Liver stiffness measurement (LSM) by transient elastography has been validated as a noninvasive method to stage liver fibrosis. Few studies have evaluated the learning curve of this method and its reproducibility has led to controversy results. We aimed to evaluate the intra- and interobserver agreement of transient elastography as well as its learning curve for definition of an experimented operator.

We retrospectively analyzed 922 examinations performed in 544 patients during a training program of transient elastography. Patients with chronic hepatitis C with or without HIV co-infection that had two examinations by the training operator (intraobserver analysis; n=125) or examination by both training and experimented operators (interobserver analysis; n=151) in the same day were included. LSM was converted to METAVIR score: <7.1 as F0F1, 7.1–9.4 as F2, 9.5–12.4, as F3 and >12.4kPa as F4.

The overall intra- and interobserver intraclass correlation coefficient [ICC 95% CI] were 0.926 (0.901–0.951) and 0.912 (0.885–0.939), respectively. Measurements were correlated [Spearman's] in intra- [0.906, P<0.0001] and interobserver [0.907, P<0.0001] analysis. Reliability values [kappa (SE)] were k=0.74 (0.09) and k=0.85 (0.08) for fibrosis stages F≥2 and k=0.77 (0.09) and k=0.75 (0.08) for cirrhosis in intra- and interobserver analysis, respectively. Agreement was improved when operator's experience was higher than 100 exams. However, it was observed discordance for fibrosis staging between examinations in a quarter of patients.

Although there was a considerable discrepancy on fibrosis staging between examinations and a small power, transient elastography had an acceptable reproducibility in our population. Performance of at least 100 examinations should be used to define an experimented operator.