504115

research-article2013
PPSXXX10.1177/1745691613504115Galatzer-Levy, BryantPTSD Heterogeneity

Perspectives on Psychological Science

636,120 Ways to Have Posttraumatic 8(6) 651­–662

© The Author(s) 2013
Reprints and permissions:
Stress Disorder sagepub.com/journalsPermissions.nav
DOI: 10.1177/1745691613504115
pps.sagepub.com

Isaac R. Galatzer-Levy1 and Richard A. Bryant2

1
New York University School of Medicine; and 2University of New
South Wales, Kensington, New South Wales, Australia

Abstract
In an attempt to capture the variety of symptoms that emerge following traumatic stress, the revision of posttraumatic
stress disorder (PTSD) criteria in the 5th edition of the Diagnostic and Statistical Manual of Mental Disorders (DSM–5)
has expanded to include additional symptom presentations. One consequence of this expansion is that it increases
the amorphous nature of the classification. Using a binomial equation to elucidate possible symptom combinations,
we demonstrate that the DSM–IV criteria listed for PTSD have a high level of symptom profile heterogeneity (79,794
combinations); the changes result in an eightfold expansion in the DSM–5, to 636,120 combinations. In this article,
we use the example of PTSD to discuss the limitations of DSM-based diagnostic entities for classification in research
by elucidating inherent flaws that are either specific artifacts from the history of the DSM or intrinsic to the underlying
logic of the DSM’s method of classification. We discuss new directions in research that can provide better information
regarding both clinical and nonclinical behavioral heterogeneity in response to potentially traumatic and common
stressful life events. These empirical alternatives to an a priori classification system hold promise for answering
questions about why diversity occurs in response to stressors.

Keywords
posttraumatic stress disorder (PTSD), diagnosis, DSM–5, heterogeneity, combinatorics, latent growth mixture
modeling

Diagnostic categories are commonly used for research the initial and evolving role the DSM has played as a
and treatment of mental illness. These categories allow research tool. We first examine the history of the DSM
clinicians and researchers to identify those who have a and the posttraumatic stress disorder (PTSD) diagnosis.
disorder on the basis of a clear set of symptoms and We discuss how attempts to clearly define diagnostic
symptom categories. By accurately defining symptom entities led to built-in measurement issues that limit the
requirements for the illness, researchers and clinicians utility of diagnoses for research or the evaluation of treat-
can identify predictors and correlates of specific mental ment effects. We focus on the underlying mathematical
illnesses and subsequently develop and apply more tar- foundations of this scheme to demonstrate that as the
geted treatments. Abstractly, this approach seems simple DSM moves through permutations, the diagnostic criteria
and clear, achieving the ongoing goal of the Diagnostic rarely become clarified and in some cases become
and Statistical Manual of Mental Disorders (DSM) frame- increasingly obscured as the diagnosis expands to
work to create well-defined, nonobscure diagnoses to aid encompass more heterogeneous presentations. Finally,
in research and treatment (Spitzer, Endicott, & Robins, we discuss recent developments that have emerged
1978). However, upon closer scrutiny, we find that many as a potentially more robust method for identifying
of the diagnostic classifications are not simple or specific. meaningful clinical outcomes without reliance on DSM
Individuals with the same diagnosis can have remarkably diagnoses.
distinct symptom presentations. This calls the use of DSM
diagnoses as research and treatment tools into question.
Corresponding Author:
This article describes the current diagnostic system, Isaac R. Galatzer-Levy, New York University School of Medicine,
which often relies on identifying combinations of subsets 1 Park Ave, New York, NY 10016
of symptoms, or criteria, to define diagnoses. We describe E-mail: isaac.gl@gmail.com
652 Galatzer-Levy, Bryant

We focus primarily on PTSD because it is a good best available evidence” with clearly defined cutoffs
example of how many factors, both historical and math- separating normality from pathology (Galatzer-Levy &
ematical, have led to a nondescript classification. To Galatzer-Levy, 2007; M. Wilson, 1993). Common symp-
demonstrate this point, we show how the PTSD diagno- toms were left out if clinicians were not reliable in record-
sis has expanded to encompass increasing numbers of ing them. For example, “blunted affect,” a commonly
populations and presentations, leading to increasing observed symptom of schizophrenia where the individ-
underlying heterogeneity in the diagnosis. Next, we dis- ual displays a limited emotional range, was not included
cuss empirical efforts to clarify the DSM PTSD diagnosis, in the diagnosis because clinicians were unreliable in
the current findings related to PTSD, and the limitations observing it (Spitzer et al., 1978).
of these efforts in the light of the measurement problems The DSM as it appears now typically offers diagnostic
intrinsic to the diagnosis. Finally, we discuss the rele- categories that are made up of what are termed “criteria.”
vance of the issues highlighted by the problems with These are symptom clusters that are heuristically grouped
diagnosis in nonclinical contexts where researchers are together. Each diagnosis has one or several criteria, each
interested in identifying discrete populations on the basis of which contains either one or several possible symp-
of behavioral characteristics. toms. If the criterion has more than one possible symp-
tom, there is often a rule that in order for said criterion to
A Brief History of DSM Diagnoses as a be met, a certain number of those symptoms must be
present. For the overall diagnosis to be given, a specified
Measurement Tool number of criteria must also be met.
Despite their use in medical, psychological, and socio-
logical research, DSM diagnoses were not initially
A Brief History of the PTSD Diagnosis
intended or designed to explore basic science hypothe-
ses (Kraemer, 2007). Some of the confusion with regard PTSD was first defined in the DSM–III (American
to their intended use may come from the word “statisti- Psychiatric Association, 1980) following intense political
cal” in the DSM title. This term was initially in reference pressure placed on the mental health field to recognize
to the new ability to count the number of individuals the psychological effects of war that were observed
meeting a specific diagnosis in mental hospitals and among Vietnam veterans as well as concentration camp
in the general population, not as an allusion to the diag- survivors (Helzer, Robins, & McEvoy, 1987). The diagno-
nosis’s value as a statistical variable (Kraemer, 2007). sis of PTSD was proposed to recognize that persistent
However, the initial iterations of the DSM (DSM–I and psychological reactions to horrific events represented an
DSM–II) were heavily criticized as being overly broad, illness requiring care and treatment, not cowardice or the
vague, and theoretically driven. In particular, these diag- manifestation of a previous psychiatric illness, such as
nostic classifications were ineffective at differentiating depression, hysteria, or psychosis. The recognition of this
mental disorders and those who were sick from those unique disorder laid the groundwork for governments to
who were healthy (M. Wilson, 1993), making them inad- provide specific mental health services to veterans who
equate for the purpose for which they were designed. had previously been ignored, court-martialed, or sent
Beginning with the DSM–III, a concerted effort was away to mental hospitals for generic treatments (Gersons
made to create a system where the diagnosis was defined & Carlier, 1992).
only in terms of observable symptoms to combat previ- PTSD was defined in the DSM–III to include a total min-
ous criticisms. The primary goal was to create a classifica- imum of 4 of a possible 12 symptoms from three symptom
tion system that was highly reliable, meaning that any criteria. The event itself had to be one that was outside of
clinician observing the same patient independently normal human experience and would be considered dis-
would reach the same diagnosis. Vague terms, such as tressing to almost anyone. Consistent with other DSM–III
“neurosis” and “psychosis,” were removed in favor of spe- diagnoses, the PTSD diagnosis was meant primarily for
cific, observable, symptom-based criteria for each diag- differential diagnosis (i.e., differentiating the disorder from
nosis (Spitzer, Endicott, & Robins, 1975). An explicit other disorders). The diagnosis was intended to identify
assumption was made that the validity of diagnoses severe and persistent fear responses and was not meant as
would follow from the establishment of reliability and an exhaustive list of possible symptoms of any posttrau-
that clear diagnostic criteria could be used as a founda- matic condition (J. P. Wilson, 1994).
tion for clinical research by relating these classifications Along with the rest of the DSM, PTSD was further
to genetics and psychobiology (Spitzer et al., 1978). The codified following a series of criticisms of the manual,
DSM task force, a select group of clinicians, administra- leading to the revised DSM–III (DSM–III–R; American
tors, and researchers, was formed with the goal of creat- Psychiatric Association, 1987; M. Wilson, 1993). The DSM–
ing a diagnostic manual that defined diagnoses only in III–R expanded the total potential symptoms from 12 to
terms of reliably observable symptoms based on “the 17 in response to criticism that the previous criteria were
PTSD Heterogeneity 653

narrowly focused on responses observed in Vietnam vet- observable symptoms with the goal of creating “the best
erans and Holocaust survivors and needed to better recog- possible classification system based on the latest avail-
nize responses of survivors of other events, such as natural able knowledge” (M. Wilson, 1993), it remains unclear
disasters (Helzer et al., 1987; McFarlane, 1988). The DSM– whether existing empirical knowledge at that time was a
III–R maintained the DSM–III’s criterion defining a trau- firm enough foundation to build from, at least in the case
matic event. Further, the diagnosis required a minimum of PTSD. Differentiating disorders from each other and
number of symptoms from each of the three other criteria: differentiating normality from pathology on the basis of
(a) reexperiencing symptoms, such as intrusive thoughts symptom presentation is central to the DSM endeavor.
or images of the event or “flashbacks,” where the individ- However, PTSD encompasses a number of common stress
ual feels as if he or she is back at the event; (b) avoidance symptoms that may typically occur in the general popula-
and numbing symptoms, such as efforts to avoid activities tion. A number of PTSD symptoms have been shown to
or situations associated with the traumatic event or dimin- emerge in response to minor oral surgery (de Jongh et al.,
ished interest in activities overall; and (c) arousal symp- 2008), routine childbirth (Olde, Hart, Kleber, & Son, 2006),
toms, such as difficulty falling or staying asleep or irritability and bad movies (Lees-Haley, Price, Williams, & Betz,
or outbursts of anger (for complete symptom listing for 2001). These findings make sense if we consider that
each criterion, see American Psychiatric Association, 1987). many of the symptoms of PTSD, such as nightmares and
Symptoms were chosen and heuristically grouped together sleep disturbances, represent common nondescript stress
by the DSM task force on the basis of “best available evi- symptoms (Bonanno, Galea, Bucciarelli, & Vlahov, 2006).
dence” consistent with the DSM scheme (J. P. Wilson, Sorting out the distribution of symptoms that is com-
1994; M. Wilson, 1993). As such, individuals were required mon in nonpathological populations from that of patho-
to display a minimum of 6 of 17 total possible symptoms logical populations is essential if decisions are to be
distributed across the three criteria. These symptoms were made about which symptoms or what level of symptoms
required to be sustained for at least 1 month, and, it is signal pathology. The first epidemiological studies of
important to note, any combination of symptoms would trauma-exposed populations meant to assess rates of
suffice as long as the required symptom count was met or clinical outcomes, to our knowledge, were the National
exceeded to meet each criterion. Vietnam Veterans Readjustment Study (Kulka et al., 1990),
Only a few alterations were made to the PTSD diag- commissioned by Congress; the Vietnam Experience
nosis for the DSM–IV. The definition of a traumatic event Study, conducted by the Centers for Disease Control
was expanded to be more inclusive and now encom- (Centers for Disease Control Vietnam Experience Study
passed experiencing, witnessing, or being otherwise Group, 1988); and the Epidemiologic Catchment Area
confronted with any event that involves actual or threat- Survey of Detroit-area youth (Breslau, Davis, Andreski, &
ened physical harm to self or others, and some symp- Peterson, 1991); all of these were conducted after the
toms switched criteria (American Psychiatric Association, establishment of the DSM–III–R PTSD criteria. As such,
1994). With DSM–5, a new criterion has been introduced the DSM committee had limited empirical data to differ-
of alterations in mood and cognition, which includes entiate abnormal from common stress response symp-
new symptoms of various states (e.g., guilt, shame, mis- toms. Salient general stress features may have been
trust). These have been added, in part, because they are chosen because they were commonly observed, not
commonly reported in PTSD presentations in military, because they were shown to separate normal from path-
emergency responder, and interpersonal violence popu- ological populations. This concern is reflected in sugges-
lations. Conceptually, this addition represents an exten- tions that future DSM and other diagnostic systems
sion from the traditional fear response focus of PTSD to identify and focus on rare rather than common symp-
encompass other affective responses. This expansion toms of posttraumatic stress (McNally, 2009).
has resulted in the DSM–5 work group defining PTSD
by four criteria and requiring a minimum of 8 of 19 pos- Attempts to Improve the Diagnosis by
sible total symptoms. It is important to note that PTSD,
along with many DSM diagnoses, requires meeting a
Empirically Identifying the Structure
final criterion that the symptoms cause significant dis- of PTSD
tress or impairment. Though criteria and symptoms have always ultimately
been decided by a DSM committee of experts who define
A Brief History of Epidemiological the diagnosis (Buckley, Blanchard, & Hickling, 1998),
attempts have been made over the years to put this prac-
Research Into PTSD
tice on stronger empirical footing. To this end, numerous
Though the modern DSM project, beginning with factor analytic studies have been conducted to empiri-
DSM–III, focused exclusively on diagnoses based on cally identify the correct criteria and symptoms of PTSD
654 Galatzer-Levy, Bryant

by determining what symptoms hang together into com- mental illness on the basis of a set of logical rules comes
mon factors along with the number of common factors. with a built-in limitation for validity that stands in con-
Factor analyses have revealed two- (Buckley et al., 1998; trast to the hopes that validity would naturally flow from
Taylor, Kuch, Koch, Crockett, & Passey, 1998), three- reliability. The DSM strives to create a set of rules that
(Larsson, 2005), and four-factor solutions (Asmundson et accurately discriminates cases of the disorder from non-
al., 2000; King, Leskin, King, & Weathers, 1998; Palmieri cases. However, it has been proven that any logical sys-
& Fitzgerald, 2005; Simms, Watson, & Doebbeling, 2002). tems that derive definitions from a set of axioms (rules)
The symptoms within a factor often vary across these are, by definition, incomplete ones (Gödel, 1992). The
studies even if the number of factors does not. The DSM– only formal logical way around this is to have an infinite
5 work group has come out in favor of four criteria based number of axioms that deterministically describe all pos-
in part on these factor analytic studies (APA, 2013). sible outcomes, past, present, and future. (The exception
The reason for the different factor solutions remains to this occurs only when the axiom is trivially defined;
unknown. However, recent findings indicate that the fac- Gödel, 1992.) Gödel’s proofs concern the limitations of
tor structure of PTSD, including the number of factors mathematical axioms for proving all mathematical results,
and symptoms that comprise them, is population depen- and as a result, the relationship between his proofs and
dent (Shevlin & Elklit, 2012). Furthermore, a number of DSM diagnoses may not be apparent. However, the DSM
these factor analytic studies have shown that the factors is a mathematical model of mental illness that utilizes a
are highly correlated, often at or exceeding a correlation series of logical statements to define discrete disorders,
of .90, bringing their uniqueness as separate diagnostic and as such, it suffers from the limitations of the model-
criteria into question. Taken together, these findings may ing approach it uses. A complete set of rules that define
indicate that the theoretically driven DSM model for mental illness a priori is not achievable according to
PTSD—one that conceptualizes separate symptom crite- Gödel’s proofs. It is important to stress that attempts to
ria representing different aspects of the disorder—may identify a set of rules a priori can lead to high levels of
not best fit the data. Additionally, defining a specific complexity without getting closer to the goal of achiev-
structure for PTSD on the basis of factor analyses may ing a consistently applicable definition.
introduce measurement error, as there is evidence that it For example, imagine we wanted to develop an algo-
is not generalizable. rithm that differentiates baseball fans from all other peo-
A second limitation of this approach is that it does not ple. We could do this easily by giving a trivial definition
provide information about the relationship between (i.e., baseball fans are those who say they are fans of
symptoms and a dependent outcome, such as level of baseball). If we wanted to be able to pick them out solely
functioning, well-being, or distress. Factor analysis can on the basis of a set of features without people telling us
provide information only about the underlying relation- whether they are fans, we could make a set of rules in
ship between the variables that are in the model. Without the same manner as the DSM, such that baseball fans are
prior information from population-based research that those who attend games. However, two types of people
has identified symptoms associated with negative out- attend games: those who are fans and those who are not.
comes, factor analysis is limited because it may retain There are an infinite number of reasons nonfans would
features that are strongly related to one another but have attend games (e.g., they are going with a friend).
limited explanatory or predictive validity as related to Ultimately, the rule is ineffective for defining fans because
clinically meaningful outcomes. As discussed above, that sometimes it will be accurate and other times it will not
kind of data was missing when symptoms were selected. be. If more rules are added, our definition of a fan could
Further, features that are strongly related to clinically be more exact, but the same problem would persist. For
meaningful outcomes but not to other features may be example, we can make a second rule that a baseball fan
excluded in a factor analysis if the goal is to identify a is defined as one who attends baseball games and
finite number of factors and the items that fall into those watches games on television. This will capture more true
factors. fans but leaves out people who either attend games or
watch them on television—any number of which could
Hidden Heterogeneity in the DSM and also be true fans. Conversely, we could state that a fan is
someone who watches games on television or goes to
PTSD games. However, the same problem persists that although
Factor analysis has been used to refine the symptom cri- we are capturing true fans and true nonfans, we are also
teria. However, defining the presence or absence of a defining people as fans who are not and defining people
mental illness on the basis of symptoms above a cutoff as nonfans who are. Further, the definition of a fan has
on multiple criteria remains problematic regardless of become more diffuse as “fan” could be indicative of dif-
how the criteria are identified or validated. Defining a ferent behaviors.
PTSD Heterogeneity 655

In the case of many DSM diagnoses, logical “and/or” where N is the number of symptoms per symptom cluster
rules are often used in an attempt to create a complete and K represents all possible number of symptoms
definition of the disorder. Continuing with the baseball needed to satisfy or exceed the diagnosis.
fan analogy, if we decided our definition was too narrow, There are different ways this can be calculated. We
we could expand it by using such “and/or” rules. will provide an example where we are drawing all com-
However, the same fundamental problem remains that binations of three of four different colored balls (red,
our definitions capture some fans and some nonfans, but green, blue, yellow) from a bag to determine how many
they also identify some nonfans as fans and some fans as different color combinations are possible. Because we
nonfans. “Or” rules are common in the DSM. When such want to know how many different combinations of three
rules are used, the definition becomes less exact and of the four balls there were, we would draw three, then
more heterogeneous. Being a fan encompasses an put them back and draw again. We would keep drawing
increasingly heterogeneous set of descriptors as more until we had pulled all combinations of three of four col-
“or” rules are applied. ored balls. This is known as drawing with replacement.
This can be demonstrated mathematically using psy- There are four combinations of three out of four balls
chiatric diagnoses, many of which are structured in the (red, green, blue; red green, yellow; red, blue, yellow;
same manner as our definition of baseball fans. When green, blue, yellow). The way we identify the possible
thinking about psychiatric diagnosis, these issues have combinations of colored balls is the same way we can
real-life implications for research and treatment. Just as identify the possible combinations of symptoms in a cri-
true fans may be misclassified by our rules because they terion. For example, one individual with PTSD could
do not display the right combination of characteristics, so present with difficulty concentrating and hypervigilance,
too do DSM diagnoses leave out individuals because they whereas another individual could present with difficulty
lack the “correct” combination of symptoms or include concentrating and exaggerated startle response. As such,
people for whom such a diagnosis is inappropriate. there are multiple combinations that involve the same
Further, as the definition becomes more heterogeneous, symptoms, and the DSM follows drawing with replace-
so does the population that it defines. In the case of fans, ment rules. It is important to note that the DSM requires
using the rules we described comes with the assumption at least a specified number of symptoms. If we stated that
that those who go to games are the same population as we want to know all combinations of at least three of
those who watch games on television. Similarly, with four colored balls, we would also include a fifth combi-
DSM diagnoses, one must assume that individuals with nation (red, green, blue, yellow). As such, there are five
different symptom presentations have the same underly- combinations of at least three of four colored balls.
ing mental illness. Diagnostic criteria that are conditional for the diagno-
sis to be reached, such as PTSD Criterion A (exposure to
a traumatic event), do not factor into the above equation
Analysis of Heterogeneity in because they represent conditions under which the diag-
Psychiatric Diagnosis nosis is applicable and do not add diversity to the diag-
Calculating heterogeneity nosis. Finally, if a diagnosis has multiple symptom criteria
that must be met, the n choose k equation would be
To elucidate the heterogeneity within psychiatric diagno- calculated for each criterion, and the number of combi-
ses, we used an n choose k binomial equation with nations for each criterion are multiplied by each other.
replacement (Equation 1). This equation provides a For example, to calculate all possible combinations of
method for identifying the number of possible combina- PTSD presentations, the n choose k equation is applied
tions of a set of objects. The equation is calculated as to each criterion (reexperiencing, avoidance, arousal),
follows: n!, which is the total number of possible symp- and then the number of symptom combinations from
toms (n) multiplied by n − 1, n − 2, . . . until 1 is reached. each criterion are multiplied.
For example, if I had four objects, I would multiply 4 * 3
* 2 * 1. This value is divided by k!, multiplied by (n − k)!.
If we wanted to know how many ways we can pick three Diagnostic permutations in various
out of our four objects, n = 4 and k = 3. As such, we disorders
would calculate: 4! / 3! * (4 − 3)!
We applied the above calculation to PTSD, social phobia,
specific phobia, obsessive–compulsive disorder, panic dis-
  n  n  order, and major depressive episode in terms of the DSM–
∏n =i  ∑    , where   = n !/ k ! ( n − k ) ! (1)
k
   k  III–R, DSM–IV, and DSM–5 criteria, where applicable.
656 Galatzer-Levy, Bryant

Multiple disorders were examined because diagnoses PTSD.  For the DSM–III–R, we found that the number of
vary in the number of criteria and symptoms. By examin- PTSD combinations is equal to the product of 15 possible
ing these disorders side by side and over time, it becomes combinations of reexperiencing symptoms, 99 possible
clearer how heterogeneity increases as a function of combinations of avoidance–numbing symptoms, and 57
symptoms and criteria. (For full symptom criteria for possible combinations of hyperarousal symptoms, result-
PTSD and other disorders from the DSM–III to the DSM– ing in 84,645 presentations. For the DSM–IV, we found
5, see American Psychiatric Association, 1980, 1987, 1994, that the product of 31 possible combinations of intrusion
2013.) For clarity, the criteria for each diagnosis from symptoms, 99 possible combinations of avoidance–
DSM–III–R to the revisions in DSM–5 are presented in numbing symptoms, and 26 possible combinations of
Table 1. Though this exercise could be conducted with hyperarousal symptoms resulted in a total of 79,794 pre-
any diagnoses, we chose Axis 1 anxiety and mood disor- sentations. For the DSM–5, we found that the product of
ders, as these may share some common underlying 31 possible combinations of intrusion symptoms, 3 pos-
causes with PTSD (Cox, Clara, & Enns, 2002). The same sible combinations of avoidance symptoms, 120 possible
binomial equation without replacement was also con- combinations of cognitive–mood symptoms, and 57 pos-
ducted to demonstrate the number of ways to meet sible combinations of hyperarousal symptoms produces
requirements for each diagnosis. Finally, all ways to have 636,120 possible presentations.
symptoms in each criterion without meeting the require-
ments were calculated. This provides information about Major depressive episode. Using the same binomial
the number of presentations that are left out of the diag- equation to analyze major depressive episode, we found
nosis, because this too is affected as the diagnosis that the combination of depressed mood along with all
changes (see Table 2). combinations of four or more of the remaining symptoms,

Table 1.  Diagnostic Criteria From DSM–III–R, DSM–IV, and DSM–5 for PTSD and Major Depressive Episode and Select Anxiety
Disorders

Disorder DSM–III–R DSM–IV DSM–5

PTSD 2 conditional criteria 4 conditional criteria 4 conditional criteria
1 of 4 reexperiencing symptoms 1 of 5 reexperiencing symptoms 1 of 5 reexperiencing symptoms
3 of 7 avoidance/numbing 3 of 7 avoidance/numbing 1 of 2 avoidance symptoms
symptoms symptoms 2 of 7 negative alterations in
2 of 6 hyperarousal symptoms 2 of 5 hyperarousal symptoms cognition symptoms
2 of 6 hyperarousal symptoms

MDE 4 conditional criteria Unchanged from prior edition Unchanged from prior edition
5 of 9 symptoms, with at least 1
being depressed mood or loss
of interest or pleasure

Specific phobia 6 conditional criteria 7 conditional criteria Unchanged from prior edition

Social phobia 7 conditional criteria 8 conditional criteria 10 conditional criteria

OCD 1 conditional criteria 4 conditional criteria 3 conditional criteria

4 of 4 obsession symptoms 4 of 4 obsession symptoms 2 of 2 obsession symptoms
and/or and/or and/or
3 of 3 compulsion symptoms 2 of 2 compulsion symptoms 2 of 2 compulsion symptoms

Panic 4 conditional criteria 3 conditional criteria 2 conditional criteria

Panic attack criteria defined as 4 Panic attack criteria defined as Panic attack criteria defined as
of 13 symptoms 4 of 13 symptoms, with 1 of 3 4 of 13 symptoms, with 1 of 2
accompanying symptoms accompanying symptoms

Note: Conditional criteria refer to diagnosis specific criteria that are invariant and must be endorsed to meet the overall criteria. For example, Cri-
terion E for PTSD, which requires that the duration of the symptoms last more than 1 month, is a conditional criterion because it does not present
with variations for diagnosis, and if it is not met, the diagnosis cannot be assigned. PTSD = posttraumatic stress disorder; DSM–III–R = Diagnostic
and Statistical Manual of Mental Disorders (3rd edition, revised); DSM–IV = DSM (4th edition); DSM–5 = DSM (5th edition); MDE = major depres-
sive episode; OCD = obsessive–compulsive disorder.
PTSD Heterogeneity 657

Table 2.  Number of Heterogeneous Symptom Combinations to Meet or Not

Meet DSM Criteria for Six Diagnoses

Disorder DSM–III–R DSM–IV DSM–5

Posttraumatic stress disorder  
  Possible combinations 84,645 79,794 636,120
  Minimum combinations 2,100 1,750 3,150
  Excluded presentations 35,370 42,253 107,973
Major depressive episode  
  Possible combinations 227 227 227
  Minimum combinations 126 126 126
  Excluded presentations 154 154 154
Specific phobia  
  Possible combinations 1 1 1
  Minimum combinations 1 1 1
  Excluded presentations 0 0 0
Social phobia  
  Possible combinations 1 1 1
  Minimum combinations 1 1 1
  Excluded presentations 0 0 0
Obsessive–compulsive disorder  
  Possible combinations 3 3 3
  Minimum combinations 2 2 2
  Excluded presentations 0 0 0
Panic disorder  
  Possible combinations 7,814 54,698 23,442
  Minimum combinations 715 715 715
  Excluded presentations 377 377 377

Note: This table presents (a) possible combinations, defined as all possible combina-
tions of symptoms that qualify for the specified diagnosis; (b) minimum combinations,
defined as all possible combinations of symptoms that result in the minimum number
of symptoms necessary to qualify for each specified diagnosis; and (c) excluded pre-
sentations, defined as all combinations of one or more symptoms in all categories that
do not meet diagnostic criteria. The number of combinations was calculated using a
binomial n choose k without replacement equation.

excluding loss of interest, produces 64 possible combina- means to have the disorder). Finally, we observed that
tions, as does the presence of symptom loss of interest panic disorder increased dramatically from the DSM–III–
excluding depressed mood. If both symptoms are present, R, with 7,814 presentations; to 54,698 in the DSM–IV with
all possible combinations of three or more of the remain- the addition of a separate criterion; and then reduced
ing seven symptoms produce 99 combinations. The sum dramatically to 23,442 presentations when the number of
of these combinations equals 227. symptoms in one criterion was reduced in the DSM–5.

Select anxiety disorders.  Social and specific phobias, Consequences of the DSM
which present with only one criterion and symptom, cre- measurement scheme: PTSD findings
ate a fully identified diagnosis. Though obsessive–com-
pulsive disorder changes in the total number of symptoms,
are inconsistent and they miss cases
the diagnosis consistently comes with only three permu- The consequences of the ignored heterogeneity in DSM
tations, as the presence of all obsessive symptoms, the diagnoses are unclear. PTSD presents with a large amount
presence of all compulsive symptoms, or the presence of of heterogeneity. Although we cannot determine whether
both represents only three possible combinations of findings in the PTSD literature have been impacted by
symptoms. These three disorders present with very little this heterogeneity, we can examine the consistency of
or no heterogeneity because they are trivially defined findings and hypothesize that inconsistency may be the
(i.e., there is a specific complete description of what it result of heterogeneity in the diagnosis.
658 Galatzer-Levy, Bryant

A number of meta-analyses have been conducted that funding research based on DSM diagnoses because of
have identified consistent social, demographic, trauma- their primary focus on reliability at the expense of validity
related, and biological correlates of PTSD, as well as leading to nonspecific findings. NIMH will refocus its
treatments that are effective for PTSD, but these same funding efforts around research seeking to understand
analyses have shown that the effect sizes of many of psychopathology based on alterations in cognition, emo-
these variables vary dramatically. In one large meta-anal- tion, and behavior and the underlying biology and neuro-
ysis of risk factors for PTSD, 11 out of the 14 common circuitry of these domains (Insel, 2013). The heterogeneity
predictors, including socioeconomic status, age, race, found in the PTSD diagnosis exemplifies this problem of
previous trauma, trauma severity, social support, life primarily using diagnosis as an outcome.
stress, and intelligence (among others) displayed signifi-
cant heterogeneity in their effect size; for example, Potential Solutions and Future
trauma severity accounts for between 0.02% to 58% of
the variance (Brewin, Andrews, & Valentine, 2000).
Directions
Similarly, though no heterogeneity statistic was offered, a Although the DSM approach presents with many limita-
second meta-analysis that examined seven pretrauma tions for identifying populations from a measurement
and peritraumatic (occurring at the time of the trauma) perspective, the field remains interested in identifying
characteristics demonstrated a wide spread in effect sizes and describing normal and pathological populations. The
across all seven predictors. For example, the strongest DSM diagnostic scheme presents with untenable limita-
predictors of PTSD, peritraumatic emotions and peritrau- tions. The DSM defines clinical outcomes a priori. This
matic dissociation, demonstrated effect sizes that ranged system often relies on complex sets of rules in an attempt
from .15 to .55 and from .14 to .94, respectively (Ozer, to capture the correct population. However, an unin-
Best, Lipsey, & Weiss, 2003). Similarly, significant levels of tended consequence of this approach is that the diagno-
heterogeneity are observed in the effect sizes of treat- sis encompasses increasingly heterogeneous symptom
ment for PTSD (Bisson, 2007; Bradley, Greene, Russ, presentations without addressing the limitations of a pri-
Dutra, & Westen, 2005; Sherman, 1998) and structural ori definitions. Conversely, other traditionally commonly
brain abnormalities associated with PTSD (Karl et al., used approaches also have limitations.
2006). The reason for these inconsistencies in findings is Typically, researchers examine population means and
not known. However, as all of the above studies depend infer their clinical meaning on the basis of clinical norms
on measurement precision within the PTSD diagnosis, of the measure. If means are charted over time, research-
measurement error in the PTSD diagnosis is worth exam- ers can make inferences about the course of the disorder
ination as a potential culprit. or the response to a treatment or life event. However, this
Sampling and measurement error have also been approach is unsatisfactory for the reasons that diagnosis
noted as significant problems with the PTSD diagnosis, is appealing. Diagnosis hopes to provide information
given that subthreshold presentations of PTSD are often about who is doing poorly, and in what ways, and who
associated with impairment in psychosocial and occupa- is doing well, whereas means can reveal only how every-
tional functioning in the clinical range (Marshall et al., one is doing together.
2001), and help-seeking behavior among such individu- A second common method for characterizing out-
als is similar to those who meet full requirements for the comes, similar to that used in psychiatric diagnosis, is to
diagnosis (Stein, Walker, Hazen, & Forde, 1997). These dichotomize a continuous variable to create two popula-
findings indicate that the combinatorial experiment pro- tions on the basis of a cutoff point. This method would
vided above does not simply represent an academic reveal heterogeneity and can be used to track the course
exercise but in fact has real-life consequences as the as one can observe whether individuals change catego-
diagnosis appears to produce false negatives, missing ries over time. However, dichotomizing variables has
cases in need of intervention. long been discouraged by methodologists because of
The concern with DSM diagnoses is well understood, evidence that the practice weakens statistical power,
and changes are pending. The National Institute of Mental increases measurement error, and can lead to falsely sig-
Health (NIMH) recently shifted its focus to the Research nificant results (Cohen, 1983), while also artificially sepa-
Domains Criteria project, which adopts a transdiagnostic rating individuals into two groups when these groups are
approach and focuses on common mechanisms that not naturally occurring (MacCallum, Zhang, Preacher, &
may underlie disorders (Insel et al., 2010). This focus Rucker, 2002). Despite these long-standing criticisms,
reflects the concern in the field of the potentially limiting researchers continue to follow this practice because it
preoccupation with diagnostic categories in conducting allows for ease of analysis and aids in interpretation
psychological and biological research into psychiatric (DeCoster, Iselin, & Gallucci, 2009). New statistical
conditions. Specifically, NIMH is moving away from approaches are warranted and necessary.
PTSD Heterogeneity 659

A number of new methods have emerged that capital- only a few transient stress symptoms. Some do poorly
ize on computational advances to empirically identify initially but get better over time, and some suffer chroni-
common outcomes. These approaches were out of reach cally with consistently elevated symptom levels. From a
when the DSM scheme was first conceived and may pro- research perspective, these distinct outcomes may be
vide some viable alternatives for studying outcome het- influenced and maintained by different factors and may
erogeneity in both clinical and nonclinical contexts. We require different forms of remediation. Latent growth
will discuss one method that appears promising, though mixture modeling and its extensions are relevant to apply
this description is not meant to be exhaustive. in any context where it is not parsimonious to assume
Relatively new data analytic approaches have recently one common trajectory across the population (Muthen,
been used to empirically identify latent (or not directly 2004), as they allow for the identification of meaningful
observable) subpopulations, by identifying finite latent subpopulations without imposing a set of a priori
mixture distributions that underlie the observed distribu- rules. Further, they account for distributions and mea-
tion. This approach can empirically identify distinct pop- surement error within the identified subpopulations
ulations on the basis of the distribution of their symptoms (Muthen, 2004). Using such approaches may significantly
(or any other outcome measure of interest). It is flexible, reduce measurement error, as clinical outcomes are
as it can be applied to cross-sectional as well as longitu- empirically defined, sidestepping many of the measure-
dinal data (Muthen, 2000). ment limitations of DSM diagnoses. Further, researchers
An increasing number of studies have utilized cross- can examine change in any domain that may be affected
sectional latent class analysis to identify distinct popula- by significant life events, such as cognitive ability, world-
tions where it is theoretically relevant to compare view, emotional valence, or any other theoretically rele-
populations on the basis of mental health status. For vant psychological domain. This can provide a broader
example, researchers have begun to identify latent sub- picture of both adaptive and maladaptive responses to
populations clustered by symptoms and have used those life stressors and may result in the identification of novel
empirically identified subpopulations as outcome vari- treatment targets that are aimed at domains other than
ables rather than using DSM diagnostic status (Breslau, symptoms.
Reboussin, Anthony, & Storr, 2005; Steenkamp et al.,
2012). This approach has also been used to empirically
identify subtypes of PTSD that are distinct in their symp-
Relevance beyond clinical research
tom presentation from others with the same diagnostic The example provided by PTSD is also relevant beyond
status (Wolf et al., 2012). Latent class analysis allows the the scope of clinical research, as it highlights generally
investigator to examine multiple features simultaneously important issues in the measurement of behavioral
to understand how they commonly cluster together into change in response to stressful life events. A number of
latent subpopulations. Change in these clusters over time studies have identified common patterns of adaptation to
can be examined using methods such as latent transition common stressful life events, examining the course of
analysis (Muthen, 2000). subjective well-being in response to marriage, divorce,
Change over two time points provides limited informa- and widowhood (Mancini, Bonanno, & Clark, 2009),
tion. Mixture distributions can be extended longitudinally unemployment (Galatzer-Levy, Bonanno, & Mancini,
to identify common populations characterized by their 2010), and becoming a parent (Galatzer-Levy, Mazursky,
latent symptom course over multiple occasions. A number Mancini, & Bonanno, 2011); or focusing on general dis-
of recent studies have used approaches such as latent tress during college (Galatzer-Levy & Bonanno, 2013;
growth mixture modeling and its extensions to identify Galatzer-Levy et al., 2012). Some of these studies have
distinct subpopulations according to longitudinal course challenged long-standing findings by identifying errors in
following potentially traumatic events. This has been estimation on the basis of the use of a single mean trajec-
shown to reveal clinically meaningful patterns in multiple tory rather than identifying heterogeneous populations.
domains, including PTSD symptom severity (Berntsen et For example, a long-standing body of literature has dem-
al., 2012; Bonanno, Mancini, et al., 2012; deRoon-Cassini, onstrated that individuals drop in their level of subjective
2010; Galatzer-Levy, Madan, Neylan, Henn-Haase, & well-being when they have children and do not recover
Marmar, 2011), anxiety and depression symptom severity until their children leave home (Twenge, Campbell, &
(Bonanno, Kennedy, Galatzer-Levy, Lude, & Elfström, Foster, 2004). However, when patterns of subjective well-
2012; Galatzer-Levy & Bonanno, 2012), and general dis- being are examined using a latent growth mixture mod-
tress that is not diagnosis specific (Galatzer-Levy et al., eling approach, we find that only a small proportion dip
2013; Galatzer-Levy, Burton, & Bonanno, 2012). significantly, pulling down the population average. In
These studies indicate that there is meaningful hetero- fact, many new parents increase significantly in their sub-
geneity in how people respond to significant life stress- jective well-being when they become parents (Galatzer-
ors and potentially traumatic events. Most cope well with Levy, Mazursky, et al., 2011).
660 Galatzer-Levy, Bryant

Increasing communication across alternative to single mean trajectory approaches that

subdisciplines assume homogeneity even when this is not a parsimoni-
ous assumption.
The above findings highlight the similarities in the course Such an empirical approach for identifying behavioral
of the stress responses when measuring outcomes such patterns both in clinical and nonclinical contexts is
as subjective well-being or symptoms. An added benefit nascent. A great deal of work is necessary to identify and
of new approaches that do not rely on diagnoses is that understand common outcomes of disparate, potentially
they can potentially increase the translatability of find- traumatic, and common stressful life events. However,
ings across multiple subdisciplines of psychology. the lesson learned from the example of PTSD is that
Although few outside of clinical psychology and psychia- empirical findings are only as strong as the clarity of the
try utilize psychiatric diagnoses as an outcome, many constructs under study. If the construct is noisy, diffuse,
fields study behavioral changes in response to common or lacking in validity, it becomes increasingly difficult to
and atypical life stressors, such as how personality study the phenomenon.
changes in response to anxiety-provoking life events Finally, new approaches hold promise for identifying
(Bolger, 1990); how significant others aid in coping with common outcomes that are heuristically meaningful both
life-threatening events, such as cancer onset (Bolger, in clinical and nonclinical contexts. One set of approaches
Foster, Vinokur, & Ng, 1996); or how contextual threat that hold promise is the empirical identification of latent
cues are encoded into memory, leading to behavior mixture distributions both cross-sectionally and longitu-
change, in animal models (LeDoux, 2012; Phillips & dinally. This approach allows for the empirical identifica-
LeDoux, 1992). It is difficult and often cumbersome to tion of populations rather than imposing rigid external
relate these findings to diagnostic entities despite the constraints based on a priori definitions. It also has the
importance of these bodies of work for understanding added benefit of increasing the translatability of findings
clinical outcomes. However, empirical approaches for across disciplines of psychology, because many subdisci-
identifying meaningful subpopulations may be appealing plines study behavioral responses to life stressors, though
to researchers across disciplines who examine behavioral few outside of clinical fields use psychiatric status as an
change. Further, clinical psychology becomes less alien- outcome. The value of these techniques and others will
ated from other subdisciplines that are studying similar be determined by their empirical utility. Regardless, new
phenomena by examining common outcomes, such as approaches that examine the heterogeneity in stress
general distress in clinical populations, rather than rely- response behavior rather than ignoring it will doubtless
ing on diagnoses that are rarely used in other research lead to stronger and clearer findings by virtue of reduc-
contexts. ing measurement error and by pursuing both reliability
and validity.
Conclusion Declaration of Conflicting Interests
This article endeavored to demonstrate the limitations of The authors declared that they had no conflicts of interest with
DSM diagnoses from a measurement perspective. We pri- respect to their authorship or the publication of this article.
marily focused on the PTSD diagnosis because it pro-
vides an example of the many factors that influenced the References
DSM to create a broad and error-prone diagnostic scheme. American Psychiatric Association. (1980). Diagnostic and sta-
We discussed how the desire for a diagnosis to encom- tistical manual of mental disorders (3rd ed.). Washington,
pass the broad spectrum of posttraumatic presentations DC: Author.
led to a diagnosis in DSM–5 that was complex, even com- American Psychiatric Association. (1987). Diagnostic and
pared with other DSM disorders. However, we further statistical manual of mental disorders (3rd ed., rev.).
discussed that these limitations may be secondary to the Washington, DC: Author.
primary measurement limitation that results from any American Psychiatric Association. (1994). Diagnostic and sta-
attempt to create an algorithm based on a priori rules. As tistical manual of mental disorders (4th ed.). Washington,
such, the modeling approach of the DSM necessarily gen- DC: Author.
erates both false positives and false negatives. Finally, we American Psychiatric Association. (2013). Diagnostic and sta-
tistical manual of mental disorders (5th ed.). Washington,
provided some possible solutions focusing on recent
DC: Author.
research that identifies clinical outcomes empirically Asmundson, G. J., Frombach, I., McQuaid, J., Pedrelli, P., Lenox,
rather than a priori and demonstrated how this approach R., & Stein, M. B. (2000). Dimensionality of posttraumatic
may ultimately provide stronger information about pre- stress symptoms: A confirmatory factor analysis of DSM-IV
dictors, correlates, and, when relevant, effective treat- symptom clusters and other symptom models. Behaviour
ments related to these outcomes. It also provides an Research and Therapy, 38, 203–214.
PTSD Heterogeneity 661

Berntsen, D., Johannessen, K. B., Thomsen, Y. D., Bertelsen, for dichotomization. Psychological Methods, 14, 349–366.
M., Hoyle, R. H., & Rubin, D. C. (2012). Peace and war tra- doi:10.1037/a0016956
jectories of posttraumatic stress disorder symptoms before, de Jongh, A., Olff, M., van Hoolwerff, H., Aartman, I. H. A.,
during, and after military deployment in Afghanistan. Broekman, B., Lindauer, R., & Boer, F. (2008). Anxiety and
Psychological Science, 23, 1557–1565. post-traumatic stress symptoms following wisdom tooth
Bisson, J. I. (2007). Psychological treatments for chronic post- removal. Behaviour Research and Therapy, 46, 1305–1310.
traumatic stress disorder systematic review and meta- doi:10.1016/j.brat.2008.09.004
analysis. British Journal of Psychiatry, 190, 97–104. deRoon-Cassini, T. A. (2010). Psychopathology and resilience
Bolger, N. (1990). Coping as a personality process: A prospec- following traumatic injury: A latent growth mixture model
tive study. Journal of Personality and Social Psychology, 59, analysis. Rehabilitation Psychology, 55, 1–11. doi:10.1037/
525–537. a0018601
Bolger, N., Foster, M., Vinokur, A. D., & Ng, R. (1996). Close Galatzer-Levy, I. R., & Bonanno, G. A. (2012). Beyond nor-
relationships and adjustments to a life crisis: The case of mality in the study of bereavement: Heterogeneity in
breast cancer. Journal of Personality and Social Psychology, depression outcomes following loss in older adults. Social
70, 283–294. Science & Medicine, 74, 1987–1994. doi:10.1016/j.socs-
Bonanno, G. A., Galea, S., Bucciarelli, A., & Vlahov, D. (2006). cimed.2012.02.022
Psychological resilience after disaster: New York City in the Galatzer-Levy, I. R., & Bonanno, G. A. (2013). Heterogeneous
aftermath of the September 11th terrorist attack [Research patterns of stress over the four years of college: Associations
Support, N.I.H., Extramural]. Psychological Science, 17, with anxious attachment and ego-resiliency. Journal of
181–186. Personality, 81, 476–486.
Bonanno, G. A., Kennedy, P., Galatzer-Levy, I. R., Lude, P., & Galatzer-Levy, I. R., Bonanno, G. A., & Mancini, A. D. (2010).
Elfström, M. L. (2012). Trajectories of resilience, depression, From marianthal to latent growth mixture modeling: A return
and anxiety following spinal cord injury. Rehabilitation to the exploration of individual differences in response to
Psychology, 57, 236–247. unemployment. Journal of Neuroscience, Psychology, and
Bonanno, G. A., Mancini, A. D., Horton, J. L., Powell, T. M., Economics, 3, 116–125.
Leardmann, C. A., Boyko, E. J., . . . Millennium Cohort Galatzer-Levy, I. R., Brown, A. D., Henn-Haase, C., Metzler, T.
Study Team. (2012). Trajectories of trauma symptoms J., Neylan, T. C., & Marmar, C. R. (2013). Positive and nega-
and resilience in deployed U.S. military service members: tive emotion prospectively predict trajectories of resilience
Prospective cohort study. British Journal of Psychiatry: The and distress among high-exposure police officers. Emotion,
Journal of Mental Science, 200, 317–323. 13, 545–553.
Bradley, R., Greene, J., Russ, E., Dutra, L., & Westen, D. (2005). Galatzer-Levy, I. R., Burton, C. L., & Bonanno, G. A. (2012).
A multidimensional meta-analysis of psychotherapy for Coping flexibility, potentially traumatic life events, and
PTSD. American Journal of Psychiatry, 162, 214–227. resilience: A prospective study of college student adjust-
Breslau, N., Davis, G. C., Andreski, P., & Peterson, E. (1991). ment. Journal of Social and Clinical Psychology, 31, 542–
Traumatic events and posttraumatic stress disorder in an 567. doi:10.1521/jscp.2012.31.6.542
urban population of young adults. Archives of General Galatzer-Levy, I. R., & Galatzer-Levy, R. M. (2007). The revolu-
Psychiatry, 48, 216–222. tion in psychiatric diagnosis: Problems at the foundations.
Breslau, N., Reboussin, B. A., Anthony, J. C., & Storr, C. L. Perspectives in Biology and Medicine, 50, 161–180.
(2005). The structure of posttraumatic stress disorder: Galatzer-Levy, I. R., Madan, A., Neylan, T. C., Henn-Haase,
Latent class analysis in 2 community samples. Archives of C., & Marmar, C. R. (2011). Peritraumatic and trait dis-
General Psychiatry, 62, 1343–1351. sociation differentiate police officers with resilient versus
Brewin, C. R., Andrews, B., & Valentine, J. D. (2000). Meta- symptomatic trajectories of posttraumatic stress symptoms.
analysis of risk factors for posttraumatic stress disorder in Journal of Traumatic Stress, 24, 557–565. doi:10.1002/
trauma-exposed adults. Journal of Consulting and Clinical jts.20684
Psychology, 68, 748–766. Galatzer-Levy, I. R., Mazursky, H., Mancini, A. D., & Bonanno,
Buckley, T. C., Blanchard, E. B., & Hickling, E. J. (1998). A G. A. (2011). What we don’t expect when expecting:
confirmatory factor analysis of posttraumatic stress symp- Evidence for heterogeneity in subjective well-being in
toms. Behaviour Research and Therapy, 36, 1091–1099. response to parenthood. Journal of Family Psychology, 25,
doi:10.1016/S0005-7967(98)00076-X 384–392.
Centers for Disease Control Vietnam Experience Study Group. Gersons, B. P., & Carlier, I. V. (1992). Post-traumatic stress dis-
(1988). Health status of Vietnam veterans: Psychosocial order: The history of a recent concept. British Journal of
characteristics. Journal of the American Medical Association, Psychiatry, 161, 742–748.
259, 2701–2707. Gödel, K. (1992). On formally undecidable propositions
Cohen, J. (1983). The cost of dichotomization. Applied of Principia Mathematica and related systems. North
Psychological Measurement, 7, 249–253. Chemsford, MA: Courier Dover.
Cox, B. J., Clara, I. P., & Enns, M. W. (2002). Posttraumatic Helzer, J. E., Robins, L. N., & McEvoy, L. (1987). Post-traumatic
stress disorder and the structure of common mental disor- stress disorder in the general population: Findings of the
ders. Depression and Anxiety, 15, 168–171. Epidemiologic Catchment Area Survey. The New England
DeCoster, J., Iselin, A.-M. R., & Gallucci, M. (2009). A Journal of Medicine, 317, 1630–1634. doi:10.1056/nejm
conceptual and empirical examination of justifications 198712243172604
662 Galatzer-Levy, Bryant

Insel, T. (2013). Transforming diagnosis. Retrieved from http:// Muthen, B. (Ed.). (2004). Latent variable analysis: Growth mix-
www.nimh.nih.gov/about/director/2013/transforming- ture modeling and related techniques for longitudinal data.
diagnosis.shtml Newbury Park, CA: Sage.
Insel, T., Cuthbert, B., Garvey, M., Heinssen, R., Pine, D. S., Olde, E., van der Hart, O., Kleber, R. J., & van Son, M. J. M.
Quinn, K., . . . Wang, P. (2010). Research domain criteria (2006). Posttraumatic stress following childbirth: A review.
(RDoC): Toward a new classification framework for research Clinical Psychology Review, 26, 1–16.
on mental disorders. American Journal of Psychiatry, 167, Ozer, E. J., Best, S. R., Lipsey, T. L., & Weiss, D. S. (2003).
748–751. doi:10.1176/appi.ajp.2010.09091379 Predictors of posttraumatic stress disorder and symptoms
Karl, A., Schaefer, M., Malta, L. S., Dörfel, D., Rohleder, N., in adults: A meta-analysis. Psychological Bulletin, 129,
& Werner, A. (2006). A meta-analysis of structural brain 52–73.
abnormalities in PTSD. Neuroscience & Biobehavioral Palmieri, P. A., & Fitzgerald, L. F. (2005). Confirmatory fac-
Reviews, 30, 1004–1031. doi:10.1016/j.neubiorev.2006.03 tor analysis of posttraumatic stress symptoms in sexually
.004 harassed women. Journal of Traumatic Stress, 18, 657–666.
King, D. W., Leskin, G. A., King, L. A., & Weathers, F. W. (1998). Phillips, R., & LeDoux, J. (1992). Differential contribution of
Confirmatory factor analysis of the clinician-administered amygdala and hippocampus to cued and contextual fear
PTSD Scale: Evidence for the dimensionality of posttrau- conditioning. Behavioral Neuroscience, 106, 274–285.
matic stress disorder. Psychological Assessment, 10, 90– Sherman, J. J. (1998). Effects of psychotherapeutic treatments
96. for PTSD: A meta-analysis of controlled clinical trials.
Kraemer, H. C. (2007). DSM categories and dimensions in clini- Journal of Traumatic Stress, 11, 413–435.
cal and research contexts. International Journal of Methods Shevlin, M., & Elklit, A. (2012). The latent structure of post-
in Psychiatric Research, 16(Suppl. 1), S8–S15. doi:10.1002/ traumatic stress disorder: Different models or different pop-
mpr.211 ulations? Journal of Abnormal Psychology, 121, 610–615.
Kulka, R. A., Schlenger, W. E., Fairbank, J. A., Hough, R. L., doi:10.1037/a0028591
Jordan, B. K., Marmar, C. R., & Weiss, D. S. (1990). Trauma Simms, L. J., Watson, D., & Doebbeling, B. N. (2002).
and the Vietnam war generation: Report of the findings Confirmatory factor analyses of posttraumatic stress symp-
from the National Vietnam Veterans Readjustment Study. toms in deployed and nondeployed veterans of the Gulf
New York, NY: Brunner/Mazel. War. Journal of Abnormal Psychology, 111, 637–647.
Larsson, G. (2005). Dimensional analysis of the Impact of Spitzer, R. L., Endicott, J., & Robins, E. (1975). Clinical criteria
Event Scale using structural equation modeling. Journal of for psychiatric diagnosis and DSM–III. American Journal of
Traumatic Stress, 13, 193–204. Psychiatry, 132, 1187–1192.
LeDoux, J. (2012). Rethinking the emotional brain. Neuron, 73, Spitzer, R. L., Endicott, J., & Robins, E. (1978). Research diag-
653–676. nostic criteria: Rationale and reliability. Archives of General
Lees-Haley, P. R., Price, J. R., Williams, C. W., & Betz, B. P. Psychiatry, 35, 773–782.
(2001). Use of the Impact of Events Scale in the assessment Steenkamp, M. M., Nickerson, A., Maguen, S., Dickstein, B. D.,
of emotional distress and PTSD may produce misleading Nash, W. P., & Litz, B. T. (2012). Latent classes of PTSD
results. Journal of Forensic Neuropsychology, 2, 45–52. symptoms in Vietnam veterans. Behavior Modification, 36,
MacCallum, R. C., Zhang, S., Preacher, K. J., & Rucker, D. D. 857–874.
(2002). On the practice of dichotomization of quantitative Stein, M. B., Walker, J. R., Hazen, A. L., & Forde, D. R. (1997).
variables. Psychological Methods, 7, 19–40. Full and partial posttraumatic stress disorder: Findings from
Mancini, A. D., Bonanno, G. A., & Clark, A. E. (2009). Stepping a community survey. American Journal of Psychiatry, 154,
off the hedonic treadmill: Latent class analyses of individual 1114–1119.
differences in response to major life events. Manuscript sub- Taylor, S., Kuch, K., Koch, W. J., Crockett, D. J., & Passey,
mitted for publication. G. (1998). The structure of posttraumatic stress symp-
Marshall, R. D., Olfson, M., Hellman, F., Blanco, C., Guardino, toms. Journal of Abnormal Psychology, 107, 154–160.
M., & Struening, E. L. (2001). Comorbidity, impair- doi:10.1037/0021-843x.107.1.154
ment, and suicidality in subthreshold PTSD. American Twenge, J. M., Campbell, W. K., & Foster, C. A. (2004).
Journal of Psychiatry, 158, 1467–1473. doi:10.1176/appi. Parenthood and marital satisfaction: A meta-analytic review.
ajp.158.9.1467 Journal of Marriage and Family, 65, 574–583.
McFarlane, A. C. (1988). The aetiology of post-traumatic stress Wilson, J. P. (1994). The historical evolution of PTSD diagnos-
disorders following a natural disaster. British Journal of tic criteria: From Freud to DSM–IV. Journal of Traumatic
Psychiatry, 152, 116–121. Stress, 7, 681–698.
McNally, R. J. (2009). Can we fix PTSD in DSM-V? Depression Wilson, M. (1993). DSM–III and the transformation of American
and Anxiety, 26, 597–600. doi:10.1002/da.20586 psychiatry: A history. American Journal of Psychiatry, 150,
Muthen, B. (2000). Integrating person-centered and variable- 399–410.
centered analyses: Growth mixture Modeling with latent Wolf, E. J., Miller, M. W., Reardon, A. F., Ryabchenko, K. A.,
trajectory classes. Alcoholism: Clinical and Experimental Castillo, D., & Freund, R. (2012). A latent class analysis of
Research, 24, 882–891. doi:10.1111/j.1530-0277.2000.tb0 dissociation and PTSD: Evidence for a dissociative subtype.
2070.x Archives of General Psychiatry, 69, 698–705.