Increased C‐reactive protein concentration and suicidal behavior in people with psychiatric disorders: A systematic review and meta‐analysis

Abstract Objective Suicide is a leading cause of death worldwide. Identifying factors associated with suicidality (suicidal ideation [SI]/suicidal behavior) could increase our understanding of the pathophysiological underpinnings of suicide and improve its prevention. Methods We conducted a systematic review (PubMed/PsycInfo/Cochrane databases, up to September 2020) and random‐effect meta‐analysis including observational studies comparing peripheral C‐reactive protein (CRP) levels in suicidal versus non‐suicidal patients affected by any psychiatric disorder and healthy controls (HC). Primary outcome was the CRP standardized mean difference (SMD) between patients with high suicidality versus those with absent or low suicidality. Secondary outcomes were SMD of CRP levels between those with suicide attempt versus no suicide attempt, as well as between those with (high) versus low or absent SI. Quality of included studies was measured with Newcastle‐Ottawa scale. Results Out of initial 550 references, 21 observational studies involving 7682 subjects (7445 with mood disorders or first‐episode psychosis, 237 HC) were included. A significant association of CRP levels with suicidality (SMD 0.688, 95% CI 0.476–0.9, p < 0.001) emerged. CRP levels were higher in individuals with high SI (SMD 1.145, 95% CI 0.273–2.018, p = 0.010) and in those with suicide attempt (SMD 0.549, 95%CI 0.363–0.735, p < 0.001) than non‐suicidal individuals (either patients or HC). Main analyses were confirmed in sensitivity analysis (removing HC), and after adjusting for publication bias. The cross‐sectional design of included studies, and the high heterogeneity of diagnosis and treatment limit the generalizability of these results. Median quality of included studies was high. Conclusion CRP is associated with higher suicidality in patients with mental disorders. Large cohort studies longitudinally monitoring CRP levels are needed to explore its longitudinal association with suicidality.


| INTRODUCTION
Suicide and suicide-related behaviors are a public health problem. Each year, approximately 800 000 people die due to suicide, while it is estimated that the number of suicide attempts is over twenty times higher. 1 Of those who die by suicide, over 90% suffer from a mental illness, most commonly major depressive disorder (MDD) in up to two-thirds of patients. 2,3 Risk of suicide is particularly high among patients diagnosed with severe depression or bipolar disorder (BD), especially when there are concomitant mixed features or agitation, or cooccurring substance use disorders. 4 Indeed, BD carries a high suicide risk; it has been estimated that around 15-20% of patients with BD die by suicide. [5][6][7] Moreover, suicidal behaviors (SBs) occur at a significantly greater rate in schizophrenia than in the general population, with a lifetime suicide rate in individuals with schizophrenia approximately of 10%. 8 There are multiple factors and potential mechanisms that contribute to the complexity of SBs and suicidal risk in these mental disorders (depressive, BD, schizophrenia). 9 Previous studies have focused on SB, suicidal ideation (SI), and non-lethal suicidal attempt (SA) as a target. Such targets are promising for suicide-related research since they strongly relate to a suicide outcome and are more frequent than suicide, with prevalence rates of 9.2% and 2.7% for SI and a non-lethal SA, respectively. 10 In addition, 29% of individuals with lifetime SI attempt suicide, suggesting that SI and SA map on a continuum of suicide risk. 10,11 Therefore, in order to address the high suicide rate worldwide, identification of contributors and markers of both SI and SA could be crucial in clinical practice.
The assessment of suicide risk based exclusively on the patient's clinical history has low specificity. 12 In addition, previous reviews have examined the predictive validity of suicide assessment instruments, demonstrating poor performance in the prediction of subsequent suicide attempt and suicide. [13][14][15][16] A complex combination of psychosocial, biological, cultural, and environmental factors can result in SB. 17,18 While none of these factors can reliably predict suicide alone, 19 suggestive evidence indicated that some biological markers are possibly related to increased SBs, including activation of immune-inflammatory pathways. [20][21][22][23][24] Two previous metaanalyses revealed that inflammatory markers' levels were increased in suicidal patients compared with healthy controls (HC), suggesting a possible important role of immuneinflammatory pathways in SB pathogenesis. 20,21 More in detail, a possible pathway implicating immune-inflammatory processes is through the release of cytokines and their consequences, which have neurotoxic effects, lead to a breakdown of the blood-brain barrier, thereby allowing activated immune cells and their products to influence brain functions. 25 C-reactive protein (CRP), a positive acute phase protein, an inflammatory protein synthesized by Kupffer cells in the liver in response to increases in pro-inflammatory cytokines Conclusion: CRP is associated with higher suicidality in patients with mental disorders. Large cohort studies longitudinally monitoring CRP levels are needed to explore its longitudinal association with suicidality.

K E Y W O R D S
C-reactive protein, inflammation, prevention, psychiatry, suicidal attempt, suicidal behavior, suicidal ideation, suicidality

Summations
• CRP levels are increased in patients affected by severe mental illness with elevated suicidality, after pooling data from 21 studies and 7682 subjects. • Specifically, our study provides evidence of a large (SMD 1.145, 95% CI 0.273-2.018) association between CRP concentrations and suicidal ideation, as well as a medium (SMD 0.549, 95% CI 0.363-0.735) association with suicide attempt.

Limitations
• The observational and cross-sectional design with small to moderate sample sizes of included studies, and the partial adjustment for clinical confounders limit the generalizability of these results.
including IL-6, IL-1, and tumor necrosis factor (TNF)-α, [26][27][28] is frequently employed in clinical and translational research given its detectability at lower levels using high-sensitivity assays. 29,30 It is easily detected both in serum and in plasma samples, 31,32 hence widely used in clinical practice to measure the presence and severity of inflammation. 33,34 While activated immune-inflammatory pathways, which imply CRP elevation, have been reported to significantly contribute to the pathophysiology (and incorporated in staging models) of mood disorders, including MDD and BD, as well as to the cognitive and symptomatic features of schizophrenia, and SBs, 20,[35][36][37][38] to date contradictory results emerged from studies focusing on CRP's role in patients with SBs. While a few studies with relatively small sample size showed no significant associations between CRP and SB, [39][40][41] other studies revealed statistically significant association in patients with mental disorders. 22,[42][43][44][45][46] In some of these studies, increased CRP and pro-inflammatory cytokines have been described in individuals following recent 43,47 and remote 42 suicide attempts.
To the best of our knowledge, so far only one metaanalysis pooled data on the association between CRP and suicidality, only focusing on people with depressive disorder. 48 It remains unclear whether CRP is associated with suicidality beyond those with depressive disorder, whether the association is valid with both suicide ideation and SB, and to what extent.

| AIMS OF THE STUDY
The aim of the present systematic review and metaanalysis is to pool data from observational studies reporting on CRP and suicidality (ie, SB/SI), without any restrictions in terms of underlying diagnosis.

| Search
A systematic review was conducted by database inception, last search on September 3, 2020, via Ovid platform in PubMed/PsycInfo/Cochrane database. Search key was "(CRP OR C-reactive protein OR hsCRP OR hs-CRP) AND (suicide OR SB OR SA OR suicidal thoughts OR Self-Mutilation OR suicide* OR SI OR commit suicide OR suicidality)." In addition, a manual search was conducted, screening the reference list of included trials and relevant reviews. Two authors independently (TT and VDP) screened the title and abstract of the initial list of articles and then collected the full text of potentially eligible ones, further assessing eligibility. The reason for the exclusion of studies was also recorded. If any disagreement emerged at this stage, it was solved by a third author (AM).
Observational studies comparing peripheral CRP concentrations between suicidal and non-suicidal patients, and between suicidal patients and HCs were included, based on the following criteria: (1) Patients could have any psychiatric disorder defined according to standardized diagnostic criteria (ie, DSM, or ICD-any version); (2) the study had to report on suicidality, defined as either SI, suicide attempt, or suicide; and (3) the study had to report CRP concentrations in peripheral sample. Excluded were studies not focusing on the patients with any psychiatric disorder and focusing on the general population, or not reporting CRP levels. We also excluded case reports, or case series, and reviews. We also excluded studies that compared CRP levels in subjects with medical disorders known to increase CRP levels. We did not apply any language restrictions.

| Primary and secondary outcomes
The primary outcome was CRP concentration in subjects with suicidality (either SI or SB) versus those without or with lowest suicidality.
Secondary outcomes were CRP levels in those with suicide attempt versus no suicide attempt, as well as in those individuals displayed highest versus without or lowest SI.

| Data extraction
Two authors (TT and VDP) independently extracted data with a third author (AM) solving persistent disagreements and making the final decision. The following variables were extracted into a predefined excel spreadsheet: author, year, country, underlying mental condition, population setting, age group, sample size, diagnosis, exclusion criteria, PCR concentrations, assay methods, outcomes, and funding.
Mean and standard deviation of baseline, endpoint, or change values of primary and secondary outcomes were extracted according to their availability in the published eligible studies. When such values were not reported in text or tables, but only in figures, data were extracted from figures via an online platform (https://autom eris.io/ WebPl otDig itize r/).

| Quality assessment
Two authors (AM and VDP) independently assessed the quality of the included studies with the Newcastle-Ottawa Scale (NOS), with an average score ≥7 (out of nine) indicating high quality. 49

| Statistical analyses
A PRISMA-compliant 50 systematic review and randomeffect 51 meta-analysis was conducted, when at least two studies reported the same outcome. When more than one outcome was reported in one study from the same sample, the mean of the effect sizes within each individual study was considered to avoid double counting and artificial narrowing of confidence intervals. Heterogeneity was assessed with the I 2 statistics for each analysis (with significant heterogeneity being indicated by I 2 ≥50%). 52 Publication bias was assessed via Egger's test. 53 We also calculated the fail-safe number (estimated number of studies needed to move the effect size from significant to non-significant), and trim and fill adjusted analysis 54 in case of publication bias (namely Egger's test p-value < 0.1). Because of different detection methods between studies, the standard mean difference (SMD) was computed as the effect size for the differences in CRP levels and calculated corresponding 95% confidence intervals. Analyses were run using comprehensive meta-analysis v2.0 (CMA, version 2 -metaanalysis.com). 55 Random-effect meta-regression was conducted to explore age, quality of included studies, and years of publications as potential moderators of the primary outcome. Sensitivity analyses were run removing HC to avoid confounding by indication and focusing on those studies comparing highest levels of SI versus lowest/no SI (secondary outcome only). Subgroup analyses were run to compare studies based on country, psychiatric condition affecting the population of interest, CRP type (high-sensitivity, normal), CRP sampling (plasma vs. serum), and the quality of the studies.

| Search results, characteristics, and quality of included studies
Search results and the study selection process are illustrated in Figure 1. Out of 550 initial hits, we screened 405 studies (after removing duplicates) at the title/abstract level, selecting 40 studies for full-text assessment. We excluded 19 studies for specific reasons after full-text assessment and ultimately included 21 studies. The complete list of the 19 studies excluded after full-text assessment, with reasons for exclusion, is reported in Table S3 (Supplementary material, page 7). Detailed characteristics and references of included studies are reported in Table 1.
Overall, this meta-analysis reports data from 7682 subjects (7445 affected by mental disorders, and 237 HC). Among persons with mental disorders, 453 were affected by BD, 2722 had a MDD, 2969 by a mood disorder, 30 had a first-episode psychosis, and 1271 by mixed disorders. Age range was from 18 to 70 years old.
Twenty-one studies included subjects with different disorders, one study included only subjects diagnosed with BD, 56 ten studies investigated patients with MDDs, 9,[22][23][24][25]39,40,[57][58][59] and four studies included people affected by mood disorders. 11,41,44,60 Moreover, one study was focused on first-episode psychosis, 61 and five studies evaluated patients with mixed disorders. 42,43,45,62,63 All continents except Africa and Oceania were represented. Specifically, four studies were conducted in Italy, three in the United States and in France, two in Brazil, one in Croatia, China, India, Iraq, Ireland, Sweden, Taiwan, and Turkey and one in The Netherlands. A detailed report on the quality of included studies according to the NOS scale is reported in Table 2.
The quality of included studies was high (NOS score ≥7) in 16 out of 21 studies, with a median score of 7.

| Main and sensitivity analyses
Results of the main comparative random-effects metaanalysis, together with publication bias, sensitivity, and subgroup analyses are reported in detail in Table 3. Regarding primary outcome, CRP was higher in subjects with suicidality (either SB or ideation) versus those without or with lowest suicidality (k = 21, SMD 0.69, 95% CI 0.48 to 0.90, p < 0.001, I 2 89.46%, medium effect size). Regarding secondary outcomes, CRP levels were elevated in those with suicide attempt versus those without suicide attempt (k = 17, SMD 0.55, 95%CI 0.36 to 0.73, p < 0.001, I 2 85.68%, medium effect size), and the difference was even larger when focusing on SI (k = 5, SMD 1.14, 95% CI 0.273 to 2.02, p = 0.010, I 2 95.43%, large effect size).
Sensitivity analyses removing HC to avoid confounding by indication substantially confirmed findings of main analyses.

| Subgroup analyses
Detailed results are reported in Table 3. In subgroup analyses, effect sizes on primary outcome differed across countries. Specifically, Turkey had a large effect size and China a very small one.
In addition, the association of CRP levels with suicidality is confirmed among the different psychiatric disorders (BD; depressive disorder; mood disorders; first-episode psychosis; and mixed disorders), suggesting the potentiality of CRP a transdiagnostic marker of suicide risk.
Moreover, the results are influenced also by the type of CRP detection (high sensitivity vs. normal). The association is confirmed with both techniques, yet the effect size is large with "regular" CRP array, but of medium magnitude with high-sensitivity measurement.
Finally, no significant difference emerged from subgroup analyses regarding studies' quality and sampling (plasma, serum).

| Meta-regression
Detailed results are reported in Table 4. Studies' quality as a continuous measure, year of publication, and age of participants did not moderate results.

| Publication bias
Detailed results are reported in Table 3. Potential publication bias was detected in all analyses, yet fail-safe-number ranged from 63 to 1151, and the significant nor the magnitude of effect sizes did not change substantially after trims-and-fill procedure.

| DISCUSSION
This meta-analysis shows a significant association of CRP levels with suicidality, and specifically a large association with SI, and medium association with suicide attempt pooling data from 21 studies and 7682 subjects (7445 with depressive disorders, BD, and schizophrenia spectrum disorders).
To the best of our knowledge, this is the first systematic review and meta-analysis pooling data from patients with depressive disorder, BD, and schizophrenia spectrum disorder including first-episode psychosis, and providing association estimates between CRP and SI and behavior separately. Results are clinically relevant for several reasons. First, suicide has a high prevalence in the longtime course of schizophrenia and affective disorders. Showing that CRP is associated with suicidality sums up to the established evidence of the association between a pro-inflammatory status and depressive disorders, 48,64,65 schizophrenia, [66][67][68][69] and BD. [70][71][72] Findings of this evidence synthesis are consistent with previous reports suggesting possible pathogenetic pathways mediating CRP-suicidality link. 21,36,73 For instance, it has been shown that the microglia in the brain of patients died by suicide is activated, and that pro-inflammatory mediators mRNA expression is increased in the prefrontal cortex of those dying by suicide, across different diagnoses. 21,36,73 More precisely, levels of mRNA or proteins expression of inflammatory markers (IL1β, IL4, IL13, IL6, and TNF-α) are higher in orbitofrontal cortex of suicide victims than in subjects died from other causes. 74,75 Also, pronounced microgliosis in specific brain regions (ie, dorsolateral prefrontal cortex, anterior cingulate cortex, and mediodorsal thalamus) of suicidal SMI patients has been observed. 73,76,77 In addition, an increased level of inflammatory markers (CRP, IL2, IFN gamma, IL4, IL5, IL6, IL10, TNF-α, IL6, IL8, and TGF-β1) in the plasma and cerebrospinal fluid of suicidal patients has been described. 78 Epidemiological evidence also points toward an association between immune-system alterations and suicidality, as cohort studies have shown a link between allergy, atopic disorders, asthma, and suicidality. [79][80][81] Converging evidence on the role of inflammatory mediators that increase suicidality also comes from evidence on the iatrogenic increased risk of SI with pro-inflammatory cytokines used to treat hepatitis C or multiple sclerosis. 82 Environmental and interpersonal stressors (eg, social and familial threats, and negative life events) play a crucial role in triggering SB. These precipitating/contributing factors may significantly interact with predisposing factors that increase vulnerability to suicide according to a stress-diathesis interplay. 84 However, the main pathophysiological mechanisms underlying this link are still poorly understood. 78 Stressful conditions can activate inflammatory signaling pathways in specific immune cells including monocytes and macrophages linked to abnormally enhanced pro-inflammatory IL-1β, IL-6, and TNF-α levels 85,86 on neuroglial cells. Pro-inflammatory cytokines released by microglia may stimulate oligodendroglia releasing cytokines involved in myelination and astrocytes having phagocytic properties and secreting abnormally elevated cytokines. 87,88 Importantly, activated immune-inflammatory pathways including elevated inflammatory cytokines and their multiple consequences have neurotoxic and excitotoxic effects on neurons leading to neurocognitive deficits associated with changes in neuroplasticity, synaptic sprouting, microglial sampling, lowered neurogenesis, changes in receptor expression and neuronal signaling, elevated nitro-oxidative stress, and finally neurodegeneration. 89 The state-of-the-art theory of mood disorders and schizophrenia is that immuneinflammatory neurotoxicity induces deficits in neuronal functioning which not only lead to depressive behaviors and psychosis but also SBs. [89][90][91][92] The routine analysis of serum CRP concentrations as a measure of inflammation and disease activity remains one of the most widely utilized assays in medicine. As a matter of fact, CRP is widely used as a marker of inflammation, infection and for risk stratification of cardiovascular events. 93

T A B L E 3 (Continued)
for research, because of the short half-life of other cytokines, and because the detectability at lower levels. 36 Several studies have shown that CRP levels were significant associated with SI in patients with MDD, [22][23][24] as well as in patients affected by generalized anxiety disorders. 94 Such association seems to be specific for SI beyond the diagnostic group. For instance, in patients with depressive disorder, those with high levels of SI tend to show significantly higher levels of inflammatory cytokines (a composite score including TNF-α, IL-6, IL10, and CRP) compared to those with low levels of SI even after adjusting for depression severity. 22,59 Recently, Gan et al. 95 found that several clinical features of BD including recent suicide attempt are associated with low-grade inflammation (defined as CRP > 3 mg/L), after adjusting for BMI. In schizophrenia, in which a pro-inflammatory status is present and is associated with clinical features, [96][97][98][99] higher suicide risk patients showed higher CRP levels than lower suicide risk patients and HCs. 61 However, some contrasting findings can also be found in the literature. For instance, Karlovic and colleagues showed that IL6 was associated with SI within patients with depression with melancholic (32 patients) and atypical (23 patients) features, yet no difference was found in the concentration of CRP. 39 In addition, CRP concentrations are increased also in suicide attempters with depressive disorder, 25,58 but no differences in the levels of serum hs-CRP among suicide attempters and among non-suicide attempters with MDD were found. 40 Moreover, such association appears to be accounted for the presence of physical disease among patients receiving care in a medical setting. 57 Conversely, a previous study showed a significant association between severe tobacco dependence and history of suicide attempt in BD patients, but not with level of CRP, independently of confounding factors. 56 Finally, one study showed that CRP levels appeared not to be related to SB nor ideation in inpatients with schizophrenia who were retrospectively categorized according to CRP at admission (CRP > 1 vs. <1 mg/dl). 100 Overall, this meta-analysis is of importance in merging evidence from several studies with small sample size, generating a more credible effect size beyond small study bias and type II error in underpowered studies.
This meta-analysis has several limitations. First, the majority of studies included in the present work adopted a cross-sectional study design with small to moderate sample sizes, and therefore, we cannot draw firm conclusions on causality. Thus, it is unclear whether increased inflammation represents a state or trait associated with SB.
Second, studies in mood disorders should examine the association between SBs and a larger panel of immuneinflammatory markers. 101 Third, CRP levels are affected by clinical confounders including age, BMI, obesity, smoking, low serum vitamin D levels, low levels of physical activity, poor diet, allergies, childhood maltreatment, stress, sleep disorders, and subclinical infections. [102][103][104][105] For example, in BD, the increases in hs-CRP are no longer significant after adjusting for the effects of BMI, age, and early lifetime trauma, which explain together 55% of the variance in hs-CRP. 106 Given that the present analysis included data published in original studies, we could not control for such confounding factors. Individual patient data meta-analyses (IPDMA) could overcome such limitation. Fourth, we were not able to control for ongoing treatment, despite the evidence indicates that treatment with antidepressants may affect cytokine and CRP levels. [107][108][109][110] IPDMA could overcome this limitation as well.
Fifth, it has been shown that the vast majority of evidence on biomarkers in the field of psychiatry is affected by several sources of bias. The present work is not exempt by different possible sources of bias, including reverse causality, small study effect, and excess of significance bias. 111 Finally, we have not identified any study specifically focusing on schizophrenia, and the subgroup analyses by diagnosis remain exploratory, as just individual studies or too few studies have been conducted in each separate diagnostic group.
Future studies should assess more comprehensive profiles of the acute phase response (ie, changes in positive and negative acute phase proteins including haptoglobin, α1-antitryprin, fibrinogen, sedimentation rate, and albumin), inflammation (acute phase proteins, IL-1β, IL-6, and TNF-α, and complement factors), and cell-mediated immune activation (M1 and Thelper-1 cytokines). 37 Since hs-CRP is strongly influenced by BMI and age, future psychiatric research should always adjust for the effects of these confounders and publish the residualized data, which should be used in CMAs.
Also, specific meta-research efforts primarily focusing on transdiagnosticity of CRP elevation as a marker or predictor of suicidality should be undertaken to test whether CRP is a transdiagnostic measure of suicidality. 112,113  Moreover, the interplay between brain-immune markers and response to treatments, particularly in the field of suicide prevention, needs further studies. 114,115 In conclusion, we show that CRP is associated with risk of suicidality in patients with mental disorders, and in particular with SI.