Letter to editor regarding the published manuscript: “A prospective comparison of the diagnostic accuracies of ultrasound and magnetic resonance imaging in preoperative staging of endometrial cancer”

The study, "A prospective comparison of the diagnostic accuracies of ultrasound and magnetic resonance imaging in preoperative staging of endometrial cancer," study aimed to compare the diagnostic performance of 2 paired radiological methods for assessing deep muscle invasion in the case of endometrial cancer. The surgical specimen defined the reference standard.

It is an important manuscript, and the methodology, presentation, and interpretation of results are essential for many researchers and clinicians. The manuscript is interesting and well written, although we believe some crucial aspects need clarification.

The authors calculated that the study required a minimum of 49 patients to detect a 10% difference in sensitivity at 80% power. They applied statistics introduced by Liu et al. [1], which is suitable for demonstrating non-inferiority or equivalence for paired data. From a general point of view, 49 patients appear a very low sample size, especially when considering that only 22 patients had the target condition, e.g., deep muscle invasion.

The exact study design is not defined in the method section. Still, applying Liu statistics and a 10% non-inferiority limit for sensitivity suggests that the authors have chosen a non-inferiority design. In other words, the authors probably wished to demonstrate that ultrasound (US) is not more than 10% less sensitive than magnetic resonance imaging (MRI) to detect deep muscle invasion.

According to the results, the sensitivity of US and MRI was 86% and 77%, respectively, but the authors conclude that the difference was not significant. What do the authors mean by this? The null hypothesis in a non-inferiority setting would state that US is inferior to MRI, and the difference exceeds the non-inferiority limit of 10%. In a superiority test setting, the null hypothesis would be that US is not significantly different from MRI. These are pretty different interpretations, and a better explanation of the intention as well as the analysis is necessary to understand the result.

In a non-inferiority study, the 95% confidence interval (CI) of the difference is of primary interest rather than the respective point estimates of the 2 methods. In the current paper, the authors do not report this difference. However, we estimate that the 95% CI of the sensitivity difference ranges between −13% and 30%. Consequently, US is possibly 13% less sensitive than MRI or 30% more sensitive. Therefore, if our assumptions are correct, the study failed to demonstrate non-inferiority.

The authors also say that the primary outcome is the comparison of accuracy between US and MRI. Since the sample size estimation focused on sensitivity, the choice of accuracy as the primary outcome is somewhat confusing. In the current paper, the accuracy of US and MRI was 75% and 76%, respectively, and the authors conclude that the accuracy is comparable. But again, the authors do not report an essential result, the 95% CI of the difference in accuracy. We estimate that the 95% CI of the accuracy difference ranges between −15% and 13%. That means US might be 15% less accurate than MRI. Using a non-inferiority margin of 10%, one can only conclude that non-inferiority is not demonstrated.

We believe the aim, study design, results, and conclusion need some clarification. Unless we misunderstand essential aspects of the methodology, the study included far too few patients to draw any conclusions. Consequently, the study is grossly underpowered and inconclusive, as indicated by the wide confidence intervals. We would appreciate the author's comments on sample size estimation, presentation, and interpretation of results.