ABSTRACT
Background
Data missingness can bias interpretation and outcomes resulting from data use. We describe data missingness in the longest‐standing US‐based youth fitness surveillance system (2006/07–2019/20).
Methods
This observational study uses the New York City FITNESSGRAM (NYCFG) database from 1,983,629 unique 4th–12th grade students (9,147,873 student‐year observations) from 1756 schools. NYCFG tests for aerobic capacity, muscular strength, and endurance were administered annually. Mixed effects models determined the prevalence of missingness by demographics, and associations between demographics and missingness.
Results
Across years, 20.1% of students were missing data from all three tests (11.7% for elementary students, 15.6% middle, and 36.3% high). Missingness did not differ by sex, but differed significantly by race/ethnicity and student home neighborhood socioeconomic status.
Conclusion
The nation's largest youth fitness surveillance system demonstrates the highest fitness data missingness among high school students, with more than 1/3 of students missing data. Non‐Hispanic Black students and those with very poor home neighborhood SES, across all grade levels, have the highest odds of missing data.
Implications for School Health
Strategies to better understand and ameliorate the causes of school‐based fitness testing data missingness will increase overall data quality and begin to address health inequities in this critical metric of youth health.
Keywords: child and adolescent health, data‐driven decision‐making in school health, organization and administration of school health programs, physical activity, physical fitness and sport
1. Introduction
Accurate assessment of child and adolescent health increasingly depends on real‐world data (RWD)—information collected in daily life settings [1, 2]. Among the most substantial, diverse, and representative of these data sources is school‐based physical fitness testing (SB‐PFT), mandated in public schools in 16 states comprising more than 60% of American school children [3]. Reliable metrics of physical fitness in youth are critical in the effort to effectively position resources to address the seemingly intractable epidemic of metabolic injury (e.g., obesity, type 2 diabetes, lifelong cardiovascular disease risk, etc.) associated with physical inactivity, sedentary lifestyles, and poor nutrition [4, 5, 6, 7].
SB‐PFT is an important surveillance tool to obtain an integrative measure of child/adolescent health [8, 9]. Monitoring population‐level youth physical fitness can provide timely data about patterns over time and across subgroups, not only to inform scientific research, but also to support fitness‐related programming and interventions in order to address deficiencies and inequities in fitness performance and improve youth health [10]. Given the high cost and administrative challenges of national surveillance systems to monitor youth fitness, SB‐PFT systems, many of which are mandated and funded in public school districts across the country, should play an important role in monitoring youth fitness over time. As such, the Institute of Medicine and the National Academies of Sciences have called for expanding and improving SB‐PFT as a validated, efficient tool for surveilling youth health at the population level [10, 11].
New York City (NYC), the nation's largest city with the biggest public school district (the NYC Department of Education; NYCDOE) serving over 1.1 million students in close to 1600 schools, has used an adapted version of the Cooper Institute's FITNESSGRAM (NYCFG), the most widely‐used tool to assess and report on school‐aged youths' fitness across the United States [12], to assess fitness and physical health annually in Kindergarten‐12th grade youth citywide [13]. The FITNESSGRAM consists of a variety of tests of physical fitness, strength, and flexibility that can be conducted in field conditions such as schools.
NYCDOE mandates schools have at least 85% of eligible students complete the assessment battery each school year. Since school year 2006/07, in partnership with the NYC Department of Health and Mental Hygiene (which holds administrative responsibility over the dataset), data have been combined from multiple sources (NYCDOE administrative databases, the American Community Survey, the US Census, and NYCFG assessments) into a single comprehensive record for each student for each year of public school attendance. This student‐level analytic dataset is used to define the NYC public school student population and can be used to construct prevalence and trend estimates for fitness outcomes. Resulting data have been used to support individual student progress and fitness goal attainment, assess NYCDOE programmatic successes [14, 15], report on secular trends in youth fitness across student subgroups [16, 17, 18], and identify interventions to address inequities in youth fitness outcomes [19, 20].
The NYCFG currently comprises the largest annual youth fitness surveillance system in the country; there are no other known jurisdictions that have been collecting objective student‐level fitness data and aggregating it with other relevant datasets this comprehensively or consistently. Despite the scope and breadth of this system, however, there are key factors that should be addressed to strengthen data quality as well as the predictive power of this dataset.
Missing data, a key component of RWD quality, is pervasive in human research, is often underreported, and, if of sufficient quantity, can compromise the reliability and validity of study findings [21, 22]. Understanding the mechanisms of missing data can also lead to novel insights [23]. New approaches to assess the quality of RWD and the real‐world evidence derived from it are emerging across a wide spectrum of clinical research as computation capacity now permits curation of large and complex data sets such as those found in the electronic health record [24, 25, 26, 27]. Threats to real‐world data quality tend to be greater when such data are collected in underserved and minoritized populations [28, 29, 30].
Addressing the missingness of NYCFG data would strengthen the inferences—and the resulting programmatic and policy changes enacted—based on related analyses. The objective of this paper is to assess NYCFG surveillance system fitness data missingness (from tests for aerobic capacity and muscular strength and endurance) and suggest potential areas of improvement and additional research necessary to strengthen this important system.
2. Methods
2.1. Participants
Data for this longitudinal study are from the NYCFG dataset, which is jointly managed by the NYCDOE and the NYC Department of Health and Mental Hygiene, for school years 2006/07 (first year available) through 2019/20 (most recent year available). This interval is particularly illustrative since the COVID‐19 pandemic disrupted attendance and SB‐PFT in most public schools, and the impact of these disruptions has yet to be completely resolved. NYCFG includes student fitness assessments collected annually by NYCDOE for public school students. This dataset included data collected by NYCDOE for 2,880,809 NYC public school students (15,819,535 student‐year observations) from kindergarten through 12th grade from 2322 schools [13]. Students were included in this study if they were considered fitness testing‐eligible by the NYCDOE, meaning students had to: (1) be in the NYCFG student population (defined as being enrolled in a traditional education district), which excluded 1,298,020 students (2,539,534 student‐year observations) from three districts educating special education, charter, and adult students and not required to administer the NYCFG; (2) be aged 9–19 years as of December 31st of the school year, which excluded 1,373,544 students (3,965,371 student‐year observations) outside this age range; and (3) be enrolled in 4th–12th grades (grades in which the aerobic capacity and muscular strength and endurance tests are conducted for the NYCFG), which excluded 154,035 students (166,757 student‐year observations). The final dataset included 1,983,629 unique students (9,147,873 student‐year observations).
2.2. Instrumentation: FITNESSGRAM Measures
The NYCFG consists of six tests administered September–May annually by trained school personnel using district‐provided equipment. Test administrators are required to attend formal training and use standardized protocols (i.e., manuals, video‐based training, site‐visits) designed to maximize testing consistency. This study reports on NYCFG aerobic capacity, curl‐up, and push‐up test data, which are demonstrated to be reliable and valid measures of cardiorespiratory fitness and abdominal and upper body strength and endurance, respectively, and which are the recommended assessments of youth fitness for health monitoring systems [12].
Cardiorespiratory fitness is an essential component in the assessment of health‐related fitness in youth, providing critical insights into the efficiency of the cardiovascular and respiratory systems during physical activity. The NYCFG uses the Progressive Aerobic Cardiovascular Endurance Run (PACER Test) to estimate student aerobic capacity, during which students run as many shuttles back and forth across a 15‐m course as possible, at pace with a recorded metronome that accelerates each minute. The number of completed shuttles is recorded and used to estimate O2max (which represents the maximum rate at which oxygen can be taken in, transported, and utilized by the body during intense exercise). Muscular strength and muscular endurance are measured via curl‐ups (conducted with flexed knees and feet unanchored) and push‐ups (performed at a 90° elbow angle), with both tests performed at a specified pace until the pace is not met or form is broken.
The primary outcomes in this study were four binary (yes/no) variables indicating whether the student had missing data for each singular test (aerobic capacity, curl‐up, push‐up) and whether data were missing for all three tests combined. Data were missing due to either challenges with test administration or data entry/management (biologically implausible data were retained). However, the reason for missingness was not recorded for the majority (93%) of observations.
2.3. Covariates
Established correlates of youth physical activity were included as covariates [31, 32]. Non‐time varying [31, 32] (fixed due to NYCFG data processing procedures) covariates included student sex (male/female) and race/ethnicity (Asian/Pacific Islander, non‐Hispanic Black, Hispanic/Latino, non‐Hispanic White, other/multiple races). Students not identifying a specific race/ethnicity (due to missing data or parent refusal to report), or reporting multiple races were classified as “other.” American Indian/Native Alaskan students are also grouped as “other” due to small sample size. Because the classification of the “other” race/ethnicity category is admittedly not coherent, we use it in descriptive analyses only (Table 1), aiming to reduce inferences.
TABLE 1.
Sociodemographic characteristics of New York City public school students (grades 4–12) eligible for school fitness testing, school years 2006/07–2019/20.
| 2006/07 | 2007/08 | 2008/09 | 2009/10 | 2010/11 | 2011/12 | 2012/13 | 2013/14 | 2014/15 | 2015/16 | 2016/17 | 2017/18 | 2018/19 | 2019/20 | |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Total, n | 673,392 | 672,245 | 666,369 | 670,522 | 668,508 | 662,793 | 654,392 | 648,624 | 643,965 | 642,784 | 639,938 | 643,362 | 636,983 | 623,996 |
| Male, % | 50.6 | 50.5 | 49.4 | 50.7 | 50.8 | 50.8 | 50.9 | 51.0 | 51.1 | 51.1 | 51.1 | 51.1 | 51.1 | 51.1 |
| Female, % | 49.4 | 49.5 | 50.6 | 49.3 | 49.2 | 49.2 | 49.1 | 49.0 | 48.9 | 48.9 | 48.9 | 48.9 | 48.9 | 48.9 |
| Elementary grades (4/5), % | 21.6 | 21.1 | 21.1 | 21.3 | 21.5 | 21.6 | 21.7 | 21.8 | 22.2 | 22.5 | 23.0 | 22.7 | 22.4 | 22.1 |
| Middle school grades (6–8), % | 33.3 | 32.5 | 32.4 | 32.2 | 32.1 | 32.2 | 32.4 | 32.4 | 32.0 | 31.9 | 32.0 | 32.0 | 32.5 | 32.8 |
| High school grades (9–12), % | 45.1 | 46.3 | 46.4 | 46.5 | 46.4 | 46.2 | 46.0 | 45.8 | 45.8 | 45.6 | 45.0 | 45.3 | 45.1 | 45.2 |
| Non‐Hispanic Asian, % | 13.9 | 14.2 | 14.7 | 15.4 | 15.8 | 16.1 | 16.5 | 16.9 | 17.1 | 17.4 | 18.0 | 18.1 | 18.2 | 18.4 |
| Non‐Hispanic Black, % | 32.9 | 32.6 | 32.0 | 31.1 | 30.3 | 29.5 | 28.6 | 27.5 | 26.6 | 25.7 | 24.7 | 24.0 | 23.3 | 22.5 |
| Hispanic/Latino, % | 39.1 | 39.4 | 39.6 | 39.6 | 39.9 | 39.9 | 40.0 | 40.2 | 40.5 | 40.6 | 40.5 | 40.8 | 41.1 | 41.3 |
| Non‐Hispanic White, % | 13.4 | 13.2 | 13.2 | 13.3 | 13.5 | 13.8 | 14.2 | 14.4 | 14.6 | 14.9 | 15.2 | 15.3 | 15.4 | 15.6 |
| Other/multiple races, % | 0.7 | 0.6 | 0.6 | 0.6 | 0.6 | 0.7 | 0.8 | 1.0 | 1.2 | 1.4 | 1.6 | 1.8 | 2.0 | 2.2 |
| Home neighborhood SES: a very‐wealthy, % | 18.6 | 18.7 | 18.8 | 18.9 | 19.0 | 19.1 | 19.4 | 19.7 | 19.9 | 20.2 | 20.6 | 20.9 | 21.1 | 21.5 |
| Home neighborhood SES: wealthy, % | 26.8 | 26.9 | 27.0 | 27.1 | 27.3 | 27.4 | 27.4 | 27.5 | 27.6 | 27.7 | 28.0 | 28.1 | 28.2 | 28.3 |
| Home neighborhood SES: poor, % | 24.6 | 24.5 | 24.5 | 24.5 | 24.5 | 24.6 | 24.5 | 24.4 | 24.3 | 24.3 | 24.2 | 24.0 | 23.9 | 23.7 |
| Home neighborhood SES: very‐poor, % | 30.0 | 29.9 | 29.7 | 29.5 | 29.2 | 28.9 | 28.8 | 28.4 | 28.2 | 27.8 | 27.2 | 27.1 | 26.8 | 26.5 |
| PACER, b % | 62.2 | 41.1 | 26.2 | 19.8 | 17.0 | 14.8 | 13.3 | 11.0 | 10.7 | 10.7 | 11.1 | 10.5 | 10.3 | 53.0 |
| push‐ups, % | 62.5 | 41.1 | 26.1 | 19.8 | 16.9 | 14.7 | 13.2 | 11.0 | 10.7 | 10.7 | 11.1 | 10.5 | 10.4 | 53.7 |
| Curl‐ups, % | 62.5 | 41.1 | 26.2 | 19.8 | 16.9 | 14.7 | 13.2 | 11.1 | 10.7 | 10.7 | 11.1 | 10.5 | 10.4 | 54.4 |
| All three tests, % | 63.9 | 42.3 | 27.1 | 20.5 | 17.5 | 15.2 | 13.7 | 11.5 | 11.2 | 11.3 | 11.7 | 11.2 | 11.1 | 58.0 |
Student's home neighborhood poverty level is used as a proxy for socioeconomic status (SES) and was determined for students' home census tracts using the 2008–2012 American Community Survey poverty data by census tract (defined according to 2010 Census boundaries) and categorized as very poor (0% to < 10% of individuals living below 100% of the Federal Poverty Level), poor (10% to < 20%), wealthy (20% to < 30%), and very wealthy (30% to 100%).
Fitness tests administered through the New York City FitnessGram (NYCFG) included the progressive aerobic cardiovascular endurance run (PACER), push‐ups, and curl‐ups.
Time‐varying covariates included age, grade type (elementary [grades 4th–5th]; middle [6th–8th]; and high school [9th–12th]), and student socioeconomic status (SES). Student eligibility for free or reduced‐price meals (FRPM) through the National School Lunch Program is not included in this longitudinal analysis, as FRPM data were not collected consistently across study years due to changes in eligibility criteria over time. Instead, we include student home neighborhood poverty level as a proxy for SES, due to its measurement stability across study years. Student home neighborhood SES was determined for students' home census tracts using the 2008–2012 American Community Survey poverty data by census tract (defined according to 2010 Census boundaries) and categorized as very poor (0% to < 10% of individuals living below 100% of the Federal Poverty Level), poor (10% to < 20%), wealthy (20% to < 30%), and very wealthy (30% to 100%). Missing values were resolved, when possible, using information for the same student from the closest year available.
2.4. Data Analysis
Annual student demographic characteristics and the unadjusted prevalence of having missing data for PACER, curl‐up, push‐up, and all three tests for each school year (2006/07–2018/19) were determined using descriptive statistics. To determine the proportion of missing data across study years 2007/08–2018/19, linear mixed effects models with random effects for school and student were used. These models were run for all students and then by key demographic characteristics: sex, grade level, race/ethnicity, and home neighborhood SES. Because of clear effect modification by grade level (p < 0.0001), additional models were run stratified by grade level. Years 2006/07 and 2019/20 were dropped from these models due to the atypically high prevalence of missing data, related to implementation issues during the first year (64% missing for all three tests) and COVID‐19‐related school closures (53% missing), respectively. To determine associations between key student demographic characteristics and missing fitness testing data, logistic mixed effects models with a random effect for school were run using data from 2018/19 (the most recent complete year available, which reflects practice after multiple years of implementation). These models were run for all students, as well as stratified by grade level. Analyses were conducted in Stata SEv16 (College Station, Texas).
3. Results
The NYCFG dataset included 1,983,629 unique students (9,147,873 student‐year observations) from 1756 NYCDOE traditional public schools, with an average of 653,420 students per year. Across school years, student demographic composition remained relatively stable (Table 1). In 2019/20 (the last year available), 51.0% of students were male; 22.0% in elementary grades, 33.0% in middle school grades, and 45.0% in high school grades; 18.4% non‐Hispanic Asian, 22.5% non‐Hispanic Black, 41.3% Hispanic/Latino, 15.6% White; and 21.5% had very wealthy, 28.3% wealthy, 23.7% poor, and 26.5% had very‐poor home neighborhood SES.
3.1. Prevalence of Missing Data
Data missingness was consistently similar across all three tests (PACER, push‐up, curl‐up) and just slightly higher for all three tests combined from 2007/08–2018/19 (excluding data from 2006/07, the first year testing began, and 2019/20 when schools closed for in‐person learning in March 2020 due to COVID‐19). On average, 20.0% of all students were missing PACER data, 19.9% were missing push‐up data, 20.0% were missing curl‐up data, and 20.1% were missing data from all three tests (Table 2).
TABLE 2.
Proportion of annually missing fitness testing data, by student demographic characteristics, school years 2007/08–2018/19.
| Number of observations | PACER | Push‐ups | Curl‐ups | All three tests | |
|---|---|---|---|---|---|
| Missing data, % ± SE | Missing data, % ± SE | Missing data, % ± SE | Missing data, % ± SE | ||
| All students | 7,850,485 | 20.0 ± 0.5 | 19.9 ± 0.5 | 20.0 ± 0.5 | 20.1 ± 0.5 |
| Male | 3,995,763 | 20.0 ± 0.5 | 20.0 ± 0.5 | 20.0 ± 0.5 | 20.1 ± 0.5 |
| Female | 3,854,722 | 20.0 ± 0.5 | 19.9 ± 0.5 | 20.0 ± 0.5 | 20.1 ± 0.5 |
| Non‐Hispanic Asian | 1,295,789 | 18.2 ± 0.5 | 18.3 ± 0.5 | 18.3 ± 0.5 | 19.0 ± 0.5 |
| Non‐Hispanic Black | 2,202,369 | 21.2 ± 0.5 | 21.2 ± 0.5 | 21.3 ± 0.5 | 22.0 ± 0.5 |
| Hispanic/Latino | 3,151,639 | 20.4 ± 0.5 | 20.3 ± 0.5 | 20.4 ± 0.5 | 21.2 ± 0.5 |
| Non‐Hispanic White | 1,118,113 | 21.0 ± 0.5 | 20.9 ± 0.5 | 20.9 ± 0.5 | 21.7 ± 0.5 |
| Home neighborhood SES: a very‐wealthy, % | 1,507,269 | 19.0 ± 0.5 | 19.1 ± 0.5 | 19.1 ± 0.5 | 19.8 ± 0.5 |
| Home neighborhood SES: wealthy, % | 2,107,577 | 19.2 ± 0.5 | 19.2 ± 0.5 | 19.3 ± 0.5 | 20.0 ± 0.5 |
| Home neighborhood SES: poor, % | 1,864,253 | 19.6 ± 0.5 | 19.6 ± 0.5 | 19.6 ± 0.5 | 20.4 ± 0.5 |
| Home neighborhood SES: very‐poor, % | 2,178,351 | 20.6 ± 0.5 | 20.5 ± 0.5 | 20.6 ± 0.5 | 21.4 ± 0.5 |
Student home neighborhood poverty level is used as a proxy for socioeconomic status (SES) and was determined for students' home census tracts using the 2008–2012 American Community Survey poverty data by census tract (defined according to 2010 Census boundaries) and categorized as very poor (0% to < 10% of individuals living below 100% of the Federal Poverty Level), poor (10% to < 20%), wealthy (20% to < 30%), and very wealthy (30% to 100%).
3.2. Prevalence of Missing Data by Demographic Characteristics
On average from 2007/08–2018/19, 20.1% of both female and male students were missing data for all three tests; 19.0% of non‐Hispanic Asian students, 22.0% of non‐Hispanic Black students, 21.2% of Hispanic/Latino students, 21.7% of non‐Hispanic White students; and 19.8% with very wealthy, 20.0% with wealthy, 20.4% with poor, and 21.4% with very‐poor home neighborhood SES were missing data for all three tests (Table 2).
3.3. Prevalence of Missing Data by Demographic Characteristics and School Level
Stratifying the data by school type, elementary students had the lowest proportion of missing data for all three tests (11.7%), followed by middle school students (15.6%) and high school students (36.3%; Figure 1). Across all three school types, non‐Hispanic Black students had the highest proportion of missing data (31.1% in elementary, 16.6% in middle, and 37.0% in high school). At the elementary level, after non‐Hispanic Black students, the highest proportion of missing data was for students with very poor home neighborhood SES (12.7%), followed by non‐Hispanic White (12.6%) and Hispanic/Latino (12.0% students). At the middle school level, after non‐Hispanic Black students, the highest proportion of missing data was for non‐Hispanic White (16.4%) students, students with very poor home neighbo SES (15.5%), and Hispanic/Latino (15.4%) students. At the high school level, after non‐Hispanic Black students, the highest proportion of missing data was for Hispanic/Latino (36.7%), female (36.5%), and both non‐Hispanic White and students with very poor home neighborhood SES (36.3%). This data is also presented in tabular format in Appendix A.
FIGURE 1.
Proportion of annually missing fitness testing data for all three fitness tests (administered through the New York City FITNESSGRAM (NYCFG) included PACER, push‐ups, and curl‐ups), by grade type and student demographic characteristics. Data are for school years 2007/08–2018/19 (NYCFG years 2006–07 and 2019–20 were excluded due to the high prevalence of missingness in those years). Grade level includes elementary grades (4 and 5), middle school grades (6–8), and high school grades (9–12). Student home neighborhood poverty level is used as a proxy for socioeconomic status (SES) and was determined for students' home census tracts using the 2008–2012 American Community Survey poverty data by census tract (defined according to 2010 Census boundaries) and categorized as very poor (0% to < 10% of individuals living below 100% of the Federal Poverty Level), poor (10% to < 20%), wealthy (20% to < 30%), and very wealthy (30% to 100%).
3.4. Associations Between Missing Data and Demographic Characteristics
When examining associations between student sociodemographic characteristics and missing fitness testing data for the 2018–19 school year (Table 3), for all students, there was no statistically significant difference in missing data between males and females for any of the individual tests or for all three tests. Compared with elementary students, high school students had 2.9 greater odds (95% CI: 2.396, 3.493) of having missing data for all three tests. Compared with non‐Hispanic Asian students (who had the lowest proportion of missing data among racial/ethnic groups), non‐Hispanic Black (OR = 2.2; 95% CI: 1.853, 2.553), Hispanic/Latino (OR = 1.9; 95% CI: 1.702, 2.216), and non‐Hispanic White (OR = 1.3; 95% CI: 1.127, 1.548) students all had higher odds of having missing data for all three tests. Compared with students with very wealthy home neighborhood SES, students with wealthy (OR = 1.2; 95% CI: 1.061, 1.295), poor (OR = 1.4; 95% CI: 1.279, 1.634), and very poor (OR = 1.8; 95% CI: 1.566, 2.121) home neighborhood SES all had higher odds of having missing data for all three tests.
TABLE 3.
Associations between student sociodemographic characteristics and missing fitness testing data, school year 2018–19.
| PACER odds ratio ± SE (95% CI) | Push‐ups odds ratio ± SE (95% CI) | Curl‐ups odds ratio ± SE (95% CI) | All three tests odds ratio ± SE (95% CI) | |
|---|---|---|---|---|
| All students | ||||
| Male | Ref | Ref | Ref | Ref |
| Female | 1.0 ± 2.4 (0.959, 1.053) | 1.0 ± 2.5 (0.961, 1.059) | 1.0 ± 2.4 (0.967, 1.063) | 1.0 ± 2.4 (0.958, 1.051) |
| Elementary grades (4/5) | Ref | Ref | Ref | Ref |
| Middle school grades (6–8) | 1.4 ± 0.2 (1.092, 1.756) | 1.4 ± 0.2 (1.062, 1.723) | 1.3 ± 0.2 (1.045, 1.697) | 1.2 ± 0.1 (0.958, 1.513) |
| High school grades (9–12) | 3.5 ± 0.3 (2.915, 4.271) | 3.4 ± 0.3 (2.804, 4.151) | 3.4 ± 0.3 (2.784, 4.121) | 2.9 ± 0.3 (2.396, 3.493) |
| Non‐Hispanic Asian | Ref | Ref | Ref | Ref |
| Non‐Hispanic Black | 2.2 ± 0.3 (1.885, 2.632) | 2.2 ± 0.2 (1.901, 2.657) | 2.3 ± 0.3 (1.912, 2.683) | 2.2 ± 0.2 (1.853, 2.553) |
| Hispanic/Latino | 1.9 ± 0.1 (1.690, 2.225) | 2.0 ± 0.1 (1.713, 2.258) | 2.0 ± 0.1 (1.714, 2.269) | 1.9 ± 0.1 (1.702, 2.216) |
| Non‐Hispanic White | 1.3 ± 0.1 (1.093, 1.521) | 1.3 ± 0.1 (1.117, 1.569) | 1.3 ± 0.1 (1.125, 1.580) | 1.3 ± 0.1 (1.127, 1.548) |
| Home neighborhood SES: a very‐wealthy, % | Ref | Ref | Ref | Ref |
| Home neighborhood SES: wealthy, % | 1.2 ± 0.1 (1.090, 1.326) | 1.2 ± 0.1 (1.055, 1.305) | 1.2 ± 0.1 (1.057, 1.306) | 1.2 ± 0.1 (1.061, 1.295) |
| Home neighborhood SES: poor, % | 1.5 ± 0.1 (1.320, 1.692) | 1.5 ± 0.1 (1.288, 1.673) | 1.5 ± 0.1 (1.293, 1.679) | 1.4 ± 0.1 (1.279, 1.634) |
| Home neighborhood SES: very‐poor, % | 1.9 ± 0.1 (1.605, 2.192) | 1.8 ± 0.1 (1.576, 2.166) | 1.8 ± 0.1 (1.578, 2.165) | 1.8 ± 0.1 (1.566, 2.121) |
| Elementary grades (4, 5) | ||||
| Male | Ref | Ref | Ref | Ref |
| Female | 1.1 ± 0.03 (1.001, 1.129) | 1.1 ± 0.03 (1.019, 1.141) | 1.1 ± 0.03 (1.029, 1.155) | 1.1 ± 0.03 (1.018, 1.127) |
| Non‐Hispanic Asian | Ref | Ref | Ref | Ref |
| Non‐Hispanic Black | 1.7 ± 0.2 (1.332, 2.174) | 1.7 ± 0.2 (1.369, 2.188) | 1.8 ± 0.2 (1.441, 2.300) | 1.7 ± 0.2 (1.335, 2.076) |
| Hispanic/Latino | 1.6 ± 0.2 (1.333, 2.025) | 1.7 ± 0.2 (1.359, 2.005) | 1.7 ± 0.2 (1.414, 2.084) | 1.7 ± 0.2 (1.364, 2.000) |
| Non‐Hispanic White | 1.0 ± 0.1 (0.772, 1.251) | 1.2 ± 0.3 (0.785, 1.900) | 1.3 ± 0.3 (0.829, 1.977) | 1.2 ± 0.2 (0.839, 1.766) |
| Home neighborhood SES: a very‐wealthy, % | Ref | Ref | Ref | Ref |
| Home neighborhood SES: wealthy, % | 1.3 ± 0.1 (1.063, 1.505) | 1.0 ± 0.2 (0.681, 1.555) | 1.0 ± 0.2 (0.691, 1.562) | 1.0 ± 0.2 (0.7413, 1.468) |
| Home neighborhood SES: poor, % | 1.6 ± 0.2 (1.268, 1.969) | 1.3 ± 0.2 (0.831, 1.987) | 1.3 ± 0.2 (0.830, 1.969) | 1.2 ± 0.2 (0.845, 1.753) |
| Home neighborhood SES: very‐poor, % | 2.2 ± 0.3 (1.639, 2.846) | 1.8 ± 0.4 (1.106, 2.796) | 1.7 ± 0.4 (1.088, 2.721) | 1.7 ± 0.3 (1.115, 2.499) |
| Middle school grades (6–8) | ||||
| Male | Ref | Ref | Ref | Ref |
| Female | 0.8 ± 0.05 (0.764, 0.943) | 0.9 ± 0.05 (0.762, 0.954) | 0.9 ± 0.05 (0.781, 0.963) | 0.9 ± 0.05 (0.766, 0.944) |
| Non‐Hispanic Asian | Ref | Ref | Ref | Ref |
| Non‐Hispanic Black | 2.8 ± 0.5 (2.059, 3.927) | 2.9 ± 0.5 (2.074, 3.938) | 2.9 ± 0.5 (2.127, 4.031) | 2.7 ± 0.5 (1.987, 3.691) |
| Hispanic/Latino | 2.1 ± 0.2 (1.748, 2.635) | 2.2 ± 0.2 (1.801, 2.713) | 2.2 ± 0.2 (1.819, 2.732) | 2.2 ± 0.2 (1.779, 2.654) |
| Non‐Hispanic White | 1.5 ± 0.3 (1.096, 2.150) | 1.6 ± 0.3 (1.107, 2.171) | 1.6 ± 0.3 (1.119, 2.200) | 1.5 ± 0.2 (1.120, 2.098) |
| Home neighborhood SES: a very‐wealthy, % | Ref | Ref | Ref | Ref |
| Home neighborhood SES: wealthy, % | 1.1 ± 0.1 (0.908, 1.408) | 1.1 ± 0.1 (0.915, 1.418) | 1.1 ± 0.1 (0.908, 1.409) | 1.1 ± 0.1 (0.921, 1.384) |
| Home neighborhood SES: poor, % | 1.5 ± 0.2 (1.156, 2.003) | 1.5 ± 0.2 (1.1551, 2.003) | 1.5 ± 0.2 (1.17205, 2.032) | 1.5 ± 0.2 (1.157, 1.931) |
| Home neighborhood SES: very‐poor, % | 2.0 ± 0.4 (1.39474, 2.859) | 2.0 ± 0.4 (1.40903, 2.894) | 2.0 ± 0.4 (1.41327, 2.895) | 2.0 ± 0.3 (1.417, 2.759) |
| High school grades (9–12) | ||||
| Male | Ref | Ref | Ref | Ref |
| Female | 1.1 ± 0.03 (0.995, 1.126) | 1.1 ± 0.03 (0.9966, 1.129) | 1.1 ± 0.03 (0.996, 1.129) | 1.1 ± 0.03 (0.992, 1.120) |
| Non‐Hispanic Asian | Ref | Ref | Ref | Ref |
| Non‐Hispanic Black | 2.1 ± 0.2 (1.644, 2.606) | 2.1 ± 0.2 (1.6537, 2.639) | 2.1 ± 0.3 (1.643, 2.635) | 2.0 ± 0.2 (1.635, 2.565) |
| Hispanic/Latino | 2.0 ± 0.2 (1.618, 2.507) | 2.0 ± 0.2 (1.631, 2.542) | 2.0 ± 0.2 (1.6184, 2.537169) | 2.0 ± 0.2 (1.621, 2.488) |
| Non‐Hispanic White | 1.4 ± 0.1 (1.114, 1.653) | 1.4 ± 0.1 (1.1091, 1.648) | 1.4 ± 0.1 (1.10740, 1.646494) | 1.3 ± 0.1 (1.113, 1.633) |
| Home neighborhood SES: very‐wealthy, % | Ref | Ref | Ref | Ref |
| Home neighborhood SES: wealthy, % | 1.1 ± 0.1 (1.060, 1.268) | 1.2 ± 0.1 (1.054, 1.265) | 1.2 ± 0.1 (1.058, 1.269) | 1.2 ± 0.1 (1.065, 1.270) |
| Home neighborhood SES: poor, % | 1.4 ± 0.1 (1.243, 1.604) | 1.4 ± 0.1 (1.253, 1.618) | 1.4 ± 0.1 (1.256, 1.623) | 1.4 ± 0.1 (1.257, 1.611) |
| Home neighborhood SES: very‐poor, % | 1.7 ± 0.1 (1.469, 2.055) | 1.8 ± 0.1 (1.486, 2.068) | 1.8 ± 0.1 (1.491, 2.078) | 1.7 ± 0.1 (1.487, 2.058) |
Student's home neighborhood poverty level is used as a proxy for socioeconomic status (SES) and was determined for students' home census tracts using the 2008–2012 American Community Survey poverty data by census tract (defined according to 2010 Census boundaries) and categorized as very poor (0% to < 10% of individuals living below 100% of the Federal Poverty Level), poor (10% to < 20%), wealthy (20% to < 30%), and very wealthy (30% to 100%).
3.5. Associations Between Missing Data and Demographic Characteristics for Elementary Students
At the elementary level, female students had slightly higher odds of having missing data for all three tests (OR = 1.1; 95% CI: 1.018, 1.127) compared with male students (Table 3). Compared with non‐Hispanic Asian students, non‐Hispanic Black (OR = 1.7; 95% CI: 1.335, 2.076) and Hispanic/Latino (OR = 1.7; 95% CI: 1.364, 2.000) students in elementary school had higher odds of having missing data for all three tests. Compared with students with very wealthy home neighborhood SES, students in elementary school with very poor (OR = 1.7; 95% CI: 1.115, 2.499) home neighborhood SES had higher odds of having missing data for all three tests.
3.6. Associations Between Missing Data and Demographic Characteristics for Middle School Students
At the middle school level, female students had slightly lower odds of having missing data for all three tests (OR = 0.9; 95% CI: 0.766, 0.944) compared with male students (Table 3). Compared with non‐Hispanic Asian students, non‐Hispanic Black (OR = 2.7; 95% CI: 1.987, 3.691), Hispanic/Latino (OR = 2.2; 95% CI: 1.779, 2.654), and non‐Hispanic White (OR = 1.5; 95% CI: 1.120, 2.098) students all had higher odds of having missing data for all three tests. Compared with students with very wealthy home neighborhood SES, students with poor (OR = 1.5; 95% CI: 1.157, 1.931) and very poor (OR = 2.0; 95% CI: 1.417, 2.759) home neighborhood SES had higher odds of having missing data for all three tests.
3.7. Associations Between Missing Data and Demographic Characteristics for High School Students
At the high school level, the pattern of missingness across race/ethnicity and home neighborhood SES was similar to that found at the elementary and middle school levels. Specifically, at the high school level, compared with non‐Hispanic Asian students, non‐Hispanic Black (OR = 2.0; 95% CI: 1.635, 2.565), Hispanic/Latino (OR = 2.0; 95% CI: 1.621, 2.488), and non‐Hispanic White (OR = 1.3; 95% CI: 1.113, 1.633) students all had higher odds of having missing data for all three tests (Table 3). Compared with students with very wealthy home neighborhood SES, students with wealthy (OR = 1.2; 95% CI: 1.065, 1.270), poor (OR = 1.4; 95% CI: 1.257, 1.611), and very poor (OR = 1.7; 95% CI: 1.487, 2.058) home neighborhood SES all had higher odds of having missing data for all three tests.
4. Discussion
This study using data from nearly 2 million unique students and approximately 10 million student‐year observations from the highly racially/ethnically and socioeconomically diverse NYC public school student population demonstrates a high prevalence (20%) of missing NYCFG data for all 4th–12th grade students. By comparison, the Nation's Report Card, which tracks data on mathematics and reading, shows data missingness around 10% [33]. In NYCFG, this missingness is driven primarily by data loss (36%) among high school students. Findings further demonstrate that these data are not missing at random; student race/ethnicity and home neighborhood SES were significantly associated with data missingness, signaling inequities in fitness testing administration and/or data maintenance that are worthy of addressing.
The robust NYCFG system has collected student‐level fitness data for nearly two decades and represents the largest and longest‐standing objective student‐level fitness data surveillance system in the US [13, 34, 35]. However, there is clear potential for systematic bias and random and/or differential measurement error, including variation across school sites where school staff may differ in their adherence to testing protocol, despite formal training and protocols which are designed to maximize consistency across administers. For example, while NYC physical education teachers receive formal NYCFG fitness testing training [12], classroom teachers who administer the test may not receive the same level of training. Understanding patterns of missingness is an important step in identifying methods for improving data quality, and thus improving data use, including individual‐, school‐, and systems‐level inferences drawn from these data.
Grade type was a clear effect modifier of NYCFG data missingness. Elementary students had the lowest proportion of missing data for all three tests (12%), followed by middle (16%) and high (36%) school students. Further, compared with elementary students, high school students had nearly three times higher odds of having missing data. Across study years, among high school observations with a reason for missingness listed (22%), data were most often listed as missing due to medical injury (34%), very low student attendance (25%), and long‐term absence (22%). Several other factors may contribute to greater missingness at the high school level, includinga higher number of PE exemptions as students age, which contribute to logistical challenges in scheduling testing; decreased interest in testing for older students; decreased oversight for testing in high schools; and/or a greater prioritization of academic learning over SB‐PFT. Additional research is necessary to determine and address the root causes of this missingness, as they apply both to fitness data and to school absences in general.
While differences in the odds of data missingness by sex were small and inconsistent, there were clear and consistent differences by race/ethnicity, with non‐Hispanic Black students demonstrating the highest proportion of missing data. Non‐Hispanic Black students (elementary OR = 1.7, middle OR = 2.7, and high OR = 2.0) and Hispanic/Latino students (elementary OR = 1.7, middle OR = 2.2, and high OR = 2.0) had higher odds of having missing data for all three tests compared with non‐Hispanic Asian students, who had the lowest proportion of missing data across all grade levels. In NYC, Non‐Hispanic Black and Hispanic/Latino students have also demonstrated lower prevalences of meeting age‐ and sex‐specific criterion referenced (Healthy Fitness Zone) cardiorespiratory fitness standards [12]; in 2018/19, 33% of non‐Hispanic Black and Hispanic/Latino students were in the Healthy Fitness Zone for aerobic capacity compared with 44% of non‐Hispanic White and 37% of non‐Hispanic Asian students (Thompson et al. under review).
There was also a clear dose–response relationship between student home neighborhood SES and having missing NYCFG data. Students from neighborhoods with very poor SES had consistently higher odds of missing data, across tests and across school years, compared with students from neighborhoods with very wealthy SES. These patterns also track with documented differences in fitness attainment in this population; in 2018/19 a statistically significantly lower proportion of students with very poor (33.4%), poor (34.2%), and wealthy (34.7%) home neighborhood SES were in the Healthy Fitness Zone for aerobic capacity compared with students with very wealthy (41.1%) home neighborhood SES (Thompson et al. under review). If students from demographic groups with historically lower fitness performance have greater odds of missing data, this could be signaling larger within‐school, within‐district, and student neighborhood‐level issues related to quality physical activity opportunities that need to be rectified.
In population surveillance programs, there are commonly employed statistical methods used to adjust for missing data in analyses. To reduce bias due to data not missing at random, the NYCFG has longitudinally assigned statistical weights to the measured population to be representative of the enrolled population, accounting for both individual‐ and school‐level characteristics. While these statistical weights can be applied by researchers with the statistical skills, they are not typically applied by schools for reports tracking school‐specific data or individual student progress over time. The NYCFG specifically states a goal of supporting students' learning about, and measurement of, health‐related fitness components [12, 36]. If data are missing for one‐fifth of eligible students, with greater missingness for students with historically lower physical activity and fitness levels [37, 38], this could limit the program's ability to serve its goal for all students and to address existing disparities. The high proportion of missing data also points to the lower standing of physical education and health in the US K‐12 educational system; if academic (Math or English Language Arts) data were missing for 20% of students, intervention would ensue.
Several study limitations warrant mention. First, and most importantly, the reason for data missingness in this dataset is unknown for most student year observations, limiting our ability to identify interventions to directly address the specific and potentially NYC‐unique source(s) of missingness. Further research is necessary to elucidate specific data collection barriers in this unique urban context. Second, the NYCFG dataset does not include students from private, charter, and special education schools (approximately 19%, 13%, and 2% of NYC school‐aged children, respectively) [39]. Because public school students are more likely to be non‐White and to have lower home neighborhood SES, these findings may not generalize to the entire NYC youth population. Third, cardiorespiratory and muscular strength and endurance data collection starts in the 4th grade, limiting our ability to include the city's K‐3rd grade population, further reducing the generalizability of these findings.
4.1. Implications for School Health Policy, Practice, and Equity
SB‐PFT is an underutilized and remarkably diverse set of pediatric health RWD that provides unique insight into youth integrative physiological well‐being. Moreover, SB‐PFT can be simultaneously used to strengthen policies, improve the school environment, and create improvement plans at the individual student level. While currently not a widespread practice, SB‐PFT also has the potential to become a part of individual students' electronic medical records, providing clinicians with valuable and actionable metrics of health. However, work to address the root causes of data missingness are necessary to strengthen the value of this important system.
Additional evidence from NYC is needed to qualitatively understand the barriers to NYCFG protocol adherence and potential data entry/maintenance issues at the teacher, site administrator, and district levels that may be driving missingness, particularly as students age. Further, engaging key stakeholders (students, teachers, and administrators) in the co‐creation of updated methodology for SB‐PFT could support higher adherence to testing best practices and decrease data missingness. Finally, working to ensure fitness testing is better embedded in physical education curriculum could give it more relevance to teachers and students, which could also decrease missingness.
5. Conclusion
Missing data is a threat to the RWD quality of SB‐PFT. We found that missing SB‐PFT data in the NYCFG—the nation's largest youth fitness surveillance system—is an enduring problem; across years, an average 20% of students are missing data for all three fitness tests, driven primarily by 36% missingness among high school students. Non‐Hispanic Black and Hispanic/Latino students and those with very poor home neighborhood SES have the highest odds of missing data for all three fitness tests. The relatively stable patterns of missingness across years suggest troubling pervasiveness of health‐related inequities. Available statistical weighting procedures can mitigate to some degree the threat of missingness to data quality. Nonetheless, overcoming the causes of missing data will improve SB‐PFT and enhance the effectiveness of this invaluable set of pediatric health RWD.
Ethics Statement
Preparation of this paper did not involve primary research or data collection involving human subjects, and therefore, no institutional review board examination or approval was required. Both the NYCDOHMH Institutional Review Board and the UC Berkeley Committee for the Protection of Human Subjects deemed this non‐human subject research.
Conflicts of Interest
REDACTED, REDACTED, and REDACTED are employed by the New York City Department of Health and Mental Hygiene Office of School Health, Data Science and Research Division, which manages the NYCFG dataset. The other authors have no conflicts of interest to declare.
Acknowledgments
The authors would like to acknowledge the tremendous work by the NYCDOE Office of School Wellness Programs and NYCDOE school administrators, teachers, and students who oversaw, implemented, and participated in the NYCFG.
Appendix A. Proportion of Annually Missing Fitness Testing Data, by School Level and Student Demographic Characteristics, School Years 2007/08–2018/19
| Number of observations | PACER | Push‐ups | Curl‐ups | All three tests | |
|---|---|---|---|---|---|
| Missing data, % ± SE | Missing data, % ± SE | Missing data, % ± SE | Missing data, % ± SE | ||
| Elementary grades (4, 5) | 1,719,243 | 10.8 ± 0.4 | 10.8 ± 0.4 | 10.9 ± 0.4 | 11.7 ± 0.4 |
| Male | 876,127 | 10.9 ± 0.4 | 10.9 ± 0.4 | 11.0 ± 0.4 | 11.9 ± 0.4 |
| Female | 843,116 | 10.5 ± 0.4 | 10.5 ± 0.4 | 10.5 ± 0.4 | 10.5 ± 0.4 |
| Non‐Hispanic Asian | 283,854 | 10.4 ± 0.5 | 10.5 ± 0.5 | 10.5 ± 0.5 | 11.3 ± 0.5 |
| Non‐Hispanic Black | 435,709 | 12.1 ± 0.5 | 12.1 ± 0.5 | 12.2 ± 0.5 | 13.1 ± 0.5 |
| Hispanic/Latino | 707,294 | 11.0 ± 0.4 | 11.0 ± 0.4 | 11.1 ± 0.4 | 12.0 ± 0.5 |
| Non‐Hispanic White | 270,435 | 11.7 ± 0.5 | 11.6 ± 0.5 | 11.6 ± 0.5 | 12.6 ± 0.5 |
| Home neighborhood SES: a very‐wealthy, % | 350,301 | 9.8 ± 0.5 | 9.9 ± 0.5 | 9.9 ± 0.5 | 10.8 ± 0.5 |
| Home neighborhood SES: wealthy, % | 454,664 | 10.5 ± 0.4 | 10.5 ± 0.4 | 10.6 ± 0.4 | 11.5 ± 0.4 |
| Home neighborhood SES: poor, % | 401,507 | 10.7 ± 0.5 | 10.8 ± 0.5 | 10.9 ± 0.5 | 11.7 ± 0.5 |
| Home neighborhood SES: very‐poor, % | 466,001 | 11.7 ± 0.5 | 11.7 ± 0.5 | 11.8 ± 0.5 | 12.7 ± 0.5 |
| Middle school grades (6–8) | 2,529,366 | 14.9 ± 0.7 | 14.8 ± 0.7 | 14.8 ± 0.7 | 15.6 ± 0.7 |
| Male | 1,292,343 | 14.5 ± 0.7 | 14.5 ± 0.7 | 14.5 ± 0.7 | 15.2 ± 0.7 |
| Female | 1,237,023 | 14.5 ± 0.7 | 14.4 ± 0.7 | 14.5 ± 0.7 | 15.2 ± 0.7 |
| Non‐Hispanic Asian | 415,166 | 13.4 ± 0.7 | 13.4 ± 0.7 | 13.4 ± 0.7 | 13.9 ± 0.7 |
| Non‐Hispanic Black | 686,790 | 15.7 ± 0.7 | 15.5 ± 0.7 | 15.6 ± 0.7 | 16.6 ± 0.7 |
| Hispanic/Latino | 1,028,538 | 14.7 ± 0.7 | 14.6 ± 0.7 | 14.7 ± 0.7 | 15.4 ± 0.7 |
| Non‐Hispanic White | 373,821 | 16.0 ± 0.7 | 15.9 ± 0.7 | 15.8 ± 0.7 | 16.4 ± 0.7 |
| Home neighborhood SES: a very‐wealthy, % | 497,755 | 14.2 ± 0.7 | 14.0 ± 0.7 | 14.1 ± 0.7 | 14.9 ± 0.7 |
| Home neighborhood SES: wealthy, % | 672,241 | 14.3 ± 0.7 | 14.2 ± 0.7 | 14.2 ± 0.7 | 14.9 ± 0.7 |
| Home neighborhood SES: poor, % | 591,804 | 13.9 ± 0.7 | 13.8 ± 0.7 | 13.8 ± 0.7 | 14.6 ± 0.7 |
| Home neighborhood SES: very‐poor, % | 697,786 | 14.7 ± 0.7 | 14.6 ± 0.7 | 14.7 ± 0.7 | 15.5 ± 0.7 |
| High school grades (9–12) | 3,601,876 | 35.8 ± 1.1 | 35.8 ± 1.1 | 35.8 ± 1.1 | 36.3 ± 1.1 |
| Male | 1,827,293 | 35.6 ± 1.1 | 35.6 ± 1.1 | 35.6 ± 1.1 | 36.2 ± 1.1 |
| Female | 1,774,583 | 35.9 ± 1.1 | 35.9 ± 1.1 | 35.9 ± 1.1 | 36.5 ± 1.1 |
| Non‐Hispanic Asian | 596,769 | 31.2 ± 1.1 | 31.2 ± 1.1 | 31.2 ± 1.1 | 31.8 ± 1.1 |
| Non‐Hispanic Black | 1,079,870 | 36.5 ± 1.1 | 36.4 ± 1.1 | 36.4 ± 1.1 | 37.0 ± 1.1 |
| Hispanic/Latino | 1,415,807 | 36.1 ± 1.1 | 36.1 ± 1.1 | 36.1 ± 1.1 | 36.7 ± 1.1 |
| Non‐Hispanic White | 473,857 | 35.7 ± 1.1 | 35.7 ± 1.1 | 35.6 ± 1.1 | 36.3 ± 1.1 |
| Home neighborhood SES: a very‐wealthy, % | 659,213 | 33.4 ± 1.1 | 33.4 ± 1.1 | 33.4 ± 1.1 | 33.9 ± 1.1 |
| Home neighborhood SES: wealthy, % | 980,672 | 33.8 ± 1.1 | 33.8 ± 1.1 | 33.8 ± 1.1 | 34.4 ± 1.1 |
| Home neighborhood SES: poor, % | 870,942 | 34.6 ± 1.1 | 34.6 ± 1.1 | 34.6 ± 1.1 | 35.2 ± 1.1 |
| Home neighborhood SES: very‐poor, % | 1,014,564 | 35.7 ± 1.1 | 35.7 ± 1.1 | 35.7 ± 1.1 | 36.3 ± 1.1 |
Student's home neighborhood poverty level is used as a proxy for socioeconomic status (SES) and was determined for students' home census tracts using the 2008–2012 American Community Survey poverty data by census tract (defined according to 2010 Census boundaries) and categorized as very poor (0% to < 10% of individuals living below 100% of the Federal Poverty Level), poor (10% to < 20%), wealthy (20% to < 30%), and very wealthy (30% to 100%).
Thompson H. R., Ricks‐Oddie J. L., Schneider M., et al., “Data Missingness and Equity Implications in the Nation's Largest Student Fitness Surveillance System: The New York City School Based Physical Fitness Testing Programs, 2006–2020,” Journal of School Health 95, no. 7 (2025): 498–509, 10.1111/josh.70021.
References
- 1. Horton D. B., Blum M. D., and Burcu M., “Real‐World Evidence for Assessing Treatment Effectiveness and Safety in Pediatric Populations,” Journal of Pediatrics 238 (2021): 312–316. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2. Zisis K., Pavi E., Geitona M., and Athanasakis K., “Real‐World Data: A Comprehensive Literature Review on the Barriers, Challenges, and Opportunities Associated With Their Inclusion in the Health Technology Assessment Process,” Journal of Pharmacy and Pharmaceutical Sciences 27 (2024): 12302. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3. Krochmal P., Cooper D. M., Radom‐Aizik S., and Lu K. D., “US School‐Based Physical Fitness Assessments and Data Dissemination,” Journal of School Health 91, no. 9 (2021): 722–729. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4. Committee on Fitness Measures and Health Outcomes in Youth; Food and Nutrition Board; Institute of Medicine , “Measuring Fitness in Youth,” in Fitness Measures and Health Outcomes in Youth, ed. Pate R., Oria M., and Pillsbury L. (National Academies Press (US), 2012), 2, https://www.ncbi.nlm.nih.gov/books/NBK241311/. [PubMed] [Google Scholar]
- 5. Stodden D., Sacko R., and Nesbitt D., “A Review of the Promotion of Fitness Measures and Health Outcomes in Youth,” American Journal of Lifestyle Medicine 11, no. 3 (2017): 232–242. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6. Kraav J., Tamme R., Remmel L., et al., “Arterial Structure in 18‐Year‐Old Males Is Dependent on Physical Activity at 12 Years and Cumulative Cardiorespiratory Fitness From Puberty to Late Adolescence,” Pediatric Exercise Science 35, no. 3 (2023): 144–154. [DOI] [PubMed] [Google Scholar]
- 7. Henriksson H., Henriksson P., Tynelius P., and Ortega F. B., “Muscular Weakness in Adolescence Is Associated With Disability 30 Years Later: A Population‐Based Cohort Study of 1.2 Million Men,” British Journal of Sports Medicine 53, no. 19 (2019): 1221–1230. [DOI] [PubMed] [Google Scholar]
- 8. Ortega F. B., Ruiz J. R., Castillo M. J., and Sjostrom M., “Physical Fitness in Childhood and Adolescence: A Powerful Marker of Health,” International Journal of Obesity 32, no. 1 (2008): 1–11. [DOI] [PubMed] [Google Scholar]
- 9. Ortega F. B., Artero E. G., Ruiz J. R., et al., “Reliability of Health‐Related Physical Fitness Tests in European Adolescents. The HELENA Study,” International Journal of Obesity 32, no. Suppl 5 (2008): S49–S57. [DOI] [PubMed] [Google Scholar]
- 10. Pate R. R., Welk G. J., and McIver K. L., “Large‐Scale Youth Physical Fitness Testing in the United States: A 25‐Year Retrospective Review,” Pediatric Exercise Science 25, no. 4 (2013): 515–523. [DOI] [PubMed] [Google Scholar]
- 11. National Academies of Sciences, Medicine, Medicine Division, and Committee on Strategies for Implementing Physical Activity Surveillance , Implementing Strategies to Enhance Public Health Surveillance of Physical Activity in the United States (National Academies Press, 2019). [PubMed] [Google Scholar]
- 12. Plowman S. and Meredith M., FitnessGram Reference Guide, 4th ed. (Cooper Institute, 2013). [Google Scholar]
- 13. Day S. E., Konty K. J., Napier M. D., Irvin E., Thompson H. R., and D'Agostino E. M., “NYC FITNESSGRAM: Population‐Level Physical Fitness Surveillance for New York City Youth,” American Journal of Epidemiology 192, no. 3 (2023): 334–341, 10.1093/aje/kwac204. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14. Thompson H. R., Madsen K. A., Nguyen C., McKenzie T. L., and Picciotto S., “Impact of New York City's 4‐Year Multi‐Component Natural Experiment to Improve Elementary School Physical Education on Student Cardiorespiratory Fitness,” BMC Public Health 24, no. 1 (2024): 3161. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15. Bleeker JM M., Fox M. K., Jean‐Louis B., and Fox M., “The Impact of Healthy Harlem on Student Fitness Outcomes After Two and Three Years,” Mathematica Policy Research, (2018). https://ideas.repec.org/p/mpr/mprres/b2d609156cb34ec49a0627e84117bef3.html. [Google Scholar]
- 16. Konty K. J., Day S. E., Larkin M., Thompson H. R., and D'Agostino E. M., “Physical Fitness Disparities Among New York City Public School Youth Using Standardized Methods, 2006‐2017,” PLoS One 15, no. 4 (2020): e0227185. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17. Neshteruk C. D., Day S. E., Konty K. J., Armstrong S. C., Skinner A. C., and D'Agostino E. M., “Severe Obesity and Fitness in New York City Public School Youth, 2010‐2018,” BMC Public Health 23, no. 1 (2023): 345. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18. Zewdie H. Y., Zhao A. Y., Ogletree S. S., et al., “Longitudinal Associations Between Neighborhood Child Opportunity and Physical Fitness for New York City Public School Youth,” American Journal of Epidemiology 192, no. 8 (2023): 1278–1287. [DOI] [PubMed] [Google Scholar]
- 19. Bezold C. P., Stark J. H., Rundle A., et al., “Relationship Between Recreational Resources in the School Neighborhood and Changes in Fitness in New York City Public School Students,” Journal of Urban Health 94, no. 1 (2017): 20–29. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20. Stiefel L., Elbel B., Pflugh Prescott M., Aneja S., and Schwartz A. E., “School Wellness Programs: Magnitude and Distribution in New York City Public Schools,” Journal of School Health 87, no. 1 (2017): 3–11. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21. Nijman S. W. J., Leeuwenberg A. M., Beekers I., et al., “Missing Data Is Poorly Handled and Reported in Prediction Model Studies Using Machine Learning: A Literature Review,” Journal of Clinical Epidemiology 142 (2022): 218–229. [DOI] [PubMed] [Google Scholar]
- 22. Heymans M. W. and Twisk J. W. R., “Handling Missing Data in Clinical Research,” Journal of Clinical Epidemiology 151 (2022): 185–188. [DOI] [PubMed] [Google Scholar]
- 23. Basiri A. and Brunsdon C., “Missing Data as Data,” Patterns 3, no. 9 (2022): 100587. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24. Pfaffenlehner M., Behrens M., Zöller D., et al., “Methodological Challenges Using Routine Clinical Care Data for Real‐World Evidence: A Rapid Review Utilizing a Systematic Literature Search and Focus Group Discussion,” BMC Medical Research Methodology 25, no. 1 (2025): 8. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25. Khare R., Utidjian L. H., Razzaghi H., et al., “Design and Refinement of a Data Quality Assessment Workflow for a Large Pediatric Research Network,” EGEMS 7, no. 1 (2019): 36. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26. Hicken V. N., Thornton S. N., and Rocha R. A., “Integration Challenges of Clinical Information Systems Developed Without a Shared Data Dictionary,” Studies in Health Technology and Informatics 107, no. Pt 2 (2004): 1053–1057. [PubMed] [Google Scholar]
- 27. Barghout S. H., Meti N., Chotai S., et al., “Adaptive Universal Principles for Real‐World Observational Studies (AUPROS): An Approach to Designing Real‐World Observational Studies for Clinical, Epidemiologic, and Precision Oncology Research,” British Journal of Cancer 132, no. 2 (2024): 139–153. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28. Nong P., Williamson A., Anthony D., Platt J., and Kardia S., “Discrimination, Trust, and Withholding Information From Providers: Implications for Missing Data and Inequity,” SSM Popul Health 18 (2022): 101092. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29. Chen Y., Lin H. Y., Tseng T. S., Wen H., and DeVivo M. J., “Racial Differences in Data Quality and Completeness: Spinal Cord Injury Model Systems' Experiences,” Topics in Spinal Cord Injury Rehabilitation 24, no. 2 (2018): 110–120, 10.1310/sci2402-110. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30. Getzen E., Ungar L., Mowery D., Jiang X., and Long Q., “Mining for Equitable Health: Assessing the Impact of Missing Data in Electronic Health Records,” Journal of Biomedical Informatics 139 (2023): 104269. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31. Biddle S. J. H., Atkin A. J., Cavill N., and Foster C., “Correlates of Physical Activity in Youth: A Review of Quantitative Systematic Reviews,” International Review of Sport and Exercise Psychology 4, no. 1 (2011): 25–49. [Google Scholar]
- 32. Van Der Horst K., Paw M. J., Twisk J. W., and Van Mechelen W., “A Brief Review on Correlates of Physical Activity and Sedentariness in Youth,” Medicine and Science in Sports and Exercise 39, no. 8 (2007): 1241–1250, 10.1249/mss.0b013e318059bf35. [DOI] [PubMed] [Google Scholar]
- 33. The National Center for Education Statistics , “National Assessment of Educational Profress. The Nation's Report Card,” 2025, https://nces.ed.gov/nationsreportcard/.
- 34. NYC , “Department of Education NYC FITNESSGRAM,” 2020, https://www.schools.nyc.gov/learning/subjects/physical‐education.
- 35. NYC FITNESSGRAM , “Assessment Protocol. FITNESSGRAM Assessment Module,” 2020, http://nycphysicaleducation.com/nyc‐fitnessgram/.
- 36. Plowman S., Sterling C., Corbin C., Meredith M., Welk G., and Morrow J., “The History of FitnessGram,” Journal of Physical Activity & Health 3, no. Supp 2 (2006): S5–S20. [Google Scholar]
- 37. Thompson H. R., Singh B. K., Ibarra‐Castro A., Woodward‐Lopez G., and Madsen K. A., “Access to Credentialed Elementary Physical Education Teachers in California and Students' Cardiorespiratory Fitness,” Preventive Medicine 121 (2019): 62–67. [DOI] [PubMed] [Google Scholar]
- 38. Whitt‐Glover M. C., Taylor W. C., Floyd M. F., Yore M. M., Yancey A. K., and Matthews C. E., “Disparities in Physical Activity and Sedentary Behaviors Among US Children and Adolescents: Prevalence, Correlates, and Intervention Implications,” Journal of Public Health Policy 30, no. Suppl 1 (2009): 309–334. [DOI] [PubMed] [Google Scholar]
- 39. Domanico R., A Statistical Profile of New York's K‐12 Educational Sector: Race, Income, and Religion (Manhattan Institute for Policy Research, 2020). [Google Scholar]