Академический Документы
Профессиональный Документы
Культура Документы
ORIGINAL ARTICLE
C-reactive protein is increased in schizophrenia but is not
altered by antipsychotics: meta-analysis and implications
BS Fernandes1,2, J Steiner3, H-G Bernstein3, S Dodd1,4, JA Pasco1,5, OM Dean1,4, P Nardin2, C-A Gonçalves2 and M Berk1,4
The inflammatory hypothesis of schizophrenia (SZ) posits that inflammatory processes and neural–immune interactions are
involved in its pathogenesis, and may underpin some of its neurobiological correlates. SZ is the psychiatric disorder causing the
most severe burden of illness, not just owing to its psychiatric impairment, but also owing to its significant medical comorbidity.
C-reactive protein (CRP) is a commonly used biomarker of systemic inflammation worldwide. There are some conflicting results
regarding the behaviour of CRP in SZ. The aims of this study were to verify whether peripheral CRP levels are indeed increased in
SZ, whether different classes of antipsychotics divergently modulate CRP levels and whether its levels are correlated with positive
and negative symptomatology. With that in mind, we performed a meta-analysis of all cross-sectional studies of serum and plasma
CRP levels in SZ compared to healthy subjects. In addition, we evaluated longitudinal studies on CRP levels before and after
antipsychotic use. Our meta-analyses of CRP in SZ included a total of 26 cross-sectional or longitudinal studies comprising 85 000
participants. CRP levels were moderately increased in persons with SZ regardless of the use of antipsychotics and did not change
between the first episode of psychosis and with progression of SZ (g = 0.66, 95% confidence interval (95% CI) 0.43 to 0.88,
P o 0.001, 24 between-group comparisons, n = 82 962). The extent of the increase in peripheral CRP levels paralleled the increase in
severity of positive symptoms, but was unrelated to the severity of negative symptoms. CRP levels were also aligned with an
increased body mass index. Conversely, higher age correlated with a smaller difference in CRP levels between persons with SZ and
controls. Furthermore, CRP levels did not increase after initiation of antipsychotic medication notwithstanding whether these were
typical or atypical antipsychotics (g = 0.01, 95% CI − 0.20 to 0.22, P = 0.803, 8 within-group comparisons, n = 713). In summary, our
study provides further evidence of the inflammatory hypothesis of SZ. Whether there is a causal relationship between higher CRP
levels and the development of SZ and aggravation of psychotic symptoms, or whether they are solely a marker of systemic low-
grade inflammation in SZ, remains to be clarified.
Molecular Psychiatry (2016) 21, 554–564; doi:10.1038/mp.2015.87; published online 14 July 2015
1
IMPACT Strategic Research Centre, Barwon Health, Deakin University, School of Medicine, Geelong, VIC, Australia; 2Laboratory of Calcium Binding Proteins in the Central Nervous
System, Department of Biochemistry, Federal University of Rio Grande do Sul, UFRGS, Porto Alegre, Brazil; 3Department of Psychiatry, University of Magdeburg, Magdeburg,
Germany; 4Department of Psychiatry and Orygen, Florey Institute for Neuroscience and Mental Health, The National Centre of Excellence in Youth Mental Health, University of
Melbourne, Parkville, VIC, Australia and 5Department of Medicine, NorthWest Academic Centre, University of Melbourne, St Albans, VIC, Australia. Correspondence:
Dr BS Fernandes, IMPACT Strategic Research Centre, Barwon Health, Deakin University, School of Medicine, Geelong, VIC, Australia.
E-mail: brisasf@gmail.com
Received 10 March 2015; revised 1 May 2015; accepted 26 May 2015; published online 14 July 2015
C-reactive protein in schizophrenia
BS Fernandes et al
555
Although much attention has been paid to the study of cyto- modulate CRP levels, and fourthly whether its levels are correlated
kines, other biomarkers of inflammation, in particular C-reactive with positive and negative symptomatology and obesity. Lastly,
protein (CRP), could also play a role in the context of SZ. CRP we wanted to find out whether CRP levels change following
possesses specific advantages over cytokines as biomarkers. CRP is antipsychotic therapy, indicating either potential harmful or
one of the most commonly used biomarkers of inflammation in protective metabolic effects of antipsychotic medication regard-
clinical practice. It is not just used as a measurement of systemic ing inflammation. With this in mind, we performed a meta-
inflammation mirroring inflammatory conditions in studies analysis of all cross-sectional studies of serum and plasma CRP
devoted to unveil altered pathophysiological states; it is also the levels in SZ compared to healthy subjects. In addition, we
most employed laboratory test in clinical settings worldwide, to evaluated longitudinal studies on CRP levels before and after
assess systemic, peripheral inflammatory status. High-sensitivity antipsychotic use. This will help to clarify the role of CRP as a
CRP is the most sensitive assay for assessing CRP levels.12 Usually, biomarker of low-grade inflammation in SZ, ultimately clarifying
in apparently healthy subjects, CRP levels are lower than 3 mg l − 1, the role of CRP as an underpinning factor involved in the
but slightly increased levels can occur in normal individuals. These inflammatory hypothesis.
indicate a state of low-grade inflammation, which is frequently
associated with obesity and metabolic syndrome, and predicts an
increased risk of depression13 and cardiovascular diseases— MATERIALS AND METHODS
common comorbidities in persons with SZ.14 CRP is mainly This study is comprised of a between-group meta-analysis comparing
produced by hepatocytes and is directly modulated by IL-6 and serum and plasma CRP levels in persons with SZ and healthy controls, and
IL-1β, both of which appear to be increased particularly during of two within-group meta-analyses of longitudinal studies of CRP levels
exacerbations of psychotic status.11,15,16 This raises the possibility before and after initiation or change of class of antipsychotics. We adhered
that CRP, in addition to inflammatory cytokines, may play a role in to the Preferred Reporting Items for Systematic Reviews and Meta-
Analyses statement and by the Cochrane Collaboration.43 The literature
the inflammatory hypothesis of SZ and therefore be involved in its search, decisions on inclusion, data extraction and quality control were all
pathophysiology in a similar manner. performed independently by two of the authors (BSF and CAG).
Several studies have explored CRP in SZ; however, the results
are inconsistent, with approximately half of the studies reporting
an association between CRP levels and SZ, whereas the other Search strategy
half reported no association.14,17–41 Furthermore, the results are We conducted a systematic search for all possibly eligible English and non-
constrained by diverging methodological characteristics among English peer-reviewed articles to avoid language publication bias44,45 using
Medline, the Cochrane Library, Scielo, Scopus and Web of Knowledge. No
the different reports, such as small sample sizes and presence of
year or country restrictions were used. The Boolean search term used for
systemic clinical diseases (for instance, diabetes and atherosclero- the electronic database search was (CRP OR C-reactive protein OR hsCRP
sis). It is also controversial whether CRP is increased in SZ per se OR hs-CRP) and (schizophrenia OR psychosis OR schizophreniform OR
or whether it is merely a marker of the associated systemic schizoaffective). The last search was performed in January 2015. We then
inflammation commonly found in this disorder, due to increased manually checked the reference sections of the publications found
prevalence of obesity and metabolic syndrome, among others. through our electronic search, to identify additional studies that may
Importantly, it is debatable whether CRP levels are correlated have been missed. Study selection eligibility and exclusion criteria were
with severity of positive and negative symptoms in SZ, which is pre-specified.
paramount to reveal a relationship between CRP and the
pathophysiology of SZ. Study selection
To make the field even more confusing, many antipsychotics, Inclusion criteria were as follows: (1) adult subjects with SZ, schizophreni-
the first and sole line of medication to treat psychosis, are form, or schizoaffective disorder, as defined by the American Psychiatric
habitually responsible for obesity, consequent insulin resistance Association, which from now on will collectively be referred to as SZ for the
and metabolic syndrome, possibly contributing to the increased sake of simplicity; (2) pairwise comparison with a control group of healthy
cardiovascular mortality in this population.22,40 Once more, the volunteers for the between-group meta-analysis or pairwise comparisons
body of evidence here is contradictory, with some longitudinal before and after antipsychotics in longitudinal studies for the within-group
clinical studies showing worsening of metabolic syndrome and, meta-analyses; and (3) studies assessing circulating serum or plasma CRP
or high-sensitivity CRP in human blood samples in vivo. Exclusion criteria
ultimately, increased levels of CRP after initiation of antipsychotics,
were as follows: (1) duplicate reports; (2) studies that employed CRP assays
some showing no change and some showing decreased levels of that provide dichotomous results (‘positive‘ or ‘negative‘); and (3) lack of a
CRP after initiation of these medications.20–23,26,27,32,37,40 On the control group for the between-group meta-analysis, or lack of follow-up for
other hand, animal studies suggest that antipsychotics may the within-group meta-analyses (Figure 1). The decision of whether to
decrease IL-6 and, as a theoretical consequence, negatively include studies in the meta-analyses was made based on the above
modulate CRP.42,43 criteria, and a consensus was reached among the authors on those
Inconsistent findings among the studies might be caused by decisions.
heterogeneous patient populations, variations in methodology or
small sample sizes lacking statistical power. Meta-analysis is a Data extraction
recognized technique used to resolve discrepancies between To avoid potential errors, two reviewers (BSF and CAG) inde-
studies. It is a quantitative method that combines results from pendently extracted data (n, mean and s.d.). The following data were
independent studies to increase statistical power in order to extracted: diagnostic status, CRP levels, sex, age, duration of illness, body
derive more solid conclusions. This allows one to distinguish small mass index (BMI), type of antipsychotics, Positive and Negative Symptoms
effects from no effect. It also helps to determine whether the Scale (PANSS) total scores and its positive and negative sub-scores.46
variation in effects between studies is merely due to the expected According to information available in the studies, sub-group analysis
random statistical fluctuation or instead to sample variations or according to antipsychotic status (that is, drug-naive or -free, or on
trait assessment. In addition, the technique of meta-regression antipsychotics) was performed. Subjects with SZ were considered drug-
free if they were off psychiatric medication before blood sampling for at
may be used to evaluate confounders and discrepancies among
least 2 weeks. Discrepancies in data entry were double-checked by the
different studies.43 reviewers with the original published data and consensus was reached.
The aims of this study were therefore firstly to verify if When the necessary data were not available from the published paper, we
peripheral CRP levels are indeed increased in SZ, secondly if the contacted the authors and requested the necessary information. If the
increases occur since the occurrence of the first episode of results were graphically presented and the authors could not provide the
psychosis, thirdly if different classes of antipsychotics divergently data, we used the method for data extraction from graphs by Sistrom
Studies excluded: 22
Duplicate report
No control group
CRP assessed before onset of disease
Figure 1. Flow diagram of the systematic review showing the study inclusion and exclusion process. Some studies included subjects in more
than one mood state, being included in more than one analysis.
et al.47 Whenever multiple reports pertained to the same groups of Statistical analysis
patients, we retained only the most comprehensive report for the meta- Comprehensive Meta-analysis Software version 2.0 (CMA, Borenstein, USA)
analytic calculations to avoid duplication of information.43 was employed in all analyses. As studies used different measurement
methods, standardized mean difference estimates of the differences in CRP
levels between subjects with SZ and healthy controls were used as the ES,
Quality assessment using Hedges’ adjusted g. This provides an unbiased ES adjusted for small
We used the Newcastle–Ottawa Scale (NOS)48 as recommended by the sample sizes. The 95% confidence interval (95% CI) of the ES was
Cochrane Collaboration,43 to assess the quality of the eligible observational also computed. An ES of 0.2 is considered as indicating a small effect,
studies. For the between-group and within-group meta-analyses, we used meaning a small difference in CRP levels between subjects with SZ and
controls, of 0.5 indicating a moderate effect, and of 0.8 indicating a
the NOS scale for case–control and for cohort studies, respectively. Overall
large effect.43
quality score was defined as the frequency of criteria that were met by the
We assessed the heterogeneity across studies using the Cochran Q
particular study. The NOS scale contains eight items in the case–control test,50 a weighted sum of the squares of the deviations of individual study
section and nine items in the cohort section, categorized into the three ESs estimates from the overall estimate, and a P-value of o0.10 was
domains of selection, comparability, exposure and outcome (the last only considered significant (that is, showing heterogeneity). Inconsistency
for the cohort scale). A series of response options is provided for each item. across studies was then quantified with the I2 metric.45,51 It can be
A star system was used to enable semi-quantitative assessment of study interpreted as the percentage of total variation across several studies due
quality, such that the highest quality studies are awarded a maximum of to heterogeneity, and it is considered moderate when between 50% and
one star for each item, with the exception of the comparability domain, 75% and high when > 75%. As the analyses showed that the studies were
which allows the assignment of two stars. As such, the NOS scale ranges heterogeneous, we pooled ES results from different studies according to
between 0 and 9 stars.48 Item ‘non-response rate’ from Exposure in the the inverse variance method of accounting for random effects, which
case–control scale was not applicable; therefore, a maximum of eight stars allows population level inferences and is more stringent than the fixed-
was considered. For the cohort scale we considered the maximum as nine effect models.52. Random-effect modelling assumes a genuine diversity in
stars. The quality score of the included studies ranged from 1 to 7. All the results of various studies and incorporates a between-study variance
studies were included in the posterior analyses. into the calculations, and is considered the best modelling in the presence
of heterogeneity. The direction of the ES was positive if subjects with SZ
showed increased CRP and negative if they showed decreased CRP levels
when compared with controls for the between-group meta-analysis, and
Publication bias
positive if it increased after the use of antipsychotics for the within-group
Studies with negative results are less likely to be published than studies meta-analyses.
with positive results. To account for significant publication bias, we Unrestricted maximum likelihood random-effects meta-regressions52 of
analyzed a funnel plot graph, a scatter plot of the effect size (ES) against a ESs were performed with mean age, length of follow-up, BMI, NOS score
measure of study size, and the Egger test.45 In addition, the Orwin’s fail-safe and severity of disease as assessed by PANSS as moderators, to determine
N-test (file drawer statistic)49 was used to quantify the number of possible whether these covariates influenced the ES. Studies were weighted such
negative omitted studies that would be required to make our results that the studies with the most precise parameters, quantified by the
nonsignificant (P40.05). sample size and 95% CI, had more influence in the regression analyses.43
Figure 2. Forest plot for random effects between-group meta-analysis on serum and plasma C-reactive protein (CRP) levels in persons with
schizophrenia (SZ) and healthy controls. The sizes of the squares are proportional to sample size. A total of 19 studies were included,
comprising 1995 subjects with SZ and 80 967 healthy controls.
on olanzapine with aripiprazole as an add-on therapy.32 The to the source of the sample, that is, population-based or clinical
follow-up varied from 4 to 52 weeks and the antipsychotic studies, heterogeneity was mostly high in the clinical studies and
treatment initiated included haloperidol, clozapine, risperidone, low in the population-based studies. In all other subgroup
olanzapine, aripiprazole, or a combination of typical and atypical analyses, the heterogeneity remained high and could not be
antipsychotics. The mean BMI of the study participants was 26.65 explained by any single variable (Table 1). Thereafter, we
at baseline and 27.63 at the end of follow-up (P = 0.901) conducted a sensitivity analysis in all sub-group meta-analyses
(Supplementary Table S2). Overall, initiation of antipsychotics excluding studies one at a time to determine the robustness of the
was not associated with an increase in CRP levels (g = 0.01, 95% CI analyses and to verify whether a particular study was responsible
− 0.20 to 0.22, P = 0.803, 8 within-group comparisons, n = 713). for the high heterogeneity. No single study thoroughly explained
When we analysed the changes of CRP levels according the type the heterogeneity and the results remained significant in all cases,
of antipsychotic initiated, we verified that our results remained even when the the larger population study conducted by Wium-
nonsignificant regardless of whether the antipsychotic was typical, Andersen et al.14 were excluded.
atypical, or a combination of both (Figure 3 and Table 1). We ran the analysis including only the 17 studies that employed
Furthermore, in the meta-regressions we verified that neither age, a highly sensitive CRP assay and the results remained significant
weeks of follow-up nor BMI at the baseline influenced these with a moderate ES (g = 0.67, 95% CI 0.43 to 0.90, P o0.001, 22
results (see Table 1 for details). between-group comparisons, n = 82 744) and the heterogeneity
In addition, we analysed if the change of type of antipsychotics did not change. In addition, CRP levels were equally increased in
altered CRP levels. Three studies22,26,40 comprising 12 pairwise serum and plasma (Table 1).
comparisons with a total of 1325 persons with SZ were included. In the within-group meta-analyses of CRP alterations after the
Again, change of antipsychotics did not produce a change in CRP initiation of antipsychotics and after a change of antipsychotics,
levels (g = − 0.03, 95% CI − 0.11 to 0.06, P = 0.059, 12 within-group the results remained nonsignificant in the sensitivity analysis
comparisons, n = 1325), irrespective of whether the change was to when one study at a time was excluded, again showing that these
a typical or atypical antipsychotic (Figure 4 and Table 1). Once negative results were not due to a particular study.
more, the meta-regressions showed that age, weeks of follow-up
and BMI at the baseline did not influence these results (Table 1). Publication bias
We found no evidence of publication bias in the funnel plot of the
Sensitivity analyses between-group meta-analysis and the Egger’s test was nonsigni-
In the between-group meta-analysis, the heterogeneity for the ficant (P = 0.11). Moreover, the Orwin’s fail-safe N-test yielded 5564
whole analysis was high. When we analysed the studies according studies as the number of negative studies necessary to turn our
All 24 1,995 80,967 0.60 0.43 0.88 0.001 91.14 259.68 0.001
Drug-naïve or free 6 348 360 0.87 0.14 1.60 0.021 94.99 99.78 0.001
All antipsychotics 18 1,647 80,607 0.58 0.36 0.79 0.001 87.42 135.22 0.001
Typical antipsychotics 2 237 177 0.87 0.67 1.08 0.001 0.00 0.08 0.928
Atypical antipsychotics 7 346 1020 0.53 0.14 0.93 0.008 76.99 26.08 0.001
Typical/atypical antipsychotics 9 1064 79,410 0.52 0.27 0.76 0.001 85.34 54.56 0.001
First episode SZ 6 348 360 0.63 0.03 1.22 0.038 86.99 30.74 0.001
Chronic SZ 18 1,647 80,607 0.76 0.43 1.08 0.001 94.93 256.67 0.001
Populational samples 4 473 79,033 0.18 0.06 0.30 0.003 14.96 3.53 0.317
Clinical samples 20 1,934 3,456 0.77 0.52 1.02 0.001 88.43 164.29 0.001
Age and BMI paired 14 1,096 79,599 0.51 0.43 0.59 0.001 93.15 189.98 0.001
hsCRP only 22 1,842 80,902 0.67 0.43 0.90 0.001 91.53 248.06 0.001
Without clinical conditions 17 1,227 961 0.78 0.69 0.87 0.001 88.42 138.19 0.001
With clinical conditions 7 768 79,736 0.45 0.36 0.54 0.001 93.62 94.14 0.001
Serum 18 1,344 1,053 0.65 0.38 0.91 0.001 87.97 141.40 0.001
Plasma 6 651 80,565 0.69 0.22 1.16 0.004 95.57 112.91 0.001
Within-Group SZ before and after No. of pairwise SZ sample size Hedges’ g 95% CI P I2 Q P
start of antipsychotics
BMI (all*) 23 1,892 80,952 0.21 0.04 0.29 0.004 − 1.16 − 0.83 0.401
BMI (drug-naïve or free*) 6 348 360 0.29 0.16 0.38 0.001 − 6.41 − 5.78 0.001
BMI (on all antipsychotics*) 17 1,544 80,592 0.11 0.03 0.20 0.009 − 2.53 − 2.14 0.032
WHR (all) 5 490 445 3.18 − 15.98 22.35 0.744 − 1.65 − 0.18 0.852
Waist circumference (all) 10 558 511 0.01 − 0.07 0.09 0.867 0.22 0.05 0.954
Age (all) 24 1,995 80,967 − 0.02 − 0.04 0.01 0.169 1.34 2.62 0.008
Age (drug-naïve or free*) 6 348 360 − 0.20 − 0.34 -0.06 0.006 6.84 3.12 0.001
Age (on all antipsychotics) 18 1,647 80,607 − 0.01 − 0.03 0.01 0.558 0.84 1.78 0.074
Age of onset (all) 10 639 592 0.05 − 0.01 0.12 0.087 − 0.76 − 0.90 0.365
Length of illness (all) 10 639 592 − 0.03 − 0.09 0.02 0.219 1.01 2.93 0.003
PANSS Total Scores (all) 12 1,047 920 − 0.01 − 0.03 0.02 0.750 0.99 1.27 0.201
PANSS Positive Scores (all*) 7 401 349 0.12 0.03 0.23 0.013 − 1.25 − 1.48 0.137
PANSS Negative Scores (all) 7 401 349 0.04 − 0.10 0.20 0.533 − 0.08 − 0.06 0.947
Sample size (all) 23 1,892 80,952 − 0.01 − 0.02 0.01 0.392 0.68 5.51 0.001
NOS (all) 23 1,892 80,952 − 0.02 − 0.04 0.16 0.189 1.39 1.89 0.392
Within-Group SZ before and after No. of pairwise SZ sample size Slope 95% CI P Intercept Z P
start of antipsychotics
Length of follow-up in weeks 8 713 − 0.01 − 0.02 0.01 0.583 0.04 0.37 0.708
Age 8 713 0.01 − 0.02 0.02 0.910 − 0.03 − 0.09 0.923
BMI in the baseline 8 602 − 0.06 − 0.14 0.02 0.144 1.58 1.42 0.154
Within-Group SZ before and No. of pairwise SZ sample size Slope 95% CI P Intercept Z P
after change of antipsychotics
Length of follow-up in weeks 12 1,325 -0.01 − 0.02 0.02 0.612 0.02 0.45 0.403
Age 12 1,325 0.02 − 0.03 0.03 0.485 − 0.03 − 0.12 0.763
BMI in the baseline 12 1,325 -0.12 − 0.19 0.12 0.319 1.93 1.95 0.261
Abbreviations: BMI, body-mass index in kg/m2; CI, confidence interval; HC, healthy control; hsCRP, high-sensitivity C-reactive protein; NOS, Newcastle-
Ottawa Scale; PANSS, Positive and Negative Symptoms Scale; SZ, schizophrenia; WHR, waist-hip ratio. *Po0.05.
Typical
Diaz et al., 2010a 0.43 0.053
0.01 0.983
Atypical
Diaz et al., 2010b -0.12 0.521
0.32 0.494
Typical/Atypical
0.01 0.849
Figure 3. Forest plot for random effects within-group meta-analysis on serum and plasma C-reactive protein (CRP) level alterations after
initiation of antipsychotics in persons with schizophrenia (SZ). The sizes of the squares are proportional to sample size. A total of 6 studies
were included, comprising 713 subjects with SZ.
Study g p
Typical
-0.08 0.312
0.06 0.058
-0.17 0.134
Figure 4. Forest plot for random effects within-group meta-analysis on serum and plasma C-reactive protein (CRP) levels alterations after
change of type antipsychotics in persons with schizophrenia (SZ). The sizes of the squares are proportional to sample size. A total of 3 studies
were included, comprising 1325 subjects with SZ.
Supplementary Information accompanies the paper on the Molecular Psychiatry website (http://www.nature.com/mp)