The present analysis was conducted based on students’ HealthCorps survey responses and their self-reported weight and height data from New York City (NYC) high schools before and after participating in HealthCorps programming during the 2013-2014 academic year, although during this year, a total of 62 schools over 13 states participated in HealthCorps program. The present study focused on implementation of the HealthCorps programming in selected NYC high schools. The selection criteria for the NYC HealthCorps schools were as follows:  (1) must be a public high school; (2) must have 50% or more students eligible for free or reduced-price meals; and (3) must lack any existing programs aimed at improving nutrition, physical activity and mental resilience behavior. Although a total of 15 schools were selected, survey data were not available from 1 school. Whites were a minority racial group in all of the selected 14 NYC schools with HealthCorps programming.
The study design was a two parallel arm quasi-experimental pre-post comparison design as neither the schools nor the students were randomly assigned to accommodate real-world settings and constraints as much as possible, and students’ participation in the HealthCorps program was voluntary. This study was approved by the Institutional Review Boards of Albert Einstein College of Medicine and NYC Department of Education.
HealthCorps programming consists of a curriculum delivered in classrooms, through mentorship, and through a variety of during- and after-school health-promoting activities (www.healthcorps.org/resources/the-curriculum). All of these activities were coordinated and delivered by intensively trained well-qualified program coordinators each of whom is designated to a single school (https://www.healthcorps.org/become-a-coordinator/). All Coordinators are trained on all HC program components for 3 weeks in the summer as well as a week of professional development in the winter recession. Through weekly check-ins and reports, program supervisors ensure all program components are being implemented at their coordinators’ schools. Each supervisor visits a site at least once a year to observe delivery of the curriculum also to ensure standardization. Depending on school policies or school culture, however, coordinators are encouraged to tailor the program to meet the needs of their specific school and community. As such, the coordinators often work in conjunction with a school wellness council and its members.
Coordinators teach approximately 10 classroom HealthCorps lessons weekly or bi-weekly and for either a semester or a year. These lesson topics include developing tools that build mental resilience, healthy eating habits, and physical fitness. The coordinators also provide mentoring to students and staff with various health-related resources and action plans that address meeting personal health goals. Coordinators also co-facilitate wellness councils, comprised of staff members and students, who utilize the Alliance for a Healthier Generation platform (www.healthiergeneration.org) and inventory to address gaps in the school’s health policies and programming in order to design action plans to improve nutrition, physical fitness and mental resilience of their students. Weekly afterschool clubs focus on nutrition, physical fitness and/or mental resilience whose contents were determined by the annual HealthCorps Community Needs Assessment instrument that systematically identifies areas of programming need at each school.
Activities outside the classroom included lunchroom “Café-o-Yeas” food samplings, Teen Battle Chef and other cooking programs, Youth Lead Action Research, and annual Highway to Health school and community-wide Festivals. The “Café-o-Yeas” is a monthly activity in that coordinator and students share samples of healthy foods with other students at lunchtime at a booth. The Teen Battle Chef program involves culinary coaching and cooking “battles” to help students learn how to cook healthy meals and build their public speaking, teamwork and leadership skills.  Finally, the Youth Lead Action Research program helps students develop research abilities so that they can identify health needs in their school or community using surveys, interviews or photovoice web tools (www.photovoice.org) and then develop specific projects that meet those needs.
Depending on restrictions or schedules of schools, lengths of the program were a semester- or year-long. Regardless of the length, however, the school-level dosage of the program was identical and the aforementioned program activities were applied to all participating schools. Implementation process of the program activities was determined specific to each school depending on its needs and overall culture identified based on several evaluations: regular wellness council meetings, Alliance for a Healthier Generation’s School Health Index Assessment evaluations (https://schools.healthiergeneration.org/dashboard/about_assessment/, www.cdc.gov/healthyschools/shi/index.htm), and HealthCorps Community Needs Assessment evaluation at the beginning of the school year. Total exposure hours for students voluntarily participating in the program could be as long as 45 h over a maximum 36 weeks for the school year. However, the individual student-level program exposure hours varied based on attendance. Nevertheless, individual-level HealthCorps class or activities attendance records were not possible for coordinators to collect, and thus not available for the present analysis.
Self-reported weight and height, and validation
Self-reported weight and height were collected and converted to sex-age-specific BMI z-scores following the 2000 CDC growth curves,  which is relevant for measuring longitudinal pediatric adiposity changes.  Classifications of weight status were also made based on sex-age-specific BMI percentiles: normal weight (BMI < 85th%-tile), overweight/obese (BMI ≥ 85th%-tile), and obese (BMI ≥ 95th%-tile).  Since only ages in years, indicated as y below, instead of months was available in the survey, we converted ages in years to ages in months by using the formula 12*y + 6 following the CDC guideline (www.cdc.gov/nccdphp/dnpao/growthcharts/resources/sas.htm). The self-reported weight and height were well validated by measured weight and height from an internal validation random sample of N = 73 students with wide ranges of height and weight: correlations between self-reported and measured weight and height were 0.89 (p < 0.001) and 0.91 (p < 0.001), respectively.
A 77-item HealthCorps survey for the 2013-14 academic year included questions on demographic factors, weight-related knowledge and behaviors. Survey data were collected prior to and post teaching of the HealthCorps curriculum. The survey items were selected from well-established questionnaire items such as the biannual Youth Risk Behavior Survey (YRBS) conducted by the CDC, reliabilities of whose items are well supported.  Reliabilities of HealthCorps survey items were assessed in our prior study .
Knowledge items were multiple-choice questions and had a single correct answer (see Additional file 1). Each knowledge item score was 1 if its answer was correct or 0 otherwise; when an answer was missing, we considered it incorrect. Increase of one score represents one more correct answer. The following domains of knowledge items were structured by the design of the survey questionnaires: nutrition (score range: 0-6), physical activities (score range: 0-5), breakfast (score range: 0-2), sleep (score range: 0-2), and mental resilience (score range: 0-3).
Behavior items included the same five domains as the knowledge items above. With exception of the breakfast and sleep behavior domains, the other three behavior domains included subscales representing distinctive constructs. Although the subscales of behavior items were also structured by design, determination of their final subscales was aided by factor analysis and construct validity analysis. Behavior item scores were assigned on ordinal Likert scales with higher scores representing higher frequency, higher confidence or higher likelihood. Detailed scoring schemes for each behavior item are listed in the Additional file 1. Behavior subscale scores were determined as the sum of the corresponding items.
When some, but not all, item responses were missing, the sum was prorated based on the number of observed items and their sums. The subscales of nutrition behavior domains included: Fruit and Vegetable (F&V) intake, High Energy Density (HED) food intake, Water and Juice (W&J) consumption, and Sugar-Sweetened Beverage (SSB) consumption. The subscales of physical activity domains were: Physical Activity (PA) Days, and Physical Activity (PA) Barriers. Finally, the subscales of mental resilience domains were: General Attitude, Confidence in Healthy Eating, Confidence in Exercising, and Future Exercise Plan.
A vast majority of the students who participated in the HealthCorps program received it as a part of their school classroom programming. Therefore, their participation was not necessarily driven only by motivations. The pre-program survey was administered during September 2013 whereas the post-program survey was administered during December 2013 for the semester-long program and during May 2014 for the year-long program. A total of 2279 students (58% males) responded to either or both the pre and post survey. Among them, 1708 of the students were enrolled in classes teaching the HealthCorps curriculum, and the remaining 571 students, who were not in classes that included teaching of the HealthCorps curriculum, served as comparison students. For the present analysis, we applied the following inclusion criteria. The study sample must have: 1) responses to both surveys; 2) biologically probable BMI values  for both surveys; and 3) absolute difference in BMI z-scores between pre and post surveys less than 2.5 (about >3SD of sample pre-post difference) to eliminate potentially improbable weight changes. This application resulted in a total of 832 students with N = 611 (57.4% males) for the HealthCorps and N = 221 (56.6% males) for the comparison students. Detailed flowchart is depicted in Fig. 1.
Baseline characteristics were summarized with descriptive statistics such as mean, standard deviations, and percentages. Comparisons of the characteristics between HealthCorps and comparison students at baseline were made using t- or Chi-square tests. To test our first study aim, we applied mixed-effects linear models to examine the pre-post effect of HealthCorps programming on BMI z-score to takes into account potential outcome correlations between pre and post periods in addition to clustering effects of schools; specifically, school- and participant-specific intercepts were considered random.
The main predictor for primary analysis testing significance of HealthCorps effect on changes in BMI z-sores was the time effect (pre vs. post) within each arm. To this end, we included arm indicator (HealthCorps vs. comparison), time indicator, and arm-by-time interaction as fixed effects. The arm-by-time interaction term was included to construct and test contrasts pertinent to the arm-specific time effects using the sex-specific entire sample. Likewise, for the analysis of changes in obesity rates, we applied mixed-effects logistic models with the same fixed and random effects. The same modeling approach was applied to testing the secondary aim to examine the effect of HealthCorps on knowledge and health behaviors. All of primary and secondary analyses included time, age and Hispanic ethnicity as additional fixed-effects covariates, and were conducted separately for males and females and further stratified by baseline weight status. SAS v9.3 was used for all analyses and results with p < 0.05 were declared statistically significant.