C-Reactive Protein, Interleukin-6, and Vascular Recurrence After Stroke: An Individual Participant Data Meta-Analysis

Background: Anti-inflammatory therapies reduce recurrent vascular events in coronary disease. Existing studies have reported highly conflicting findings for the association of blood inflammatory markers with vascular recurrence after stroke leading to uncertainty about the potential of anti-inflammatory therapies after stroke and no consensus about the utility of measurement of inflammatory markers in current guidelines. Methods: We investigated the association between hsCRP (high-sensitivity C-reactive protein), IL-6 (interluekin-6), and recurrent major adverse cardiovascular events (MACE), and stroke from individual participant data from 8420 patients with ischemic stroke/transient ischemic attack from 10 prospective studies. We did within-study multivariable regression analyses and then combined adjusted risk ratio (RR) by random-effects meta-analysis. Results: During 18 920 person-years of follow-up, 1407 (16.7% [95% CI, 15.9–17.5]) patients had MACE and 1191 (14.1% [95% CI, 13.4–14.9]) patients had recurrent stroke. On bivariate analysis, baseline IL-6 was associated with MACE (RR, 1.26 [95% CI, 1.10–1.43]) and recurrent stroke (RR, 1.18 [95% CI, 1.05–1.32]), per unit increase logeIL-6. Similar associations were observed for hsCRP (MACE RR, 1.19 [95% CI, 1.09–1.29]; recurrent stroke RR, 1.12 [95% CI, 1.04–1.21], per unit increase logehsCRP). After adjustment for vascular risk factors and treatment, independent associations remained with MACE (IL-6, RR, 1.12 [95% CI, 1.04–1.21]; hsCRP, RR, 1.09 [95% CI, 1.04–1.15]) and recurrent stroke (IL-6, RR, 1.09 [95% CI, 1.00–1.19]; hsCRP, RR, 1.05 [95% CI, 1.00–1.11]). Comparing the top with the bottom quarters (Q4 versus Q1), IL-6 (RR, 1.35 [95% CI, 1.09–1.67]) and hsCRP (RR, 1.31 [95% CI, 1.07–1.61]) were associated with MACE after adjustment. Similar results were observed for recurrent stroke for IL-6 (RR, 1.33 [95% CI, 1.08–1.65]) but not hsCRP (RR, 1.16 [95% CI, 0.93–1.43]). Conclusions: Blood markers of inflammation were independently associated with vascular recurrence after stroke, strengthening the rationale for randomized trials of anti-inflammatory therapies for secondary prevention after ischemic stroke/TIA.


See related article, p 1300
Inflammation plays a central role in atherosclerotic plaque development and rupture, and promotes thrombosis, leading to thromboembolic events. 4,5 Evidence from laboratory studies, imaging studies with plaque 18-F fluorodeoxyglucose positron emission tomography, and genetic epidemiological studies support a role of inflammation in the pathogenesis of stroke. [6][7][8] In randomized controlled trials (RCTs) of patients with coronary artery disease (CAD), inhibition of IL (interleukin)-1β with canakinumab reduced vascular recurrence with greatest benefit in responders measured by on-treatment blood IL-6 and hsCRP (high-sensitivity C-reactive protein) levels, and the pleiotropic anti-inflammatory agent colchicine reduced major vascular events, including stroke. 9,10 Despite evidence from patients with CAD, it is unclear whether blood inflammatory markers are associated with recurrent vascular events after stroke. This may be related to differences in pathophysiology between stroke and CAD. Consequently, current guidelines do not recommend the routine measurement of inflammatory blood markers in patients with stroke. In a recent systematic review of over 2500 reports, only half of 30 published articles reported associations between hsCRP, IL-6, and recurrent stroke. Furthermore, pooled analysis of available studies was not possible due to variability in the definition of inflammatory marker exposure and adjustment for potential confounding variables. 11 Studies to clarify the potential prognostic information provided by blood inflammatory markers are an important step to improve identification of high-risk patients most likely to benefit from anti-inflammatory therapy in randomized trials and clinical practice. We aimed to investigate if circulating hsCRP and IL-6 are associated with vascular recurrence after ischemic stroke in an individual participant data (IPD) meta-analysis of prospective studies.

Systematic Review
We did a systematic review of studies investigating the association between blood inflammatory markers, measured after ischemic stroke or transient ischemic attack (TIA), and the risk of recurrent stroke and major adverse cardiovascular event (MACE). Details of eligibility criteria, search strategy, study selection and data extraction have been previously described 11 and information is also available in the Supplemental Material. Briefly, 2 reviewers searched Embase, Ovid Medline, and conference proceedings for eligible studies published up to January 10, 2019, using a combination of medical subject headings and free text search terms (Supplemental Material). Key eligibility criteria were studies of ≥50 patients with ischemic stroke/TIA, with blood analyzed for hsCRP and IL-6 measured after the index event, and reported associations with recurrent stroke or MACE. The protocol was published in advance on PROSPERO (International Prospective Register of Systematic Reviews; URL: https://www.crd.york.ac.uk/ PROSPERO/; Unique identifier: CRD42018116190). This article written according to PRISMA-IPD guidelines (Preferred Reporting Items for Systematic Reviews and Meta-Analyses of Individual Participant Data; Supplemental Material). 12

Collection and Harmonization of IPD
After publication of a study-level systematic review, 11 authors of eligible studies were invited to provide IPD. Previously unpublished data were also considered. If studies were unable to provide individual-level data for centralized analysis due to restrictions on data transfer, study inclusion was enabled by remote analysis using identical methodology before 2-stage meta-analysis of these results with those from the other studies. A Project Steering Committee was established and a prespecified protocol for data collection, harmonization, and quality assurance was agreed (Supplemental Material). Key individual patient data requested from participating studies included demographic details, index event type (stroke or TIA), National Institutes for Health Stroke Scale, stroke risk factors, statin and antithrombotic medication use, timing of blood sampling and measurements obtained for baseline blood inflammatory markers, follow-up duration, and outcome events. Further information is provided in the Supplemental Material.

Outcomes
The prespecified outcomes were (1) any MACE (first event of recurrent stroke, major coronary events, and vascular death) and (2) first recurrent stroke (ischemic, hemorrhagic, or unspecified) which occurred after blood sampling. For standardization and generalisability, stroke was defined as a new focal  13 Coronary events were defined as myocardial infarction, unstable angina requiring hospitalization, cardiac arrest, or sudden cardiac death, confirmed by clinical information, cardiac enzymes, and electrocardiographs. Vascular death was defined as a fatal stroke or coronary event attributed as the cause of death. One included study used a broader definition of coronary events and vascular death (Table S1). 14 Recurrences that occurred prior to blood sampling for inflammatory markers were excluded.

Risk of Bias
The Quality in Prognosis Studies tool was used to assess the risk of bias of included studies and was adjudicated by 2 reviewers (Supplemental Material). 15

Statistical Analysis
A prespecified statistical analysis plan was agreed. Synthesis of IPD involved a 2-stage approach. First, estimates of association (hazard or odds ratios) of the baseline levels of hsCRP and IL-6 with recurrent events were calculated separately within each study after identical adjustment for potential confounding variables. Second, the adjusted effect estimates were pooled using random-effects meta-analysis with inverse variance weighting. For prospective cohort studies, adjusted hazard ratios were estimated from multivariable Cox proportional hazards regression, with time to first recurrence as the dependent variable. For the single nested case-control study within a randomized trial (PROGRESS [Perindopril Protection Against Recurrent Stroke Study]), 16 the adjusted odds ratio was calculated using a conditional logistic regression model. Hazard ratios and odds ratios were assumed to approximate the same RR (because the outcome is rare) and were collectively summarized as risk ratios (RRs). Both hsCRP and IL-6 were positively skewed and transformed to the natural logarithm (log e ) scale to stabilize the variance. All analyses were done per unit and per quarter increase in the log e -transformed inflammatory marker, after adjustment. One unit increase in log e of the biomarker corresponds to an ≈3-fold increase on the natural scale. Each analysis was performed using prespecified increasing degrees of adjustment. Model 1 adjusted for age and sex (and trial arm where appropriate). Model 2 adjusted for variables in model 1 plus hypertension, smoking, diabetes, atrial fibrillation, CAD, and antithrombotic use. Model 3 adjusted for variables in model 2 plus statin therapy. We did not include both hyperlipidemia and statin therapy in the same model due to collinearity between these variables. In exploratory analyses, we further analyzed the association between each inflammatory marker at baseline (hsCRP or IL-6), adjusted for variables in model 2 plus the other inflammatory marker. To examine the shape of the relationship between each inflammatory marker and outcome, we plotted pooled adjusted RRs against quarters of the biomarker distributions.
Prespecified subgroup analyses were performed according to age, sex, index event severity, and time to phlebotomy with differences between subgroups assessed by visual inspection of pooled effect estimates and Wald tests. The relationship between inflammatory markers and covariates (including demographics, cardiovascular risk factors, and medications) was examined from the 9 studies where individual-level data was provided centrally. The relationship between the inflammatory marker and continuous covariate data was analyzed by plotting the mean concentrations of each biomarker within each tenth of the covariate distribution and the degree of association examined using Spearman rank correlation. For categorical variables, differences in concentrations of either blood marker were analyzed using the Mann-Whitney U test. We investigated potential influence of outlying studies by removing one study at a time. All analyses were done using R (R Project for Statistical Computing, Vienna, Austria).

Study Selection, Characteristics, and IPD
The systematic literature search identified 2515 nonduplicate records. After title and abstract screening, 130 studies were selected for full text review. This identified 30 eligible studies. 11 After contacting corresponding authors, 8 studies agreed to participate in the IPD analysis ( Figure 1). 14,[17][18][19][20][21][22] Two additional studies with previously unpublished data were included after they were identified through correspondence. 23, 24 In total, 10 studies comprising 8420 patients and 18, 920 person-years of follow-up were included (Table S2; 9 prospective cohort studies and 1 nested case-control study within an RCT [PROGRESS]). Nine studies provided IPD for centralized analysis. One study group (PROGRESS) did prespecified identical analyses locally and provided summary aggregate data for meta-analysis with other data. 16 A summary of participating studies is provided in the Supplemental Material.

Clinical Characteristics
Clinical characteristics of included patients are provided in Table 1. The qualifying event was ischemic stroke in 6911 (82.1%) and TIA in 1509 patients (17.9 %). Four studies excluded severe strokes, [17][18][19]23 including 2 which included only patients with TIA. 18,19 The remaining 6 studies included patients with ischemic stroke regardless of severity. High rates of treatment with antithrombotic medication at hospital discharge (or randomization for the single RCT) were observed. Greater variation between studies was observed for rates of statin treatment at discharge (or at randomization for PROGRESS), likely reflecting different study eras. The median time from index stroke/TIA to blood sampling was <1

Study Quality and Follow-Up
No study was considered at high risk of bias in any domain of the Quality in Prognosis Studies tool (Table S4)

Associations of Inflammatory Markers With Other Variables
In the pooled data set, direct linear associations were observed between hsCRP and age, body mass index, stroke severity (measured by National Institutes for Health Stroke Scale), IL-6, and fibrinogen, with inverse associations observed for LDL (low-density lipoprotein) and total cholesterol, and triglycerides ( Figure 2). No associations were evident with sex, systolic or diastolic blood pressure, current smoking, diabetes, antiplatelet, or statin therapy. Both IL-6 and hsCRP were higher in patients with atrial fibrillation compared with those without atrial fibrillation (P<0.001). hsCRP increased rapidly in early time-intervals after stroke/TIA onset and remained elevated in the postacute phase compared with the earliest time intervals. For IL-6, direct linear associations were also observed with age, hsCRP, and fibrinogen, with a similar association also observed for systolic blood pressure ( Figure 2). Inverse linear relationships were apparent for LDL and total cholesterol, and triglycerides. Curvilinear relationships were present for stroke severity (positive association) and time interval (negative). A U-shaped relationship with body mass index was observed. No relationships were observed between IL-6 and sex, diastolic blood pressure, smoking, antiplatelet, or statin therapy.  Figure 3). These associations remained consistent after further adjustment for each inflammatory marker (Table S6).

Inflammatory Markers and Risk of Recurrent Vascular Events
When analyzed by quarters of the baseline inflammatory marker distribution, a dose-dependent increase in risk of recurrent MACE was observed for increasing hsCRP. Compared with quarter 1 (hsCRP <1.03 mg/L, referent), the RR of MACE in quarter 4 (hsCRP >7.38

Sensitivity Analysis
The main results were not driven by the influence of any single study ( Figure S1). However, there was no association between inflammatory markers and MACE when the analysis was restricted to patients with hsCRP below ≤3 mg/L or below the median IL-6 level (<4.5 pg/mL; Table S7).

Subgroup Analyses
In subgroup analyses, consistent directions of association were observed for hsCRP/IL-6 and MACE stratified by age (≤65 versus >65 years) and sex. Although of borderline statistically significance (P=0.07 for interaction), hsCRP appeared to be more strongly associated with recurrence in patients with nonsevere stroke/TIA (defined by National Institutes for Health Stroke Scale ≤5) compared with patients with more severe stroke (Figure S2). Similarly, the associations were consistent for both hsCRP and IL-6 when stratified by time of outcome events (defined as MACE ≤90 days versus MACE >90 days), by time to phlebotomy (defined as blood sampling

DISCUSSION
Our study provides new high-quality evidence that inflammation is associated with vascular recurrence after stroke. Despite convincing data in coronary disease, previous studies examining the prognostic role of blood inflammatory markers after stroke have reported conflicting findings and a study-level systematic review of this question was inconclusive. 11 In this context, in an IPD meta-analysis of 9 prospective cohort studies and one RCT, we found that hsCRP and IL-6 were associated with recurrent vascular events after stroke. For every 3-fold increase in hsCRP or IL-6 concentration, there was a 20% to 25% increase in the risk of major cardiovascular events. These associations were attenuated but remained after adjustment for established cardiovascular risk factors and contemporary secondary prevention therapy. For patients in the highest quarter of the hsCRP and IL-6 distributions, the adjusted risk of recurrent MACE was increased by about one-third compared with the lowest quarter. Similarly, the risk of recurrent stroke was greatest in the highest IL-6 quarter, but not in the highest hsCRP quarter. As expected, we found linear associations between inflammatory markers and age, body mass index, and stroke severity. However, contrary to previous reports no association was observed between CRP/IL-6 and smoking, diabetes, blood pressure, or statin therapy. The precise explanation for this is unclear.
In response to stimuli, such as cholesterol microcrystals, plaque macrophages assemble intracellular NLRP3 (NOD-like receptor family pyrin domain-containing 3) inflammasomes which activate IL-1β and IL-18 from their inactive proforms. Interleukin-1β is a potent stimulus for local and systemic IL-6 expression, which amplifies the inflammatory response, promotes atherothrombosis, and stimulates hepatic expression of the downstream marker CRP. Our findings are consistent with previous meta-analyses which demonstrated that hsCRP 25 and IL-6 26 are associated with first stroke in apparently healthy adults, and with Mendelian randomization studies which reported that lifelong impaired IL-6 signaling is protective against stroke and coronary events. 7 Our study is also consistent with meta-analyses in coronary disease 27 and recent data from the Cardiovascular Inflammation Reduction Trial, which demonstrated that hsCRP and IL-6 were independently associated with MACE outcomes in coronary patients with diabetes/metabolic syndrome treated with contemporary secondary prevention medications. 28 In the CANTOS (Canakinumab Anti-Inflammatory Thrombosis Outcomes Study) randomized trial, IL-1β inhibition reduced recurrent vascular events in coronary patients, with the treatment benefit mainly apparent in those with on-treatment hsCRP or IL-6 suppression. 9 Colchicine, a broad-spectrum antiinflammatory agent, substantially reduced recurrent MACE and stroke in coronary trials. 29 Taken together, our findings and these studies provide a strong rationale for targeting the NLRP3/IL-1β/IL-6 pathway in future randomized trials for secondary stroke prevention. Current stroke guidelines do not recommend the routine measurement of blood inflammatory markers after stroke. Our study raises the hypothesis that measurement of hsCRP, IL-6, or both may improve identification of patients with stroke at greatest risk of vascular recurrence and prediction of recurrent events. In coronary patients, the strength of association between hsCRP, IL-6, and MACE recurrence is similar to that observed with LDL-cholesterol, which is routinely measured in practice. 30 Inflammatory markers also identified patients with high residual vascular risk, even after lipid reduction to unprecedented low levels in coronary trials of PCSK-9 (proprotein convertase sutilisin/kexin type 9) inhibitors. Although specific anti-inflammatory stroke prevention medications are not currently available, measurement of inflammatory markers might identify subgroups of patients at greater risk, and who may benefit more intensive modification of risk factors. Our study suggests that early measurement of inflammatory markers after stroke or TIA provides useful prognostic information. Further studies are needed to investigate the prognostic role of such markers measured late after stroke, associations in  stroke subtypes, the relative utility of hsCRP compared with IL-6, their use for prediction of recurrent stroke versus poststroke coronary events, and the additional information provided when added to clinical prognostic scores (eg, Essen Risk Score). The strengths of our study include the large sample size, which provided greater statistical power and precision than was previously available. All studies were identified after a standardized systematic review process, thereby minimizing selection bias, and used validated methods for inflammatory biomarker measurement. A prespecified assessment of study quality demonstrated the internal validity of included cohorts. Access to IPD allowed for standardized definitions of covariates across studies and a uniform and comprehensive adjustment for potential confounders, unlike many previous studies. 11 Although not all identified studies from our systematic review provided IPD, the direction of our findings is concordant with those from several large studies which were not included, supporting the validity of our results. 31,32 Our findings are biologically plausible and consistent with other data from epidemiological and genetic studies, and RCTs in CAD.
We acknowledge some limitations. We cannot entirely exclude the possibility of residual confounding. However, we believe that this is unlikely as we adjusted for a wide range of established cardiovascular risk factors. Our study included mainly patients from European and Asian ethnicity so may not be externally generalized to other ethnic groups. Our study was not designed to identify the source of blood inflammatory markers, which may partly reflect the poststroke acute-phase response in addition to background inflammation from atherosclerotic plaque. 11 However, Figure 4. A through D, IL-6 (interleukin-6), hsCRP (high-sensitivity C-reactive protein) and risk of major adverse cardiovascular event (MACE) by quarter. Dose-dependent relationship per quarter of the baseline distribution of biomarker between (A) IL-6 and MACE; (B) hsCRP and MACE; (C) IL-6 and recurrent stroke; (D) hsCRP and recurrent stroke. P values represent difference between quarter 1 (referent) and 4. For hsCRP, quarter 1 defined as <1.03 mg/L, quarter 4 >7.38 mg/L. For IL-6, quarter 1 defined as <2.11 ng/L, quarter 4 >12.76 ng/L. All analyses adjusted for age, sex, trial arm (where appropriate), smoking, hypertension, diabetes, atrial fibrillation, coronary artery disease (CAD), antithrombotic therapy, and statins. our findings were consistent in studies which obtained blood samples at early and later time points, and when stratified by stroke severity. We cannot fully exclude the possibility that inflammation caused by cerebral infarction, or infection may have contributed to the risk of recurrent events (reverse causation). We believe that this is unlikely, as other studies have demonstrated that circulating hsCRP and IL-6 are associated with incident stroke, and Mendelian randomization studies have shown that reduced incident stroke risk is associated with lifelong exposure to genetic variants which down-regulate IL-6 signaling. We did not analyze the associations of hsCRP and IL-6 with recurrent events stratified by stroke subtype. However, in addition to atherosclerosis, 4 evidence also supports an important role of inflammation in pathogenesis of atrial fibrillation, 33,34 coagulation system activation, 35 and vascular recurrence after lacunar stroke. 31,36 Subtype-specific analyses will be the subject of a further report. Intracerebral hemorrhage accounted for just 6% of stroke recurrences in this analysis, which is lower than expected, and the relative contribution of inflammation to these events is unknown.
To date, there are no completed RCTs of anti-inflammatory therapies for secondary prevention after stroke. 37 One ongoing trial, CONVINCE (Colchicine for Prevention of Vascular Inflammation in Non-Cardioembolic Stroke) is comparing low-dose colchicine added to standard care to standard care alone. 38 Other trials are beginning or planned. 39 In high-risk patients with chronic kidney disease and raised hsCRP, the ZEUS trial (Effects of Ziltivekimab Versus Placebo on Cardiovascular Outcomes in Participants With Established Atherosclerotic Cardiovascular Disease, Chronic Kidney Disease and Systemic Inflammation Trial) is investigating targeted IL-6 inhibition with ziltivekimab for prevention of vascular events including stroke. 40 Similar randomized trials in patients with stroke are urgently needed.