A meta-analysis of profile and time-course of
symptom change in acute schizophrenia treated
with atypical antipsychotics
Megan Sherwood1,2, Allen E. Thornton3and William G. Honer1
1Centre for Complex Disorders, Department of Psychiatry, University of British Columbia, Vancouver, BC, Canada
2Department of Psychiatry, St. Paul’s Hospital, Vancouver, BC, Canada
3Department of Psychology, Simon Fraser University, Burnaby, BC, Canada
The profile and time-course of symptom response in acute schizophrenia is unclear. For the present study,
we hypothesized that the time-course would be nonlinear. A meta-analysis was performed using ran-
domized, controlled clinical trials of five atypical antipsychotics reported in nine electronic databases.
Studies were of subjects experiencing an acute exacerbation of illness, with multiple BPRS or PANSS data-
points as outcome measures. A mixed factorial repeated-measures ANOVA was used. Twenty-one pub-
lished clinical trials were identified. Reduction in total symptoms from baseline to 4 wk was associated
with a linear decline in symptomatology (F=23.4, d.f.=1, 7, p=0.002) without attenuation of effect. In
contrast, from baseline to 6 wk the linear symptom reduction (F=76.5, d.f.=1, 12, p<0.001) eventually
flattened at the end of the trial (F=87.2, d.f.=1, 12, p<0.001). Secondary analyses showed a similar pattern
for typical antipsychotics, and the same profile for risperidone and olanzapine as for atypical agents as a
whole. Inclusion of LOCF data altered the results at 4 wk, but not 6 wk; completion rates had no effect on
results. In conclusion, this meta-analysis confirms our hypothesis for 6-wk data. The profile of symptom
change is one of linear symptom reduction until 4 wk, with a flattening of treatment effects by 6 wk.
A curvilinear profile of schizophrenia symptom reduction has possible implications with respect to trial
design and clinical decision-making.
Received 8 January 2005; Reviewed 17 March 2005; Revised 12 June 2005; Accepted 16 June 2005;
First published online 5 September 2005
Key words: Adult, antipsychotic agents, clinical trials, mental disorders, meta-analysis, psychiatric status
rating scales, psychotic disorders, randomized controlled trials, schizophrenia.
Long-standing conventional wisdom holds that while
the side-effects of antipsychotic drugs may be appar-
ent even after the first dose, the response of symptoms
to treatment may take weeks (Barbaran, 2000; Gelder
et al., 2001; Marder, 2000; Miyamato et al., 2003). This
belief has considerable implications for research and
practice. To a great extent, clinical research in psy-
chotic disorders has begun to include a longitudinal
component, designed to address the dynamics of
illness progression or recovery (Breier et al., 1991;
Carone et al., 1991; Fenton and McGlashan, 1991;
Heinrichs and Awad, 1993). Accurate timing of
sampling is important for such studies. Similarly, the
timing of measures of response in clinical trials
depends on the profile of treatment response over
time. Increasingly sophisticated analyses are being
applied to clinical trial outcome data; however, the
validity of models of treatment response over time is
uncertain (Gueorguieva and Krystal, 2004; Lieberman
et al., 2003). In regular clinical practice, knowledge of
the actual time-course of response to antipsychotic
drugs will influence decisions of how long to maintain
patients on an individual drug before considering
changes to treatment (APA, 2004).
Keck et al. (1989) summarized the early literature
tom improvement over time were observed. However,
these studies included heterogeneous diagnoses,
treatment with typical antipsychotics, and use of dis-
similar outcome measures. Only two of these five
studies addressed the time-course over several weeks
Address for correspondence: Dr W. G. Honer, Centre for Complex
Disorders, Vancouver Coastal Health Research Institute, #203-828
West 10th Ave., Vancouver, BC, Canada, V5Z 1L8.
Tel.: 604-875-4827Fax: 604-875-4376
International Journal of Neuropsychopharmacology (2006), 9, 357–366. Copyright f 2005 CINP
for acute schizophrenia. The authors concluded that
the timing and rate of response to treatment remained
uncertain, in part due to the small number of trials
In their recent meta-analysis, Agid et al. (2003)
described an early response to treatment with anti-
sampling 42 studies and over 7000 patients, argues
strongly against a delayed response or non-response
model. Agid et al. sampled broadly, including trials of
medication in both acute exacerbation and chronic
states of illness, as well as subjects treated with typical
and atypical agents by both oral and intramuscular
routes. The authors concluded that antipsychotic
response is greater in the first two weeks compared to
the third and fourth weeks. The approach to analysis
was designed to understand the mechanism of action
of antipsychotic drugs. Questions remain unanswered
concerning responses in acute vs. chronic illness, the
pattern of response beyond 4 wk, and possible biases
introduced by inclusion of studies with only baseline
to end-point values.
The profile and time-course of response for acute
schizophrenia remain unclear. An acute worsening
of symptoms in non-treatment-resistant populations
may approximate the first-episode population, which
is of particular interest to researchers and clinicians
(Lieberman et al., 2003). For the present analysis,
we hypothesized that the time-course of treatment
response to atypical antipsychotic medication would
be nonlinear, with different results at 4 and 6 wk.
We expected continued improvement after 4 wk,
and compared the shape of the time-response curve
at 4 and 6 wk. Our study criteria were more narrowly
defined to reflect current clinical and research popu-
lations, notably with an acute exacerbation of illness,
treatment with an atypical agent, and with multiple
data-points to capture the shape of the time-response
curve. Although this meta-analysis was not designed
to address the following specific areas of interest,
secondary analyses were added with respect to the
symptom change profile of typical antipsychotics and
the profiles of risperidone and olanzapine.
Electronic searches included MEDLINE on OVID from
1966 to September 2003,
September 2003, CINAHL from 1982 to September
2003, a combination of three databases: Cochrane
Database of Systematic Review (CDSR), ACP Journal
EMBASE from 1980 to
Club (ACP), Database of Abstracts of Review of
Effectiveness (DARE) from September 2003, Cochrane
Central Register of Controlled Trials from September
2003, PsychINFO 1974 to September 2003, and LILACS
on line (www.bireme.br/bvs/l/ibd.htm) to December
For the MEDLINE search, key words and phrases used
were ‘schizophrenia’, ‘psychiatric status rating scales’
and the atypical antipsychotic drug names ‘risper-
idone’, ‘olanzapine’, ‘quetiapine’, ‘ziprasidone’, and
‘aripiprazole’. Where possible, these key words were
exploded. Results were limited in the searching
process to ‘randomized controlled trial’, ‘human’,
and ‘English language’. All abstracts were reviewed
from each search, and methods and results from the
References from included and excluded studies from
the Cochrane schizophrenia group systematic reviews
of these five atypical agents were similarly reviewed.
A similar approach was used with the remainder of
the databases, but modifications included ‘rating
scales’ in place of ‘psychiatric status rating scales’,
randomized control trials was used as a key word,
or clinical trials was used as a limit and results were
searched by hand for the double-blind, randomized
We selected randomized, controlled clinical trials of
patients with an acute exacerbation of schizophrenia
treated with an atypical antipsychotic agent. We
further selected trials that reported at least three
data-points between weeks 0 and 6. Selected studies
reported a Brief Psychiatric Rating Scale (BPRS) total
score or a Positive and Negative Syndrome Scale
(PANSS) total score as an outcome measure; there was
no restriction on publication date or sample size. We
did not search for the placebo arms of clinical trials,
unless these were comparison arms for one of the five
atypical agents. Heterogeneity between trials, includ-
ing older trials of typical antipsychotic agents vs.
placebo, was felt to limit the applicability of including
additional placebo results from trials of typical agents.
The inclusion criteria were:
(1) Subjects were diagnosed with schizophrenia,
(2) Trial medication included risperidone, olanzapine,
quetiapine, ziprasidone, or aripiprazole.
(3) BPRS or PANSS total scores were reported for at
least three data-points between baseline and 6 wk.
(4) Double-blind, randomized control trial.
358M. Sherwood et al.
(5) Subjects defined as experiencing an acute exacer-
bation of illness; or a minimum of 80% of subjects
were required to be experiencing a worsening of
symptoms from their usual baseline.
(6) English language journal.
The exclusion criteria were:
(1) Subjects with a primary mood disorder, or a diag-
nostically heterogeneous group of subjects.
(2) Subjects received more than one primary trial
medication such as olanzapine augmented with
(3) Treatment resistant, stable chronic, or residual
phase of illness patients, or unable to determine
presence of acute exacerbation from published
description or by contacting authors.
(4) Patients under the age of 15 yr.
(5) Studies of less than 4 wk in duration.
(6) Route of administration other than oral.
(7) Insufficient data-points. Only 1 wk of missing data
could be extrapolated between two points, there-
fore some trials with three or more data-points
(8) Second publication if data were published more
After review of published materials and methods and
any related publications, the first author was contacted
for clarification and information where it was unclear
if the subjects were acute exacerbation (n=3 potential
studies). A response was obtained from one query,
resulting in exclusion of one study (Conley and
Mahmoud, 2001). In the other two cases, studies were
excluded for lack of information. Pharmaceutical
companies were contacted in two cases for clarifi-
cation of reporting of last observation carried forward
(LOCF) vs. observed cases.
Data extraction and standardization
PANSS and BPRS total scores were converted to a per-
item score with a baseline of zero. A PANSS per-item
score was calculated as (baseline – 30)/30. BPRS scores
were assumed to be scored 1–7 on an 18-item scale,
unless otherwise specified, and the per-item calculated
as (baseline – 18)/18. Adjustments were made for 0–6
scoring as (baseline)/18. No trial specified a 16-item
BPRS scale. Where BPRS or PANSS data were avail-
able in graphic form, these were enlarged then
extracted by paper-and-ruler method by one of us
(M.S.). We coded demographic variables where these
were available for a majority of studies.
We based the meta-analysis on placebo vs. active
treatment. Thus each study arm was treated as a
distinct ‘subject’, but within the same study, arms
with different doses of the same medication were
combined. Arms with doses of medication at or
lower than the manufacturers’ starting doses were
excluded (f5 mg olanzapine, n=1 arm; f50 mg
quetiapine, n=1 arm). Two trials were excluded due
to inadequate consecutive data-points between base-
line and 4 wk (Plesnicar et al., 2003; Sechter et al.,
Quantitative data synthesis
We used SPSS software (SPSS Inc., Cary, NC, USA) for
a mixed factorial repeated-measures analysis of vari-
ance (ANOVA), including a within-subject component
of weekrweek data, and a between-subject compo-
nent of active vs. placebo treatment. We also used an
approach of controlling for the effects of the placebo
condition on active treatment by using a subtraction
technique. The average BPRS per-item placebo value
for each data-point was subtracted from the BPRS per-
item active treatment value at the corresponding point
for each active treatment arm. An alternative placebo
subtraction technique was also done to compare with
the first technique; where a specific study had a
matched placebo arm for the active treatment arm, this
was subtracted directly. Where no matched placebo
arm existed, the averaged placebo result for placebo
arms as a whole was used for the subtraction. To cor-
rect for potentially positively biased probability values
secondary to violations of the sphericity assumption,
the Greenhouse-Geisser corrected probability value
was reported for the within-subjects factor. In three
studies, we interpolated single data-points from two
If there was a significant effect for treatment group,
week, and a grouprweek interaction, polynomial
follow-up contrasts were applied. The contrasts were
used to identify different patterns of response for
placebo alone, active treatment alone, and active
treatment corrected by subtraction technique. We
limited analysis to linear and quadratic polynomial
terms, which represent the most clinically meaningful
components of the symptom-response curve. A stat-
istically significant linear term indicated a monotonic
decrease in severity of symptoms over time, while
statistically significant linear and quadratic terms
together indicated a curvilinear response over time,
with lessening amounts of change as time increased.
Results were compared using 4 and 6 weeks’ worth
of data. The impact of using data reported as
LOCF vs. observed cases was also determined for each
Symptom change in acute schizophrenia359
As well as the primary analyses, we performed a series
of four separate secondary analyses. The first two
included rates of study completion above vs. below
60%, and use of the BPRS vs. the PANSS as the out-
come measure to determine if additional potentially
confounding variables altered the result. In the third
of these analyses, to explore the pattern of treatment
response for typicals, data from the corresponding
typical arms of these 21 studies were extracted, where
available. Within the same study, arms with different
typical agents were combined (n=1). Both placebo
accounting techniques were used. The same mixed
ing the use of observed cases at 4 wk and LOCF plus
observed cases at 6 wk according to the application
from the original atypical analysis. In the last of
these secondary analyses, a mixed factorial repeated-
measures ANOVA was applied to risperidone and to
olanzapine arms, matching the use of observed cases
at 4 wk and observed plus LOCF cases at 6 wk from
the main analysis.
From the electronic databases, MEDLINE identified 96
trials, EMBASE 76, CINAHL 8, CDSR+ACP+DARE 49,
Cochrane Central Register of Controlled Trials 66, and
LILACS 1. Following application of inclusion and
exclusion criteria, 21 double-blind, randomized con-
trolled trials (Beasley et al., 1996, 1997; Blin et al., 1996;
Borison et al., 1992, 1996; Casey et al., 2003; Ceskova
and Svestka, 1993; Copolov et al., 2000; Daniel et al.,
1999; Hirsch et al., 1996; Hoyberg et al., 1993; Kane,
2002; Keck et al., 1998; King et al., 1998; Klieser et al.,
1995; Lieberman et al., 2003; Martin et al., 2002; Muller
et al., 2002; Peuskens et al., 1999; Potkin et al., 2003;
Small et al., 1997) with 31 separate condensed arms
were identified. This represents 3552 trial subjects. The
studies were published between 1992 and 2003, and
ranged in duration from 4 to 12 wk. Refer to Table 1 for
further sample characteristics.
Results at 4 wk
To evaluate whether active treatment was associated
with a differential symptom recovery in the initial
4 wk, we conducted a two (active treatment vs.
placebo)r five (baseline,
measures ANOVA that included samples containing
both LOCF and observed data. The results indicated a
significant treatment r week interaction (F=13.8,
d.f.=4, 116, p<0.001). As can be seen in Figure 1,
there was a greater linear decline (F=15.8, d.f.=1, 29,
p<0.001) in the symptoms of subjects in the active
treatment arms compared with placebo arms. This
difference is further characterized by the additional
analyses reported below.
Analyses of the eight placebo arms failed to reveal the
presence of significant symptom improvement over
4 wk of follow-up (F=2.7, d.f.=4, 28, p>0.1). As seen
in Figure 1, the treatment response profile was essen-
tially flat. Because the number of samples containing
actual observed cases was small, we were unable to
evaluate differences between LOCF and observed
cases. Nonetheless, the shape of the line was similar
whether observed cases alone were used or whether
LOCF and observed cases were combined.
Active treatment arms
Twenty-three active treatment samples were ident-
ified in total. Fifteen of the samples contained LOCF
data; eight samples were comprised of observed cases
only. To ensure that we accurately characterized
treatment response over time, we evaluated the extent
to which a sample composed of LOCF data differed
from a sample with only observed cases data. While a
main effect for week was detected (F=97.0, d.f.=4, 84,
p<0.001), the LOCF and observed samples showed
differences in their response over the 4-wk follow-
up (F=6.60, d.f.=4, 84, p=0.01). Consequently, the
LOCF samples were excluded as these samples may
artificially flatten the pattern of treatment response
over time; however, both LOCF and observed cases
have limitations in the accuracy of describing the pat-
tern of response. Subsequent analyses examining the
observed samples only revealed that active treatment
was associated with marked symptom decline across
the interval (F=26.8, d.f.=4, 28, p=0.001). As seen in
Figure 1, there was a linear reduction in symptoms
over 4 wk (F=32.0, d.f.=1, 7, p=0.001). There was no
evidence of a statistically significant quadratic term
(all p values >0.25).
Corrected active treatment arms
Use of the placebo subtraction technique to control
for non-specific effects produced essentially the same
result as analyses of uncorrected data. While we
again observed a main effect for weeks of treatment
(F=64.0, d.f.=4, 84, p<0.001), a significant difference
emerged between LOCF and observed cases (F=6.60,
360M. Sherwood et al.
d.f.=4, 84, p=0.01). This difference indicated that in-
clusion of LOCF samples might obscure the pattern of
symptom reduction. Nonetheless, for the eight sam-
ples containing only observed outcomes, a significant
effect for week emerged (F=19.5, d.f.=4, 28, p<0.001),
which was characterized by a linear reduction in total
symptoms from baseline to 4 wk (F=23.4, d.f.=1, 7,
p=0.002). There was no evidence for a statistically
significant quadratic term (all p values >0.4).
As an additional check for potentially confounding
was used (see Methods). This did not change the
main effect for week (F=15.8, d.f.=4, 28, p<0.001),
which is most appropriately characterized as a linear
decline in symptoms (F=18.9, d.f.=1, 7, p=0.003).
Once again, there was no evidence for a statistically
significant quadratic term (all p values >0.5). Indi-
vidual secondary analysis on a small number of cases
(risperidone, n=5; olanzapine, n=3) indicates that
risperidone and olanzapine show the same profile as
atypical antipsychotics as a whole, and risperidone
and olanzapine did not differ from one another (all
p values >0.18). Data are available on request.
Results at 6 wk
To evaluate whether active treatment was associated
with a differential symptom recovery in data extend-
ing out to 6 wk, we conducted a two (active treat-
ment vs. placebo)rseven (baseline, weeks 1–6)
repeated-measures ANOVA that included samples
containing both LOCF and actual observed cases.
The results (see Figure 1) indicated a significant
treatment r week interaction (F=12.5, d.f.=6, 102,
p<0.001). There was a significantly greater linear
decline (F=12.9, d.f.=1, 17, p<0.001) in the symp-
toms of active treatment samples relative to placebo
samples,withan eventual attenuationof this
Table 1. Demographic characteristics
Studies whose subjects received
active treatment (n=23)
Studies whose subjects received
placebo treatment (n=8)
Number of subjects per study
Age at first diagnosis or treatment (yr)
Completion per study (%)
Studies reporting LOCF (%)
Oral washout (minimum number
Depot washout (wk)
Protocol allowed maximum dose
to be achieved by 2 wk
Receiving a benzodiazepine (%)
Receiving an anticholinergic (%)
Studies treating with (%)
Symptom change in acute schizophrenia361
differential effect (F=24.5, d.f.=1, 17, p<0.001). The
pattern of 6-wk follow-up data is further characterized
by the additional analyses reported below.
In the five placebo samples that were reported for
6 wk, there was no effect for weeks of treatment
(F=1.9, d.f.=6, 24, p>0.2). The placebo results are a
combination of LOCF and observed cases; this LOCF
plus observed case placebo is a match for the reported
active results (See Figure 1).
Active treatment arms
Fourteen active treatment studies extended through
6 wk; six of these samples were observed cases, eight
involved LOCF. Analyses failed to suggest a differen-
tial response in the LOCF vs. observed cases (F=1.90,
d.f.=6, 72, p>0.15). However, a significant effect was
observed for treatment weeks (F=127.9, d.f.=6, 72,
p<0.001). As seen in Figure 1, without correction
for placebo effect, the findings indicated a significant
linear (F=166.6, d.f.=1, 12, p<0.001) decrease in
symptoms. In contrast to data limited to 4 wk, the
6-wk samples also demonstrated a significant flatten-
ing of response towards the end of the trial (F=80.7,
d.f.=1, 12, p<0.001).
Corrected active treatment arms
The pattern of 6-wk findings remained similar for data
that was derived using the placebo subtraction tech-
nique to control for non-specific effects. There was
no divergence in response between the LOCF and
observed cases (F=1.9, d.f.=6, 72, p=0.19); however,
a main effect for week was observed (F=66.2,
d.f.=6, 72, p<0.001). From baseline to 6 wk there
was a linear decrease in symptomatology (F=76.5,
d.f.=1, 12, p<0.001) that flattened with ongoing
treatment (F=87.2, d.f.=1, 12, p<0.001). As with the
4-wk data, the alternative placebo subtraction tech-
nique was used as an additional check for confound-
ing variables. This did not change findings at 6 wk
with respect to the apparent significant changes
in symptomatology (F=56.0, d.f.=6, 72, p<0.001)
that involve a linear decline (F=98.0, d.f.=1, 12,
p<0.001), and eventual attenuation of effect (F=71.4,
d.f.=1, 12, p<0.001). Individual secondary analysis
Placebo – all
Mean item score BPRS or PANSS
Placebo – all
Placebo – observed
Figure 1. Decline from baseline in standardized BPRS or PANSS per-item scores (0–6 scale) in study arms of subjects with acute
exacerbations of schizophrenia treated with atypical antipsychotics or with placebo. Panel (a) shows the response of active
treatment arms (atypical, n=8; typical, n=8) with observed cases data for 4 wk; the placebo data is from observed plus LOCF
cases (n=8) or observed cases only (n=2). Significant effects of treatment over time were observed for active treatment but not
placebo treatment (see text). Panel (b) shows the response of observed cases plus LOCF for active treatment arms (atypical,
n=14; typical, n=8) or for placebo arms (n=5) with data for 6 wk. Significant effects of treatment over time were observed for
active treatment but not placebo treatment (see text).
362M. Sherwood et al.
on a small number of cases (risperidone, n=4;
olanzapine, n=4) indicates that risperidone and
olanzapine show the same profile as atypical anti-
psychotics as a whole, and risperidone and olanzapine
did not differ from one another (all p values >0.16).
Data are available on request.
After initial analyses showed a linear effect for atypi-
cals to 4 wk and a more attenuated effect to 6 wk, a
secondary analysis was done for typical antipsychotic
agents where results were reported in the same 21
trials. For 4-wk data, 10 active typical treatment
cases. Following the same procedure as for atypical
agents at 4 wk, analysis of observed cases without
placebo accounting showed a significant effect for
week (F=35.0, d.f.=4, 24, p<0.001), characterized by
a linear (F=49.8, d.f.=1, 6, p<0.001) and quadratic
(F=10.0, d.f.=1, 6, p=0.02) reduction in total symp-
toms from baseline to 4 wk. A similar result was
seen for observed cases with placebo accounting with
week effect (F=22.4, d.f.=4, 24, p<0.001) and linear
reduction over the 4 wk (F=32.4, d.f.=1, 6, p=0.001).
Alternative placebo accounting produced essentially
the same result, with a week effect (F=17.3, d.f.=4, 24,
p=0.002) and linear reduction (F=25.4, d.f.=1, 6,
p=0.002). Again, as with the atypical data, there was
no evidence of a statistically significant quadratic term
(all p values >0.06).
For 6-wk data, six active typical treatment samples
were identified for LOCF plus observed cases extend-
ing to 6 wk. Following the same procedure as for
atypical agents at 6 wk, a similar pattern of 6-wk
findings was determined for typicals as for atypicals.
For LOCF plus observed cases without placebo
correction, findings show a significant week effect
(F=75.1, d.f.=6, 42, p<0.001) with a linear decrease
in symptoms (F=108.8, d.f.=1, 7, p<0.001) as well as
a quadratic effect (F=67.4, d.f.=1, 7, p<0.001). This
is maintained for LOCF plus observed cases with
placebo accounting, similarly showing effect by week
(F=39.9, d.f.=6, 42, p<0.001) and linear (F=50.7,
d.f.=1, 7, p<0.001) and quadratic (F=72.3, d.f.=1, 7,
likewise showed a week effect (F=23.5, d.f.=6, 42,
p<0.001), with linear (F=26.8, d.f.=1, 7, p=0.001)
and quadratic terms (F=168.0, d.f.=1, 7, p<0.001).
Potentially confounding factors
The included studies were separated into two
groups: those with completion rates at or above
60%, and those with completion rates below 60%.
Differential completion rates did not modify the
pattern of recovery across 4 wk (F=0.4, d.f.=4, 80,
p>0.5) or at 6 wk (F=1.6, d.f.=6, 66, p>0.2). There
was no difference in the recovery curve if the PANSS
(n=12) or the BPRS (n=11) was used as the outcome
measure at 4 wk (F=0.11, d.f.=4, 80, p>0.8) or at 6 wk
(F=0.11, d.f.=6, 72, p>0.8). Baseline BPRS values
were slightly lower than PANSS values, but this did
not affect the recovery pattern.
The present study demonstrated a change in the
time-course of symptom response to antipsychotic
medication from 4 to 6 wk. The shape of the symptom-
response curve is linear from baseline to 4 wk,
representing steady reduction in symptoms. However,
between 4 and 6 wk of treatment, there is a significant
attenuation of treatment response. In contrast, the
placebo treatment response profile is essentially flat
over the full 6-wk interval. These results are based on
a defined group of patients studied during an acute
exacerbation of a primary psychotic disorder and
treated with an atypical medication. Patterns of
symptom change related to active treatment and to
placebo may be different in chronic or stable patients,
particularly if high-dose typical antipsychotic medi-
cations are withdrawn prior to entering studies.
At 4 wk, our results are largely similar to those of
Agid et al. (2003). Despite our extensive exclusion
criteria, designed to allow only acute exacerbation
medication, and requiring multiple data-points, we
included 21 studies and 3552 trial subjects, compared
with 42 studies and 7450 subjects (Agid et al., 2003).
Only eight of the present trials were included in the
Agid et al. meta-analysis (see Table 2). Of their sub-
jects, 34.1% were treated with typical antipsychotics
vs. 62.4% treated with olanzapine or risperidone and
3.5% receiving placebo. While the Agid et al. meta-
analysis showed attenuation of treatment response
between 2 and 4 wk, we found a steady linear decrease
over the full 4 wk, with no attenuation over this
period. This may reflect our exclusion of LOCF data,
although LOCF was found not to be a confounding
factor in the Agid et al. analysis. Moreover, the linear
decrease following the initial weeks was also apparent
in the analysis of the 6-wk data, which included LOCF
results. An additional consideration is our focus on
trials requiring an acute exacerbation of schizo-
phrenia. In the Agid et al. meta-analysis approxi-
mately one-third of the trials of atypical antipsychotics
Symptom change in acute schizophrenia363
included subjects in the chronic rather than the acute
phase of illness. Such subjects could have an earlier
attenuation of response.
At 6 wk, our results indicate a slowing of the treat-
ment response, with attenuation of response following
a linear decline. This result extends our understanding
of the symptom response profile. This is consistent
with the first-episode data of Lieberman et al. (2003),
giving some weight to our assumption that the acute
exacerbation patients may reflect some aspects of the
first-episode patients. This large controlled trial of
first-episode psychosis treated with olanzapine or
haloperidol showed a significant early decline in
symptoms. The authors report different results from
LOCF analysis vs. a random regression coefficients
model, but their observed case results show continued
symptom improvement, albeit at a slower rate, well
past 6 wk. In the present meta-analysis, extending the
length of time of assessment changes the result. Flat-
tening of the curve from 4 to 6 wk cannot be attributed
to the inherent linearization of results by the LOCF
method. Since each included study had at least five
original data-points, the linearization does not result
from averaging of baseline to end-point analyses.
There are several sources of potential bias in the
present study, including possible lack of publication of
negative data and authors’ choice to include weekly
rather than baseline to end-point data. Limitations of
the meta-analysis included a broad definition of what
constitutes an acute exacerbation. Recent work (Davis
et al., 2003) indicates that some atypical agents may be
more efficacious than others; the generally accepted
dose ranges of some atypical agents have changed
since publication of these trials predominantly in the
1990s (APA, 2004). Our approach of condensation of
active treatments may obscure subtle differences in
response based on agent and dose. Future analyses
of larger numbers of studies of individual agents, in-
cluding a range of doses may be fruitful. Concomitant
medication use is a potentially confounding effect,
and a variety of sedatives, anticholinergic agents,
and beta-blockers were permitted in these studies.
Unfortunately, reporting of the use of these drugs is
non-standardized, which precludes analysis. Com-
pletion rates were somewhat low in several of these
studies, however, this did not appear to influence the
results. There are suggestions from the available
studies of different time-response curves for positive
and negative symptom data, and this could be usefully
examined in the future when more complete datasets
may be available. Use of BPRS or PANSS subscale data
would allow separation of reduction in agitation from
Table 2. Comparison of methodology and data between Agid et al. (2003) and the present study
CharacteristicsAgid et al. Present
Number of studies
Number of data-points
Study publication years
Condensed study arms across doses
Number of time-points
Weighted according to number of patients
Studies included in both meta-analyses
2 or more
Percentage change week to week vs. baseline score
(Agid et al. references: 33, 35–37, 41, 46, 56, 66)
Chronic phase of illness (n=7)
(Agid et al. references: 34, 42–44, 57, 69–70)
Typical antipsychotic arms only (n=26)
(Agid et al. references: 31–32, 38–40, 45, 47–55,
58–65, 67–68, 71)
Intramuscular only (n=1)
(Agid et al. reference: 72)
3 or more
Studies from Agid et al. excluded
in present analyses
Atypical arm required
Number of subjects receiving typicals
Number of subjects receiving atypicals
Number of subjects receiving placebo
Number of data-points for typicals
Number of data-points for atypicals
Number of data-points for placebo
364M. Sherwood et al.
more specifically psychotic symptoms. In future, a
more subtle analysis of the time-course of symptom
change could be obtained from individual fitted lines
per research subject. This would be more reflective of
the variation in response compared to an averaged
result, and may identify subpopulations of subjects
with very different rates of response to treatment.
The present findings have implications for timing of
investigations during clinical research, particularly
functional brain imaging of treatment response. From
the perspective of clinical practice, the results suggest
that some response to antipsychotic medication should
be expected within the first several weeks. This is con-
sistent with recent research on typical antipsychotics
suggesting that lack of improvement in positive
symptoms during the first week of treatment with
fluphenazine predicted lack of response at 4 wk
(Correll et al., 2003). Another recent study also sug-
gests reduction of psychotic symptoms in acutely ill
patients can be detected within the first 2–24 h after
intramuscular injection of olanzapine or haloperidol
(Kapur et al., 2005). Careful prospective studies
of atypical antipsychotics to clarify the predictive
value of early symptom reduction for longer term
treatment response may be of value.
Supported by Canadian Institutes of Health Research
grant NET-54013. Dr Robert Conley provided helpful
information concerning a clinical trial, Dr Robert
Zipursky and Dr Shitij Kapur provided preprints
of papers. Janssen and AstraZeneca also provided
clarification on inquiries concerning trials.
Statement of Interest
Dr Honer has served as a consultant for AstraZeneca,
Brainworks Inc., In-silico, Janssen Ortho, and Novartis
AG; and has served on the speakers’ bureau for Eli
Lilly & Company and Pfizer.
Agid O, Kapur S, Arenovich T, Zipursky RB (2003).
Delayed-onset hypothesis of antipsychotic action: a
hypothesis tested and rejected. Archives of General
Psychiatry 60, 1228–1235.
APA (2004). Practice Guideline for the Treatment of Patients
with Schizophrenia. American Journal of Psychiatry
Supplement. American Psychiatric Association.
Barbaran JM (2000). Intraneuronal signalling pathways.
In: Sadock BJ, Sadock VA (Eds.), Kaplan & Sadock’s
Comprehensive Textbook of Psychiatry (7th edn), vol. 1
(pp. 78). Philadelphia: Lippincott Williams & Wilkins.
Beasley Jr. CM, Hamilton SH, Crawford AM, Dellva MA,
Tollefson GD, Tran PV, Blin O, Beuzen JN (1997).
Olanzapine versus haloperidol: acute phase results of
the international double-blind olanzapine trial. European
Neuropsychopharmacology 7, 125–137.
Beasley Jr. CM, Tollefson G, Tran P, Satterlee W, Sanger T,
Hamilton S (1996). Olanzapine versus placebo and
haloperidol: acute phase results of the North American
double-blind olanzapine trial. Neuropsychopharmacology
Blin O, Azorin JM, Bouhours P (1996). Antipsychotic and
anxiolytic properties of risperidone, haloperidol, and
methotrimeprazine in schizophrenic patients. Journal
of Clinical Psychopharmacology 16, 38–44.
Borison RL, Arvanitis LA, Miller BG (1996). ICI 204,636,
and atypical antipsychotic: efficacy and safety in a
multicenter, placebo-controlled trial in patients with
schizophrenia. Journal of Clinical Psychopharmacology 16,
Borison RL, Pathiraja AP, Diamond BI, Meibach RC (1992).
Risperidone: clinical safety and efficacy in schizophrenia.
Psychopharmacology Bulletin 28, 213–218.
Breier A, Schreiber J, Dyer J, Pickar D (1991). National
Institute of Mental Health longitudinal study of
chronic schizophrenia. Archives of General Psychiatry 48,
Carone B, Harrow M, Westermeyer J (1991). Posthospital
course and outcome in schizophrenia. Archives of General
Psychiatry 48, 247–253.
Casey DE, Daniel DG, Wassef AA, Tracy KA, Wozniak P,
Sommerville KW (2003). Effect of divalproex combined
with olanzapine or risperidone in patients with an acute
exacerbation of schizophrenia. Neuropsychopharmacology
Ceskova E, Svestka J (1993). Double-blind comparison of
risperidone and haloperidol in schizophrenic and
schizoaffective psychoses. Pharmacopsychiatry 26, 121–124.
Conley RR, Mahmoud R (2001). A randomized double-
blind study of risperidone and olanzapine in the
treatment of schizophrenia or schizoaffective disorder.
American Journal of Psychiatry 158, 765–774.
Copolov DL, Link CGG, Kowalcyk B (2000). A multicentre,
double-blind, randomized comparison of quetiapine
(ICI 204,636, ‘Seroquel’) and haloperidol in schizophrenia.
Psychological Medicine 30, 95–105.
Correll CU, Malhotra AK, Kaushik S, McMeniman M, Kane
JM (2003). Early prediction of antipsychotic response in
schizophrenia. American Journal of Psychiatry 160,
Daniel DG, Zimbroff DL, Potkin SG, Reeves KR,
Harrigan EP, Lakshminarayanan M (1999). Ziprasidone
80 mg/day and 160 mg/day in the acute exacerbation of
schizophrenia and schizoaffective disorder: a 6-week
placebo-controlled trial. Ziprasidone Study Group.
Neuropsychopharmacology 20, 491–505.
Davis JM, Chen N, Glick I (2003). A meta-analysis of the
efficacy of second-generation antipsychotics. Archives of
General Psychiatry 60, 553–564.
Symptom change in acute schizophrenia 365
Fenton WS, McGlashan TH (1991). The natural history of
schizophrenia subtypes: II. Positive and negative
symptoms and long term course. Archives of General
Psychiatry 48, 978–986.
Gelder M, Mayou R, Cowen P (2001). Shorter Oxford Textbook
of Psychiatry. Oxford: Oxford University Press.
Gueorguieva R, Krystal JH (2004). Progress in analyzing
repeated-measures data and its reflection in papers
published in the Archives of GeneralPsychiatry. Archives of
General Psychiatry 61, 310–317.
Heinrichs RW, Awad AG (1993). Neurocognitive sub-
types of chronic schizophrenia. Schizophrenia Research
Hirsch SR, Link CG, Goldstein JM, Arvanitis LA (1996).
ICI 204,636: a new atypical antipsychotic drug. British
Journal of Psychiatry 168 (S29), 45–56.
Hoyberg OJ, Fensbo C, Remvig J, Lingjaerde O, Sloth-
Nielsen M, Salvesen I (1993). Risperidone versus
perphenazine in the treatment of chronic schizophrenic
patients with acute exacerbations. Acta Psychiatrica
Scandinavica 88, 395–402.
Kane JM, Carson WH, Saha AR, McQuade RM, Ingenito
GG, Zimbroff DL, Ali MW (2002). Efficacy and safety
of aripiprazole and haloperidol versus placebo in patients
with schizophrenia and schizoaffective disorder.
Journal of Clinical Psychiatry 63, 763–771.
Kapur S, Arenovich T, Agid O, Zipursky R, Lindborg S,
Jones B (2005). Immediate onset of antipsychotic
effect: the early onset hypothesis. American Journal of
Psychiatry 102, 939–946.
Keck PE, Cohen BM, Baldessarini RJ, McElroy SL (1989).
Time course of antipsychotic effects of neuroleptic drugs.
American Journal of Psychiatry 146, 1289–1292.
Keck PJ, Buffenstien A, Ferguson J, Feighner J, Jaffe W,
Harrigan EP, Morrissey MR (1998). Ziprasidone 40 and
120 mg/day in the acute exacerbation of schizophrenia and
schizoaffective disorder: a 4-week placebo-controlled trial.
Psychopharmacology 140, 173–184.
King DK, Link CG, Kowalcyk B (1998). A comparison of
bd and tid dose regimens of quetiapine (Seroquel) in the
treatment of schizophrenia. Psychopharmacology 137,
Klieser E, Lehmann E, Kinzler E, Wurthmann C, Heinrich K
(1995). Randomized, double-blind, controlled trial of
risperidone versus clozapine in patients with chronic
schizophrenia. Journal of Clinical Psychopharmacology 15,
Lieberman JA, Tollefson GD, Tohen M, Green AI, Gur RE,
Kahn R, McEvoy J, Perkins D, Sharma T, Zipursky RB,
Wei H, Hamer R (2003). Comparative efficacy and safety of
atypical and conventional antipsychotic drugs in first-
episode paychosis: a randomized, double-blind trial of
olanzapine versus haloperidol. American Journal of
Psychiatry 160, 1396–1404.
Marder SR (2000). Schizophrenia: Somatic treatment. In:
Sadock BJ, Sadock VA (Eds.), Kaplan & Sadock’s
Comprehensive Textbook of Psychiatry (7th edn), vol. 1
(pp. 1208). Philadelphia: Lippincott Williams & Wilkins.
Martin S, Loo H, Peuskens J, Thirumalai S, Giudicelli A,
Fleurot O, Rein W (2002). A double-blind, randomised
comparative trial of amisulpiride versus olanzapine in
the treatment of schizophrenia. Current Medical Research
and Opinion 18, 355–362.
Miyamato S, Lieberman JA, Fleishhacker WW, Aoba A,
Marder SR (2003). Therapeutics. In: Tasman A, Kay J,
Lieberman JA (Eds.), Psychiatry (2nd edn) (pp. 2552).
London: John Wiley & Sons.
Muller N, Riedel M, Scheppach C, Brandstatter B,
Sokullu S, Krampe K, Ulmschneider M, Engel RR,
Moller HJ, Schwarz MJ (2002). Beneficial antipsychotic
effects of celecoxib add-on therapy compared to
risperidone alone in schizophrenia. American Journal
of Psychiatry 159, 1029–1034.
Peuskens J, Bech P, Moller HJ, Bale R, Fleurot O, Rein W
(1999). Amisulpride vs. risperidone in the treatment of
acute exacerbations of schizophrenia. Psychiatry Research
Plesnicar BK, Zalar B, Tomori M, Kranjc I (2003).
Measurement of simple reaction time in antipsychotic
treatment of patients with schizophrenia. Middle European
Journal of Medicine 115, 59–62.
Potkin SG, Saha AR, Kujawa MJ, Carson WH, Ali M,
Stock E, Stringfellow J, Ingenito G, Marder SR (2003).
Aripiprazole, an antipsychotic with a novel mechanism
of action, and risperidone vs placebo in patients with
schizophrenia and schizoaffective disorder. Archives of
General Psychiatry 60, 681–690.
Sechter D, Peuskens J, Fleurot O, Rein W, Lecrubier Y
(2002). Amisulpiride vs. risperidone in chronic
schizophrenia: results of a 6-month double-blind study.
Neuropsychopharmacology 27, 1071–1081.
Small JG, Hirsch SR, Arvanitis LA, Miller BG, Link CGG
(1997). Quetiapine in patients with schizophrenia.
Archives of General Psychiatry 54, 549–557.
366M. Sherwood et al.