In recent years, there has been an increased need to understand new quantitative analysis methodologies applied in key works within the cardiology and cardiac surgery literature. Discussions regarding prospective or retrospective study designs, bias control, confounders, and the statistical tests applied have given way to the primacy of randomized controlled trials as the highest quality evidence, enhanced by methodologies addressing, among others, non-inferiority and Bayes’ theorem. 

Although the results of the present analysis are relevant, they do not require extensive discussion. Instead, it is the methodology that necessitates careful consideration to understand the basis for the authors’ conclusions. In brief, this meta-analysis aggregates 5-year mortality outcomes from key comparative studies of TAVR and surgical aortic valve replacement (SAVR), all of which employed non-inferiority methodologies. This meta-analysis encompassed a full risk spectrum, including the PARTNER 1A, PARTNER 2A, PARTNER 3, CoreValve US, SURTAVI, NOTION, Evolut Low Risk, and UK-TAVI trials. Consequently, 8698 patients were included, with 4443 undergoing TAVR and 4255 undergoing SAVR. Following a Bayesian meta-analysis, the authors concluded that TAVR cannot be considered non-inferior to SAVR based on 5-year mortality outcomes; in other words, SAVR was significantly superior in this mid-term follow-up. 

COMMENTARY: 

In traditional statistical inference, we determine that the occurrence of a phenomenon differs significantly between groups when, with a margin of error below a predetermined criterion (alpha error, typically 5%), it is deemed not due to chance. The absence of differences, or the probability of a phenomenon occurring above this 5% threshold, would classify it as equivalent. Non-inferiority studies focus on this concept of equivalence, but from a different perspective. Simply stated, they aim to determine whether the difference in the occurrence of a phenomenon between two groups falls within a pre-specified interval, outside of which it would be considered clinically relevant. The objective, therefore, is not to establish the degree to which the phenomenon occurs nor the differences between groups but to assess its clinical impact according to a defined non-inferiority margin. The boundaries of this margin are determined by differences in occurrence rates reported between groups. Exceeding this margin leads to a rejection of non-inferiority or, conversely, to a finding of superiority for one group over the other. With this type of analysis, it is reasonable to assume that they are less stringent than “traditional” inference methods, as achieving non-inferiority, the primary goal, is easier than identifying statistically significant differences. Additionally, they are less dependent on statistical power, which is inherently tied to sample size, and whose insufficiency is one of the main reasons for failing to detect statistically significant differences in “traditional” statistics. 

If non-inferiority analysis appears convoluted, Bayesian statistics may seem even more “alien.” This approach involves calculating the probability of a future event through a predictive distribution based on Bayes’ theorem, all within a known and real initial probability margin. Essentially, it serves as a statistical “crystal ball.” However, it sometimes merely amplifies future trends based on current, non-significant or non-inferior results. Ultimately, it is clear that both concepts are well-suited to each other. With the former, it is easy to conclude “equivalence,” and with the latter, future trends can be predicted based on that initial result. 

To introduce a personal opinion that many share, these methodologies have been incorporated into various clinical trials in controversial fields due to their suitability for producing favorable results: false equivalence in the former, and false anticipation of benefits in the latter, thereby forming a rapid, industry-serving body of evidence. Applying this rationale to TAVR versus surgical aortic valve replacement makes the trajectory clear: establish short-term non-inferiority to justify a seemingly less invasive technique and, subsequently, with known early benefits in the first one or two years due to reduced surgical burden, anticipate future results that reinforce the indication even in clinical guidelines. 

Thus, the proposed analysis, having employed actual mid-term results, once past the well-known two-year survival curve threshold, seems to offer results “outside the script,” as it amplifies trends that, in many studies, are already significantly superior for the surgical option. We must also consider that, despite randomization, patients undergoing TAVR in high- and moderate-risk studies were especially comorbid, which could have impacted the results when considered in a combined analysis. 

Although this study draws attention to the unchecked advance of interventional procedures by yielding positive results for surgery, perhaps its primary relevance is demonstrating that these methodologies, so distinct from “traditional” ones, skew the data and introduce biases. Put simply, the “hunter has been hunted,” or performing Bayesian analysis with data beyond two years of follow-up has given the TAVR industry a dose of “its own medicine.” In a consensus document previously reviewed on this blog regarding left main coronary artery disease treatment, certain types of analyses, such as all-cause mortality or composite events, were proscribed. Let us hope that, at last, the EBM principle will be embraced in structural interventions: “evidence-based medicine” rather than “evidence-biased medicine.” 

REFERENCE: 

Heuts S, Kawczynski MJ, Sardari Nia P, Maessen JG, Biondi-Zoccai G, Gabrio A. Bayesian interpretation of non-inferiority in transcatheter versus surgical aortic valve replacement trials: a systematic review and meta-analysis. Interdiscip Cardiovasc Thorac Surg. 2023 Nov 2;37(5):ivad185. doi: 10.1093/icvts/ivad185.