Research Article 

Clusters of Healthy and Unhealthy Eating Behaviors Are Associated With Body Mass Index 
Among Adults William J. Heerman, MD, MPH1,2,3; Natalie Jackson, MPH2,3; Margaret 
Hargreaves, MD4; Shelagh A. Mulvaney, PhD2,5,6; David Schlundt, PhD3; Kenneth A. 
Wallston, PhD5; Russell L. Rothman, MD, MPP2,3 
ABSTRACT Objective: To identify eating styles from 6 eating behaviors and test their association with body mass 
in- dex (BMI) among adults. Design: Cross-sectional analysis of self-report survey data. Setting: Twelve primary 
care and specialty clinics in 5 states. Participants: Of 11,776 adult patients who consented to participate, 9,977 
completed survey questions. Variables Measured: Frequency of eating healthy food, frequency of eating unhealthy 
food, breakfast frequency, frequency of snacking, overall diet quality, and problem eating behaviors. The primary 
depen- dent variable was BMI, calculated from self-reported height and weight data. Analysis: k-Means cluster 
analysis of eating behaviors was used to determine eating styles. A categorical variable representing each eating 
style cluster was entered in a multivariate linear regression predicting BMI, controlling for covariates. Results: Four 
eating styles were identified and defined by healthy vs unhealthy diet patterns and engage- ment in problem eating 
behaviors. Each group had significantly higher average BMI than the healthy eating style: healthy with problem 
eating behaviors (b 1⁄4 1.9; P < .001), unhealthy (b 1⁄4 2.5; P < .001), and unhealthy with problem eating behaviors 
(b 1⁄4 5.1; P < .001). Conclusions and Implications: Future attempts to improve eating styles should address not only 
the consumption of healthy foods but also snacking behaviors and the emotional component of eating. Key Words: 
obesity, body mass index, nutrition, behavioral science, eating patterns (J Nutr Educ Behav. 2017;-:1-7.) Accepted 
February 5, 2017. 
INTRODUCTION 
Fighting  obesity  requires  more  than  eatinglessandexercisingmore.Certainly  caloric  intake  and  physical  activity  fre-  quency  and 
intensity are the primary determinants of energy balance. Howev- er, both weight loss and maintenance 
of  healthy  weight  are  best  achieved  through  sustained  adherence  to  a  broader  range  of  healthy  eating  (eg,  increased 
fruit/vegetable  intake)  and  physical  activity  (eg, reduced seden- tary time) behaviors. Based on a foun- dational understanding of 
the complex and multilevel determinants of healthy 
eating and healthy physical activity, much work has been done to develop interventions that facilitate these healthy behaviors.1,2 
In the US obesity affects 36.5% of adults and 17.0% of children.3 With the long- term health complications of obesity, including 
diabetes, heart disease, and cancer, the continued effort to under- stand which eating behaviors support 1General Pediatrics, 
Vanderbilt University Medical Center, Nashville, TN 
achieving a healthy weight is of 
para- 2Department of Internal Medicine and Public Health, Vanderbilt University, Nashville, TN 
mount importance.4,5 3Center 
for Health Services Research, Vanderbilt University, Nashville, TN 
Changes in eating behaviors 
have 4Department of Internal Medicine, Meharry Medical College, Nashville, TN 
been independently associated 
with 5School of Nursing, Vanderbilt University, Nashville, TN 
long-term changes in weight.6 In 
6Department of Biomedical Informatics, Vanderbilt University, Nashville, TN 
particular, behaviors such as 
skipping Conflict of Interest Disclosure: The authors’ conflict of interest disclosures can be found online 
meals, snacking, drinking 
sweetened with this article on www.jneb.org. 
beverages, and eating fast food 
were Address for correspondence: William J. Heerman, MD, MPH, General Pediatrics, Vander- 
frequently studied as potential 
contrib- bilt University Medical Center, 2146 Belcourt Ave, 2nd Fl, Nashville, TN 37212; Phone: 
utors to obesity.7-9Strong 
evidence links (615) 322-7080; Fax: (615) 343-7650; E-mail: Bill.Heerman@vanderbilt.edu 
skipping breakfast with obesity, 
particularly ©2017 Society for Nutrition Education and Behavior. Published by Elsevier, Inc. All rights 
among children.10,11However, a 
recent reserved. 
meta-analysis of 153 articles 
exam- http://dx.doi.org/10.1016/j.jneb.2017.02.001 
ining the association between eating 
Journal of Nutrition Education and Behavior Volume -, Number -, 2017 1 
 
behaviors  and obesity among adults and children concluded that evidence was insufficient to draw meaningful conclusions owing 
to  2  major  limita-  tions.12 First, most existing studies did not adequately consider potential confounders of the proposed associa- 
tions.  Second,  most  existing  studies  considered  only  how  a  single  eating  behavior  was  associated  with  obesity  and  did  not 
account for the contribu- tion of other potentially related eating behaviors. 
Therefore,  the  objective  of  this  study  was  to  address  the  existing  gap  in  the  literature  by  considering  how  a  broad  range  of 
eating  behaviors  relate  to body mass index (BMI), controlling for po- tential confounders including socio- demographics (ie, age, 
gender,  race/  ethnicity)  and  physical  activity  in  a  large  sample  of  adults.  This  was  accom-  plished  by  first  considering  how  6 
measures  of  diet  quality  and  eating  be-  haviors  clustered  together  into  patterns  of  eating  styles,  and  then  testing  whether  those 
eating styles were asso- ciated with BMI. 

METHODS Participants and Procedures 

The  researchers  conducted  analyses  on  cross-sectional  survey  data  collected  be-  tween  August,  2014 and November, 2015. The 
primary  mode  of  survey  administration  was  via  an  electronic  survey  delivered  on  a  tablet  computer  or  an  e-mailed  survey  link 
using  REDCap.13Participants  recruited  in  per-  son  also  had  the  option  to  complete  a  paper  survey.  The  survey  consisted  of 72 
items  that  queried  participants  about  demographic  and  background  informa-  tion  as  well  as  health  behaviors,  and  took 
participants  an  average  of  17  mi-  nutes  to  complete.  This  study  received  approval  from the Vanderbilt Univer- sity Institutional 
Review  Board.  All  par-  ticipants  signed  informed  consent  before  participating  and  received  a  $10  gift  card  for  completing  the 
survey. 
Patients  were  recruited  from  medical  clinics  in  3  health  networks  from  the  Pa-  tient  Centered  Outcomes  Research 
Institute–funded  Mid-South  Clinical  Data  Research  Network  (CDRN).14  The  Mid-South  CDRN  integrates  a  clinical  data 
infrastructure across the US, con- sisting of: (1) Vanderbilt University Medical Center partnering with 
2 Heerman et al Journal of Nutrition Education and Behavior Volume -, Number -, 2017 
Meharry  Medical  College,  (2)  the  Van-  derbilt  Healthcare  Affiliated  Network,  (3)  Greenway  Health,  and  (4)  the  Caro-  linas 
Collaborative,  a  consortium  of  4  aca-  demic  health  systems  and  multiple  community  health  systems  across  North  Carolina and 
South  Carolina.  This  study  recruited  participants  from  Vanderbilt,  the  Vanderbilt Healthcare Affiliated Network, and Greenway 
Health.  The  final  sample  of  partici-  pants  was  recruited  from  12  medical  clinics,  a  mixture  of  primary  care  and  subspecialty 
clinics.  Recruitment  occurred  using  4  principal  approaches:  face-to-face  recruitment  in  medical  clinics,  an  e-mail  sent  directly 
from  a  patient's  medical  provider,  an  e-mail  sent  from  the  research team, or an e- mail sent from a clinic's medical direc- tor. To 
be  eligible,  participants:  (1)  were  aged  $18  years;  (2)  had  $1  clinic  note  in  the  electronic  health  record  (EHR)  since  April  30, 
2009;  (3)  had  $2  weight  measures  in  the  EHR  since  April 30, 2009; and (4) had $1 height measurement in the EHR after age 18 
years.  The  survey  was  conducted  exclusively  in  English.  Participants  were  excluded  if  they  had  a  mental  condition  or  visual 
acuity  that  pre-  cluded  participation  (assessed  by  the  research  team  at  the  time  of  face-to-  face survey administration). Figure 1 
shows a diagram of recruitment flow for the study. 

Measures 
The  primary  independent  (predictor)  variables consisted of 6 self-report eating behaviors: (1) frequency of eating healthy food (2 
items);  (2)  frequency  of  eating  unhealthy  food  (3  items);  (3)  breakfast  frequency  (1  item);  (4) fre- quency of snacking (1 item); 
(5)  overall  diet  quality (1 item); and (6) problem eating behaviors (4 items). A complete list of survey items included in this anal- 
ysis and how they were scored is avail- able in the Supplementary Material. The frequency of healthy and un- healthy food scores 
as  well  as  the  mea-  sure  of  problem  eating behaviors were developed based on a factor analysis of 9 items from the survey, each 
of  which  was answered on a 4-point response scale from 1 1⁄4 never to 4 1⁄4 often; once a day or more. The healthy food variable 
was a composite frequency of eating fruits and vegetables (range, 2–8). The unhealthy food variable was a 
composite  frequency  of  consuming  fast  food,  sugary  drinks,  and  desserts  (range,  3–12).  Self-reported  frequencies  of  eating 
breakfast  and  snacking  were  scored  on  a  3-point  response  scale  from  0  1⁄4  never  to  2  1⁄4  daily.  Self-reported  diet  quality  was 
scored  on  a  5-point  response  scale  from  1  1⁄4  poor  to  5  1⁄4  excellent.  These  items  were  used  in  previous  survey  research  and 
demon-  strated  validity  in those contexts (ie, a prospective cohort of 3,000 hospital- ized patients with acute coronary syn- drome 
or  heart  failure15;  and  67,928  participants  enrolled  in  the  Southern  Community  Cohort  Study).16  The  4-  item  measure  of 
problem  eating  be-  haviors  was  based on a previously vali- dated scale; in this study it had high internal consistency (Cronbach a 
1⁄4 .83).17 
The  primary  dependent  variable  was  BMI, calculated from self-reported height and weight data. Demographic variables were 
all  self-reported and included age, gender, race/ethnicity, and household income. Self-reported physical activity was assessed as a 
potential confounder with a single item that ranged from 11⁄4 I am very inactive to 5 1⁄4I am active most days. 
Demographiccharacteristicsincluding age, gender, household income, and race/ethnicity were summarized using mean and SD 
for  continuous  variables  and  proportions  for  categorical  vari-  ables.  Annual  household  income  was  categorized  into  3 mutually 
exclusive  categories  (<$35,000,  $35,000  to  $75,000,  and>$75,000).  Race/ethnicity  was  categorized  into  4  mutually  exclu-  sive 
categories (white, non-Hispanic; black, non-Hispanic; Hispanic; and other, non-Hispanic). 

Analytic Strategy 
k-Means  cluster  analysis  was  used  for  the  6  independent  eating behavior variables to determine eating styles. A scree plot of the 
weighted  sum  of  squares  and  the  proportional  reduc-  tion  of  the  weighted  sum  of  squares  against  a  range  of  potential  clusters 
(range,  1–20)  was  used  to  identify  the  optimal  number  of  clusters neces- sary to prespecify in the k-means algo- rithm.18 Based 
on  the  scree  plot,  4  initial  centroids  were  randomly iden- tified and 10,000 repetitions of the k- means algorithm were conducted 
to achieve convergence of the solution 
 
(ie,  no  observations  changed clusters for the final centroids).19 A k-means cluster analysis was first conducted on roughly half of 
the  sample  (re- cruited from 1 of the 3 health systems in the CDRN). The analysis was then repeated on the other half of the sam- 
ple. These 2 independent cluster ana- 
Journal of Nutrition Education and Behavior Volume -, Number -, 2017 Heerman et al 3 
Figure  1.  Of  the  11,776  participants  who  consented  to  participate,  10,807  (91.8%)  completed  at  least  70%  of  survey  items and 
9,977  had  complete  data  for  analysis.  The  response rate as defined by the Council of American Survey Research Organiza- tions 
was 16.4%.23 
lyses  generated  similar  cluster  centers  with  nearly  identical  relationships  with  BMI.  Consequently,  the  k-means cluster analysis 
was  repeated  on  the  total  sample  (n  1⁄4  9,977), which is re- ported here. Because k-means cluster analysis is sensitive to missing 
data, only individuals with complete data 
on BMI and the 6 predictor variables were included. 
To  represent  the  patterns  of  eating  behaviors in each cluster, each of the 6 predictor variables was standardized and presented 
as  the  mean  z-score  of  each  variable  as  it  was  grouped  per  cluster  (this was also done within the k-means clustering algorithm). 
This allowed for comparison across scales with different numbers of items and score ranges. 
A  categorical  variable  with  levels  rep-  resenting  each  replicable  cluster  served  as  the  primary  independent  variable  for  the 
subsequent  set  of  analyses.  The  primary  analysis  used  multiple  linear  regression  with  BMI  as  the  dependent variable, adjusting 
for  age  (continuous),  gender  (dichotomous),  household  in-  come  (ordinal),  race/ethnicity  (categori-  cal),  and  physical  activity 
(continuous).  A  secondary  analysis  in  which  3  sepa-  rate  logistic  regressions  compared  cate-  gorical  weight  status  as  the 
dependent  variable  and  eating  cluster  as the inde- pendent variable (controlling for the same covariates) was completed to clarify 
the  clinical  interpretation  of  the  results.  Weight  status  was  defined  using  the  following  cut  points: normal weight (reference)1⁄4 
BMI$18.5  and<25  kg/m2;  overweight1⁄4BMI$25  and<30  kg/m2;  obesity  1⁄4  BMI  $30  and  <40  kg/m2; morbid obesity 1⁄4 BMI 
$40 kg/m2.3 Statistical significance was set at a < .05. 

RESULTS 
Of  the  11,776  respondents  who  con-  sented  to  participate in the survey, 9,977 (84.7%) had complete data on study variables and 
were  included  in  this  analysis.  The  Table  lists  full  demo-  graphic  characteristics  of  the  overall  sample.  The  majority  of 
respondents  were  female  and  self-identified  as non-Hispanic white; 41.0% had an annual household income >$75,000. Although 
the  average  BMI  was  29.3  (SD  7.4),  over  one third of respondents were obese (BMI $30 kg/m2) and almost 20% were morbidly 
obese (BMI $40 kg/m2). 

Eating Style Determination 

k-Means cluster analysis revealed 4 coherent patterns of eating behaviors (Figure 2). Items related to healthy diet 
 
clustered together, so a healthy diet pattern was defined by higher scores on the healthy food scale, lower scores on the unhealthy 
food scale, higher scoresonthegenerallyhealthydietqual- ity item, and higher frequency of eating breakfast. More frequent 
snacking be- haviors clustered with an unhealthy diet pattern. The greatest discrimination between the clusters was based on the 
combination of healthy/unhealthy diet pattern and problem eating behavior scores, which led to 4 patterns of eating styles: 
healthy, healthy with problem eating behaviors, unhealthy, and un- healthy with problem eating behaviors. Healthier clusters of 
eating styles were associated with more frequent vegetable intake, with nearly 75% of participants in the 2 healthy clusters re- 
porting daily vegetable intake compared with just over 50% of those in the 2 un- healthy clusters (P < .001). Healthier clusters of 
eating styles were also asso- ciated with more frequent fruit intake; about 60% of participants in the healthy clusters reported 
daily fruit intake, compared with about 40% in the unhealthy clusters (P < .001). In addition, healthier clusters of eating styles 
were associated with more frequent physical activity; 40% of participants in the healthy cluster reported daily physical activity, 
com- pared with 28.5% in the healthy with problem eating behaviors cluster, 23.8% in the unhealthy cluster, and 
4 Heerman et al Journal of Nutrition Education and Behavior Volume -, Number -, 2017 
Table. Demographic Characteristics of Survey Participants Overall and by Eating Style 
Demographic Characteristics 
Eating Style 
Healthy (n 1⁄4 3,687) 
Unhealthy With Problem Eating Behaviors (n 1⁄4 1,907) Age, y (mean [SD]) 50.6 15.9 54.8 16.1 48.4 16.0 50.1 15.4 46.6 14.2 
Body mass index, kg/m2 
(mean [SD]) 
Healthy With Overall 
Problem Eating (n 1⁄4 9,977) 
Behaviors (n 1⁄4 2,219) 
Unhealthy (n 1⁄4 2,164) 
29.3 7.4 27.3 6.5 28.7 7.0 30.2 7.4 32.9 8.0 
Gender, female (% [n]) 71.8 7,141 63.4 2,329 78.6 1,740 70.1 1,513 82.3 1,559 Race/ethnicity 
White, non-Hispanic 84.2 8,256 81.5 2,948 87.1 1,898 84.8 1,803 85.3 1,607 Black, non-Hispanic 10.2 1,003 12.0 435 6.2 135 
11.7 249 9.8 184 Hispanic 1.9 184 2.3 84 1.9 42 1.3 28 1.6 30 Other, non-Hispanic 3.7 365 4.2 152 4.8 104 2.2 47 3.3 62 
Household income (% [n]) 
<$35,000  23.2  2,289  24.5  804  16.9  345  23.1  457  24.9  440  $35,000–$75,000  35.9  3,545  33.2 1,089 34.1 695 41.4 820 37.5 665 
>$75,000 41.0 4,055 42.4 1,392 49.0 999 35.6 706 37.6 665 
only 14.3% in the unhealthy with prob- lem eating behaviors cluster (P < .001). 

Correlates of Eating Styles 

There  were  significant  differences  in  the  demographic  characteristics  of  in-  dividuals  based  on  the  cluster  of  feeding  style  to 
which  they  were as- signed (Table). Respondents in the un- healthy with problem eating behaviors cluster had the youngest mean 
age and a higher percentage of women. 
The  average  (SD)  BMI  for  individuals  in  the  healthy  eating cluster was 27.3 (6.5) kg/m2, compared with 28.7 (7.0) kg/m2 in 
the  healthy  with  problem  eating,  30.2  (7.4)  kg/m2  in  the  un- healthy cluster and 32.9 (8.0) kg/m2 in the unhealthy with problem 
eating.  Using  the  healthy  cluster  as  the  referent  group  in  an  adjusted  linear  regression  model,  each  of  the  other  clusters  was 
significantly  associated  with  higher  average BMI (Figure 3): unhealthy with problem eating (b 1⁄4 5.0; 95% confi- dence interval 
[CI],  4.66–5.50;  P  <  .001);  unhealthy  (b  1⁄4  2.5;  95%  CI, 2.07– 2.85; P< .001), and healthy with prob- lem eating (b1⁄41.9; 95% 
CI,  1.50–2.27;  P  <  .001).  In  the  secondary  analysis  using  multivariable  adjusted  lo-  gistic  regression,  the  odds  of  being 
overweight,  obese,  and  morbidly  obese  were  all  higher  based  on  cluster  of  eating style. Odds ratios (ORs) for being overweight 
by eating style 
were:  healthy  (reference),  healthy  with  problem  eating  (OR,  1.6;  95%  CI,  1.4–1.8;  P  <  .001),  unhealthy  (OR,  2.0;  95%  CI, 
1.70–2.34;  P  <  .001),  and  unhealthy  with  problem  eating  (OR,  2.8;  95%  CI,  2.3–3.4;  P  <  .001).  The  ORs  for  being  obese  by 
eating  styles  were:  healthy  (reference),  healthy  with  problem  eating  (OR,  2.1;  95%  CI,  1.76–2.55;  P  <  .001),  un-  healthy  (OR, 
2.5;  95%  CI,  2.09–3.07;  P  <  .001),  and  unhealthy  with  prob-  lem  eating  (OR,  5.8;  95%  CI,  4.70–  7.2;  P  <  .001).  The  ORs  for 
being  morbidly  obese by eating styles were: healthy (reference), healthy with prob- lem eating (OR, 2.4; 95% CI, 1.93– 2.89; P < 
.001),  unhealthy (OR, 3.0; 95% CI, 2.44–3.65; P < .001), and un- healthy with problem eating (OR, 8.9; 95% CI, 7.17–11.06; P < 
.001). 

DISCUSSION 
These  data  demonstrate  4  coherent  pat-  ternsofbehaviorsreflecting  eating  styles  associated  with  BMI  independent  of  other 
demographic  covariates.  The  most notable characteristic of these clusters is the discriminant value of combining the frequency of 
eating  healthy  and  unhealthy  foods  with  problem  eating  behaviors  (ie,  overeating,  impulsive  eating,  and  emotional  eating). 
Specifically, respondents who reported a healthy diet pattern had higher average BMI as well as a higher frequency of obesity 
 
and  morbid  obesity  when  they  also  re-  ported  engaging  in  more frequent problem eating behaviors. Respon- dents who reported 
an  unhealthy  diet  pattern  even  in  the  absence  of  problem  eating  behaviors  had  a  higher  BMI  than  did  those  with  a healthy diet 
who  also  avoided  problem  eating  be-  haviors.  Respondents'  BMI  and  those  for  the  proportion  of  individuals  with  obesity  or 
morbid obesity was highest if respondents had an unhealthy diet pattern and simultaneously engaged in problem eating behaviors. 
Thus  these  results  demonstrate  that  both  unhealthy  eating  patterns  and  problem  eating  be-  haviors  independently  increased  the 
odds  of  a  person  being  obese  or  morbidly  obese.  When  unhealthy  eating patterns and problem eating behaviors were combined, 
the  OR  of  obesity was 5.8 and the OR of morbid obesity was 8.9. Thesefindings suggested that clus- ters of eating behaviors may 
be more clinically relevant than any single component dietary assessment. 
These  data  have  direct  implications  for  developing  interventions  and  clin-  ical  recommendations  for  eating  behav-  iors  to 
support  healthy  weight.  One  main  advantage  of  the  clustering  approach  used  for  this  analysis  is  that  it  moved  past  the 
examination of a sin- gle behavior. For example, skipping 
Journal of Nutrition Education and Behavior Volume -, Number -, 2017 Heerman et al 5 
Figure  2.  Four  clusters  of  eating  styles  were  identified  using  k-means  cluster  analysis.  Standardized  z-scores  for  each  of  the 
scales  included  in  the k-means analysis are shown within the final cluster assignment. Scales are oriented so that healthier behav- 
iors have a positive z-score and unhealthier behaviors have a negative z-score. 
breakfast  and  eating  at  fast-food  restau-  rants  each  were associated with higher BMI, both cross-sectionally and in longi- tudinal 
data.12These  data  were  consistent  with  those  findings,  in  that  unhealthy  diet  patterns  including  consumption  of  fast  food  (eg, 
pizza),  sugary  drinks,  and  desserts  (including  candy)  were  associated  with  higher  BMI.  These data also suggested that snacking 
behaviors  clustered  with  unhealthy  eating  pat-  terns,  which  is  important  for  address-  ing  the  potential  effect  of  snacking  on  an 
unhealthy  weight  trajectory.  In  addition,  these  data  highlighted  the importance of problem eating behav- iors such as overeating, 
impulsive eating, and emotional eating. Regard- less of diet quality or the frequency of healthy food consumption, problem eating 
behaviors  confer  an  additional  risk  of  higher  BMI.  Because  the prob- lem eating behaviors questionnaire used in this study (The 
Personal  Dia-  betes Questionnaire-4) had only 4 items, it may be possible to incorpo- rate screening for these types of behav- iors 
into routine clinical practice. 
One  main  limitations  of  this  study  was  that  it  was  cross-sectional,  which  limited  causal  inference.  This  was  partic-  ularly 
important in this context because it was unclear whether unhealthy eating patterns precede obesity or vice versa.20 
In  addition,  all  of  the data were self- reported and therefore could have been affected by a social desirability bias, the direction of 
which  was  uncer-  tain  because the outcome (BMI) was likely related to the bias. However, chart review on approximately half of 
those  surveyed  showed  a  high  correlation  (r  1⁄4  0.94)  between  self-reported  and  objectively  measured weight, which somewhat 
alleviated  the  concern  about  social  desirability  bias.  Although  the  items dealing with diet and exercise were used successfully in 
other  studies,  it  is  a limitation in that the researchers did not revalidate them with this spe- cific population before including them 
in  this  survey,  although  the  pattern  of  associations  of  those  items  with  BMI  in  this  study  provides  strong  evidence  of  their 
construct  validity.  To  minimize  participant  burden,  the  measure  of  physical  activity was limited to a single item, which may not 
have  been  suffi-  ciently sensitive to detect variation in the behavior. Selection bias may also havebeenaconcern,giventhelargeper- 
centage  of  individuals  recruited  online  (or  electronically).  Consequently,  the  sample  studied  was  from  a  predomi-  nantly older, 
female  population  and  may  not  be  generalizable  to  a  broader,  more  racially or ethnically diverse popu- lation. This may suggest 
that  this type of analysis be confirmed in additional de- mographic groups before broad policy or program implications are imple- 
mented.  However,  the geographic varia- tion of the sample mitigated this concern to a degree, and the percentage of patients with 
obesity  in  this  sample  (38%)  closely  mirrored  the  recently  re-  ported  national  prevalence.3  Although  the  items  included  in this 
survey  were  validated  in  other  populations,  and  although  the  strong  associations  with  BMI  in  this  study  provided  preliminary 
evidence  of  construct  validity,  they  were  not  validated  in  this  specific  popu-  lation,  which  may  have led to a misclas- sification 
bias. 

IMPLICATIONS FOR RESEARCH AND PRACTICE 

This  large-sample,  cross-sectional  anal-  ysis  described  unique  coherent  eating  styles,  recognizing  the  importance  of  both 
snacking and problem eating be- haviors within the context of healthy 
 
or  unhealthy  food  choices.  These distinct eating styles were strongly asso- ciated with BMI. This study provided the groundwork 
for  future  longitudinal studies to investigate these potential causal relationships in the service of obesity prevention. Furthermore, 
these  data  were  consistent  with  recommenda-  tions  from  multiple  professional  organi-  zations21,22  and  indicated  that  future 
attempts  to  improve  healthy  eating  behaviors  through  individual  dietary  counseling  and  public  health  initiatives  should address 
not only the consumption of healthy foods but also snacking behaviors and the affective (ie, emotional) component of eating. 

ACKNOWLEDGMENTS 
The Mid-South CDRN was initiated and funded by the Patient-Centered Outcomes Research Institute through 
6 Heerman et al Journal of Nutrition Education and Behavior Volume -, Number -, 2017 
Figure  3.  Distribution  of  body  mass  index  (BMI)  as  measured  by  kernel  density  according  to  cluster  of  eating  behavior.  The 
default  STATA  (version  14.2;  StatCorp,  College  Station,  TX)  settings for kernel density were used, including the Epanechnikov 
kernel  and  the  calcu-  lated  optimal  bandwidth  of  the  kernel.  The  inset  figure  shows  the adjustedband 95% con- fidence interval 
(CI)  from  the  linear  regression evaluating the association between eating cluster and BMI, controlling for age, gender, household 
income, race/ethnicity, and phys- ical activity. The referent group is the healthy eating cluster. 
contract  CDRN-1306-04869,  the  Van-  derbilt  Institute  for  Clinical  and  Trans-  lational  Research  with  support  grant  from 
ULTR000445,  from  the  National  Center  for  Advancing  Translational  Sci-  ences/National  Institutes  of  Health;  and  institutional 
funding.  Dr  Heerman's  time  was  supported  by  a  K12  grant  fr-  om  the  Agency  for  Healthcare  Research  and  Quality 
(K12HS022990) and a K23 grant from the National Heart, Lung, and Blood Institute (K23 HL127104). 

SUPPLEMENTARY DATA 
Supplementary data related to this article can be found online at http:// dx.doi.org/10.1016/j.jneb.2017.02.001. 

REFERENCES 
1. Huang TT, Drewnosksi A, Kumanyika S, Glass TA. A systems-oriented multilevel 
framework for addressing obesity in the 21st century. Prev Chronic Dis. 2009;6:A82. 2.Waters E, de Silva-Sanigorski A, Hall BJ, 
et al. Interventions for prevent- ing obesity in children. Cochrane Data- base Syst Rev. 2011:CD001871. 3. Ogden CL, Carroll 
MD, Fryar CD, Flegal KM. Prevalence of obesity among adults and young: United States, 2011- 2014. Hyattsville, MD: National 
Center for Health Statistics; 2015. NCHS Data Brief No. 219. 4. Bhupathiraju SN, Hu FB. Epidemi- ology of obesity and 
diabetes and their cardiovascular complications. Circ Res. 2016;118:1723-1735. 5.Calle EE, Rodriguez C, Walker- Thurmond K, 
Thun MJ. Overweight, obesity, and mortality from cancer in a prospectively studied cohort of U.S. adults. N Engl J Med. 
2003;348: 1625-1638. 6. Mozaffarian D, Hao T, Rimm EB, Willett WC, Hu FB. Changes in diet and lifestyle and long-term 
weight gain in women and men. N Engl J Med. 2011;364:2392-2404. 7. Pereira MA, Kartashov AI, Ebbeling CB, et al. Fast-food 
habits, weight gain, and insulin resistance (the CARDIA study): 15-year prospective analysis. Lancet. 2005;365:36-42. 8. 
Rosenheck R. Fast food consumption and increased caloric intake: a system- atic review of a trajectory towards weight gain and 
obesity risk. Obesity Rev. 2008;9:535-547. 9. Barnes TL, French SA, Mitchell NR, Wolfson J. Fast-food consumption, diet 
quality and body weight: cross- sectional and prospective associations in a community sample of working adults. Pub Health 
Nutr. 2016;19:885-892. 10. Berkey CS, Rockett HR, Gillman MW, Field AE, Colditz GA. Longitudinal study of skipping break- 
fast and weight change in adolescents. Int J Obesity Relat Metab Disord. 2003; 27:1258-1266. 11. van der Heijden AA, Hu FB, 
Rimm EB, van Dam RM. A prospective study of breakfast consumption and weight gain among U.S. men.Obesity (Silver 
Spring). 2007;15:2463-2469. 12. Mesas AE, Munoz-Pareja M, Lopez- Garcia E, Rodriguez-Artalejo F. Selected 
eatingbehavioursandexcessbodyweight: a systematic review.Obesity Rev. 2012;13: 106-135. 13. Harris PA, Taylor R, Thielke R, 
Payne J, Gonzalez N, Conde JG. Research elec- tronic data capture (REDCap)—a 
 
metadata-driven methodology and workflow process for providing transla- tional research informatics support. J Bio- med 
Inform. 2009;42:377-381. 14. Rosenbloom ST, Harris P, Pulley J, et al. The Mid-South clinical Data Research Network. J Am 
Med Inform Assoc. 2014;21:627-632. 15. Meyers AG, Salanitro A, Wallston KA, et al. Determinants of health after hos- pital 
discharge: rationale and design of the Vanderbilt Inpatient Cohort Study (VICS). BMC Health Services Res. 2014;14:10. 16. 
Signorello LB, Munro HM, Buchowski MS, et al. Estimating nutrient intake from a food frequency question- naire: incorporating 
the elements of race 
Journal of Nutrition Education and Behavior Volume -, Number -, 2017 Heerman et al 7 
and geographic region. Am J Epidemiol. 2009;170:104-111. 17. Stetson B, Schlundt D, Rothschild C, Floyd JE, Rogers W, 
Mokshagundam SP. Development and validation of The Personal Diabetes Questionnaire (PDQ): a measure of diabetes self-care 
behaviors, perceptions and barriers. Diabetes Res Clin Practice. 2011;91:321-332. 18. Makles A. Stata tip 110: how to get the 
optimal k-means cluster solution. Stata J. 2012;12:347-351. 19. Tan P-N, Steinbach M, Kumar V. Introduction to Data Mining. 
1st ed. Bos- ton, MA: Pearson Addison Wesley; 2006. 20. Ludwig DS, Friedman MI. Increasing adiposity: consequence or cause 
of 
overeating? JAMA. 2014;311: 2167-2168. 21. Raynor HA, Champagne CM. Posi- tion of the Academy of Nutrition and 
Dietetics: interventions for the treat- ment of overweight and obesity in adults. J Acad Nutr Diet. 2016;116: 129-147. 22. 
American Diabetes Association. Obe- sity management for the treatment of type 2 diabetes. Diabetes Care. 2017;40: S57-S63. 
23. Council of American Survey Research Organizations. On the Definition of Response Rates, 1982. http://c.ymcdn. 
com/sites/www.casro.org/resource/resmgr/ docs/casro_on_definitions_of_resp.pdf. Accessed March 6, 2017. 
 
7.e1 Heerman et al Journal of Nutrition Education and Behavior Volume -, Number -, 2017 
CONFLICT OF INTEREST 
The authors have not stated any con- flicts of interest.