Quality assessment and quality control of echocardiographic performance in a large multicenter international study: Valsartan in Heart Failure Trial (Val-HeFT)

Objective: To qualify 302 multinational echocardiography sites to record and read serial studies and to monitor quality in 5010 patients randomized into Valsartan in Heart Failure Trial (Val-HeFT). Background: Decentralized echocardiography reading is unprecedented in large clinical trials. Methods: Single and duplicate recordings, and triplicate readings of echocardiographic variables were submitted to 3 core laboratories. Quality of recording was defined with a 16-point scoring system; accuracy of reading by agreement with core readings; reproducibility by agreement between the duplicate studies. Results: Seventy-five percent of initial submissions were approved for recording, and 50% for reading. Resubmissions were evaluated until approval. Initial scores of sites approved with 1 versus 2 submissions differed, 13.8 ± 1.4 versus 10.6 ± 2.0, P <.001; final score was similar, 13.4 ± 1.6, P = ns. Initial score of sites approved after 3 or more submissions differed, 9-5 ± 1.9, P <.001; final score improved to 12.7 ± 1.9, but remained lower, P <.001. Expressed as 95% limits of agreement (mean difference ± 1.96 × SD), accuracy = -0.04 × 0.74 cm for left ventricular internal end-diastolic diameter (LVIDd); -0.29 ± 14.3% for ejection fraction (EF); reproducibility = 0.00 ± 0.53 cm for LVIDd; -0.25 ± 8.3% for EF. Quality of random sampling at baseline, 4, 12, and 18 months showed recording scores of 11.7 ± 2.7, 12.3 ± 2.4, 12.1 ± 2.2, and 11.4 ± 2.0, P = .24. Power analysis revealed differences of 0.09 cm for LVIDd, and 0.86% for EF detectable with a power of 90% and alpha of 5%. Conclusion: The qualifying process improved echocardiography recording and reading to bring 95% of the sites to an equivalent level of quality. Monitoring quality found that recording quality and reading accuracy were maintained 18 months into the trial. Reproducibility, given the large sample size, will be able to detect small changes in LVIDd and EF.