Accepted Manuscript The association of fitness and school absenteeism across gender and poverty: A prospective multilevel analysis in New York City middle schools Emily M. D’Agostino, DrPH, Sophia E. Day, MA, Kevin J. Konty, PhD, Michael Larkin, MA, Subir Saha, PhD, Katarzyna Wyka, PhD PII:
S1047-2797(17)30336-8
DOI:
10.1016/j.annepidem.2017.12.010
Reference:
AEP 8339
To appear in:
Annals of Epidemiology
Received Date: 12 June 2017 Revised Date:
8 December 2017
Accepted Date: 21 December 2017
Please cite this article as: D’Agostino EM, Day SE, Konty KJ, Larkin M, Saha S, Wyka K, The association of fitness and school absenteeism across gender and poverty: A prospective multilevel analysis in New York City middle schools, Annals of Epidemiology (2018), doi: 10.1016/ j.annepidem.2017.12.010. This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
ACCEPTED MANUSCRIPT Fitness, poverty and gender on absenteeism
RI PT
Emily M. D’Agostino, DrPHa; Sophia E. Day, MAb; Kevin J. Konty, PhDb; Michael Larkin, MAc; Subir Saha, PhDc; Katarzyna Wyka, PhDa CUNY Graduate School of Public Health and Health Policy, 55 W 125th Street, New York, NY 10027, USA b NYC Department of Health and Mental Hygiene, Office of School Health, 1 Court Square, 20th Floor Queens, NY 11101, USA c NYC Department of Education, Office of School Wellness, 335 Adams Street, 28th Floor Brooklyn, NY 11201, USA
SC
a
Present address
TE D
1
M AN U
Address for correspondence: Emily D’Agostino, DrPH, MS, MA Miami-Dade Department of Parks, Recreation and Open Spaces1 275 NW 2nd Street Suite 416 Miami, Florida 33128 email:
[email protected] Office: 305-755-7938 | Mobile: 646-853-1223| Fax: 305-755-7864
EP
4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27
The association of fitness and school absenteeism across gender and poverty: A prospective multilevel analysis in New York City middle schools
AC C
1 2 3
1
ACCEPTED MANUSCRIPT Fitness, poverty and gender on absenteeism Abstract
29
Purpose
30
One-fifth to one-third of students in high-poverty, urban school districts do not attend school
31
regularly (missing ≥6 days per year). Fitness is shown to be associated with absenteeism,
32
although this relationship may differ across poverty and gender subgroups.
33
Methods
34
Six cohorts of New York City public school students were followed from grades 5-8 during
35
2006/7-2012/13 (n=349,381). Stratified three-level longitudinal generalized linear mixed models
36
were used to test the association of fitness changes and one-year lagged child-specific days
37
absent across gender and poverty.
38
Results
39
In girls attending schools in high/very high poverty areas, greater improvements in fitness the
40
year prior were associated with greater improvements in attendance (p=0.034). Relative to the
41
reference group (>20% decrease in fitness composite percentile scores from the prior year), girls
42
with a large increase in fitness (>20%) demonstrated 10.3% fewer days absent (Incidence Rate
43
Ratio (IRR) 95% CI: 0.834, 0.964), followed by those who had a 10-20% increase in fitness
44
(9.2%, IRR 95% CI: 0.835, 0.987), no change (5.4%, IRR 95% CI: 0.887, 1.007) and a 10-20%
45
decrease in fitness (3.8%, IRR 95% CI: 0.885, 1.045). In girls attending schools in low/mid
46
poverty areas, the fitness-attendance relationship was also positive, but no clear trend emerged.
47
In boys, the fitness-attendance relationship was also positive but not significant in either poverty
48
group.
49
Conclusions
AC C
EP
TE D
M AN U
SC
RI PT
28
2
ACCEPTED MANUSCRIPT Fitness, poverty and gender on absenteeism Fitness improvements may be more important to attendance improvements in high/very high
51
poverty girls compared with low/mid poverty girls, and both high/very high and low/mid poverty
52
boys. Expanding school-based physical activity programs for youth particularly in high-poverty
53
neighborhoods may increase student attendance.
RI PT
50
54 Key words
56
Poverty Areas; Gender; Schools; School-Age Population; Cardiorespiratory Fitness; Fitness
57
Testing; Youth; Absenteeism
M AN U
58
SC
55
List of abbreviations
60
NYC, New York City; DOE, Department of Education; Fitnessgram, NYC FITNESSGRAM;
61
DOHMH, Department of Health and Mental Hygiene; PACER, Progressive Aerobic
62
Cardiovascular Endurance Run; BMI, Body Mass Index; CDC, Centers for Disease Control and
63
Prevention; ICC, Intraclass Correlation; IRR, Incidence Rate Ratio.
EP AC C
64
TE D
59
3
ACCEPTED MANUSCRIPT Fitness, poverty and gender on absenteeism Introduction
66
Only 25% of youth ages 12-15 meet the recommended daily 60 minutes or more of moderate to
67
vigorous physical activity [1]. Schools in the US have replaced physical education and recess
68
with non-physical instructional time due in part to an increasing emphasis on high-stakes testing
69
[2]. This raises particular concern given a well-established link between school-based physical
70
activity and health [3-7], and potential associations with academics [8-11] and attendance [8, 10-
71
12].
SC
RI PT
65
72
Schools in high-poverty areas, in particular, may hold a critical role in providing youth with safe
74
and accessible physical activity [13, 14]. Environmental factors, including area poverty and the
75
built environment, are shown in the literature to be associated with children’s tendency to
76
participate in school- and neighborhood-based physical activity [13-16]. Neighborhood factors
77
may contribute to opportunities for safe, attractive, and accessible physical activity [15, 17].
78
School contextual factors may also impact community norms and attitudes pertaining to
79
children’s school absences [18, 19]. Given one-fifth to one-third of students in high-poverty,
80
urban school districts do not attend school regularly (missing ≥6 days per year), high-poverty
81
subgroups may benefit from physical activity interventions. However, no literature has examined
82
the potential modification of area poverty on the youth fitness-school absenteeism relationship.
TE D
EP
AC C
83
M AN U
73
84
Disparities in both physical activity and absenteeism also persist across gender. Only 18% of
85
high school girls compared with 37% of boys are reported to meet physical activity guidelines
86
[20]. Numerous studies demonstrate low self-esteem in adolescent girls is significantly
87
associated with both lower physical activity levels [21] and absenteeism [22, 23]. However, no
4
ACCEPTED MANUSCRIPT Fitness, poverty and gender on absenteeism 88
papers were identified that examined gender differences in the fitness-absenteeism relationship.
89
Given this literature, nuanced research on gender differences in the association of fitness and
90
attendance are warranted.
RI PT
91
This study examined gender differences by area poverty on the longitudinal fitness-attendance
93
association in 6 cohorts of New York City (NYC) Department of Education (DOE) middle
94
school students each followed consecutively over 4 years. It was hypothesized that
95
improvements in fitness (cardiorespiratory, muscular endurance, and muscular strength fitness
96
composite percentile scores) would predict lower subsequent absenteeism (one-year lagged days
97
absent) after accounting for potential individual- and school-level confounders, clustering, and
98
time-dependent interactions. It was further hypothesized that higher area poverty and female
99
gender would predict a stronger magnitude of association of fitness and absenteeism.
M AN U
SC
92
TE D
100 Methods
102
Data sources and study population
103
Data for this study were drawn from the NYC FITNESSGRAM (Fitnessgram) dataset jointly
104
managed by NYC Department of Education (DOE) and Department of Health and Mental
105
Hygiene (DOHMH) [24]. It is comprised of annual fitness assessments collected by DOE for
106
approximately 870,000 NYC public school students per year (grades K-12) starting in 2006-07.
107
This study was approved by the City University of New York and DOHMH IRBs and was
108
determined by these boards to be public health surveillance that is not research, and therefore
109
exempt from the requirement for obtaining written informed consent.
AC C
EP
101
110
5
ACCEPTED MANUSCRIPT Fitness, poverty and gender on absenteeism The Fitnessgram is based on the Cooper Institute’s FitnessGram®, which is demonstrated to
112
have both strong reliability and validity [25]. Fitnessgram performance tests provide a health
113
assessment related to present and future health outcomes. NYC schools are mandated to have
114
≥85% of eligible students complete the Fitnessgram assessment each year. Inclusion criteria for
115
this study included enrollment in a NYC public school for ≥2 consecutive years while in grades
116
6-8 during the study period (2006/07-2012/13) while attending a school that collected
117
Fitnessgram measurements (for Sample Selection Flow Chart see Figure A.1). Student cohorts
118
were defined based on year of initiating grade 6. Fitness-change data from grades 5-6, 6-7 and
119
7-8 was paired with days absent per year for grades 6, 7, and 8, respectively. The final sample of
120
6-8th graders included 349,381 unique students nested in 624 schools (mean school
121
population=541 students; SD=632). Students in 6th, 7th and 8th grades contributed 177,281,
122
220,769, and 186,135 student-years, respectively.
M AN U
SC
RI PT
111
TE D
123 Measures
125
The primary exposure was a categorical variable representing age- and gender-specific percent
126
change in fitness composite percentile scores based on the sum of percentile scores for the
127
Progressive Aerobic Cardiovascular Endurance Run (PACER), muscle strength and endurance
128
(curl-up and push-up) tests [24]. Scores were converted to percentiles to account for expected
129
improvements in performance with increasing age and gender. The fitness variable was
130
categorized as >20% decrease, 10-20% decrease, 20%
131
increase in performance from the year prior consistent with longitudinal research on fitness and
132
academic outcomes drawing from the Fitnessgram dataset [9].
AC C
EP
124
6
ACCEPTED MANUSCRIPT Fitness, poverty and gender on absenteeism 133
The primary outcome for this analysis was child-level number of days absent per year. Annual
134
enrollment and attendance records were matched to their Fitnessgram results by a unique student
135
identifier.
RI PT
136
Gender was based on parent report. Consistent with NYC DOHMH guidelines [26], a school
138
neighborhood’s socioeconomic status was defined according to American Community Survey
139
2007-2012 data as the percentage of households in the school zip code living below the federal
140
poverty threshold (low (30%)
141
area poverty) [27].
M AN U
SC
137
142
Covariates included time, race/ethnicity, place of birth and school size. These covariates are
144
shown in the literature to predict both fitness and absenteeism [8, 28-30]. Time (calendar year) at
145
height and weight measurement was treated as a continuous variable. Race/ethnicity was based
146
on parent report and was grouped into 5 categories: Hispanic, non-Hispanic black, non-Hispanic
147
white, Asian/Pacific Islander, and other (including multiple races). A binary school size variable
148
was based on count of individuals attending each student’s school. Per the literature, schools
149
with less than 400 students were considered small schools, and schools with 400 or more
150
students were considered non-small schools [30].
EP
AC C
151
TE D
143
152
Change in obesity status from the year prior (obese to not obese, consistently not obese,
153
consistently obese, not obese to obese) was also included as a potential confounder based on the
154
literature [8]. BMI is collected annually as part of the Fitnessgram curriculum. Obesity was
155
defined as having a BMI ≥ 95th percentile for youth in the same gender and age group according
7
ACCEPTED MANUSCRIPT Fitness, poverty and gender on absenteeism 156
to CDC guidelines [31]. Change in obesity status category was chosen as a covariate in lieu of
157
changes in BMI percentile to capture meaningful shifts in body composition associated with
158
school outcomes [32].
RI PT
159 Statistical methods
161
Descriptive statistics were computed to summarize sample characteristics. Next, trends in
162
absenteeism (days absent) by fitness, grade, and demographic characteristics were examined.
SC
160
163
Mixed models methodology was used to assess between-school variability in student
165
absenteeism. First, in order to determine the extent of variation in absenteeism at the school-
166
level, unconditional three-level longitudinal linear mixed models with random-intercepts were fit
167
to the data for all students and stratified by gender. The school-level intraclass correlation (ICC)
168
was calculated as the ratio of the variance for the school, divided by the sum of the 3 variance
169
parameter estimates, represented as: σ2school / (σ2student+ σ2school+ σ2ε) . Although univariate
170
distributions for the days absent variable demonstrated a long right-tailed Poisson distribution,
171
the ICC was calculated based on a linear mixed model given the ICC is not well defined for
172
Poisson models [33].
TE D
EP
AC C
173
M AN U
164
174
Next, the longitudinal association of change in fitness and lagged number of days absent per year
175
was assessed by fitting stratified negative binomial longitudinal mixed models [34, 35] with
176
random-intercepts, the exposure, child-specific change in fitness from the year prior, and an
177
offset term representing total instructional days per school year. Crude and adjusted models
178
were stratified by school-area poverty (low/mid vs. high/very high poverty) and gender (girls vs.
8
ACCEPTED MANUSCRIPT Fitness, poverty and gender on absenteeism boys) (models 1-8). Negative binomial models were used because data were overdispersed.
180
Statistical significance of the overall association of the exposure, change in fitness and outcome,
181
one-year lagged number of days absent per year was assessed using Type III Tests of Fixed
182
Effects. Exponentiated beta coefficients (incidence rate ratios (IRR)) represented the association
183
of each level of exposure (>20% increase, 10-20% increase, 20% decrease in fitness). A two
185
sided P-value of 20% decrease (19%), 10-20% increase (12%), and 10-20% decrease
197
(12%). There were 624 schools included in the analysis, including 365 (58%) small schools.
EP
AC C
198
TE D
190
9
ACCEPTED MANUSCRIPT
Fitness, poverty and gender on absenteeism
Table 1. Demographic and fitness-change characteristics of in New York City public school students in grades 6-8 across gender and poverty subgroups (Nstudents=349381, Nobs=675266a), 2006/7-2012/13. Males n % 177355 51
High Povertyb
nc,d
nc,d
%e
nc,d
%e
nc,d
59134
34
43462
25
38611
19277 15888 10063
33 27 17
19332 13373 4266
45 31 10
21081 14652 947
13686
23
6331
15
1785
31756 12235 15143
54 21 26
47160 11965 20624 13364 44020 14233 23880
Low Povertyb
Medium Povertyb
High Povertyb
%e
nc,d
%e
nc,d
%e
nc,d
%e
nc,d
%e
23
31422
18
60085
34
45945
26
39899
22
20 21 39
19497 14652 11081
33 24 18
20810 13152 4807
45 29 11
22241 14538 1060
56 36 3
6597
21
14609
24
6979
15
1896
5
TE D
55 6118 38 6484 3 12096 5
Very High Povertyb
22161 13488 7813
51 31 18
22630 13440 2541
59 21905 35 3035 7 6482
70 10 21
31564 12192 16329
53 20 28
22976 14386 8583
50 31 19
23161 14136 2602
58 35 7
EP
30816 18 Overall Race/Ethnicity 6095 20 Hispanic 6619 22 Non-Hispanic Black 11537 37 Non-Hispanic White Asian and/or Pacific 6412 21 Islander Language Spoken at Home 21568 70 English 3139 10 Spanish 6109 20 Other Language Place of Birth 27180 88 US 3628 12 Foreign f Change in Fitness (all years) 9957 16 20% Decrease 6935 11 10-20% Decrease 24291 40 1 race/ethnicity=177. dStudents in 6th, 7th and 8th grades contributed 177,281, 220,769, and 186,135 student-years, respectively. ePercents may not total 100 due to rounding. fChange in fitness composite percentile scores from the year prior was based on PACER (Progressive Aerobic Cardiovascular Endurance Run), Push-up and Curl-up Fitnessgram tests. gChange in obesity status from the year prior was defined according to CDC growth chart-derived norms for gender and age in months based on a historical reference population, and used to compute the BMI percentile for each child. Obesity was defined as having a BMI ≥ 95th percentile for youth in the same gender and age in months group.
AC C
EP
TE D
M AN U
a
RI PT
2407 50299 7299 1941
SC
Obese to not obese Consistently not obese Consistently obese Not obese to obese Attending Small Schools (20% decrease 22573 12.02(12.36) 10-20% decrease 10526 11.63(11.91) 20% increase 23836 10.93(11.24)
n 22532 16414 55658 17418 23946
Grade 8 (nobs=213482)
n 18416 12475 40911 12386 17602
Mean(SD) 12.85(13.06) 11.40(11.69) 10.51(10.79) 10.49(11.04) 10.79(10.85) Mean(SD) 14.87(15.40) 13.81(14.73) 13.62(14.37) 12.64(13.39) 12.99(14.31)
Tabulated mean estimates based on days absent per year. bNmale students=177355, Nfemale students=172026, Nlow/mid c poverty students=181457, Nhigh/very high poverty students=167,917. Observations account for 1-3 years of fitness-change per d student. Based on percentage of households in the school zip code living below the federal poverty threshold (low (30%) area poverty) drawing from the e American Community Survey (ACS) 2007-2012. Nmissing Area Poverty=52. fChange in fitness composite percentile scores from the year prior was based on PACER (Progressive Aerobic Cardiovascular Endurance Run), Push-up and Curl-up Fitnessgram tests.
199
TE D
a
n 17493 13707 48590 14108 17794
SC
Low/Mid Povertyde (nobs=355951) Change in Fitnessf n >20% decrease 20949 10-20% decrease 10937 20% increase 29954
Grade 7 (nobs=257201)
RI PT
Grade 6 (nobs=204583)
Twenty six percent and 22% of students attended a school in a high- or very high-poverty area,
201
respectively. Table 2 shows higher absenteeism with increasing grade level and increasing
202
poverty for students who had a decrease or no improvement vs. increase in fitness from the year
203
prior. Students in the 6th grade show inverse dose-response trends between fitness and
204
absenteeism, although higher grades do not show this trend.
AC C
205
EP
200
206
Table 3 shows higher absenteeism with decreasing fitness and increasing grade level across both
207
genders. Boys showed more mean days absent than girls across all grades, although girls
12
ACCEPTED MANUSCRIPT Fitness, poverty and gender on absenteeism
Table 3. Mean in absenteeisma for New York City public school students in grades 6-8, by level of fitness-change from the previous year across gender (Nstudents=349381b, Nobs= 675318c) Grade 7 (nobs=257203)
Grade 8 (nobs=213526)
22053 10446 29804 11509 27202
10.23(10.78) 9.75(10.57) 9.55(10.66) 9.32(10.06) 9.14(9.86)
23640 15492 50242 15536 22613
11.32(12.49) 10.30(11.77) 9.61(11.34) 9.40(10.91) 9.53(10.50)
Boys (nobs=340982) Change in Fitnessd >20% decrease 10-20% decrease 20% increase
n 21471 11017 32325 12174 26588
Mean (SD) 11.06(11.76) 10.60(11.20) 10.43(11.07) 10.08(10.65) 9.97(10.39)
n 23035 16023 51272 16401 22949
Mean (SD) 11.82(12.80) 10.94(12.00) 10.52(11.98) 10.02(11.04) 10.13(11.13)
n 17982 13129 44723 13593 18300
M AN U
a
17934 13065 44795 12904 17101
13.51(13.77) 12.33(13.19) 11.60(12.28) 11.27(12.15) 11.80(12.85)
SC
Girls (nobs=334336) Change in Fitnessd >20% decrease 10-20% decrease 20% increase
RI PT
Grade 6 (nobs=204589)
Mean (SD) 14.27(14.93) 12.76(13.36) 12.27(13.00) 11.71(12.32) 11.97(12.65)
Tabulated mean estimates based on days absent per year. bNmale students=177355, Nfemale students=172026, Nlow/mid poverty c students=181457, Nhigh/very high poverty students=167917. Observations account for 1-3 years of fitness-change per student. d Change in fitness composite percentile scores from the year prior was based on PACER (Progressive Aerobic Cardiovascular Endurance Run), Push-up and Curl-up Fitnessgram tests.
showed greater increases in absenteeism with increasing grade and decreasing fitness. The overall
209
mean absentee rate across all schools (n=624) based on the unconditional longitudinal
210
model intercept was 12 days per year. ICC estimates demonstrated a large degree of clustering at
211
the school-level (9% for models including all students; 10% for gender-stratified models; see
212
Table A.1).
EP
AC C
213
TE D
208
214
Longitudinal association of fitness-change, absenteeism and school-area poverty across gender
215
Crude models examining the fitness-absenteeism association stratified by school-area poverty
216
and gender showed significant associations between change in fitness and lagged absenteeism for
217
all poverty levels across both genders (p20% Increase 10-20% Increase 20% Decrease
IRR (model 1)e,f
BOYS (nobs=340982) Change in Fitnessd >20% Increase 10-20% Increase 20% Decrease
IRR (model 3)e,h
95% CI lower upper 0.821 0.912 0.831 0.936 0.850 0.933 0.888 1.003 . .
TE D
0.866 0.882 0.890 0.944 .
M AN U
SC
Table 4. Crude longitudinal association of fitness-change and absenteeism in New York City public school students in grades 6-8 (Nstudents=349381a, Nobs=675266b, 624 schools) across school area poverty and gender. Low/Mid School-Area Povertyc High/Very High School-Areac (nobs=355951) Poverty (nobs=319315)
AC C
EP
0.877 0.885 0.916 0.936 .
95% CI lower upper 0.817 0.941 0.818 0.958 0.860 0.974 0.864 1.015 . .
IRR (model 2)e,g 0.897 0.915 0.966 0.971 .
IRR (model 4)e,i 0.891 0.906 0.962 0.971 .
95% CI lower upper 0.847 0.950 0.856 0.979 0.919 1.016 0.909 1.037 . .
95% CI lower upper 0.819 0.969 0.823 0.997 0.894 1.036 0.882 1.069 . .
Abbreviation: IRR, Incidence rate ratios. aNmale students=177355, Nfemale students=172026, Nlow/mid poverty b students=181457, Nhigh/very high poverty students=167917. Observations account for 1-3 years of fitness-change per c student. Based on percentage of households in the school zip code living below the federal poverty threshold (low (30%) area poverty) drawing from the American Community Survey (ACS) 2007-2012. dChange in fitness composite percentile scores from the year prior was based on PACER (Progressive Aerobic Cardiovascular Endurance Run), Push-up and Curl-up Fitnessgram tests. eExponentiated beta coefficients (IRR) represented the association of each level of exposure (>20% increase, 10-20% increase, 20% decrease in fitness). P-values were calculated using adjusted negative binomial longitudinal mixed models; Statistical significance of number of days absent per year was assessed using Type III Tests of Fixed Effects. A two sided P-value of 20% Increase 10-20% Increase 20% Decrease
RI PT
Table 5. Adjusteda longitudinal association of fitness-change and absenteeism in New York City public school students in grades 6-8 (Nstudents=349381b, Nobs=675266c, 624 schools) across school-area poverty and gender. Low/Mid School-Area Povertyd High/Very High School-Area (model 5) Povertyd (model 6) IRRf,g 95% CI IRRf,h 95% CI Girls (nobs=334336)
upper 0.975 0.862 0.979 1.039 .
IRRf,j
0.893 0.898 0.941 0.957 .
95% CI
lower 0.786 0.775 0.840 0.826 .
upper 1.016 1.041 1.060 1.110 .
225
AC C
EP
TE D
Abbreviation: IRR, Incidence rate ratios. aAdjusted for individual-level race/ethnicity, grade, calendar year, change in obesity status from the year prior, place of birth (US (not NYC), NYC, or foreign), group-level school size, and including interactions (Grade*Ethnicity and Grade*Place of Birth). bNmale students=177355, Nfemale students=172026, Nlow/mid poverty students=181457, Nhigh/very high poverty students=167917. c Observations account for 1-3 years of fitness-change per student. dBased on percentage of households in the school zip code living below the federal poverty threshold (low (30%) area poverty) drawing from the American Community Survey (ACS) 2007-2012. eChange in fitness composite percentile scores from the year prior was based on PACER (Progressive Aerobic Cardiovascular Endurance Run), Push-up and Curl-up Fitnessgram tests. fExponentiated beta coefficients (IRR) represented the association of each level of exposure (>20% increase, 10-20% increase, 20% decrease in fitness). P-values were calculated using adjusted negative binomial longitudinal mixed models; Statistical significance of number of days absent per year was assessed using Type III Tests of Fixed Effects. A two sided P-value of 20% Increase
10-20% Increase
20% decrease in fitness)
Note. The graph illustrates the relationship between fitness change and absenteeism in girls by
517
school-area poverty. As shown above, relative to girls attending schools in high/very high
518
poverty areas who had a large decrease in fitness (>20% from the year prior), those who had a
519
large increase (>20%), small increase (10-20%),