CURRENT DIRECTIONS IN PSYCHOLOGICAL S CIENCE

Empirically Supported
Complexity
Rethinking Evidence-Based Practice in Psychotherapy
Drew Westen1,2 and Rebekah Bradley2
1
Department of Psychology, Emory University, and 2Department of Psychiatry and Behavioral Sciences, Emory University

ABSTRACT—Over the last 10 years, evidence-based practice Over the last 10 years, evidence-based practice has become
in psychology has become synonymous with a particular synonymous in psychology with a particular operationalization
operationalization of it aimed at developing a list of em- of it, the empirically supported therapies movement (defined by
pirically supported therapies. Although much has been an attempt to develop a list of empirically supported treatments
learned since the emergence of the empirically supported for specific disorders; Chambless & Ollendick, 2000). Psycho-
therapies movement, its restrictive definition of evidence therapy research (and hence the effort to base psychotherapy on
(excluding, for example, basic science as a source of evi- evidence) actually existed long before this movement. In 1977,
dence to be used by clinicians) is problematic, and the as- Smith and Glass demonstrated meta-analytically (by combining
sumptions inherent in its nearly exclusive focus on brief, data across hundreds of studies) that ‘‘generic’’ psychotherapy
focal treatments for specific disorders are themselves not yields a very large effect (in the average study, patients who had
generally supported by the available data. Recent meta- received psychotherapy fell in the 85th percentile of patients in
analytic data support a more nuanced view of treatment the control condition on measures of mental health). Subsequent
efficacy than one that makes dichotomous judgments of research has corroborated these findings (Wampold, 2001),
empirically supported or unsupported, suggesting the need documenting that most psychotherapies ‘‘work’’ for most pa-
for a more refined concept of evidence-based practice in tients. In this sense, the practice of generic psychotherapy has
psychology. been evidence based at least since 1977.
What distinguishes the empirically supported therapies
KEYWORDS—evidence based practice; empirically support-
movement from prior efforts to place psychotherapy on terra
ed therapy; multidimensional meta-analysis; psychothera-
firma are two interrelated features. First, to counter psychiatric
py; treatment research; exclusion criteria; generalizability;
dissemination practice guidelines that tended to understate the effects of
psychological treatments, the empirically supported therapies
movement adopted a U.S. Food and Drug Administration (FDA)
The last several years have seen a worldwide movement toward model in which treatments are classified as either supported (and
evidence-based practice in medicine (Sackett, Straus, Ri- hence to be used) or unsupported (and hence not to be used).
chardson, Rosenberg, & Haynes, 2000). It is difficult to imagine Evidence was thus limited, in this model, primarily to data ob-
how anyone could object to such a movement (or what an alter- tained in randomized controlled trials, which compare outcomes
native might be). What is too seldom appreciated, however, is between groups of patients who have been randomly assigned to
that evidence-based practice is a construct (i.e., an idea, ab- either active or control treatments.
straction, or theoretical entity) and thus must be operationalized Second, the empirically supported therapies movement fo-
(i.e., turned into some concrete form that comes to define it). The cused on brief, focal treatments for specific disorders (e.g., major
way it is operationalized is not incidental to whether its net depressive disorder) as defined by the Diagnostic and Statistical
effects turn out to be positive, negative, or mixed. Manual of Mental Disorders (DSM-IV; American Psychiatric
Association, 1994). Although researchers converged on the
study of brief treatments (rather than the longer-term treatments
that are the norm in clinical practice) for many reasons, brevity is
Address correspondence to Drew Westen, Departments of Psychology
and Psychiatry, Emory University, 532 Kilgo Circle, Atlanta, GA advantageous from a researcher’s perspective because of a
30322; e-mail: [email protected]. simple fact of experimental method as applied to psychotherapy:

266 Copyright r 2005 American Psychological Society Volume 14—Number 5

 Drew Westen and Rebekah Bradley

The longer the treatment, the more within-group variability; the ments researchers imagine a priori are best, finding that
more variability, the less one can draw precise causal conclu- they outperform interventions intended to be inert (which they
sions. Testing treatments of brief and fixed duration for patients inevitably will, if they are as effective as the generic psycho-
who meet diagnostic criteria for a specific disorder thus allows therapies studied by Smith & Glass, 1977), and declaring them
researchers to minimize within-group variability in both patient the treatment or treatments of choice. From a strictly statistical
characteristics and treatment delivered. The typical empirically perspective, it seems very unlikely that the one or two treatments
supported therapy for children and adolescents as well as adults per disorder that researchers bet their money on at the outset
lasts 8 to 10 weeks; we know of only one that exceeds 4 months’ would consistently outperform the interventions obtained by,
duration.1 say, the 20% of experienced, full-time, doctoral-level clinicians
The empirically supported therapies movement has been as- in the community who get the best results with their patients.
sociated with impressive advances in the treatment of several It might thus be worth taking the time to identify and study the
disorders, such as panic disorder, obsessive-compulsive disor- practices of such clinicians. Wampold (personal communica-
der, and posttraumatic stress disorder (PTSD). Yet a new body of tion, December, 2004) has recently found, in fact, that 60% of
evidence suggests that dichotomous, either–or judgments of therapists in an outpatient HMO sample actually obtain results
empirically supported versus unsupported may seriously mis- comparable to the results obtained in relevant randomized
represent the state of the science of psychotherapy (Westen & controlled trials.
Morrison, 2001; Westen, Novotny, & Thompson-Brenner, 2004, Second, researchers have routinely compared their treatments
2005). to no treatment (control groups of patients on waiting lists for
treatment); treatments provided by underskilled, underpaid,
A RESTRICTED VIEW OF EVIDENCE bachelors-level counselors with enormous caseloads; and other
intent-to-fail conditions (e.g., pseudotreatments designed spe-
Elsewhere we have described a number of problems with ope- cifically as control groups to prove the superiority of the inves-
rationalizing evidence-based practice as a list of approved tigator’s preferred treatment and that have no theoretical
(proven) treatments. Here we note three. rationale or are delivered by graduate students who know they
First, empirically supported therapies research arose as a are administering treatment that is not supposed to work). Un-
psychotherapeutic analog to drug trials and thus is implicitly fortunately, such control conditions do not control for any of the
predicated on a pharmaceutical model: Researchers in the most important confounds that threaten causal inference—
laboratory create new psychological ‘‘compounds,’’ compare namely, the factors common to most therapies (i.e., talking with a
them to placebos, and then ‘‘disseminate’’ them to clinicians. presumed expert who offers hope, support, empathy, a coherent
The problem is that no one knows whether the psychological theoretical rationale, advice, etc.) that we know produce large
compounds researchers have assembled, based on the same effects. We know of no published study in the last decade pur-
kinds of intuitive hunches that Meehl (1954) warned against, are ported to demonstrate specific efficacy of a treatment that has
superior to the home-brewed compounds assembled by clini- ruled out the most parsimonious rival hypothesis: that something
cians in the community, particularly those who get the best re- intended to be effective works better than something intended
sults. to be ineffective.2
Unlike medications, which require years of research and de- Third, science is about examining all the evidence, not just the
velopment in the laboratory, many therapeutic interventions results from particular designs. Experimental designs are the
(e.g., cognitive therapy) emerge from practice, not the laboratory. best method we have for assessing causal relations, and for this
As psychotherapy researchers, we would do well to use clinical reason randomized controlled trials are of particular importance
practice as a natural laboratory for identifying promising treat- in evidence-based practice. However, the tradeoff of external to
ment approaches, which we could then compare to laboratory- internal validity in randomized controlled trials—that is, the
derived interventions—rather than testing the one or two treat- emphasis on really ‘‘tight,’’ well-controlled designs as opposed to
generalizability to the real world—tends to be very high, ren-
1
Whether these characteristics are inherent in the methodology of empirically dering imperative the use of multiple forms of evidence. Para-
supported therapies is a matter of some debate, but they are, empirically, char-
acteristic of nearly all empirically supported therapies. The only such therapies doxically, by exhorting clinicians to apply treatment manuals
not characterized by these features have obtained some of the most impressive usually developed 10 to 20 years earlier (the time it takes for one
results in the literature, but they are multifaceted treatment packages whose
elements have never been dismantled, and hence whose active ingredients (and or two National Institute of Mental Health-funded replications),
efficacy relative to the multifaceted, often integrative or eclectic treatments the empirically supported therapies movement has minimized
widely practiced in the community) are unknown. New treatments are also on the the importance of basic science in evidence-based practice.
horizon for addressing some of the limitations of single-disorder manuals for
mood, anxiety, and eating disorders, which is a very promising development. Surely, developments over the last decade in the understanding
However, to our knowledge, these treatments are maintaining the 20-session
upper limit characteristic of the empirically supported therapies literature,
2
which seems contrary to their goals of addressing broader targets than treatments A study by Marsha Linehan on dialectical behavior therapy will, to our
of similar length aimed at single disorders. knowledge, be the first.

Volume 14—Number 5 267

 Evidence-Based Practice in Psychotherapy

of the neural mechanisms of learning, personality diatheses (i.e., manuals sequentially, beginning with the most distressing or
underlying vulnerabilities) for psychiatric disorders, comor- debilitating disorder first and then working one’s way through
bidity (co-occurrence of multiple disorders), implicit (uncon- manuals for the other disorders. That this is the most effective
scious) processes, attitude formation and maintenance, emotion, way to address the complex, multifaceted concerns that bring
developmental psychopathology, and other psychological pro- most patients into treatment is, however, an untested assump-
cesses should be considered by clinicians delivering evidence- tion. This assumption is likely to be problematic when seemingly
based interventions, whether or not these developments have yet distinct syndromes reflect common underlying causes or when
been assimilated by applied researchers into particular treat- the presence of multiple syndromes has emergent properties not
ment manuals. reducible to each alone.
In fact, shared diatheses (i.e., two putatively distinct disorders
BRIEF TREATMENTS FOR DISCRETE DISORDERS having related causes) and emergent properties appear to be
more the norm than the coincidental occurrence of unrelated
The empirically supported therapies movement is predicated on syndromes, for which a sequential-manuals approach might be
the assumption that most patients either have, or can be treated appropriate. Paradoxically, the movement to codify a list of
as if they have, one primary syndrome, for which a single empirically supported therapies for specific disorders emerged
treatment package can be designed. Without this assumption, at precisely the same time (and often in the same laboratories) as
researchers would need to test dozens of manuals to address all basic-science research demonstrating that vulnerabilities based
the possible interactions among disorders for even a handful of in personality and temperament account for much of the reason
common disorders (e.g., the interaction of major depression and patients tend to come in with multiple mood, anxiety, or other
panic disorder, of major depression and substance abuse, or of disorders (e.g., substance use disorders, eating disorders; Brown,
major depression and both panic and substance abuse). This Chorpita, & Barlow, 1998; Krueger, 2002). Research using
assumption is not inherent in the use of randomized controlled statistical methods like structural equation modeling and other
trials; indeed, it is precisely what distinguishes randomized multivariate techniques consistently finds that personality
controlled trials in the empirically supported therapies era from variables such as negative affectivity (the tendency to experi-
randomized controlled trials in the past, which often used mixed ence unpleasant feelings) underlie an internalizing spectrum of
patient samples. pathology—that is, a spectrum of symptoms and syndromes
In retrospect, the effort to create a list of empirically supported reflecting problems with anxiety and depression. From this point
therapies without first empirically addressing the question of of view, treatment researchers might do well to develop inter-
what brings most patients in for treatment (i.e., conducting a ventions for personality diatheses (e.g., negative affectivity,
needs assessment) was a serious error. Many patients in clinical impulsivity, maladaptive ways of trying to regulate emotions, and
practice do not meet the thresholds imposed by DSM-IV for repetitive interpersonal patterns and views of the self and rela-
‘‘having’’ a specific disorder (e.g., a patient may frequently binge tionships that underlie many of these patterns) and spend less
and purge but not twice a week every week for 3 months, hence time developing treatments to address the residual variance
receiving the diagnosis of eating disorder not otherwise specified accounted for by specific symptom clusters. More likely, treat-
instead of bulimia nervosa). After 20 years of research, we know ment research should focus on both the person and the specific
little about the generalizability of research on disorders such as symptoms with which the person presents.
major depressive disorder, generalized anxiety disorder, or The growing recognition of the role of personality in psycho-
PTSD for the vast numbers of patients who present with de- pathology suggests substantial limits to the brief, focal treat-
pression, high trait anxiety, or a history of childhood trauma, ments that have been the near-exclusive focus of empirically
respectively. supported therapies research (see Westen, Gabbard, & Blagov,
Further, patients who present with a single syndrome are the in press). The primary treatments tested in randomized con-
exception rather than the rule in both clinical practice and re- trolled trials to date were never intended to bring about per-
search settings. Indeed, the correlation between self-report sonality change. The interpersonal therapy manual for
measures of anxiety and depression tends to be large in both depression, for example, is explicit in its focus on current rather
normal and clinical samples, raising questions about focusing than recurrent interpersonal patterns. The cognitive-behavioral
primary attention on discrete syndromes. Studies consistently therapy manual for depression focuses the first 5 to 6 sessions on
find that most clinical syndromes occur together with other psychoeducation and behavioral activation, leaving little time to
syndromes or with personality pathology and that the presence of explore anything but explicit or readily accessible automatic
co-occurring conditions, like the presence of significant per- thoughts that help maintain the present episode.
sonality pathology, often (but not always) augurs poorly for Equally problematic is the assumption, implicit in the modest
therapy outcome in brief clinical trials. Advocates of empirically treatment lengths characteristic of empirically supported ther-
supported therapies usually suggest that the way to handle such apies, that psychopathology is relatively mutable over brief in-
co-occurring symptoms or syndromes in practice is to apply tervals. This assumption appears to be valid for some disorders,

268 Volume 14—Number 5

 Drew Westen and Rebekah Bradley

notably anxiety syndromes characterized by an association 100

% Improved of completed treatment
between an identifiable stimulus or small set of stimuli and a 90
% Improved of entered treatment
conditioned emotional response. For most disorders, however, 80
studies of both naturalistic and treatment samples find that re- 70
lapse and residual impairment are the rule rather than the ex- 60

ception. The natural course of major depression, for example, is 50

one of repeated reoccurrences with residual symptoms often 40

persisting between major depressive episodes. 30

20
10
EMPIRICALLY UNDERQUALIFIED CONCLUSIONS
0
OCD Panic GAD PTSD Depression
We turn now to the findings of studies widely viewed as vali-
Fig. 1. Percent of patients improved of those who entered or completed
dating empirically supported therapies. To what extent a treat- psychotherapy for five disorders: obsessive-compulsive disorder (OCD),
ment can be considered empirically supported depends on what panic disorder, generalized anxiety disorder (GAD), posttraumatic stress
one considers a positive outcome. In fact, a treatment can look disorder (PTSD), and depression. The data come from randomized con-
trolled trials for each disorder summarized meta-analytically (i.e. across
like it is empirically supported using one set of criteria and studies). Across disorders, around half of those who completed these
empirically unsupported using another. Over the last several treatments tended to improve; less than half of patients who entered
years, we have used meta-analysis to summarize findings about treatment (including those who, for whatever reasons, did not choose to
complete treatment) typically improved. (Comparable data were not
both treatment outcome and generalizability of results to eve- available for bulimia nervosa, hence their absence here, although recovery
ryday practice across studies of empirically supported therapies rates for bulimia tend to resemble improvement rates for other disorders.)
for a number of disorders. In these investigations, we have de-
scribed the state of the art using multiple metrics that provide To index generalizability, one can describe the common inclu-
distinct, often nonoverlapping indicators of empirical support in sion/exclusion criteria used in studies of a given disorder, the
what we have called a multidimensional meta-analysis. The goal number of exclusion criteria used to weed out patients whose
is to provide a more nuanced view of treatment efficacy than a complexities may complicate treatment, and the percent of
simple thumbs up or thumbs down. patients excluded of those screened.
The first and most familiar metric, effect size, describes the Thus far we have completed multidimensional meta-analyses
magnitude of the effect the average patient can expect to receive, on six widely studied disorders: major depressive disorder, panic
usually relative to a control group (i.e., how much benefit does disorder, generalized anxiety disorder, obsessive-compulsive
the average patient get from this treatment?). A treatment could, disorder, bulimia nervosa, and PTSD (Bradley, Greene, Russ,
however, obtain a moderate effect size by producing a very large Dutra, & Westen, 2005; Eddy, Dutra, Bradley, & Westen, 2004;
effect for a small subset of patients (or a moderate but incomplete Thompson-Brenner, Glass, & Westen, 2003; Westen & Morrison,
reduction in symptoms for many). Thus, two other useful metrics 2001). Across disorders, treatment-versus-control effect sizes
are percent recovered and percent improved. In calculating im- tend to be moderate to large (generally within the 95% confi-
provement and recovery rates, however, the devil is in the de- dence interval identified by Smith & Glass, 1977), providing a
nominator: percent improved or recovered out of what number? compelling portrait of the efficacy of these treatments.
A liberal estimate limits the denominator to patients who com- Data on percent of patients improved and recovered paint a
pleted the treatment (completer analyses). A more conservative more variable picture. Figure 1 describes improvement rates in
estimate uses the number of patients who actually began treat- randomized controlled trials for patients receiving active treat-
ment (intent-to-treat analyses), on the logic that patients who ments widely described as empirically supported for a number of
drop out should be factored into the results. Neither metric is disorders. On average, roughly half of patients who complete
more definitive than the other; each provides incremental in- treatments in such trials improve. Recovery is harder to come by,
formation—that is, information that the other does not provide. with both completer and intent-to-treat analyses showing re-
Complicating matters further, a treatment could lead to sub- covery rates in the 30s or low 40s for most disorders.
stantial improvement in most patients but nevertheless leave The good news is that these response rates are generally
them highly symptomatic. Thus, another important index is re- substantially higher than for patients in the control conditions.
sidual post-treatment symptomatology. Also important is sus- However, we suspect most patients would be chagrined to learn
tained efficacy over time, particularly in disorders characterized that what is widely described in the research literature as the
by frequent recurrence. And a final set of variables quantifies treatment of choice for their disorder gives them a 1 in 3 chance
generalizability. If researchers screen substantially to maximize of recovery, and that in the last 20 years of psychotherapy re-
the homogeneity of samples (e.g., eliminating patients with search virtually no one has considered doubling or tripling the
certain troublesome forms of comorbidity), they need to qualify ‘‘dose’’ of these therapies. Nor has anyone compared the im-
the population to whom their results are intended to generalize. provement and recovery rates obtained in randomized trials of

Volume 14—Number 5 269

 Evidence-Based Practice in Psychotherapy

100 The data for panic disorder are very impressive. The data for
90
depression, for which researchers have made broad claims about
efficacy, are poor. The most recently published, high-quality
80
randomized controlled trial for major depression yielded similar
70 findings: Only one in four patients treated with cognitive be-
60 havioral therapy recovered and remained recovered 2 years later
50 (Hollon et al., 2005).
With respect to generalizability, the vast majority of patients
40
who appear for an initial interview (let alone of those who call
30
and are screened out on the phone) are excluded from most
20 studies for most disorders, raising questions about the extent to
10 which these treatments can be generalized to patients in the
0
community. The universal exclusion of patients who are thinking
Panic Depression Bulimia about suicide from treatment studies of major depression is a
Fig. 2. Percent of psychotherapy patients treated for panic, depression, or troubling example (in both psychotherapy and medication tri-
bulimia who remained improved at 12- to 24-month follow-up. The data als), given that the alternative is to send these patients off for
come from randomized controlled trials for each disorder summarized therapies that are empirically untested.
meta-analytically (i.e. across studies), where follow-up data were available.

CONCLUSIONS
brief therapies to results obtained by experienced clinicians,
some percent of whom, either by chance, trial and error (i.e., The empirically supported therapies movement provided one
operant conditioning), or creativity, experimentation, and ex- approach to operationalizing evidence-based practice—an ap-
perience, are likely to have happened upon something more proach that made considerable sense in light of the context in
effective. 1995. At that time, prominent psychiatric treatment guidelines,
Data on residual post-treatment symptomatology tell a similar often influenced by the pharmaceutical industry, were placing
story of a glass both half empty and half full. Treatments for psychotherapy on the back burner. However, it has become clear
bulimia nervosa have produced some of the most impressive over time that this operationalization of evidence-based practice
recovery rates of the disorders we have studied meta-analyti- is limited by the evidence it includes and excludes; by the as-
cally: Roughly 40% of patients who enter treatment and 50% sumptions inherent in focusing on brief treatments for discrete
who complete it no longer binge and purge at the end of treat- syndromes, many of which are themselves empirically prob-
ment. However, the average rates of bingeing in patients in lematic; and by the assumption that treatments can be readily
randomized controlled trials, including both those patients who classified as either supported or unsupported, which seems
have stopped completely and the remainder of patients treated untenable in light of meta-analytic data from randomized con-
with empirically supported therapies, is once to twice a week, trolled trials. If the goal is to identify best practice, researchers
and the average rate of purging is twice to three times a week— should consider all available data (e.g., basic research on psy-
just short of the threshold for being diagnosed with the disorder. chopathology), not just data using one research design; they
(In this case, as in the other disorders we have studied, the should identify promising treatment strategies empirically
numbers are not substantially different for specific treatments rather than testing only those treatments preferred a priori by
widely believed to be the treatment of choice than for other researchers that can be readily tested in brief versions; they
treatments intended to work.) These data provide two very im- should impose only those exclusion criteria in randomized
portant take-home messages, neither of which is ‘‘more true’’ controlled trials that a sensible clinician would impose in every-
than the other. The first is that the average patient experiences a day practice (e.g., psychosis or severe brain damage in treating
dramatic reduction in binge and purge episodes after treatment depression) to maximize generalizability; and they should rou-
(many of these patients begin treatment bingeing and purging tinely compare laboratory-derived treatments to empirically
daily). The second is that the average patient is not restored to identified best practice in the community rather than to waitlist
health in 18 to 20 weeks. controls and intent-to-fail conditions (e.g., treatments for buli-
For most disorders, the data on sustained efficacy are more mia nervosa in which the therapist is forbidden from discussing
troubling. Researchers generally have not followed patients the symptom). The burden of proof for a new treatment should be
suffering from chronic or recurrent disorders beyond 6 to 9 that its outcomes compare favorably to the outcomes obtained by
months, and when they have, the results have often been so- experienced clinicians, not that it survives a test of the null
bering. Figure 2 shows the meta-analytic data on the percentage hypothesis (i.e., that it works better than nothing, or better than
of patients in active treatments in randomized controlled trials something intended to fail), given the consistent finding over
showing sustained improvement or recovery at 12 to 24 months. 30 years that most psychotherapies for most patients produce a

270 Volume 14—Number 5

 Drew Westen and Rebekah Bradley

large effect relative to controls. Indeed, researchers would likely Bradley, R., Greene, J., Russ, E., Dutra, L., & Westen, D. (2005). A
find experienced clinicians much more receptive to dissemi- multidimensional meta-analysis of psychotherapy for PTSD.
nation of their treatments if they could demonstrate that these American Journal of Psychiatry, 162, 214–227.
Brown, T.A., Chorpita, B.F., & Barlow, D.H. (1998). Structural rela-
treatments outperform what experienced clinicians currently do.
tionships among dimensions of the DSM-IV anxiety and mood
We should repeat that we are not arguing against the impor- disorders and dimensions of negative affect, positive affect, and
tance of randomized controlled trials. No other method is as autonomic arousal. Journal of Abnormal Psychology, 107, 179–
definitive in demonstrating causation. We are arguing, rather, 192.
that the utility of such trials is being unnecessarily compromised Chambless, D., & Ollendick, T. (2000). Empirically supported psy-
by systematic biases in both the range and duration of inter- chological interventions: Controversies and evidence. Annual
ventions being tested and the targets of intervention (single, Review of Psychology, 52, 685–716.
Eddy, K.T., Dutra, L., Bradley, R., & Westen, D. (2004). A multidi-
categorically defined disorders, which defy everything we know
mensional meta-analysis of pharmacotherapy for obsessive-com-
about psychopathology). pulsive disorder. Clinical Psychology Review, 24, 1011–1030.
We can summarize our argument in a single equation: EBP > Hollon, S., DeRubeis, R.J., Shelton, R.C., Amsterdam, J.D., Salomon,
EST (i.e., evidence-based practice [EBP] is more than a list of R.M., O’Reardon, J.P., Lovett, M.L., Young, P.R., Haman, K.L.,
empirically supported therapies [ESTs] for discrete disorders). Freeman, B.B., & Gallop, R. (2005). Prevention of relapse fol-
The empirically supported therapies movement has helped lowing cognitive therapy versus medications in moderate to severe
depression. Archives of General Psychiatry, 62, 417–422.
foster a more empirical attitude toward psychotherapy that was
Krueger, R.F. (2002). The structure of common mental disorders.
long overdue in clinical psychology. However, as an operation- Archives of General Psychiatry, 59, 570–571.
alization of evidence-based practice, its definition of evidence Meehl, P.E. (1954). Clinical vs. statistical prediction. Minneapolis:
eliminates too much of the available science; its focus on brief University of Minnesota Press.
treatments for discrete disorders reflects a set of assumptions Sackett, D.L., Straus, S.E., Richardson, W.S., Rosenberg, W., & Haynes,
that are not themselves empirically supported; and its tendency R.B. (2000). Evidence based medicine: How to practice and teach
to foster dichotomous, ‘‘either–or’’ thinking about empirical EBM (2nd ed.). London: Churchill Livingstone.
support is neither scientifically useful nor justified by the Smith, M., & Glass, G. (1977). Meta-analysis of psychotherapy outcome
studies. American Psychologist, 32, 752–760.
available evidence.
Thompson-Brenner, H., Glass, S., & Westen, D. (2003). A multi-
dimensional meta-analysis of psychotherapy for bulimia nervosa.
Clinical Psychology: Science and Practice, 10, 269–287.
Recommended Reading Wampold, B.E. (2001). The great psychotherapy debate: Models, meth-
Barlow, D. (2002). Anxiety and its disorders (2nd ed.). New York: ods, and findings. Mahwah, NJ: Erlbaum.
Guilford Press. Westen, D., Gabbard, G., & Blagov, P. (in press). Back to the future:
Borkovec, T.D., & Castonguay, L.G. (1998). What is the scientific Personality structure as a context for psychopathology. In R.F.
meaning of empirically supported therapy? Journal of Consulting Krueger & J.L. Tackett (Eds.), Personality and psychopathology.
and Clinical Psychology, 66, 136–142. New York: Guilford.
Westen, D., & Morrison, K. (2001). (See References) Westen, D., & Morrison, K. (2001). A multidimensional meta-analysis
Westen, D., Novotny, C., & Thompson-Brenner, H. (2004). (See Refer- of treatments for depression, panic, and generalized anxiety dis-
ences) order: An empirical examination of the status of empirically sup-
ported therapies. Journal of Consulting and Clinical Psychology,
69, 875–899.
Westen, D., Novotny, C.M., & Thompson-Brenner, H. (2004). The em-
Acknowledgments—Preparation of this article was supported
pirical status of empirically supported psychotherapies: As-
in part by NIMH Grants MH62377 and MH62378. sumptions, findings, and reporting in controlled clinical trials.
Psychological Bulletin, 130, 631–663.
REFERENCES Westen, D., Novotny, C.M., & Thompson-Brenner, H. (2005). EBP6¼
EST: Reply to Crits-Christoph, Wilson, and Hollon (2005) and
American Psychiatric Association (2004). Diagnostic and statistical Weisz, Weersing, and Henggeler (2005). Psychological Bulletin,
manual of mental disorders (4th ed.). Washington, DC: Author. 13, 427–433.

Volume 14—Number 5 271