NCBI Bookshelf. A service of the National Library of Medicine, National Institutes of Health.

Whitlock EP, O'Conner EA, Williams SB, et al. Effectiveness of Primary Care Interventions for Weight Management in Children and Adolescents: An Updated, Targeted Systematic Review for the USPSTF [Internet]. Rockville (MD): Agency for Healthcare Research and Quality (US); 2010 Jan. (Evidence Syntheses, No. 76.)

Cover of Effectiveness of Primary Care Interventions for Weight Management in Children and Adolescents

Effectiveness of Primary Care Interventions for Weight Management in Children and Adolescents: An Updated, Targeted Systematic Review for the USPSTF [Internet].

Show details

Appendix ADetailed Methods

Key Questions and Analytic Framework

Using the methods of the USPSTF,70 we developed three key questions (KQ) (with six sub-key questions) and an analytic frame work (Figure 3) in conjunction with members of the USPSTF to update its 2005 recommendation on Screening for Childhood Overweight and Obesity2. These KQs were designed to evaluate the effectiveness and safety of behavioral and pharmacological treatments for overweight and/or obese children. Each KQ focused on a different area of the evidence. KQ1 evaluates the effectiveness of interventions in reducing or stabilizing weight using short-term (6–12 months since enrolling in treatment), while KQ2 focuses on the maintenance of BMI improvements through medium-term (between 1 to 5 years since enrollment and at least 12 months since treatment ended). KQ3 assesses adverse effects of behavioral and pharmacological interventions. KQ1a and KQ2a consider other beneficial outcomes arising from the interventions. KQ1b, KQ2b, KQ1c, and KQ2c consider whether specific program components and population or environmental factors can be identified for short-or longer-term effective weight management programs.

Literature Search Strategy

We searched for systematic reviews in Ovid MEDLINE®, PsycINFO, Database of Abstracts of Reviews of Effects (DARE), the Cochrane Database of Systematic Reviews (CDSR), Cochrane Central Register of Controlled Trials (CCRCT), and Education Resources Information Center (ERIC) 2004 to 2007. We selected relevant, good quality systematic reviews where available to assist in conducting our literature search. Quality criteria were based on USPSTF methods,70 supplemented by NICE methodology9(see Appendix A Table 3). A 2006 comprehensive NICE report was based on a series of systematic reviews and addressed the prevention and management of obesity in adults and children.9 Relevant portions of this report served as a basis for the primary search for the literature included in the current report. The NICE report only included orlistat and sibutramine. Therefore, we used another good-quality review of pharmacological treatments54 as the basis for our search for pharmacological treatments. We conducted update searches in Ovid MEDLINE®, PsycINFO, Database of Abstracts of Reviews of Effects, the Cochrane Database of Systematic Reviews, Cochrane Central Register of Controlled Trials, and Education Resources Information Center from 2005 (2003 for pharmacological treatments) to June 10, 2008, to identify literature that was published after the search dates of these reports (Appendix A Table 1). The literature search and reports9,54 were supplemented by hand-searching the reference lists of other good-quality reviews of childhood obesity treatment,2,7,117–119 suggestions from experts, and reviewing reference lists of included trials. We did not search for data from non-peer-reviewed sources.

Appendix A Table 3. Quality rating criteria.

Appendix A Table 3

Quality rating criteria.

Appendix A Table 1. Exact search string.

Appendix A Table 1

Exact search string.

Article Review and Data Abstraction

Two investigators independently reviewed 2786 abstracts and 369 articles. Every abstract was considered for inclusion in each key question. Discrepancies were resolved by consensus. Detailed inclusion/exclusion criteria can be found in Appendix A Table 2. Briefly, the study population included overweight or obese 2 to 18 year-olds. We excluded studies of children with idiosyncratic weight management issues due to behavioral, cognitive, or medical factors. Trials were required to be designed to promote weight loss or maintenance and report weight outcomes of at least 6 months, although we included immediate harms when these were also reported. Interventions using mazindol were excluded because it is no longer used in current practice. Trials were required to have a minimal intervention or control group and randomize at least 10 participants in each arm. Only controlled trials (RCTs and CCTs) were included for efficacy (short-term and maintenance) of behavioral and pharmacological treatments. Weight management programs reporting pre-specified adverse events resulting in death, hospitalization, or need for urgent medical or psychiatric treatment were included to assess harms (KQ3) for all treatment modalities, even if they did not report one of our specified weight outcomes or did not meet the minimum 6-month followup required for the other key questions. In addition, we abstracted all reports of harms or potential harms in included studies.

Appendix A Table 2. Study eligibility criteria.

Appendix A Table 2

Study eligibility criteria.

We limited our consideration of behavioral interventions to those published in or after 1985. We did this because the dramatic increases in overweight in children that occurred during the 1980s and 1990s and changes in environmental and social factors related to weight gain, such as types and quantities of food readily available to children (e.g., fast food purveyors in school cafeterias, vending machines with soft drinks and candy widely available in schools) and the increased availability of sedentary activities in the home (such as computers, home DVD/video players, and video games) made the generalizability of studies to the current environment questionable.

We only examined other beneficial outcomes (KQ1a & KQ2a), important components of care (KQ1b & KQ2b) and population or environmental factors (KQ1c & KQ3c) using trials that were included for KQ1 (short-term efficacy) or KQ2 (maintenance efficacy). When reported, we abstracted data on beneficial outcomes, including impact on co-morbidities.

We used a two-step process to determine which specific intervention components we examined for KQ1b and KQ2b. First, we examined prior literature and identified several factors that may affect weight outcomes in behavioral interventions. These include whether or not studies included organized physical activity sessions,73 behavioral management techniques2,10 (for dietary and physical activity), or involved parents or families in addition to the child (clarifying extent to which parental involvement is important, for what ages).10,74,119 Second, we examined the distribution of treatment elements between successful and unsuccessful treatment trials. To do this, we coded the age of the participants (C=Children only (only included children aged 12 and under); A=Adolescents only (only included those aged 10 and older); B=Both age groups (age range included both younger children and adolescents)). We coded the three main components of behavioral interventions as follows: (1) presence of organized physical activity sessions (0=did not provide organized physical activity session, 1=provided organized physical activity); (2) used of behavioral modification principles (0=no or minimal use of behavioral modification principles,1=applied behavioral modification principles in treatment); (3) family involvement (0=no parental involvement beyond consent/receiving materials; 1=parent attended 1 to 3 sessions, less intensive involvement than child; 2=parent was also a primary recipient of treatment).

One investigator abstracted data from included studies into evidence tables. A second investigator verified the evidence tables’ content. Two investigators independently quality rated all studies using established design-specific criteria (Appendix A Table 3). Discrepancies were resolved by consensus or consultation with a third investigator. Poor-quality studies were excluded. Eight trials of behavioral interventions120–127 and one of pharmacological treatment128 were excluded because they did not meet our quality criteria.

Treatment intensity was categorized by hours of contact as follows: very low intensity (less than 10 hours); low (10 to 25 hours); medium (26 to 75 hours), high (over 75 hours). Thus, at the least, a high-intensity program would amount to twice-weekly hour-long meetings for 6 months and once-weekly hour-long meetings for the next 6 months, assuming no more than 2 sessions are missed. The lowest end of the medium intensity range would involve weekly hour-long meetings for 6 months. Weight outcomes were categorized as short-term (6 to 12 months since beginning treatment) or medium-term (between 1 and 4 years after beginning treatment and at least 12 months after ending active treatment). The longest followup reported in any of the included trials was 4 years. Maintenance was evaluated where possible using multiple measurements in the same individuals at least 12 months after an active intervention ended, or by using single post-baseline-measurements in the medium term. Weight outcomes were abstracted as reported, and included many different measures: endpoint BMI, absolute change in BMI from baseline, percent change in BMI from baseline, absolute change in BMI SDS from baseline, endpoint weight, and absolute change in weight from baseline.

In addition, we evaluated whether or not a treatment was comprehensive. Interventions were considered comprehensive if they included all of the following elements: (1) counseling for weight loss or healthy diet, (2) counseling for physical activity or a physical activity program, and (3) instruction in and support for the use of behavioral management techniques to help make and sustain changes in diet and physical activity were considered comprehensive. An intervention was considered to use behavioral management techniques if any of the following elements were described: self-monitoring (having the child document diet-related behaviors or physical activity), stimulus control (modifying factors that appear to serve as cues leading to inappropriate eating, such as while watching television); eating management (techniques specifically aimed at modifying the act of eating, such as eating slowly); contingency management (contingency contracting, where rewards are given for desired eating or exercise behaviors, weight loss, or treatment adherence); cognitive-behavioral techniques (the attempt to alter maladaptive cognitions related to health behaviors, or use cognitive approaches to enhance behavior change, such as problem-solving to cope with high-risk situations).

Literature Synthesis

This review included studies of both behavioral interventions and pharmacological agents. We address each type of intervention for each of the six key questions listed in our analytic framework. We discuss each pharmacological agent as a separate intervention.

Where possible, data were synthesized using quantitative methods. For most questions, however, we relied on qualitative synthesis due to significant heterogeneity in setting, age range, intervention approach, weight outcome reported, and timing of outcome reporting among the limited number of studies available for each overall type of intervention. We modeled typical cases to more clearly articulate the magnitude of weight or weight change in pounds. In these cases, we used growth charts published by the Centers for Disease Control and Prevention (CDC)14 to estimate average height for age and to translate between percentile scores, BMI, and percent overweight (based on CDC-published 50th percentile scores for weight or BMI). We also employed on-line calculators provided at the CDC web site75,76 for calculating BMI and BMI percentiles. We used the following formula to convert BMI to pounds for an illustrative child of a given age and height: Pounds = (BMI*inches2)/703.

Studies reported a variety of weight outcomes including BMI, BMI percentile scores, BMI standard deviation or z-scores, and percent overweight. All of these measures have strengths and limitations. While BMI is reliably measured and widely used, it can be problematic when averaging BMI change over a wide age range where younger children would naturally show smaller changes. Percentile scores are helpful when describing weight change in children of many ages because they are a measure of relative overweight, rather than absolute weight. The limitation of percentile scores, however, is that there can be a large range in the highest extremes (above the 99th percentile).

To avoid the difficulties with an limited upper range of BMI percentile scores, many researchers report BMI standard deviation scores (SDS, also known as z-scores) or measures of “percent overweight.” Both of these are measures of the relative degree of overweight similar to percentile scores, but without a truncated upper limit. BMI SDS is calculated as the number of standard deviation units above or below the median, based on statistically derived curves.129 BMI SDS requires the use of published computer programs that access reference data and formulae, such as that published by the CDC130 Percent overweight is calculated by the simple formula:

100*(childs BMI/50thpercentile BMI for childs age and sex).

This method was used chiefly in earlier studies, published before computer programs were available to calculated BMI SDS. The disadvantage of using percent overweight scores is that they do not account for the known weight distribution.

Quantitative Synthesis

For the behavioral interventions, we conducted meta-analyses of short-term and maintenance outcomes separately. Most trials reported weight outcomes as post-intervention BMI or changes in BMI from baseline and compared these changes between intervention and control groups. Among trials that did not report BMI or change in BMI, three trials reported weight outcomes as changes in BMI standard deviation scores (SDS),78,80,87 and one trial reported changes in percent overweight.84 Three79,80,87 of the trials that reported BMI or related measures between groups at followup statistically tested only whether shape and slope of the curves from baseline through followup were significantly different. For one of these trials87 we used 24-month outcomes as an estimate for 12-month outcomes, which were shown graphically, but did not report means and standard deviations. The 24-month outcome is a slight underestimate of the 12-month effect, and although the 24-month effect was not statistically significant cross-sectionally, we show it as being statistically significant in Table 3 and in the text descriptions since the graphical display in the article indicated non-overlapping confidence intervals at 12-month followup

We focused on the change in BMI from baseline as the preferred measure of weight change when it was available. If BMI change was unavailable and could not be calculated, we used change in BMI SDS as our second choice, and change in percent overweight as the third choice. Because we combined different outcomes, we analyzed standardized effect sizes. We also ran a meta-analysis examining only those reporting BMI change and found that that pattern of results and magnitude of effects were very similar to those seen in the primary meta-analysis that included all trials (and allowed different measures of weight change). We report the more comprehensive results in the meta-analysis including all trials.

The number of observations included in the analysis of interest to this review (as opposed to the number randomized, or the number with complete data, for example) was used as the n in the meta-analysis. If both intention-to-treat (ITT) and completers-only analyses were reported, we selected the ITT analysis for inclusion in the meta-analysis. If a trial involved two active treatment arms, the arm with a greater number of treatment hours or that was judged to be most comprehensive was selected for the meta-analysis. If outcomes were reported at multiple time points in the short-term, we chose the one closest to 12 months post-baseline. No trials reported maintenance outcomes at more than one time point for both intervention and control groups. We used random effects models because the trials varied considerably along many dimensions that would impact both baseline BMI (e.g., age, minimum overweight inclusion criteria) and change in BMI (e.g., intensity of intervention, comprehensiveness of treatment program). All meta-analyses were conducted using RevMan 4.2.

Trials were grouped according to comprehensiveness and intensity into the following categories: (1) comprehensive, medium (26–75 hours of contact) to high (76 or more hours) intensity; (2) comprehensive, low intensity (11–25 hours); (3) comprehensive, very low intensity (fewer than 10 hours); (4) focused interventions. Interventions were considered to be comprehensive if they provided dietary counseling, physical activity counseling, and employed behavior modification principles to assist with behavior change. Trials were only statistically combined within category. All trials reporting maintenance outcomes (KQ2) fell into different categories, and were therefore not statistically combined, though the forest plot is presented to facilitate comparison with across trials.

If mean change scores from baseline for each group were not reported, we calculated an unadjusted difference between the mean baseline and mean followup scores for each group using simple subtraction. Standard deviations (SDs) of the change scores were reported in five trials with post-treatment outcomes and one trial with followup outcome. In addition, two authors who did not report them in published articles provided us with these unpublished data.82,88 We calculated standard deviations for trials that did not report them. Baseline BMI is highly correlated with post-treatment and follow-up BMI, and we had to take this correlation into account when calculating the standard deviations of the change scores. In order to estimate the degree of correlation, we examined data from a recently published trial in a school setting131 that reported both the SDs of the change scores (which we were attempting to calculate) and the SDs of the baseline and post-treatment BMIs (which we would use to calculate of the SDs of the change scores). Although this trial was excluded from the current review due to setting, it used an intervention approach and population comparable to those targeted by this review. From this trial, we ascertained that the correlation between the baseline and post-treatment BMI was approximately 0.90. Therefore, we assumed a correlation of 0.90 for the remaining trials and calculated SDs of BMI change using the following formula:


When given standard errors rather than standard deviations, we calculated standard deviations by multiplying the standard error by the square root of n. When given symmetric confidence limits rather than standard deviations, we determined the standard deviation using the following formula:

Std Dev=(CI width)(sqrt(n))2*(1.96)
Appendix A Figure 1. Search results and article flow.

Appendix A Figure 1Search results and article flow

*Includes bariatric surgery articles.

PubReader format: click here to try


  • PubReader
  • Print View
  • Cite this Page
  • PDF version of this title (2.1M)
  • Disable Glossary Links

Recent Activity

Your browsing activity is empty.

Activity recording is turned off.

Turn recording back on

See more...