• Pretrauma Cognitive Ability and Risk for Posttraumatic Stress ...

Sponsored Links

Download the ebook

Pretrauma Cognitive Ability and Risk
for Posttraumatic Stress Disorder
A Twin Study
William S. Kremen, PhD; Karestan C. Koenen, PhD; Corwin Boake, PhD; Shaun Purcell, PhD;
Seth A. Eisen, MD; Carol E. Franz, PhD; Ming T. Tsuang, MD, PhD; Michael J. Lyons, PhD
Context: Cognitive deficits are associated with post- PTSD. After controlling for confounders, the highest cog-
traumatic stress disorder (PTSD), but whether such defi- nitive ability quartile had a 48% lower risk than the low-
cits reflect sequelae or risk factors is not fully resolved. est ability quartile (P .001). Non≠PTSD-concordant pairs
had the highest scores; PTSD-concordant pairs had the
Objective: To determine, in a representative sample, lowest scores; and PTSD-discordant pairs had interme-
whether preexposure cognitive ability is associated with diate scores. Differences in Armed Forces Qualification
risk for PTSD, and whether that risk is genetically me- Test scores within twin pairs were significant only in
diated. PTSD-discordant pairs (P=.04) and were accounted for
specifically by the discordant dizygotic pairs (P=.002).
Design, Setting, and Participants: The co-twin≠
Genetic influences on preexposure cognitive ability ex-
control study involved 2386 male Vietnam-era twin vet-
plained 5% of the variation in PTSD, but 100% of that
erans with a mean (SD) age of 41.9 (2.7) years, a popu-
relationship was explained by common genes.
lation-based sample of men who were in military service
during this era. Cognitive ability scores were obtained
Conclusions: Preexposure cognitive ability is a risk or
just before military induction at a mean (SD) age of 19.7
(1.5) years. Participants included only individuals who a protective factor for PTSD. The variance in PTSD ex-
were exposed to potentially traumatic events and under- plained by preexposure cognitive ability is accounted for
went preexposure cognitive testing. entirely by common genetic factors. Lower cognitive abil-
ity may be a marker of less adaptive coping against ad-
Main Outcome Measures: Armed Forces Qualifica- verse mental health consequences of exposure to poten-
tion Test (of cognitive ability) percentile scores and PTSD tially traumatic events. Further study of the potential
diagnosed by means of structured interviews. mechanisms through which cognitive ability confers risk
is needed.
Results: We found a significant dose-response relation-
ship between preexposure cognitive ability and risk for Arch Gen Psychiatry. 2007;64:361-368
OSTTRAUMATIC STRESS DISOR- quences or sequelae of the disorder, but
der (PTSD) is a frequent con- lower cognitive ability might be an indica-
sequence of war or disaster. tor of increased risk for PTSD as well.6 There
Estimates from the current have been several longitudinal studies of
Iraq war indicate that as many premorbid cognitive ability in psychiatric
as 30% of combat veterans have signifi- illness; most of them have examined schizo-
cant mental health problems, including phrenia, whereas a few have examined
PTSD, thus making it a serious public health mood disorders. As reviewed by Zammit et
issue.1 Even if this figure is an overesti- al,7 these studies provide generally consis-
mate owing to its reliance on self-report tent evidence that lower premorbid cogni-
measures, the data still strongly suggest that tive ability is associated with a greater risk
PTSD is a serious public health issue. Cog- for schizophrenia, whereas the evidence has
nitive deficits have been frequently ob- been mixed for mood disorders. Premor-
served in PTSD, but this relationship has bid cognitive ability in PTSD has been far
almost always been based on studies that less well studied.
have assessed cognition after PTSD has de- We are aware of 3 reports in which pre-
Author Affiliations are listed at veloped.2-5 For the most part, cognitive defi- exposure cognitive ability has been as-
the end of this article. cits have been assumed to be conse- sessed, each of which contained several
Downloaded from www.archgenpsychiatry.com on April 28, 2011
©2007 American Medical Association. All rights reserved.
dicator of risk for PTSD and to determine the strength
Table 1. Demographic and Clinical Characteristics of this effect. We addressed this issue through our ex-
of 2386 Twin Veterans amination of a large sample of male twins who served in
the US military during the Vietnam era (1965-1975). We
Characteristic Finding
controlled for the following potential confounders: age
Continuous Variables, Mean (SD) and education at entry into the military, combat expo-
Age at study, y 41.9 (2.6) sure, and parental education. Use of a co-twin≠control
Age at military induction, y 19.7 (1.5)
Education at military induction, y 12.1 (1.3)
design ensured maximal similarity of the groups being
AFQT percentile score* 55.2 (23.4) compared and enabled us to make inferences about the
Mother's years of education 11.7 (2.5) presence of genetic influences. For example, given lower
Father's years of education 11.10 (3.2) preexposure cognitive ability as a risk factor for PTSD,
Categorical and Ordinal Variables, No. (%) we hypothesized that cognitive test scores in twins with
Combat exposure index PTSD would be lower than those of their trauma-
No SEA service 1231 (51.6) exposed co-twins without PTSD.
SEA service without combat exposure 170 (7.1)
Low combat exposure 262 (11.0)
Medium combat exposure 283 (11.9) METHODS
High combat exposure 440 (18.4)
PTSD diagnosis
Yes 436 (18.3)
No 1950 (81.7)
The study was approved by the institutional review boards of
Abbreviations: AFQT, Armed Forces Qualification Test; PTSD,
the participating institutions, and all participants gave in-
posttraumatic stress disorder; SEA, Southeast Asia. formed consent to participate. Participants were recruited from
*The mean is slightly above average (the 50th percentile) because the Vietnam Era Twin Registry16 as part of the Harvard Twin
candidates scoring below the 10th percentile were statutorily excluded from Study of Drug Abuse.17 Efforts to contact all available registry
military service. members resulted in interviews of 8269 (80.3%) of 10 300 eli-
Includes only trauma-exposed twins; 9.6% of the entire sample was gible men that were completed in 1992. Twins for the present
diagnosed as having PTSD.
report were drawn from the 3322 pairs in which both mem-
bers participated.17 Interviews were conducted by trained re-
search assistants. Registry members are demographically rep-
methodological limitations. Two studies--which had rela- resentative of military personnel from that era.18 Zygosity was
tively small sample sizes--found that lower precombat determined on the basis of questionnaire and blood group meth-
ods, an approach that achieves approximately 95% accuracy
cognitive ability scores in Vietnam veterans were asso- compared with DNA analysis.19 The racial/ethnic breakdown
ciated with a greater likelihood of developing PTSD.8,9 of the sample was 90.4% non-Hispanic white, 4.9% African
As noted by the authors, the first8 did not control for the American, 2.7% Hispanic, 1.3% Native American/Alaskan na-
level of combat exposure, which could by itself account tive, and 0.7% other. One third of the sample (33.3%) were high
for the increased risk of PTSD, and neither used repre- school graduates; 38.6% were college graduates; 92.6% were
sentative case or control samples.8,9 In the more recent employed full-time; and 1.8% were employed part-time.
study by Macklin et al,9 the comparison group of veter- Participants were excluded from the present analyses if they
ans without PTSD had cognitive ability scores approxi- (1) reported no history of exposure to a traumatic event (be-
mately 1 SD above average. Age and education at entry cause, by definition, they could not be at risk for PTSD); (2)
into the military and childhood socioeconomic status tend had missing or invalid cognitive ability scores; or (3) reported
having experienced a traumatic event before military service
to be associated with cognitive ability and risk for (ie, before their cognitive testing). The remaining sample con-
PTSD,10,11 but these previous studies were also unable to sisted of 1328 monozygotic (MZ) and 1058 dizygotic (DZ) twins;
control for those variables. their demographic and clinical characteristics are shown in
An Israeli study that used an epidemiological sample Table 1.
also found that individuals who developed PTSD had
lower premorbid intellectual ability than those who did MEASURES
not; however, there was no control for exposure to com-
bat in that study, and control participants were not nec- Age and education at military induction were abstracted from
essarily exposed to trauma at all.12 Also, clinical-- the original military personnel file, including the DD4 and DA-20
rather than research-interview≠based--diagnoses were forms. Trauma history was assessed using the Diagnostic In-
used in that study,12 and the diagnoses were based on terview Schedule Version III≠Revised.20 Diagnoses were based
DSM-III-R criteria13 for some individuals and DSM-IV cri- on DSM-III-R criteria. The interview includes inquiries about
teria14 for others. Use of these different DSM editions may the occurrence of a number of traumatic events. If a respon-
be an important issue because the DSM-III-R stressor cri- dent endorses one of these events or another qualifying expe-
terion is more stringent and results in a much lower life- rience, the interviewer gathers information about age at expo-
sure, age at the onset of symptoms, symptom type, and duration.
time prevalence of PTSD than does that of DSM-IV. Con-
Premilitary trauma history was recorded if a twin reported a
cerns have been raised about overclassification of people traumatic event with an age of onset that preceded age of in-
as trauma survivors on the basis of DSM-IV criteria.15 duction into the military. This information is collected on up
A large and representative epidemiological sample in to 3 traumatic events. Only a minority of registry twins re-
which such factors are taken into account is necessary ported more than 1 trauma and, of those, PTSD was almost al-
to confirm lower preexposure cognitive ability as an in- ways associated with the earliest trauma. Approximately 34.7%
Downloaded from www.archgenpsychiatry.com on April 28, 2011
©2007 American Medical Association. All rights reserved.
of those reporting trauma exposure cited combat as their worst Cognitive Ability in Twin Pairs Concordant
trauma. A total of 3065 participants, consisting of 46.1% of the and Discordant for PTSD
sample, reported exposure to 1 or more traumatic events, of
whom 649 (21.2%) received lifetime diagnoses of PTSD.21 De- Our second set of analyses examined risk for PTSD in twin pairs
tails of the interview procedure, types of traumatic events re- concordant and discordant for the disorder. Eligibility for this
ported, and PTSD diagnostic data were reported previously.10 analysis was the same as for the previous analysis except that
Combat exposure was based on the number of reported com- both members of a twin pair had to meet the criteria. The final
bat experiences with distinct characteristics on the Combat Ex- sample consisted of 616 eligible twin pairs, including 43 pairs
posure Index.22 Ratings on this scale were as follows: 0 (no South- concordant for having PTSD, 185 pairs discordant for PTSD,
east Asia service); 1 (Southeast Asia service without combat and 383 pairs concordant for not having PTSD. If lower pre-
exposure); 2 or 3 (low combat exposure); 4 to 6 (medium com- exposure cognitive ability is a risk factor for PTSD (rather than
bat exposure); and 7 or higher (high combat exposure). Pre- a sequela of illness), then AFQT percentiles in discordant pairs
vious work has shown that the ordinal combat categories have should be lower in the twins with PTSD compared with their
good internal consistency ( =0.86) and test-retest reliability trauma-exposed co-twins without PTSD. The discordant co-
( =0.84).22 twin≠control comparison is particularly useful because co-
The preexposure cognitive ability measure was the Armed twins are far more closely matched than other individuals. Be-
Forces Qualification Test (AFQT),23 a group-administered tween-pair comparisons should yield higher scores in non≠
screening test used to determine whether draftees and enlist- PTSD-concordant pairs (where neither twin had PTSD) and
ment applicants met the minimal standards for military ser- lower scores in PTSD-concordant pairs. Neither of the concor-
vice. These scores were obtained from the DD4 and DA-20 forms. dant-pair groups should manifest significant within-pair dif-
The AFQT consisted of 100 multiple-choice items with the fol- ferences. Given the clear directional hypothesis, we used 1-tailed
lowing 4 equal components: vocabulary, arithmetic, spatial vi- matched-pairs t tests to assess within-pair differences. We then
sualization, and tool identification/understanding. Total raw used ordinary least squares regression, controlling for con-
scores were corrected for guessing and converted to percentile founders, to test the association between preexposure cogni-
scores referenced to the male population who had been mobi- tive ability and risk for PTSD within and between twin pairs.
lized for military service during World War II. Separate scores
for the 4 components are not available. Individuals scoring be-
low the 10th percentile were excluded from the military; thus, Genetic Influences on the Association
the AFQT percentile scores in this sample range from 10 to 99. Between PTSD and Preexposure Cognitive Ability
The AFQT was designed to measure military trainability rather
than intelligence, but it does appear to be a highly g-loaded mea- Our third set of analyses involved testing whether there were
sure (ie, an index of general cognitive ability) that correlates genetic influences underlying the association between PTSD
well with traditional IQ measures.24 After correcting for restric- and preexposure cognitive ability by adding a term for the in-
tion of range, McGrevy et al25 reported a correlation of 0.84 be- teraction of zygosity and the within-pair effect to the regres-
tween scores on the AFQT and the Wechsler Adult Intelli- sion model. If preexposure cognitive ability is more strongly
gence Scale26 in a sample that was demographically similar to associated with PTSD within PTSD-discordant DZ pairs than
the present sample. More recently, Orme et al27 found that the within PTSD-discordant MZ pairs (ie, a significant interaction
correlation between the AFQT and the Multidimensional As- is present), it suggests that genetic mediation is likely. This logic
sessment Battery (an intelligence measure very similar to the may not be readily apparent to someone who is unfamiliar with
Wechsler Adult Intelligence Scale [r=0.91]) was 0.85 after cor- behavioral genetics. If any trait or association between traits is
recting for restriction of range. due to genes, then MZ twins cannot differ because they are ge-
netically identical. Therefore, if the aforementioned associa-
tion is genetically mediated, differences would have to be ac-
counted for by DZ twin pairs, who share on average only 50%
STATISTICAL ANALYSIS of their genes. The interaction being tested is, thus, one in which
there is a significant difference within DZ pairs, but not within
Phenotypic Analysis of Preexposure Cognitive MZ pairs.
Ability and Risk for PTSD We then tested whether the association of cognitive ability
with PTSD was accounted for by shared genetic etiology by using
We first conducted phenotypic analyses using logistic regres- maximum likelihood estimation techniques to fit different struc-
sion to test the relation of preexposure cognitive ability with tural equations to raw data. Structural equation models were
the risk for PTSD before and after controlling for potential fit in Mplus version 3.11 statistical software30 (estimator,
confounders. Twins were treated as singletons for this analy- weighted least squares; parameterization, theta). The full sample
sis, and the sandwich variance estimator was used to correct of twin pairs (N=3322 pairs) was included in this analysis. Be-
for nonindependence of data from twins in the same fam- cause exposure to a potentially traumatic event is a criterion
ily.28,29 Participants were excluded from these phenotypic for the diagnosis of PTSD, veterans who did not report trauma
analyses if they (1) reported no history of exposure to a po- exposure were coded as missing. Participants were not ex-
tentially traumatic event (because, by definition, they could cluded owing to missing data; Mplus uses full information maxi-
not be at risk for PTSD; n=3678); (2) had missing or invalid mum likelihood estimation to retain the complete sample size
cognitive ability scores (n=702); or (3) reported having expe- for each analysis.
rienced a potentially traumatic event before military service We first computed correlations for MZ and DZ pairs to as-
(ie, before their cognitive testing; n = 376). The remaining sess the cross-twin cross-trait correlations for the AFQT score
sample for phenotypic analyses consisted of 1328 monozy- and PTSD. In the univariate twin model, the variance for the
gotic (MZ) and 1058 dizygotic (DZ) twins. The sample was AFQT score or PTSD is partitioned into the variance due to ad-
divided into quartiles based on AFQT scores, and odds ratios ditive genetic (A), common environmental (C), and individual-
(ORs) with 95% confidence intervals (CIs) were calculated to specific environmental influences (E), including error (the ACE
indicate risk for PTSD relative to the reference group, which model). In the bivariate twin analysis, MZ and DZ correla-
was the lowest-scoring AFQT quartile. tions are compared across traits, ie, one twin's AFQT score is
Downloaded from www.archgenpsychiatry.com on April 28, 2011
©2007 American Medical Association. All rights reserved.
Table 2. Preexposure Cognitive Ability (AFQT Percentile Scores) and Risk for PTSD Compared With the Reference Group*
AFQT Group, Quartiles AFQT Percentile Score Range No. of Subjects OR (95% CI) P Value
Comparison With Reference Group
Low (reference group) 10-33 574 ... ...
Medium-low 34-55 606 0.74 (0.56-0.98) .04
Medium-high 56-75 612 0.56 (0.42-0.75) .001
High 76-99 594 0.42 (0.31-0.58) .001
Comparison With Reference Group After Controlling for Confounders
Low (reference group) 10-33 574 ... ...
Medium-low 34-55 606 0.87 (0.64-1.17) .35
Medium-high 56-75 612 0.69 (0.50-0.95) .02
High 76-99 594 0.52 (0.37-0.73) .001
Abbreviations: AFQT, Armed Forces Qualification Test; CI, confidence interval; OR, odds ratio; PTSD, posttraumatic stress disorder; ellipses, not applicable.
*The risk of PTSD relative to the reference group is 1 minus the odds ratio. For example, an odds ratio of 0.74 indicates a 26% lower risk relative to the
reference group.
correlated with the co-twin's PTSD diagnosis. If the cross-trait ticipants were more likely to have served in Southeast
twin correlations are greater for MZ than for DZ twins, this im- Asia, to have been exposed to combat, and to have been
plies that genetic factors contribute to the phenotypic correla- younger and have had less than a high school education
tion between the 2 traits. A significant path from additive ge- on entry into the military (P .001 for all). These com-
netic influences (A) on the AFQT to PTSD indicates the extent
parisons strongly argue against any selection bias that
to which genetic influences on the AFQT score also influence
variation in PTSD. A significant path from shared environmen- would have influenced the results.
tal (C) influences on the AFQT score to PTSD indicates the ex-
tent to which shared environmental influences on the AFQT PHENOTYPIC ANALYSIS OF PREEXPOSURE
score also influence PTSD. A significant path from nonshared COGNITIVE ABILITY AND RISK FOR PTSD
environmental influences (E) on the AFQT score to PTSD in-
dicates the extent to which nonshared or unique environmen-
Table 2 indicates that lower preexposure cognitive abil-
tal influences on the AFQT score also influence PTSD.
We assessed model fit via 3 fit statistics. When the models ity was significantly associated with increased risk of PTSD
are nested (ie, identical with the exception of constraints that in a dose-response fashion. For example, the OR of 0.42
involve setting specific parameters to 0), the difference in fit in the highest cognitive ability group indicates that in-
between models can be tested by 2
, using as its degrees of dividuals in this group are at 58% lower risk of develop-
freedom the difference in degrees of freedom of the 2 models. ing PTSD compared with those in the lowest cognitive
If the is statistically nonsignificant, the more parsimoni- ability (reference) group. We then tested whether the cog-
ous nested submodel is selected because the test indicates that nitive ability≠PTSD relation was attenuated after adjust-
the model fit does not deteriorate with the additional con- ing for confounders. Several potential confounders were
straints. The second model-selection statistic was the compara- associated with risk for PTSD, including combat expo-
tive fit index, for which values greater than 0.95 are indicative
of good-fitting models.31 The third model-selection statistic was
sure (OR, 1.30 [95% CI, 1.21-1.41; P .001]), age at mili-
the root mean square error of approximation, which is an in- tary entry (OR, 0.88 [95% CI, 0.81-0.96; P=.005]), and
dex of the model discrepancy, per degree of freedom, from the having less then a high school education at military en-
observed covariance structure.32 Values less than 0.05 indi- try (OR, 1.54 [95% CI, 1.13-2.10; P=.006]). However, as
cate close fit, and values less than 0.08 indicate fair fit to the can be seen in Table 2, the strength of the association
data.33 between cognitive ability and risk for PTSD remained simi-
lar after adjusting for confounders. Figure 1 illustrates
RESULTS the clear linear relationship between preexposure cog-
nitive ability and PTSD prevalence.
The demographic and clinical characteristics for partici-
pants included in the phenotypic analyses are shown in COGNITIVE ABILITY IN TWIN PAIRS
Table 1. Participants included in this analysis did not dif- CONCORDANT AND DISCORDANT FOR PTSD
fer from those excluded on zygosity (MZ, 57.7% for those
included and 55.9% for those excluded; 1 =0.02 [P=.88]), Members of discordant pairs differed significantly in pre-
maternal education of less than high school (26.0% for exposure cognitive ability; PTSD was associated with lower
those included and 27.2% for those excluded; 1 =1.19 scores in twins with PTSD compared with their non-
[P=.28]), or paternal education (39.0% for those in- PTSD co-twins (t185 =1.79; P=.04, 1-tailed test). Differ-
cluded and 38.6% for those excluded; 1 = 0.09 [P=.77]). ences within both types of concordant pairs were non-
The AFQT scores also did not differ when those in- significant (P .43 for all). As seen in Figure 2, the
cluded were compared with those excluded for reasons pattern of preexposure cognitive ability was such that it
other than invalid or missing AFQT data (b=-0.005 [SE, was also highest in non≠PTSD-concordant pairs and low-
0.61; P=.99]). Because selection criteria were based on est in PTSD-concordant pairs (described in the next para-
exposure to a potentially traumatic event, included par- graph). Results were similar if the AFQT score was stan-
Downloaded from www.archgenpsychiatry.com on April 28, 2011
©2007 American Medical Association. All rights reserved.
30 60
P = .48
55 P = .04
n = 144 P = .44
AFQT Percentile Score
PTSD Prevalence, %
n = 121
15 n = 97
n = 74
0 25
Group 1 Group 2 Group 3 Group 4
(n = 574) (n = 606) (n = 612) (n = 594)
Non-PTSD Concordant PTSD Discordant PTSD Concordant
(n = 383 Pairs) (n = 185 Pairs) (n = 43 Pairs)
Figure 1. Prevalence of posttraumatic stress disorder (PTSD) by
Armed Forces Qualification Test (AFQT) score quartile obtained before the Figure 2. Mean Armed Forces Qualification Test (AFQT) scores before
veteran twins' exposure to trauma. Sample includes all individuals exposed trauma exposure in twin pairs. Sample includes only pairs in which both
to trauma after having taken the AFQT. Group 1 has low (range, 10-33 twins were exposed to trauma after having taken the AFQT. PTSD indicates
points); group 2, medium to low (range, 34-55 points); group 3, medium to posttraumatic stress disorder. *Statistically significant P value.
high (range, 56-75 points); and group 4, high (range, 76-99 points) AFQT
dardized and treated as a continuous variable in the logistic
regression analysis (unadjusted OR, 0.73 [95% CI, 0.65- 55
0.80; P .001]; adjusted OR, 0.77 [95% CI, 0.69-0.88; P = .02 P = .86
AFQT Percentile Score 50
P .001]).
In the ordinary least squares regression, the within- 45
pair components test for the association between preex-
posure cognitive ability and PTSD diagnosis after con- 40
trolling for between-pair confounders (eg, parental
education). After controlling for confounders, both the
between-pair (b=-4.96 [SE, 2.55; P=.05]) and the within- 30
pair (b = -3.78 [SE, 1.71; P=.03]) effects were signifi-
cant. For the between-pair effect, PTSD-concordant pairs 25
had AFQT scores that averaged 4.96 percentile points
DZ Twin Pairs MZ Twin Pairs
lower than those of non≠PTSD-concordant pairs; discor- (n = 94 Pairs) (n = 91 Pairs)
dant pairs had AFQT scores that were 2.48 percentile
points lower than those of non≠PTSD-concordant pairs. Figure 3. Mean Armed Forces Qualification Test (AFQT) scores before
Within-pair comparisons of PTSD-discordant twins trauma exposure in monozygotic (MZ) vs dizygotic (DZ) twin pairs
showed that twins with PTSD had AFQT scores that av- discordant for posttraumatic stress disorder (PTSD). Sample includes only
pairs in which both twins were exposed to trauma after having taken the
eraged 3.87 percentile points lower than their co-twins AFQT. For the interaction of zygosity and the within-pair AFQT effect,
without PTSD. The between-pair and within-pair ef- P = .002. *Statistically significant P value.
fects were not significantly different from one another
(Wald test, F1,709 = 0.15 [P= .70]).
Table 3 presents the correlations for AFQT score and
GENETIC INFLUENCES ON PTSD by zygosity. The higher cross-twin cross-
THE ASSOCIATION BETWEEN PTSD phenotype correlations among MZ twins compar

Use: 0.0838