Complex Comorbidity Clusters in OEF/OIF VeteransThe Polytrauma Clinical Triad and Beyond

Mary Jo V. Pugh, PhD, RN,*wz Erin P. Finley, PhD, MPH,*y Laurel A. Copeland, PhD,z8Chen-Pin Wang, PhD,*w Polly H. Noel, PhD,*y Megan E. Amuan, MPH,z

Helen M. Parsons, PhD, MPH,w Margaret Wells, BS,*w Barbara Elizondo, BA,*wand Jacqueline A. Pugh, MD*#

Background: A growing body of research on US Veterans from

Afghanistan and Iraq [Operations Enduring and Iraqi Freedom, and

Operation New Dawn (OEF/OIF)] has described the polytrauma

clinical triad (PCT): traumatic brain injury (TBI), posttraumatic

stress disorder (PTSD), and pain. Extant research has not explored

comorbidity clusters in this population more broadly, particularly

co-occurring chronic diseases.

Objectives: The aim of the study was to identify comorbidity

clusters among diagnoses of deployment-specific (TBI, PTSD, pain)

and chronic (eg, hypertension, diabetes) conditions, and to examine

the association of these clusters with health care utilization and

adverse outcomes.

Research Design: This was a retrospective cohort study.

Subjects: The cohort comprised OEF/OIF Veterans who received

care in the Veterans Health Administration in fiscal years (FY)

2008–2010.

Measures: We identified comorbidity using validated ICD-9-CM

code–based algorithms and FY08–09 data, followed by which we

applied latent class analysis to identify the most statistically distinct

and clinically meaningful patterns of comorbidity. We examined the

association of these clusters with process measures/outcomes using

logistic regression to correlate medication use, acute health care

utilization, and adverse outcomes in FY10.

Results: In this cohort (N = 191,797), we found 6 comorbidity

clusters. Cluster 1: PCT + Chronic Disease (5%); Cluster 2: PCT

(9%); Cluster 3: Mental Health + Substance Abuse (24%); Cluster 4:

Sleep, Amputation, Chronic Disease (4%); Cluster 5: Pain, Mod-

erate PTSD (6%); and Cluster 6: Relatively Healthy (53%). Sub-

sequent health care utilization patterns and adverse events were

consistent with disease patterns.

Conclusions: These comorbidity clusters extend beyond the PCT

and may be used as a foundation to examine coordination/quality of

care and outcomes for OEF/OIF Veterans with different patterns of

comorbidity.

Key Words: Afghan Campaign 2001 (Operation Enduring Free-

dom), comorbidity, Iraq War 2003 (Operation Iraqi Freedom),

veterans

(Med Care 2014;52: 172–181)

The years since 2001 have presented extraordinary chal-lenges to the US Armed Forces and the Veterans Health

Administration (VA) in their efforts to provide high-qualitycare to Veterans returning from deployment to Afghanistanand Iraq (Operations Enduring and Iraqi Freedom, and Op-eration New Dawn, hereafter OEF/OIF). Studies in thispopulation have identified high rates of combat-related in-juries and mental health conditions that commonly co-occurand that have implications for health care utilization, treat-ment, and long-term outcomes.1–4 The majority of priorstudies have focused on the signature injuries of traumaticbrain injury (TBI), posttraumatic stress disorder (PTSD), andpain, also known as the polytrauma clinical triad (PCT),individually or in combination (eg, TBI and PTSD, PTSDand pain, etc.).5–10

Although trauma-based injuries and mental healthproblems dominate research in this population,10–13 Veterans

From the *South Texas Veterans Health Care System; wDepartment ofEpidemiology and Biostatistics, University of Texas Health ScienceCenter San Antonio, San Antonio; zTexas A&M Health Science Center,Bryan; yDepartment of Medicine, Division of Clinical Epidemiology,University of Texas Health Science Center San Antonio, San Antonio;8Center for Applied Health Research, jointly sponsored by CentralTexas Veterans Health Care System, and Scott and White HealthcareSystem, Temple, TX; zCenter for Health Quality, Outcomes andEconomic Research, Edith Nourse Rogers Memorial VA Hospital,Bedford, MA; and #Department of Medicine, Division of HospitalMedicine, University of Texas Health Science Center San Antonio,San Antonio, TX.

The content of this article is solely the responsibility of the authors and doesnot necessarily reflect the official views of the Veterans’ Health Ad-ministration.

Funded by the Department of Veterans Affairs, Office of Research andDevelopment, VA Health Services Research and Development Service(DHI 09-237).

The authors declare no conflict of interest.Reprints: Mary Jo V. Pugh, PhD, RN, South Texas Veterans Health Care

System, 7400 Merton Minter Blvd, San Antonio, TX 78229. E-mail:Maryjo.pughm@uthscsa.edu.

Supplemental Digital Content is available for this article. Direct URL cita-tions appear in the printed text and are provided in the HTML and PDFversions of this article on the journal’s Website, www.lww-medicalcare.com.

Copyright r 2013 by Lippincott Williams & WilkinsISSN: 0025-7079/14/5202-0172

ORIGINAL ARTICLE

172 | www.lww-medicalcare.com Medical Care � Volume 52, Number 2, February 2014

of OEF/OIF also experience a significant chronic diseaseburden14,15 and non–combat-related injuries (eg, motor ve-hicle accidents), musculoskeletal issues, and medical con-ditions (eg, digestive, genitourinary, and circulatory).16 As aresult, studies have begun to describe comorbidity of mentalhealth and physical health conditions in this population.Cohen et al17 found that OEF/OIF Veterans with PTSD aremore likely to have cardiovascular disease–related con-ditions, and Frayne et al found that those with PTSD hadmore medical comorbidities diagnosed than those withoutPTSD.18 Nonetheless, studies of patterns of mental andphysical comorbidity and associated outcomes beyond thePCT are not available. Thus, despite growing emphasisacross the medical literature on providing integrated treat-ment approaches for patients with multiple chronic con-ditions,19 we lack information regarding whether patterns ofmultiple chronic conditions in this relatively young pop-ulation exist that can inform recommendations for improvingtreatment and services and support more comprehensiveevaluation of processes and outcomes of care.18

Our goal was to identify patterns of co-occurringphysical and mental health conditions among OEF/OIFVeterans using latent class analysis (LCA), and to examinethe association of those patterns with health care utilizationand adverse events. We hypothesized that (1) previouslyunidentified clusters of comorbidity among OEF/OIFVeterans exist and include a variety of combinations of TBI,PTSD, pain, and chronic disease, and (2) variation in co-morbidity clusters are associated with differences in pro-cesses of care and adverse outcomes. We specificallyexpected the following: 2a/Mental Health—comorbidityclusters with higher likelihood of depression and/or othermental health disorders have higher utilization of psycho-tropic medications, emergency care, inpatient mental health/substance abuse treatment, and adverse outcomes such assuicide-related behavior (SRB); 2b/Pain—comorbidity clus-ters with higher likelihood of pain have higher utilization ofmedications commonly used for pain (eg, anticonvulsants,opioid pain relievers), and inpatient neurosurgery/orthopediccare; and 2c/Chronic Disease—comorbidity clusters withhigher likelihood of chronic disease states have higher uti-lization of related medications (eg, antihypertensive agents),inpatient medical hospitalizations, and mortality rate.

METHODS

Data SourcesAfter securing IRB approval, we obtained data for

Fiscal Years (FY) 2008 to 2010 (FY08–10; October 1, 2007to September 30, 2010) from the VA Decision SupportSystem National Data Extracts (DSS inpatient, outpatient,and pharmacy data), the OEF/OIF roster, and the VA VitalStatus file. Inpatient and outpatient files included ICD-9-CMcodes, clinic or hospital bedsection in which care was re-ceived, and demographic characteristics. Outpatient phar-macy data included the generic drug name. We identifiedlatent classes in diagnostic data from FY08–09; health careutilization and adverse outcomes were identified in FY10,after comorbidity latent classes (clusters) were identified.

ParticipantsParticipants flagged as OEF/OIF Veterans were se-

lected for inclusion if they received VA inpatient oroutpatient care at least 1 time yearly in each of FY08 andFY09.

Measures

Comorbid ConditionsInpatient and outpatient diagnoses from face-to-face

clinic/hospital care during FY08–09 were used to identifycomorbid conditions for subsequent LCA; diagnoses fromancillary care such as laboratory, radiology, etc. were ex-cluded. We used ICD-9-CM codes previously validated foruse in administrative data (including Charlson, Elix-hauser)20,21 to create dichotomous indicators for 32 physical/mental health and postdeployment conditions including TBI,PTSD, pain, and other conditions commonly identified inthis population.22–24 ICD-9-CM codes for each condition arepresented in Appendix 1, Supplemental Digital Content 1,http://links.lww.com/MLR/A637.

Demographic CharacteristicsDemographics included age, sex, race/ethnicity, and

marital status, obtained from inpatient and outpatient data.Individuals with missing race were included as a separatecategory. We also identified individuals who deployed withthe National Guard/Reserve (per OEF/OIF roster).

Health Care UtilizationWe identified process measures describing pharmaco-

therapy, emergency care, and inpatient care during FY10.We created dichotomous variables for 10 theoretically rele-vant drug groups related to pain, mental health, and chronicdisease management (Table 2). Individuals with at least 1prescription for each drug group during FY10 were classifiedas receiving a prescription in that drug group (eg, opioids).We created indicators of health care utilization using out-patient identifiers for emergency/urgent care (130, 131) andbedsection (treating clinician’s specialty; see Table 3).

Adverse OutcomesAdverse outcomes included proxies for homelessness,

diagnoses of drug overdose, suicidal ideation or attempt(hereafter SRB), and death. ICD-9-CM and clinic stop codesfor SRB and drug overdose are provided in Appendix 1,(Supplemental Digital Content 1, http://links.lww.com/MLR/A637). Mortality through FY10 was defined using the VAVital Status File.

AnalysisWe used LCA, a method of structural equation mod-

eling that identifies unobservable subgroups (latent classes)within a population on the basis of the distribution of binarydiagnosis outcomes to identify patterns of comorbidity in thispopulation. Thus, each latent class represents a “cluster” ofcomorbidities that occur most commonly in a subgroup ofthe population. Like other latent variable modeling tech-niques, LCA is an iterative process that assesses the fit ofthe model to the data. Our analysis started with a 2-class

Medical Care � Volume 52, Number 2, February 2014 Beyond the Polytrauma Clinical Triad

r 2013 Lippincott Williams & Wilkins www.lww-medicalcare.com | 173

solution, modeling the joint probability of the conditionsunder the assumption that they are correlated through classmembership. We then conducted LCA with 3–7 classes, al-lowing 20 different start values for each LCA to ensure thatthe global maximum was reached and results were consistentregardless of the starting point. We identified the model with

the best fit on the basis of reliability of the estimates, Akaikeinformation criterion, and Bayesian information criterion.

Once the best fitting model was identified, we used thepseudoclass technique to estimate each individual’s class(hereafter cluster) membership25 by drawing a randomsample from the multinomial distribution based on posterior

0102030405060708090

100

Pro

bab

ility

of

Hav

ing

Dia

gn

osi

s

Comorbid Conditions

Cluster 1 Polytrauma Clinical Triad, Depression,Chronic Disease

0102030405060708090

100

Pro

bab

ility

of

Hav

ing

Dia

gn

osi

s

Comorbid Conditions

Cluster 2 Polytrauma Clinical Triad

0102030405060708090

100

Pro

bab

ility

of

Hav

ing

Dia

gn

osi

s

Comorbid Conditions

Cluster 3 Mental Health, Substance Abuse

0102030405060708090

100

Pro

bab

ility

of

Hav

ing

Dia

gn

osi

s

Comorbid Conditions

Cluster 4 Sleep, Amputation, Chronic Disease

0102030405060708090

100

Pro

bab

ility

of

Hav

ing

Dia

gn

osi

s

Comorbid Conditions

Cluster 5 Pain, Moderate PTSD

0102030405060708090

100

Pro

bab

ility

of

Hav

ing

Dia

gn

osi

s

Comorbid Conditions

Cluster 6 Relatively Healthy

FIGURE 1. Distribution of comorbidity types by cluster.

Pugh et al Medical Care � Volume 52, Number 2, February 2014

174 | www.lww-medicalcare.com r 2013 Lippincott Williams & Wilkins

class probabilities of membership. We then examined sub-sequent health care utilization and adverse outcomes for eachcluster.

We used descriptive statistics to examine differences indemographic characteristics using the w2 test and analysis ofvariance. Finally, we used logistic regression models to ex-amine the relationship between cluster and (1) medicationsprescribed, (2) acute health care utilization, and (3) adverseoutcomes adjusting for multiple comparisons using theBonferroni adjustment. Confidence intervals were used tocompare nonreference group clusters with each other (non-overlapping intervals indicate significant difference).

RESULTSOur analysis of care-seeking Veterans (N = 191,797)

found that the LCA with 6 classes obtained global maximumlikelihood estimates and had the smallest Akaike informationcriterion and Bayesian information criterion compared withother LCAs whose results converged. Figure 1 summarizesthe probability of an individual in each cluster having spe-cific conditions and the general distribution of comorbiditytypes across the 6-class model, with descriptions of the 21conditions that were most notable in each cluster.

Clusters 1 and 2 exhibited characteristics of the PCT(TBI, PTSD, Pain). Cluster 1 also had significantly higherprobabilities of low back pain, chronic pain, insomnia, de-pression, hypertension, and osteoarthritis, and lower proba-bilities of TBI than Cluster 2 (Cluster 1: PCT, Depression,Chronic Disease; Cluster 2: PCT). Cluster 3 was charac-terized by depression, PTSD, anxiety, and substance abuse(Mental Health, Substance Abuse). Cluster 4 was charac-terized by insomnia, amputation, and chronic disease (Sleep,Amputation, Chronic Disease). Cluster 5 was characterizedby back pain and other pain (Pain, Moderate PTSD). Cluster6 was characterized by low probabilities of all conditions(Relatively Healthy).

Although all bivariate analyses of demographic char-acteristics demonstrated statistically significant differences(Table 1), the most clinically relevant differences were re-lated to age. In Clusters 1 and 4 the mean age was over 40,whereas Clusters 2 and 3 had mean ages in the early 30s[F = 1731 (df = 5, 191,723), P < 0.01]. Clusters 1 and 4 hadsignificantly higher proportions of African Americans andlower proportions of whites than expected [w2 = 2367 (df =20), P < 0.001]. Individuals in Clusters 1, 4, and 5 weresignificantly more likely to be married and to be deployed asNational Guard/Reservists than expected [w2 = 2686 (df = 5),P < 0.001; w2 = 1013 (df = 5), P<0.001].

Health Care Utilization and Adverse OutcomesTable 2 displays the proportion of individuals from

each cluster who received specific medication types, and theodds ratios from individual logistic regression models com-pared with Cluster 6. The final column of Table 2 shows theclusters with similar results, identified by examining theoverlap of confidence intervals from the logistic regressionanalysis.When confidence intervals overlap, those clustersare not significantly different from each other, allowingcomparisons between clusters beyond the reference group. T

AB

LE

1.

Dem

og

rap

hic

Ch

ara

cteri

stic

sb

yC

om

orb

idit

yC

lust

er

Dem

ographic

Variables

Cluster

1:PolytraumaClinical

Triad+

Depression,Chronic

Disease

N=

8801(4.6%

)

Cluster

2:Polytrauma

ClinicalTriad

N=

16,585(8.6%

)

Cluster

3:MentalHealthand

Substance

Abuse

N=

45,242(23.6%

)

Cluster

4:Sleep,

Amputation,ChronicDisease

N=

8158(4.3%

)

Cluster

5:

Pain,PTSD

N=

10,769

(5.6%

)

Cluster

6:

Relatively

Healthy

N=

102,242

(53.3%

)

Ag

e[M

ean

(SD

)]4

0.2

(10

.2)

33

.6(8

.4)

33

.9(8

.6)

41

.4(9

.8)

38

.9(1

0.0

)3

6.7

(10

.2)

Mal

e(%

)8

8.3

91

.48

3.9

92

.98

6.8

86

.5R

ace/

ethnic

ity

Afr

ican

Am

eric

an(%

)

19

.21

1.4

14

.62

3.0

16

.61

5.3

His

pan

ic(%

)1

2.5

10

.29

.41

0.6

11

.09

.6W

hit

e(%

)5

5.1

59

.35

7.2

47

.65

4.5

50

.3O

ther

(%)

3.1

3.2

3.1

3.5

3.2

3.3

Un

kn

ow

n(%

)1

0.1

15

.91

5.8

15

.21

4.6

21

.5M

arri

ed(y

es)

(%)

58

.34

8.8

40

.16

5.1

56

.54

8.5

Nat

ion

alG

uar

d/

Res

erve

(%)

59

.04

4.6

41

.85

1.7

51

.44

6.5

All

com

par

isons

had

stat

isti

call

ysi

gnifi

cant

dif

fere

nce

sat

the

P<

0.0

01

level

.P

TS

Din

dic

ates

post

trau

mat

icst

ress

dis

ord

er.

Medical Care � Volume 52, Number 2, February 2014 Beyond the Polytrauma Clinical Triad

r 2013 Lippincott Williams & Wilkins www.lww-medicalcare.com | 175

TA

BLE

2.

Typ

es

of

Pre

scri

pti

on

Med

icati

on

sRece

ived

inFY

10:

Com

pari

son

sW

ith

the

Rela

tive

lyH

ealt

hy

Clu

ster

(Clu

ster

6)

%

OR

(95%

CI)

Cluster

1:PolytraumaClinical

Triad,Depression,Chronic

Disease

N=

8801(4.6%

)

Cluster

2:

PolytraumaClinical

Triad

N=

16,585(8.6%

)

Cluster

3:Mental

Health,Substance

Abuse

N=

45,242(23.6%

)

Cluster

4:Sleep,

Amputation,Chronic

Disease

N=

8158(4.3%

)

Cluster

5:Pain,

Moderate

PTSD

N=

10,769

(5.6%

)

Cluster

6:

Relatively

Healthy

N=

102,242

(53.3%

)

GroupsWith

Sim

ilarResults

Pai

nre

late

dA

nti

convuls

ants

41.5

16.6

(15.7

–17.5

)20.2

5.9

(5.6

–6.2

)14.2

3.9

(3.7

–4.0

)14.4

3.9

(3.7

–4.2

)21.1

6.3

(5.9

–6.6

)4

.1R

efer

ence

1>

2,5

>3,4 6

Opio

idpai

nre

liev

er48.5

7.1

(6.8

–7.5

)25.1

2.5

(2.4

–2.6

)19.3

1.8

(1.8

–1.9

)25.3

2.6

(2.4

–2.7

)35.9

4.2

(4.0

–4.4

)11.7

Ref

eren

ce1

>5

>2,4

>3

> 6N

onst

eroid

alan

ti-

infl

amm

atory

dru

g

56.0

4.8

(4.6

–5.0

)36.7

2.2

(2.1

–2.3

)28.8

1.5

(1.5

–1.6

)39.6

2.5

(2.4

–2.6

)45.3

3.1

(3.0

–3.3

)21.0

Ref

eren

ce1

>5

>4

>2

>3

> 6S

kel

etal

musc

lere

laxan

ts37.8

6.6

(6.3

–6.9

)18.7

2.5

(2.4

–2.6

)12.9

1.6

(1.6

–1.7

)18.8

2.5

(2.4

–2.7

)30.6

4.8

(4.6

–5.0

)8

.4R

efer

ence

1>

5>

2,4

>3

> 6P

sych

iatr

icA

nti

psy

choti

c:at

ypic

al26.7

16.9

(15.9

–18.0

)12.8

6.8

(6.4

–7.3

)14.8

8.1

(7.7

–8.5

)9

.04.6

(4.2

–5.0

)8

.54.3

(4.0

–4.7

)2

.1R

efer

ence

1>

3>

2>

4,5

>6

Anti

psy

choti

c:ty

pic

al7.5

7.1

(6.4

–7.8

)4.0

3.7

(3.3

–4.1

)3.4

3.1

(2.9

–3.4

)2

.62.3

(2.0

–2.7

)3

.02.7

(2.4

–3.1

)1

.1R

efer

ence

1>

2,3

,5>

4> 6

Anti

dep

ress

ant

75.2

15.9

(15.1

–16.7

)52.5

5.8

(5.6

–6.0

)51.5

5.6

(5.4

–5.7

)46.6

4.6

(4.4

–4.8

)40.6

3.6

(3.4

–3.7

)16.1

Ref

eren

ce1

>2,3

>4

>5

> 6C

hro

nic

dis

ease

Anti

hyper

tensi

ve

agen

ts40.2

4.4

(4.2

–4.6

)9.4

0.7

(0.6

–0.7

)12.9

1.0

(0.9

–1.0

)37.8

4.0

(3.8

–4.2

)19.6

1.6

(1.5

–1.7

)13.3

Ref

eren

ce1

,4>

5>

3,

6>

2O

ral

hypogly

cem

icag

ent

or

insu

lin

8.7

3.5

(3.2

–3.8

)0.8

0.3

(0.2

–0.4

)2.0

0.7

(0.7

–0.8

)10.3

4.2

(3.8

–4.5

)3

.01.1

(1.0

–1.3

)2

.7R

efer

ence

1,

4>

5,

6>

3> 2

Lip

idm

edic

atio

ns

30.6

3.5

(3.4

–3.7

)10.7

1.0

(0.9

–1.0

)10.5

0.9

(0.9

–1.0

)29.4

3.3

(3.2

–3.5

)18.8

1.9

(1.8

–2.0

)11.1

Ref

eren

ce1,

4>

5>

2,

3,

6

CI

indic

ates

confi

den

cein

terv

al;

FY

10,

fisc

alyea

r2010;

OR

,odds

rati

o;

PT

SD

,post

trau

mat

icst

ress

dis

ord

er.

Pugh et al Medical Care � Volume 52, Number 2, February 2014

176 | www.lww-medicalcare.com r 2013 Lippincott Williams & Wilkins

Medications commonly used for pain were far more likely tobe received by individuals in Cluster 1 than by patients inother clusters, with the lowest utilization of pain medicationsbeing in Cluster 6. Utilization of pain medications was alsohigher for Cluster 5 compared with most other groups.Similar patterns were found for psychotropic medications,with those in Cluster 1 having the highest odds of receivingeach type of psychotropic medication, whereas individuals inCluster 6 had the lowest odds. Those in Clusters 2 and 3 alsowere consistently more likely to receive atypical anti-psychotics and antidepressants than Clusters 4, 5, and 6.With regard to medications for chronic disease, individualsin Clusters 1 and 4 had the highest likelihood of receivingantihypertensives, hypoglycemics, and lipid medications,whereas individuals in Cluster 2 consistently had the lowestlikelihood of receiving these medications. Individuals inCluster 6 were more likely than Clusters 2 and 3 to receivehypoglycemic medications.

Table 3 shows the prevalence of acute care utilizationin FY10 and odds ratios comparing Cluster 6 with otherclusters. The final column of Table 3 lists groups with similarresults, identified by examining the overlap of confidenceintervals from the logistic regression analysis. Individuals inCluster 1 had the highest levels of acute care utilizationacross the board, whereas those in Cluster 6 had the lowestlevels. For inpatient psychiatric/substance abuse care, thosein Clusters 3 and 2 were significantly higher than all re-maining clusters. Finally, individuals in Cluster 5 wereequally likely to have orthopedic/neurosurgery inpatientcare as those in Cluster 1, and individuals in Clusters 2 and 4were less likely to have orthopedic/neurosurgery inpatientcare than Clusters 1 and 5, but more likely than those inClusters 3 and 6.

Adverse outcomes were consistently highest in Clus-ters 1 and 3 and lowest for individuals in Cluster 6, with theexception of overdose, in which Cluster 4 was not differentfrom Cluster 6. Of interest, individuals in Cluster 4, who hada higher likelihood of chronic disease, were at lower like-lihood of mortality than those in Cluster 3, which includedindividuals with primarily mental health/substance use dis-order (Table 4).

DISCUSSIONThis study is the first to describe broad comorbidity

clusters in Veterans from the conflicts in Afghanistan andIraq, accounting for conditions related to TBI, mental health,chronic pain, and chronic disease. As predicted (hypothesis1), we found 6 clusters, 2 of which were characterized by all3 signature PCT injuries—TBI, PTSD, and pain. However,individuals in Cluster 1 had lower probability of TBI andhigher probability of headache, back pain, other pain thanthose in Cluster 2. Individuals in Cluster 1 had higherprobabilities of conditions associated with the PCT such assleep disturbance, depression, and anxiety. Finally, in-dividuals in Cluster 1 were older and more likely to havechronic disease diagnoses including hypertension, diabetes,osteoarthritis, and obesity.

Although the PCT was not evident in Clusters 3, 4, and5, other conditions, including individual PCT conditions,emerged. Cluster 3 was characterized by multiple mentalhealth conditions including PTSD, depression, anxiety, andsubstance use disorder, whereas Cluster 5 was characterizedprimarily by back pain and other pain, with a likelihood ofPTSD similar to our overall cohort (32%). Although thelikelihood for PTSD was similar between Clusters 4 and 5,all individuals in Cluster 4 had diagnoses of sleep disorder,and higher probabilities of hypertension, obesity, osteo-arthritis, and amputation diagnoses. These findings are con-sistent with PCT literature, and recent work that foundVeterans with PTSD were more likely to have chronic dis-ease than those without PTSD.9,19 However, they suggestthat there are distinct clusters of individuals with PTSD,rather than 1 prototypical PTSD patient. Additional researchusing these clusters may help us identify early on those in-dividuals with PTSD who may progress more quickly tochronic disease versus those with less chronic disease bur-den. Such research may also allow us to develop inter-ventions that ultimately reduce disease burden in the PTSDchronic disease clusters.

Finally, despite the plethora of literature identifyingprofound disease burden in this population,5,23,26–29 wefound a cluster of relatively healthy individuals who ac-counted for about half of the cohort (53%). Although theseindividuals had a likelihood of hypertension slightly higherthan the population prevalence, the likelihood of otherconditions was significantly lower than the respective pop-ulation prevalences. Across clusters, examination of socio-demographic characteristics suggests that clusters witha higher mean age (Clusters 1, 4, 5, and 6) were more likelyto have associated chronic disease. Age was not associatedwith mental health conditions, however, as Clusters 1 and3 had the highest likelihood of a variety of psychiatricdiagnoses.

Analyses of health care utilization and adverse out-comes also generally supported our second hypothesis. Withregard to psychiatric outcomes (Hypothesis 2a/MentalHealth), those clusters with high probability of multiplepsychiatric diagnoses (Clusters 1 and 3) were consistentlyamong those with the highest likelihood of psychotropicmedications, emergency care, inpatient psychiatric/substanceabuse care, and adverse outcomes of homelessness, over-dose, SRB, and mortality. Cluster 1 in particular had a higherlikelihood of psychotropic medications and inpatient psy-chiatric care than Cluster 3. Other findings suggest that theremay be a “dose response” effect for psychiatric comorbidity.For example, Clusters 2, 4, and 5, each characterized by alower likelihood of PTSD and other psychiatric conditionsthan Clusters 1 and 3, had psychiatrically related outcomeshigher than Cluster 6 but lower than Clusters 1 and 3. Thisfinding echoes similar findings in studies of suicidal idea-tion30 and suggests that the impact of PTSD-depression co-morbidity in this population extends beyond suicidalideation.

With regard to Hypothesis 2b/Pain, we found thatclusters characterized by pain (Clusters 1, 2, and 5) had thehighest likelihood of anticonvulsants and opioids; Clusters 1

Medical Care � Volume 52, Number 2, February 2014 Beyond the Polytrauma Clinical Triad

r 2013 Lippincott Williams & Wilkins www.lww-medicalcare.com | 177

TA

BLE

3.

Acu

teC

are

Rece

ived

inFY

10:

Com

pari

son

sW

ith

the

Rela

tive

lyH

ealt

hy

Clu

ster

(Clu

ster

6)

%

OR

(95%

CI)

Cluster

1:PolytraumaClinical

Triad,Depression,Chronic

Disease

N=

8801(4.6%

)

Cluster

2:

Polytrauma

ClinicalTriad

N=

16,585(8.6%

)

Cluster

3:Mental

Health,Substance

Abuse

N=

45,242(23.6%

)

Cluster

4:Sleep,

Amputation,Chronic

Disease

N=

8158(4.3%

)

Cluster

5:Pain,

Moderate

PTSD

N=

10,769

(5.6%

)

Cluster

6:

Relatively

Healthy

N=

102,242

(53.3%

)

GroupsWith

Sim

ilarResults

Em

erg

ency

/urg

ent

care

39

.43

.9(3

.7–

4.1

)2

5.5

2.0

(2.0

–2.1

)2

5.2

2.0

(2.0

–2.1

)2

2.8

1.8

(1.7

–1.9

)2

5.1

2.0

(1.9

–2.1

)1

4.4

Ref

eren

ce1

>2

,3,5

>4

> 6In

pat

ient

Neu

rolo

gy

1.3

12

.6(9

.6–

16

.4)

0.6

6.0

(4.6

–7.9

)0

.66

.3(5

.0–

7.9

)0

.33

.1(2

.0–

4.7

)0

.44

.1(2

.9–

5.8

)0

.1R

efer

ence

1>

2,3

>4

,5>

6In

pat

ien

tm

edic

ine

3.8

6.2

(5.4

–7.1

)1

.21

.9(1

.6–

2.2

)1

.52

.3(2

.1–

2.6

)1

.62

.6(2

.1–

3.1

)1

.32

.0(1

.7–

2.4

)0

.6R

efer

ence

1>

2,3

,4,5

>6

Inp

atie

nt

psy

chia

tric

/su

bst

ance

abu

se

6.5

13

.4(1

1.9

–1

5.1

)3

.16

.2(5

.5–

7.0

)4

.59

.2(8

.4–

10

.2)

1.7

3.3

(2.8

–4.0

)1

.93

.7(3

.1–

4.3

)0

.5R

efer

ence

1>

3>

2>

4,5

>6

Inp

atie

nt

ort

ho

ped

ic/

neu

rosu

rger

y

1.0

7.3

(5.6

–9.6

)0

.21

.7(1

.2–

2.5

)0

.21

.3(1

.0–

1.7

)0

.42

.7(1

.8–

4.0

)0

.96

.8(5

.2–

8.8

)0

.1R

efer

ence

1,5

>2

,4>

3,6

Oth

erin

pat

ien

tsu

rger

y1

.23

.3(2

.7–

4.1

)0

.61

.6(1

.3–

2.0

)0

.51

.4(1

.2–

1.7

)0

.92

.4(1

.8–

3.1

)0

.82

.1(1

.7–

2.7

)0

.4R

efer

ence

1>

2,4

,5>

3> 6

CI

indic

ates

confi

den

cein

terv

al;

OR

,odds

rati

o;

PT

SD

,post

trau

mat

icst

ress

dis

ord

er.

Pugh et al Medical Care � Volume 52, Number 2, February 2014

178 | www.lww-medicalcare.com r 2013 Lippincott Williams & Wilkins

and 5 were also more likely to receive nonsteroidal anti-inflammatory drugs and skeletal muscle relaxants than oth-ers, especially Cluster 1. Of interest, opioid use was 10%even in Cluster 6, suggesting that opioid exposure extendsbeyond risk factors identified by Seal et al31 and that addi-tional research may be needed to examine opioid exposuresand outcomes in this “healthy” sector of the OEF/OIF pop-ulation. Clusters 1 and 5, which had the highest likelihood ofback pain, predictably also had the highest likelihood ofinpatient care for orthopedics/neurosurgery. With regard toadverse outcomes, we saw no consistent pattern with pain,unless pain was combined with significant psychiatric co-morbidity. Cluster 5 was characterized primarily by highprobability of back pain and other pain, but was lower inrankings of homeless care, overdose, and SRB than Cluster1, in which pain was accompanied by high rates of PTSD anddepression.

Finally, Hypothesis 2c/Chronic Disease was alsolargely supported. Clusters 1 and 4, in which chronic diseasewas a distinguishing feature, were most likely to receiveantihypertensive, hypoglycemic, and lipid medications.Cluster 1 had the highest likelihood of medical hospital-izations, but findings were not as consistent for Cluster 4.Unexpectedly, Cluster 6 was more likely to receive anti-hypertensive and hypoglycemic medications than Cluster 2;however, Cluster 6 had the lowest odds of medical hospi-talization and mortality. The finding for antihypertensivemedications is consistent with the probability of hyper-tension in Cluster 6, suggesting that, although these in-dividuals do have some chronic disease, they are treated anddo not yet demonstrate adverse outcomes as defined bymortality.

In addition to these hypotheses, several findings werenotable. First, despite the emphasis in the literature on thecomorbidity of TBI, PTSD, and pain, only about 13% of thecohort was defined by this pattern. We found that those inCluster 1, who also had a higher probability of depressionand chronic disease, had significantly more health care uti-lization and adverse outcomes than any of the other clustersincluding Cluster 2 with only PCT. Moreover, whereasPTSD was prominent in 4 distinct clusters (Clusters 1, 2, 3,and 5), TBI was prominent only in conjunction with PTSDand pain. Although not identified as a signature injury of war,this analysis found that depression was a significant co-morbidity, which characterized 2 clusters (Clusters 1 and 3)both of which were also characterized by PTSD. Findingsindicate that these 2 clusters were most likely to have ad-verse outcomes and were also more likely to have a varietyof different forms of acute care including emergency care,and neurology, medicine, and psychiatric inpatient care.Thus, depression is also an important comorbidity that con-tributes substantially to health care utilization and adverseoutcomes.

Second, although clusters characterized by high like-lihood of pain (Clusters 1 and 5) had the highest prevalenceof opioid exposure (49% and 36%, respectively), opioidexposure in Cluster 6 was nearly 12%. Because we did nothave access to the number of days of medication received,we do not know whether these were short-term prescriptionsT

AB

LE

4.

Ad

vers

eO

utc

om

es

inFY

10:

Com

pari

son

sW

ith

the

Rela

tive

lyH

ealt

hy

Clu

ster

(Clu

ster

6)

%

OR

(95%

CI)

Cluster

1:PolytraumaClinicalTriad,

Depression,Chronic

Disease

N=

8801(4.6%

)

Cluster

2:

PolytraumaClinical

Triad

N=

16,585(8.6%

)

Cluster

3:Mental

Health,Substance

Abuse

N=

45,242(23.6%

)

Cluster

4:Sleep,

Amputation,Chronic

Disease

N=

8158(4.3%

)

Cluster

5:Pain,

Moderate

PTSD

N=

10,769

(5.6%

)

Cluster

6:

Relatively

Healthy

N=

102,242

(53.3%

)

GroupsWith

Sim

ilarResults

Ho

mel

ess

spec

ific

care

3.9

6.6

(5.8

–7.5

)2

.13

.5(3

.1–

4.0

)3

.66

.1(5

.6–

6.7

)1

.11

.8(1

.5–

2.3

)1

.32

.1(1

.8–

2.6

)0

.6R

efer

ence

1,

3>

2>

4,

5>

6O

ver

do

se0

.68

.4(5

.6–

12

.6)

0.2

2.5

(1.5

–4

.1)

0.3

4.5

(3.2

–6.4

)0

.11

.2(0

.52

–2

.9)

0.2

2.7

(1.5

–4.7

)

0.1

Ref

eren

ce1

,3

>(3

),2

,5

>4

,6

Su

icid

e-re

late

db

ehav

iors

1.4

13

.3(1

0.3

–1

7.2

)0

.87

.2(5

.6–

9.3

)1

.09

.3(7

.6–

11

.5)

0.4

3.6

(2.4

–5.3

)0

.43

.3(2

.3–

4.8

)0

.1R

efer

ence

1,

3>

2,

3>

4,

5>

6M

ort

alit

y0

.72

.9(2

.1–

3.8

)0

.31

.4(1

.1–

1.9

)0

.62

.5(2

.1–

2.9

)0

.41

.7(1

.2–

2.5

)0

.31

.4(1

.0–

2.0

)0

.2R

efer

ence

1,

3>

2,

4,

5>

6

CI

indic

ates

confi

den

cein

terv

al;

N,

clust

ersi

mil

arto

>1

gro

up;

OR

,odds

rati

o;

PT

SD

,post

trau

mat

icst

ress

dis

ord

er.

Medical Care � Volume 52, Number 2, February 2014 Beyond the Polytrauma Clinical Triad

r 2013 Lippincott Williams & Wilkins www.lww-medicalcare.com | 179

after a minor procedure or injury. We do, however, knowthat the majority of the prescriptions were forhydrocodone—the most commonly prescribed drug in theUnited States overall.32 In addition, the extremely highprevalence of opioids, skeletal muscle relaxants, and anti-depressants, etc. in Clusters 1 and 5 suggests that a sig-nificant number of individuals are receiving multiple centralnervous system drugs simultaneously, which may lead toadverse drug side effects and adverse outcomes.

Finally, it is notable that veterans of the NationalGuard/Reserve were overrepresented in several clusters.These clusters were also characterized by older age andhigher likelihood of being married, and 2 were characterizedby age-related chronic disease. This finding is consistentwith reports of service members, many of whom (66%) wereNational Guard/Reserve, seeking treatment for cardio-vascular disease while deployed in Iraq.14

LimitationsThese data provide a unique perspective on patterns of

comorbidity among Veterans of the Afghanistan and Iraqwars but have limitations that must also be considered. First,our data are derived from VA health care databases. As such,only those who have received care in the VA are included.There is no way to identify individuals who may also havebeen deployed to Desert Storm. Thus, Gulf War Illness fromprior deployments may also be represented in these clusters.This concern can be addressed, however, in future compo-nents of the study using survey data.

Moreover, the participants received VA care for sev-eral years. The extent to which those receiving care withinthe VA differ from those who opt to receive care elsewhereis not well understood, although Veterans in general arebetter off in terms of health and wealth compared withVeterans seeking VA care.33 Our data do not reflect non-VAcare or diagnoses received by individuals in non-VA settings,and our measure of mortality does not reflect cause of death,which was not available for this study. Thus, mortality mayhave been due to SRB, accidents, or chronic disease. Futureresearch is needed to explicate cause of death.

Moreover, use of dichotomous variables that only ac-count for whether a condition, service, or prescription waspresent/absent during the period examined does not provideinformation about disease severity or intensity of care re-ceived across clusters. These data were also cross-sectionalin nature. Some individuals had been in VA care for 8 years,whereas for others these were the first years of care.Longitudinal analysis beginning in the first year of VA careand continuing over an extended period of time is needed toidentify characteristics associated with development of newcomorbidities and/or increasing severity of initial co-morbidities.

Finally, these clusters were statistically derived, and itis very likely that these clusters may differ in a number ofways that could not be measured given the data available.34

Future research adding survey and medical chart abstractiondata will allow closer comparison of some currently un-known characteristics, and longitudinal analyses will allowus to determine whether these clusters represent stable

trajectories of comorbidity or whether there is change inpatterns of comorbidity over time.

This study provides an initial perspective to assistclinicians both in and outside the VA system in under-standing patterns of comorbidity clusters among OEF/OIFVeterans and subsequent care patterns and adverse outcomesassociated with these clusters. Information from these datamay inform basic science and stimulate research that in-vestigates common mechanisms for specific conditions (eg,epilepsy and depression), defines integrated treatment ap-proaches for Veterans with complex mental and physicalhealth comorbidities, and examines quality of care for Vet-erans with different constellations of comorbidity.

REFERENCES1. Gailey R, McFarland LV, Cooper RA, et al. Unilateral lower-limb loss:

prosthetic device use and functional outcomes in service members from

Vietnam War and OIF/OEF conflicts. J Rehabil Res Dev. 2010;47:317–331.2. Sayer N, Chiros C, Sigford B, et al. Characteristics and rehabilitation

outcomes among patients with blast and other injuries sustained during

the Global War on Terror. Arch Phys Med Rehabil. 2008;89:163–170.3. Hoge CW, Toboni HE, Messer SC, et al. The occupational burden of

mental disorders in the U.S. Military: psychiatric hospitalizations,

involuntary separations, and disability. Am J Psychiatry. 2005;162:

585–591.4. Hoge CW, Castro CA, Messer SC, et al. Combat duty in Iraq and

Afghanistan, mental health problems, and barriers to care. N Engl J

Med. 2004;351:13–22.5. Lew HL, Otis JD, Tun C, et al. Prevalence of chronic pain,

posttraumatic stress disorder, and persistent postconcussive symptoms

in OIF/OEF veterans: polytrauma clinical triad. J Rehabil Res Dev.

2009;46:697–702.6. Gironda R, Clark ME, Ruff RL, et al. Traumatic brain injury,

polytrauma, and pain: challenges and treatment strategies for the

polytrauma rehabilitation. Rehabil Psychol. 2009;54:247–258.7. Carlson KF, Kehle SM, Meis LA, et al. Prevalence, assessment, and

treatment of mild traumatic brain injury and posttraumatic stress

disorder: a systematic review of the evidence. J Head Trauma Rehabil.

2011;26:103–115.8. Taylor BC, Hagel EM, Carlson KF, et al. Prevalence and costs of co-

occurring traumatic brain injury with and without psychiatric dis-

turbance and pain among Afghanistan and Iraq War Veteran VA users.

Med Care. 2012;50:342–346.9. Nazarian D, Kimerling R, Frayne SM. Posttraumatic stress disorder,

substance use disorders, and medical comorbidity among returning U.S.

Veterans. J Trauma Stress. 2012;25:220–225.10. Milliken CS, Auchterlonie JL, Hoge CW. Longitudinal assessment of

mental health problems among active and reserve component soldiers

returning from the Iraq war. JAMA. 2007;298:2141–2148.11. Clark ME, Bair MJ, Buckenmaier CC 3rd, et al. Pain and combat

injuries in soldiers returning from Operations Enduring Freedom and

Iraqi Freedom: implications for research and practice. J Rehabil Res

Dev. 2007;44:179–194.12. White JM, Stannard A, Burkhardt GE, et al. The epidemiology of

vascular injury in the wars in Iraq and Afghanistan. Ann Surg.

2011;253:1184–1189.13. Pietrzak RH, Whealin JM, Stotzer RL, et al. An examination of the

relation between combat experiences and combat-related posttraumatic

stress disorder in a sample of Connecticut OEF-OIF Veterans. J

Psychiatr Res. 2011;45:1579–1584.14. Sullenberger L, Gentlesk PJ. Cardiovascular disease in a forward

military hospital during Operation Iraqi Freedom: a report from

deployed cardiologists. Mil Med. 2008;173:193–197.15. Defense Science Board Task Force. Deployment of Members of the

National Guard and Reserve in the Global War on Terrorism.

Washington, DC: Office of the Under Secretary of Defense for

Acquisition, Technology, and Logistics; 2007.

Pugh et al Medical Care � Volume 52, Number 2, February 2014

180 | www.lww-medicalcare.com r 2013 Lippincott Williams & Wilkins

16. Zouris JM, Wade AL, Magno CP. Injury and illness casualtydistributions among U.S. Army and Marine Corps personnel during

Operation Iraqi Freedom. Mil Med. 2008;173:247–252.17. Cohen BE, Marmar C, Ren L, et al. Association of cardiovascular risk

factors with mental health diagnoses in Iraq and Afghanistan War

Veterans using VA health care. JAMA. 2009;302:489–492.18. Frayne SM, Chiu VY, Iqbal S, et al. Medical care needs of returning

veterans with PTSD: their other burden. J Gen Intern Med. 2011;26:33–39.19. Agency for Healthcare Research and Quality. Multiple chronic

conditions. Available at: http://www.ahrq.gov/professionals/prevention-

chronic-care/decision/mcc. Accessed July 21, 2013.20. Deyo RA, Cherkin DC, Ciol MA. Adapting a clinical comorbidity index

for use with ICD-9-CM administrative databases. J Clin Epidemiol.

1992;45:613–619.21. Elixhauser A, Steiner C, Harris DR, et al. Comorbidity measures for use

with administrative data. Med Care. 1998;36:8–27.22. Ruff RL, Ruff SS, Wang XF. Headaches among Operation Iraqi

Freedom/Operation Enduring Freedom Veterans with mild traumatic

brain injury associated with exposures to explosions. J Rehabil Res Dev.

2008;45:941–952.23. Wallace DM, Shafazand S, Ramos AR, et al. Insomnia characteristics and

clinical correlates in Operation Enduring Freedom/Operation Iraqi Free-

dom Veterans with post-traumatic stress disorder and mild traumatic brain

injury: an exploratory study. Sleep Med. 2011;12:850–859.24. Jakupcak M, Luterek J, Hunt S, et al. PTSD symptom clusters in

relationship to alcohol misuse among Iraq and Afghanistan war veterans

seeking post-deployment VA health care. Addict Behav. 2010;35:840–843.

25. Bandeen-roche K, Miglioretti DL, Zeger SL, et al. Latent variableregression for multiple discrete outcomes. J Am Stat Assoc.1997;92:1375–1386.

26. Carter AC, Capone C, Short EE. Co-occurring posttraumatic stressdisorder and alcohol use disorders in Veteran populations. J DualDiagn. 2011;7:285–299.

27. Helmer DA, Chandler HK, Quigley KS, et al. Chronic widespread pain,mental health, and physical role function in OEF/OIF veterans. PainMed. 2009;10:1174–1182.

28. Petrakis IL, Rosenheck R, Desai R. Substance use comorbidity amongveterans with posttraumatic stress disorder and other psychiatric illness.Am J Addict. 2011;20:185–189.

29. Seal KH, Bertenthal D, Miner CR, et al. Bringing the war back home:mental health disorders among 103,788 US Veterans returning from Iraqand Afghanistan seen at Department of Veterans Affairs facilities. ArchIntern Med. 2007;167:476–482.

30. Lemaire CM, Graham DP. Factors associated with suicidal ideation inOEF/OIF Veterans. J Affect Disord. 2010;130:231–238.

31. Seal KH, Shi Y, Cohen G, et al. Association of mental health disorderswith prescription opioids and high-risk opioid use in US Veterans ofIraq and Afghanistan. JAMA. 2012;307:940–947.

32. Manchikanti L, Helm S 2nd, Fellows B, et al. Opioid epidemic in theUnited States. Pain Physician. 2012;15:ES9–38.

33. Morgan RO, Teal CR, Reddy SG, et al. Measurement in VeteransAffairs Health Services research: Veterans as a special population.Health Serv Res. 2005;40:1573–1583.

34. Shepperd S, Lewin S, Straus S, et al. Can we systematically review studiesthat evaluate complex interventions? PLoS Med. 2009;6:e1000086.

Medical Care � Volume 52, Number 2, February 2014 Beyond the Polytrauma Clinical Triad

r 2013 Lippincott Williams & Wilkins www.lww-medicalcare.com | 181