As for any commonly used biomarker, whether in clinical practice or clinical research, accuracy and reproducibility is of the utmost importance. Echocardiography is no exception. In fact, echocardiography is viewed by some as having a higher degree of variability than other imaging modalities, and so every possible effort to reduce that variability should be implemented. In clinical research, that variability from one clinical site to the next means that simply pooling data from echo reports is no longer an acceptable practice. Echo Core Labs (ECLs) provide an opportunity for standardized acquisition and centralized, blinded, independent image analysis which results in reduced variability and, for these reasons, have become the gold standard in clinical trials.

However, is one ECL ‘gold standard’ comparable to another ECL ‘gold standard’? The current report by Khouri MG et al. compares echo analysis from two prestigious academic ECLs and finds that there is significant variability across labs. Despite a desire to combine data across both of these ECLs, and despite conducting web-based training sessions across both ECLs, the reproducibility of mitral valve inflow velocities, left ventricular ejection fraction (LVEF), and a simplified global longitudinal strain (GLS) were worse between the ECLs than within an ECL. Inter-lab reproducibility was improved in some measurements by retraining but was minimally improved for GLS. While several limitations could be ascribed to this investigation (and indeed the authors thoughtfully acknowledge most of them), this article provides important lessons in our pursuit of the “perfect way of measuring LV function” in clinical trials.

In our need to find an optimal method of performing measurements (accuracy and reproducibility), there are many preconceptions. In fact, we are often tempted to think that

• –
  

  any Core Lab measurement is more accurate than those performed at clinical centers;

  
  
• –
  

  newer quantitative variables such as GLS are better than the “old-fashioned” LVEF; and

  
  
• –
  

  automated (or semi-automated) measurements are more accurate than those fully performed by humans.

While these are reasonable assumptions based on our experience, the investigation by Khouri et al. in this Issue of the Journal provides us with some humbling results. After reading this paper, some basic (but important) questions need to be addressed.