Homework Help
UNDERSTANDING EFFECTS OF EXERCISE AND DIET TO IMPROVE MENTAL AND
PHYSICAL HEALTH IN CHILDREN WITH BEHAVIORAL HEALTH DISORDERS
APRIL B. BOWLING
A Dissertation Submitted to the Faculty of
The Harvard T.H. Chan School of Public Health
in Partial Fulfillment of the Requirements
for the Degree of Doctor of Science
in the Department of Nutrition
Harvard University
Boston, MA
March 2017
ProQuest Number:
All rights reserved
INFORMATION TO ALL USERS The quality of this reproduction is dependent on the quality of the copy submitted.
In the unlikely event that the author did not send a complete manuscript and there are missing pages, these will be noted. Also, if material had to be removed,
a note will indicate the deletion.
Published by ProQuest LLC (
ProQuest
). Copyright of the Dissertation is held by the Author.
All Rights Reserved. This work is protected against unauthorized copying under Title 17, United States Code
Microform Edition © ProQuest LLC.
ProQuest LLC 789 East Eisenhower Parkway
P.O. Box 1346 Ann Arbor, MI 48106 - 1346
28223221
28223221
2020
ii
Dissertation Advisor: Dr. Kirsten K. Davison April B. Bowling
Understanding Effects of Exercise and Diet to Improve Mental and Physical Health in Children
with Behavioral Health Disorders
Abstract
Approximately 13%–20% of children living in the United States experience a
diagnosable behavioral health disorder (BHD) in a given year. Children with BHDs have
elevated risk of poor mental and physical health outcomes. It is established that diet quality and
physical activity (PA) levels are predictors of both chronic disease risk and behavioral health.
Unfortunately, children with BHDs experience increased risk of poor diet and low PA relative to
typically developing peers.
Given the high prevalence of BHDs and associated health disparities, this dissertation
aims to help address major gaps in existing research. Chapter one investigates the effects on
behavior of an RCT using cybercycling at a therapeutic school serving children with BHDs. We
used mixed-effects logistic regression to assess relationships between intervention exposures and
behavioral outcomes. We found that students successfully engaged in and had significantly
lower risk of poor behavioral outcomes during the intervention, particularly on days they
participated in aerobic exercise.
Chapter two examines exercise dose and changes in behavior, to better inform exercise
prescription and interventions for children with a variety of BHDs. We used mixed-effects linear
regression to examine dose-response relationships between aerobic exercise duration and
intensity, and minutes of disciplinary time out of class and self-regulation scores. We found
duration had an inverse, linear relationship with negative outcomes, and effects were amplified
among children with an Attention Deficit-Hyperactivity Disorder (ADHD) diagnosis.
iii
Chapter three considers roles played by ADHD diagnosis, stimulant use, and health
behaviors in predicting children’s BMI change. We applied linear regression to longitudinal data
from a nationally-representative cohort of children to assess associations between duration of
stimulant use, ADHD diagnosis, and BMI change, and logistic regression to assess odds of poor
diet and PA patterns between groups. We found stimulant use predicted greater BMI change in
pre-/early adolescence, and children with ADHD were at elevated risk of poor dietary patterns
regardless of medication status. Diet and PA did not mediate relationships between stimulant
use and BMI change.
Each chapter provides important evidence to address chronic disease and behavioral
health risks among children with BHDs in order to reduce existing health disparities.
iv
Table of Contents
Abstract ii
List of Figures v
List of Tables vi
Acknowledgements vii
Chapter 1 Cybercycling Effects on Classroom Behavior in Children with
Behavioral Health Disorders: an RCT
1
Chapter 2 Dose-Response Effects of Exercise on Behavioral Health in
Children and Adolescents
24
Chapter 3 ADHD Medication, Dietary Patterns, Physical Activity, and
BMI in Children: A Longitudinal Analysis of the ECLS-K Study
47
v
List of Figures
Figure 1.1. Overview of Manville Moves cybercycling physical education
curriculum.
6
Figure 1.2. Enrollment, randomization, and attrition flow diagram
(CONSORT).
11
Figure 1.3. Percentage of participants exceeding CATRS-10 screening
thresholds as a function of study condition.
13
Figure 2.1. Effect modification of the relationship between exercise duration
and disciplinary time out of class (TOC) by ADHD status.
39
Figure 2.2. Effect modification of the relationship between exercise duration
and self-regulation score (Conners Abbreviated Teacher Rating
Scale-10 Item) by ADHD status.
39
Figure 3.1. Conceptual model of relationships between ADHD, medication,
health behaviors, and BMI change in children.
52
vi
List of Tables
Table 1.1. Baseline participant characteristics. 12
Table 1.2. Overall intervention and cybercycling PE effects on behavioral
health outcomes compared with control condition.
15
Table 2.1. Description of baseline participant characteristics. 34
Table 2.2. Regression modeling results for continuous exercise exposures
and measures for behavioral outcomes.
36
Table 3.1. Characteristics of the study population. 58
Table 3.2. Linear regression modeling results for BMI and BMI z-score
outcomes.
61
Table 3.2. Logistic regression results for dietary patterns and physical
activity outcomes in 5 th
grade.
62
vii
Acknowledgements
I owe a tremendous debt of gratitude to all of my family, friends and colleagues who
have supported, encouraged, and not held my absences against me over the last three years. I am
especially grateful to my friend, Dr. Kara Zivin, who provided invaluable insight and helped me
navigate doctoral studies while being a parent of young children, and when I was notably less
successful than I wished to be, provided a model of humor and grace which I could emulate. My
greatest thanks go to my family, particularly my husband, Peter, and my children, Morgan and
Sam, who have shown great independence and growth while proudly supporting my research.
This dissertation would not have been possible without the invaluable mentoring and
support of my advisor, Dr. Kirsten Davison. In addition to providing me with critical research
opportunities, she also helped me assemble a very strong, cross-disciplinary dissertation
committee. I am indebted to its members, Dr. William Beardslee, Dr. Sebastien Haneuse, and
Dr. Daniel P. Miller, all of whom provided keen insight and important feedback on my
dissertation papers throughout the process. I am also grateful to Dr. Eric Rimm and Dr. James
Slavet for sharing their expertise, as well as to the Manville School research team for their
dedication and quality of work.
Finally, I would like to dedicate this dissertation to two people. First, to the memory of
my friend, Kimberly Whitmore, who initially engaged me in this field of research and remains
one of the kindest and most hopeful people I have ever known. Secondly, it is dedicated to my
friend and former athlete, Patrick Cogan, who has not let a diagnosis of Friedreich’s Ataxia stop
him from amazing feats of athletics and friendship, and who provided me with daily inspiration
as I undertook my doctoral studies and completed this dissertation.
1
Chapter 1:
Cybercycling effects on classroom behavior in children with
behavioral health disorders: an RCT
April Bowling, James Slavet, Daniel P. Miller, Sebastien Haneuse,
William Beardslee, Kirsten Davison
2
Abstract
Background and Objectives: Exercise is linked with improved cognition and behavior in
children in clinical and experimental settings. This translational study examined if an aerobic
cybercycling intervention integrated into physical education (PE) resulted in improvements in
behavioral self-regulation and classroom functioning among children with mental health
disabilities attending a therapeutic day school.
Methods: Using a 14-week crossover design, students (N=103) were randomly assigned by
classroom (k=14) to receive the 7-week aerobic cybercycling PE curriculum during fall 2014 or
spring 2015. During the intervention, children used the bikes 2 times per week during 30-40
minute PE classes. During the control period children participated in standard non-aerobic PE.
Mixed effects logistic regression was used to assess relationships between intervention exposures
and clinical thresholds of behavioral outcomes, accounting for both individual and classroom
random effects.
Results: Children experienced 32-51% lower odds of poor self-regulation and learning-
inhibiting disciplinary time out of class when participating in the intervention; this result is both
clinically and statistically significant. Effects were appreciably more pronounced on days that
they participated in the aerobic exercise, but carryover effects were also observed.
Conclusions: Aerobic cybercycling PE shows promise for improving self-regulation and
classroom functioning among children with complex behavioral health disorders. This school-
based exercise intervention may significantly improve child behavioral health without increasing
parental burden or health care costs, or disrupting academic schedules.
3
Introduction
The Centers for Disease Control and Prevention reports 13%–20% of children living in
the US experience behavioral health disorders (BHD) in a given year; such disorders are among
the most costly conditions to treat in children. 1 Those experiencing BHD have other chronic
health conditions (e.g., asthma, diabetes) more often than children without BHD. 1,2
Meanwhile, there is growing evidence that children with BHD are less likely to engage in
aerobic exercise/physical activity than their typically developing peers. 3,4
This can occur for
many reasons including exclusion from sports due to behavioral problems, comorbid sensory
issues, delayed motor skills, and anxiety. 4,5
Low engagement in aerobic activity is linked with
lower fitness, which may then further discourage exercise participation. 3
Low engagement in exercise is particularly troubling given these children’s increased risk
for chronic diseases and evidence that exercise may have cognitive, behavioral and emotional
benefits. Relatively short bouts have been shown to improve impulsivity and mood state among
typically developing children and those with single BHD such as attention deficit hyperactivity
disorder (ADHD), autism (ASD), or depression. 6-8
Intensity has also been linked to cognitive
effects in children; one study found that vigorous intensity exercise resulted in better executive
function outcomes than moderate/light intensity exercise. 9
These findings emphasize the importance of finding aerobic exercise modalities that
overcome the engagement challenges facing this population. However, this research has not
been translated into special education settings, or extended to examine effects in children with
heterogeneous BHD. Thus, in this study we examine if a cybercycling PE intervention, which
4
successfully engages children with BHD in aerobic exercise, 11
is linked to improvements in
behavioral self-regulation and classroom functioning relative to standard non-aerobic PE.
Methods
Setting and Participants
The study was conducted at a therapeutic day school affiliated with Harvard Medical
School. The school enrolls about 110 children each year in kindergarten through 10 th
grade year
with diagnosed BHD, many of whom have learning disabilities, but does not serve children with
intellectual disabilities.
Students are predominantly male and have multiple diagnoses; in an average school year,
about 40% of enrolled students are diagnosed with ASD, 60% with ADHD, 40% with an anxiety
disorder, and 30% with a mood disorder. There are 14 classrooms, each with a head teacher, an
assistant teacher, and a classroom counselor. Students are engaged with a variety of in-school
service providers including psychologists, occupational therapists, and speech pathologists.
Intervention Design
A 14-week crossover design was utilized. Children were randomly assigned by classroom
to receive the 7-week intervention during fall or spring, with a 10-week washout period between
treatment arms. A simple random number generator was used to assign 7 classes to the fall
treatment; the remainder served as the fall control group and received treatment in the spring.
Detailed information on the intervention and its development is published elsewhere. 10
The intervention, known as Manville Moves, featured a progressive and aerobically
challenging PE curriculum using virtual-reality exergaming stationary bicycles (cybercycles).
5
The curriculum overview is contained in Figure 1.1. Existing PE efforts at the school were not
successful at engaging the majority of children in extended bouts of aerobic exercise. Therefore,
the exercise modality was selected to optimally engage children with complex BHD who, again,
often face exercise engagement challenges such as sensory processing disorders, low fitness
levels, socialization challenges, and motor delays; the curriculum was designed to gradually
accustom participants to riding for extended durations and higher intensities.
During the intervention, children used the bikes 2 times per week during 30-40 minute
PE classes, starting at 10 minutes riding duration and building to over 20 minutes over the 7
week period. During the control period, children continued to participate in standard PE
programming (2 times/week x 30-40 mins). Standard PE is focused on games to build
socialization and team skills as well as motor skill acquisition through activities such as
basketball shooting; thus there are only short bouts of aerobic exercise and many students
struggle to remain engaged even with extensive staff attention.
The study protocol was reviewed/approved by the Harvard T.H. Chan School of Public
Health Institutional Review Board. The intervention was co-designed by school personnel and
the research team and implemented during the 2014-2015 academic year. The school elected to
integrate the intervention into school programming; thus an opt-out consent process was utilized.
Demographic/baseline data were obtained using an online caregiver survey after an active
consent process.
6
CURRICULUM WEEK #1
PE Class #1 PE Class #2
Minimum Duration Target Intensity Course Type Minimum Duration Target Intensity Course Type
10 Minutes Light to Moderate Flat None Choice Choice
CURRICULUM WEEK #2
PE Class #1 PE Class #2
Minimum Duration Target Intensity Course Type Minimum Duration Target Intensity Course Type
12 Minutes Light to Moderate Moderate Hills,
Flats 5 Minutes Choice Choice
CURRICULUM WEEK #3
PE Class #1 PE Class #2
Minimum Duration Target Intensity Course Type Minimum Duration Target Intensity Course Type
15 Minutes Moderate Rolling 7 Minutes Choice Choice
CURRICULUM WEEK #4
PE Class #1 PE Class #2
Minimum Duration Target Intensity Course Type Minimum Duration Target Intensity Course Type
18 Minutes Moderate to
Vigorous Varying Hills,
Flats 8 Minutes Choice Choice
CURRICULUM WEEK #5
PE Class #1 PE Class #2
Minimum Duration Target Intensity Course Type Minimum Duration Target Intensity Course Type
15 Minutes Moderate Mostly flat 10 Minutes Choice Choice
CURRICULUM WEEK #6
PE Class #1 PE Class #2
Minimum Duration Target Intensity Course Type Minimum Duration Target Intensity Course Type
20 Minutes Moderate to
Vigorous Hilly 5 Minutes Choice Choice
CURRICULUM WEEK #7
PE Class #1 PE Class #2
Minimum Duration Target Intensity Course Type Minimum Duration Target Intensity Course Type
15 Minutes Vigorous Time Trial None Choice Choice
Figure 1.1. Overview of Manville Moves cybercycling physical education curriculum.
7
Objectives
The study’s first research aim was to determine if students had improved classroom
functioning demonstrated through reduced disciplinary time out of class, as well as improved
self-regulation scores when participating in the intervention PE curriculum compared with
standard non-aerobic PE. Thus, the first aim was to assess the combined acute and chronic
effects of exercise, along with programmatic effects of the intervention, over the duration of 7-
week intervention compared with the 7-week control condition.
The second research aim was to specifically assess the acute effects of exercise by
examining behavioral changes only on days that children participated in the aerobic cybercycling
PE intervention compared with the control condition.
Exposure and Outcomes
Exercise exposure was measured using the data captured by the bicycles via student-
specific login codes. Data compiled for each riding bout included timestamp, average heart rate,
and minutes of riding. Student refusals to ride, as well as any occurrences of mechanical or
electrical failure, were documented on paper. It was not feasible to collect objective exercise
data for the control condition, since children would not tolerate wearing heart rate monitors or
accelerometers. However, the standard PE curriculum did not include programming targeting
aerobic exercise while children were in the control condition. Additional information on fidelity
of implementation and student engagement is described elsewhere. 11
Behavioral self-regulation was operationalized using the Conners Abbreviated Teacher
Rating Scale (CATRS-10), a validated screening instrument for behavioral problems related to
inattention, impulsivity/hyperactivity, and emotional lability. 12,13
Classroom counselors
8
completed the CATRS-10 at the end of each school day for each student. The instrument consists
of 10 statements regarding the child’s behavior rated on a 4-point Likert scale, with a total score
ranging from 0 to 30. 13
A score of 15 or higher has been the standard for screening children with
symptomatology at a level of clinical concern. 14-16
Equivalent screening thresholds were used for
the emotional lability subscale (4 questions, ≥6 out of possible 12) and impulsivity subscale (6
questions, ≥9 out of possible 18).
Because the classroom counselors accompanied students to PE, it was not possible to
blind them to participants’ treatment group assignment. However, the counselors were not
explicitly informed of the study objectives and received no incentives dependent on the
participants’ treatment group assignment. They were also prevented from viewing previous
recordings of CATRS-10 in order to help prevent manipulations of variability based on prior
measurements. The study coordinator for the school reviewed counselor reporting records on a
weekly basis to ensure compliance. A research assistant was also assigned to check CATRS-10
scores for variability on a bi-weekly basis; no statistically significantly decrease in reporting
variability was observed during the course of the study. In addition, no differences in reporting
patterns were observed on days when floating counselors filled in for classroom counselors who
were absent. Of note, the floating counselors did not attend PE sessions with the class and
therefore were not aware of their treatment status.
Classroom functioning was operationalized based on disciplinary time out of class
(TOC). When teachers determined a child must leave class due to unacceptable behavior,
counselors recorded the event using the mobile survey platform to enter the student identification
code and number of minutes for each TOC event as it occurred. This measure yields minutes of
TOC per day and number of TOC events per day for each student. Teachers follow a school-
9
wide policy when determining whether a child receives TOC and unlike counselors, were
blinded to treatment status. Because recording TOC has been a longstanding procedure, and any
TOC must be reported to parents, we feel that this measure is less vulnerable to subjective
interpretation or bias if treatment status was learned by teachers.
School clinicians established a priori thresholds for clinically relevant TOC per day that
constituted either a disruption to a student’s ability to learn classroom material (defined as 1 or
more events per day regardless of cumulative time or 10 or more minutes per day regardless of
total number of events), or prevented meaningful learning for that day (defined as 5 or more
events per day or 90 or more minutes per day).
Sample Size and Analytical Plan
Because exposures and outcomes were measured each day, the relevant unit of analysis
for the study is a child-day. A logistic-normal mixed effects regression model 17
was used to
assess relationships between intervention exposures and clinical thresholds of behavioral
outcomes accounting for both individual and classroom random effects. Outcome variables were
dichotomous indicators of whether or not a child exceeded clinical or screening thresholds for
either classroom functioning, measured by TOC, or self-regulation using the CATRS-10.
The maximum possible sample size was 109 based on school enrollment, and a priori
power calculations indicated that a sample of at least 75 students would provide power > 90% to
detect the small to moderate effect size (d=.3 to d=.5) in executive functioning evident in
literature. 6,18,19
Given nearly universal participation, daily outcome measurements, cross-over
design, and minimal attrition, the sample size was more than adequate. The number of child-days
varied by outcome and are described in the results section.
10
The primary models tested aim 1 (i.e., the overall treatment effect of the intervention) for
each outcome measure. Treatment status and treatment order were included in each model as
independent variables to evaluate overall treatment effects while accounting for potential
seasonality/contamination effects. The secondary models assessing aim 2 (i.e., the acute exercise
effects of the intervention) additionally included terms indicating whether a child participated in
the cybercycling PE class intervention on that day. Finally, the models were also run introducing
a dichotomous variable representing whether the students used the bikes in the virtual course
mode or the video gaming mode to test for potential exergaming effects. All statistical analyses
were conducted in STATA 13.1, using a two-tailed significance level of P = 0.05.
Results
Demographic characteristics
Figure 1.2 shows the CONSORT enrollment, randomization and attrition flow diagram.
The final enrollment was N=103 students; the fall intervention arm included n=51 students in 7
randomly assigned classrooms, and the spring intervention arm included n=52 students in the
remaining 7 classrooms. Baseline demographic data is contained in Table 1.1. Participants were
83.5% males, and ages ranged from 7 to 16 years of age with a mean of 11.8 years of age. No
adverse events occurred during the intervention.
11
Figure 1.2. Enrollment, randomization, and attrition flow diagram (CONSORT).
Total School Population
N=109
Students Excluded Due to
Medical Exemption
n=1
Final Sample Size
N=103 students
k=14 classrooms
(3 Lower, 7 Middle, 4 Upper)
Students Transferring Out
n=4
Spring Treatment Group
n=52 students
k=7 classrooms
(2 Lower, 3 Middle, 2 Upper)
Fall Treatment Group
n=51 students
k=7 classrooms
(1 Lower, 4 Middle, 2 Upper)
Student Inadvertently Omitted
from Data Recording System
n=1
12
Table 1.1: Baseline participant characteristics.
Overall
(N=103)
Male (N, %) 86 (83.5)
Age (mean, range) 11.8 (7-16)
Multiple Diagnoses (N, %) 1
46 (57.5)
Taking Medication (N, %) 1
39 (50.7)
Race/Ethnicity (N, %) 1,2
Black
White
Other
7 (8.8)
68 (85.0)
4 (5.0)
Eligibility Free/Reduced Cost Lunch (N, %) 1
23 (29.9)
Met Fitness Standards at Baseline (N, %) 3
11 (17.8)
1 Subset of students for whom a guardian returned a baseline survey (n=80).
2 One family declined to answer so categories do not add to 100%.
3 Subset of students who participated in baseline fitness testing (n=62).
Documentation of the exposure and outcome variables
There were 913 cybercycling PE days; average cycling duration in class was 16.1 (±5.3)
minutes, while average heart rate was 146.6 (±25.3) beats per minute. This indicates that
students achieved sustained aerobic exercise 20
during cybercycling PE classes on average.
For the classroom functioning outcome of TOC there were 6,419 observations recorded
over both intervention periods (fall n= 3318, �̅�=17.44±50.11; spring n= 3101, �̅�=17.48±54.18),
of which there were 2,122 instances of students having one or more removals from class in a day
(fall n=1205, spring n= 917). There were 5,252 observations of CATRS-10 scores recorded (fall
n= 2378, �̅�=9.03±7.02; spring n=2874, �̅�=8.50±6.52); of those, 955 exceeded the clinical
screening threshold for disruptive behavior (fall n=475, spring n=480). Treatment order was not
13
found to be significant in any of the models. Individual and classroom random effects are
controlled for in all analyses; children serve as their own controls and group level differences in
outcome variables between fall and spring do not affect the validity of analyses. Compared to
days in the control condition, percentages of children exceeding screening thresholds for overall
CATRS-10 score and both impulsivity and emotional lability sub-scores declined on days in the
intervention condition, and were lowest on days when the children participated in cybercycling
PE class (Figure 1.3). Post-hoc tests found no evidence of time of day effects for morning versus
afternoon PE classes.
Figure 1.3. Percentage of participants exceeding CATRS-10 screening thresholds as a function
of study condition.
0
5
10
15
20
25
Total CATRS-10 Score Impulsivity/Hyperactivity Sub- score
Emotional Lability Sub-score
% o
f S
tu d
e n
ts E
x ce
e d
in g
S cr
e e
n in
g C
u to
ff
Control Condition (Average of All Days)
Intervention Condition (Average of All Days)
Intervention Condition (Average of Cybercycling PE Days Only)
14
Models testing overall intervention effect (aim 1)
Results (Table 1.2, column 1) show clinically significant intervention effects. While in
the 7-week intervention, students experienced significantly reduced odds of exceeding screening
thresholds for total CATRS-10 score (OR=0.68, 95% CI: 0.57, 0.81), emotional lability sub-
score (OR=0.64, 95% CI: 0.52, 0.77), and impulsivity/hyperactivity sub-score (OR=0.49, 95%
CI: 0.36, 0.67), relative to when they participated in standard non-aerobic PE (i.e., the control
condition). Students also experienced significantly lower odds of having 5+ TOC events
(OR=0.54, 95% CI: 0.32, 0.91). The overall treatment effect was not significant for learning
disruptive TOC outcomes (1+ events, 10+ minutes) or preclusive to learning TOC minutes (90+
minutes).
15
Table 1.2: Overall intervention and cybercycling PE effects on behavioral outcomes compared
with control condition.
Outcome
Overall Intervention
Effect a
(Adjusted OR, CI)
Intervention Effect on
Cybercycling PE Days b
(Adjusted OR, CI)
(1) (2)
Exceeds Total CATRS-10 Screening
Threshold
0.68 (0.57, 0.81) 0.29 (0.14, 0.61)
Exceeds Impulsivity/Hyperactivity
Threshold
0.49 (0.36, 0.67) 0.28 (0.13, 0.59)
Exceeds Emotional Lability Threshold 0.64 (0.52, 0.77) 0.24 (0.11, 0.53)
1+ TOC events/day 1.04 (0.92, 1.18) 0.43 (0.26, 0.72)
5+ TOC events/day 0.54 (0.32, 0.91) 0.10 (0.02, 0.61)
10+ TOC minutes/day 1.03 (0.89, 1.18) 0.50 (0.29, 0.89)
90+ TOC minutes/day 1.04 (0.84, 1.31) 0.34 (0.14, 0.84)
a Adjusted for treatment order and accounting for random effect of individual and random effect
of classroom assignment. Odds relative to any day in the control condition. b Adjusted for treatment order, elective biking days, and non-adherent PE class days in the
treatment condition. Odds relative to non-biking days in the control condition.
TOC=Disciplinary Time Out of Class
OR=Odds Ratio
CI=Confidence Interval
Models testing acute exercise effects (aim 2)
The effects of acute exercise were significantly more pronounced than the overall
intervention effects (Table 1.2, column 2). On days that students participated in an intervention
PE class, they experienced clinically and statistically significantly reduced odds of exceeding
screening thresholds for total CATRS-10 score (OR=0.29, 95% CI: 0.14, 0.61), emotional
16
lability sub-score (OR=0.24, 95% CI: 0.11, 0.53), and impulsivity/hyperactivity sub-score
(OR=0.28, 95% CI: 0.13, 0.59), relative to the control condition. Acute effects of cybercycling
PE class also resulted in significantly reduced odds of both learning-disruptive and preclusive
TOC events (OR=0.43, CI:0.26, 0.72; OR=0.10, CI: 0.02, 0.61) and minutes (OR=0.50, CI: 0.29,
0.89; OR=0.34, CI: 0.14, 0.84). Treatment order and video gaming mode were not significant in
any of the models.
Discussion
This study provides compelling evidence that children and adolescents with multiple,
heterogeneous BHD in a school setting can successfully engage in and experience behavioral
benefits from an aerobic, cybercycling PE curriculum. Across the intervention period, odds that
children would display clinically disruptive behaviors including impulsivity and emotional
lability were 32% to 51% lower than during the control condition. These effects strengthened on
days when children participated in an intervention cybercycling class; here odds of disruptive
levels of behavioral dysregulation declined between 71% and 76% relative to the control
condition. Acute exercise led to significant declines in odds of receiving learning disruptive and
preclusive amounts of disciplinary TOC. These results build on previous research showing
positive effects of aerobic exercise on mood and impulsivity in children, and translate those
findings into a program implemented in a real world setting with children with multiple BHD.
The primary research aim was to determine whether intervention participation resulted in
improved behavioral outcomes for students; this was the case, however, effects were more
pronounced on days when the children participated in the structured, aerobic cybercycling PE
classes. So while there seem to be chronic exercise and programmatic effects of this intervention
17
on behavioral self-regulation and classroom functioning even on days when children did not
bike, acute exercise is the primary driver of the intervention effect. This finding is consistent
with prior studies and proposed mechanisms by which neuroendocrine and reticular-activating
systems affect mood and functioning in areas of the brain related to executive function and
impulse control. 21
Although it was impossible to blind counselors to the intervention condition, we feel risk
of bias in recording student behaviors due to knowledge of treatment status was low. Neither
teachers nor counselors were aware of the primary study hypotheses. Also, the outcome of
disciplinary TOC is determined by classroom teachers who were not aware of treatment
condition. Post-hoc tests for bias included comparison of CATRS-10 by blinded (floating) and
non-blinded counselors, which indicated no reporting differences, and examination of scores on
days students electively rode. If counselors were aware of the hypothesis/biased in reporting,
scores should have been lower for children on those days as they were for cybercycling PE days
since counselors accompany them to ride; in fact point estimates indicated worse scores than on
non-riding days (not statistically significant).
In addition to aerobic exercise, the cybercycling PE classes may hold several other
advantages over standard PE programming. They require fewer transitions, which are often
challenging to children with BHD. Also, the cybercycling PE allowed students to avoid peer
judgements of performance, since other students could not see their performance data unless they
share it. It was also less noisy and chaotic than standard PE classes. 11
However, it is important to
note that standard PE class may confer its own benefits, including motor skill acquisition, team
sports practice, and socialization. Cybercycles are also relatively expensive; since the key is
18
overcoming aerobic exercise engagement barriers, other modalities should be explored with
similar populations.
Because the intervention was implemented as part of school programming, participation
was nearly universal; therefore, selection biases that generally accompany active participant
recruitment and consent protocols were avoided. The generalizability of the results is enhanced
because participants had a wide variety of diagnoses, including complex comorbidities. Finally,
this study demonstrates strong ecological validity because the intervention was conducted within
existing school schedule and staffing. This improves potential for scale-up/dissemination in more
diverse settings catering to children with BHD.
Despite these strengths, the generalizability of the results is limited by the setting and
population targeted. Therapeutic day schools serve children who have been unsuccessful in
public school special education environments. Thus, findings from this study are limited to
students with substantial BHD. Future research could test the intervention in special education
programs in public schools is needed to determine if a broader group of children may benefit
from the program.
Despite overall cuts to PE class time allocation 22
, efforts are taking place across the
United States to allow movement in classrooms and facilitating improved learning and
behavior. 23
However, while children with BHD may most benefit from the effects of aerobic
exercise, they are least likely to be easily engaged in such activities. This study shows that a
cybercycling PE curriculum can successfully engage children with a variety of complex BHD in
high-quality aerobic exercise, and as a result, they experience significant improvements in
19
important behavioral measures. Critically, such a curriculum can successfully affect student
behavior within existing school programming with short durations and low frequency.
Acknowledgements
The authors would like to thank the students, families, and staff at the Manville School
for their participation in the implementation and evaluation of Manville Moves, in particular
Brian Wood, Bobby Hermesch, Jim Prince, and Amanda Hayes. We would also like to thank
Tom McCarthy for providing equipment technical support and expertise.
Funding Source
Harvard University philanthropic funding; TriROK Foundation grant; Ascend Capital
grant.
Financial Disclosure
All authors have indicated they have no financial relationships relevant to this article to
disclose.
Conflict of Interest
All authors have indicated they have no potential conflicts of interest to disclose.
Clinical Trials Registration
Clinical Trials Registry, United States National Institutes of Health. Clinical trials
number NCT02766101. Web link:
https://clinicaltrials.gov/ct2/show/NCT02766101?term=NCT02766101&rank=1
20
References
1. Perou R, Bitsko RH, Blumberg SJ, et al. Mental health surveillance among children—
United States, 2005–2011. MMWR Surveill Summ. 2013;62(Suppl 2):1-35.
2. Merikangas KR, He J-p, Burstein M, et al. Lifetime prevalence of mental disorders in US
adolescents: results from the National Comorbidity Survey Replication–Adolescent Supplement
(NCS-A). Journal of the American Academy of Child & Adolescent Psychiatry.
2010;49(10):980-989.
3. Rimmer JH, Rowland JL, Yamaki K. Obesity and secondary conditions in adolescents
with disabilities: Addressing the needs of an underserved population. Journal of Adolescent
Health. 2007;41(3):224-229.
4. Bandini LG, Curtin C, Hamad C, Tybor DJ, Must A. Prevalence of overweight in
children with developmental disorders in the continuous national health and nutrition
examination survey (NHANES) 1999-2002. J Pediatr. 2005;146(6):738-743.
5. Mangerud WL, Bjerkeset O, Lydersen S, Indredavik MS. Physical activity in adolescents
with psychiatric disorders and in the general population. Child and adolescent psychiatry and
mental health. 2014;8(1):2.
6. Best JR. Effects of physical activity on children’s executive function: Contributions of
experimental research on aerobic exercise. Developmental Review. 2010;30(4):331-351.
7. Hansen CJ, Stevens LC, Coast JR. Exercise duration and mood state: How much is
enough to feel better? Health Psychology. 2001;20(4):267-275.
21
8. Pontifex MB, Saliba BJ, Raine LB, Picchietti DL, Hillman CH. Exercise improves
behavioral, neurocognitive, and scholastic performance in children with attention-
deficit/hyperactivity disorder. The Journal of pediatrics. 2013;162(3):543-551.
9. Castelli DM, Hillman CH, Hirsch J, Hirsch A, Drollette E. FIT Kids: Time in target heart
zone and cognitive performance. Preventive Medicine. 2011;52, Supplement:S55-S59.
10. Davison KK, Bowling A, Garcia J, Wood B, Hermesch R, Prince J, Hayes A, Kow R,
Newlan S, Slavet J. A cybercycling intervention to improve behavioral regulation and classroom
functioning among children with behavioral health disorders: Pragmatic randomized trial design
for Manville Moves. Contemporary Clinical Trials. Contemporary Clinical Trials. 2016; 49:40-
46.
11. Bowling AB, Slavet J, Garcia J, Wood B, Miller D, Hermesch R, Kow R, Newlan S,
Davison KK. Implementation Fidelity of a Cybercycling Curriculum among Children with
Behavioral Health Disorders. Translational Journal of the American Council on Sports
Medicine. In press.
12. Conners C. Rating scales for use in drug studies with children Psychopharmacology
Bulletin, special issue on pharmacotherapy of children. 1973.
13. Conners C. Manual for Conners’ rating scales. North Tonawanda, NY: Multi-Health
Systems. 1989.
14. Holmberg K, Sundelin C, Hjern A. Screening for attention‐deficit/hyperactivity disorder
(ADHD): can high‐risk children be identified in first grade? Child: care, health and
development. 2013;39(2):268-276.
22
15. Jones K, Daley D, Hutchings J, Bywater T, Eames C. Efficacy of the Incredible Years
Programme as an early intervention for children with conduct problems and ADHD: long‐term
follow‐up. Child: care, health and development. 2008;34(3):380-390.
16. Rowe KS, Rowe KJ. Norms for parental ratings on Conners' Abbreviated Parent-Teacher
Questionnaire: implications for the design of behavioral rating inventories and analyses of data
derived from them. Journal of Abnormal Child Psychology. 1997;25(6):425-451.
17. Fitzmaurice GM, Laird NM, Ware JH. Applied longitudinal analysis. Vol 998: John
Wiley & Sons; 2012.
18. Biddle SJH, Asare M. Physical activity and mental health in children and adolescents: a
review of reviews. British Journal of Sports Medicine. 2011;45(11):886-895.
19. Tomporowski PD. Cognitive and behavioral responses to acute exercise in youths: A
review. Pediatric Exercise Science. 2003;15(4):348-359.
20. Washington RL, van Gundy JC, Cohen C, Sondheimer HM, Wolfe RR. Normal aerobic
and anaerobic exercise data for North American school-age children. The Journal of pediatrics.
1988;112(2):223-233.
21. McMorris T, Tomporowski PD, Audiffren M. Exercise and cognitive function. Wiley
Online Library; 2009.
22. Taber DR, Chriqui JF, Perna FM, Powell LM, Slater SJ, Chaloupka FJ. Association
between state physical education (PE) requirements and PE participation, physical activity, and
body mass index change. Preventive medicine. 2013;57(5):629-633.
23
23. Lonsdale C, Rosenkranz RR, Peralta LR, Bennie A, Fahey P, Lubans DR. A systematic
review and meta-analysis of interventions designed to increase moderate-to-vigorous physical
activity in school physical education lessons. Preventive Medicine. 2013;56(2):152-161.
24
Chapter 2:
Dose-Response Effects of Exercise on Behavioral Health in Children and Adolescents
April Bowling, James Slavet, Daniel P. Miller, Sebastien Haneuse,
William Beardslee, Kirsten Davison
25
Abstract
Purpose: Aerobic exercise appears to positively affect behavioral health in children. However,
little research has been conducted on dose-response effects among those with behavioral health
disorders (BHD). This study uses data collected from an RCT to test the effects of cybercycling
on behavioral outcomes in children with BHD attending a therapeutic school. We examine dose-
response relationships between duration and intensity of cybercycling and minutes of
disciplinary time spent out of class and self-regulation scores; additionally we examine potential
effect modification by Attention-Deficit/Hyperactivity Disorder (ADHD) diagnosis.
Methods: A 14-week crossover design was used. Children (N=103, 83.5% male, age 11.8±2.3)
were randomly assigned by classroom to receive the 7-week intervention during fall or spring,
during which they used the bikes 2 times per week in physical education class. Real-time data on
exercise duration was collected via the bicycles. The Conners Abbreviated Teacher Rating Scale
for self-regulation and minutes of disciplinary time out of class (TOC) were recorded daily for
each child. Ride duration and average heart rate were treated as continuous predictors of
outcomes using mixed effects linear regression.
Results: For every 10 minutes of riding, children had an associated decline of 10.7 minutes of
TOC (p<0.001) and 1.2 points improvement in self-regulation score (p<0.001). For each increase
of 10 beats per minute average heart rate children had an associated decline of 1.3 minutes
(p=0.05) and 0.21 points (p<0.05). The effect of duration on behavioral outcomes was
significantly modified by ADHD diagnosis.
26
Conclusion: Both duration and intensity of aerobic exercise appeared to have significant, linear
relationships with improved behavioral outcomes among children with a variety of BHD;
children with ADHD may experience the greatest benefits of riding for longer durations.
27
Introduction
Childhood behavioral health disorders are common and increasing in prevalence in the
United States and many other countries (15, 18). While the neurodevelopmental diagnosis
Attention- Deficit/Hyperactivity Disorder (ADHD) is one of the most common, behavioral health
diagnoses in children also include autism spectrum disorders, depression, bipolar, and disruptive
behavior disorders among many others (17). According to the Center for Disease Control’s
most recent report on pediatric mental health surveillance, 13%–20% of children living in the
United States experience a diagnosable behavioral health disorder in a given year, and lifetime
rates of comorbid mental health diagnoses may exceed 40% (17).
There is a growing body of evidence that exercise exposures, particularly bouts of aerobic
exercise, may positively affect behavior, mood and cognition in children both with and without
behavioral health disorders (12). A recent systematic review of 8 randomized controlled trials
found generally positive, if somewhat weak effects of exercise on psychosocial functioning in
children (12). A meta-analysis of 19 studies investigating effects of physical activity on
executive function in children, adolescents and adults found a significant moderate effect size for
acute exercise exposures (23).
While a comprehensive review of pathways is beyond the scope of this article, research
suggests that acute aerobic exercise affects neurotransmitter secretion and reuptake and is
associated with changes in neural activation patterns, while chronic exercise appears to promote
neurogenesis (14, 22). Although studies are sparse, there is also evidence that children with
ADHD may selectively experience certain benefits to executive function and decreased
impulsivity from exercise (19, 20). This would make sense, given that exercise affects the same
28
catecholamine pathways targeted by stimulant medication to produce attentional and behavioral
improvements.
Unfortunately, there is also mounting evidence that like children with other types of
disabilities, those with behavioral health disorders are less likely to engage in aerobic exercise
and/or moderate to vigorous physical activity (MVPA; hereafter simply described as exercise)
and thus are less likely to experience these potential benefits (13). A variety of barriers to
exercise have been documented in these populations including exclusion from traditional
sporting leagues due to behavioral problems, oppositional responses to attempts by parents to
manage health behaviors, sensory disorders and anxiety that discourage exercise engagement,
comorbid gross motor delays, and weight gain and perceived exertion changes associated with
certain psychiatric medications (21).
Given these challenging barriers to exercise engagement, it is crucial to examine whether
exposures of shorter duration and lower intensity improve behavioral health outcomes in order to
improve both efficacy and feasibility of interventions and exercise prescriptions. While many
studies have shown overall treatment effects on behavioral outcomes from specific exercise and
physical activity interventions among children with behavioral health disorders, effects have
varied significantly across different treatment modalities and intervention approaches (1, 3, 10).
For example, Davis et al. (7) and Hillman et al. (11) both used 20 minute bouts of MVPA
and found significant effects on self-regulation among typically developing children. Among
children with ADHD, Gawrilow et al. (9) found improvements to response inhibition after acute
trampoline jumping bouts of only 5 minutes, Pontifex et al. (20) found improvements to a variety
or neurocognitive and executive function measures after 20 minutes of MVPA on a treadmill,
29
and Chang et al. (4) found similar results after 30 minutes of running. In contrast, Oriel et al.
(16) found that 15 minutes of running was not associated with improved classroom functioning
in children with autism spectrum disorders. We are not aware of other studies of exercise and
classroom behavioral outcomes that used exercise exposures of less than 45 minutes, a difficult
amount to achieve given the barriers to exercise engagement in this population.
Also of importance, most current research has primarily focused on typically developing
children or those with a single behavioral health diagnosis, and thus lacks generalizability to
children with multiple behavioral health challenges. This also contributes to the existing
inability to advise parents and clinicians whether shorter and less intense exercise exposures
elicit significant behavioral improvements for diverse populations, despite a demand for such
knowledge (26). This lack of understanding of real-world dose-response relationships
constitutes a critical evidence gap that negatively affects the treatment of these disorders and
inhibits the design of effective programming in therapeutic schools, special education classrooms
and clinical settings.
This article reports an exploratory follow-up investigation of the dose-response
relationship between acute exercise duration and intensity and behavioral outcomes post-exercise
among children with heterogeneous behavioral health disorders who participated in the Manville
School Cybercycling and Behavior Randomized Controlled Trial (Manville Moves). Manville
Moves was designed to examine if an aerobic cybercycling PE intervention was linked to
improvements in behavioral self-regulation and classroom functioning relative a control
condition among 103 children with behavioral health disorders. The RCT found that
participation in the intervention resulted in greatly reduced clinically disruptive behaviors and
30
learning disruptive disciplinary time out of class among children compared to the control
condition (reference pending, under review at Pediatrics).
Building on these results, in the current study we examine relationships between
behavioral outcomes and continuous measures of exercise duration and intensity. Based on
previous research, we hypothesized that both increased exercise duration and intensity would
have inverse relationships with negative behaviors. Also, while Manville Moves was not
designed or powered to examine modification by individual child behavioral health
characteristics, it provides an opportunity to conduct hypotheses-generating analyses into
possible effect modification of exercise-behavior relationships by children’s ADHD diagnosis
status, which as described previously, may plausibly alter neurological and endocrine responses
to exercise.
Methods
Setting and Participants
The study design, fidelity of implementation, and treatment effects of the Manville
Moves RCT has been described extensively in other publications (2, 8). A total of 103 children
attending a therapeutic day school participated in the original study, from which this
investigation draws its data. Participants ranged from 7 to 16 years of age, and all had at least
one diagnosed behavioral health disorder, with the most prevalent diagnoses being ADHD,
autism and anxiety. The study protocol was reviewed and approved by the Harvard T.H. Chan
School of Public Health Institutional Review Board. The school integrated the PE curriculum
into school programming; thus, the study used an opt-out consent process, and all children
enrolled in the school participated in the study with the exception of one child who had a medical
31
exemption from PE class. Following an active consent process, caretakers of the participants
filled out a survey, offered both online and via hard copy, which collected data on their child’s
baseline physical activity levels, diagnoses, medication types, and family demographics.
Surveys were returned for 80 (77.7%) participants.
Design and Procedures
The RCT utilized a crossover, group-randomized controlled design. Children were
randomly assigned by classroom to receive the 7-week intervention during fall or spring, with a
10-week washout period between treatment arms. The intervention featured an aerobically
challenging PE curriculum using virtual-reality exergaming stationary bicycles (cybercycles).
During the intervention, children used the bikes 2 times per week during 30-40 minute PE
classes. Figure 1.1 (Chapter 1) contains an overview of the curriculum. Because of the inclusion
of “choice days” in the curriculum, which had very short or no minimum riding duration or
intensity targets, there was not a floor effect on riding exposures presented by the curriculum.
This study utilizes the exercise exposure and behavioral outcome data measured on cybercycling
PE days during the RCT. Thus, while the original study utilized cross-over control groups, as a
dose-response investigation the current study only utilizes data from the intervention groups.
Exercise Duration and Intensity
Data on continuous measures of exercise duration (minutes of riding) and intensity
(average HR) were collected via the cybercycles during PE class (Expresso HD Upright Bike™,
Interactive Fitness, Santa Clara, CA) using unique login codes for each participant. Participants
entered their codes prior to each ride under the direction of staff, and data were directly uploaded
into the database. HR was measured continuously via handlebar sensors and recorded as an
32
average for each riding bout. While the preferred measurement approach would have been the
use of chest-strap HR monitors, these were not tolerated by the study population, and thus the
handlebar sensors were used instead.
Behavioral Outcomes
Behavioral self-regulation score (SRS): Classroom counselors completed the Conners
Abbreviated Teacher Rating Scale – 10 Item at the end of each school day for each student. This
instrument is a commonly used, well-validated measurement instrument for behavioral problems
related to hyperactivity/impulsivity and emotional lability, and is utilized to screen for ADHD.(6,
24) Higher scores on the Conners scale indicate worse behavior problems. Over the 14-week
study, 5,250 Conners ratings were recorded using a mobile survey platform.
Disciplinary Time Out of Class (TOC): Time spent out of class is negatively related to
learning outcomes, and was considered an important indicator of overall classroom behavior and
functioning.(8) Staff recorded all events where students were asked to leave the academic
classroom by a teacher due to unacceptable behavior, including the number of minutes that the
child was out of the classroom. Classroom counselors used a mobile survey platform to enter the
student identification code and number of minutes for each disciplinary event as it occurred.
Over the 14-week study, 6,489 observations of TOC were recorded, including one recording of
zero minutes for each day that a student did not have time out of class.
Statistical Analyses
Analyses examined the effect of exercise duration and intensity on behavioral self-
regulation and classroom functioning. As previously noted, exercise duration and intensity were
modeled as continuous predictors of the outcomes. Multilevel mixed-effect linear regression
33
was conducted to assess relationships between behavioral outcomes and both duration of
exercise and intensity of exercise while accounting for both random individual and classroom
effects, since individuals consistently biked within their assigned class.
Separate models were constructed for each outcome and included both exposure types to
control for covariation of duration and intensity. The primary models were unadjusted (model
1). Models were tested for non-linear effects, and secondary analyses were also conducted to
test for effects of time and increased fitness over time on relationships between exercise
exposure and behavior. This was accomplished by including an interaction term between time
(week of study) and exercise exposure. Models were then adjusted for demographic and health
characteristics of children including race, sex, age, ADHD diagnosis, and psychiatric medication
use (model 2). Finally, models were examined for effect modification by ADHD (model 3). All
statistical analyses were conducted in STATA 13.1, using a two-tailed significance level of P =
0.05.
Results
Participants’ engagement in aerobic exercise through cybercycling PE classes during the
intervention is detailed extensively elsewhere (2). There were N=103 participants who took part
in both the control and the intervention condition; demographic and health characteristics at
baseline are shown in Table 2.1. Over 779 PE class riding bouts, the average cycling duration in
class was 16.1 (±5.3) minutes, while average heart rate was 146.6 (±25.3) beats per minute
(approximately 77% of theoretical maximum threshold based on studies in similar populations)
(3). Modeling results showed no evidence of non-linearity, and there was no significant effect of
fitness on behavioral outcomes for either exercise duration (SRS: β=-0.03, 95%CI: -0.16 – 0.10;
34
TOC: β=-0.84, 95%CI: -1.86 – 0.18) or intensity (SRS: β=0.07, 95%CI: -0.05 – 0.18; TOC: β=-
0.55, 95% CI: -1.49 – 0.39). Thus fitness effect (i.e., week of study) was omitted from the final
models.
Table 2.1. Description of baseline participant characteristics.
Overall
(N=103)
Male (N, %) 86 (83.5)
Age (mean, range) 11.8 (7-16)
Multiple Diagnoses (N, %) 1
46 (57.5)
ADHD Diagnoses (N, %) 1,2
Taking Medication
Not Medicated
46 (57.5)
24 (52.2)
22 (47.8)
Taking Medication (N, %) 1 39 (50.7)
Race/Ethnicity (N, %) 1,3
Black
White
Other
7 (8.8)
68 (85.0)
4 (5.0)
1 Subset of students for whom a guardian returned a baseline survey (n=80)
2 Either singular or comorbid diagnosis
3 One family declined to answer so categories do not add to 100%
35
Dose-Response Relationship Modeling
Multilevel mixed-effects linear regression results are shown in Table 2.2. In both the
crude (model 1) and adjusted models (model 2), minutes of riding had a significant, inverse
linear relationship with both self-regulation score and disciplinary time out of class minutes. In
the adjusted model, holding average HR constant, each additional 10 minutes of riding were
associated with an 10.7 minute reduction in disciplinary time out of class (95%CI: -16.8, -5.2)
and 1.2 (95%CI: -1.9, -0.5) point decrease in self-regulation score.
Intensity showed a similar, but weaker association with outcomes. In the adjusted model,
each additional 10 bpm was associated with a 1.3 minute decline in disciplinary time out of class
(95%CI: -2.7, 0.0) and a 0.2 (95%CI: -0.4, 0.0) point decrease in self-regulation score.
36
Table 2.2. Regression modeling results for continuous exercise exposures and measures of
behavioral outcomes.
Model 1
Unadjusted
(N=103)
Model 2
Adjusted
(N=80)
Model 3 Effect
Modification by
ADHD
(N=80)
Outcome: Self-regulation Score
β (95% CI)
Per 10 Minutes of
riding 1 (duration)
-1.06 (-1.74, -0.40) -1.18 (-1.90, -0.46) -0.50 (-1.50, 0.50)
Per 10 BPM Average
HR (intensity) -0.14 (-0.30, 0.02) -0.21 (-0.40, -0.02) -0.31 (-0.65, 0.04)
Week of Study (time) -- 0.23 (0.06, 0.41) 0.24 (0.07, 0.42)
Race (Reference: White)
African
American -- 1.88 (-1.30, 5.05) 1.23 (-1.50, 4.95)
Other -- -0.79 (-7.81, 6.24) -0.87 (-8.00, 6.26)
Male -- 0.66 (-1.77, 3.08) 0.44 (-2.03, 2.91)
Age
-- -0.48 (-0.98, 0.03) -0.49 (-0.98, 0.02)
Has ADHD
-- 0.63 (-1.30, 2.55) 0.41 (-6.21, 7.04)
On Medication
-- -0.22 (-2.02, 1.57) -0.32 (-2.14, 1.50)
ADHD*Duration -- -- -1.37 (-2.76, 0.03)
ADHD*Intensity -- -- 0.15 (-0.25, 0.55)
Intercept
11.89 (9.08, 14.70) 16.39 (8.78, 23.99) 17.12 (8.03, 26.21)
37
Table 2.2 (continued). Regression modeling results for continuous exercise exposures and
measures of behavioral outcomes.
Outcome: Minutes of Disciplinary Time Spent Out of Class
β (95% CI)
Per 10 Minutes of
riding (duration) -8.18 (-13.36, -3.01) -10.69 (-16.78, -5.21) -4.41 (-11.91, 3.10)
Per 10 BPM Average
HR (intensity) -1.15 (-2.32, 0.02) -1.32 (-2.65, 0.02) -1.10 (-3.47, 1.27)
Week of Study (time) -- 1.08 (-0.29, 2.45) 1.18 (-0.18, 2.55)
Race
(White=reference)
African
American -- 1.75 (-13.99, 17.48) 0.76 (-15.14, 16.66)
Other -- -14.63 (-47.42, 18.17) -16.24 (-49.38, 16.89)
Male
-- 11.23 (-0.49, 22.95) 9.94 (-2.01, 21.90)
Age -- 1.27 (-1.02, 3.55) 1.28 (-1.04, 3.61)
Has ADHD -- 4.05 (-5.38, 13.49) 26.50 (-19.02, 72.01)
On Medication
-- 1.16 (-7.67, 9.49) -0.08 (-9.02, 8.86)
ADHD*Duration -- -- -12.91 (-23.46, -2.37)
ADHD*Intensity -- -- -0.21 (-3.01, 2.58)
Intercept 40.10 (20.92, 59.24) 14.31 (-25.34, 53.97) 2.45 (-49.89, 54.78) 1 No children had zero minutes of riding in PE classes, therefore, duration is rescaled to reflect
the minimum riding time of 1.67 minutes.
38
Effect Modification Modeling
The models showed evidence of effect modification of the association between
continuous minutes of riding and both self-regulation score and minutes of disciplinary time out
of class by ADHD diagnosis (Table 2.2, Model 3). The main effect of riding duration was no
longer significant in either case, but children with ADHD experienced 12.9 minutes (95%CI: -
23.5, -2.4) less time out of class and a self-regulation score improvement of 1.4 points (-4.7%,
95%CI: -2.8, 0.0) for each 10 additional minutes of riding. ADHD diagnosis did not appear to
modify the relationship between intensity and outcomes. A graphical depiction of the
modification of the relationship between exercise duration and outcomes by ADHD diagnosis is
shown in Figures 2.1 (TOC) and 2.2 (self-regulation score). Three-way interactions testing
differences in exposure-outcome relationships between children with ADHD both taking and not
taking medication were not significant in any of the models.
39
Figure 2.1. Effect modification of the relationship between exercise duration and disciplinary
time out of class (TOC) by ADHD status.
Figure 2.2. Effect modification of the relationship between exercise duration and self-regulation
score (Conners Abbreviated Teacher Rating Scale – 10 Item) by ADHD status.
40
Discussion
Few studies to date have examined dose-response relationships between exercise duration
and intensity and self-regulation in children. What is more, to our knowledge no studies have
been conducted among children with heterogeneous behavioral health disorders in real-world
settings. This exploratory follow-up assessment of the Manville Moves RCT showed significant,
inverse linear relationships between duration and intensity of cybercycling during PE classes and
poor self-regulation scores, and between duration of exercise and minutes of disciplinary time
out of class experienced outside of PE classes.
From a clinical standpoint, effects were more pronounced for exercise duration than for
intensity. Dose-response models showed that each additional 10 minutes of riding was linked to
more than 10 minutes less disciplinary time out of class, and a 3.9% reduction in self-regulation
score, an indicator of hyperactivity, impulsivity and emotional lability. It is worth noting again
that the amount of time that children spent riding did not affect the amount of time available for
them to be asked to leave the academic classroom; children had PE class for the same amount of
time regardless of how long they spent riding the cybercycles during that PE class.
Importantly, the effect modification models showed that the main effect of exercise
duration on both outcomes was mainly driven by children with ADHD, but this was not the case
for intensity. After controlling for medication status, children with ADHD experienced almost
13 minutes less disciplinary time out of class and a 4.7% improvement in self-regulation score
for every 10 additional minutes of riding, although the latter was only borderline significant,
likely due to a lack of power. That modification was not present for exercise intensity, which
41
remained borderline significant and strengthened in its association with improved self-
regulations scores.
There are a variety of plausible explanations for these effect modification findings. One
is that the physiological effects of acute exercise have been shown to influence the same
catecholaminergic systems that stimulant medications for ADHD target to improve attention and
decrease impulsivity (25). In this study, it appeared that improvements in self-regulation score
among children with ADHD were primarily driven by the amount of time that they spent
exercising, not the intensity at which they exercised. There are a number of possible
explanations for the lack of effect of exercise intensity. It may reflect the low baseline fitness of
this population, which might limit variation in intensity more than might be the case in adults or
typically developing children. It may also reflect the inherent limitations of using average HR as
a measure of intensity. Additionally, we cannot rule out the possibility that it reflects increased
measurement error inherent with hand-grip based HR monitoring. Future research seeking to
explore these interactions could utilize objective measurement of catecholamine secretion with
exercise of differing duration and intensity in children with ADHD in controlled laboratory
settings using gold standard intensity measurement instruments.
A limitation of this study is the lack of data on types of medication used, which prevents
evaluation of effect modification by medication class. However, children with ADHD diagnoses
whose parents reported they were talking medication did not experience different exercise effects
when three way interactions were tested. In addition, 72% of children with an ADHD diagnosis
in this sample had at least one comorbid behavioral health disorder, and effect modification by
differing constellations of diagnoses could not be examined due to small cell sizes. However,
while effects of exercise may differ across children with ADHD who also have comorbidities,
42
we show that the common diagnosis of ADHD is a significant effect modifier of outcomes,
despite any such variation. Future research should focus on quantifying effects of aerobic
exercise duration in combination with differing doses of stimulant medication in children with
ADHD, in order to understand the feasibility and limitations of using exercise as a primary
therapeutic approach for managing ADHD symptoms.
This study takes important first steps in identifying potential dose-response relationships
between exercise duration and intensity and behavioral health outcomes in children with
behavioral health disorders. Our findings indicate that aerobic exercise of even relatively short
duration and moderate intensity is likely behaviorally beneficial to children with a variety of
behavioral health disorders, and that there is a particularly a strong positive dose-response
relationship between duration of aerobic exercise and self-regulation and classroom functioning
among children with ADHD.
Acknowledgements
The authors would like to thank the students, families, and staff at the Manville School
for their participation in the implementation and evaluation of Manville Moves, in particular
Brian Wood, Bobby Hermesch, Jim Prince, and Amanda Hayes. We would also like to thank
Tom McCarthy for providing equipment technical support and expertise, as well as Richard
Kow, Sami Newlan, and Jeanette Garcia for their assistance with the data collection and review
process. This study was supported by a private grant to Harvard T.H. Chan School of Public
Health.
43
Conflict of interest
The authors declare that they have no conflict of interest. The results of the study are
presented clearly, honestly, and without fabrication, falsification, or inappropriate data
manipulation.
References
1. Best JR. Effects of physical activity on children’s executive function: Contributions of
experimental research on aerobic exercise. Developmental Review. 2010;30(4):331-51. doi:
http://dx.doi.org/10.1016/j.dr.2010.08.001.
2. Bowling AB, Slavet J, Miller, DP, et al. Cybercycling effects on classroom behavior in
children with behavioral health disorders: an RCT. Pediatrics. In Press.
3. Castelli DM, Hillman CH, Hirsch J, Hirsch A, Drollette E. FIT Kids: Time in target heart
zone and cognitive performance. Preventive Medicine. 2011;52, Supplement:S55-S9. doi:
http://dx.doi.org/10.1016/j.ypmed.2011.01.019.
4. Chang Y-K, Liu S, Yu H-H, Lee Y-H. Effect of acute exercise on executive function in
children with attention deficit hyperactivity disorder. Archives of Clinical Neuropsychology.
2012;27(2):225-37.
5. Conners C. Manual for Conners’ rating scales. North Tonawanda, NY: Multi-Health
Systems. 1989.
6. Conners C. Rating scales for use in drug studies with children Psychopharmacology
Bulletin, special issue on pharmacotherapy of children. 1973; 3:217.
44
7. Davis CL, Tomporowski PD, McDowell JE, et al. Exercise improves executive function
and achievement and alters brain activation in overweight children: a randomized, controlled
trial. Health Psychology. 2011;30(1):91.
8. Davison K, Bowling A, Garcia J, et al. A cybercycling intervention to improve
behavioral regulation and classroom functioning among children with behavioral health
disorders: Pragmatic randomized trial design for Manville Moves. Contemporary clinical trials.
2016; 49:40-46.
9. Gawrilow C, Stadler G, Langguth N, Naumann A, Boeck A. Physical Activity, Affect,
and Cognition in Children With Symptoms of ADHD. Journal of attention disorders.
2016;20(2):151-162. doi: 10.1177/1087054713493318. PubMed PMID: 23893534.
10. Hansen CJ, Stevens LC, Coast JR. Exercise duration and mood state: How much is
enough to feel better? Health Psychology. 2001;20(4):267-75. doi: 10.1037/0278-6133.20.4.267.
11. Hillman CH, Pontifex MB, Raine LB, Castelli DM, Hall EE, Kramer AF. The effect of
acute treadmill walking on cognitive control and academic achievement in preadolescent
children. Neuroscience. 2009;159(3):1044-54.
12. Lees C. Effect of aerobic exercise on cognition, academic achievement, and psychosocial
function in children: a systematic review of randomized control trials. Preventing chronic
disease. 2013;10:E174.
13. Mangerud WL, Bjerkeset O, Lydersen S, Indredavik MS. Physical activity in adolescents
with psychiatric disorders and in the general population. Child and adolescent psychiatry and
mental health. 2014;8(1):2.
45
14. McMorris T, Tomporowski PD, Audiffren M. Exercise and cognitive function.
Chichester, UK ;Hoboken, NJ: Wiley-Heinrich; 2009. xii, 377 p. p.
15. Merikangas KR, He J-p, Burstein M, et al. Lifetime prevalence of mental disorders in US
adolescents: results from the National Comorbidity Survey Replication–Adolescent Supplement
(NCS-A). Journal of the American Academy of Child & Adolescent Psychiatry.
2010;49(10):980-9.
16. Oriel KN, George CL, Peckus R, Semon A. The effects of aerobic exercise on academic
engagement in young children with autism spectrum disorder. Pediatric physical therapy : the
official publication of the Section on Pediatrics of the American Physical Therapy Association.
2011;23(2):187-93. Epub 2011/05/10. doi: 10.1097/PEP.0b013e318218f149. PubMed PMID:
21552085.
17. Perou R, Bitsko RH, Blumberg SJ, et al. Mental health surveillance among children—
United States, 2005–2011. MMWR Surveill Summ. 2013;62(Suppl 2):1-35.
18. Polanczyk GV, Salum GA, Sugaya LS, Caye A, Rohde LA. Annual Research Review: A
meta‐analysis of the worldwide prevalence of mental disorders in children and adolescents.
Journal of Child Psychology and Psychiatry. 2015;56(3):345-65.
19. Pontifex MB. Transient modulations of inhibitory control in children with ADHD: The
effect of a single bout of physical activity. University of Illinois at Urbana-Champaign. US:
ProQuest Information & Learning; 2013. 103 p. p.
20. Pontifex MB, Saliba BJ, Raine LB, Picchietti DL, Hillman CH. Exercise improves
behavioral, neurocognitive, and scholastic performance in children with attention-
46
deficit/hyperactivity disorder. The Journal of Pediatrics. 2013;162(3):543-51. doi:
10.1016/j.jpeds.2012.08.036. PubMed PMID: 2013-06916-035. PMID: 23084704. First Author
& Affiliation: Pontifex, Matthew B.
21. Rimmer JH, Rowland JL, Yamaki K. Obesity and secondary conditions in adolescents
with disabilities: Addressing the needs of an underserved population. Journal of Adolescent
Health. 2007;41(3):224-9.
22. Tomporowski PD, Davis CL, Miller PH, Naglieri JA. Exercise and children’s
intelligence, cognition, and academic achievement. Educational psychology review.
2008;20(2):111-31.
23. Verburgh L, Königs M, Scherder EJ, Oosterlaan J. Physical exercise and executive
functions in preadolescent children, adolescents and young adults: a meta-analysis. British
journal of sports medicine. 2013:bjsports-2012-091441.
24. Westerlund J, Ek U, Holmberg K, Näswall K, Fernell E. The Conners' 10‐item scale:
findings in a total population of Swedish 10–11‐year‐old children. Acta Paediatrica.
2009;98(5):828-33.
25. Wigal SB, Emmerson N, Gehricke J-G, Galassetti P. Exercise: Applications to childhood
ADHD. Journal of Attention Disorders. 2013;17(4):279-90. doi: 10.1177/1087054712454192.
PubMed PMID: 2013-12985-001. PMID: 22863768. First Author & Affiliation: Wigal, Sharon
B.
26. Williams J, Klinepeter K, Palmes G, Pulley A, Foy JM. Diagnosis and treatment of
behavioral health disorders in pediatric practice. Pediatrics. 2004;114(3):601-6.
47
Chapter 3:
ADHD Medication, Dietary Patterns, Physical Activity, and BMI in Children:
A Longitudinal Analysis of the ECLS-K Study
April Bowling, Kirsten Davison, Sebastien Haneuse, William Beardslee, Daniel P. Miller
48
Abstract
Purpose: We examined relationships between ADHD, stimulant use and BMI change in a
nationally-representative cohort of children, as well as differences in diet/physical activity that
may mediate associations between stimulant use and BMI change.
Methods: Using the Early Childhood Longitudinal Study-Kindergarten cohort 1998-1999
(N=8250), we modeled BMI/z-score change by ADHD and stimulant start time, odds of
unhealthy diet/physical activity predicted by ADHD and stimulant use, and performed mediation
analysis assessing indirect effects of health behaviors.
Results: Early stimulant use predicted short-term BMI reductions, but any stimulant use
predicted increased BMI growth between 5 th
and 8 th
grade. Children with ADHD had higher
odds of poor diet regardless of medication. Health behaviors did not mediate associations
between stimulants and BMI change.
Conclusions: Stimulant use predicted higher BMI trajectory between 5 th
and 8 th
grade but not
improved health behaviors. Future research should explore potential mechanisms by which
early/long-term stimulant use may affect metabolism.
49
Background and Rationale
The Centers for Disease Control and Prevention estimates that about 5.1 million children
were diagnosed with Attention Deficit/Hyperactivity Disorder (ADHD) in 2011, representing
about 9% of all children between the ages of 4 and 17 in the United States (Visser et al.,
2014). Of those diagnosed, approximately 3.5 million (69%) were taking at least one
medication, usually a stimulant, intended to treat ADHD symptoms (Visser et al.,
2014). Stimulant use is greatest between 6 and 12 years of age, although use among adolescents
is increasing at the highest rate (Zuvekas & Vitiello, 2012).
A recent meta-analysis of 42 studies conducted by Cortese et al. (Cortese et al., 2015)
found that children ages 6 to 17 who were reported as having an ADHD diagnosis had 1.20 times
the odds of obesity (95% CI: 1.05-1.37) as those without the disorder, while adults with ADHD
had 1.55 times the odds (95% CI: 1.32-1.81). Several mechanisms have been identified that may
explain this association. Firstly, ADHD is highly comorbid with conditions that may be
independently associated with weight gain including depression, anxiety, substance abuse,
conduct disorder, and oppositional defiant disorder (Larson, Russ, Kahn, & Halfon, 2011; Pauli-
Pott, Neidhard, Heinzel-Gutenbrunner, & Becker, 2014). Secondly, the impulsivity aspect and
brain chemistry pathology of ADHD are hypothesized to combine to contribute to disordered
eating patterns, including binge eating and food addiction (Cortese & Morcillo, 2010; Davis,
Levitan, Smith, Tweed, & Curtis, 2006). Finally, genetic and fetal programming may contribute
to underlying biological mechanisms that drive both ADHD development and metabolic
dysregulation (Rodriguez et al., 2008), and a bidirectional association has been suggested via
neurotransmitter and endocrine changes associated with increased adiposity (Cortese & Morcillo,
2010).
50
Until recently, however, stimulant medication treatment of ADHD was not considered as
a potential contributory factor to increased overweight and obesity risk among children with
ADHD. This was because only cross-sectional and short-term longitudinal studies of 3 months to
three years had been conducted. These studies either found no association, or found that short-
term stimulant use was associated with growth deficits and lower body mass index (BMI)
(Charach, Figueroa, Chen, Ickowicz, & Schachar, 2006; Spencer, Biederman, & Wilens, 1998;
Zachor, Roberts, Hodgens, Isaacs, & Merrick, 2006), likely varying by developmental period and
due to short-term endocrine and anorexigenic side effects of stimulants (Leonard, McCartan,
White, & King, 2004).
Cortese et al.’s meta-analysis pooled results from 12 such studies of obesity risk in both
medicated and unmedicated children, adolescents and adults with ADHD, and found no evidence
that individuals with medicated ADHD were at increased risk of obesity (odds ratio 1.00,
95%CI: 0.87-1.15), but higher odds of obesity for those with unmedicated ADHD (odds
ratio=1.43, 95% CI: 1.23-1.67) (Cortese et al., 2015). However, of the 11 studies conducted in
children and/or adolescents, 10 were cross-sectional, and none stratified point estimates of
associations between BMI and medication status by duration of medication use to examine
associations between stimulant use and obesity risk after the initial weight loss and growth
retardation period of 1 to 3 years (Cortese et al., 2015). Also, none of the studies looked at
differences in associations by developmental periods, which could be critical, since some studies
have shown that children with ADHD exhibit delayed physical maturation independent of
medication status, perhaps due to ADHD’s association with low birthweight (Van Mil et al.,
2015).
51
The only long-term, prospective longitudinal study examining BMI trajectory differences
by stimulant use and duration was not included in Cortese et al.’s meta-analysis (Schwartz et al.,
2014). The Schwartz study used longitudinal electronic health record data from Pennsylvania’s
Geisinger Health System on 163,820 children ages 3 to 18 years, and modeled BMI trajectories
with increasing age in relation to ADHD diagnosis, age at first stimulant use, and stimulant use
duration. It found associations between early and long-term stimulant use in childhood and
subsequent increased BMI trajectory into overweight and obesity in adolescence and early
adulthood, but did not find such an association in individuals with unmedicated
ADHD (Schwartz, Bailey-Davis et al. 2014).
While this study design was rigorous and advanced understanding of the complex
relationships between ADHD diagnosis, stimulant use, and obesity, the study population was
primarily white and higher income, and a similar exploration has not been made to date in a more
nationally representative sample with racial/ethnic, socioeconomic, and geographic
diversity. Importantly, the Schwartz study also left open the question of mechanisms; thus even
if the results are replicable it is unclear if the positive relationship between long-term stimulant
use and increase in BMI trajectory occurs because of differences in health behaviors such as
dietary intake and physical activity among children taking medication. A conceptual model of
these relationships is shown in Figure 3.1.
52
Figure 3.1. Conceptual model of relationships between ADHD, medication, health behaviors,
and BMI change in children.
In order to help address these gaps in our understanding, this study will use the Early
Childhood Longitudinal Study 1998-99 Kindergarten Cohort (ECLS-K98) dataset to
investigate two primary aims. First, we will examine whether the positive association between
long-term ADHD medication use and increased BMI trajectory during early adolescence found
by Schwartz et al. is evident in a nationally representative sample of children, and if such a
relationship changes by developmental stage or becomes more pronounced with longer durations
of reported use (Aim 1). Secondly, we will examine differences in 5 th
grade dietary patterns and
physical activity levels between children without ADHD, those with unmedicated ADHD, and
those with medicated ADHD, and test whether any relationship between long-term stimulant use
and increases in BMI and BMI z-score between 5 th
and 8 th
grade are mediated by parent-reported
physical activity levels and child-reported unhealthy diet score (Aim 2).
ADHD BMI
Trajectory
Stimulant
Medication
Use
Diet & Physical
Activity
Potential Confounders SES, Sex, Race/Ethnicity
Comorbid Behavioral Health
Diagnoses and Psychiatric
Medications Early Externalizing Symptoms
53
Methods
Study Design
This study uses data from the ECLS-K98, a prospective observational cohort of
kindergarteners and their parents, teachers, and schools that began in 1998-1999, and was conducted
by the National Center for Education. Data collection occurred in kindergarten, 1st grade, 3rd grade,
5th grade, and 8th grade (5 waves). Including the base year and sample refreshening in 1 st grade,
21,410 children were enrolled and 9,360 had data collected on them through 8 th grade. Much of the
decrease in sample size is a consequence of planned attrition, as the ECLS-K98 only followed
children who remained in their initial school, as well as a small subsample of children who switched
schools during the study period.
We used multiple imputation with chained equations to account for missing values in
covariates, accommodating arbitrary missing value patterns (White, Royston, & Wood, 2011).
Imputations were based on available information on all variables included in the study, as well as
additional baseline demographic variables. Following recommendations in the literature on
multiple imputation, this study excludes children for whom BMI was not collected in 5 th or 8
th grade
(Von Hippel, 2007), as well as those starting stimulants between 5 th and 8
th grade since the exposure
would not temporally precede the outcome, for a final complete case study population of N=8,250 1 .
Measures:
ADHD Diagnosis and Medication Use: In each wave, parents were asked if their child had
received a diagnosis of ADHD by a health care professional. In the 5 th and 8
th grade waves, all
11
All sample sizes have been rounded to the nearest 10 to comply with ECLS-K98 reporting and
privacy requirements.
54
parents who reported a diagnosis of ADHD for their child in the current or previous waves were
asked the question, “Is your child now taking any prescription medicine for the condition related to
his or her ADD, ADHD, or hyperactivity?” If they reported medication use, they were then asked
regarding the duration of use (less than 1 year, 1-2 years, 3-4 years, 5 or more years), and the name
of the medication. Those answers were used to code dummy variables representing whether
children began a stimulant medication prior to 1 st grade, between 1
st grade and 3
rd grade, or between
3 rd
grade and 5 th grade.
BMI and BMI z-score: Trained ECLS-K98 staff members measured children's height and
weight in the spring of each data collection wave using a Shorr board (accuracy: 0.01 cm) and a
Seca digital bathroom scale (model 840 [Seca, Hanover, MD]; accuracy: 0.1 kg), respectively.
Children were asked to remove shoes and heavy clothing prior to measurement. Height and weight
were measured two times consecutively in order to minimize errors, and averaged when the
discrepancy between replicates was less than 5.1 cm for height and 2.27 kg for weight (98% of
cases); otherwise, the value closest to the grade-level median was used (Tourangeau, Nord, Lê,
Sorongon, & Najarian, 2009). BMI was calculated using weight in kilograms divided by the square
of height in meters (kg/m 2 ). BMI z-scores were then calculated using CDC 2000 Growth Reference
standards for age and gender.
Health Behaviors: Health behavior questions on the ECLS-K98 were mainly taken from
two existing surveys conducted by the Centers for Disease Control and Prevention (CDC)/Division
of Adolescent and School Health Surveys: the Youth Risk Behavior Surveillance Survey (YRBSS)
and the School Health Programs and Policies Survey (SHPPS) (Tourangeau et al., 2009). One food
consumption question on fast-food meals was taken from the California Children’s Healthy Eating
and Exercise Practices Survey (CalCheeps) (Tourangeau et al., 2009).
55
Physical activity was operationalized as parent-reported number of days per week where the
child experienced 20 or more consecutive minutes of moderate to vigorous intensity exercise. Diet
score is generated as a summary measure of 6 questions on dietary intake asked of students in 5 th
and 8 th grades. These questions (3 on vegetable and salad intake, 2 on sugar sweetened beverage
(SSB) intake, 1 on fast food intake) were used to construct an “unhealthy diet pattern” score.
Children reporting less than one serving of salad or vegetables per day, those reporting more than 1
SSB serving per day, and those eating fast food more than 3 times per week all received 1 point
each, for a possible score ranging from 0 (most healthy) to 3 (most unhealthy).
Covariates: Covariates included parent-reported child characteristics of age, race/ethnicity,
sex, socioeconomic status (SES), and comorbid behavioral health diagnoses, as well as teacher-
reported externalizing symptoms in kindergarten as measured by the externalizing sub-scale of the
Child Behavior Checklist (CBCL) (Ebesutani et al., 2010). The components used to create the
continuous SES variable are father/male guardian’s education, mother/female guardian’s
education, father/male guardian’s occupational prestige, mother/female guardian’s occupational
prestige, and household income. In households with two mothers or two fathers, education and
occupational prestige for both mothers/fathers was used. Each parent’s occupation was scored
using the average of the 1989 General Social Survey (GSS) prestige scores for the 1980 census
occupational category codes that correspond to the ECLS-K98 occupation code. The
demographic covariates were collected from parents during baseline interviews, while data on
parental report of comorbid diagnoses was obtained at each wave.
56
Statistical Analyses
All analyses were conducted using STATA 13. ANOVA and chi-square tests were used to
evaluate differences in demographic variables among the three groups of interest (children without
ADHD, those with unmedicated ADHD, those with ADHD medicated by 5 th
grade). For Aim 1, we
used multiple linear regression to first model BMI change by developmental period (1 st to 3
rd grade,
3 rd
to 5 th grade, and 5
th and 8
th grade) using dummy variables representing medication start time
(before 1 st grade, between 1
st and 3
rd grade, and between 3
rd and 5
th grade) as the outcome predictor.
We then adjusted the models to control for baseline confounders identified in the literature and
conceptual modeling including race/ethnicity, socioeconomic status, sex, birthweight, comorbid
behavioral health diagnoses, and teacher reported externalizing behavior at baseline (Cortese &
Morcillo, 2010; van Egmond-Fröhlich, Widhalm, & De Zwaan, 2012; Van Mil et al., 2015). We
then replicated both the unadjusted and confounder adjusted models using change in BMI z-score,
which provides age and gender standardized context for effect size and may aid clinical
interpretability. Finally, we tested equality of coefficients to evaluate if BMI effects differed
significantly for the various medication start times.
For Aim 2, we dichotomized dietary scores and physical activity levels and used logistic
regression to model the odds of unhealthy behaviors (unhealthy diet score ≥ 2 and physical activity
level < 3 days per week) as predicted by ADHD diagnosis and medication category, in both
unadjusted and confounder adjusted models (Howard et al., 2011). Finally, we conducted a
mediation analysis utilizing both Sobel-Goodman tests and calculation of indirect effects post-
imputation (Baron & Kenney 1986, Hayes 2013) to determine if physical activity levels or diet
scores are significant mediators of any effects of medication on BMI change.
57
Results
Table 3.1 shows baseline characteristics of the children in the study sample (n=8250), and
differences across the groups of interest. Parents reported a diagnosis of ADHD by a health care
provider for approximately 8% of the children in the sample (n=650). Among children diagnosed
with ADHD by 5th grade, almost 60% were taking medications to manage symptoms (n=380).
Children taking medication were more likely to be white (p=0.03) than children with ADHD who
remained unmedicated by the 5 th
grade. Of children who were medicated, 60 were taking
medication by 1 st grade, 120 were reported as having started medication between 1
st and 3
rd grade,
and 190 were reported as starting between 3 rd
and 5 th grade. Bivariate results suggest that mean
BMI was not different between children without ADHD and those with unmedicated ADHD in 5th
grade, however, it was significantly lower for children with medicated ADHD (p<0.001). By 8th
grade, no significant difference among the three groups remained.
58
Table 3.1. Characteristics of the study population. 1
Child Characteristics
Total
Sample
N=8250
No
ADHD
n=7600
Unmedicated
ADHD
n=270
Medicated
ADHD
n=380
Test
Statistic 2
Sex (M) (%) 49.9 48.0 70.4 73.7 χ
2 =138.2,
p<0.001
Mean age at start of 5 th
grade (years) 11.2±.4 11.2±.4 11.2±.4 11.3±.4
F=1.17,
p=0.31
Mean BMI in 5 th
grade
(kg/m 2 )
20.5±4.7
20.6±4.7 20.2±4.6 19.1±4.3 F=20.2,
p<0.001
Mean BMI in 8 th
grade
(kg/m 2 )
22.8±5.4 22.9±5.4 22.6±5.3 22.3±5.1 F=2.3,
p=0.11
Mean BMI change
from 5 th
to 8 th
grade
(kg/m 2 )
2.3±2.8
2.3±2.8 2.4±2.6 3.2 ± 2.4 F=23.2,
p<0.001
Mean BMI z-score
change 5 th
to 8 th
grade
-0.01
±0.58
-0.02±
0.57 0.00 ± 0.61
0.29 ±
0.67
F=55.3,
p<0.001
Physical activity level
(days/week) 3.7±1.9 3.7± 1.8 3.8± 2.0 3.6 ±2.0
F=0.83,
p=0.44
Unhealthy diet score a
0.93±0.81 0.90 ±
0.81 1.23 ± 0.82 1.20 ±0.82
F=43.64,
p<0.001
Early externalizing
symptoms b 1.6±0.6
1.6± 0.6 2.0 ± 0.8 2.2± 0.7
F=245.9,
p<0.001
Comorbid behavioral
health diagnosis (%) 3.2 2.0 11.1 21.1
χ 2 =495.2,
p<0.001
59
Table 3.1 (continued). Characteristics of the study population. 1
Family
Characteristics
SES c (%)
χ 2 =13.8,
p=0.09
Quintile 1 14.5 14.6 14.8 13.2
Quintile 2 17.1 16.8 22.2 21.1
Quintile 3 19.2 18.9 18.5 23.7
Quintile 4 21.6 21.7 18.5 21.1
Quintile 5 24.5 24.6 22.2 23.7
Ethnicity (%) χ
2 =96.7,
p<0.001
African-
American 10.4 10.4 14.8 7.9
Asian 5.6 5.9 <3.7 <2.6
Hispanic 17.1 17.8 14.8 7.9
White 61.9 60.8 66.7 81.6
Other 5.0 5.1 3.7 2.6 1 Descriptive statistics are for observed data (unimputed); some observations were missing data so
sample sizes varied for physical activity (n=7810), diet (n=7910), externalizing symptoms
(n=7810), comorbid diagnoses (n=7930), and SES (n=7980). Sample sizes for all descriptive
statistics are rounded to the nearest 10 before conversion to percentages to comply with National
Center for Education’s privacy requirements. Therefore, not all categories will add to 100%. 2 Degrees of freedom are redacted as required by the National Center of Education’s privacy
requirements. a Out of a possible 3 points (0=most healthy, 3=most unhealthy).
b On a 1-4 point Likert Scale.
c SES quintiles as calculated by the National Center for Education Statistics from the full ECLS-K98
sample at baseline.
60
Aim 1: Modeling results of medication start time as the predictor of BMI and z-score
change are shown in Table 3.2. In both unadjusted and confounder adjusted models, unmedicated
ADHD was negatively associated with BMI change up to 3 rd
grade, after which the association
became non-significant. In the unadjusted model, starting medication before 1 st grade was
negatively associated with BMI change from 1 st to 3
rd grade, not significantly associated with BMI
change from 3 rd
to 5 th
grade, and positively associated with change from 5 th to 8
th grade. Once the
model was adjusted for confounders, starting medication by 1 st grade remained inversely associated
with BMI change through 5 th grade, and still became positively associated with increased BMI
change from 5 th
to 8 th
grade. Both models found that starting medication between 1 st and 3
rd grade
was not significantly associated with BMI change from 3 rd
to 5 th grade, but was associated with a
nearly 1.2 kg/m 2 greater increase in BMI change from 5
th to 8
th grade. Finally, beginning medication
between 3 rd
and 5 th grade was positively associated with BMI increases from 5
th to 8
th grade in both
unadjusted and adjusted models. For all three medication start times, this association translates to
about a 0.3 SD increase in children’s BMI z-score from 5 th to 8
th grade, while children with
unmedicated ADHD experience the same rate of growth as those without an ADHD diagnosis.
Tests of the equality of coefficients showed no significant difference among the increased BMI and
BMI z-score changes from 5 th to 8
th grade experienced by children with different medication start
times.
61
T a b
le 3
.2 . L
in e a r
re g re
ss io
n m
o d e li
n g r
e su
lt s
fo r
B M
I a n d B
M I
z -s
c o re
o u tc
o m
e s .
Δ
B M
I
1 st -3
r d G
r a d
e
Δ B
M I
3 r d -5
th G
r a d
e
Δ B
M I
5 th
-8 th
G r a d
e
Δ z B
M I
1 st -3
r d G
r a d
e
Δ z B
M I
3 r d -5
th G
r a d
e
Δ z B
M I
5 th
-8 th
G r a d
e
β (
9 5 %
C I)
M o d e l 1 :
U n a d ju
st e d
1
U n m
e d ic
a te
d b
e fo
re
5 th g
ra d e
2
-0 .3
7
( -0
.6 7 , -0
.0 7 )
0 .2
2
(- 0 .0
2 , 0 .4
7 )
0 .1
3
(- 0 .2
1 , 0 .4
8 )
-0 .0
2
(- 0 .1
2 , 0 .0
9 )
0 .0
7
(- 0 .0
0 , 0 .1
3 )
0 .0
2
(- 0 .0
5 , 0 .0
9 )
S ta
rt in
g M
e d ic
a ti
o n
b y 1
st G
ra d e
2
-1 .2
8
(- 1 .8
8 , -0
.6 9 )
-0 .4
1
(- 0 .9
0 , 0 .0
7 )
1 .0
2
(0 .3
4 , 1 .7
0 )
-0 .3
8
(- 0 .5
9 , -0
.1 7 )
-0 .0
3
(- 0 .1
6 , 0 .1
0 )
0 .3
3
(0 .1
8 , 0 .4
7 )
S ta
rt in
g M
e d ic
a ti
o n
b e tw
e e n 1
st a
n d 3
rd
G ra
d e
2
--
-0 .2
6
(- 0 .6
1 , 0 .0
9 )
1 .1
7
(0 .6
8 , 1 .6
6 )
--
-0 .0
0
(- 0 .1
0 , 0 .0
9 )
0 .3
6
(0 .2
6 , 0 .4
6 )
S ta
rt in
g M
e d ic
a ti
o n
b e tw
e e n 3
rd a
n d 5
th
G ra
d e
2
--
--
0 .8
3
(0 .4
4 , 1 .2
2 )
--
--
0 .2
8
(0 .2
0 , 0 .3
6 )
M o d e l 2 : C
o n fo
u n d e r
a d ju
st e d
1 ,3
U n m
e d ic
a te
d b
e fo
re
5 th g
ra d e
2
-0 .4
3
(- 0 .7
3 , -0
.1 3 )
0 .1
3
(- 0 .1
2 , 0 .3
8 )
0 .0
7
(- 0 .2
8 , 0 .4
1 )
-0 .0
4
(- 0 .1
4 , 0 .0
7 )
0 .0
7
(- 0 .0
0 , 0 .1
4 )
0 .0
0
(- 0 .0
7 , 0 .0
8 )
S ta
rt in
g M
e d ic
a ti
o n
b y 1
st G
ra d e
2
-1 .3
0
(- 1 .9
0 , -0
.7 1 )
-0 .5
0
(- 0 .9
9 , -0
.0 1 )
0 .9
4
(0 .2
5 , 1 .6
2 )
-0 .4
2
(- 0 .6
4 , -0
.2 0 )
-0 .0
5
(- 0 .1
9 , 0 .0
9 )
0 .3
0
(0 .1
6 , 0 .4
4 )
S ta
rt in
g M
e d ic
a ti
o n
b e tw
e e n 1
st a
n d 3
rd
G ra
d e
2
--
-0 .2
9
(- 0 .6
4 , 0 .0
7 )
1 .1
9
(0 .7
0 , 1 .6
8 )
--
-0 .0
3
(- 0 .1
3 , 0 .0
7 )
0 .3
5
(0 .2
5 , 0 .4
5 )
S ta
rt in
g M
e d ic
a ti
o n
b e tw
e e n 3
rd a
n d 5
th
G ra
d e
2
--
--
0 .7
8
(0 .3
9 , 1 .1
8 )
--
--
0 .2
7
(0 .1
8 , 0 .3
5 )
1 T
h e r
e fe
re n c e g
ro u p i
s c h il
d re
n w
it h o u t
A D
H D
.
2 C
h il
d re
n i
n b
o th
u n m
e d ic
a te
d a
n d m
e d ic
a te
d c
a te
g o ri
e s
h a v e b
e e n r
e p o rt
e d a
s d ia
g n o se
d w
it h A
D H
D b
y a
h e a lt
h c a re
p ro
v id
e r.
3 B
M I
c h a n
g e m
o d
e l
is a
d ju
st e d f
o r
so c io
e c o n o m
ic s
ta tu
s (S
E S
), r
a c e /e
th n ic
it y , se
x , a g
e , c o m
o rb
id b
e h a v io
ra l
h e a lt
h d
ia g n o se
s, b
ir th
w e ig
h t,
a n d e
x te
rn a li
z in
g s
y m
p to
m s
in 1
st g
ra d
e ;
z -s
c o re
m o d e l
is a
d ju
st e d f
o r
S E
S , ra
c e /e
th n ic
it y ,
c o m
o rb
id b
e h a v io
ra l
h e a lt
h
d ia
g n o se
s, b
ir th
w e ig
h t,
a n d e
x te
rn a li
z in
g s
y m
p to
m s
in 1
st g
ra d e a
s z -s
c o
re s
a re
s ta
n d a rd
iz e d f
o r
a g e a
n d s
e x
.
62
Aim 2: Logistic modeling results for the dichotomous outcomes of unhealthy physical
activity levels and dietary scores are shown in Table 3.3. The adjusted model showed that the odds
of having low levels of parent-reported physical activity levels in 5 th grade were significantly higher
for children who had ADHD and were medicated by 5 th
grade (adjusted OR=1.27, 95% CI: 1.02,
1.59) relative to children without ADHD; this was not the case for children with unmedicated
ADHD (OR=1.02, 95% CI: 0.79, 1.32). In the both the unadjusted and confounder adjusted
models, odds of having an unhealthy diet score were higher for children with ADHD, regardless of
whether they were medicated (adjusted OR= 2.17, 95% CI: 1.39-3.38) or not (adjusted OR=1.88,
95% CI: 1.13, 3.10).
Table 3.3. Logistic regression results for dietary patterns and physical activity outcomes in 5 th
grade.
Unhealthy Physical Activity
Level (Days per week<3)
Unhealthy Dietary Pattern
(Score≥2))
Odds Ratio (95% CI)
Model 1: Unadjusted 1
Not medicated (ADHD
diagnosis) 0.90 (0.70, 1.16) 1.81 (1.39, 2.36)
Medicated (ADHD diagnosis) 1.08 (0.88, 1.33) 1.62 (1.29, 2.02)
Model 2: Confounder adjusted 1,2
Not medicated (ADHD
diagnosis) 1.02 (0.79, 1.32) 1.52 (1.15, 2.00)
Medicated (ADHD diagnosis) 1.27 (1.02, 1.59) 1.41 (1.11, 1.81) 1
Reference category is children without ADHD. 2 Adjusted for SES, race/ethnicity, sex, comorbid behavioral health diagnoses, and externalizing
symptoms in 1 st grade.
63
Both Sobel-Goodman tests and calculation of indirect effects post-imputation found no
significant mediation of the relationship between BMI change between 5 th and 8
th grade and
medication use by either child-reported diet score or parent-reported physical activity level in 5 th
grade. The indirect effect of medication use through diet was very small (β=0.01, 95% CI: -0.01,
0.03), representing only 1.2% of the total effect of medication use on BMI change. The indirect
effect of medication use through physical activity was also negligible (β= -0.01, 95% CI: -0.02,
0.00) at 0.8% of the total effect.
Discussion
Mean BMI in 5 th grade was significantly lower among children taking medication to manage
their ADHD symptoms as compared to children who were not taking medication. Although this
cross-sectional finding cannot be causally interpreted, it was expected given the evidence that
stimulant use can reduce growth rates for up to two to three years in young children and the median
medication start time for the sample was 3 rd
grade (Charach et al., 2006). However, these children
experienced greater increases in BMI between 5 th
and 8 th grade; thus by 8
th grade their mean BMI
was no longer significantly different from either typically developing children, or those with ADHD
who remained unmedicated. While early medication use prior to 3 rd
grade appears to predict greater
BMI increases between 5 th and 8
th grade, this predicted increase was not significantly different from
that associated with a later start time. Children with unmedicated ADHD did not experience either
higher BMI or greater BMI change during any developmental period. This was the case whether we
controlled for early externalizing symptom severity or not, which demonstrates the associations are
unlikely to be confounded by indication.
When viewed in context with previous studies, our findings appear very consistent with
Schwartz et al.’s results, and provide further insight among a geographically and
64
demographically diverse sample. Certainly, our study provides additional evidence of BMI
rebound during pre- and early adolescence among children who have ADHD and begin stimulant
use before 5 th
grade. By contrast, after controlling for confounders and medication use, we found
a negative association between unmedicated ADHD diagnosis and BMI increases up to 3 rd
grade,
and no association after that time. While this appears to conflict with the findings of the Cortese
et al. meta-analysis, it is likely due to the fact that the meta-analysis relied on cross-sectional data
in medicated children with ADHD, and could not distinguish between developmental periods or
short- versus long-term medication use. However, we did find that children with ADHD
diagnoses have worse diet scores regardless of medication use, and there is an indication of
slightly increased BMI trajectory (non-significant) among these children between 5 th
and 8 th
grade, which may not manifest in significant differences until later in the life course.
While there was an increased risk of low physical activity levels among children taking
medication, neither physical activity levels nor dietary score acted as mediators of the relationship
between medication use and BMI change. This may provide some evidence that the positive
association between increased BMI trajectory and long-term stimulant use could be due not to
differences in health behaviors among children with medicated and unmedicated ADHD, but
instead to physiological changes or metabolic compensation associated with early and long-term use
of stimulants.
Unfortunately, this analysis is limited by the lack of follow-up after 8 th grade; thus we
cannot confirm that these BMI trajectories are maintained after the study period, and so we cannot
determine from this data whether the increased BMI gains among stimulant medicated children
simply represent growth catch up after initial stimulant use, or as Schwartz et al. found, predicate a
continued increased trajectory into higher risk of overweight and obesity (Schwartz et al., 2014).
65
However, it is important to note that no studies have found large changes in established population
level BMI trajectory after 8 th grade (Williams & Goulding, 2009).
Though we used multiple imputation to account for missing data, we dropped cases where
data was missing on the dependent variables. Therefore, the significant attrition in the study means
that children with more severe ADHD symptoms may be disproportionately lost to follow up,
which may influence the finding of no association between unmedicated ADHD and BMI increases.
However, this possibility is made less likely because there were no significant differences between
the proportion of overall attrition and attrition by children with diagnosed ADHD, as well as no
differences in externalizing symptoms between those leaving the study and those remaining.
Finally, physical activity and diet are not objectively measured, which increases
measurement error, decreases the likelihood of detecting differences among the groups, and
increases the likelihood that that the indirect effect is attenuated in mediation analyses (Baron &
Kenny, 1986). However, the very small indirect effects indicates that while future research should
focus on objectively measuring health behaviors and clarifying these relationships during later
adolescence, it is critical to develop a better evidence base regarding potential physiologic
mechanisms by which stimulants may affect long-term metabolism. Also, given that children with
ADHD were found to have significantly worse diets despite measurement limitations and regardless
of medication status, this finding should be taken seriously.
Associations between ADHD diagnosis, stimulant medication use and BMI changes have
important clinical implications. Longitudinal evidence of higher pre- and early adolescent BMI
trajectory among children using stimulant medications in childhood may simply reflect growth
catch-up, but it may also indicate that metabolic compensation is occurring associated with early
66
and extended stimulant use. These weight gain impacts may only be exacerbated if children are co-
pharmacologically treated with second generation antipsychotics, which are increasing in pediatric
use and known to causally increase adiposity (Penzner et al., 2009).
Given the longitudinal evidence, stimulants should not be viewed by physicians as a tool to
reduce obesity risk among children with ADHD, at least until additional research is conducted. By
the same token, there is not enough evidence to suggest that future obesity risk should deter
prescriptions indicated to treat ADHD’s behavioral symptoms. Since children with ADHD do
appear to be at higher risk of poor dietary intake, clinicians should consider lifestyle counseling
following an ADHD diagnosis regardless of medication prescription.
References
Baron, R. M., & Kenny, D. A. (1986). The moderator–mediator variable distinction in social
psychological research: Conceptual, strategic, and statistical considerations. Journal of
personality and social psychology, 51(6), 1173.
Charach, A., Figueroa, M., Chen, S., Ickowicz, A., & Schachar, R. (2006). Stimulant treatment
over 5 years: effects on growth. Journal of the American Academy of Child & Adolescent
Psychiatry, 45(4), 415-421.
Cortese, S., & Morcillo, P. C. (2010). Comorbidity between ADHD and obesity: exploring
shared mechanisms and clinical implications. Postgraduate medicine, 122(5), 88-96.
Cortese, S., Moreira-Maia, C. R., St. Fleur, D., Morcillo-Peñalver, C., Rohde, L. A., & Faraone,
S. V. (2015). Association between ADHD and obesity: a systematic review and meta-
analysis. American journal of psychiatry, 173(1), 34-43.
67
Davis, C., Levitan, R. D., Smith, M., Tweed, S., & Curtis, C. (2006). Associations among
overeating, overweight, and attention deficit/hyperactivity disorder: a structural equation
modelling approach. Eating behaviors, 7(3), 266-274.
Ebesutani, C., Bernstein, A., Nakamura, B. J., Chorpita, B. F., Higa-McMillan, C. K., Weisz, J.
R., & The Research Network on Youth Mental, H. (2010). Concurrent Validity of the
Child Behavior Checklist DSM-Oriented Scales: Correspondence with DSM Diagnoses
and Comparison to Syndrome Scales. Journal of Psychopathology and Behavioral
Assessment, 32(3), 373-384. doi: 10.1007/s10862-009-9174-9
Howard, A. L., Robinson, M., Smith, G. J., Ambrosini, G. L., Piek, J. P., & Oddy, W. H. (2011).
ADHD is associated with a “Western” dietary pattern in adolescents. Journal of attention
disorders, 15(5), 403-411.
Larson, K., Russ, S. A., Kahn, R. S., & Halfon, N. (2011). Patterns of comorbidity, functioning,
and service use for US children with ADHD, 2007. Pediatrics, peds. 2010-0165.
Leonard, B. E., McCartan, D., White, J., & King, D. J. (2004). Methylphenidate: a review of its
neuropharmacological, neuropsychological and adverse clinical effects. Human
Psychopharmacology: Clinical and Experimental, 19(3), 151-180.
Pauli-Pott, U., Neidhard, J., Heinzel-Gutenbrunner, M., & Becker, K. (2014). On the link
between attention deficit/hyperactivity disorder and obesity: do comorbid oppositional
defiant and conduct disorder matter? European child & adolescent psychiatry, 23(7),
531-537.
Penzner, J. B., Dudas, M., Saito, E., Olshanskiy, V., Parikh, U. H., Kapoor, S., . . . Sheridan, E.
M. (2009). Lack of effect of stimulant combination with second-generation
antipsychotics on weight gain, metabolic changes, prolactin levels, and sedation in youth
68
with clinically relevant aggression or oppositionality. Journal of child and adolescent
psychopharmacology, 19(5), 563-573.
Rodriguez, A., Miettunen, J., Henriksen, T. B., Olsen, J., Obel, C., Taanila, A., . . . Järvelin, M.
(2008). Maternal adiposity prior to pregnancy is associated with ADHD symptoms in
offspring: evidence from three prospective pregnancy cohorts. International Journal of
Obesity, 32(3), 550-557.
Schwartz, B. S., Bailey-Davis, L., Bandeen-Roche, K., Pollak, J., Hirsch, A. G., Nau, C., . . .
Glass, T. A. (2014). Attention Deficit Disorder, Stimulant Use, and Childhood Body
Mass Index Trajectory. Pediatrics, 133(4), 668-676.
Spencer, T., Biederman, J., & Wilens, T. (1998). Growth deficits in children with attention
deficit hyperactivity disorder. Pediatrics, 102(Supplement 3), 501-506.
Tourangeau, K., Nord, C., Lê, T., Sorongon, A. G., & Najarian, M. (2009). Early Childhood
Longitudinal Study, Kindergarten Class of 1998-99 (ECLS-K): Combined User's Manual
for the ECLS-K Eighth-Grade and K-8 Full Sample Data Files and Electronic
Codebooks. NCES 2009-004. National Center for Education Statistics.
van Egmond-Fröhlich, A., Widhalm, K., & De Zwaan, M. (2012). Association of symptoms of
attention-deficit/hyperactivity disorder with childhood overweight adjusted for
confounding parental variables. International Journal of Obesity, 36(7), 963-968.
Van Mil, N. H., Steegers-Theunissen, R. P., Motazedi, E., Jansen, P. W., Jaddoe, V. W.,
Steegers, E. A., . . . Tiemeier, H. (2015). Low and high birth weight and the risk of child
attention problems. The Journal of pediatrics, 166(4), 862-869. e863.
Visser, S. N., Danielson, M. L., Bitsko, R. H., Holbrook, J. R., Kogan, M. D., Ghandour, R. M., .
. . Blumberg, S. J. (2014). Trends in the parent-report of health care provider-diagnosed
69
and medicated attention-deficit/hyperactivity disorder: United States, 2003–2011.
Journal of the American Academy of Child & Adolescent Psychiatry, 53(1), 34-46. e32.
Von Hippel, P. T. (2007). Regression with missing Ys: An improved strategy for analyzing
multiply imputed data. Sociological Methodology, 37(1), 83-117.
White, I. R., Royston, P., & Wood, A. M. (2011). Multiple imputation using chained equations:
issues and guidance for practice. Statistics in medicine, 30(4), 377-399.
Williams, S. M., & Goulding, A. (2009). Patterns of growth associated with the timing of
adiposity rebound. Obesity, 17(2), 335-341.
Zachor, D. A., Roberts, A. W., Hodgens, J. B., Isaacs, J. S., & Merrick, J. (2006). Effects of
long-term psychostimulant medication on growth of children with ADHD. Research in
developmental disabilities, 27(2), 162-174.
Zuvekas, S. H., & Vitiello, B. (2012). Stimulant medication use in children: a 12-year
perspective. American Journal of Psychiatry.