This study followed the Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) reporting guideline (13). Studies were included if the following criteria were fulfilled: (1) Population comprised patients who underwent TAVI; (2) Reported cardiac-specific biomarker elevation within 72 h; (3) Standardized thresholds cut-points for PPMI based on VARC-2 (peak troponin ≥15 × ULN or CK-MB ≥5 × ULN) or similarly comparable definitions (4); (4) KM curves of all-cause mortality; (5) Fully published status; and (6) Written in English.

PubMed, EMBASE, and Cochrane Center databases were systematically searched for articles that met our inclusion criteria and were published by February 25, 2023. Additionally, we reviewed the references of the included articles and previous reviews to identify relevant texts. We utilized the following search strategy: (“Myocardial injury” OR “PPMI” OR “Troponin” or “Troponin I [TnI]” or “Troponin T [TnT]” or “High-Sensitivity Troponin I [hsTnI]” or “High-Sensitivity Troponin T [hsTnT]” OR “CKMB” OR “CK-MB” OR “Creatine kinase” OR “Creatine phosphokinase” OR “CPK” OR “phosphocreatine kinase”) AND (“Transcatheter aortic valve implantation” OR “TAVI” OR “Transapical aortic valve replacement” OR “TAVR”).

The following steps were taken for study selection: (1) identification of titles of records through database search; (2) removal of duplicates; (3) screening and selection of abstracts; (4) assessment for eligibility through full-text papers; and (5) final inclusion in the study. Two independent reviewers (P.C. and M.M.G.) selected the studies. When there was disagreement, a third reviewer (M.F.S.M.) decided to include or exclude the study. Ethical approval was not applicable to this study as it consisted of a systematic review and meta-analysis.

VARC-2 standardized thresholds were utilized in 13 out of 18 studies (3, 12, 14–24). The other 5 studies utilized the following criteria: CK-MB and/or TnT rise > 5 ULN (25); hsTnT ≥ 166 pg/ml (26); CK-MB > 7 ng/ml (27); TnT increase > 3 ULN (28); and hsTnT rise ≥ 18.3 ULN (29).

The risk of bias was evaluated using the Risk of Bias In Non-randomized Studies of Interventions (ROBINS-I) (30). The studies and their characteristics were classified as having low, moderate, serious or critical risk of bias. Two independent reviewers (P.C. and M.M.G.) assessed risk of bias.

Time-to-event outcomes are not amenable to the standard statistical procedures. For meta-analyses, pooling the treatment effect over several studies must either use estimates of median survival and event rates assessed from survival estimates at given time points, or fall back on direct estimates of the hazard ratio. These approaches are unsatisfactory since they fail to consider the central principles of survival analysis, such as censoring and the proportional hazards assumption (11, 31). As a consequence, the “curve approach” has emerged as the current gold standard for meta-analysis of time-to-event data (32). This approach reconstructed individual patient data (IPD) based on published KM graphs from the included studies. In this meta-analysis, we used the R package “IPDfromKM’’ version 0.1.10 (33).

Raw data coordinates (time and survival probability) for each treatment arm were extracted from published KM survival curves using dedicated software. Subsequently, data coordinates were processed based on the raw data coordinates from the first stage in conjunction with the numbers at risk at given time points when available, and IPD was reconstructed using the IPDfromKM software.

Quality assessment of KM derived IPD data was performed graphically by comparing the derived KM curves with the original curves. The reconstructed IPD was then merged to create the study dataset.

We visually assessed the outcomes of interest in both arms using KM estimates, next, hazard ratios (HRs) with 95% CIs for the difference between both arms were calculated using a Cox frailty model. The inclusion of a γ frailty term was used to account for heterogeneity between studies, with studies modelled as a random effect using random intercepts. The proportional hazards assumption of the Cox model was checked using the Grambsch-Therneau test and diagnostic plots based on Schoenfeld residuals (34).

To deal with proportional hazards assumption violations and assess how the prognostic value of post-TAVI PPMI changed over time, we performed two complementary techniques. First, we fitted a flexible parametric survival model with B-splines. The baseline hazard rate was modeled on a spline with four degrees of freedom. Interactions between the treatment arm and time were added by using a second spline function. We also added a γ frailty term to account for heterogeneity between studies. This technique allowed us to estimate time-varying hazard ratios for our analyses of interest. Finally, we performed landmark analyses to further discriminate short- and long-term PPMI prognostic values.

Subgroup analyses were performed for VARC-2-defined PPMI, and for the VARC-2 cutoff of both troponin and CK-MB-defined PPMI, to further assess the differences in mortality according to the different biomarkers and to investigate the effect of the VARC-2 cutoffs on mortality over the long-term follow-up.

All analyses were performed using R Statistical Software (version 4.2.2, Foundation for Statistical Computing, Vienna, Austria).

The ROBINS-I tool identified ten studies as having a low risk of bias and eight studies as having a moderate risk of bias (Supplementary Figure S2). Quality assessment of KM-derived IPD data by visual comparison of the derived KM curves with the original curves did not show any relevant differences. This analysis, therefore, indicates the results derived from this meta-analysis are grounded on well-conducted observational studies and, therefore, should be regarded as more reliable compared with a scenario in which the aggregated studies were mostly at high-risk for bias.

The pooled analysis for overall survival (OS) comparing patients who had PPMI with those who did not have PPMI after two years revealed that PPMI after TAVI associated with lower cumulative OS (HR = 1.46, 95% CI 1.30–1.65, p < 0.01). Likewise, when only VARC-2 criteria were considered, patients with PPMI also had a lower OS than those without PPMI (HR = 1.23, 95% CI 1.07–1.40, p < 0.01). To investigate the effects of different myocardial injury biomarkers on PPMI, we performed the same analysis with VARC-2 troponin-defined PPMI and VARC-2 CK-MB-defined PPMI. In both cases, OS at two years was lower in the PPMI group (HR = 1.16, 95% CI 1.01–1.33, p = 0.04, and HR = 1.59, 95% CI 1.20–2.09, p < 0.01, respectively), yet the association was much stronger with CK-MB than with troponin (Figure 1).

Using flexible parametric models with B-splines, we estimated time-varying HRs for VARC-2-defined PPMI, as well as for VARC-2-troponin and CK-MB-defined PPMI. This revealed that VARC-2 and troponin-PPMI were associated with lower OS in the initial two months (Figures 2A, 3A), whereas CK-MB-PPMI was associated with lower OS in the first month only (Figure 3C).

Furthermore, landmark analysis was performed using cutoff values determined by time-varying HRs. In the first two months (Figure 2B), VARC-2 PPMI was significantly associated with lower OS (HR = 1.64, 95% CI 1.31–2.07, p < 0.01). However, this was no longer observed after 2 months in the landmark analysis (HR = 0.98, 95% CI 0.83–1.14, p = 0.75). The same trend was observed in the subgroup of troponin-only defined PPMI (Figure 3B), and CK-MB only defined PPMI (Figure 3D). In the first two months, troponin-defined PPMI was significantly associated with lower OS (HR = 1.32 95% CI 1.05–1.67, p = 0.02), but no longer after the 2 month-landmark (HR = 1.00, 95% CI 0.85–1.17, p = 0.98). Finally, in the first month, CK-MB-defined PPMI was strongly associated with lower OS (HR = 7.44, 95% CI 4.76–11.66, p < 0.01), but this association was not statistically significant after 1 month (HR = 0.73, 95% CI 0.50–1.07, p = 0.11).

TAVI is a minimally invasive procedure that does not involve aortic cross clamping and cardioplegia, which are established factors for increased cardiac biomarkers release after valvular surgical procedures (35). Nevertheless, prior studies have demonstrated some degree of elevation of both CK-MB and troponin after the procedure in up to two-thirds of TAVI patients (22). Interestingly, PPMI incidence differs according to the cardiac injury biomarker analyzed and the cutoff point used; although troponin elevation > 15 ULN is of common occurrence during the first 72 h post-TAVI, only 10% of patients experience CK-MB elevation > 5 times the ULN (18). Our pooled analysis corroborates these findings, as the incidence of troponin PPMI was 61% vs. 9% for CK-MB defined PPMI, according to VARC-2 criteria of >5 times the ULN for CK-MB and >15 times the ULN for troponin. This difference in the incidence can be partially explained by the fact that CK-MB elevation requires a greater myocardial injury compared with troponin. For instance, as previously shown CK-MB VARC-2 cutoff threshold of >5 ULN displayed a better correlation with troponin levels of >75 ULN, which is much higher than the established VARC-2 recommendation of >15 times (4, 18). Therefore, the optimal PPMI cutoff point remains a matter of debate and, with the advent of ultra- and of high-sensitivity biomarkers assay kits even lower thresholds of myocardial injury can be measured, potentially overestimating the incidence of PPMI, ultimately jeopardizing its clinical relevance (36). Nonetheless, due to the new VARC-3 definition (≥70 times the ULN of troponin), we hypothesize that PPMI incidence will decrease in future studies while its prognostic significance will rise. This was recently demonstrated in a study by Real et al., in which PPMI incidence using troponin was 14% based on the VARC-3 criteria vs. 59% with VARC-2 (12).

PPMI is likely the result of several factors, such as transient hypotension during ventricular rapid pacing, distal microembolization of calcium particles during balloon dilatation and valve manipulation, mechanical compression of left ventricular outflow, subclinical ventricular trauma due to the wire, coronary artery disease that increases oxygen supply-demand mismatch, and coronary artery occlusion (3, 14, 29, 37). Several procedural predictors of post-TAVI PPMI are also known, such as early experience, first generation valves and transapical (TA) approach (5, 22). TA access is not only associated with PPMI, but is also a known factor for apical myocardial necrosis (38, 39). These findings further corroborate the use of alternatives accesses other than the TA, whenever TF access is not feasible (40). Regarding valve types and PPMI, self-expanding valves (SEV) were previously associated with a two-fold higher incidence of PPMI as compared with balloon-expandable valves (BEV), even after adjusting for several possible confounders (3, 21). This might be explained by various reasons, such as balloon pre-dilatation and after SEV deployment, which can lead small calcium particles to embolize to the coronary arteries, myocardial stunning triggered by more events of rapid pacing performed during the additional balloon dilations in comparison with BEV and perivalvular myocardial compression (41, 42).

Previously published meta-analyses found that PPMI was associated with an increased risk of early and late overall mortality (6–8). Our meta-analysis supports these findings and contributes to the existing literature by aggregating a significantly larger number of patients than previous analyses, indicating that most of the prognostic value of troponin-defined PPMI occurred within the first two months after TAVI, and even earlier for CK-MB-defined PPMI (first month). Furthermore, CK-MB was a better prognostic marker of short and 2-year mortality in comparison with troponin. Two important messages from these results are that first CK-MB using the VARC-2 definition of 5× the ULN is a valuable prognostic tool for mortality. Second, the VARC-2 definition for troponin-defined PPMI of 15× the ULN could overestimate the prevalence of PPMI and hinder its prognostic capacity. VARC-3 definition of 70× the ULN of troponin perhaps is a more suitable value and this is also corroborated by the recent publication of Real et al. which showed no association between VARC-2-troponin defined PPMI with the 1-year mortality (12). Yet, when the analysis was repeated using the VARC-3 cutoff, a statistically significant association was found (12). Importantly, whether the new cutoff is optimal remains uncertain and further studies with larger number of patients, using various assays, and with longer-term follow-up are required to confirm such findings. However, no study to date has specifically indicated potential measures which could improve PPMI patient's prognosis. Still, postprocedural cardiac biomarkers levels evaluation should be used to enhance early months risk assessment, indicating those in need for intensive postprocedural care such as a closer follow-up, possibly within a dedicated TAVI Heart Team, with intensive treatment of risk factors (8, 29).