressure; HS, high school; SBP, systolic blood pressure; T2D, type 2 diabetes.

²Mean ± SE by study group (all such values).

³Expressed as a standard z score relative to CDC age- and sex-specific norms.

⁴P < 0.05.

Seventy-nine percent of LGD and LFD enrollees completed the 3-mo study, 61% of enrollees completed 1 y of follow-up, and 54.9% enrollees completed 2 y of follow-up (Figure 1). Noncompleters were not significantly different from completers in their baseline clinical characteristics. There were no significant differences in completion rates between study groups (P = 0.9) or in the total number of intervention sessions attended [74.6% (95% CI: 66.9, 82.3) for the LGD group; 69.9% (95% CI: 61.9, 77.9) for the LFD group; P > 0.6]. The loss to follow-up occurred throughout the 2-y study period as shown in Figure 1, with 21.2% (24 of 113 subjects) occurring during the first 12 wk of intervention after subjects began attempts to follow the diets, 22.5% (20 of 89 subjects) after the first 12 wk and before the end of the 12-mo follow-up, and 7.2% (5 of 69 subjects) after the 12-mo follow-up and before the end of the 24-mo study period. Reasons for loss to follow-up are shown in Figure 1.

Dietary changes during the intervention

LGD participants reported significantly greater decrease in the GL per kilocalories of their meals after intervention compared with that of LFD participants at 3 mo after intervention (P = 0.018) and a trend toward a greater decrease in relative GI (Table 2; P = 0.06). Overall, at 3 mo after intervention, LGD participants reported consuming 49.6% of their calories from carbohydrates, 20.8% of their calories from protein, and 30.3% of their calories from fat, whereas the corresponding consumption for LFD was 52.6% from carbohydrates, 20.6% from protein, and 26.8% from fat. However, at the 12- and 24-mo follow-up, there were no significant differences in reported changes in GL or the percentage energy intake from macronutrients between the 2 dietary groups (Table 2).

TABLE 2.

Changes in reported dietary intake from baseline to postintervention¹

	Baseline		3 mo postintervention		12 mo postintervention		24 mo postintervention
	Low–glycemic load diet (n = 57)	Low-fat diet (n = 56)	Low–glycemic load diet (n = 39)	Low-fat diet (n = 32)	Low–glycemic load diet (n = 36)	Low-fat diet (n = 33)	Low–glycemic load diet (n = 27)	Low-fat diet (n = 24)
Total energy consumed (kcal/d)	1209 ± 74	1223 ± 64	1227 ± 78	1086 ± 113	1203 ± 92	1042 ± 71	1148 ± 75	1146 ± 135
Percentage of energy from carbohydrates	53.6 ± 1.5	51.4 ± 1.2	49.6 ± 1.7	52.6 ± 1.5	53.0 ± 1.9	54.9 ± 2.2	56.4 ± 3.1	53.5 ± 1.8
Percentage of energy from protein	18.0 ± 0.9	17.6 ± 0.7	20.8 ± 1.12	20.6 ± 0.92	17.3 ± 0.8	19.2 ± 1.1	17.1 ± 1.4	18.6 ± 1.3
Percentage of energy from fat	28.4 ± 1.2	31.1 ± 1.1	30.3 ± 1.6	26.8 ± 1.52	30.4 ± 1.6	26.4 ± 1.7	26.4 ± 2.8	28.3 ± 1.7
Relative GI3	57.7 ± 0.9	57.7 ± 0.9	51.3 ± 1.25524	55.0 ± 1.04	54.6 ± 1.4	56.5 ± 0.9	55.5 ± 1.0	54.4 ± 1.5
Glycemic load (g/1000 kcal)	78.3 ± 2.5	74.5 ± 2.2	63.8 ± 2.625	73.8 ± 2.55	73.3 ± 3.2	78.2 ± 3.3	77.2 ± 3.5	73.6 ± 3.4

¹All values are means ± SEs. P values were calculated by using 2-factor ANOVA to determine differences between dietary groups and the t test to determine within–dietary group differences (P < 0.05 indicates significance). Individual P values are available from the authors.

²Significant within–dietary group difference from baseline (P < 0.05).

³GI, glycemic index. Glucose was used as a reference.

⁴Change-from-baseline between-group difference (P < 0.06).

⁵Change-from-baseline between-group difference (P < 0.05).

Anthropometric changes

Weekly BMI and BMI z score by dietary group during the 3-mo intervention showed no significant between-group differences (P = 0.4 for BMI and P = 0.96 for BMI z score). With the baseline BMI z score controlled for, the effect of the intervention on the BMI z score at 3, 12, and 24 mo postintervention by using multiple imputation as well as completers-only analyses is shown in Table 3. With intervention, both dietary groups significantly decreased their BMI z scores at 3, 12, and 24 mo postintervention (P < 0.0001, P = 0.003, and P = 0.002, respectively; see Table 1 under “Supplemental data” in the online issue). However, there were no significant differences in the time averaged mean BMI z score (P = 0.26; Table 3) or the decrease in overall BMI z score between the 2 dietary groups (P = 0.83; see Table 1 under “Supplemental data” in the online issue) or differences in the interactions of dietary group and time (see Table 1 under “Supplemental data” in the online issue). The BMI z score adjusts for age and, therefore, values may be more comparable over time than for BMI, which normally increases with age in growing children; hence, some increase in BMI with time is expected even in children who participate in a weight-loss program (Figure 2). BMI z scores were lower postintervention at all time points compared with at baseline, which indicated that, although BMI may have increased somewhat with age, after adjustment for age and sex, the study children were at a lower point compared with at baseline.

TABLE 3.

Intervention effect on BMI z score¹

	Low–glycemic load diet	Low-fat diet	P
All participants by using multiple-imputation analyses2
Baseline BMI z score	2.25 (2.16, 2.34)	2.24 (2.17, 2.31)	0.818
3-mo postintervention time	2.12 (2.08, 2.17)	2.13 (2.09, 2.18)	0.824
12-mo postintervention time	2.10 (2.05, 2.16)	2.16 (2.10, 2.11)	0.185
24-mo postintervention time	2.10 (2.02, 2.16)	2.16 (2.09, 2.22)	0.199
Time-averaged mean estimate from 3 to 24 mo	2.11 (2.06, 2.16)	2.15 (2.10, 2.20)	0.256
Completers analyses3
Baseline BMI z score	2.25 (2.16, 2.34)	2.24 (2.17, 2.31)	0.818
3-mo postintervention time	2.11 (2.06, 2.16)	2.11 (2.06, 2.16)	0.982
12-mo postintervention time	2.07 (2.00, 2.14)	2.16 (2.09, 2.23)	0.075
24-mo postintervention time	2.05 (1.95, 2.14)	2.16 (2.05, 2.26)	0.130
Time-averaged mean estimate from 3 to 24 mo	2.08 (2.03, 2.13)	2.13 (2.08, 2.19)	0.167

¹All values are means; 95% CIs in parentheses. Three-, 12-, and 24-mo measures were adjusted for the baseline measure. The BMI z score was expressed as a standard z score relative to CDC age- and sex-specific norms. The effectiveness of the 2 dietary interventions was assessed by using an intention-to-treat analysis. Differences between groups were assessed by using ANCOVA models controlled for baseline measures of BMI z score. A multiple-imputation methodology was used to estimate missing BMI z scores on the basis of the assumption that data losses occurred at random, and thus, missing observations could be estimated from concurrent and previous data on study participants. Results were unchanged when the missing-at-random assumption was relaxed.

²Total of 452 measurements for the entire study, and 117 of 452 measurements were imputed.

³Total of 335 measurements for the entire study.

FIGURE 2.

Effect of the intervention on BMI. Means (± SDs) for imputed BMI (A) and completers BMI (B) for the entire study period for the LGD (closed circles) and LFD (open circles). Effects of group assessed by using mixed-effects multiple linear regression for the entire 2-y study period were as follows: for imputed analysis, P = 0.59; for completer analysis, P = 0.44. For imputed analysis, n = 57 in the LGD and n = 56 in the LFD. For completer analysis, n = 57 (baseline), n = 45 (3 mo), n = 36 (12 mo), and n = 33 (24 mo) in the LGD; n = 56 (baseline), n = 44 (3 mo), n = 33 (12 mo), and n = 31 (24 mo) in the LFD. LFD, low-fat dietary group; LGD, low–glycemic load dietary group.

The effect of the intervention on insulin resistance, which was analyzed by using both multiple imputation and completer analyses, is shown in Table 4 (see Table 2 under “Supplemental data” in the online issue). There were no significant changes in HOMA-IR at any of the measured time points either between or within the 2 dietary groups.

TABLE 4.

Intervention effect on insulin resistance (HOMA-IR)¹

	Low–glycemic load diet	Low-fat diet	P
All participants by using multiple-imputation analyses2
Baseline HOMA-IR	2.59 (2.11, 3.19)	2.77 (2.24, 3.41)	0.663
3-mo postintervention time	2.78 (2.33, 3.31)	3.03 (2.54, 3.62)	0.483
12-mo postintervention time	2.42 (2.00, 2.94)	3.01 (2.48, 3.66)	0.121
24-mo postintervention time	2.44 (2.04, 2.92)	3.12 (2.60, 3.75)	0.060
Time-averaged mean estimate from 3 to 24 mo	2.54 (2.19, 2.95)	3.06 (2.63, 3.560	0.091
Completers analyses3
Baseline HOMA-IR	2.59 (2.11, 3.19)	2.77 (2.24, 3.41)	0.663
3-mo postintervention time	2.88 (2.35, 3.53)	2.95 (2.40, 3.61)	0.877
12-mo postintervention time	2.51 (1.93, 3.26)	2.87 (2.18, 3.78)	0.480
24-mo postintervention time	2.53 (1.97, 3.24)	2.77 (2.15, 3.56)	0.617
Time-averaged mean estimate from 3 to 24 mo	2.63 (2.24, 3.09)	2.90 (2.47, 3.41)	0.406

¹All values are means; 95% CIs in parentheses. Three-, 12-, and 24-mo measures were adjusted for the baseline measure. The effectiveness of the 2 dietary interventions was assessed by using an intention-to-treat analysis on the basis of multiple imputation. HOMA-IR was log transformed to permit parametric analysis. Differences between groups were assessed by using ANCOVA models controlled for baseline HOMA-IR concentrations. A multiple-imputation methodology was used to estimate missing HOMA-IR values on the basis of the assumption that data losses occurred at random, and thus, missing observations could be estimated from concurrent and previous data on study participants. Results were unchanged when the missing-at-random assumption was relaxed.

²Total of 452 measurements for the entire study, and 117 of 452 measurements were imputed.

³Total of 335 measurements for the entire study.

Components of metabolic syndrome

Both dietary groups showed a decrease in the prevalence of metabolic syndrome from baseline but no differences between the LGD and LFD (Table 5). Risk of developing metabolic syndrome over time if the participant had no metabolic syndrome at baseline did not differ by dietary group (HR for LGD: 0.89; 95% CI: 0.33, 2.40; P = 0.8), and risk of no longer having metabolic syndrome if the participant had metabolic syndrome at baseline did not differ between groups (HR for LGD: 1.15; 95% CI: 0.53, 2.48; P = 0.7). We examined the interaction of dietary group, previous metabolic syndrome, and time on the development or resolution of metabolic syndrome but showed no significant differences for this 3-way interaction. In addition, there was no evidence that the OR of having metabolic syndrome was altered by the 2-way interactions of group by previous status or group by time (results not shown). When all interactions were removed, we showed no evidence of either a time or study group effect.

TABLE 5.

Prevalence of metabolic syndrome at baseline and after intervention¹

	Percentage of participants with metabolic syndrome
	Baseline	3 mo	12 mo	24 mo
All participants2
Low-GL3 group	40.4 ± 6.5 [57]	22.2 ± 6.2 [45]	27.8 ± 7.5 [36]	30.3 ± 8.0 [33]
Low-fat group	32.1 ± 6.2 [56]	29.6 ± 6.9 [44]	27.3 ± 7.8 [33]	25.8 ± 7.9 [31]
Participants without metabolic syndrome at baseline4
Low-GL group	0.0 [34]	12.0 ± 6.5 [25]	15.0 ± 8.0 [20]	11.1 ± 7.4 [18]
Low-fat group	0.0 [38]	17.2 ± 7.0 [29]	19.1 ± 8.6 [21]	9.5 ± 6.4 [21]
Participants with metabolic syndrome at baseline5
Low-GL group	100.0 [23]	35.0 ± 10.7 [20]	43.8 ± 12.4 [16]	53.3 ± 12.9 [15]
Low-fat group	100.0 [18]	53.3 ± 12.9 [15]	41.7 ± 14.2 [12]	60.0 ± 15.5 [10]

¹All values are frequency percentages ± SDs; n in brackets. Metabolic syndrome was defined by using the criteria of Cook et al (2) modified to use nationally representative waist-circumference percentiles (37) and the currently recommended cutoffs for dysglycemia (38) and hypertension (39). Metabolic syndrome was diagnosed when ≥3 of the following conditions were present: 1) hypertriglyceridemia (triglyceride concentration ≥110 mg/dL); 2) low HDL cholesterol (HDL cholesterol ≤40 mg/dL); 3) abdominal obesity (waist circumference ≥90th percentile for age and sex); 4) hyperglycemia defined as impaired fasting glucose concentration (≥100 mg/dL); and 5) hypertension (systolic or diastolic blood pressure ≥90th percentile adjusted for height, age, and sex).

²RR of developing metabolic syndrome by treatment group was compared at 3, 12, and 24 mo by taking into account the status at the time just before (transition modeling). There were no significant differences by dietary group.

³GL, glycemic load.

⁴RR of developing metabolic syndrome in the low-GL group relative to the low-fat group over time in participants who did not have metabolic syndrome at baseline was determined by using survival analysis. There were no significant differences between dietary groups.

⁵RR in participants in the low-GL group relative to the low-fat group of no longer having metabolic syndrome over time in subjects who had metabolic syndrome at baseline was determined by using survival analysis. There were no significant differences between dietary groups.

The analysis of both dietary groups pooled together showed significant effects of the intervention program on the BMI z score, waist circumference, and systolic blood pressure (see Table 3 under “Supplemental data” in the online issue). There were no significant differences in primary or secondary outcomes by sex in the pooled or group analysis.

Adverse events

There were no serious adverse events reported for either treatment group. One participant in the LFD group experienced a feeling of faintness during the blood draw at the 3-mo postintervention assessment.

DISCUSSION

Both LGD and LFD were associated with significant reductions in the BMI z score as well as improvements in some aspects of metabolic syndrome. Participants in both dietary groups showed reductions in waist circumference and blood pressure with intervention. Both the LFD and the LGD were well-tolerated, with no reported serious adverse effects. Although significant differences in GLs between the 2 dietary groups were observed immediately after intervention, they did not persist beyond the 3-mo period. Consistent with some previous pediatric studies (40), we did not find differences in changes in fasting glucose or insulin between the LGD and LFD intervention groups. The magnitude of the pooled change in BMI and BMI z score for all participants was similar in comparison with other pediatric interventions (41). A systematic review and meta-analysis of 6 pediatric interventions that included LFD, low-carbohydrate diet, and LGD showed the pooled effect across all diets of −0.22 (95% CI: −0.56, 0.11), which represented a BMI loss of 2.4 over a 6-mo period (41).

A number of other pediatric studies have also shown no differences according to dietary intervention (40, 42–44). Demol et al (42) reported significant reductions in BMI and body fat percentage in 55 adolescents but no differences between groups assigned to low-carbohydrate or high-fat diets. Kirk et al (44) randomly assigned 100 obese 7–12-y-olds to low-carbohydrate, reduced-GL, or standard portion-controlled diets and showed that all 3 dietary interventions were equally effective. In a review study, Gibson et al (43) concluded that LGI and low-carbohydrate dietary interventions were no more effective than conventional energy-restricted LFDs.

However, other pediatric studies have reported contrary evidence that suggested that low-glycemic diets might result in greater reductions in BMI (22, 45, 46). In a randomized, controlled, proof-of-principle pilot study of 16 obese adolescents, Ebbeling et al (22) showed that GL was a strong predictor for a change in body fat, with a significant decrease in BMI and body fat in the LGD group but no significant change in the conventional LFD group. Spieth et al (45) retrospectively analyzed results from 107 children who attended an outpatient clinic and showed a greater decrease in BMI in children prescribed an LGI diet than in children prescribed an LFD (45). However, this study was limited because of its retrospective analysis of nonrandomly assigned participants. A third study carried out in 8 European centers over a 6-mo period evaluated 4 dietary interventions. The study showed that neither GI nor protein alone had a significant effect on body composition, but a combination of high protein and LGI was protective against overweight and obesity (46).

One possible explanation of the lack of greater improvement in the LGD group in our study might have been insufficient program fidelity if, for example, there were insufficient differences in the GL of consumed diets. However, the data did not support such a conclusion, particularly immediately after the intensive weekly intervention phase when the maximal effect was expected. Both LGD and LFD groups reported the consumption of foods with the desired macronutrient composition. Participants may not have accurately reported their actual macronutrient consumption, although, because BMI was reduced, it seems most likely that participants followed their prescribed regimens. Because, in general, without caloric restriction, weight loss is not often observed, the ingestion of fewer calories likely explains the observed BMI z scores and waist-circumference reductions. The adherence to the LGD and LFD waned over time, despite regular, although less-intense, follow-up and reinforcement of the dietary prescription. However, in itself, this outcome argued against reliance on any particular regimen. Weight management in children is challenging, and outside of a sustained, intense intervention, either LGD or LFD prescriptions may be difficult to maintain.

Inconsistent findings in clinical trials of dietary interventions may be explained by methodologic problems or biases, such as differences in program delivery or treatment intensity. However, the program components in the current study were carefully matched for their intensity, and the methods used for program delivery were carefully monitored. Inconsistencies between trials could also be due to phenotypic differences in study participants. In a study of overweight young adults, Ebbeling et al (40) showed that subjects with elevated insulin secretion had a greater weight loss with a low-GL diet and inferred that the variability in dietary weight-loss trials might be partially explained by hormonal profiles (40). The diet-phenotype interaction has also been suggested by Cornier et al (47) as an explanation for greater success with LFDs in insulin-sensitive subjects, whereas subjects who were insulin-resistant had more success with high-fat diets (47). The current study showed no variation in response by baseline insulin concentrations, although baseline insulin concentrations may not be predictive of this interaction (48).

The strengths of this study included the randomized design, use of culturally appropriate materials, preintervention and postintervention assessments performed under the same conditions for both dietary arms, and maintenance of high intervention fidelity during the first 3 mo of the intervention. The adherence to a dietary group was assessed by using dietary recall, which could have potentially introduced bias because participants may have underreported foods that they were counseled to limit and over-reported foods that they were counseled to consume. To overcome this potential bias, dietary recalls were performed by GCRC dietitians who had not been involved in delivering the dietary intervention at the community clinic. There was always the possibility of contamination of dietary arms by the interventionist or families. To minimize contamination, we did not run intervention groups concurrently and did not recruit relatives or friends who resided in the same household. The assignment of GI values of the dietary recall food items was done by using tabulated GI values (32). Some of these GI values were derived from studies done in countries where the GI of foods may have differed from those consumed in North America. To minimize the potential error in GI and GL estimates, we relied, when possible, on values derived from studies done in the United States or Canada. To mitigate a potential bias in the GI and GL calculations, the assignment of GI values and calculation of the GL of meals was done by a dietitian blinded to dietary group assignments. The fairly high attrition rate in the current study was also a limitation. However, the 3-, 12-, and 24-mo attrition rates of 21%, 39%, and 43%, respectively, were similar to those reported in other community pediatric obesity–intervention programs (42, 44, 49, 50) and in a recently published, long-term, ethnically diverse study (51). The loss to follow-up was higher than projected; however, even if the sample size was reduced to 27 subjects per group by attrition, it still retained 80% power to detect a moderate (0.5-SD) effect-size difference between groups. In most trials, 80% power is considered adequate. Therefore, by this standard, losses to follow-up alone were not sufficient to explain the lack of difference in BMI between dietary groups. Although both arms of the intervention received identical physical activity interventions and all participating children had to be accompanied by a family member at each intervention session, we do not have data regarding any changes in physical activity that may have taken place in the trial or any information on family participation in the program at home. Finally, because we had no control group in which no intervention was given, our study design did not allow us to measure the intervention effect.

In conclusion, our study shows that a comprehensive intervention program by using either an LGD or an LFD targeted at obese Hispanic children and adolescents can reduce the BMI z score, waist circumference, and blood pressure over a 2-y interval. This study indicates that it is feasible and practical to induce short-term dietary changes in Hispanic youth by using a family-based treatment approach. No significant advantage of the LGD over the LFD or vice versa was shown with regard to any outcome, although additional longer-term studies are needed to examine the sustainability of weight and cardiovascular changes in obese Hispanic youth treated with an LGD or an LFD.