Scopus, Embase, Web of Science, and PubMed databases were searched systematically from inception to April 2025 for RCTs comparing TAVR outcomes with and without CEPD. Search terms are presented in Supplementary Table 1. The retrieved studies’ reference lists and related articles were examined to ensure comprehensive coverage. No language restrictions were applied.

RCTs that compared clinical outcomes of TAVR with routine CEPD vs no CEPD were included. Case reports and nonrandomized studies were excluded. The search results were imported into Covidence software (Melbourne, Australia), and duplicate entries were removed. Authors screened all titles and abstracts, removing studies that failed to meet inclusion criteria. Two authors (I.H. and H.E.) screened the full text of the selected studies’. Disagreements were resolved in consultation with the senior author (I.Y.E).

Authors (I.H., H.E., M.H., S.A., M.F., K.A.M., H.E., L.A., R.I., B.M.H., M.E, A.A., E.H., and M.Z.) independently extracted data onto a predefined standardized sheet. Extracted information included study characteristics (first author’s name, publication year, aim, and inclusion/exclusion criteria), participant information (number, demographics, and clinical characteristics), procedural characteristics, and study outcomes. Disagreements were resolved in consultation with the senior author (I.Y.E.).

The primary outcome of interest was any stroke following TAVR. Secondary outcomes included disabling stroke, new brain lesions on magnetic resonance imaging (MRI), all-cause mortality, acute kidney injury, vascular complications, and worsening Montreal Cognitive Assessment (MoCA) score.

Two authors (S.A. and M.S.A.) independently assessed the risk of bias for each included study using the Cochrane RoB-2 tool for randomized trials. The risk of bias was categorized as low, some concerns, or high. Additionally, the same investigators independently evaluated certainty of evidence using the GRADE (Grading of Recommendations Assessment, Development, and Evaluation) framework, categorizing outcomes as very low, low, moderate, or high certainty. Disagreements were resolved in consultation with the senior author (I.Y.E).

Pooled risk ratios (RRs) with 95% confidence intervals (CIs) were calculated for dichotomous outcomes, while mean differences were calculated for continuous outcomes. Random-effects models were used for all outcomes, with results confirmed by fixed-effects model sensitivity analyses for robustness. Statistical heterogeneity was assessed using the I ² statistic and the Chi-squared test, with I ² > 50% indicating substantial heterogeneity. Subgroup analyses were conducted for each outcome to evaluate the efficacy of CEPDs by device mechanism (filter-based vs shield-based). Additionally, meta-regression analysis was performed to assess effect modification by the following covariates: mean age, percentage of females, and percentage of patients with diabetes mellitus, history of prior stroke, or atrial fibrillation. A leave-one-out sensitivity analysis was conducted to evaluate the impact of individual studies on overall estimates. All analyses followed Cochrane guidelines, and a P -value <.05 was considered statistically significant. Meta-analyses were conducted using the R software (version 4.4.2, http://www.r-project.org ) meta package.

Trial sequential analysis (TSA) was performed using Trial Sequential Analysis software (version 0.9.5.9 Beta, Copenhagen, Denmark) to calculate the required information size (RIS) and to control type I (α = 5%) and type II (β = 20%) error. A random-effects model (DerSimonian–Laird estimator) with 95% CI was applied, incorporating diversity-adjusted RIS to account for between-study heterogeneity. The RIS was derived from a predefined 35% relative risk reduction (RRR) based on the recent BHF PROTECT-TAVI trial. As a sensitivity analysis, we additionally assumed a smaller, clinically meaningful RRR of 20% to assess whether a more modest treatment effect could be detected or ruled out with the available evidence. Sequential monitoring boundaries were defined by the O’Brien–Fleming α-spending function. Z-curves were constructed to assess the effectiveness, potential harm, or lack of benefit of CEPD by categorizing the outcomes into 4 distinct groups based on statistical significance and futility boundaries: i) true positives: wherein the z-curve surpasses both conventional and TSA thresholds, thereby indicating a statistically significant benefit or harm; ii) false positives: where the z-curve exceeds the conventional threshold but not the TSA threshold, suggesting the observed effect may be attributable to random error; iii) false negatives: characterized by the z-curve falling below both thresholds without intersecting the futility boundary, implying a potential true effect may have been missed; and iv) true negatives: wherein the z-curve falls below both thresholds and intersects the futility boundary, indicating the absence of a significant benefit or harm.

The reverse fragility index (RFI) was calculated for meta-analytic primary outcomes to quantify the robustness of statistically nonsignificant results. The RFI represents the minimum number of patients in one or more trials within a meta-analysis that would need to have their event status changed (i.e., event to nonevent or vice versa) to alter the statistical significance of the pooled estimates ( P <.05). RFI categories included: 1-4 (not robust), 5-7 (somewhat robust), 8-14 (robust), and >14 (highly robust), providing a framework to interpret the stability of nonsignificant results.

The study selection process is summarized in Figure 1 . From 2,870 records identified by the initial search, 8 RCTs with 11,596 patients (5,933 with CEPD and 5,663 without CEPD) met the inclusion criteria. Seven studies followed patients for 30 days, while the BHF PROTECT-TAVI trial had a follow-up of 3 days. Transfemoral access was the predominant approach, used in 98.9% of patients. The mean age of study participants ranged from 78.1 to 84.2 years, with a slightly higher proportion of male patients in most trials (female representation ranged from 33.3% to 58%). The rate of successful CEPD deployment was consistently high, ranging from 81.2% to 100%. CEPD were categorized into 2 mechanism types: filter-based (capture) systems and shield-based (deflection) systems. The MISTRAL-C, CLEAN-TAVI, PROTECTED TAVR, SENTINEL, and BHF PROTECT-TAVI trials employed filter-based devices, while DEFLECT III, REFLECT I, and REFLECT II trials used shield-based devices ( Table 1 ).

Flow diagram for study selection

This figure illustrates the search strategy and study selection process for the meta-analysis.

All included studies, except the PROTECTED TAVR trial (17), used the Valve Academic Research Consortium-2 (VARC-2) definition of stroke, which is based on clinical evaluation and classifies strokes as disabling or nondisabling using the modified Rankin Scale. In contrast, the PROTECTED TAVR trial used the Neurologic Academic Research Consortium (NeuroARC) definition, which combines symptom-based and tissue-based criteria and incorporates systematic neuroimaging. Unlike VARC-2, which relies mainly on clinical assessment and disability grading, the NeuroARC offers a more comprehensive classification of neurologic events. It categorizes neurologic injury into overt (acutely symptomatic) central nervous system (CNS) injury (Type 1), covert (acutely asymptomatic) CNS injury (Type 2), and neurologic dysfunction without imaging evidence of CNS injury (Type 3), thus capturing a broader range of neurologic complications.

Primary outcome

The outcome of any stroke was available in all studies. There was no difference in the incidence of any stroke with CEPD compared with no CEPD (2.73% vs 2.66%; RR 0.92, 95% CI 0.75–1.14; P =.40). No heterogeneity was observed across the studies ( I ² = 0%, tau2 = 0) ( Figure 2 A).

Efficacy of cerebral embolic protection devices on stroke risk during TAVR, stratified by device type forest plots showing the effect of cerebral embolic protection devices (CEPD) versus control on A, primary outcome: any cause stroke and B, disabling stroke in patients undergoing transcatheter aortic valve replacement (TAVR). Trials are grouped by CEPD mechanism: shield-based and filter-based. Squares denote individual study risk ratios (RR) with 95% confidence intervals (CI); diamonds represent pooled estimates under fixed- and random-effects models. Red bars beneath diamonds indicate the prediction intervals for random effects. Subgroup effects and overall heterogeneity statistics are displayed at the bottom of each panel.

Secondary outcomes

The incidence of disabling stroke was also comparable between the CEPD and no CEPD groups (1.11% vs 1.37%; RR 0.80; 95% CI 0.55–1.15; P =.18). No heterogeneity was observed across the studies ( I ² = 0%, tau2 = 0) ( Figure 2 B).

The incidence of new MRI-detected brain lesions was similar between the CEPD and no CEPD groups (85.86% vs 87.9%; RR: 1.00; 95% CI: 0.93–1.07; P =.95). No heterogeneity was observed across studies ( I ² = 0%). Cognitive decline, measured by worsening MoCA score, showed no statistically significant difference between groups (19.07% vs 25%; RR: 0.95, 95% CI: 0.55–1.65; P =.80), with no heterogeneity ( I ² = 0%). These findings should be interpreted cautiously because both outcomes were evaluated in relatively small subgroups.

All-cause mortality was comparable between the CEPD and no CEPD groups (0.82% vs 0.73%; RR: 1.04; 95% CI: 0.71–1.51; P =.81). No heterogeneity was detected across studies ( I ² = 0, tau2 = 0). The incidence of acute kidney injury was similar between the CEPD and no CEPD groups, with no statistically significant difference (2.73% vs 2.39%; RR: 1.18; 95% CI: 0.95–1.48; P =.11). No heterogeneity was observed across studies ( I ² = 0%). Vascular complications were comparable between the groups (3.43% vs 2.22%; RR: 1.16, 95% CI: 0.56–2.41; P =.64), with moderate heterogeneity ( I ² = 37.1%, tau2 = 0.241) ( Figure 3 , Supplementary Figures 1-5).

Forest plot of primary and secondary outcomes comparing cerebral embolic protection devices (CEPDs) vs control.

Risk ratios were calculated using random-effects models. Horizontal lines represent 95% CI. The vertical dotted line represents a risk ratio of 1.0 (no difference between CEPD and Control).

A subgroup analysis was conducted to evaluate the efficacy of CEPDs based on device type mechanism (filter-based vs shield-based). A statistically significant subgroup interaction was observed, indicating a differential effect between the 2 device types, with shield-based devices exhibiting a higher RR for both any stroke ( P =.008) and disabling stroke ( P =.002) compared to filter-based devices ( Figure 2 ). However, this subgroup finding should be interpreted with caution and considered hypothesis-generating rather than confirmatory. No significant subgroup differences were observed for mortality, acute kidney injury, new MRI lesions, or MoCA score worsening ( P >.05). Sensitivity analysis was conducted to compare the findings of a random-effect and fixed-effects model for each outcome, which yielded consistent findings (Supplementary Figures 1-5).

Despite the clinical heterogeneity across trials, including different CEPD mechanisms (filter-based vs shield-based), valve types, follow-up, and imaging protocols, statistical heterogeneity was consistently low ( I ² = 0% for most outcomes).

Leave-one-out sensitivity analyses demonstrated that no single trial disproportionately influenced the pooled estimates, with consistent results across all outcomes confirming the reliability and robustness of the findings ( Figure 4 , Supplementary Figures 6-11).

Leave-one-out sensitivity analysis for primary outcome (any stroke)

Leave-one-out sensitivity analysis for any stroke. Each row shows the pooled risk ratio and 95% confidence interval when the specified trial is excluded from the meta-analysis. The vertical dashed line represents a risk ratio of 1.0 (no effect).

Stay updated, free articles. Join our Telegram channel

Join