Cardiac function by MRI in congenital heart disease: impact of consensus training on interinstitutional variance

To investigate the impact of interinstitutional variance (=interobserver variance between institutions) for volumetric and flow cardiac MR (CMR) data and if training on image reading could improve bias.

In a three-center study, a total of 32 adults with repaired Tetralogy of Fallot and 23 controls underwent CMR using standardized protocols for ventricular volumes/mass (by transverse and short-axis cine-MRI) and pulmonary/aortic blood flow by velocity-encoded MRI (VEC-MRI). Data were analyzed blinded and independently in each institution by experienced readers. Interinstitutional variance was determined before/after training on consented guidelines for image analysis.

In patients, initial interinstitutional variability of right ventricular parameters was substantial but decreased by training. On transverse planes, variation coefficient for end-diastolic/systolic volumes and ejection fraction decreased from 22%, 19%, and 19% to 7%, 10%, and 8%, respectively (P < 0.025). Left-ventricular variation coefficients improved for end-diastolic and stroke volumes from 8% and 15% to 4% and 6%, respectively (P < 0.007). For short-axis volumetry training resulted in narrowed limits of confidence. Variability did not significantly change in the controls. There was no significant difference between transverse/short-axis MRI. Interinstitutional variance for VEC-MRI in patients/controls was low (<4%).

Interinstitutional variance is an important source of variability in volumetric but not in flow CMR. Such variance can be reduced effectively by consented training.