Network Meta-Analysis of Ticagrelor for Stroke Prevention in Patients at High Risk for Cardiovascular or Cerebrovascular Events

Alexandra Bálint, MD; Dániel Tornyos, MD; Oumaima El Alaoui El Abdallaoui, MS; Péter Kupó, MD; András Komócsi, MD, DSc


Stroke. 2021;52(9):2809-2816. 

In This Article


This systematic review was performed per the standards outlined in the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) Extension Statement for Reporting of Systematic Reviews Incorporating NMAs of Health Care Interventions[12] and registered with the International Prospective Register of Systematic Reviews under number CRD42020170746. The data that support the findings of this analysis are available from the corresponding author upon reasonable request.

The authors collected data from 3 online databases: MEDLINE (PubMed), Cochrane Collaboration of Clinical Trials, and EMBASE until August 1, 2020, from articles reporting randomized clinical trials with ticagrelor antiplatelet therapy. No language restriction was used. Broad search terms (ticagrelor, AZD 6140, and stroke) were used and combined using the boolean operator AND.

Studies were included if the following criteria were fulfilled: (1) randomized controlled trials, (2) assessing the clinical efficacy or safety of an antiplatelet regime including ticagrelor alone or as part of a dual antiplatelet therapy strategy with ticagrelor plus aspirin, and (3) reported on the occurrence of stroke in minimum duration of 30 days (4) in patients with cerebrovascular, coronary, or peripheral artery disease.

We excluded studies if any of the following criteria were applied: (1) nonrandomized studies, (2) single-arm studies, (3) outcomes of interest were not reported or were impossible to extract or calculate from published results, (4) comparing merely the biological efficacy of the antiplatelet treatment, or (5) duplicate publications.

Two investigators (A.B. and D.T.) independently evaluated record titles and abstracts of all citations in line with the PICO criteria (patient/population, intervention, comparison and outcomes); any discrepancies were resolved by a third investigator (A.K.).

For definitions of stroke, the internal definitions of the included trials were used if compliant with focal loss of neurological function caused by an ischemic or hemorrhagic event, with residual symptoms lasting at least 24 hours or leading to death.

The primary efficacy outcome of our analysis was the occurrence or recurrence of stroke including ischemic or hemorrhagic forms. Major bleeding and all-cause mortality were assessed as main safety end points. Secondary outcomes included the individual end points of ischemic stroke, hemorrhagic stroke, and TIA, myocardial infarction, major cerebral or cardiovascular event defined as the composite of death, MI, and stroke, and cardiovascular death. Additionally, data of disabling stroke (defined as death or modified Rankin Scale score >1) were also collected. Furthermore, safety outcomes as the frequency of major and minor bleeding complications and intracranial bleeding were also evaluated. In the case of the availability of multiple major bleeding definitions, we extracted the Thrombolysis in Myocardial Infarction (TIMI) major bleeding. The data from the intention-to-treat analyses were extracted, and the end points of interest were collected until the longest follow-up available.

The methodological qualities of the studies were also assessed using the Cochrane Collaboration tool for assessing the quality of randomized controlled trials.

NMA Modeling

Considering that the trials used different control groups for comparing outcomes of ticagrelor-medicated patients and that the study arms included combination as well as monotherapy with different antiplatelets, we prespecified the use of multiple-treatment NMA supplemented with component NMA modeling.

At the first level, each potential antiplatelet combination was entered as an individual study arm, and data were pooled in a multiple-treatment NMA that allows integration of direct and indirect comparisons. We calculated the risk ratio (RR) and its SE using a frequentist approach to construct an NMA model accounting for the correlated treatment effects.[13] A random-effects model was applied by adding the estimated heterogeneity to the variance of each comparison using an adaptation of the DerSimonian-Laird estimator. The choice of the random-effects model was made based on the consideration that the true preventive effect of antithrombotic treatment may vary from study to study influenced by the heterogeneity of the included trials.

Values of I2 representing the amount of inconsistency and Cochran Q statistics and its corresponding P measuring the heterogeneity in the network were also calculated.[13,14]

A special case encountered in our network was that treatment arms may be combinations of other treatments or have common components. Therefore, the influence of individual components was intended to be evaluated in an additive model assuming that the effect of treatment combinations is the sum of the effects of its components.[15]

For easier interpretation, effect sizes are depicted in the form of forest plots with aspirin set as reference. Furthermore, a comparative ranking of the treatments according to the P-scores method (a frequentist analog of the surface under the cumulative ranking curve) was performed.[16]

The assumption of consistency that the direct evidence in a network for the effect size between two treatments does not differ from the indirect evidence was assessed by net heat plots and by net splitting. The latter method splits our network estimates into the contribution of direct and indirect evidence, which allows controlling for inconsistency in specific comparisons.

To assess publication bias, a comparison-adjusted funnel plot—an extension of the common funnel plot in cases of multiple-treatment comparison—was used displaying Egger test results in support[17] with the additional use of the Cochrane Collaboration assessment tool.

The clustering of the treatment arms was assessed using the estimated relative risk compared with aspirin in the nearest neighbor analysis. An explorative analysis was performed to assess the potential impact of background risk on the estimated treatment effect. Within this, risk of stroke of the study population using clopidogrel plus aspirin therapy was calculated, and this continuous variable was used to construct regressor in a Bayesian metaregression analysis. Additional exploratory analyses included stratification and subgrouping based on the included patient population, multilevel meta-analysis, and multivariate meta-analysis of direct comparisons using structural equation modeling.

All calculations were performed with the R statistical software package, version 3.6.3 (R Development Core Team, 2010), software using the packages meta 4.11–0, netmeta 1.2–0, and gemtc 0.8–4.[18] P<0.05 was considered to represent statistical significance.