Psychological Medicine, 2002, 32, 763–782. " 2002 Cambridge University Press
DOI: 10.1017\S0033291702005895 Printed in the United Kingdom
REVIEW ARTICLE
Psychological treatments in schizophrenia: I. Meta-analysis of
family intervention and cognitive behaviour therapy
S. PILLING,
"
P. BEBBINGTON, E. KUIPERS, P. GARETY, J. GEDDES, G. ORBACH
C. MORGAN
From the Centre for Outcomes, Research and Effectiveness, Department of Psychology, University
College London, Royal Free and University College Medical School, Department of Psychiatry and
Behavioural Sciences, UCL, Department of Psychology, Institute of Psychiatry and Department of
Psychiatry, St Thomas’ Hospital, London; and Department of Psychiatry, University of Oxford,
Warneford Hospital, Oxford
ABSTRACT
Background. While there is a growing body of evidence on the efficacy of psychological
interventions for schizophrenia, this meta-analysis improves upon previous systematic and meta-
analytical reviews by including a wider range of randomized controlled trials and providing
comparisons against both standard care and other active interventions.
Method. Literature searches identified randomized controlled trials of four types of psychological
interventions: family intervention, cognitive behavioural therapy (CBT), social skills training and
cognitive remediation. These were then subjected to meta-analysis on a variety of outcome
measures. This paper presents results relating to the first two.
Results. Family therapy, in particular single family therapy, had clear preventative effects on the
outcomes of psychotic relapse and readmission, in addition to benefits in medication compliance.
CBT produced higher rates of ‘important improvement’ in mental state and demonstrated positive
effects on continuous measures of mental state at follow-up. CBT also seems to be associated
with low drop-out rates.
Conclusions. Family intervention should be offered to people with schizophrenia who are in contact
with carers. CBT may be useful for those with treatment resistant symptoms. Both treatments, in
particular CBT, should be further investigated in large trials across a variety of patients, in various
settings. The factors mediating treatment success in these interventions should be researched.
INTRODUCTION
The effectiveness of antipsychotic medication
has made it central to the treatment of schizo-
phrenia (Schwartz et al. 1993). However, there is
an increasing acknowledgement that pharma-
cological treatment on its own is rarely sufficient
for the best outcome in this disabling condition.
"
Address for correspondence: Mr Stephen Pilling, Centre for
Outcomes, Research and Effectiveness (CORE), Department of
Psychology, UCL, 1–19 Torrington Place, London WC1E 6BT.
There are a number of reasons for this. First, the
issue of compliance has made it clear that the
social and cognitive context in which phar-
macological treatment is delivered has a major
effect on its success (Bebbington & Kuipers,
1994). Secondly, the effectiveness of anti-
psychotic medication has to some extent been
called into question. This came about because of
the interest in treatment resistance fostered by
the introduction of clozapine. Thus, it is now
generally held that a significant proportion of
patients, perhaps up to 40 %, have a poor
763
764 S. Pilling and others
response to antipsychotic medication and con-
tinue to show moderate to severe psychotic
symptoms (Kane, 1996). The final element in
the re-evaluation of the treatment of schizo-
phrenia is that there has been a change in per-
ception of psychological interventions, such
that they are now recognized as an important
component of a comprehensive therapeutic ap-
proach (Department of Health, NHS Executive,
1999).
Antipsychotic drugs have a limited impact on
the negative symptoms of schizophrenia, and
thus do not contribute to the development of the
skills necessary for successful transition back
into the community (Liberman, 1994). A recent
meta-analysis has suggested that the benefit even
of the new atypical antipsychotics is less than
previously thought (Geddes et al. 2000). There
is thus a clear requirement for the develop-
ment of new approaches if the wide ranging
needs of people with schizophrenia are to be
met. This emphasis is new. In comparison
to those channelled into pharmacological in-
terventions, relatively few resources have been
spent on evaluating and developing other
treatments.
The analysis of psychological interventions in
schizophrenia presented here is the product of a
joint British Psychological Society and Royal
College of Psychiatrist’s working party on the
development of Psychosocial Guidelines for the
Treatment of Schizophrenia. It draws largely on
the Cochrane methodology (Mulrow & Oxman,
1997), with certain modifications. In identifying
areas for review, the working party had two
guiding principles. To warrant the inclusion of
the type of psychological intervention, we
required there should be considerable uncer-
tainty about both the interpretation of the
existing research findings and the application
of these findings in routine practice. There
also had to be a sufficient level of evidence (in
the form of randomized controlled trials) to
support systematicreview andmeta-analysis. The
current paper describes the methodology used
and presents results relating to family inter-
ventions (FIs) and cognitive behaviour therapy
(CBT). It is the first of a pair intended to assess
whether the current enthusiasm for psycho-
logical interventions can be justified by the
evidence. The second paper (Pilling et al. 2001)
covers social skills training and cognitive
remediation.
Talking to people with schizophrenia and their
relatives
Modern psychological treatments for schizo-
phrenia originate in studies of the impact
of the social environment on mental illness. In
the United Kingdom there was an interest in the
effect of the family environment in the main-
tenance of major mental disorders, in particular
schizophrenia (Brown et al. 1962). This led to
the development of the concept of expressed
emotion (Brown & Rutter, 1966; Brown et al.
1972), and the establishment of family inter-
ventions for the treatment of schizophrenia (Leff
et al. 1982). These revolutionized the way that
clinicians viewed family members, and had a
general effect of improving communication
between clinicians and informal carers. The
application of CBT for depression (Beck et al.
1979) to psychotic disorders has resulted in a
complex treatment package with a range of
techniques and targets (Fowler et al. 1995).
There is considerable variation in the content
and application of these psychological inter-
ventions. The effectiveness of all psychological
interventions for schizophrenia depends on the
establishment of a positive therapeutic alliance
with the patient (Roth & Fonagy, 1996). Work-
ing with people with schizophrenia presents par-
ticular difficulties, and the pace and development
of the therapeutic alliance demands great flex-
ibility. However, the requirement for such an
alliance is clear.
There have been several systematic reviews
and meta-analytical studies of psychological
interventions in schizophrenia in the past 10
years that cover this area in whole or in part
(Mari & Streiner, 1994; Mojtabi et al. 1998;
Adams, 2000; Dixon et al. 2000; Gould et al.
2001; Bustillo et al. 2001; Pitschel-Walz et al.
2001; Rector & Beck, 2001 ; Cormac et al. 2002).
These vary considerably in range, depth and
focus. In the current paper, we have confined
our review to high quality randomized controlled
trials (RCTs) providing comparisons either with
standard care or with other active interventions.
We feel that this is an advance on previous
work. We have been able to identify more RCTs
than have been reported in previous systematic
Psychological treatments in schizophrenia: I 765
reviews. The meta-analyses are based on the
examination of original data and not taken from
other systematic reviews or meta-analytical
studies.
METHOD
Research strategy
Electronic searches for both family interventions
and cognitive behavioural interventions were
undertaken using Biological Abstracts (1980–
1999), CINAHL (1982–1999), the Cochrane
Library (Issue 2, 1999), the Cochrane Schizo-
phrenia Group’s Register of Trials (August,
1999), EMBASE (1980–1999), MEDLINE
(1966–1999), PsycLIT (1887–1999), SIGLE
(1990–1999), and Sociofile (1980–1999). (More
detailed descriptions of both search strategies
are available from the authors.) All reference
lists of the articles selected were searched for
further relevant trials. Review articles were also
scanned.
The basis of study selection
Papers were checked for methodological rigour
and validity by two reviewers (S.P. and G.O.),
who independently inspected all citations, ad-
hering to guidelines for conducting literature
reviews (Mulrow & Oxman, 1997). When dis-
putes arose about which category a citation
should be allocated to, or its relevance to the
report, we attempted to resolve them by dis-
cussion. If this failed, a further reviewer (P. B. or
P.G.) was asked to review the article and decide.
Only RCTs were considered for inclusion in the
analysis.
Given the problems in the literature sur-
rounding the definitions of psychosocial inter-
ventions and of diagnosis, explicit inclusion
criteria were specified.
For an intervention to be classed as ‘family
intervention’ it had to include family sessions
with a specific supportive and treatment func-
tion, and a minimum of one of the following
treatment components: psycho-educational in-
tervention; problem solving\crisis management
work; or, intervention with the identified patient.
In addition, interventions were required to be
for at least 6 weeks.
In order to be classified as ‘cognitive be-
haviour therapy ’ interventions had to have a
component which involved recipients estab-
lishing links between their thoughts, feelings or
actions with respect to the target symptoms ; and
the correction of their misperceptions, irrational
beliefs or reasoning biases related to those
symptoms. At least one of the following was also
required: self-monitoring of the treated person’s
thoughts, feelings or behaviours with respect to
the target symptoms; and the promotion of
alternative ways of coping with the target
symptoms.
The included studies were based on samples of
people with schizophrenia or related disorders,
including delusional disorder, schizophreniform
disorder and schizoaffective disorder (basically
ICD-10 F2 ; WHO, 1992). Trials where par-
ticipants were not restricted to people with
schizophrenia and from which it was impossible
to extract results for this group were not
included. Many participants were also reported
to have co-morbid mental disorders, such as
depression or anxiety disorder.
The individual trials excluded participants for
a variety of reasons such as organic brain
syndromes, substance misuse and failing to reach
a minimum IQ score. Outcomes were death,
mental state, relapse, re-admission, burden,
expressed emotion, medication compliance and
acceptability of treatment. These outcomes were
chosen because they were thought to be good
indicators of treatment effectiveness, clinically
important, and common to most studies.
Analysis of data
Intention-to-treat analysis was performed on all
data i.e. on a ‘ once randomized always analyse’
basis. This assumes that those participants who
ceased to engage in the study – from whatever
group – had an unfavourable outcome (with the
exception of the outcome of ‘ natural death’).
While recognizing that most psychosocial ther-
apies focus on those with severe illness and
thereby risk high attrition rates, the reviewers
felt that attrition of 50% would call into
question the value of the study. Studies losing
50 % of people were therefore excluded, even
if they reported relevant outcomes.
Analysis of dichotomous outcomes (e.g. re-
lapse, readmission) was performed using odds
ratios. An odds ratio is calculated by dividing
the probability of an event in a treatment group
766 S. Pilling and others
by its odds in the comparison group. Clinical
trials typically look for treatments that reduce
‘bad event’ rates, and therefore aim at an odds
ratio of less than one. For example, a treatment
that caused a 7 % reduction in suicide would
have an OR of 0n93. Ninety-five per cent
confidence intervals are reported with the odds
ratios. The primary analysis employed the fixed
effects method of Mantel & Haenzel (1959). This
method assumes that a single underlying treat-
ment effect is present across all studies. However,
in reality this may not be the case, particularly
for psychological treatments. To account for the
potential heterogeneity in treatment effects, a
random effects analysis was therefore also
undertaken (DerSimonian & Laird, 1986). In
the random effects analysis, heterogeneity is
accounted for both in the width of confidence
intervals and the estimate of the treatment
effect. With decreasing heterogeneity the random
effects approach moves asymptotically towards
a fixed effects model.
A similar analysis was undertaken for con-
tinuous data, using an effect size (Cohen, 1977).
Effect sizes are typically the difference between
the mean in the experimental group and the
mean in the control group, divided by a pooled
standard deviation. Thus, the effect size expresses
the difference between means relative to within-
group variation. The fixed effects model we used
in these analyses was that advocated by Hedges
& Olkin (1985), and the reported statistics were
either standardized effect sizes or weighted mean
differences. Where different measures were used
in different trials (e.g. where the Brief Psychiatric
Rating Scale (BPRS) (Overall & Gorham, 1962)
and the Comprehensive Psychopathological
Rating Scale (CPRS) (A
H
sberg et al. 1978) were
used to estimate the same underlying effect), or
where there was a likelihood of poor inter-rater
reliability, standardized effect sizes were cal-
culated, based on the procedures of Glass et al.
(1981). However, as advocated by Hedges &
Olkin (1985), a pooled standard deviation was
used instead of the comparison group standard
deviation. For the standardized effect size (d), a
value of 1n0 indicates that the mean of the
treatment group is 1 standard deviation higher
than the mean of the comparison group. Each
effect size was calculated using StatsDirect
(2000), and was corrected for bias using cal-
culations from the gamma distribution. A
DerSimonian\Laird random effect size was also
calculated to account for heterogeneity between
studies included in the meta-analysis (Der-
Simonian & Laird, 1986). To avoid applying
parametric tests to data that do not meet their
requirements, a standard was applied to all
continuous data before inclusion. Only papers
where standard deviations or standard errors
and means were reported were included in the
review. When continuous measures started from
a finite number (such as 0), data were included
only if the standard deviation, multiplied by 2,
was less than the mean. Otherwise, the mean was
unlikely to be an appropriate measure of the
centre of the distribution (Altman & Bland,
1996).
As well as inspecting the graphical presen-
tations, reviewers checked whether the differ-
ences between the results of trials were greater
than would be expected by chance alone, using
tests of heterogeneity. In this case, we used a Q
statistic (Hedges & Olkin, 1985). A significance
level less than 0n05 on the Q statistic was
interpreted as evidence of heterogeneity. When
heterogeneity was present, a sensitivity analysis
was undertaken. Outlying studies were then
removed if they caused a substantive change in
the overall findings. Random effects models
were also analysed to take into account het-
erogeneity of treatment.
In addition to the above statistics, the ‘ number
need to treat’ (NNT) was calculated. This
number is the inverse of the absolute risk
reduction i.e. the inverse of the difference in the
proportion of events in the control group and in
the treatment group. It refers to the number of
patients it is necessary to treat in order to
prevent one bad outcome (e.g. relapse) that
would not have been prevented in the control
group. NNTs in this paper are reported rounded
up, in accordance with the general consensus
(Cook & Sackett, 1995). Confidence intervals
for absolute risk reduction and the number
needed to treat are based on the iterative method
of Miettinen & Nurminen (1985).
Where a sufficient number of trials was
available, data were entered into a funnel graph
(with trial effect plotted against trial size or
‘precision ’) to assess the presence of publication
bias. A formal test of funnel plot asymmetry was
undertaken where appropriate (Egger, 1997).
Significance levels of P 0n05 were set a priori
Psychological treatments in schizophrenia: I 767
for accepting the presence of asymmetry. Where
only 3–4 studies reported an outcome, or there
was little variety in sample size (or precision
estimate) between studies, tests of asymmetry
were not appropriate.
Family interventions
From this search, a total of 33 trials of family
interventions for schizophrenia were identified,
although 15 of these were excluded from the
meta-analysis for a variety of reasons. The main
reasons for exclusion were: the intervention
comprised less than six sessions ; methods of
randomization were inadequate; participants
did not exclusively have schizophrenia and
related disorders; there was no appropriate
control group; or there were no usable data. The
18 remaining studies involved a total of 1467
patients.
Cognitive behavioural interventions
A total of 22 trials concerning cognitive behav-
ioural therapy were identified. However, 14 of
these were excluded. The main reasons for
exclusion were: methods of randomization were
inadequate; the intervention did not meet criteria
for ‘cognitive behavioural therapy ’ ; or there
were no usable data. The eight remaining studies
included 528 patients. A full list of all trials
identified for both interventions is available
from the authors.
Comparisons
Due to the rather diverse nature of the psycho-
social treatments and their comparison groups,
it was necessary to analyse several different
comparisons. However, a global comparison
was initially conducted, including all studies
that reported outcomes regardless of the com-
parison group. Then, dependent on the presence
of heterogeneity, we analysed studies that com-
pared the intervention only with standard care,
or only with other active treatments. Family
therapy was also divided into two particular
types of intervention, single family and group
family interventions. Group family interventions
were defined as those where the primary compo-
nent of the treatment was a regular group
session including more than one family. Separate
comparisons are reported where appropriate.
RESULTS
Family interventions
The main variables available to form the basis of
comparison in family treatment comprised re-
lapse in intervals of various duration dating
from the onset of treatment, relapse during
intervals commencing from the end of treatment,
readmission rates during similar sets of intervals,
rates of dropouts, rates of suicide, the effects on
burden and expressed emotion, and the level of
medication compliance.
Characteristic of participants (Table 1)
A total of 1467 patients were included in the 18
family intervention trials analysed in this review.
All studies reported the ages of patients, the
mean being 31n2 years. Fourteen studies reported
the sex of participants, with 31 % female overall.
The mean number of prior admissions, as
reported in 13 studies, was 2n7. Seven studies
report data on mean duration of illness, which
was 6n3 years. Various criteria were used to
provide a diagnosis of schizophrenia; three
studies used DSM-III, seven studies DSM-III-R,
one study DSM-IV, one the New Haven index,
one the Research Diagnostic Criteria (RDC),
and one the Chinese Medical Association’s
Criteria. Four used the Present State Examin-
ation (PSE) to provide an ICD-9 diagnosis.
Outcomes have been grouped into 12-month
time periods for convenience and consistency of
presentation. However, it should be stressed
that measurements were made at some point in
that time period, and it may be of greater
relevance whether measurements were made
during, or after the end of treatment.
Relapse
Of a total of 765 patients for whom relapse rates
were reported, 144 out of 381 receiving family
interventions and 206 out of 384 receiving other
treatments, including standard care, relapsed
within a period of 4 years. The results of the
relapse analysis are reported in Table 2.
Eleven studies compared relapse with all other
treatments over the first 12 months of treatment
(Goldstein et al. 1978; Falloon et al. 1982;
Leff et al. 1982, 1989; Hogarty et al. 1986,
1997a ; Tarrier et al. 1988; Glynn et al. 1992 ;
Xiong et al. 1994 ; McFarlane et al. 1995a, b;
