Reducing Childhood Obesity through U.S. Federal Policy

Partnership for Prevention (Kristensen, Jenson), Washington, District of Columbia; HealthPartners Institute for Research and Education (Flottemesch, Maciosek), Minneapolis, Minnesota; Aetna Foundation Inc. (Barclay), Hartford, Connecticut; ChangeLab Solutions (Ashe), Oakland, California; Los Angeles County Department of Public Health (Teutsch, retired), Los Angeles, California; American Heart Association (Sanchez), Dallas, Texas; Community and Family Medicine and Global Health (Story), Duke University, Durham, North Carolina; and Brown School and Division of Public Health Sciences (Brownson), Washington University in St. Louis, St. Louis, Missouri

Address correspondence to: Alyson Kristensen, MPH, Partnership for Prevention, 1015 18th St. NW, Ste 300, Washington DC 20036. gro.tneverp@nesnetsirka.

Abstract

Background

Childhood obesity prevalence remains high in the U.S., especially among racial/ethnic minorities and low-income populations. Federal policy is important in improving public health given its broad reach. Information is needed about federal policies that could reduce childhood obesity rates and by how much.

Purpose

To estimate the impact of three federal policies on childhood obesity prevalence in 2032, after 20 years of implementation.

Methods

Criteria were used to select the three following policies to reduce childhood obesity from 26 recommended policies: afterschool physical activity programs, a $0.01/ounce sugar-sweetened beverage (SSB) excise tax, and a ban on child-directed fast food TV advertising. For each policy, the literature was reviewed from January 2000 through July 2012 to find evidence of effectiveness and create average effect sizes. In 2012, a Markov microsimulation model estimated each policy’s impact on diet or physical activity, and then BMI, in a simulated school-aged population in 2032.

Results

The microsimulation predicted that afterschool physical activity programs would reduce obesity the most among children aged 6–12 years (1.8 percentage points) and the advertising ban would reduce obesity the least (0.9 percentage points). The SSB excise tax would reduce obesity the most among adolescents aged 13–18 years (2.4 percentage points). All three policies would reduce obesity more among blacks and Hispanics than whites, with the SSB excise tax reducing obesity disparities the most.

Conclusions

All three policies would reduce childhood obesity prevalence by 2032. However, a national $0.01/ounce SSB excise tax is the best option.

Introduction

Although recent data suggest that childhood obesity has plateaued or begun to decline, prevalence remains high. 1,2 In 2009–2010, nearly one in three U.S. youth aged 2–19 years were overweight or obese and 17% were obese. 3 Significant disparities in obesity prevalence persist among racial/ethnic groups and by SES. More Hispanic (21.2%) and non-Hispanic black (24.3%) youth were obese in 2009–2010 than non-Hispanic white youth (14.0%). 3 Obesity is also higher among lower-income children than higher-income children. 4 Further, obese adolescents tend to remain obese as adults, 5,6 making childhood the ideal time to prevent obesity. For these reasons, policymakers are interested in effective programs and policies to reduce childhood obesity.

States and localities are increasingly using laws, regulations, and other policy tools to promote healthy eating and physical activity (PA). 7 However, federal policies can reach larger populations and fund programs that benefit populations at risk for obesity, and thus play an essential role in improving public health. Information is needed about which federal policies could reduce childhood obesity rates and by how much. The purpose of this study is to estimate the impact of three federal policies on childhood obesity prevalence in 2032, after 20 years of implementation.

Methods

The methods used in this analysis are summarized below; see the Appendix for more details. Microsimulation models are useful in informing health policy decision making. 8,9 In 2012, a microsimulation model (developed in TreeAge, TreeAge Software, Inc., Williamstown MA) examined how three federal policies affect obesity-related behaviors (PA and diet), BMI, and obesity prevalence in a simulated school-aged U.S. population. The model generated annual values for these measures based on demographic and behavioral variables, and then aggregated individual estimates to create population-level results. The initial population was drawn randomly from a sample of simulated school-aged children (6–12 years) and adolescents (13–18 years) with demographic characteristics matching that of the U.S., using 2010 U.S. Census data.

The model’s primary outcome variables were BMI and changes in the percentage of overweight or obese youth. Obesity and overweight were determined by comparing BMI values from the model to BMI values from CDC growth charts. The current CDC definitions of obesity (BMI at or above the 95th percentile for age and sex) and overweight (BMI at or above the 85th percentile and below the 95th percentile for age and sex) were used. 10

The microsimulation model estimated yearly changes in PA, diet, and BMI using multivariable equations developed using 2001–2010 continuous National Health and Nutrition Examination Survey (NHANES) data. The equations included measures of PA expressed in METs and dietary recall measures, including total daily calories and grams of fat, carbohydrates, and sugar. NHANES data were used to assign initial health and BMI measurements to the simulated population, and then to estimate the impact of changes in health behaviors on changes in BMI over time. Each simulated agent was initialized using a two-step process. First, age, sex, and ethnicity were assigned with the distribution of these demographic variables across the simulated population set equal to those in the 2010 U.S. Census. Second, each agent’s initial BMI, level of PA, and diet were set conditioned on that agent’s demographics. The distribution of each factor across the simulated population was set equal to the distribution observed for that agent’s corresponding demographic group in the NHANES sample, scaled to the U.S. child and adolescent population.

The model represents changes in BMI for a simulated population over time, but the NHANES provides a series of cross-sectional estimates. To account for this difference, annual changes in BMI were estimated based on age, sex, and ethnicity trends. The relative BMI remains constant until a policy intervention causes it to trend downward to a new level consistent with expected changes in behavior from the intervention. Each policy was introduced in the model independently.

Next, a systematic process was used to search and review the literature on 26 recommended policies for preventing childhood obesity. Senior authors (RB, MA, ES, MS, and ST) narrowed the list to three policies in a two-step process using multiple criteria, including effectiveness, potential reach into the general population and high-risk groups, feasibility, acceptability, precision of information for modeling, and potential impact on childhood obesity. Effectiveness was rated as “unknown,” “emerging,” “promising,” “effective (second tier),” or “effective (first tier)”. 11 The other criteria were rated low, medium/moderate, or high. The policies would: (1) strengthen and expand federally funded afterschool programs to promote PA; (2) enact a $0.01/ounce excise tax on sugar-sweetened beverages (SSBs); and (3) ban fast food TV advertising targeting children aged 12 years and under. These policies target key obesity-related behaviors through the federal policy mechanisms of appropriation, taxation, and regulation. Table 1 summarizes the policies.

Table 1

Summary of U.S. federal policies to reduce childhood obesity

For children (ages 6-12), whose
parents purchase most SSBs, the
tax results in a 25% average
decrease in SSB consumption.

For children (ages 6-12), whose
parents purchase most fast food,
the advertising ban results in 2-3
fewer fast food meals consumed
per week.

Estimating Effect Sizes

PubMed and journal article references were searched from January 2000 through July 2012 to find effectiveness data for the three policies and create average effect sizes. A systematic strategy and policy definition limited the scope of the literature search and identified key data elements for the population groups targeted by each policy. The literature search and abstraction process followed methods previously described. 12 Owing to the varied nature of evidence, the general process to determine each policy’s average effect size was modified as needed. Table 2 lists model inputs. Effects of interventions enter the model through increases in PA or reductions in calories, both of which are determinants of BMI z-scores.

Table 2

Microsimulation model parameters

a Listed value is the average across the entire simulated population. Unique values for each simulated individual are drawn from age-, gender-, and ethnicity-specific distributions fit to the 2001-2010 continuous NHANES data.

c Expressed in metabolic equivalents (METs) derived from CDC tables and assuming a sedentary floor of 1200 (8 sleeping hours and 16 inactive hours).

d Abstracted beverages were: Coke, Sprite, Snapple, Juicy Juice, Hi-C, Capri Sun, Country Time Lemonade, Powerade, Gatorade, and Sunny D.

e Abstracted restaurants were: Arby’s, Burger King, McDonald’s, Wendy’s, Taco Bell, Dairy Queen, Subway, Hardee’s, Carl's Jr., In-N-Out Burger, Jack in the Box, White Castle, Krystal, Popeyes, Chick-fil-A, KFC, Chipotle, Sonic, Domino’s Pizza, Pizza Hut, Little Caesars Pizza, Papa John's Pizza, and CiCi's Pizza.

f The average for the entire modeled population where almost 2/3 (64.6%- Table 3 ) consumed fewer than 2 fast food meals/week.

g Those averaging fewer fast food meals per week than the modeled effect size experienced a complete elimination of fast food consumption.

Afterschool Physical Activity Programs

The literature search generated 55 potentially relevant articles. Inclusion and exclusion criteria identified 16 articles suitable for abstraction. To be abstracted, studies needed to evaluate an afterschool intervention or program, report measures that could be used in the model, have a sample size of at least 50, describe an intervention that resembled a typical afterschool program, use randomization or a pre-test/post-test design, require participants to engage in PA, and be conducted in an Organisation for Economic Co-operation and Development country. One article was eliminated during abstraction because the intervention was unclear. Effect sizes for the 15 articles 13–27 were summarized by sample size, sex, race, location, age (school grade), baseline BMI (normal or obese), program adherence, intervention design, and intervention intensity. Because most studies did not provide sample demographics, effect sizes were averaged across all racial/ethnic groups. The policy was modeled such that all programs were offered to all youth for the entire year. The modeled intervention combined individual interventions using sample sizes and demographic breakdowns as weighting factors.

Sugar-Sweetened Beverage Excise Tax

No well-controlled trials were found that directly assessed an excise tax’s impact on youth SSB consumption or the relationship between SSB consumption and childhood obesity. Instead, the literature search yielded well-controlled, econometric studies using observational data that showed a negative association between increased taxes on SSBs and SSB purchases. Few studies quantified the impact of an SSB tax or price increase on childhood obesity prevalence and those that did were based on econometric forecasts. However, two key associations were found in the literature 28–33 : (1) between SSB consumption and per-ounce excise taxes; and (2) between BMI and existing state-level soda taxes. From these studies, it was estimated that SSB consumption decreased by 25% among children and 35% among adolescents due to reduced purchasing resulting from the excise tax. The policy’s impact on obesity resulted from reduced consumption of sugar and total calories, assuming complete substitution with a zero-calorie beverage. A composite SSB was created based on the nutritional content of ten calorically-sweetened beverages.

Ban on Fast Food Television Advertising Targeting Children

No studies were found that directly evaluated the impact of restricting advertising on youth consumption of high-calorie, low-nutrient foods/beverages. The literature search also did not identify any observational or experimental studies in the U.S. that examined the same issue. Instead, it yielded observational studies that explored the association between exposure to advertising and consumption of certain types of food, food purchases, or BMI. Direct evidence appears limited to observational data assessing the impact of Quebec’s ban on child-directed advertising on fast food purchases. 34 Therefore, this analysis limits the impact of an advertising ban to fast food and uses age 12 years because research shows children under this age do not understand advertising’s persuasive intent and cannot cognitively defend themselves from its effects. 35,36

Two estimates of changes in purchasing following the advertising ban were derived and applied to modeled individuals consuming two or more fast food meals per week. The ban’s impact was assumed to be two to three fewer meals per week for children, whose parents purchase most fast food, and three to five fewer meals per week for adolescents, who can purchase fast food themselves. Although the policy targets children, the analysis included adolescents to show that the effect continues into adolescence. The policy’s impact on obesity was modeled through dietary change (reduced sugar, carbohydrates, fat, and total calories). A composite fast food meal was created by averaging adult or children’s portions of main dishes, sides, and beverages from 23 large fast food chain restaurants. Adult portions were included because adolescents eat adult meals. These restaurant-specific meals were then averaged to create the composite meal. A meal meeting the 2012 National School Lunch Program standards was assumed to be substituted for each fast food meal not eaten, which would result in an average of 504 fewer calories and 94 fewer grams of sugar consumed per week.

Results

Table 3 summarizes baseline obesity rates and behaviors for the simulated school-aged population. Consistent with the underlying data, nearly 20% are obese and considerable racial and ethnic disparities exist. Just over one quarter of children get the recommended 1 hour of daily PA, whereas only one in five adolescents do. Approximately one third of the simulated youth consume SSBs at least twice per day (32.3%) and fast food at least twice a week (35.4%).

Table 3

Baseline weight and behavior distributions of simulated population

Table 4 summarizes each policy’s predicted impact on PA or diet in 2032. Afterschool PA programs would increase the number of children and adolescents who met the daily PA recommendation by 7.7% and 7.4%, respectively. This change would represent an additional 25 minutes of moderate-to-vigorous PA per day. A $0.01/ounce SSB excise tax would reduce the number of children and adolescents consuming two or more SSBs per day by 11.4% and 16.6%, respectively, which translates into an average daily reduction of 1.5 beverages in children and 2.2 beverages in adolescents. Of the three policies, the microsimulation predicted that the ban on child-directed fast food TV advertising would impact behaviors the most. The number of children eating two or more fast food meals per week would drop by almost 20%, and for adolescents by 18%. The policy would reduce consumption in children more than adolescents because children consumed more fast food at baseline and thus would have a greater potential for change.

Table 4

Change in childhood obesity-related behaviors in 2032

Table 5 summarizes the policies’ potential impact on rates of overweight and obesity in 2032. Obesity would decline, but overweight would increase slightly as obese individuals lose weight. The afterschool policy would reduce obesity prevalence in children by 1.8 percentage points and by 1.9 points in adolescents. The microsimulation predicted that disparities would decrease for black and Hispanic adolescents, with obesity decreasing by 2.3 and 2.4 percentage points, respectively, compared to 1.6 for whites. The change would be similar for children. The predicted impact of afterschool PA programs on disparities reflects higher baseline rates of obesity among blacks and Hispanics and the non-linear relationship between behavior change and BMI. For children, afterschool PA programs would have the largest impact on obesity of the three policies.

Table 5

Change in childhood weight distributions in 2032

For adolescents, the $0.01/ounce SSB excise tax had the largest predicted overall impact on obesity, resulting in a 2.4 percentage point decrease. Obesity would decrease by 1.6 percentage points for children. The tax would also reduce disparities, especially in adolescents. Obesity in blacks and Hispanics would drop by 3.0 and 2.9 percentage points, respectively, compared to 2.0 percentage points in whites. The greater predicted impact for black and Hispanic adolescents was despite whites’ higher baseline SSB consumption. This was due to two factors: compared to blacks and Hispanics, whites had higher income levels and are therefore less affected by a price increase, and had a lower baseline rate of obesity.

The child-directed ban on fast food TV advertising has the greatest predicted behavioral impact, but would reduce obesity prevalence the least. This is due to the substitution effect and the policy’s narrow focus on fast food. Although its predicted impact on obesity is small, the large behavioral result shows that TV advertising affects what children eat. Like the other policies, its estimated impact is greater for blacks and Hispanics who watch more TV and are more heavily targeted by food marketers than whites, and thus are more impacted by reduced advertising. 37 Although this policy targets children aged 12 years and younger, it would also reduce obesity in adolescents because their baseline consumption of fast food is decreased because of less exposure to advertising as children.

The Appendix describes the key univariate results of a sensitivity analysis.

Discussion

This microsimulation analysis suggests that long-term implementation of three federal policies could reduce childhood obesity in the U.S. To our knowledge, this study provides the first quantitative estimate of the potential impact of afterschool PA programs on U.S. childhood obesity prevalence. The use of microsimulation contributes to the childhood obesity literature because behavior change can be modeled over time in the simulated population. This approach differs from models that derive estimates from population-level trends, and provides valuable information as to how policies may impact known disparities in health behaviors and obesity in different populations.

For the SSB and fast food advertising policies, this study’s effect sizes are consistent with, but smaller than, prior work owing to the use of microsimulation, narrower policy definitions, and different assumptions. In this study, a $0.01/ounce excise tax on SSBs would reduce obesity by 1.6 percentage points among 6–12-year-olds and 2.4 percentage points among 13–18-year-olds in 2032. Smith et al. 28 estimated that a 20% price increase (roughly equivalent to a $0.01/ounce excise tax) on all SSBs could reduce childhood obesity prevalence by 2.9 percentage points. Sturm and colleagues 29 estimated that an 18% differential soda tax (i.e., the difference between the regular state sales tax and the higher soda tax) would correspond to a 20% reduction of the excess BMI gain seen between the third and fifth grades if the tax impact is linear.

According to the present study, a ban on child-directed fast food TV advertising would reduce obesity among children and adolescents by nearly 1 percentage point in 2032. Chou et al. 38 estimated that banning all TV fast food advertising would reduce the number of obese children aged 3–11 years by 18%. The model of Veerman and colleagues 39 predicts that a ban on all TV food advertising would reduce obesity prevalence among U.S. children aged 6–12 years by 2.5 percentage points. The bans in both studies are much broader than in this study.

All three policies could reduce childhood obesity prevalence, particularly among blacks and Hispanics, who have higher rates of obesity than whites, thus demonstrating that federal policy could alter the childhood obesity epidemic. Although the microsimulation predicts that each policy would reduce obesity in children and adolescents, the $0.01/ounce SSB excise tax has characteristics that make it the best option. It reduces obesity while generating significant revenue for additional obesity prevention activities. Andreyeva et al. 40 estimated that a national $0.01/ounce SSB excise tax would have generated $13.25 billion in 2010. It would also reduce obesity among adults who consume SSBs, does not require substantial federal funding to implement (unlike the afterschool policy), and would not face the legal hurdles that new regulations often encounter. Unfortunately, reduced federal spending, industry lobbying, a contentious political environment, and legal protection for commercial speech hinder near-term implementation of any of these policies. However, over the long timeframe included in this analysis, the infeasible may become feasible as the evidence base for these policies grows and changes in public knowledge increase calls for stronger governmental action. Research showing the harms of consuming SSBs coupled with the need for new revenue sources may spur Congress to consider a national SSB excise tax. The courts may recognize that young children need protection against the damaging influence of junk food advertising, as was done previously for tobacco and alcohol advertising, or the federal Interagency Working Group’s voluntary marketing guidelines could be implemented. In the meantime, the findings support state- and local-level action to enact SSB excise taxes, promote PA in afterschool settings, and reduce marketing and advertising of unhealthy foods and beverages in public schools.

Limitations

This study has several limitations. Modeling childhood obesity is challenging and others believe attempts should stop at energy balance owing to insufficient data on the association between changes in behaviors and changes in BMI z-scores. 41 We agree that the challenges are significant, but attempts to examine policy impact on childhood obesity have relevance. These results are only as accurate as the method used for translating short-term study results into multiple-year effects and the reliability of cross-sectional data in determining how changes in PA and diet impact BMI z-scores. Although strong survey surveillance systems allow robust estimation of baseline trends, there are little effectiveness data for the SSB and advertising policies, particularly in children, and existing data often come from observational studies. To broaden the evidence base, international studies were included in this analysis, which may limit applicability in the U.S. In addition, the estimated policy impact is sensitive to the model assumptions. For instance, substituting a caloric beverage, rather than a zero-calorie beverage, can reduce the SSB policy’s estimated impact by over 60%. In the absence of data on substitution effects in food consumption resulting from an advertising ban, it was assumed that a lower-calorie meal would be available and consumed instead. Another limitation is the inability to assess interaction effects among the three policies or with existing policies, such as state-level physical education policies. In this analysis, policies were assessed independently, but to reverse the childhood obesity epidemic, a comprehensive set of national policies would need to be implemented.

Conclusions

The three federal policies in this analysis could each reduce childhood obesity prevalence by 2032. However, a national $0.01/ounce SSB excise tax is the best option given its ability to generate revenue for additional obesity prevention activities and reduce obesity among SSB-consuming adults.

Acknowledgments

The authors thank Jud Richland, MPH, for conceptualizing this work and developing the proposal; Ana Lindsay, DrPH, MPH, for contributing to policy selection; Dana McGree for providing administrative and project management support; and Sheena De Freitas, MPH and Amanda Asgeirsson, MPH for their research assistance.

The work of each author was supported entirely by the Aetna Foundation.

Appendix

This appendix provides detailed information about the methods used in this analysis.

I. Modeling Approach and Structure

Overview

This analysis used a microsimulation model originally developed to estimate the health impact of physical activity interventions recommended by the Community Preventive Services Task Force. In this study, the Obesity-Related Behavior Microsimulation Model (ORB) evaluated policies to reduce or prevent obesity in children and adolescents. The model is designed to examine how the policies affect obesity-related behaviors (physical activity and diet), and in turn, how changes in these behaviors affect BMI and obesity prevalence.

The model was developed with TreeAge 2012 to track the state (age, diet, physical activity, body mass index (BMI), and health status) of simulated individuals over time. TreeAge Pro 2013 R2. was used to create the estimates. Within the model, each agent is uniquely defined by a set of heterogeneous characteristics. All simulated agents are independent, in that the actions of one individual do not impact those of another. The distribution of demographic characteristic across individuals is determined by data abstracted from the literature and the population being modeled.

Demographic characteristics (sex, race and ethnicity) were assigned proportional to those of the U.S. school-age population using 2010 Census data. These individuals are then “aged” in yearly increments. Appendix Figure 1 summarizes the flow of the microsimulation.

Appendix Figure 1

Flow of the Microsimulation Model

There are two discreet processes within the model: Initiation and Progression. The following section describes how agents are initialized and then aged.

Agent Initiation

At initiation, or baseline, demographics and health behaviors are set in a sequential manner as illustrated in Appendix Figure 1 . This begins with the assumed time-invariant demographics of age, sex, and race/ethnicity. Appendix Table 1 lists the agent-level parameters tracked within the model, their conditioning factors, assumed distribution across the population, and the timing of their initialization in the model.

Initialization of time-invariant factors begins with the basic demographics of age, sex, and race/ethnicity. For each agent, these are determined by a random draw from a multinomial distribution set proportional to the population being modeled. The four racial/ethnic groups included were White, Black, Hispanic, and Other. (Note: Other was omitted from the results because it is impossible to draw policy conclusions when race is unknown).

Time varying agent-level behavioral factors are initialized by random draws from joint probability distributions. The time-varying factors that are initialized and tracked are: body mass index (BMI), physical activity level (expressed in metabolic equivalents of task or METs), and diet (total kilocalories and grams of protein, carbohydrates, dietary fiber, fat, and sugar). This begins with initialization of BMI whose conditioning factors are sex, ethnicity, and initial age. The other behavioral factors (physical activity and dietary factors), which are the target of the modeled policy interventions, are further conditioned upon initial BMI. Behavioral factors were conditioned upon BMI in order to sharpen focus upon each policy’s impact upon rates of childhood obesity.

Model Progression

Youth obesity is a relative concept. Considerable changes in BMI are anticipated as a normal part of growth and development. A BMI of 20 is considered high for a 6-year-old, but normal for a 16-year-old. Thus, obesity and obesity risk are determined by comparing a youth’s BMI to standardized BMI growth charts using standardized percentiles. The Centers for Disease Control and Prevention define childhood obesity as a BMI at or above the 95th percentile for age and sex, and overweight as a BMI that is at or above the 85th percentile and below the 95th percentile for age and sex.

The microsimulation reflects this growth pattern by modeling the change in BMI over time in terms of percentiles of the standardized BMI distribution conditioned upon agent age, sex, ethnicity, physical activity, and diet. As noted in Appendix Table 1 , initial BMI is determined by a random draw from a distribution fit to continuous National Health and Nutrition Examination Survey (NHANES) data from 2001-2010 and conditioned on age, sex, and ethnicity. Similarly, the time-variant behaviors of physical activity and diet are determined from a random draw from a distribution conditioned on age, sex, ethnicity, and BMI. Each cycle, BMI, or more precisely BMI percentile, is adjusted according to age-, sex-, and ethnicity-based trends, as well as changes in diet and physical activity attributable to the policy intervention. For instance, an individual initialized into the model at the 60th percentile of the BMI distribution will stay at that percentile unless there is a significant change in their behavior.

Appendix Figure 2 illustrates the BMI growth path of a potential agent and plots that agent’s BMI against CDC BMI growth charts. This chart pertains to the life of a simulated non-Hispanic white male from ages 2-18 and illustrates how baseline BMI and the impact of interventions are modeled. The individual is introduced into the model at age 2 with a BMI of 19.5. While this BMI corresponds to the 84th percentile of a BMI distribution conditioned upon sex and race (i.e. 2-year-old, non-Hispanic, white males), it corresponds to the 89th percentile of the population-wide distribution represented by the CDC BMI growth charts.

Appendix Figure 2

BMI Growth and Policy Impact

At age 3, his BMI decreases to 18.5, which corresponds to the 84th percentile of the sex and ethnicity conditioned distribution at age 3. This process is repeated until age 18 and, assuming no significant behavioral changes, results in the BMI path shown by the black, triangled line.

The dotted line reflects the impact of a modeled policy intervention. In this instance, the intervention was an afterschool physical activity program initiated at age 9 and continuing until age 18. Thus, this individual has 10 years of exposure to the intervention, and the intervention resulted in a sustained, elevated level of physical activity. As a result, the individual’s BMI at age 10 was reduced from 21.9 to 21 with the illustrated change in BMI percentile. At age 11, there is an additional decrease in BMI and the pattern continues until age 14 where a BMI path consistent with the increased level of physical activity is reached.

A more detailed example is provided in Appendix Table 2 . There, a six-year-old experiences a dietary intervention that entails fewer calories, increased protein, and reduced sugar and fat. As the data in the table illustrates, the impact of the dietary change occurs across several dimensions, but there is a net decrease in BMI z-score. This leads to a lower starting BMI z-score in year two, and the pattern continues with the intervention’s impact decreasing over time until a new steady-state is reached at approximately the 81 st percentile.

A weighted regression using the NHANES continuous data (2001-2010) was used to estimate the relationship between youth BMI percentile (BMI z-score) and targeted youth behaviors (physical activity and diet). The results of this analysis were incorporated into the microsimulation. This analysis used a log specification in order to estimate BMI z-score sensitivity to a percentage change in caloric consumption and physical activity, respectively. In addition, it estimated BMI z-score sensitivity to discreet changes to the following macronutrients whose daily consumption is measured in grams: protein, carbohydrates, fiber, sugar, and fat. The following general specification was used:

BMI Z − Score = α + β ag ∗ age + β r ∗ race + β s ∗ sex + β M ∗ ln ( M E T S ) + β K ∗ ln ( K c a l ) + β p ∗ protein + β c ∗ carb + β SF ∗ SFat + β TF ∗ TFat + β sug ∗ sugar + β f ∗ fiber + β ∗ INTERACTIONS

where the term INTERACTIONS represents several interactions found to be significant at the .05 level. All of these are included in Appendix Table 3 . Two of the covariates, METs and kilocalories, were heavily skewed and of several orders of magnitude larger than the other covariates. To protect against outlying values and put all covariates in the equation on a similar scale, both of these were log transformed. The log transformations, applied to both physical activity (measured in metabolic equivalents or METs) and kilocalories, allows their corresponding coefficient estimates to be interpreted as the change in BMI z-score that corresponds to a given percentage change in either physical activity or total energy consumption.

Appendix Table 4 contains estimates that are incorporated into the microsimulation model. As noted, the data used came from the continuous National Health and Nutrition Examination Survey (NHANES). Separate intercepts were included for each survey in order to filter mean level shifts due to variation attributable to question rewording or measurement change. For instance, in 2008-09, a two-day food frequency questionnaire replaced a one-day questionnaire. Similarly, questions regarding types of physical activity varied slightly from year to year.

Similarly, an indicator, or dummy, variable for ages less than 12 was incorporated to adjust for known differences in how NHANES collects data regarding physical activity and diet. Prior to age 12, physical activity and diet were measured through parental interview and certain questions, such as the detailed daily activity questionnaire, were not used.

No significant interactions with either physical activity or energy consumption (kilocalories) were found.

II. Selecting the Policies

Once the microsimulation model had been adapted for this study, a comprehensive list of obesity prevention policies was developed and then narrowed in a two-stage process to the final three policies.

First, policy recommendations to improve nutrition, increase physical activity, or promote breastfeeding were collected. These policies were found by researching recommendations issued by health organizations (e.g., Yale Rudd Center, ChangeLab Solutions, the Center for Science in the Public Interest), federal task forces (the Community Preventive Services Task Force, The White House Task Force on Childhood Obesity), and entities such as the Institute of Medicine. THOMAS was also used to search for recently introduced federal legislation that pertained to children’s nutrition, physical activity, or obesity. The policies were categorized by approach, as seen in the childhood obesity literature (e.g., environmental, economic, education/information). The study’s senior authors (RB, MA, ES, MS, ST) refined the list and approved 26 policies.

Next, criteria were used to narrow the list to 7 policies to consider further. For each of the 26 policies, the literature was briefly reviewed to better understand the policy, identify possible federal policy mechanisms, and provide a rating for each criterion to be used in the selection process. The criteria used to select the 7 policies were: extent of the evidence base, effectiveness, reach (into the general population), reach (into high risk populations), feasibility (political), and feasibility (of implementation). Tables were created that included the information gathered from the brief literature review, plus a rating for each criterion. The senior authors (RB, MA, ES, MS, ST) reviewed those tables and chose the 7 policies each thought most worthy of further consideration. The selections were then tallied and discussed. The 7 policies were determined by consensus from those receiving either the most or next-to-most number of votes:

Subsidize fruits and vegetables for recipients of the Supplemental Nutrition Assistance Program (SNAP).

Provide funding for communities to build or make improvements to public parks, playgrounds, and other safe spaces for youth to be physically active.

Require states and metropolitan planning organizations to adopt “Complete Streets” principles into all federally-funded transportation projects.

Enact a 1-cent-per-ounce excise tax on sugar-sweetened beverages, and earmark revenue for obesity prevention strategies in high risk communities.

Increase students’ fruit and vegetable consumption by expanding federal programs, such as the Fresh Fruit and Vegetable Program and the Department of Defense Fresh Fruit and Vegetable Program.

Issue briefs were then written to summarize additional research conducted on the 7 policies. Each issue brief contained pertinent history and background information, plus a rating for the same criteria as those used to select the 7 policies, with the addition of acceptability (to the general public and policymakers), precision of information for modeling, and potential impact on childhood obesity. RB, MA, ES, MS, and ST reviewed the issue briefs and selected the final three policies to be analyzed. The afterschool policy was selected because it had the strongest evidence of effectiveness, due to a large number of intervention studies. The sugar-sweetened beverage excise tax had also been well-studied and would have very broad reach into both the general population and populations at-risk for obesity. The policy to regulate advertising, although controversial, was selected to address the known negative influence of advertising on children’s diets. Although this policy posed challenges, it was selected to illustrate what could happen if the federal government regulated advertising of unhealthy foods to children.

III. Estimating Policy Impact

Once the final three policies were selected, each one was defined and a Boolean search strategy was developed using key search terms specific to each policy. The search strategies were tested and refined to weed out unrelated literature threads. The policy definitions clarified the scope of the literature search, the key data elements that were to be abstracted from identified published manuscripts, and the population groups targeted by the policy. As noted in the manuscript, the literature search and abstraction process followed methods for article identification and standards for abstraction established in prior projects.

Structured literature reviews were then conducted for the three policies to find evidence of effectiveness. Titles and abstracts were scanned for relevance to the policy. Full articles were retrieved and key information identified. There were considerable differences in the literature base for each policy, and the development of interventions is discussed on a policy-by-policy basis.

Afterschool Physical Activity

Key search terms included “activity,” “physical,” “fitness,” “school,” “children,” and “adolescen*.” The literature search generated 55 potentially relevant articles. Inclusion and exclusion criteria were developed and applied to each relevant study to identify those to abstract. Studies were eliminated if the intervention was: school-based (i.e., occurred primarily during school hours), part of a larger multi-component intervention or initiative, primarily educational, or dissimilar from a typical afterschool program in the United States (e.g., included an intensive family component, specifically targeted overweight/obese children or adolescents). Sixteen articles were identified for abstraction. During abstraction, one study was eliminated because the intervention was unclear.

Next, effect sizes for the 15 abstracted articles were compiled by sample type, sex, race/ethnicity, location, age (school grade), baseline BMI (normal, obese), adherence, intervention design, and intervention intensity (frequency and level of physical activity). The term “sample type” refers to the different groups targeted by the intervention and how they were reported. For instance, Burguera reported on two intervention groups and Howe reported on Attendees and Non-Attendees. Appendix Table 5 summarizes studies by primary author and reported study group.

Effect sizes and intervention intensity were also reported in various ways. Obesity effect sizes were reported as average change in BMI, average weight loss, and percentage sample shift in obesity category (normal, overweight, and obese). Similarly, intervention design and intensity was reported variously and included: minutes in moderate-to-vigorous physical activity, minutes relative to a control group, and percentage change in physical activity.

Most studies did not provide complete sample demographics unless a particular group was targeted. In such cases, effect sizes were averaged across all the groups. The final modeled intervention was constructed by combining individual interventions using sample sizes and demographic breakdowns as weighting factors. This intervention included weighted estimates of both adherence and physical activity change.

Excise Tax on Sugar-Sweetened Beverages

Key search terms included “sugar-sweetened beverages,” “soft drinks,” and “soda.” The literature search failed to identify an evaluation of a $0.01/ounce sugar-sweetened beverage excise tax on consumption of sugar-sweetened beverages. As mentioned in the manuscript, the evidence base linking tax policy to sugar-sweetened beverage consumption consists of well-controlled econometric studies. To develop a reliable estimate of the policy’s impact, a two-step approach was used. First, from the existing literature, a relationship between price and consumption of sugar-sweetened beverages was developed. Appendix Table 6 summarizes the articles used to establish this relationship.

The second step was to create a composite serving of a sugar-sweetened beverage. This was based upon the average caloric content and grams of sugar of 10 leading sugar-sweetened beverages. These beverages and the composite beverages are listed in Appendix Table 7 .

Ban on Fast Food Television Advertising Targeting Children

Key search terms included “brand,” “marketing,” “advertis*,” media,” “commercials,” “children,” and “food.” Consistent with the approach to assess the impact of an excise tax on sugar-sweetened beverages, a two-step approach was followed when examining the impact of reduced food advertising on obesity. First, a relationship between advertising and consumption of nutrient-poor, calorie-dense fast food was developed. Then, based upon the menus of 23 fast food restaurants, a composite meal was developed and compared to an alternative healthy meal, which was the 2012 National School Lunch Program standards. As noted in the manuscript, the relationship between consumption and advertising is based upon observational studies that explore the association between exposure to advertising and consumption.

The assumed effect size of 2-3 fewer meals per week for children from fast food-purchasing families and 3-5 fewer meals per week among adolescents was developed from Dhar’s analysis of a child-directed advertising ban in Quebec, Canada. The key impact of that advertising ban appeared to be a significant reduction in the number of households with children purchasing any fast food (13.4%). This value was applied to households with children purchasing fast food. Dhar also indicated a greater impact upon households with individuals under age 26 (38.7%). This adjustment was applied to adolescent consumption.

Next, a composite fast food meal was created. Appendix Table 8 lists, by restaurant, the composition of an average fast food meal.

IV. Sensitivity Analysis

Appendix Table 9 summarizes key univariate results for a sensitivity analysis on changes in obesity from the three policies, given different values for model parameters. For afterschool PA programs, three factors influenced changes in childhood obesity rates: baseline PA level, the intervention’s intensity (indicated by a percentage change in overall physical activity level), and participants’ adherence to the program. Obesity rates fell more under the following model assumptions: lower baseline physical activity, higher program intensity, and higher program adherence. The SSB excise tax and fast food television advertising ban, which affect obesity through dietary change, were sensitive to changes in baseline diet and assumptions regarding the composition of the SSB or fast food meal. The underlying factor for both policies is the number of calories consumed. For the excise tax, obesity rates fell more under the following assumptions: more baseline servings of SSBs and greater price sensitivity. For the advertising policy, obesity rates were most sensitive to the average number of fast food servings per week.

Appendix Table 1

Agent-level parameters within the model

Appendix Table 2

Illustration of a dietary intervention experienced by a simulated 6-year-old child

Appendix Table 3

Summary of NHANES datasets, variables, a and formulas used

a = Specific NHANES variable names are listed as used, and composite variables are carried to create new row names.

Appendix Table 4

BMI transition equations for youth model

Appendix Table 5

Articles used to develop the modeled afterschool physical activity intervention

Appendix Table 6

Articles used to develop the relationship between price and SSB consumption

Adults: avg reduction of 37
calories per day, 3.8 lbs per
year.
Children: avg reduction of
43 calories per day, 4.5 lbs
per year.

Low-income households:
reduced SSB demand is
118-135 12-ounce cans/year