FormalPara Key Points

Both antiplatelet (AP) and anticoagulant (AC) drugs can be used in the management of cervical artery dissection due to similar rates of recurrent stroke, transient ischaemic attack, haemorrhage, mortality, and complete recanalization. Despite this similarity, it is possible that the statistically equal AC/AP profiles may be due to significant conflict zones among the AC–AP comparative studies.

We urge further well-designed clinical trials to determine whether or not unnecessary anticoagulation may be avoided in cervical artery dissection.

1 Introduction

Constituting 1–2% of all ischaemic stroke aetiologies [1], cervical artery dissection (CeAD) (i.e., extracranial internal carotid artery/vertebral artery dissection) is by far a significant contributor to almost one-fifth of ischaemic stroke cases in the younger age population [2,3,4,5]. For every 100,000 people, the incidence of CeAD is about 2.6 (95% confidence interval (CI) 1.86–3.33) per year in the general population [6]. Whether traumatic or spontaneous, CeAD carries a considerable risk of a recurrent ischaemic stroke (more common in the first 2 weeks of onset) [7, 8], intracranial haemorrhage (ICH), transient ischaemic attacks (TIAs) and, rarely, death [9,10,11]. Therefore, multiple studies were conducted over the last two decades aiming to minimize these potential risks via either medical or endovascular therapies [2, 9]. The choice of anticoagulation (AC) versus antiplatelet therapy (AP) has been the most demanding question to be addressed.

Unfortunately, current evidence substantially lacks proper conclusions as to CeAD acute and maintenance management with regard to AP/AC. The heterogenous nature of the previous CeAD studies poses a significant obstacle to their external validities [2,3,4,5, 7,8,9,10,11,12,13]. Specifically, the study design (retrospective or prospective, per-protocol/intention-to-treat analysis), time to randomization, the choice of AP or AC—based on the primary physician choice (either solely or months after randomization), the outcome definitions (symptomatic vs. major intracranial haemorrhage, possible vs. definite cause of ischaemic stroke, early vs. late stroke), time to follow-up and the follow-up methodology (whether ultrasonography (US)-based or angiographically based, subclinical vs. clinically oriented using magnetic resonance imaging (MRI) of the brain) constitute the most common areas of vast methodologic variations [8,9,10,11,12,13,14].

In addition, the wide CIs of many outcome findings in these studies stem from a limited sample size [9]. For example, an estimated number of 913 patients with an 80% power and a 5% significance level has been set since 2003 for addressing the AC/AP choice [15]. However, even the CADISS trial has failed to achieve this postulated count (n = 250) [10]. Moreover, the answer to the futility of AC/AP remains perplexing in the context of a low risk of recurrent stroke [10, 11, 14] . As a result, there is only a Class of Recommendation IIA and a Level of Evidence B for either therapy based on a single randomized controlled trial (RCT) carrying substantial limitations [10, 16]. Finally, the maximum estimated follow-up period for dissection patients was 28 months, which again adds temporal blurring in addition to the relatively low case load, as opposed to atherosclerotic aetiologies being studied, for instance [17]. In this regard, 58 months was the mean follow-up period in only one study [18]. As reported by Chowdhury et al. [19], clinical follow-up averaged 17.7 months, while Kennedy et al. [14] and Sarikaya et al. [20] only included 3-month follow-up results.

In view of the described methodologic heterogeneity, temporal limitations, and low sample sizes, we aimed, through this systemic review and meta-analysis, to mitigate these barriers in order to reach a possible consensus regarding the choice of AC versus AP strategy with an emphasis on the indications for both, the relative merit of either therapies, as well as the dose and duration of therapy.

2 Methods

2.1 Study Design

The current study is a systematic review and meta-analysis of the literature. The Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020 standards were strictly followed in the course of conducting the study [21]. Our study’s PROSPERO Registration number is CRD42022338493.

2.2 Eligibility Criteria and Study Selection

We included peer-reviewed RCTs and observational studies that compared AP therapy with AC in patients with CeAD. The main exclusion criteria were non-English studies, single-arm studies, conference abstracts, reviews and studies without separate data for the intervention and comparison groups.

2.3 Literature Search

We systematically searched the following electronic databases: PubMed, Web of Science, Cochrane, Scopus, Embase and OVID till September 2021. Across all databases, the most recent search was conducted on 30 September 2021. The search strategy used was (Aspirin OR Acetylsalicylic Acid OR 2 Acetyloxy benzoic Acid OR Acylpyrin OR Aloxiprimum OR Colfarit OR Dispril OR Easprin OR Ecotrin OR Endosprin OR Magnecyl OR Micristin OR Polopirin OR Polopiryna OR Solprin OR Solupsan OR Zorprin OR Acetysal OR Antiplatelet Therapy) AND (Anticoagulants OR Anticoagulant OR Anticoagulation Agent OR Indirect Thrombin Inhibitors) AND (cervical artery dissection). The search strategy was adjusted according to the database, for example, we used quotation marks for searching on Scopus. The selection process was carried out by two independent pairs of authors (B.K.A, M.A, M.T.H, A.A.A); another author (A.I.H) was consulted to resolve any conflicts, with an average disagreement of 4%.

2.4 Quality Assessment

To assess the quality of the included cohort studies, we used the Newcastle–Ottawa Scale (NOS) for assessing the quality of non-randomised studies [22]. The domains include selection, comparability and exposure. Each domain is assessed using stars, with a maximum of nine stars. Meanwhile, to assess the risk of bias of the included RCTs, we used the Cochrane risk of bias tool for RCTs [23]. The domains include random sequence generation, allocation concealment, blinding of participants and personnel, blinding of outcome assessment, incomplete outcome data, selective reporting, and other bias. The assessment is based upon the judgment of low, high or unclear risk of bias. Two independent authors (M.A and B.K.A) assessed the quality of included studies; another author (A.I.H) was consulted to resolve any conflicts, with a disagreement of 3%.

2.5 Data Extraction

We extracted data, in an excel sheet, on the following: (1) Summary of included studies: study ID (first author/year of publication), title, study design, country and date of implementation, participants and key inclusion/exclusion criteria, intervention group, control group, duration of follow-up, and conclusion. (2) Baseline characteristics of the included population: study ID (first author/year of publication), title, study groups and the total number on the group, age, gender, dissection type, admission mRs 0–2, and ≥3, hypertension, current smoking, alcohol use, stenosis, occlusion, aneurysmal, complete recanalization, partial recanalization, prior TIAs, prior treatment, and other relevant characteristics. (3) Study outcomes as described below. Data were extracted by two independent pairs of authors (B.K.A, M.A, M.T.H, A.A.A); another author (A.I.H) was consulted to resolve any conflicts, with an average disagreement of 4%.

2.6 Study Outcomes

The study outcomes included death, primary ischaemic stroke, primary TIA, recurrent stroke, recurrent TIA, symptomatic intracranial haemorrhage (sICH), major extra-cranial bleeding, excellent outcome at 6 months (modified Rankin scale (mRS) score = 0–2), poor outcome at 6 months (mRS score ≥ 3), complete recanalization, total adverse events, major haemorrhagic complications (bleeding), seizure, admission mRS score 0–2, admission mRS score ≥ 3, all haemorrhagic complications, time from baseline MRI to follow-up MRI (days), patients with new diffusion-weighted imaging (DWI) lesions, occlusion and stenosis.

2.7 Data Synthesis

We performed the analysis using Review Manager (RevMan) version 5.4. Data were pooled as odds ratio and 95% confidence Interval (CI) (dichotomous) or mean difference (MD) and 95% CI (continuous). Data were considered significant if p < 0.05. We measured heterogeneity using the I-square (I2) test and chi-square test. We considered the data heterogeneous with chi-square P < 0.1 and the I2-value was larger than 50%. When heterogeneity was found, we used the random-effect model, other than that we used the fixed-effect model. We performed subgroup analyses according to the study design (cohort or RCTs), or according to the follow-up duration (3 months, or more than 3 months) when available.

2.8 Publication Bias

Even though the total number of included studies was 15, the maximum number of included studies in an outcome was less than 10. Therefore, we did not assess the publication bias using funnel plots, following Cochrane handbook recommendation 10.4.3.1 [24].

2.9 Meta-regression Analysis

Meta-regression requires a minimum of ten studies to be included in each outcome analysis, as stated in the Cochrane Handbook for Systematic Reviews of Interventions [25], which did not exist in the context of our research. As a direct consequence of this, a meta-regression analysis cannot be constructed on the basis of the data that are currently available.

3 Results

3.1 Literature Search

Based on our systemic search, we initially had 1230 records, and 277 of them were removed as they were duplicates. We screened 953 records, of which 144 were suitable for full-text screening. Finally, we included 15 studies [8,9,10,11, 13,14,15, 17, 26,27,28,29,30,31,32] (Fig. 1).

Fig. 1
figure 1

PRISMA flow diagram

Full size image

3.2 Characteristics of the Included Studies and Population Baselines

The included studies were mainly observational (cohort) studies, but there were also two RCTs [9, 10]. The follow-up duration of the included studies ranged from 1 month to 12 months [8,9,10,11, 13,14,15, 17, 26,27,28,29, 31, 32]. Two studies reported data on carotid dissection only [17, 29], one reported data on vertebral dissection [13], one reported data on both [8], and the other studies did not specified the type. The mean age of patients in these studies [8,9,10,11, 13,14,15, 17, 26,27,28,29,30,31,32 ranged between 34.4 and 50.4 years (Tables 1, 2).

Table 1 Summary of included studies
Full size table
Table 2 Baseline for included studies
Full size table

3.3 Risk of Bias

The included cohort studies had a score range of 8–9 stars out of 9, with the majority of studies scoring 8. Therefore, all studies were classified as high quality (Table 3). On the other hand, the included RCTs [9, 10] had a high risk of bias for allocation concealment and blinding of participants and personnel (Fig. 2).

Table 3 Risk of bias for cohort studies
Full size table
Fig. 2
figure 2

Risk of bias graph and summary

Full size image

3.4 Outcomes

3.4.1 Death During Short-Term Follow-Up (3 Months)

Five studies were included in the analysis with a total of 1,029 patients. The analysis showed insignificant statistical differences between the two groups (OR 0.47, 95% CI 0.03–7.58, p = 0.59). A subgroup analysis was performed. In cohort studies (N = 585 patients), the analysis also showed no significant differences between the two groups (OR 0.47, 95% CI 0.03–7.58, p = 0.59) (Fig. 3a).

Fig. 3
figure 3

a Death during short-term follow-up. b Death during long-term follow-up

Full size image

3.4.2 Death During Long-Term Follow-Up (More Than 3 Months)

Five studies were included in the analysis with a total of 507 patients. The analysis showed no significant differences (OR 1.63, 95% CI 0.40–6.56, p = 0.49), and the results were homogenous (p = 0.84; I2 = 0%). A subgroup analysis was performed. In an RCT by Markus et al. [10] (N = 250 patients), the analysis also showed no significant differences between the two groups (OR 2.98, 95% CI 0.12–73.76, p = 0.51). In cohort studies (N = 257 patients), the analysis also showed no significant differences between the two groups (OR 1.37, 95% CI 0.28–6.64, p = 0.7), and the results were homogenous (p = 0.72; I2 = 0%) (Fig. 3b).

3.4.3 Primary Ischaemic Stroke

Five studies were included in the analysis with a total of 935 patients. The analysis showed insignificant statistical differences between the groups (OR 2.63, 95% CI 0.96–7.25, p = 0.06), and the results were homogenous (p = 0.42; I2 = 0%). For the subgroup analysis, in the RCTs (N = 444 patients), there was a significant favouring of AC over AP (OR 6.97, 95% CI 1.25–38.83, p = 0.03), and the results were homogenous (p = 0.37; I2 = 0%). For the cohort studies, there were no significant differences (OR 0.77, 95% CI 0.15–3.98, p = 0.76), and the results were homogenous (p = 0.74; I2 = 0%) (Fig. 4a).

Fig. 4
figure 4

a Forest plot of primary ischaemic stroke. b Forest plot of primary transient ischaemic attack (TIA)

Full size image

3.4.4 Primary Transient Ischaemic Attack (TIA)

Four studies were included in the analysis (N = 836 patients). The analysis showed insignificant statistical differences between the groups (OR 0.54, 95% CI 0.20–1.44, p = 0.22), and the results were homogenous (p = 0.57; I2 = 0%). For the subgroup analysis in the RCTs (N = 444 patients), there were no statistically significant differences between the two groups (OR 0.37, 95% CI 0.08–1.59, p = 0.18), and the results were homogenous (p = 0.59; I2 = 0%). For the cohort studies, there still were no significant differences (OR 0.74, 95% CI 0.20–2.76, p = 0.66), and the results were homogenous (p = 0.27; I2 = 19%) (Fig. 4b).

3.4.5 Recurrent Stroke

Eight studies were included in the analysis (N = 1,140 patients). The analysis showed no significant statistical differences between the two groups (OR 0.97, 95% CI 0.46–2.02, p = 0.93), and the results were homogenous (p = 0.91; I2 = 0%) (Fig. 5a).

Fig. 5
figure 5

a Forest plot of recurrent stroke. b Forest plot of recurrent transient ischaemic attack (TIA)

Full size image

3.4.6 Recurrent TIA

Five studies were included in the analysis (N = 580 patients). The analysis showed non-significant statistical differences between the groups (OR 0.93, 95% CI 0.44–1.98, p = 0.85), and the results were homogenous (p = 0.78; I2 = 0%) (Fig. 5b).

3.4.7 Symptomatic Intracranial Haemorrhage (sICH)

Three studies were included in the analysis (N = 692 patients). For the cohort studies (N = 498), there were no significant differences (OR 0.38, 95% CI 0.12–1.19, p = 0.1), and the results were homogenous (p = 0.94; I2 = 0%). In Engelter's RCT [9], there were no incidences of sICH (Fig. 6a).

Fig. 6
figure 6

a Forest plot of symptomatic Intracranial haemorrhage (ICH). b Forest plot of major extra-cranial bleeding

Full size image

3.4.8 Major Extra-Cranial Bleeding

Three studies were included in the analysis (N = 539 patients). There were no significant statistical differences between the groups (OR 0.45, 95% CI 0.09–2.30, p = 0.34), and the results were homogenous (p = 0.64; I2 = 0%). For the subgroup analysis, in the RCT by Engelter et al. [9] (N = 194 patients), there were no significant statistical differences between AP and AC (OR 0.31, 95% CI 0.01–7.71, p = 0.48). For the cohort studies, there were no significant statistical differences between AP and AC (OR 0.52, 95% CI 0.08–3.45, p = 0.5), and the results were homogenous (p = 0.37; I2 = 0%) (Fig. 6b).

3.4.9 Good Functional Outcome at 6 Months (mRS Score 0–2)

Six studies were included with a total of 803 patients. There were no significant statistical differences between the groups (OR 0.68, 95% CI 0.44–1.06, p = 0.09), and the results were homogenous (p = 0.59; I2 = 0%) (Fig. 7a).

Fig. 7
figure 7

a Forest plot of good functional outcome at 6 months (mRs = 0–2). b Forest plot of poor outcome at 6 months (mRs ≥ 3)

Full size image

3.4.10 Poor Outcome at 6 Months (mRS Score ≥ 3)

Six studies were included in the analysis with a total of 803 patients. The analysis showed marginally significant statistical differences between AC and AP (OR 1.56, 95% CI 1.00–2.42, p = 0.05), and the results were homogenous (p = 0.49; I2 = 0%) (Fig. 7b).

3.4.11 Total Adverse Events

Two studies were included in the analysis, accounting for 492 patients. The results showed no significant statistical differences between AP and AC (OR 0.60, 95% CI 0.32–1.15, p = 0.12), and the results were homogenous (p = 0.89; I2 = 0%). For the subgroup analysis, in the RCT by Engelter et al. [9], there were no significant statistical differences between AP and AC (OR 0.61, 95% CI 0.31–1.20, p = 0.16). For the cohort study by Georgiadis et al. [29], there were no significant differences (OR 0.52, 95% CI 0.06–4.73, p = 0.56). (see Online Supplementary Material (OSM) 1 and Fig. 8).

3.4.12 Complete Recanalization

Five studies were included with a total of 438 patients. There were no significant statistical differences between AP and AC (OR 0.70, 95% CI 0.46–1.06, p = 0.09), and the results were homogenous (p = 0.23; I2 = 29%). For the subgroup analysis, in the RCT by Markus et al. [10] (N = 181 patients), there were no significant statistical differences between AP and AC (OR 0.94, 95% CI 0.52–1.69, p = 0.84). For the cohort studies, there was a significant favouring of AC over AP (OR 0.52, 95% CI 0.29–0.94, p = 0.03), and the results were homogenous (p = 0.35; I2 = 8%) (OSM 1 and Fig. 9).

3.4.13 Other Outcomes

The analysis showed no significant statistical differences between the two groups in any of the major haemorrhagic complications (bleeding) (p = 0.30), all haemorrhagic complications (p = 0.23), admission mRS score 0–2 (p = 0.97), admission mRS score ≥ 3 (p = 0.09), seizures (p = 1.00), occlusion (p = 0.30), stenosis (p = 0.60), time from baseline MRI to follow-up MRI (days) (p = 0.75), and patients with new DWI lesions (p = 0.66). (OSM 1 and Figs. 10–14, respectively).

4 Discussion

CeAD is generally an uncommon cause of ischaemic stroke [8, 29], recurrent ischaemic events [15], and sICH [28], as compared with other aetiologies. However, it has been considered to be one of the common causes of cerebrovascular accidents in the younger population [10, 33]. Over the last two decades, advances in research have elucidated the natural history, pathophysiology and interventional management of CeAD [2, 14, 20, 34]. Nevertheless, clinical trials, reviews and metanalyses have failed to demonstrate the optimal medical management of CeAD [2, 14, 19, 20, 34, 35].

Evolving rationales in medical treatment of CeAD have varied from the expert-based trend of discouraging AP in preference for AC (based on the assumption that stroke occurs via thromboembolic phenomena) [36], to metanalysis and observational study-based favouring of AP (due to the observed haemorrhagic events induced by the AC and hypothesized mural wall haematoma progression because of the AC) [20, 37], versus fairly treating AP/AC on a neutral basis due to lack of any superior evidence for either therapy [14, 34]. A paradigm shift in AC/AP practice has taken place following the first randomized controlled and multicentre CADISS trial [2].

To date, there has been no change in clinical evidence from the CADISS trial, which concluded that AC and AP were not statistically different in terms of adverse effects and clinical outcomes, namely death, ischaemic stroke, TIA or haemorrhage [2, 10, 35].

In our study, we concluded that—at 6 months—there was no statistically significant difference between AC and AP in terms of morbidity (TIA, recurrent strokes, recurrent TIAs, new strokes, sICH, major intracranial bleeding), total adverse events, complete recanalization and mortality, apart from the significant favouring of AC for cases of ischaemic stroke, which is considered a relatively different finding from other trials and was in agreement with findings from Vineetha et al. [11].

Although the consistently negative results may carry an apparent reproducibility among the AC–AP comparative studies, it is very important to recall similar consistently negative trials that later revealed superiority outcomes, such as the WASID trial (Warfarin-Aspirin Symptomatic Intracranial Disease) [38]. Although the scope of the WASID trial is not related to the context of dissection, we speculate that some analogy could be learnt in terms of the research methodology. Prior to the WASID trial, studies testing AC versus AP for intracranial artery stenosis had initial spurious non-significant results due to multiple flaws not initially taken into consideration [39, 40]. The global versus restricted definition of ICH as well as the lack of specification for the exact dosage and scheduling of both AC/AP were among the most striking issues that the WASID investigators utilized to turn the management from a negative trial to supportive evidence for AP in patients suffering from intracranial arterial stenosis (besides increasing the sample size) [38].

Similarly, there are noteworthy areas of conflict that could possibly contribute to the ‘statistically’ comparable AC/AP profiles in CeAD patients, regardless of the inadequate sample size. As mentioned earlier, the included RCTs had a high risk of bias for allocation concealment, and blinding of both participants and personnel, in addition to high risk of bias for outcome assessment blinding and high risk of other bias [9, 10].

We assume that the following two potential domains for bias need to be addressed more precisely in future trials:

  1. 1)

    The definition of true dissection-related stroke (DRS):

    In most of the studies, the definition of DRS was vague, and in some of the cases they were diagnosed merely on the basis of clinical evaluation or telephone surveys [10], in addition to the fact that many of the participants had co-morbidities that could also have caused the stroke, so this may have masked the real picture of genuine dissection-caused strokes [10, 29]. In a few studies, for example in the study of Geniscke et al. [28], the DRS was clearly defined, yet the sample size was low, giving a spuriously elevated percentage of stroke in the 74 patients in addition to the widened confidence intervals [11]. Added to that is the fact that the short-term 5-day MRI-based follow-up and the sole clinical follow up at 6 months have contributed to the vague nature of the stroke (whether ischaemic or haemorrhagic) [28], raising the need for larger population studies with a combined clinical-radiologic definition of ischaemic stroke spectra (territory, size and angiographic data) along the whole follow-up duration [9].

  2. 2)

    The AP/AC scheduling:

    Though there was a statistically significant AC–AP difference in terms of ischaemic stroke occurrence in our study, this finding should be interpreted with caution, as it is very important to recall the myriad of medication-related biases in terms of incomplete adjustment for the confounding effect of AP–AC doses, frequencies and compliance with administration, which should be taken into consideration as they may greatly influence the outcome difference. There are few studies that clearly stated these exact AC/AP full details, including the dose, duration of therapy, use of dual versus single AP therapy, and the compliance profile [9, 34]. Engelter et al. [9] were among the few investigators who clearly revealed the baseline medication data, including doses of 100–300 mg for the APs and the 2–3 INR-guided ACs.

    Moreover, there were methodologic variations with regard to the inclusion criteria for AP/AC selective administration. In some studies, APs were prescribed for sizable stroke cases mainly, in contrast to the TIA-presenting patients who were prescribed ACs [34]. In other trials, most patients with severe carotid stenosis were enrolled in the non-vitamin K-dependent new oral anticoagulants (NOACs) group [26]. To complicate the situation more, some studies, including the CADISS trial itself, had a final treatment design based on the treating physician’s discretion [8, 10]. To sum up, fully detailed medication scheduling data are of utmost importance in future randomized studies, being as important as the consideration of a statistically powerful sample size.

    Other areas of conflict that need to be considered in future trials include: (1) the homogenization and prolongation of follow-up duration for AC/AP arms [15, 17, 27], as a relatively short and variable follow-up duration could act as a possible cause of conflicting results [10, 15, 17, 27]; (2) a more structured inclusion of relevant study subsets, such as paediatric populations [27], patients presenting with non-stroke-related local symptoms, and cases suffering from dissection-related retinal ischaemia [29]; and (3) suggesting more individualized AP/AC management paradigms for traumatic versus spontaneous aetiologies [27], patients presenting with vascular occlusion [28], patients with aspirin intolerance [10], and intracranial versus extracranial dissection locations [10, 27].

4.1 Strengths

Despite the existence of published meta-analyses across observational studies that yielded inconclusive results, our findings appeared to be more conclusive. We aimed to perform a high-quality study to avoid previous limitations. These include, and are not limited to, our inclusion of RCTs as subgroup analysis. Others include: (1) Our inclusion of high-quality observational studies, as low quality or variation in quality appeared to affect the results of previous studies. This is supported by what Sarikaya et al. [20] stated, "Results of our overall analysis suggest an advantage of antiplatelets over anticoagulants on nearly all outcomes,… Stratified analyses according to methodological quality showed less pronounced advantages of antiplatelets in studies of higher methodological quality, with point estimates nearer the line of no difference at 1 and credibility intervals compatible with both relevant advantages and disadvantages of antiplatelets over anticoagulants". (2) The previous studies had small sample sizes and were subjected to selection bias, which is also stated in one of those studies. (3) Some of the previous meta-analyses [41, 42] included studies that had different main aims, therefore, their results and conclusions were not as focused and conclusive as ours.

A meta-analysis across both RCTs and observational studies has recently been published [43]. Their results were consistent with ours, therefore, supporting our methodology and data. However, what differs and adds more value to our study is that we analysed new outcomes, which included the intervention's influence on imaging-based outcomes (such as new ischaemic lesions on DWI sequences). Moreover, we analysed death in a long-term follow-up, which was previously reported as a limitation in some studies.

To recapitulate, our study did not show a considerable difference between the two groups. However, some important variables were reported in a few studies; for example, death and primary ischaemic stroke were only reported in five studies. Also, primary TIA was reported in only four studies. However, all of them exceeded the minimal sample size needed to meet the required statistical constraints. Still, we recommend more well-designed clinical trials to assess if unnecessary anticoagulation can be avoided in CeAD.