HT_JSL_09_FINAL.pdf

27 

Full text

(1)

COMPARISON OF TWO STATISTICAL METHODS TO ADDRESS MISSING ACCELEROMETER DATA IN HCHS/SOL YOUTH

by

Jose Santos Lopez

A paper submitted to the faculty of the University of North Carolina in fulfilment of Undergraduate Honors Thesis

in the Department of Biostatistics

Chapel Hill

April 2020

Approved by:

Advisor, Daniela Sotres-Alvarez, DrPH

Reader, Jianwen Cai, PhD

Reader, Kelly R. Evenson, PhD

Oral Presentation on: April 21 2020

(2)

1

TABLE OF CONTENTS

I. INTRODUCTION 2

II. METHODS 4

III. RESULTS 12

IV. DISCUSSION 15

V. APPENDIX 17

VI. ACKNOWLEDGEMENTS 24

VII. REFERENCES 25

(3)

2

I. INTRODUCTION

Opposed to children and adolescents of different ethnicities and or races,

Hispanic/Latino youth are at a higher risk for illnesses attributable to obesity. Across a span of

ages 2 to 19 years, children ages 6 to 11 years and adolescents ages 12 to 19 years have 2.29

times the odds and 2.82 times the odds of obesity, respectively, compared to children ages 2 to

5 years.1 Further, Hispanic children/adolescents have 1.48 (1.23-1.78, 95%CI) times the odds of

obesity compared to Non-Hispanic white children/adolescents. 1 Of concern is that obesity

during childhood and adolescence creates a precedent for obesity in adulthood.2

There is some existing research which has investigated physical activity (PA) and

sedentary behavior (SB) in Hispanic/Latino youth that corroborates Hispanic/Latino youth

engaging in below recommended amounts of PA. For example, when compared to non-Hispanic

youth, researchers have shown that Hispanic youth take part in less physical activities.3

Comparable to other large studies with accelerometer data, in SOL Youth there are large

amount of missing data, which may result in biased accelerometer PA and SB estimates if there

are distinct differences between children with accelerometer data and children with missing

accelerometer data. Based on the Hispanic Community Health Study/Study of Latinos’

(HCHS/SOL) Ancillary Study of Latino Youth’s (SOL Youth’s) accelerometer collected data, a

complete case analysis (CCA) – analysis of cases with non-missing data – of accelerometer data

yielded the overall mean PA in minutes per day by intensity as: 178.9, light; 25.4, moderate;

and 10.2 vigorous adjusted by accelerometry wear time.4,5 Physical Activity Guidelines for

(4)

3

overall wear time adjusted mean SB was 604.6 mins per day, where higher levels of SB was

observed among older individuals (age 15-16 year range in this case).5

In this paper, we aim to use data from the HCHS/SOL’s SOL Youth to assess whether

there are differences amongst the estimates of PA and SB utilizing two different methods to

address missing data compared to complete case analysis (CCA) (restricting analysis to

observations without missing data). The two methods explored in this study are multiple

imputation (MI) and inverse probability weights (IPWs).7,8We will compare the results between

these two methods treating PA and SB as either outcomes, main exposures of interest, or as

covariates, to evaluate the benefits and drawbacks of the approaches to addressing missing

(5)

4

II. METHODS

Study population

HCHS/SOL is an ongoing and the largest multi-site community-based prospective cohort

study of Hispanic/Latinos in the United States (N=16,415), where participants ages 18-74 years

were enrolled from 2008 to 2011 at the Bronx, Chicago, Miami, and San Diego study sites. 9 The

study aims to be the basis for the development of innovative hypotheses and research that will

lead to findings that will improve the health of the Hispanic/Latino community.

SOL Youth is an ancillary study to HCHS/SOL with a total of 1,466 children ages 8-16

years old recruited from 2012 to 2014 – all children of at least one parent participating in the

HCHS/SOL.4 The study aims to assess how acculturation, family behaviors, and psychosocial

factors associate with youth's lifestyle behaviors and cardiometabolic risk factors. An

approximately 3.5-hour clinical examination, included questionnaires, anthropometry, blood

collection, and wearing a physical activity monitor for 7-days. The study was approved by the

Institutional Review Boards (IRBs) from all participating institutions. All study parents or

guardians gave written informed consent for children 12 years old or younger, and children

older than 12 years old gave assent.

Data Measures

The primary measurement of physical activity and sedentary behavior was collected

over a week using an Actical accelerometer (version B-1; model 198-0200-03). Specifically,

participants were asked to go about their lives normally wearing the accelerometer above the

iliac crest secured to a belt, removing only when swimming, showering, and sleeping.5 The unit

(6)

5

epochs (intervals). SB and moderate to vigorous physical activity (MVPA) were defined as

follows: sedentary, <18 counts/15 seconds, and moderate to vigorous, ≥441-872 counts/15

seconds – based on a prior calibration study.10 The measures of interest in this study are total

counts per day, minutes of MVPA per day, and minutes of SB per day. In our case, missing

accelerometer data can arise in one of two ways: (1) participants do not return the Actical

accelerometer, and (2) participants did not meet the criteria of having eight hours or more of

wear time for at least three of the seven days. Such criteria and specifications are typical of

defining valid days of accelerometer data and the minimum number of valid wear days required

during the period of data collection.11 In conjunction, returning the accelerometer and having

eight hours or more of wear time for at least three of the seven days will constitute adherence

in this paper. On the basis that children were expected to be sleeping, counts from 12am to

5am were excluded, while non-wear time outside the specified interval was defined with Choi’

algorithm.12 The algorithm specifies that consecutive zero counts for 90 minutes or more

constitutes non-wear time, allowing for up to two minute intervals of nonzero counts given

there was a 30-minute upstream and downstream of consecutive zero counts.

Self-reported physical and sedentary behavior were asked using a SOL Youth designed

physical activity questionnaire to measure the frequency of 68 activities.5 Response options

were never, one to two times per month, one to two times per week, three to four times per

week, five to six times per week, and daily. The activities used in our analysis include transport,

sport, leisure non-sport, school, screen time (a summation of all screen related activities in

(7)

6

Acculturation was defined by the Acculturation, Habits, and Interests Multicultural Scale

for Adolescents (AHIMSA) which includes 5 groups: assimilated, separated, integrated,

marginalized and unclassified.13 Immigration generation and child language preference were

also included to gauge acculturation.

The 5-level body mass index (BMI) group variable used is based on percentiles

respective of a child’s age and the child’s actual BMI to distinguish between 5 groups:

underweight (BMI <5th percentile), normal weight (BMI-84th percentile), overweight (BMI

85-94th percentile), moderate obesity (BMI >95th percentile and BMI < 35), and severe obesity (BMI

>95th percentile and BMI ≥ 35).14 We further consolidated the variable to 4 levels — combining

underweight and normal weight groups. Additionally, an obesity binary variable was created

where being obese encompasses those classified as moderately and severely obese and

non-obese encompasses the underweight, normal weight, and overweight.

Dyslipidemia was defined as having adverse levels of lipids indicated by a total

cholesterol ≥ 200 mg/dL; low-density lipoprotein cholesterol (LDL-C) ≥ 130 mg/dL; triglycerides

≥ 150 mg/dL; or high-density lipoprotein cholesterol (HDL-C) ≤ 40 mg/dL.15,16 Each of the

corresponding measures (total cholesterol, LDL-C, triglycerides, and HDL-C) were processed via

blood samples at a central laboratory (Fairview Laboratories, University of Minnesota) using

common practices.15

Demographic covariates used included 4-level age (8-10, 11-12, 13-14, and 15-16 years

of age), gender, 7-level Hispanic/Latino background classification (Central and South American;

Cuban; Dominican; Mexican; Puerto Rican; mixed; and other), and 3-level annual family income

(8)

7 Statistical Analysis

SAS software version 9.4 (SAS Institute Inc., Cary, NC) was used for all analyses, which all

incorporated sampling weights, stratification, and clustering to account for the HCHS/SOL and

SOL Youth complex survey design. To begin, we compared those who adherent versus

nonadherent using the Wald Chi-squared test for categorical variables and t statistics with

regards to the difference of the domain means for sociodemographic and acculturation

characteristics, health characteristics, and self-reported physical activity.

The two methods explored in this study to account for missing data are MI and IPWs.

We compare the results between these two methods creating PA and SB as either outcome

variables (dependent variable), main exposures of interest (independent variable), or

covariates. We determined the variables that would be used for IPW and MI through bivariate

analyses and literature review.15,17,18,19,20,21

Multiple Imputation of All Data

To adjust for missing and incomplete accelerometer data using multiple imputation,

fully conditional specification (FCS) was used.6 Ten imputed complete datasets were created

using SAS PROC MI with the FCS method including accelerometer variables, self-reported

physical activity, socio-demographic and acculturation characteristics, and health

characteristics (Supplement Table 1), and the sampling weights and stratum to account for SOL

Youth’s complex survey design. Site was not included the FCS models because stratum is

formed by crossing site, high/low proportion if Hispanic/Latinos and a high/low socioeconomic

status from census tracks in the target areas. We specified logistic regression models for binary

(9)

8

continuous variables. For the accelerometer variables we included minutes in sedentary, light,

moderate, and vigorous not including any variables that could be derived from these variables

like MVPA and hours of wear time per day. Then, following imputation MVPA (min/day) was

calculated by adding minutes in moderate and vigorous activity. Accelerometer wear time in

hours per day calculated as the sum of sedentary (min/day), light physical activity (min/day),

moderate physical activity (min/day), and vigorous physical activity (min/day) divided by 60.

Concerning multiple imputation as the method for addressing missing data, 34.8% (N=510) of

participants had at least one missing value for any of the respective variables listed in

Supplement Table 1. However, only 8.6% (N=2266) of the total individual values (1466

observations multiplied by the 18 variables with at least one missing value) were imputed,

where each imputed dataset has 1,466 observations.

Inverse Probability Weighting Derivation with Multiple Imputation of Covariates

To adjust for missing and incomplete accelerometer data using inverse probability

weighting, participants who meet adherent criteria were weighted by the inverse of their

probability of being adherent. We fit a logistic regression model to estimate the probability of

adherence (i.e., returning the accelerometer and having 3 or more days with a minimum of 8

hours/day wear time – yes or no).

However, 13.4% (N=196) of participants had at least one missing covariate value for any

of the respective variables listed in Supplement Table 1 not including accelerometer variables

and the indicator variable of accelerometer data being collected on at least one weekend day.

Thus, predicted probabilities could not be predicted for participants with at least one missing

(10)

9

carried out first to have complete covariate data. Specifically, five imputed data sets were

generated using SAS PROC MI FCS method – only assuming there is a joint distribution for all

variables used. Included in the FCS models are age, gender, BMI, annual family income level,

background, ethnicity identity, child language preference, acculturation, immigrant generation,

dyslipidemia, self-reported physical activity, and to account for the complex survey design we

included the sampling stratum, and sampling weight. Again, site was not included as it would

only provide redundant information being part of stratum. We note that despite 196

participants having missing information that only 1.3% (N=228) of the total individual values

(1446 observations multiplied by 12 covariates with at least one missing observation) were

imputed.

Concerning the logistic regression model that will predict the probability of adherence,

convergence issues due to insufficient cells counts were avoided as we only included pairwise

interactions for the variables gender, BMI, age, ethnicity score, annual family income, language

preference, AHISMA, immigration generation, dyslipidemia and all self-reported physical

activity and sedentary behavior. Numerous models were evaluated by including different

combinations of covariates and their two-way interactions for optimal fit and convergence. The

final model had reasonable prediction properties as indicated by the averaged C-statistic of

0.79 across the five imputations. We then fit a logistic regression model for each of the 5

imputed datasets and average linear predictors of adherence to obtain a single mean predictor

that is used to calculate the IPW that is multiplied with the HCHS/SOL Youth sampling weight to

obtain the new sampling weight accounting for missing accelerometer data.

(11)

10

To create our basis for comparison with MI and IPW we conducted the analyses ignoring

missing data (N=1,104) which is known as complete case analyses.

Comparison procedure

CCA and the two methods for addressing missingness (MI and IPW) were implemented

in three different scenarios for how accelerometer data can be used for cross sectional analysis

of various questions of interest: (1) accelerometer variables as outcomes and 4-level BMI

grouping as main exposure variable of interest, (2) accelerometer variables as the exposures of

interest on the effect on obesity, and (3) dyslipidemia as an outcome, BMI as the exposure of

interest while adjusting for accelerometer variables (Figure 1). Exploration of these scenarios

are later used to evaluate the ease and difficulties of applicability for other HCHS/SOL’s SOL

Youth investigators.

Accelerometer data as outcome of interest

For illustration purposes, we assessed the cross-sectional association of BMI group with

accelerometer data (SB min/day; MVPA min/day; total counts per day) using linear regression

models adjusting for covariates. Specifically, we fit three models. The minimally adjusted model

adjusted only for study site and accelerometer wear time; model 1 further adjusted for sex and

age; and model 2 further adjusted for annual family income and immigration generation.

Accelerometer data as main exposure of interest

We estimated the association of accelerometer data (SB min/day; MVPA min/day; total

counts per day separately) with obesity using logistic regression models. The minimally adjusted

(12)

11

sex and age; and model 2 further adjusted for annual family income and immigration

generation.

Accelerometer data as covariate

We estimated the association of the 4-level BMI group on dyslipidemia adjusting for

accelerometer data and other relevant covariates. Similar models to previous two scenarios

(13)

12 III. RESULTS

In the SOL Youth study (N=1,466) there was approximately 25% missing accelerometer

data because only 1,238 participants returned the accelerometer, and of those, just 1,104 met

the adherence criteria (≥ 3 days each with ≥ 8 hrs/day wear time).

Table 1.1 presents socio-demographic characteristics, anthropometrics, and

dyslipidemia overall and by adherence to protocol. With a specified significance level of 0.05,

we observe no indication of significant differences between adherent and non-adherent

children except for study site (p-value 0.008). Among adherent children, 29.9% were from San

Diego site, whereas San Diego children comprises 46.0% of the non-adherent. Overall, 51.2% of

the children were girls, 80.9% preferred English as their first language, 48.5% integrated, 39.7%

assimilated, and 11.8% separated or marginalized. Although not significant, there were modest

differences for annual family income by adherence. Among adherent children, 55.0% have an

annual family income of less than $20,000, whereas among non-adherent children only 45.7%.

Table 1.2 presents self-reported physical activity by adherence. There are no differences by

adherence for minutes/day of screen time, nor for times per month spent on activities involving

physical activity, like spots/exercise, leisure non-sport, school, household, and transportation.

Accelerometer data as outcome of interest

Regression coefficient estimates and standard error for the association of BMI group

and accelerometer data sedentary behavior (min/day), MVPA (min/day), and total counts per

day analyzed using CCA, IPW, and MI are shown in Table 2. Sedentary behavior (min/day) tends

to be greater for moderately and severely obese compared to normal weight across all models,

(14)

13

adjusted model compared to the adjusted models. Standard errors decrease when adjusted

models are used, without notably differences between the standard errors of the two adjusted

models. MVPA (min/day) and total counts per day are lower for overweight moderately and

severely obese compared to underweight & normal weight, although not significant. Regarding

the methods of addressing missing data, we observed similar regression coefficient estimates

for model 2 except for estimates for overweight participants which show some difference

between IPW and MI estimates.

Accelerometer data as main exposure of interest

Odds ratio estimates and 95% confidence intervals (CI) of obesity for a 10 and 60

min/day increment of SB and MVPA are shown in Table 3, using CCA, IPW, and MI. Given that

an increment of only one minute in activity is not relevant, we decided to present the OR for

two meaningful increments of time (10 minutes and one hour). There are null differences in the

OR estimated by CCA, IPW, and MI. In the adjusted models, the odds of obesity are larger when

time in sedentary behavior is longer. For example, the odds of obesity are 1.3 times more for

every hour of sedentary behavior in the fully adjusted model (model 3) using CCA, IPW, and MI.

In the case of MVPA, the odds of obesity are lower when time in MVPA is longer. For example,

when using model 1 with multiple imputation, the odds of obesity are 0.37 times less for every

hour of MVPA in the full adjusted model.

Accelerometer data as covariate

Odds ratio estimates and 95% confidence intervals (CI) of dyslipidemia by BMI group

adjusting for SB min/day, wear time and other covariates are shown in Table 4, using CCA, IPW

(15)

14

underweight, normal weight, and overweight across all models. For model 2 we observed

similar IPW and MI odds ratios estimates and in contrast, IPW and MI odds ratio estimates

differ in unadjusted model and model 1. For example, model 1 are 14.4 (95% CI: 7.1, 29.3) and

9.4 (95% CI: 5.7, 15.5) respectively. Note the wide 95% CI calculated for all severely obese

(16)

15 IV. DISCUSSION

In our study, we saw differences in SB and PA adjusted means estimated using CCA, MI or IPW,

despite study site being the only statistical difference between those with and without

accelerometry data. Regarding SB and PA as the main exposure we found that odds ratios

estimates for obesity were similar across all models and usage of CCA, IPW, and MI. However,

for SB and PA as covariates, we found differences between IPW and MI in the minimally

adjusted model and model 1, but not in model 2 which further adjusted for income and

nativity. Thus, model misspecification may be an issue as implementing IPW and MI should

yield similar results.

Because IPWs can be created once and distributed to all investigators, its ease of

implementation proves more advantageous given that HCHS/SOL has a complex survey design

and investigators must use sampling weights in the analyses anyway. IPW adjusted sampling

weights such that those adherent children represent all children in the target population. In

contrast when using multiple imputation, it must be carried out by investigators each time in

correspondence to their aims to align the covariates in the imputation model to the ones in the

outcome model. When investigators are not necessarily trained to conduct multiple imputation,

the method may prove to be cumbersome and leave investigators less inclined to address data

missingness.

Strengths of the study include exploring two modes of addressing missing data and its

discussion of which is easier to implement use. Furthermore, we used a rich and diverse sample

consisting of individuals from four US urban areas, which can be weighted to reflect the

(17)

16

Nonetheless, there are also limitations that should be considered. Primarily that the study

compared CCA, MI, and IPW on a specific data set, where we do not know the truth. Moreover,

when the results from the different analysis methods are dissimilar, we do not know which is

closer to the truth. Further evaluation is needed by conducting simulation studies, where we

know the underlying truth, and can compare the performance of these methods. Another

limitation is that we only used one definition of adherence (i.e., at least three days each with 8

hours of wear time). For example, adherence could have been defined by requiring more

number of wear days or including a weekend day.

In conclusion, further research is needed to better understand whether general

recommendations could be made on which method to use to account for missing

accelerometer data in SOL Youth. Several different variables and models need to be explored

(18)

V. APPENDIX

(19)

Table 1.1 Characteristics of U.S. Hispanic/Latino Youth by Accelerometry Adherence (unweighted N = 1466), SOL Youth 2012-2014.

Characteristic N

Overall N = 1466

Adherent (>=3 adherent

days) N = 1104

Non-adherent (failed to return or

<3 adherent days) N = 362

P-value Weighted % (95% CI) Weighted % (95% CI) Weighted % (95% CI) Gender

Male 728 51.2 (48.0, 54.4) 50.1 (46.4, 53.9) 54.0 (47.3, 60.6) 0.335 Female 738 48.8 (45.6, 52.0) 49.9 (46.1, 53.6) 46.0 (39.4, 52.7)

Age, yrs

8-10 494 32.4 (29.5, 35.4) 33.9 (30.5, 37.4) 28.4 (23.0, 34.5) 0.194 11-12 350 21.8 (19.0, 24.9) 22.4 (19.5, 25.6) 20.2 (14.8, 27.0)

13-14 371 22.3 (19.9, 24.9) 21.3 (18.7, 24.2) 24.7 (19.7, 30.5) 15-16 251 23.6 (20.6, 26.9) 22.4 (19.0, 26.2) 26.7 (20.8, 33.5) Language of preference

English 1175 79.5 (75.5, 83.0) 80.9 (76.5, 84.7) 75.8 (67.9, 82.2) 0.208 Spanish 287 20.5 (17.0, 24.5) 19.1 (15.3, 23.5) 24.2 (17.8, 32.1)

Body mass index Underweight & Normal

Weight 749 53.2 (49.6, 56.8) 52.4 (48.2, 56.6) 55.2 (48.8, 61.4) 0.352 Overweight 305 20.1 (17.5, 22.9) 21.3 (18.2, 24.7) 16.9 (12.8, 21.9)

Obese 262 16.8 (14.5, 19.5) 16.1 (13.4, 19.1) 18.8 (14.3, 24.2) Severely Obese 150 9.9 (8.1, 12.2) 10.2 (8.1, 12.9) 9.2 (5.8, 14.3) Obesity Status

Obese 412 26.7 (23.8, 29.9) 26.3 (22.9, 30.0) 28.0 (22.2, 34.5) 0.650 Non-Obese 1054 73.3 (70.1, 76.2) 73.7 (70.0, 77.1) 72.0 (65.5, 77.8)

Dyslipidemia

Yes 327 21.6 (19.1, 24.5) 20.1 (17.4, 23.2) 25.7 (20.2, 32.1) 0.101 No 1088 78.4 (75.5, 80.9) 79.9 (76.8, 82.6) 74.3 (67.9, 79.8)

Generational Status

1st (Foreign-born) 290 19.1 (16.2, 22.3) 17.7 (14.8, 21.0) 22.7 (16.4, 30.6) 0.386 2nd (U.S.-born)

928 64.1 (60.2, 67.8) 65.7 (61.1, 70.1) 59.9 (52.5, 66.9) 3rd+ (U.S.-born)

224 16.8 (13.8, 20.3) 16.6 (13.2, 20.6) 17.4 (12.5, 23.6) Annual family income

<=$20,000 750 52.4 (47.6, 57.2) 55.0 (49.7, 60.2) 45.7 (36.9, 54.7) 0.110 $21,000-$40,000 457 31.7 (27.7, 36.1) 29.2 (25.0, 33.8) 38.3 (30.5, 46.7)

>$40,000 210 15.9 (12.8, 19.5) 15.8 (12.4, 19.9) 16.1 (9.7, 25.3) AHISMA

(20)

19

Characteristic N

Overall N = 1466

Adherent (>=3 adherent

days) N = 1104

Non-adherent (failed to return or

<3 adherent days) N = 362

P-value Weighted %

(95% CI)

Weighted % (95% CI)

Weighted % (95% CI)

Assimilated (US) 624 39.7 (36.2, 43.3) 40.2 (36.4, 44.1) 38.2 (31.7, 45.2) Separated

or Marginalized 167 11.8 (9.7, 14.3) 11.8 (9.5, 14.5) 11.9 (8.1, 17.0) Site

Bronx 422 35.9 (31.7, 40.3) 37.5 (33.0, 42.2) 31.7 (24.7, 39.6) 0.008 Chicago 372 16.2 (13.5, 19.2) 18.1 (14.9, 21.9) 11.0 (7.6, 15.5)

Miami 263 13.6 (11.0, 16.7) 14.5 (11.6, 17.9) 11.3 (7.5, 16.8) San Diego 409 34.3 (29.5, 39.5) 29.9 (25.3, 35.0) 46.0 (37.0, 55.3) Background

Central and South

American 180 10.0 (8.0, 12.4) 10.2 (7.9, 12.9) 9.6 (6.6, 13.8) 0.937 Cuban 103 5.4 (4.0, 7.2) 5.8 (4.3, 7.9) 4.3 (2.3, 7.6)

(21)

Table 1.2Self-Reported Physical Activity of U.S. Hispanic/Latino Youth by Accelerometry Adherence (unweighted N = 1466), SOL Youth 2012-2014.

Self-reported Physical Activity

N Overall

Overall

Mean (95% CI)

Adherent

(>=3 adherent days)

Mean (95% CI)

Non-adherent

(failed to return or <3 adherent days)

Mean (95% CI) P-value

Screen activities (TV, video, computer, phone), min/day

1393 476 (451, 500) 469 (445, 493) 493 (438, 548) 0.4114

Household activity, times/month

1455 20 (18, 21) 20 (18, 21) 20 (18, 23) 0.7455

Sports/exercise activity, times/month

1463 197 (186, 207) 198 (185, 210) 194 (172, 216) 0.7870

Leisure non-sport activity, times/month

1463 88 (85, 90) 88 (85, 92) 85 (79, 91) 0.3514

School activity, times/month

1463 48 (46, 49) 48 (46, 50) 46 (43, 50) 0.2699

Transportation activity, times/month

(22)

21 Table 2. Regression Coefficients Estimates (Standard Error) for the Association between BMI and Accelerometer Outcomes, SOL Youth 2012-2014.

Minimally Adjusted Model 1 Model 2

Outcome BMI Group CCA N=1104 IPW N=1104 MI N=1466 CCA N=1104 IPW N=1104 MI N=1466 CCA N=1104 IPW N=1104 MI N=1466

Se d en ta ry B eh av io r (mi n s/ d ay )

Under / normal

weight (REF) 0 0

0 0 0 0 0 0 0.00

Overweight -1.92

(6.85) 6.30 (7.87) -3.52 (7.10) -1.43 (5.77) -1.03 (5.51) -3.90 (6.38) -1.03 (5.77) -0.18 (5.45) -3.16 (6.37) Moderately

Obese (7.35)4.71 13.23 (8.42) (7.71)3.91 (5.57)11.74 (5.60)12.05 (6.32)11.30 11.36 (5.74) (6.02)12.29 (6.34)11.64

Severely Obese 9.88

(8.92) 15.88 (9.31) 11.90 (10.20) 17.01 (7.63) 14.36 (7.24) 15.20 (9.25) 16.29 (7.77) 14.39 (7.43) 15.96 (8.96) M o d er ate to Vig o ro u s Ph ys ic al A cti vi ty (mi n s/ d ay )

Under / normal weight (REF) 0 0 0 0 0 0 0 0 0

Overweight -1.22

(2.30) 0.45 (2.83) -0.90 (2.15) -1.61 (2.15) 0.11 (2.37) -0.81 (2.03) -2.02 (2.08) -1.99 (2.22) -0.98 (2.06) Moderately Obese -5.74 (2.01) -3.87 (2.13) -4.54 (2.20) -7.15 (1.95) -5.33 (2.07) -5.87 (2.10) -7.44 (1.97) -7.93 (2.07) -5.93 (2.13)

Severely Obese -8.02

(2.61) -5.53 (2.92) -7.49 (2.68) -9.56 (2.37) -6.69 (2.61) -8.23 (2.55) -9.43 (2.42) -8.72 (2.64) -8.60 (2.53) To ta l C o u n ts pe r da

y Under / normal weight (REF) 0 0 0 0 0 0 0 0 0

Overweight 153.73

(13055.51) 1134.10 (14008.50) 5240.19 (12601.69) -1272.24 (12264.42) 4102.08 (12388.86) 5767.74 (11937.67) -7236.83 (10225.53) -8727.67 (10252.49) 4906.81 (12199.39) Moderately Obese -25046.63 (9589.08) -21937.33 (9511.66) -15429.29 (11995.26) -33844.72 (9070.26) -26133.49 (9748.10) -24407.71 (11376.45) -34900.99 (8978.82) -38051.32 (9490.40) -24832.52 (11544.83)

Severely Obese -29477.99

(14153.58) -23290.32 (15477.97) -27254.67 (13745.36) -38693.76 (12693.65) -26314.50 (13902.33) -31674.55 (12642.80) -37383.82 (12911.68) -34594.71 (14036.85) -33322.50 (12420.41)

* Minimally Adjusted: Outcome = 4-Level BMI + Site + Hrs/day monitor worn * Model 1: Outcome = 4-Level BMI + Site + Hrs/day monitor worn + Age + Sex

(23)

22 Table 3. Odds Ratio (95% CI) for the Association between Accelerometer Sedentary Behavior and MVPA with Obesity, SOL Youth 2012-2014.

Minimally Adjusted Model 1 Model 2

Exposure Increment CCA N=1104 IPW N=1104 MI N=1466 CCA N=1104 IPW N=1104 MI N=1466 CCA N=1104 IPW N=1104 MI N=1466

Se d e n tar y B e h av io r ( x m in s/d ay

) 10 MINUTES

1.02 (0.99, 1.04) 1.03 (1.00, 1.07) 1.02 (0.99, 1.04) 1.05 (1.01, 1.08) 1.05 (1.01, 1.09) 1.04 (1.01, 1.07) 1.05 (1.02, 1.08) 1.05 (1.01, 1.08) 1.04 (1.01, 1.07)

60 MINUTES 1.10

(0.95, 1.28) 1.22 (0.98, 1.52) 1.10 (0.94, 1.29) 1.32 (1.09, 1.59) 1.36 (1.09, 1.70) 1.27 (1.05, 1.52) 1.32 (1.10, 1.59) 1.33 (1.09, 1.62) 1.28 (1.06, 1.54) M o d e rate t o Vi go ro u s Ph ysi ca l Ac ti vi ty ( x m in s/d ay

) 10 MINUTES (0.78, 0.94)0.86 (0.80, 0.98)0.89 (0.81, 0.97)0.88 (0.74, 0.90)0.81 (0.76, 0.94)0.85 (0.77, 0.93)0.85 (0.73, 0.90)0.81 (0.73, 0.91)0.82 (0.77, 0.93)0.85

60 MINUTES 0.40

(0.22, 0.70) 0.48 (0.26, 0.90) 0.48 (0.28, 0.81) 0.29 (0.16, 0.53) 0.37 (0.20, 0.71) 0.38 (0.22, 0.65) 0.28 (0.16, 0.52) 0.29 (0.16, 0.55) 0.37 (0.21, 0.64)

* Minimally Adjusted: logit{Obesity} = Accelerometer exposure + Site + Average Hrs/day monitor worn * Model 1: logit{Obesity} = Accelerometer exposure + Site + Average Hrs/day monitor worn + Age + Sex

* Model 2: logit{Obesity} = Accelerometer exposure + Average Hrs/day monitor worn + Age + Sex + Site + 3-Level Annual Family Income + Nativity

Table 4. Odds Ratio (95% CI) for the Association between BMI and Dyslipidemia, SOL Youth 2012-2014.

Minimally Unadjusted Model 1 Model 2

BMI Group CCA N=1104 IPW N=1104 MI N=1466 CCA N=1104 IPW N=1104 MI N=1466 CCA N=1104 IPW N=1104 MI N=1466 Under / normal

weight (REF) 1 1 1 1 1 1 1 1 1

Overweight 2.5 (1.5, 4.1) 3.9 (1.8, 8.2) 2.2 (1.4, 3.6) 2.4 (1.5, 4.0) 3.5 (1.8, 6.8) 2.2 (1.4, 3.6) 2.4 (1.4, 4.1) 2.4 (1.3, 4.3) 2.2 (1.4, 3.6) Moderately Obese 7.3 (4.4, 12.0) 12.3 (6.0, 25.6) 6.7 (4.3, 10.6) 7.2 (4.4, 11.9) 11.8 (6.0, 22.9) 6.8 (4.3, 10.8) 7.2 (4.2, 12.3) 7.7 (4.3, 13.9) 6.8 (4.2, 10.8)

Severely Obese 11.6 (6.7, 20.0) 16.1 (7.3, 35.3) 9.5 (5.8, 15.5) 11.0 (6.3, 19.1) 14.4 (7.1, 29.3) 9.4 (5.7, 15.5) 11.5 (6.5, 20.3) 9.7 (5.2, 18.4) 9.3 (5.7, 15.4) * Minimally Adjusted: logit{Dyslipidemia} = Sedentary Behavior min/day + 4-Level BMI + Site + Average Hrs/day monitor worn

* Model 1: logit{Dyslipidemia} = Sedentary Behavior min/day + 4-Level BMI + Site + Average Hrs/day monitor worn + Age + Sex

(24)

23 Supplement Table 1. Number of Missing Observations for Variables used in Imputation and Inverses Probability Weighting

Covariate Label Scale N N Miss

GENDER Gender (Male/Female) (Child) Binary 1466 0

BMIGRP Body Mass Index Percentile Groups (Child) Ordinal 1466 0

AGE_C4 4-Level Age Groups (Child) Ordinal 1466 0

BACKGROUND_C7 Background 7-Category Classification (Child) Nominal 1466 0

AHISMA_C3 AHISMA 3-Categry Classification Nominal 1465 1

PAE_TRANSPORT Transportation activity, times/mo Continuous 1464 2

PAE_SPORT Sports/exercise activity, times/mo Continuous 1463 3

PAE_LEISURENONSPORT Leisure non-sport activity, times/mo Continuous 1463 3

PAE_SCHOOL School activity, times/mo Continuous 1463 3

ETHID_C4_QUART Ethnic identity categories based upon variable quartile cutpoints Ordinal 1462 4

LANG_PREF Language Preference (Child) Binary 1462 4

PAE_HOUSEHOLD Household activity, times/mo Continuous 1455 11

IMGEN_PR Immigrant generation (Child) Ordinal 1442 24

INCOME_C3_PARENTˠ Annual Household Income – Parent Ordinal 1417 49

DYSLIPIDEMIA Dyslipidemia (Child) (1=Yes, 0=No) Binary 1415 51

PAE_SCREEN_MIN Screen activities (TV, video, computer, phone), min/day Continuous 1393 73

WKENDDAY_INCLUDED* Summarized activity includes at least one adherent weekend (1=Yes, 0=No) Binary 1238 228

CNTS_MIN_DAY* Counts/min (average) per day monitor was worn Continuous 1104 362

SED_DAY* Min/day (average) of sedentary activity (<17 counts/15 sec) Continuous 1104 362 LIGHT_DAY* Min/day (average) of light activity (18-440 counts/15 sec) Continuous 1104 362 MOD_DAY* Min/day (average) of moderate activity (441-872 counts/15 sec) Continuous 1104 362 VIG_DAY* Min/day (average) of vigorous activity (>=873 counts/15 sec) Continuous 1104 362

* Variables excluded from multiple imputation for inverse probability weighting

(25)

VI. Acknowledgments

This senior honors thesis is the culmination of a collaborative effort with professors from various statistical and epidemiological backgrounds. I would like to express my sincerest appreciation to Dr. Daniela Sotres-Alvarez, not only for serving as my primary thesis advisor in the Department of Biostatistics but also for being a superb mentor. I thank Dr. Jianwen Cai and Dr. Kelly Evenson for being readers and providing relevant feedback. I thank Jeff Oberhaus for his significant and lasting impact on my undergraduate path, linking me to the Hispanic

(26)

25 VII. REFERENCES

1. Ogden, C. L., Carroll, M. D., Lawman, H. G., Fryar, C. D., Kruszon-Moran, D., Kit, B. K., & Flegal, K.

M. (2016). Trends in obesity prevalence among children and adolescents in the United States, 1988-1994 through 2013-2014. Jama, 315(21), 2292-2299.

2. Simmonds, M., Llewellyn, A., Owen, C. G., & Woolacott, N. (2016). Predicting adult obesity from

childhood obesity: a systematic review and meta‐analysis. Obesity reviews, 17(2), 95-107.

3. Singh, G. K., Stella, M. Y., Siahpush, M., & Kogan, M. D. (2008). High levels of physical inactivity and

sedentary behaviors among US immigrant children and adolescents. Archives of pediatrics & adolescent medicine, 162(8), 756-763.

4. Isasi, C. R., Carnethon, M. R., Ayala, G. X., Arredondo, E., Bangdiwala, S. I., Daviglus, M. L.,

Delamater, A. M., Eckfeldt, J. H., Perreira, K., Himes, J. H., Kaplan, R. C., & Van Horn, L. (2014). The Hispanic Community Children's Health Study/Study of Latino Youth (SOL Youth): design, objectives, and procedures. Annals of epidemiology, 24(1), 29–35.

https://doi.org/10.1016/j.annepidem.2013.08.008

5. Evenson, K. R., Arredondo, E. M., Carnethon, M. R., Delamater, A. M., Gallo, L. C., Isasi, C. R., ... &

Sotres-Alvarez, D. (2019). Physical Activity and Sedentary Behavior among US Hispanic/Latino Youth: The SOL Youth Study. Medicine and science in sports and exercise, 51(5), 891-899.

6. Piercy, K. L., Troiano, R. P., Ballard, R. M., Carlson, S. A., Fulton, J. E., Galuska, D. A., ... & Olson, R. D. (2018). The physical activity guidelines for Americans. Jama, 320(19), 2020-2028.

7. Van Buuren, S. (2007). Multiple imputation of discrete and continuous data by fully conditional

specification. Statistical methods in medical research, 16(3), 219-242.

8. Seaman, S. R., & White, I. R. (2013). Review of inverse probability weighting for dealing with missing

data. Statistical methods in medical research, 22(3), 278-295.

9. Sorlie, P. D., Avilés-Santa, L. M., Wassertheil-Smoller, S., Kaplan, R. C., Daviglus, M. L., Giachello, A. L., Schneiderman, N., Raij, L., Talavera, G., Allison, M., Lavange, L., Chambless, L. E., & Heiss, G. (2010). Design and implementation of the Hispanic Community Health Study/Study of

Latinos. Annals of epidemiology, 20(8), 629–641. https://doi.org/10.1016/j.annepidem.2010.03.015

10.Romanzini, M., Petroski, E. L., Ohara, D., Dourado, A. C., & Reichert, F. F. (2014). Calibration of

ActiGraph GT3X, Actical and RT3 accelerometers in adolescents. European journal of sport science, 14(1), 91-99.

11.Cain, K. L., Sallis, J. F., Conway, T. L., Van Dyck, D., & Calhoon, L. (2013). Using accelerometers in youth physical activity studies: a review of methods. Journal of physical activity & health, 10(3), 437– 450. https://doi.org/10.1123/jpah.10.3.437

12.Choi, L., Liu, Z., Matthews, C. E., & Buchowski, M. S. (2011). Validation of accelerometer wear and

nonwear time classification algorithm. Medicine and science in sports and exercise, 43(2), 357.

13.Unger, J. B., Gallaher, P., Shakib, S., Ritt-Olson, A., Palmer, P. H., & Johnson, C. A. (2002). The

AHIMSA acculturation scale: A new measure of acculturation for adolescents in a multicultural society. The Journal of Early Adolescence, 22(3), 225-251.

14.Barlow, S. E. (2007). Expert committee recommendations regarding the prevention, assessment, and

treatment of child and adolescent overweight and obesity: summary report. Pediatrics, 120(Supplement 4), S164-S192.

(27)

26

16.American Diabetes Association. (2017). 2. Classification and diagnosis of diabetes. Diabetes care,

40(Supplement 1), S11-S24.

17.Qi, Q., Hua, S., Perreira, K. M., Cai, J., Van Horn, L., Schneiderman, N., Thyagarajan, B., Delamater, A. M., Kaplan, R. C., & Isasi, C. R. (2017). Sex Differences in Associations of Adiposity Measures and Insulin Resistance in US Hispanic/Latino Youth: The Hispanic Community Children's Health Study/Study of Latino Youth (SOL Youth). The Journal of clinical endocrinology and metabolism, 102(1), 185–194. https://doi.org/10.1210/jc.2016-2279

18.Strizich, G., Kaplan, R. C., Sotres-Alvarez, D., Diaz, K. M., Daigre, A. L., Carnethon, M. R., Vidot, D. C., Delamater, A. M., Perez, L., Perreira, K., Isasi, C. R., & Qi, Q. (2018). Objectively Measured Sedentary Behavior, Physical Activity, and Cardiometabolic Risk in Hispanic Youth: Hispanic Community Health Study/Study of Latino Youth. The Journal of clinical endocrinology and metabolism, 103(9), 3289–3298. https://doi.org/10.1210/jc.2018-00356

19.Palta, P., McMurray, R. G., Gouskova, N. A., Sotres-Alvarez, D., Davis, S. M., Carnethon, M., Castañeda, S. F., Gellman, M. D., Hankinson, A. L., Isasi, C. R., Schneiderman, N., Talavera, G. A., & Evenson, K. R. (2015). Self-reported and accelerometer-measured physical activity by body mass index in US Hispanic/Latino adults: HCHS/SOL. Preventive medicine reports, 2, 824–828.

https://doi.org/10.1016/j.pmedr.2015.09.006

20.Qi, Q., Strizich, G., Merchant, G., Sotres-Alvarez, D., Buelna, C., Castañeda, S. F., ... & Moncrieft, A. E. (2015). Objectively measured sedentary time and cardiometabolic biomarkers in US

Hispanic/Latino adults: the Hispanic Community Health Study/Study of Latinos (HCHS/SOL). Circulation, 132(16), 1560-1569.

21.Arredondo, E. M., Sotres-Alvarez, D., Stoutenberg, M., Davis, S. M., Crespo, N. C., Carnethon, M. R., Castañeda, S. F., Isasi, C. R., Espinoza, R. A., Daviglus, M. L., Perez, L. G., & Evenson, K. R. (2016). Physical Activity Levels in U.S. Latino/Hispanic Adults: Results From the Hispanic Community Health Study/Study of Latinos. American journal of preventive medicine, 50(4), 500–508.

https://doi.org/10.1016/j.amepre.2015.08.029

Figure

Updating...

Related subjects :