This meta-analysis was carried out following the guidelines set by the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) and Cochrane Handbook for Systematic Reviews (20, 21). A thorough literature search was executed across various databases, including PubMed, Scopus and, Web of Science from their inception up to September 15, 2024. Several keywords were searched, including terms such as “Radiation Therapy,” “Radiotherapy,” “Radiation-exposed,” “Chest Radiation,” “Thoracic Radiation,” “Transcatheter Aortic Valve Replacement,” “TAVI,” and “TAVR.” No limitations were imposed regarding language or geographical location. The detailed search strategy is presented in Supplementary Table S1. The protocol of the study is registered in PROSPERO (CRD42024593497).

The study selection process involved two independent reviewers (PS and DA) who screened the titles and abstracts of all identified studies. Full-text articles were retrieved for further evaluation if they appeared to meet the inclusion criteria or if there was uncertainty. Discrepancies between reviewers were resolved through discussion or consultation with a third reviewer (EA-S).

Studies were included if they compared clinical outcomes of patients undergoing TAVR with and without prior chest radiation therapy. Eligible studies were required to be peer-reviewed, published in scientific journals, and provide quantitative data on primary and secondary outcomes, such as mortality, stroke, myocardial infarction, bleeding, and heart failure exacerbation. Additionally, studies had to report data with adequate follow-up periods, including in-hospital, 30-day, 1-year, and, where available, follow-up periods extending beyond 1 year. Commentaries, case reports, case series, editorials, and books were excluded. Furthermore, studies that did not report sufficient data on relevant clinical outcomes were excluded.

The quality of included studies was assessed using the Joanna Briggs Institute (JBI) checklist, which is a comprehensive tool for evaluating the methodological quality of studies. The checklist includes 8 items for cross-sectional studies, 10 items for case-control studies, and 11 items for cohort studies. Each item can be answered with “yes,” “no,” “unclear,” or “not applicable” (22–24). Two independent reviewers (PS and DA) assessed each study using the JBI checklist, with disagreements resolved through consensus with a third reviewer (EA-S).

Two reviewers (PS and DA) independently extracted data from the included studies. Extracted data included author names, year of publication, study design, sample size, details of radiation therapy exposure, TAVR procedural details, and clinical outcomes. For studies that did not report all necessary data, corresponding authors were contacted for additional information. Outcomes of interest included in-hospital, 30-day, and 1-year mortality, cardiovascular events (e.g., stroke, myocardial infarction), bleeding complications, heart failure exacerbation, and need for pacemaker implantation post-TAVR. Disagreements resolved through consensus with a third reviewer (EA-S).

The statistical analysis was conducted using STATA 18. For binary outcomes, the odds ratio (OR) was reported as the effect measure. For continuous outcomes, the standardized mean difference (SMD) was chosen as the summary statistic. A leave-one-out sensitivity analysis was performed to assess the impact of each individual study on the overall effect. A random-effects model with restricted maximum likelihood estimation was selected for the analysis to account for potential variability across studies. In cases where the number of included studies was fewer than 10, the Knapp-Hartung correction was applied to adjust for the small sample size. Prediction interval analysis was also conducted to estimate the range of effects in future studies. Heterogeneity across studies was evaluated using the I² statistic and the Cochrane Q test. Heterogeneity was considered significant if I² exceeded 50% and the P-value was below 0.1. Publication bias was assessed using contour-enhanced funnel plots, Egger's test, Begg's test, and the trim-and-fill method (25, 26). The certainty of evidence for each outcome was assessed using the GRADE (Grading of Recommendations Assessment, Development and Evaluation) approach. Evaluations were conducted within the GRADE Pro software, which applies standardized criteria for downgrading and upgrading evidence quality. Downgrading was based on five domains: risk of bias, inconsistency, indirectness, imprecision, and publication bias. Upgrading was considered based on three domains: large magnitude of effect, plausible confounding, and evidence of a dose-response gradient. Based on these domains, the certainty of evidence was rated as high, moderate, low, or very low.

A comprehensive search of databases including PubMed, Web of Science, and Scopus was conducted, yielding a total of 1,151 records (PubMed = 282, Web of Science = 297, Scopus = 572). After removing 467 duplicate records, 684 unique studies were screened for eligibility. Following the screening process, 650 records were excluded based on their relevance to the study's criteria. The full texts of 34 studies were sought for further retrieval and assessment. Upon reviewing the 34 full-text reports, 25 studies were excluded for the following reasons: 12 studies compared TAVR with Surgical Aortic Valve Replacement (SAVR), seven studies focused on patients undergoing TAVR who were actively receiving cancer treatment, four studies were systematic reviews without meta-analyses, and two studies did not provide sufficient data for inclusion in the meta-analysis. After applying these criteria, nine studies were eligible and included in the final review (Figure 2).

The included studies in this meta-analysis evaluated the clinical outcomes of patients undergoing TAVR with prior chest radiation therapy (C-XRT) compared to those without radiation exposure. Most of the studies were conducted in the United States (27–33), with two originating from France (34, 35).

The total sample sizes across the studies ranged from 52 to 296,670 patients. The number of patients with a history of chest radiation therapy ranged from 16 to 2,780 across the included studies. All the included studies were cohort studies and median follow-up periods across the studies ranged from 1 to 60 months. All the included studies had high quality (Supplementary Table S2; Table 1).

The meta-analysis comparing 30-day mortality between patients receiving C-XRT and those not undergoing XRT showed no significant difference (OR: 1.59, 95% CI: 0.71–3.55, P = 0.20) (Figure 6A). The 95% prediction interval ranged from 0.09 to 11.11, indicating considerable variability in potential future outcomes (Figure 6A). A leave-one-out sensitivity analysis demonstrated that omitting any single study did not substantially alter the overall result (Figure 6B). Heterogeneity across studies was minimal (I² = 1.14%, H² = 1.01, τ² = 0.01), suggesting consistent results across the studies (Figure 6C). The contour-enhanced funnel plot showed an asymmetrical pattern, suggesting potential bias (Figure 6D). Egger's regression test was significant (P < 0.01), and Begg's regression test showed marginal significance (P = 0.07). The trim-and-fill analysis, which added two imputed studies on the right side, resulted in an adjusted OR of 1.68 (95% CI: 0.50–6.56) (Figure 6E).

The analysis of 1-year mortality revealed no significant difference between patients receiving C-XRT and those not undergoing XRT (OR: 1.15, 95% CI: 0.52–2.54, P = 0.72) (Figure 7). However, since only two studies were available for this outcome, conducting a sensitivity analysis was not feasible. Furthermore, the limited number of studies also made it impossible to properly assess publication bias.

The meta-analysis comparing in-hospital cardiovascular mortality between patients receiving C-XRT and those not undergoing XRT showed no significant difference (OR: 1.09, 95% CI: 0.25–4.83, P = 0.86) (Figure 8A). The 95% prediction interval was broad, ranging from 0.03 to 33.39, indicating considerable variability in potential future outcomes (Figure 8A). A leave-one-out sensitivity analysis demonstrated that excluding any individual study did not significantly alter the overall results (Figure 8B). Heterogeneity was not significant (I² = 31.26%, H² = 1.45, τ² = 0.35) (Figure 8C). The contour-enhanced funnel plot appeared relatively asymmetrical, suggesting potential publication bias (Figure 8D). However, publication bias was significant based on Egger's test (P = 0.07) but not significant according to Begg's test (P = 0.70). The trim-and-fill analysis imputed two studies on the right side, adjusting the odds ratio to 1.84 (95% CI: 0.68–4.96) (Figure 8E).

The meta-analysis comparing 30-day cardiovascular mortality between patients receiving C-XRT and those not undergoing XRT revealed no significant difference (OR: 1.01, 95% CI: 0.28–3.61, P = 0.99) (Figure 9). However, with only two studies available, it was not possible to conduct a sensitivity analysis. Moreover, the assessment of publication bias and prediction was also not feasible due to the limited number of studies.

The meta-analysis assessing in-hospital infarction between patients receiving C-XRT and those not undergoing XRT found no significant difference (OR: 0.54, 95% CI: 0.03–9.20, P = 0.67) (Figure 10A). A leave-one-out sensitivity analysis showed that removing any single study did not notably change the overall findings (Figure 10B). There was significant heterogeneity (I² = 72.69%, H² = 3.66, τ² = 4.54), indicating substantial variability between the included studies (Figure 10C). The contour-enhanced funnel plot was fairly symmetrical, pointing to a lack of potential publication bias (Figure 10D). Publication bias was further ruled out by Egger's test (P = 0.99) and Begg's test (P = 1.00). Additionally, the trim-and-fill analysis did not add any imputed studies (Figure 10E).

The meta-analysis evaluating in-hospital major bleeding incidents among patients receiving C-XRT compared to those not undergoing XRT revealed no significant difference (OR: 0.80, 95% CI: 0.40–1.59, P = 0.38) (Figure 11A). The prediction interval ranged from 0.30 to 2.11 (Figure 11A). A leave-one-out sensitivity analysis indicated that the exclusion of any single study did not significantly alter the overall results (Figure 11B). There was minimal heterogeneity observed (I² = 17.53%, H² = 1.21, τ² = 0.03) (Figure 11C). The contour-enhanced funnel plot displayed asymmetry, suggesting potential publication bias (Figure 11D). This was further supported by Egger's test, which indicated a significant result (P = 0.04), while Begg's test did not show significance (P = 0.30). Additionally, the trim-and-fill analysis incorporated two imputed studies, resulting in an odds ratio of OR = 0.90 (95% CI: 0.74–1.09) (Figure 11E).

The meta-analysis investigating 30-day major bleeding events in patients receiving C-XRT compared to those not undergoing XRT found no significant difference (OR: 0.99, 95% CI: 0.53–1.86, P = 0.98) (Figure 12A). The prediction interval spanned from 0.28 to 3.35 (Figure 12A). A leave-one-out sensitivity analysis demonstrated that the removal of any single study did not significantly affect the overall results (Figure 12B). There was no significant heterogeneity detected (I² = 43.85%, H² = 1.78, τ² = 0.14) (Figure 12C). The contour-enhanced funnel plot exhibited a relatively symmetrical pattern, indicating no apparent publication bias (Figure 12D). This finding was corroborated by Egger's test, which yielded a non-significant result (P = 0.35), and Begg's test also showed no significance (P = 0.70). Furthermore, the trim-and-fill analysis added two imputed studies, resulting in an odds ratio of OR = 1.13 (95% CI: 0.71–1.79) (Figure 12E).

The meta-analysis evaluating in-hospital stroke rates between patients treated with C-XRT and those not undergoing XRT revealed no statistically significant difference (OR: 1.21, 95% CI: 0.11–13.20, P = 0.81) (Supplementary Figure S1A). The 95% prediction interval extended from 0.02 to 811.54, indicating considerable variability in potential outcomes (Supplementary Figure S1A). Sensitivity analysis, performed using the leave-one-out method, confirmed that the exclusion of any individual study did not substantially alter the overall results (Supplementary Figure S1B). However, there was considerable heterogeneity across the included studies (I² = 85.37%, H² = 6.84, τ² = 1.75) (Supplementary Figure S1C). The contour-enhanced funnel plot displayed a generally symmetrical distribution (Supplementary Figure S1D). Both Egger's test (P = 0.87) and Begg's test (P = 1.00) demonstrated no significant signs of bias. Furthermore, the trim-and-fill analysis did not impute any missing studies (Supplementary Figure S1E).

The meta-analysis comparing 30-day stroke incidence between patients receiving C-XRT and those not undergoing XRT showed no significant difference (OR: 0.71, 95% CI: 0.34–1.51, P = 0.25) (Supplementary Figure S2A). The 95% prediction interval ranged from 0.15 to 3.17 (Supplementary Figure S2A). A leave-one-out sensitivity analysis demonstrated that removing any single study did not substantially alter the overall results (Supplementary Figure S2B). There was no significant heterogeneity (I² = 0.00%, H² = 1.00, τ² = 0.00) (Supplementary Figure S2C). The contour-enhanced funnel plot appeared symmetrical, implying no indication of publication bias (Supplementary Figure S2D). Publication bias was further ruled out with Egger's test (P = 0.81) and Begg's test (P = 1.00). The trim-and-fill analysis, which imputed one study on the right side, adjusted the OR to 0.71 (95% CI: 0.37–1.42) (Supplementary Figure S2E).

The meta-analysis that examined acute kidney injury (AKI) following the procedure in patients who received C-XRT compared to those who did not reveal no significant differences (OR: 0.42, 95% CI: 0.08–2.15, P = 0.22) (Supplementary Figure S3A). The 95% prediction interval ranged from 0.01 to 17.00. A sensitivity analysis revealed that after the removal of Kherallah et al. (32), patients without a history of C-XRT experienced significantly lower rates of AKI following TAVR (OR: 0.30, 95% CI: 0.11–0.81, P = 0.01) (Supplementary Figure S3B). The level of heterogeneity was high (I² = 69.29%, H² = 3.26, τ² = 1.03 (Supplementary Figure S3C). The contour-enhanced funnel plot was symmetrical, suggesting minimal publication bias (Supplementary Figure S3D). Additionally, both Egger's test (P = 0.22) and Begg's test (P = 0.46) supported the absence of significant publication bias. The trim and fill analysis did not insert any study (Supplementary Figure S3E).

The meta-analysis assessing heart failure exacerbation between patients receiving C-XRT and those not undergoing XRT found no significant overall difference (OR: 1.49, 95% CI: 0.96–2.33, P = 0.07) (Supplementary Figure S4A). The 95% prediction interval ranged from 0.64 to 3.46 leave-one-out sensitivity analysis demonstrated that removing any single study did not substantially alter the overall results (Supplementary Figure S4B). Heterogeneity was low (I² = 27.42%, H² = 1.38, τ² = 0.07), (Supplementary Figure S4C). The contour-enhanced funnel plot showed a symmetrical pattern, indicating no significant evidence of publication bias (Supplementary Figure S4D). Both Egger's test (P = 0.48) and Begg's test (P = 0.54) also showed no significant publication bias. Furthermore, the trim-and-fill analysis did not add any imputed studies (Supplementary Figure S4E).

The meta-analysis evaluating PPM implantation after TAVR between patients receiving C-XRT and those not undergoing XRT found no significant overall difference (OR: 1.46, 95% CI: 0.70–3.05, P = 0.26). The 95% prediction interval ranged from 0.38 to 5.49 (Supplementary Figure S5A). leave-one-out sensitivity analysis demonstrated that removing any single study did not substantially alter the overall results (OR: 1.83, 95% CI: 1.27–2.61, P < 0.01) (Supplementary Figure S5B). Heterogeneity was not significant (I² = 44.84%, H² = 1.81, τ² = 0.20) (Supplementary Figure S5C). The contour-enhanced funnel plot displayed an asymmetrical pattern, suggesting potential publication bias (Supplementary Figure S5D). Both Egger's test (P < 0.01) and Begg's test (P = 0.07) further confirmed the presence of significant publication bias. Moreover, the trim-and-fill analysis, which added two imputed studies, adjusted the OR to 1.98 (95% CI: 0.92–3.87) (Supplementary Figure S5E).

In the meta-analysis, the GRADE assessment revealed that the evidence quality for several outcomes was predominantly low or very low. Specifically, in-hospital stay, in-hospital mortality, 30-day mortality, in-hospital cardiovascular mortality, in-hospital myocardial infarction, in-hospital major bleeding, PPM after TAVR, and in-hospital stroke were all assigned a very low grade. Conversely, 30-day major bleeding, 30-day stroke, AKI after TAVR and, HF exacerbation after TAVR, were graded as having low evidence quality (Table 2).

Our meta-analysis includes data from a wide range of follow-up durations, from as short as 1 month to as long as 60 months to identify the absence of significant differences in short-term outcomes while highlighting the potential for increased long-term risks. The inclusion of long-term follow-up periods adds depth to our understanding of the effects of radiation therapy on TAVR outcomes, providing new insights that were previously underreported. Furthermore, it integrates data from a broader pool of patients, including several large cohorts such as Gajjar et al. and Agrawal et al. (29, 30) with a larger number of patients undergoing TAVR in both the radiation-exposed and non-exposed groups, we provide a more statistically powerful comparison, minimizing bias that might arise from smaller studies or limited patient populations. This enhances the generalizability of our findings to the broader patient population undergoing TAVR. The use of sensitivity analyses, prediction intervals, and publication bias assessments (e.g., Egger's and Begg's tests) throughout the meta-analysis ensures the robustness of our results. In addition, we conducted leave-one-out sensitivity analyses to verify the stability of our findings, and the consistent results across studies reinforce the reliability of our conclusions. These rigorous methods provide confidence that our findings are not skewed by small-study effects or publication bias.

Despite the mentioned strength and novelty of our study, there are some limitations: first, there was significant heterogeneity across the included studies, driven by a variety of factors. One major source of heterogeneity stems from the observational study design and various follow-up durations across the included. The second possible cause of heterogeneity originates from different radiation therapy dosages received by patients (15, 41) and studies included in our analysis did not always provide detailed information on the specific dosages, making it difficult to account for this variation. Third, there was variability in the timeframe between radiation therapy and TAVR across the included studies. Some patients had undergone radiation decades before TAVR, while others received treatment more recently. This variability may influence the extent of radiation-induced cardiovascular damage and could have impacted the outcomes. Fourth, the indications for radiation therapy varied widely, with some patients receiving treatment for breast cancer, while others were treated for lymphoma or lung cancer. These differences in underlying cancer types alongside possible adjuvant therapies besides radiation, such as chemotherapy can also have cardiovascular effects. These concomitant treatments were not uniformly reported or controlled for included studies, making it challenging to isolate the impact of radiation therapy on TAVR outcomes. Finally, there was a lack of data regarding the long-term durability of surgical aortic valves in patients with prior chest radiation. Future studies comparing long-term structural valve deterioration between TAVR and SAVR in this specific population are warranted. Additionally, although we used several statistical tools to decrease bias, such as sensitivity analyses, prediction intervals, and publication bias assessments (e.g., Egger's and Begg's tests), we did not employ bootstrapping of the data. Bootstrapping could have improved the confidence intervals (CIs) and provided a more robust estimate, especially given the high variability in the data. We acknowledge that incorporating bootstrapping in future studies could further enhance the precision of the results and minimize bias due to variability.