ABSTRACT
This letter evaluates whether overestimation of control event rates and the use of absolute noninferiority margins influenced conclusions of randomized trials comparing transcatheter and surgical aortic valve replacement. By reanalyzing published noninferiority trials using observed event rates and corresponding relative margins, we assessed the robustness of trial conclusions to alternative margin specifications. Our findings inform interpretation of noninferiority evidence underlying contemporary clinical decision-making in aortic valve replacement.
Background
Noninferiority trial design is increasingly applied in clinical research to determine if a new treatment performs at least as well as an established treatment within an acceptable clinical margin The noninferiority margin defines the maximum allowable difference by which the new treatment can be less effective than the established 1 while still being deemed clinically acceptable. This margin can be expressed either on a relative scale, eg, preestablished relative risk or hazard ratio, or an absolute scale, eg, preestablished absolute risk increase.
The use of absolute margins has recently been criticized, as a lower-than-expected event rate increases the likelihood of achieving noninferiority, despite the fact that the results are less certain when event rates are low In contrast, trials with relative noninferiority margins, similar to superiority trials, are less likely to achieve noninferiority when event rates are lower than anticipated and the estimated treatment effects are uncertain. Since the observed event rate (OER) in clinical trials is often considerably lower than the assumed event rate (AER), noninferiority trials using absolute margins are often biased in favor of noninferiority. ,
Absolute noninferiority margins have been used by most trials that compared transcatheter versus surgical aortic valve replacement (TAVR vs SAVR). The aim of this analysis is to critically evaluate the event rate assumptions and use of the noninferiority approach in recent TAVR vs SAVR trials.
Methods
A systematic search was conducted to identify clinical trials comparing SAVR vs TAVR (Ovid Medline (1946 to present), Ovid Embase (1946 to present) and Cochrane Library without time limits). Expected and observed event rates, noninferiority margins, and hypothesis test results were collected for each powered endpoint. The ratio of assumed to observed event rate (AER/OER) was calculated as a descriptive measure of the magnitude and direction of event rate misestimation (ratio >1 indicating overestimation and < 1 underestimation), consistent with prior methodological analyses For noninferiority trials using absolute margins, the corresponding assumed relative margin (CARM) and relative margin that corresponded to the absolute margin based on the observed control event rate (corresponding observed relative margin, CORM), and these were compared by plotting observed versus assumed values For example, in PARTNER 3, the assumed control event rate was 16.6% with a prespecified absolute noninferiority margin of 6%. The corresponding assumed relative margin (CARM) was therefore 22.6/16.6 = 1.36. Using the observed control event rate of 15.1%, the same absolute margin corresponds to a relative margin (CORM) of 21.1/15.1 = 1.39. To assess whether using a relative margin would have changed the trial outcome, relative risks and confidence intervals were calculated using the Farrington–Manning method with a 1-sided α level of 0.025, and noninferiority was determined based on the P value relative to the prespecified 1-sided α level.
Results
Eight trials were included for a total of 5,006 patients undergoing TAVR and 4,925 SAVR. ,,,,,,, Only 1 trial used relative noninferiority margin In trials using absolute noninferiority margin the median absolute margin was 6% (interquartile range [IQR] 5.5-7.25; Table ). The control even event rate was overestimated in 7 trials (87.5%) with an average overestimation ratio of 1.33 and was underestimated in 1 (12.5%). Despite overestimation, the corresponding CORM was not significantly higher (1.37, IQR 1.30-1.41 vs 1.40, IQR 1.34-1.63; P =.54). Taking into account the newly calculated noninferiority relative margin, only 1 study was found to be noninferior by absolute margin but not noninferior by relative margin while the majority of trials were noninferior by both the absolute and relative margin ( Figure ).
Table
Trial characteristics.
| Trial | Primary endpoint | Type of NI margin | Control AER (%) | Absolute NI margin (%) | Expected CRM | Control OER (%) | Observed CRM | Upper 95% bound for observed CRM | Conclusions |
|---|---|---|---|---|---|---|---|---|---|
| PARTNER 1A | Death from any cause | Absolute | 32 | 7.5 | 1.23 | 26.8 | 1.27 | 1.23 | Noninferiority criterion met using absolute margin but not CRM |
| PARTNER 2A | Death from any cause or disabling stroke | Relative | 30 | NA | 1.20 | 21.1 | 1.20 | 1.15 | Noninferior by both absolute margin and CRM |
| PARTNER 3 | Composite of death, stroke, or rehospitalization | Absolute | 16.6 | 6 | 1.36 | 15.1 | 1.39 | 0.81 | Noninferior by both absolute margin and CRM |
| CoreValve | Death from any cause | Absolute | 20 | 7.5 | 1.37 | 19.1 | 1.39 | 0.99 | Noninferior by both absolute margin and CRM |
| SURTAVI | Composite of death from any cause or disabling stroke | Absolute | 17 | 7 | 1.41 | 14.0 | 1.50 | 1.16 | Noninferior by both absolute margin and CRM |
| EVOLUT | Composite of death from any cause or fatal or nonfatal stroke | Absolute | 15 | 6 | 1.40 | 6.7 | 1.89 | 1.19 | Noninferior by both absolute margin and CRM |
| UK-TAVI | Death from any cause | Absolute | 7.5 | 5 | 1.67 | 6.6 | 1.75 | 1.19 | Noninferior by both absolute margin and CRM |
| DEDICATE | composite of death from any cause or fatal or nonfatal stroke | Absolute | 6.2 | 1 | 1.16 | 10.0 | 1.10 | 0.81 | Noninferior by both absolute margin and CRM |
Stay updated, free articles. Join our Telegram channel
Full access? Get Clinical Tree
