Long-Term Outcomes of Transcatheter Aortic Valve Replacement in Low-Flow Low-Gradient Aortic Stenosis: A Reconstructed Time-to-Event and Multivariate Meta-Analysis.
This systematic review and meta‐analysis adhered to the Preferred Reporting Items for Systematic Reviews and Meta‐Analysis standards and was registered in the International Prospective Register of Systematic Reviews (CRD42025636782). No ethical approval and considerations were required because only previously published study‐level data were used in this meta‐analysis. The data supporting this study's findings are available from the corresponding author upon reasonable request.PubMed, Embase, Scopus, and the Cochrane Library databases were systematically searched to identify relevant studies from inception until January 10, 2025. The search was conducted using the following key terms and their equivalent terms, without any language restriction: [“High gradient” OR “Low‐flow low‐gradient”] AND [“Aortic stenosis”] AND [“Transcatheter aortic valve replacement”]. Additionally, references of relevant original and review articles were reviewed to identify all potentially relevant studies. Our detailed search strategies for each database are available in the Supplemental Material.The eligibility criteria for inclusion in this analysis were the following: (1) original research articles, including observational studies and randomized controlled trials; (2) studies comparing clinical or echocardiographic outcomes in patients with confirmed severe AS between groups with LFLG—classical LFLG or paradoxical LFLG—or with HG; and (3) studies including patients who underwent TAVR or those reported separate results for TAVR group. We excluded case reports, single‐arm studies, and review articles. Furthermore, each study's population was carefully reviewed for overlapping patient populations. Two studies were excluded due to potential population overlap.12, 13We classified AS hemodynamic subtypes based on current guidelines as follows: HG AS: AVA ≤1.0 cm2 and MG ≥40 mm Hg; classical LFLG AS: AVA ≤1.0 cm2, MG <40 mm Hg, stroke volume index ≤35 mL/m2, and LVEF <50%; and paradoxical LFLG AS: AVA ≤1.0 cm2, MG <40 mm Hg, stroke volume index ≤35 mL/m2, and LVEF ≥50% (Table S1).4, 11After removing duplicates, 4 authors (S.K., S.S.M., P.N., A.M.) independently screened all titles and abstracts based on the prespecified eligibility criteria. Any discrepancies were resolved by consensus with a fifth author (A.S.M.). Subsequently, full texts were reviewed and the following variables were extracted in an electronic data sheet: study design, study year, country, hemodynamic group (HG, classical LFLG, paradoxical LFLG), eligibility criteria, number of patients, baseline characteristics (country, study design, Society of Thoracic Surgeons Predicted Risk of Mortality %, age, sex, body mass index, diabetes, hypertension, hyperlipidemia, prior myocardial infarction, prior percutaneous coronary intervention, prior coronary artery bypass graft, prior stroke, peripheral artery disease, coronary artery disease, atrial fibrillation, chronic kidney disease, chronic obstructive pulmonary disease), pre‐ and postprocedural echocardiographic parameters (valve type, AVA, aortic valve mean gradient [AVMG]), and clinical outcomes (all‐cause mortality, cardiovascular mortality, stroke, heart failure hospitalization, myocardial infarction, permanent pacemaker implantation, acute kidney injury [AKI], major bleeding, and life‐threatening bleeding).
The primary outcome was all‐cause mortality. Secondary outcomes included cardiovascular mortality, heart failure hospitalization, AKI, major bleeding, life‐threatening bleeding, stroke, myocardial infarction, permanent pacemaker implantation, and change from baseline in AVA and post‐TAVR AVMG.
Two independent authors (P.N. and S.S.M.) assessed the quality of included studies using the methodological index for non‐randomized studies tools (MINORS).14 Any discrepancies were resolved by mutual consensus with a third author (A.S.M.).
Individual patient data (IPD) from published Kaplan–Meier (KM) curves for all‐cause mortality were reconstructed using the Liu et al. method.15 Details of the IPD reconstruction process are explained in the Supplemental Methods. We used the ScanIt software to digitize published KM curves by importing the survival (y axis) and time (x axis) values. Then, the extracted values were combined with the number of patients at risk for each hemodynamic group at the time points indicated in the KM curve to reconstruct the IPD. The accuracy of each IPD reconstruction was evaluated using root mean square error, mean absolute error, and maximum absolute error statistics. All reconstructed IPD from each hemodynamic group were merged, and the pooled KM curves were estimated. Cox proportional hazards models stratified by study were fitted to estimate hazard ratios (HRs with 95% CIs). Robust SEs clustered at the study level were used to provide valid statistical inference by accounting for within‐study correlation and differences in baseline hazard functions across studies. This stratified modeling approach ensures that treatment effect estimates are derived from within‐study comparisons while holding study membership fixed. The proportional hazards assumption was evaluated using Schoenfeld residual plots and the Grambsch–Therneau test as a diagnostic assessment.16
For secondary outcomes, where KM curves were unavailable, data were synthesized by calculating effect sizes using pooled adjusted and unadjusted HR or risk ratios (RR) for binary outcomes and standardized mean differences (SMD) for continuous outcomes. Adjusted HRs or RRs were preferentially used when available; otherwise, unadjusted estimates were included. All relative effect measures were analyzed on the logarithmic scale, and corresponding SEs were extracted directly or derived from reported CIs.
Because most studies included 3 comparison arms (classical LFLG, paradoxical LFLG, and HG) multiple treatment comparisons from the same study are statistically correlated. These studies were therefore analyzed using a fixed‐effect multivariate meta‐analysis model, which jointly estimated the 3 pairwise treatment comparisons while accounting for within‐study correlations. Variance–covariance matrices were constructed from the reported standard errors of the log‐transformed treatment effects, assuming a constant within‐study correlation coefficient of 0.5 between comparisons. The multivariate model estimated pooled log‐transformed HRs/RRs for the 3 pairwise contrasts (classical LFLG versus HG AS, paradoxical LFLG versus HG, and classical LFLG versus paradoxical LFLG AS), which were subsequently exponentiated to obtain HRs with 95% CIs. To incorporate 2‐arm studies, additional fixed‐effect pairwise meta‐analyses were performed using the inverse‐variance method, combining relevant comparisons from both 3‐arm and 2‐arm studies. Heterogeneity across studies was assessed using Higgins and Thompson's I2 statistic. Publication bias was examined visually using comparison‐adjusted funnel plots and statistically using Egger's test. All analyses were performed using the mvmeta, meta, metafor, and ggplot2 packages in R version 4.2.1 (R Foundation, Vienna, Austria). Statistical significance was determined at a 2‐sided P value threshold of <0.05.
RESULTS
Our initial systematic search identified 4681 studies. Following duplicate removal, 3465 studies were screened based on title and abstract. The full text of 38 studies was evaluated for eligibility, and ultimately, 19 studies were included in the final analysis (Figure 1).17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35 These observational studies included 20 493 patients with severe AS (12 237 with HG AS, 3274 with classical LFLG AS, and 4982 with paradoxical LFLG AS) undergoing TAVR. Studies were conducted in Germany (7 studies), the United States (3 studies), Switzerland (2 studies), Canada (2 studies), Italy (1 study), France (1 study), Portugal (1 study), Poland (1 study), and Japan (1 study). Details about baseline characteristics, procedural characteristics and outcomes, and the type and generation of devices in each study are reported in Table 1 and Tables S2 through S5. Furthermore, the results of the risk of bias assessment using methodological index for nonrandomized studies criteria are summarized in Table S6.
C‐LFLG indicates classical low‐flow low‐gradient; NR, not reported; P‐LFLG, paradoxical low‐flow low‐gradient; and STS‐PROM, Society of Thoracic Surgeons Predicted Risk Of Mortality.
A total of 12 studies, comprising 19 502 patients (11 791 with HG AS, 2995 with classical LFLG AS, and 4716 with paradoxical LFLG AS), were included in the all‐cause mortality reconstructed IPD meta‐analysis. Classical LFLG AS was associated with a significantly higher risk of 5‐year all‐cause mortality than the HG AS (HR, 1.92 [95% CI, 1.62–2.27], P<0.001). Similarly, paradoxical LFLG AS was associated with a significantly higher risk of 5‐year all‐cause mortality than HG AS (HR, 1.20 [95% CI, 1.07–1.34], P=0.002). The pooled reconstructed KM curve is presented in Figure 2.
HG indicates high gradient; and LFLG, low‐flow low‐gradient.
Accuracy tests for the reconstructed IPD indicated that the extracted data points were captured with high precision across all studies, as presented in Table S7. Additionally, no statistically significant evidence of deviation from the proportional hazards assumption was detected based on Schoenfeld residual plots and the Grambsch–Therneau test (Figures S1 through S3). The multivariate meta‐analysis of all‐cause mortality after pooling 3‐arm and 2‐arm studies (17 studies; 19 721 patients), showed similar findings with a higher risk of all‐cause mortality in classical LFLG versus HG AS (HR, 1.68 [95% CI, 1.56–1.82], P<0.001), paradoxical LFLG versus HG AS (HR, 1.23 [95% CI, 1.14–1.34], P<0.001), and classical LFLG versus paradoxical LFLG (HR, 1.32 [95% CI, 1.22–1.42], P<0.001; Figure 3).
HG indicates high gradient; HR, hazard ratio; and LFLG, low‐flow low‐gradient. Results with a 2‐sided P*‐value <0.05 were considered statistically significant and bolded.
Our pooled analysis of 3‐arm and 2‐arm studies revealed that both classical and paradoxical LFLG AS groups were associated with a significantly higher risk of cardiovascular mortality compared with HG AS (HR, 1.94 [95% CI, 1.74–2.16], P<0.001 and HR, 1.40 [95% CI, 1.25–1.57], P<0.001, respectively). The risk of cardiovascular mortality was higher in classical LFLG AS versus paradoxical LFLG AS (HR, 1.46 [95% CI, 1.29–1.65], P<0.001). We observed a significantly higher risk of heart failure rehospitalization in classical LFLG versus. HG AS (HR, 4.12 [95% CI, 2.16–7.83], P<0.001), and also paradoxical LFLG versus. HG AS (HR, 1.80 [95% CI, 1.12–2.85], P = 0.012). No significant difference was found in the risk of heart failure hospitalization between groups with classical and paradoxical LFLG AS (HR, 1.11 [95% CI, 0.78–1.58], P 0.55; Figure 3).
In addition, classical LFLG AS was associated with a higher risk of AKI compared with HG AS (HR, 1.21 [95% CI, 1.04–1.41], P = 0.013). The risk of AKI was comparable in paradoxical LFLG versus. HG AS (HR, 1.12 [95% CI, 0.97–1.30], P = 0.12), and classical versus. paradoxical LFLG AS (HR, 0.92 [95% CI, 0.76–1.10], P = 0.35; Figure 4). There were no significant differences in the risk of major bleeding, life‐threatening bleeding, stroke, myocardial infarction, or permanent pacemaker implantation across AS hemodynamic groups except for higher risk of stroke in paradoxical LFLG versus. HG AS (RR, 1.22 [95% CI, 1.03–1.45], P = 0.020; Figure 4). Moreover, the pooled outcomes of the 3‐arm studies are reported in te Table S8.
HG indicates high gradient; LFLG, low‐flow low‐gradient; and RR, risk ratio. Results with a 2‐sided P* value <0.05 were considered statistically significant and bolded.
The analysis of post‐TAVR echocardiographic indices showed that both groups with classical and paradoxical LFLG AS had significantly lower differences in the change from baseline in AVA compared with HG AS (SMD, −0.10 [95% CI, −0.15 to −0.04], P<0.001 and SMD, −0.16 [95% CI, −0.20 to −0.11], P<0.001, respectively). Change from baseline in AVA was higher in classical LFLG AS versus. paradoxical LFLG AS (SMD, 0.13 [95% CI, 0.07–0.19], P<0.001; Figure 5). Additionally, post‐TAVR AVMG was lower in both classical LFLG AS (SMD, −0.40 [95% CI, −0.45 to −0.35], P<0.001) and paradoxical LFLG AS (SMD, −0.36 [95% CI, −0.53 to −0.19], P<0.001; Figure 5) compared with HG AS.
AVA indicates arotic valve area; AVMG, aortic valve mean gradient; HG, high gradient; LFLG, low‐flow low‐gradient; SMD, standardized mean difference; and TAVR, transcatheter aortic valve replacement. Results with a 2‐sided P* value <0.05 were considered statistically significant and bolded.
Visual assessment of the funnel plots and Egger's test did not indicate evidence of publication bias for clinical and echocardiographic outcomes, except for the change in AVA following TAVR (Egger's test P: 0.006). The funnel plots corresponding to each outcome are presented in Figures S4 through S14.
DISCUSSION
In this meta‐analysis of 19 observational studies comprising 20 493 patients with severe AS undergoing TAVR, we report the following important findings across different flow‐gradient hemodynamic subtypes: (1) time‐to‐event reconstructed IPD analysis demonstrated that classical and paradoxical LFLG AS was associated with a higher risk of 5‐year all‐cause mortality following TAVR compared with the HG AS; (2) both groups with classical and paradoxical LFLG AS exhibited an increased risk of cardiovascular mortality versus. the group with HG AS; (3) patients with classical and paradoxical LFLG AS had a higher risk of heart failure hospitalization compared with those with HG AS; and (4) post‐TAVR echocardiographic findings showed a lower change from baseline in AVA and post‐TAVR AVMG in classical and paradoxical LFLG AS compared with HG AS.
Our findings highlight the long‐term prognostic implications of flow‐gradient subtypes in severe AS and emphasize that TAVR outcomes are not uniform across hemodynamic subtypes. The increased all‐cause and cardiovascular mortality observed in classical and paradoxical LFLG AS underscores the importance of understanding the distinct pathophysiological mechanisms underlying these subgroups. Although previous studies have suggested that paradoxical LFLG AS may represent a less severe phenotype, our results indicate that this group still faces significant long‐term mortality risks, despite lower than those seen in classical LFLG AS.36
In this study, we found that both classical and paradoxical LFLG AS subtypes were associated with a significantly higher risk of long‐term all‐cause and cardiovascular mortality following TAVR compared with HG AS. These findings contrast with the most recent meta‐analysis on flow gradient subtypes in severe AS, which evaluated the risk of 1‐year major adverse cardiovascular events following TAVR.37 The meta‐analysis by Ullah et al., identified a significantly higher risk of 1‐year major adverse cardiovascular events in classical LFLG AS compared with HG AS but found no significant difference between paradoxical LFLG AS and HG AS.37 However, our study extends these findings in several ways. First, whereas prior meta‐analyses focused primarily on 1‐year outcomes, we analyzed 5‐year all‐cause mortality, providing a more comprehensive assessment of long‐term risks. Second, we conducted a time‐to‐event reconstructed IPD analysis, allowing for a more granular assessment of survival.15 Third, we applied stricter eligibility criteria, ensuring that all included studies incorporated stroke volume index to define flow‐gradient subtypes. By excluding normal‐flow LG AS cases, our analysis achieved greater precision in defining the prognostic impact of these subtypes. Fourth, unlike prior studies that combined patients undergoing surgical and transcatheter aortic valve replacement, our study was restricted to TAVR‐only populations, eliminating the heterogeneity introduced by differences in procedural risks and patient selection.38 Lastly, we excluded studies that combined classic and paradoxical LFLG AS into a single LFLG category, a methodological limitation in previous analyses that may have resulted in misclassification of distinct hemodynamic subtypes.8, 39
The increased risk of heart failure hospitalization and AKI in classical LFLG AS further reflects the greater burden of myocardial dysfunction in this group. Classical LFLG AS is typically associated with chronic afterload mismatch, impaired contractility, and extensive myocardial fibrosis, all of which contribute to poor post‐TAVR recovery despite valve replacement.40, 41 The predominantly older and frailer population within this subgroup may also contribute to higher rates of post‐TAVR complications. In contrast, paradoxical LFLG AS, despite its preserved LVEF, is characterized by concentric LV remodeling, increased stiffness, and restrictive filling dynamics.42 These structural and functional abnormalities likely explain the continued risk of adverse cardiovascular outcomes post TAVR.
The echocardiographic findings from our study demonstrated significantly lower change from baseline in AVA post TAVR among patients with classical LFLG and paradoxical LFLG AS compared with HG AS, suggesting that valve expansion and anatomical relief of stenosis were better in patients with the HG hemodynamic subtype. In addition, both classical and paradoxical LFLG AS exhibited significantly lower post‐TAVR AVMG compared with HG AS. This likely reflects the persistence of low transvalvular flow following TAVR, a shared physiological feature of LFLG subtypes.
Results from the risk‐of‐bias assessment using the methodological index for nonrandomized studies criteria indicated moderate methodological quality across 19 included studies (median score: 15 [interquartile range, 13–17]); 63.2% were comparative studies. In terms of reporting quality, all included studies stated aims clearly, and 68.4% had appropriate follow‐up, yet only 47.4% reported adequate prospective data collection. We must emphasize that the results should be interpreted with extra caution because limited prospective data collection increases susceptibility to information bias and unmeasured confounding factors. Moreover, accuracy metrics of the reconstructed IPD indicate high‐fidelity capture of KM time‐to‐event data across studies, supporting the reliability of pooled survival estimates while not mitigating study‐level bias.
The findings of this study underscore the need for future research to refine risk stratification and management strategies to improve long‐term outcomes in LFLG AS following TAVR. Although TAVR provides a survival advantage over conservative management, the risk of mortality remains elevated in both classical and paradoxical LFLG AS, warranting investigations into adjunctive therapies targeting myocardial fibrosis, diastolic dysfunction, and LV remodeling.36 Invasive cardiac catheterization is pivotal in evaluating LFLG AS, as it allows definitive differentiation between true‐severe and pseudo‐severe AS when imaging data are inconclusive.43 Patients with LFLG AS may benefit from a structured post‐TAVR follow‐up program that includes close heart failure monitoring with optimization of guideline‐directed medical therapy according to LVEF, together with routine renal function monitoring to identify and mitigate the risk of AKI.44 The absence of randomized controlled trials comparing flow‐gradient subtypes in TAVR limits definitive conclusions, highlighting the need for future trials to better understand underlying mechanisms and optimize treatment approaches. Further research is also needed to better characterize long‐term hemodynamic and survival outcomes in LFLG AS, which could guide the development of more consistent, evidence‐based risk assessment frameworks.
This meta‐analysis has several limitations that need to be considered. First, the inclusion of only observational studies may introduce potential selection bias and residual confounding. Second, despite rigorous eligibility criteria, inconsistencies in flow‐gradient classification across studies remain a concern due to heterogeneity in diagnostic criteria and variability in echocardiographic assessment protocols. Third, the use of IPD reconstruction for all‐cause mortality time‐to‐event analysis, while validated, may introduce approximation errors related to the extraction and digitization of survival data from KM curves. In addition, reconstructed IPD lacks patient‐level covariates and therefore does not allow adjustment for individual confounders. To mitigate this limitation, Cox proportional hazards models were stratified by study with robust standard errors, enabling within‐study comparisons while accounting for between‐study heterogeneity in baseline risk. Furthermore, all‐cause mortality multivariate meta‐analysis using adjusted HRs showed no statistically significant differences compared with the reconstructed IPD analysis, supporting the robustness of the findings despite the absence of individual‐level adjustment for potential confounders. Fourth, the absence of long‐term echocardiographic follow‐up limits the ability to evaluate post‐TAVR myocardial remodeling, structural valve degeneration, and late hemodynamic changes, all of which are critical for assessing long‐term procedural success. Lastly, the lack of data on frailty, diastolic dysfunction, and myocardial fibrosis restricts our ability to determine their prognostic significance, particularly in paradoxical LFLG AS. Future studies should incorporate standardized echocardiographic protocols, frailty assessments, and myocardial tissue characterization to better elucidate the mechanisms underlying poor outcomes across these hemodynamic subgroups.
CONCLUSIONS
In this meta‐analysis of 19 studies (20 493 patients) comparing outcomes following TAVR across hemodynamic subtypes of severe AS, both classical and paradoxical LFLG AS were associated with a higher risk of long‐term all‐cause and cardiovascular mortality compared with HG AS. Additionally, classical and paradoxical LFLG AS were associated with a higher risk of heart failure hospitalization compared with HG AS. Future studies are needed to compare long‐term clinical and echocardiographic outcomes across flow‐gradient subtypes in a randomized setting to confirm these findings and ensure a more precise evaluation of treatment effects.
Sources of Funding
None.
Disclosures
Dr Kolte has received research funding from the National Institutes of Health/National Heart, Lung, and Blood Institute and Medtronic, Inc. outside of the submitted work. All other authors have no conflicts of interest to declare.
Supporting information
PRISMA 2020 Checklist
Tables S1–S8
Figures S1–S14
Search Strategy
Supplemental Methods
AS
Dr. Andrea Scotti
Cardiology · Interventional Cardiology & StructuralMontefiore Medical Centerandreas