Abstract
Background:
Cerebral embolic protection devices (EPDs) were developed to mitigate the risk of stroke during transcatheter aortic valve replacement (TAVR), but their benefit remains unproven. In the PROTECTED-TAVR trial, EPD use did not reduce periprocedural stroke (primary study outcome) but led to a 62% reduction in the secondary endpoint of disabling stroke. Given these results, the impact of EPDs during TAVR remains unclear.
Methods:
We used STS/ACC TVT registry data to examine the association between EPD use and a proxy for disabling stroke among transfemoral TAVR patients between 1/2018–6/2023. The primary outcome was in-hospital disabling stroke—defined as stroke associated with either in-hospital death or discharge to a non-home location. We evaluated the association between EPD use and disabling stroke using instrumental variable (IV) analysis with site-level preference for EPD use as the instrument—a quasi-experimental approach that can support causal inference. In addition, we performed a propensity-score based comparison using overlap weighting as a secondary analysis.
Results:
The study population consisted of 414,649 patients of whom 53,389 (12.9%) received an EPD. The unadjusted rate of in-hospital disabling stroke was 0.7% among the EPD group and 0.9% in the no EPD group. EPD use was associated with a reduction in disabling stroke in both IV analysis (RR 0.87; 95% CI: 0.73–1.00) and propensity-weighted (PW) analysis (OR 0.79; 95% CI: 0.70–0.90) but was not associated with a reduction in non-disabling stroke. In subgroup analyses, the benefit of EPD was greater among those with versus without prior stroke (interaction p<0.05 for IV and PW).
Conclusions:
In the largest study to date, among patients undergoing TAVR, EPD use was associated with a small, borderline significant reduction in stroke associated with death or a discharge to a non-home location (a proxy for disabling stroke) that is likely to be causal in nature. Taken together with previous mechanistic and clinical studies, these findings provide credible evidence that EPDs benefit patients undergoing TAVR.
Keywords: TAVR, stroke, outcomes research, instrumental variable, outcomes
Graphical Abstract
Over the last 5 years, transcatheter aortic valve replacement (TAVR) has become the dominant form of aortic valve replacement in the United States.1 Nonetheless, despite continued technologic iteration and procedural refinements, TAVR-related stroke remains a feared procedural complication with potentially devastating consequences.2,3 Cerebral embolic protection devices (EPDs) have been developed in an attempt to mitigate this risk.
Although the first EPDs were approved more than 6 years ago, data on the ability of EPDs to improve clinical or patient-centered outcomes remain inconclusive. A prior observational study of 123,186 patients in the TVT registry who underwent TAVR between 2018 and 2019 did not find an association between EPD use and in-hospital stroke in the primary instrumental variable (IV) analysis, and found only a modestly lower risk of in-hospital stroke in the secondary propensity-weighted analysis.4 More recently, the PROTECTED-TAVR trial (NCT04149535) randomized 3000 patients undergoing transfemoral TAVR to use of the Sentinel EPD (Boston Scientific, Natick, MA) vs. no EPD and found that the use of an EPD did not reduce periprocedural stroke (the primary study outcome), but was associated with a reduction in the exploratory secondary endpoint of disabling stroke (which occurred in 0.5% of the patients in the EPD group vs. 1.3% of those in the control group).5 In light of these conflicting results, there remains considerable controversy on how to interpret this trial and inform use of EPDs in practice.
In particular, given the exploratory nature of the disabling stroke benefit in the PROTECTED-TAVR trial, it is important to understand if its findings regarding disabling stroke can be replicated in other high-quality datasets. To address this gap in knowledge, we performed a rapid real-world analysis of the association between EPD use and in-hospital disabling stroke-- defined as a stroke with either in-hospital death or discharge to a location other than home-- in the TVT Registry to serve as a complement to the published results of this randomized trial and to further guide clinical practice.
METHODS
Study overview.
The methods of this study are virtually identical to our previous TVT study on this topic,4 except for the use of in-hospital disabling stroke as the primary outcome, the inclusion of additional years of data, and inclusion of whether a site was a stroke center as a covariate. We used data from the Society of Thoracic Surgeons/American College of Cardiology TVT Registry, which collects data on patient demographics, comorbidities, and outcomes using standardized definitions on nearly all TAVR procedures performed in the US outside of clinical trials.6 The TVT Registry undergoes a 10% random audit on a yearly basis by a third party. Data will not be made available as per data use agreements with the TVT Registry.
Patient population.
The study cohort included consecutive adults who underwent isolated elective or urgent first native or valve-in-valve transfemoral TAVR between January 2018 and June 2023. As in our previous study, we excluded patients from hospitals that performed <20 TAVR procedures in a given calendar year and patients with missing data on EPD use or the primary endpoint. We also excluded patients who received an EPD other than Sentinel, as Sentinel was the only approved device on the market, and these patients were likely in clinical trials. Finally, we excluded patients with discharge location listed as ‘against medical advice’ or ‘other location,’ owing to uncertainty regarding stroke severity for these patients. This study was granted a waiver of written informed consent and authorization by Advarra, a central Institutional Review Board, as it was a retrospective analysis of de-identified patient data from the TVT registry.
Outcomes.
The prespecified primary outcome was in-hospital disabling stroke, which was defined as in-hospital stroke associated with either in-hospital death or discharge to a location other than home. This novel definition was derived from pooled patient-level data from 6 large TAVR trials that incorporated systematic neurologic assessment to screen for stroke (PARTNER 2A (NCT01314313), PARTNER 3 (NCT02675114), Sapien S3i Registry (NCT03222128), SURTAVI (NCT01586910), Evolut Low Risk (NCT02701283), and PROTECTED-TAVR(NCT04149535)).7 When limited to trials conducted after 2017, the sensitivity of this definition was 73% and the specificity was 90%.
Secondary outcomes included in-hospital stroke and in-hospital non-disabling stroke, which was defined as any in-hospital stroke that did not meet the definition of a disabling stroke. We also examined a prespecified falsification endpoint of in-hospital gastrointestinal (GI) or genitourinary (GU) bleeding to assess for residual confounding in our analyses, since it would be unlikely to differ because of EPD use but could differ due to persistent unmeasured differences between treatment groups despite statistical adjustment.
Statistical Analysis.
The main exposure was use of an EPD, which consisted exclusively of the Sentinel device as it was the only EPD approved for use in the United States at the time of this study. We examined trends in the proportion of sites using an EPD and the proportion of TAVR procedures performed using an EPD by calendar quarter graphically. We then examined variation in the proportion of TAVR procedures using an EPD across hospitals over the study period and in the first two quarters of 2023. Baseline characteristics of patients treated with and without an EPD were compared using standardized mean differences, with a threshold of at least 0.10 considered to define a meaningful difference.8
As in our previous study, we evaluated the association between EPD use and outcomes using 2 complementary techniques: (1) instrumental variable (IV) analysis with site-level preference for EPD use as the instrument; and (2) a propensity-score based comparison using overlap weighting. In contrast to propensity-score weighting, IV analysis can address both measured and unmeasured confounding by taking advantage of naturally occurring variation in treatment patterns in practice that more closely approximates formal randomization.9 We defined site-level preference for use of EPD as a hospital’s proportion of use of an EPD for all qualifying procedures during the calendar quarter of a procedure, which was recalculated for each site each quarter.
To perform the IV analysis, we used standard two-stage least squares regression. Each stage included all of the variables included in the previously validated TVT registry in-hospital stroke model along with several additional patient-level and hospital covariates (Table S1).10 Absolute risk estimates and relative risk ratios were calculated based on the average of the predicted event rates with and without EPD across all patients in the study cohort regardless of treatment received. The Wald F-test was used to assess the strength of an instrument (an F-statistic greater than 10 is generally considered to represent a strong instrument).11 To assess the exogeneity of the instrument (i.e. whether the instrument is correlated with the regression model error term), we compared baseline patient characteristics according to whether the treating hospital used any EPD during the calendar quarter of the procedure. We calculated standardized mean differences between these two groups with a threshold of at least 0.10 used to define a meaningful difference.8
For our secondary analysis, we performed a propensity score-based analysis using overlap weights with logistic regression to evaluate the association of EPD use with clinical outcomes. This analysis included the same variables as the IV model in both the calculation of the propensity score and as covariates in the logistic regression model (doubly robust). The analysis also accounted for clustering of outcomes within hospitals by means of the generalized estimating equations framework.12,13 We calculated absolute risk estimates and risk differences based on the mean predicted probability of an outcome when all patients were assumed to receive an EPD or to not receive an EPD.
In exploratory analyses, we repeated the IV and propensity score-based analyses for subgroups based on history of prior stroke, STS surgical mortality risk categories (high/extreme, intermediate, low), history of peripheral arterial disease, age (<80 vs. ≥80), valve morphology (bicuspid, tricuspid, or other), and whether valve-in-valve TAVR was performed.
In sensitivity analyses, we repeated the IV and propensity score-based analyses after reclassifying patients who experienced an in-hospital stroke along with a major vascular complication, unplanned vascular surgery or intervention, unplanned cardiac surgery or intervention, or retroperitoneal hematoma and discharge to a non-home location as a non-disabling stroke instead of disabling stroke as in our primary analysis. These concomitant events were chosen because they could also explain discharge to a non-home location and could confound ascertainment of disabling stroke using our primary definition.
Single imputation using the sample median for continuous variables and mode for categorical variables was used to address missing data in baseline characteristics (missing in <2% for all covariates). A 2-sided p-value <0.05 was considered statistically significant without adjustment for multiple comparisons. Analyses were performed using SAS 9.4 (Cary, NC, USA).
RESULTS
Patient Population and EPD Use.
Between January 2018 and June 2023, 438,479 adult patients underwent elective or urgent first isolated TAVR and were included in the TVT registry. After excluding patients meeting 1 or more of the exclusion criteria, our final analytic sample included 414,649 patients from 808 sites (Figure S1).
The proportion of all TAVR procedures using EPD increased steadily from 5% in Q1 2018 to 15% in Q3 2020. It then stabilized until Q4 2022 when it dropped to 12% after publication of the PROTECTED-TAVR trial (Figure 1). The proportion of sites using at least one embolic protection device during the calendar quarter generally increased from 7% in Q1 2018 to 39% in Q2 2022 and subsequently decreased to 34% after the PROTECTED-TAVR trial results were known. There remained substantial variation in EPD use across hospitals throughout the study period, with 6.3% of sites performing >50% of TAVR procedures with an EPD and 39.8% performing no procedures with an EPD (Figure 2A). Such variation persisted in Q1–2 2023, during which 7.6% of sites performed >50% of TAVR procedures with an EPD, while 60.5% performed no TAVR procedures with an EPD (Figure 2B).
Figure 1.
Embolic protection device use during TAVR by calendar quarter. The blue line represents the proportion of sites using an embolic protection device at least once during TAVR by calendar quarter. The orange line represents the proportion of TAVR patients receiving an embolic protection device during valve implantation by calendar quarter.
Figure 2.
Variation in use of embolic protection devices across hospitals. A) Variation in use of embolic protection devices across hospitals among patients undergoing transfemoral TAVR over the entire study period (January 2018-June 2023). B) Variation in use of embolic protection devices across hospitals among patients undergoing transfemoral TAVR in Q1-Q2 2023. X-axis represents unique sites sorted by use of EPD from highest to lowest.
Patient Characteristics.
Baseline characteristics of patients receiving an EPD and not receiving an EPD were generally similar, although patients who received an EPD were more likely to have had a bicuspid aortic valve or prior aortic valve procedure or replacement and more likely to have received conscious sedation (rather than general anesthesia) for their TAVR procedure (Table 1). Most hospital characteristics differed between patients who did and did not receive EPD during TAVR. Patients who received an EPD were more likely to have been treated at a teaching hospital, a hospital designated as a stroke center, or a hospital located in an urban area in the West or Midwest, with a greater number of beds and higher annualized TAVR volumes. Hospital-level historical stroke rates in 2017 (prior to meaningful adoption of EPD) were also slightly higher at sites that treated patients with EPDs.
Table 1.
Baseline characteristics
| EPD Usage (N = 53,389) | No EPD Usage (N = 361,260) | Standardized mean difference | |
|---|---|---|---|
| Demographics | |||
| Age, years | 78.0±8.8 | 78.7±8.6 | −0.0818 |
| Female | 20775/53388 (38.9%) | 158265/361244 (43.8%) | −0.0996 |
| Clinical characteristics | |||
| Body surface area, m2 | 1.9±0.3 | 1.9±0.3 | 0.0439 |
| Current/recent smoker (< 1 year) | 2978/53252 (5.6%) | 21323/360024 (5.9%) | −0.0142 |
| Porcelain aorta | 651/53358 (1.2%) | 4373/360975 (1.2%) | 0.0008 |
| Peripheral arterial disease | 10195/53372 (19.1%) | 69174/361073 (19.2%) | −0.0014 |
| Glomerular filtration rate ml/min/1.73m2 | 64.7±24.1 | 62.7±25.0 | 0.0819 |
| NYHA Class IV | 4837/52918 (9.1%) | 32882/358257 (9.2%) | −0.0013 |
| Hemodialysis | 1170/53350 (2.2%) | 13209/360878 (3.7%) | −0.0871 |
| Severe/chronic lung disease | 2479/51205 (4.8%) | 21160/347267 (6.1%) | −0.0551 |
| Aortic valve morphology | 0.1299 | ||
| Bicuspid | 4710/52986 (8.9%) | 19861/358765 (5.5%) | |
| Tricuspid | 45372/52986 (85.6%) | 318213/358765 (88.7%) | |
| Other/uncertain | 2904/52986 (5.5%) | 20691/358765 (5.8%) | |
| Past medical history | |||
| Prior transient ischemic attack | 3949/53358 (7.4%) | 25153/360930 (7.0%) | 0.0167 |
| Prior stroke | 6007/53375 (11.3%) | 36788/361118 (10.2%) | 0.0345 |
| Prior TAVR | 284/53360 (0.5%) | 1290/360956 (0.4%) | 0.0263 |
| Prior AVR | 4428/53376 (8.3%) | 19432/361076 (5.4%) | 0.1156 |
| Prior aortic valve procedure | 6183/53377 (11.6%) | 29351/361136 (8.1%) | 0.1162 |
| Prior non-aortic valve procedure | 1078/53389 (2.0%) | 6285/361260 (1.7%) | 0.0206 |
| Procedural characteristics | |||
| Surgical risk category | 0.0423 | ||
| Inoperable/Extreme Risk | 2837/53110 (5.3%) | 20376/359508 (5.7%) | |
| High risk | 18413/53110 (34.7%) | 120001/359508 (33.4%) | |
| Intermediate risk | 19285/53110 (36.3%) | 136878/359508 (38.1%) | |
| Low risk | 12575/53110 (23.7%) | 82253/359508 (22.9%) | |
| Days between procedure date and November 1, 2011 | 3418.0±514.7 | 3338.3±572.2 | 0.1463 |
| Pre-procedural shock | 335/53361 (0.6%) | 2486/360990 (0.7%) | −0.0075 |
| Inotropes | 754/52902 (1.4%) | 6988/359325 (1.9%) | −0.0404 |
| Pre-procedure mechanical assist device | 103/53346 (0.2%) | 994/360920 (0.3%) | −0.0170 |
| Anesthesia type | 0.1573 | ||
| Conscious Sedation | 38381/53327 (72.0%) | 233692/360836 (64.8%) | |
| General Anesthesia | 14841/53327 (27.8%) | 125726/360836 (34.8%) | |
| Other (epidural/combination) | 105/53327 (0.2%) | 1418/360836 (0.4%) | |
| Hospital characteristics | |||
| Annualized TAVR volume (median, IQR) | 210.2 (122.5, 332.2) | 134.3 (89.1, 222.8) | 0.5958 |
| Hospital stroke rate in 2017, % | 1.9±1.4 | 1.7±1.5 | 0.1553 |
| Hospital disabling stroke rate in 2017, % | 1.3±1.1 | 1.2±1.3 | 0.0904 |
| Stroke center | 21008/53389 (39.3%) | 97193/361260 (26.9%) | 0.2668 |
| Teaching hospital | 41138/53389 (77.1%) | 231420/361260 (64.1%) | 0.2880 |
| Location | 0.1880 | ||
| Urban | 37451/53389 (70.1%) | 229650/361260 (63.6%) | |
| Suburban | 11013/53389 (20.6%) | 103491/361260 (28.6%) | |
| Rural | 4925/53389 (9.2%) | 28119/361260 (7.8%) | |
| Type | 0.1752 | ||
| Private/Community | 37243/53389 (69.8%) | 279666/361260 (77.4%) | |
| University | 15132/53389 (28.3%) | 77149/361260 (21.4%) | |
| Government | 1014/53389 (1.9%) | 4445/361260 (1.2%) | |
| Region | 0.2969 | ||
| Northeast | 9290/53272 (17.4%) | 74660/359891 (20.7%) | |
| West | 15113/53272 (28.4%) | 76276/359891 (21.2%) | |
| Midwest | 15751/53272 (29.6%) | 80733/359891 (22.4%) | |
| South | 13118/53272 (24.6%) | 128222/359891 (35.6%) | |
| Number of beds | 647.8±329.4 | 537.4±287.8 | 0.3572 |
Unless otherwise indicated, mean (standard deviation) is displayed for the continuous variables. EPD = embolic protection device; NYHA = New York Heart Association; TAVR = transcatheter aortic valve replacement; AVR = aortic valve replacement; IQR = interquartile range.
When patients were compared according to whether their treating hospital used an EPD during the same quarter as the TAVR, there were residual differences in hospital characteristics, type of anesthesia, and procedure date, but all other patient characteristics were similar (Tables S2–S3). These findings suggest that the instrument (site proportion of EPD use within the calendar quarter) adequately categorized patients into groups independent of baseline patient characteristics. The Wald F-statistic was 414,826, consistent with a strong instrument.
Unadjusted Analyses.
In unadjusted analyses, EPD use was associated with lower rates of in-hospital disabling stroke (0.7% vs 0.9%, p<0.01) and in-hospital stroke (1.2% vs 1.4%, p=0.03; Table 2). However, EPD use was not associated with in-hospital non-disabling stroke (0.5% for both groups, p=0.54) or GI/GU bleeding (0.3% vs 0.4%, p=0.19).
Table 2.
Unadjusted outcomes
| EPD Usage (N = 53,389) | No EPD Usage (N = 361,260) | Odds Ratio (95% CI) | P-value | |
|---|---|---|---|---|
| Primary endpoint | ||||
| In-hospital disabling stroke | 383/53389 (0.7%) | 3336/361260 (0.9%) | 0.78 (0.66, 0.91) | 0.002 |
| Secondary endpoints | ||||
| In-hospital stroke | 655/53389 (1.2%) | 5092/361260 (1.4%) | 0.87 (0.76, 0.99) | 0.031 |
| In-hospital non-disabling stroke | 272/53389 (0.5%) | 1756/361260 (0.5%) | 1.05 (0.9, 1.22) | 0.543 |
| Falsification Endpoint | ||||
| GI/GU bleeding | 176/53388 (0.3%) | 1356/361245 (0.4%) | 0.88 (0.72, 1.07) | 0.194 |
Odds ratios, 95% CIs, and Wald Chi-square p-values obtained from unadjusted generalized estimating equations, accounting for clustering within site.
EPD = embolic protection device
Adjusted Analyses.
In the primary IV analysis, EPD use was associated with a borderline significant reduction in disabling stroke (adjusted relative risk (RR) 0.87; 95% CI: 0.73–1.00; Table 3). However, there was no significant difference in the adjusted rate of in-hospital stroke (RR 0.90, 95% CI: 0.79–1.01), in-hospital non-disabling stroke (RR 0.97, 95% CI: 0.78–1.16), or GI/GU bleeding (RR: 1.04, 95% CI: 0.82–1.26).
Table 3.
Adjusted association of in-hospital outcomes with use of cerebral embolic protection during TAVR
| Instrumental variable analysis model | Propensity score-based model | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| EPD (%) | No EPD (%) | Absolute Risk Diff., % (95% CI) | Adjusted Rel. Risk (95% CI) | p-value | EPD (%) | No EPD (%) | Absolute Risk Diff., % (95% CI) | Adjusted Odds Ratio (95% CI) | p-value | |
| Primary endpoint | ||||||||||
| Disabling stroke | 0.79 | 0.91 | −0.12 (−0.25, 0) | 0.87 (0.73, 1.00) | 0.056 | 0.76 | 0.95 | −0.19 | 0.79 (0.70, 0.90) | < 0.001 |
| Secondary endpoints | ||||||||||
| Any stroke | 1.27 | 1.40 | −0.14 (−0.29, 0.02) | 0.90 (0.79, 1.01) | 0.088 | 1.24 | 1.47 | −0.23 | 0.84 (0.76, 0.93) | 0.001 |
| Non-disabling stroke | 0.48 | 0.49 | −0.01 (−0.11, 0.08) | 0.97 (0.78, 1.16) | 0.78 | 0.48 | 0.52 | −0.04 | 0.92 (0.79, 1.08) | 0.304 |
| Falsification Endpoint | ||||||||||
| GI/GU bleeding | 0.38 | 0.37 | 0.01 (−0.07, 0.09) | 1.04 (0.82, 1.26) | 0.738 | 0.39 | 0.38 | 0.02 | 1.05 (0.86, 1.27) | 0.643 |
EPD = embolic protection device; Rel risk= relative risk; CI = confidence interval; GI = gastrointestinal; GU = genitourinary
For the instrumental variable model, predicted event rates were calculated based on the average of the predicted event rates with and without EPD across all patients in the study cohort (regardless of treatment received).
For the propensity score-based model, predicted event rates were calculated based on the average predicted probability assuming all patients received an EPD or all patients did not.
In the propensity weighted analysis, EPD use was associated with a significant reduction in both disabling stroke (OR 0.79; 95% CI: 0.70–0.90) and in-hospital stroke (OR 0.84, 95% CI: 0.76–0.94). There was no difference in in-hospital non-disabling stroke (OR 0.92, 95% CI: 0.79, 1.08) or GI/GU bleeding (OR: 1.05, 95% CI: 0.86–1.27).
Subgroup and Sensitivity Analyses.
In exploratory subgroup analyses, there was a significant interaction between EPD use and the risk of in-hospital disabling stroke when the analyses were stratified according to history of prior stroke (Figure 3). In IV analysis, EPD use was associated with a significant reduction in disabling stroke among patients with prior stroke (RR 0.59, 95% CI: 0.27–0.92), but not among those without prior stroke (RR 0.92, 95% CI: 0.77–1.08; p=0.03 for interaction). In propensity-weighted analysis, the findings were similar. EPD use was associated with a greater reduction in disabling stroke among patients with prior stroke (OR 0.59, 95% CI: 0.43–0.81) compared with those without prior stroke (OR 0.85, 95% CI: 0.74–0.96; p=0.04 for interaction). There were no significant interactions between EPD use and risk of in-hospital disabling stroke when stratified according to STS surgical risk categories, history of peripheral arterial disease, age <80 versus ≥80, aortic valve morphology, or whether valve-in-valve TAVR was performed in either IV or propensity score-based models. Results were largely unchanged when patients with other complications that could have led to discharge to a non-home location were no longer classified as a disabling stroke (IV absolute risk difference: 0.90, 95% CI: 0.75–1.05; propensity-weighted OR 0.81, 95% CI: 0.71–0.92).
Figure 3.
Subgroup analyses of the primary endpoint. A) Instrumental variable analysis. B) Propensity score weighted analysis. OR = odds ratio. CI = confidence interval. PAD = peripheral arterial disease.
DISCUSSION
In this updated analysis of data from the TVT registry involving more than 400,000 patients treated with transfemoral TAVR, we found that in our primary IV analysis, EPD use was associated with a reduction in risk of stroke associated with death or a discharge to a non-home location (a proxy for disabling stroke), although the degree of benefit was small and of borderline statistical significance—results that were supported by our secondary propensity-score weighted analysis. In addition to demonstrating a treatment benefit in the overall TAVR population, we also identified a patient subset that seemed to derive preferential benefit from embolic protection—patients with a history of prior stroke. Depending on the specific analysis, we found that this subgroup (which represents ~10% of all patients undergoing TAVR) experienced a relative and absolute risk reduction that was 3–5x larger than for the overall study population.
These findings extend and reinforce existing evidence from the PROTECTED-TAVR trial regarding the benefit provided by EPDs in patients undergoing TAVR.5 In particular, our findings help to address the concern that PROTECTED-TAVR may have suffered from selection bias if patients with anatomic features suggestive of a high risk of procedure-related stroke were systematically excluded from randomization. By including all patients undergoing commercial transfemoral TAVR in the US, our study suggests that it is unlikely that such “cherry picking” was a major factor in the PROTECTED-TAVR results.
The 13% relative reduction in disabling stroke in our primary IV analysis among all patients is much smaller than the 62% relative reduction found in PROTECTED-TAVR (disabling stroke with EPD use 0.5%, no EPD use 1.3%), although the 95% confidence interval for this effect extended to an 8% relative risk reduction. 5 Of note, we did not find a significant difference in non-disabling stroke with EPD use in either IV or propensity-weighted analyses, which is congruent with the results of PROTECTED-TAVR, in which the benefit of EPD use was limited to disabling stroke. Notably, both the IV and propensity-weighted effect size for any stroke in our study is similar to our prior TVT Registry analysis examining EPD use and any stroke after TAVR, which did not include patients after 2019.4 Thus, the main difference between the current study and our previous study is that with the inclusion of additional years of data, our confidence intervals around these point estimates are now much tighter. Based on the point estimates from our primary IV analysis, the number needed to treat (NNT) to prevent one disabling stroke is 833, which decreases to 526 when based on the results of our propensity-weighted analysis.
Our study is the first to identify a subgroup of patients that is more likely to benefit from EPD use. We found a significant interaction between the reduction in disabling stroke with EPD use and history of prior stroke. Among patients with prior stroke, EPD use was associated with a 40% lower risk of disabling stroke in both IV and propensity-weighted analyses. In this subset, the NNT to prevent one disabling stroke was 155 based on the point estimates from our primary IV analysis. It is possible that risk factors that predispose to prior stroke, such as vascular calcification, may also predispose to greater valvular debris that could be captured by an EPD during TAVR. Alternatively, it is possible that a prior stroke could alter neuroplasticity such that the impact of a subsequent stroke during TAVR may be more profound. Understanding the mechanisms underlying the differential benefit of EPD use in the subgroup of patients with prior stroke remains a rich area for future inquiry.
Given the modest overall benefit of EPD in our study, our findings support ongoing clinical equipoise for the UK-based BHF PROTECT-TAVI trial (ISRCTN16665769) but suggest that a clinical benefit is unlikely to be found with its intended sample size. PROTECT-TAVI is randomizing patients undergoing TAVR to EPD use vs. no use and is expected to complete enrollment in mid-2026.14 PROTECT-TAVI aims to recruit 7730 patients and has 80% power to detect a 33% relative risk reduction in any stroke assuming an event rate of 3% in the control arm. Based on the effect size observed in our study, however, PROTECT-TAVI, as well as a planned patient-level pooled analysis of PROTECTED-TAVR and PROTECT-TAVI (PROSPERO CRD42022324160), are also likely to be underpowered. Assuming a 3% stroke rate as anticipated and a relative risk reduction of 10% based on our IV analysis, a randomized trial would have to enroll ~97,000 patients to detect a significant difference in the rate of in-hospital stroke with 80% power. Although the primary endpoint of PROTECT-TAVI is all stroke, it will also collect information regarding disabling stroke. Assuming a 1.3% disabling stroke rate as found in PROTECTED-TAVR and a relative risk reduction of 13% based on our IV analysis, a randomized controlled trial would need to enroll ~132,000 patients to detect a significant difference in rate of in-hospital disabling stroke with 80% power.
Another unique aspect of our study is the insights it provides on how clinical trials impact clinical practice. Specifically, we noted that EPD use stabilized after its initial rapid uptake from 2017–2020 but subsequently declined after publication of the PROTECTED-TAVR results. This temporal change demonstrates how clinical practice was meaningfully influenced rapidly after publication of randomized clinical trial data. In 2023, although a small percentage of centers use EPD for most of their cases, most centers still do not use EPD for any TAVR procedures. This persistent variation may reflect differences in how specific centers weigh the uncertain benefits of EPD use against the additional cost of using this device for TAVR procedures.
Limitations.
The findings of this study should be interpreted in light of several important limitations. First, this is an observational study, and the results may suffer from residual confounding, although we used a quasi-experimental design for our primary analysis that should account for both measured and unmeasured confounding. Second, disabling stroke was not captured directly in the TVT registry so we used discharge location as a proxy for stroke severity, which is an imperfect definition. As such, it is possible that our approach misclassified the severity of stroke in patients who were discharged to a non-home location after stroke for another reason. However, this definition of disabling stroke had reasonable sensitivity and high specificity when validated against neurologically adjudicated disabling stroke from 6 large TAVR trials.7 Moreover, results were directionally similar in a sensitivity analysis in which we reclassified patients with other complications that may have led to discharge to a non-home location.
Third, it is possible that the TVT registry underreported stroke given the lack of formal neurologic assessment in routine practice. However, we would expect that this issue would be less likely to affect ascertainment of disabling stroke, since these events are more likely to be clinically apparent than more subtle non-disabling strokes. Moreover, unless underreporting was also correlated with use/non-use of EPD, this would be expected to affect statistical precision but not the relative risk reduction. Finally, these results only apply to the Sentinel EPD as it was the only EPD approved for use in the United States during the study period. As such, our findings may not extend to other EPDs that protect all 4 major cerebral vessels. Notably, in PROTECTED-TAVR, only 1 of 6 disabling strokes in the EPD use group could be localized to the protected territory whereas the other 5 occurred in the distribution of the left vertebral artery or in patients in whom effective cerebral embolic protection could not be maintained throughout the procedure.5
Conclusions.
In conclusion, in the largest study of EPD use during TAVR to date, we found that EPD use was associated with a small reduction in stroke associated with death or a discharge to non-home location (a proxy for disabling stroke) in routine clinical practice that may be enhanced in patients with prior stroke. The fact that these results were derived from an analytic approach (instrumental variable analysis) that can support causal inference and are congruent with those from the PROTECTED-TAVR trial suggest that this benefit is real but the effect size may be relatively small. Nonetheless, we believe these findings provide much needed evidence that EPDs benefit patients undergoing TAVR.
Supplementary Material
Clinical Perspective.
What is Known
The effect of cerebral embolic protection devices (EPDs) during TAVR on stroke and other outcomes after TAVR is unclear.
In the PROTECTED-TAVR trial, EPD use did not demonstrate a reduction in periprocedural stroke (primary study outcome) but led to a 62% reduction in the exploratory secondary endpoint of disabling stroke.
What the Study Adds
In this large observational study, we found that EPD use was associated with a small, borderline statistically significant reduction in the risk of stroke associated with death or a discharge to a non-home location (a proxy for disabling stroke) that was at the lower bound of what was observed in the PROTECTED-TAVR trial.
The relative risk reduction in disabling stroke with EPD use was amplified among the subset of patients with prior stroke.
These findings provide evidence that supports a true clinical benefit of EPD use for patients undergoing TAVR, limited to prevention of disabling stroke
SOURCES OF FUNDING
This work was funded in part by an investigator-initiated grant from Boston Scientific. Boston Scientific had no role in the study design, analysis, interpretation of study results, writing the manuscript, or decision for publication.
Non-standard Abbreviations and Acronyms
- TAVR
transcatheter aortic valve replacement
- EPD
embolic protection device
- IV
instrumental variable
- GI
gastrointestinal
- GU
genitourinary
- NNT
number needed to treat
Footnotes
DISCLOSURES
The views expressed in this article are those of the authors and do not necessarily reflect the position or policy of the Department of Veterans Affairs, the U.S. Government, or National Institutes of Health.
Dr. Butala is supported by the Boettcher Foundation Webb-Waring Biomedical Research grant and reports consulting fees from Shockwave Medical and consulting fees and ownership interest in HiLabs, outside the current work.
Dr. Yeh is a Special Government Employee of the US Food and Drug Administration. He has institutional research grants with the US FDA, Boston Scientific, Abbott Vascular and Medtronic. He is a consultant for Abbott Vascular, Boston Scientific, Elixir Medical, InfraRedx, Medtronic, Shockwave Medical, and Zoll.
Dr. Messenger reports institutional grant support from Philips Medical Systems and Medtronic.
Dr. Cohen reports institutional research grants from Boston Scientific, Edwards Lifesciences, Medtronic, Abbott, Zoll, IRhythm, Corvia, Philips, CathWorks, and Ancora. He is a consultant to Abbott, Boston Scientific, Edwards Lifesciences, and Heartbeam.
Dr. Vemulapalli reports grants / contracts from: American College of Cardiology, Society of Thoracic Surgeons, National Institutes of Health (R01 and UG3), Cytokinetics, Abbott Vascular, Boston Scientific. Consulting / Advisory Board: Astra Zeneca, Medtronic, Boehringer Ingelheim, Veralox Therapeutics, Icon, HeartFlow, Total CME
REFERENCES
- 1.Otto CM. Informed Shared Decisions for Patients with Aortic Stenosis. N Engl J Med. 2019;380:1769–1770. doi: 10.1056/NEJMe1903316 [DOI] [PubMed] [Google Scholar]
- 2.Muralidharan A, Thiagarajan K, Van Ham R, Gleason TG, Mulukutla S, Schindler JT, Jeevanantham V, Thirumala PD. Meta-Analysis of Perioperative Stroke and Mortality in Transcatheter Aortic Valve Implantation. Am J Cardiol. 2016;118:1031–1045. doi: 10.1016/j.amjcard.2016.07.011 [DOI] [PubMed] [Google Scholar]
- 3.Eggebrecht H, Schmermund A, Voigtlander T, Kahlert P, Erbel R, Mehta RH. Risk of stroke after transcatheter aortic valve implantation (TAVI): a meta-analysis of 10,037 published patients. EuroIntervention. 2012;8:129–138. doi: 10.4244/EIJV8I1A20 [DOI] [PubMed] [Google Scholar]
- 4.Butala NM, Makkar R, Secemsky EA, Gallup D, Marquis-Gravel G, Kosinski AS, Vemulapalli S, Valle JA, Bradley SM, Chakravarty T, et al. Cerebral Embolic Protection and Outcomes of Transcatheter Aortic Valve Replacement: Results From the Transcatheter Valve Therapy Registry. Circulation. 2021;143:2229–2240. doi: 10.1161/circulationaha.120.052874 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Kapadia SR, Makkar R, Leon M, Abdel-Wahab M, Waggoner T, Massberg S, Rottbauer W, Horr S, Sondergaard L, Karha J, et al. Cerebral Embolic Protection during Transcatheter Aortic-Valve Replacement. New Eng J Med. 2022; 387:1253–1263. doi: 10.1056/NEJMoa2204961 [DOI] [PubMed] [Google Scholar]
- 6.Carroll JD, Edwards FH, Marinac-Dabic D, Brindis RG, Grover FL, Peterson ED, Tuzcu EM, Shahian DM, Rumsfeld JS, Shewan CM, et al. The STS-ACC transcatheter valve therapy national registry: a new partnership and infrastructure for the introduction and surveillance of medical devices and therapies. J Am Coll Cardiol. 2013;62:1026–1034. doi: 10.1016/j.jacc.2013.03.060 [DOI] [PubMed] [Google Scholar]
- 7.Butala NM, Yeh RW, Kapadia S, Cohen DJ. Use of Discharge Disposition to Determine Stroke Severity after TAVR. Manuscript in process. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Austin PC. Using the standardized difference to compare the prevalence of a binary variable between two groups in observational research. Commun Stat-Simul C. 2009;38:1228–1234. [Google Scholar]
- 9.Cameron AC, Trivedi PK. Microeconometrics: Methods and Applications. Cambridge University Press; 2005. [Google Scholar]
- 10.Thourani VH, O’Brien SM, Kelly JJ, Cohen DJ, Peterson ED, Mack MJ, Shahian DM, Grover FL, Carroll JD, Brennan JM, et al. Development and Application of a Risk Prediction Model for In-Hospital Stroke After Transcatheter Aortic Valve Replacement: A Report From The Society of Thoracic Surgeons/American College of Cardiology Transcatheter Valve Therapy Registry. Ann Thorac Surg. 2019;107:1097–1103. doi: 10.1016/j.athoracsur.2018.11.013 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Staiger DO, Stock JH. Instrumental variables regression with weak instruments. In: National Bureau of Economic Research; Cambridge, Mass., USA; 1994. [Google Scholar]
- 12.Thomas LE, Li F, Pencina MJ. Overlap Weighting: A Propensity Score Method That Mimics Attributes of a Randomized Clinical Trial. JAMA. 2020;323:2417–2418. doi: 10.1001/jama.2020.7819 [DOI] [PubMed] [Google Scholar]
- 13.Li F, Morgan KL, Zaslavsky AM. Balancing covariates via propensity score weighting. JASA. 2018;113:390–400. [Google Scholar]
- 14.Kharbanda RK, Perkins AD, Kennedy J, Banning AP, Baumbach A, Blackman DJ, Dodd M, Evans R, Hildick-Smith D, Jamal Z, et al. Routine cerebral embolic protection in transcatheter aortic valve implantation: rationale and design of the randomised British Heart Foundation PROTECT-TAVI trial. EuroIntervention. 2023;18:1428–1435. doi: 10.4244/eij-d-22-00713 [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.