Interobserver variability in assessment of cranial ultrasound in very preterm infants

Summary Background Cranial ultrasound (cUS) findings help doctors in the clinical management of preterm infants and in their discussion with parents regarding prediction of outcome. cUS is often used as outcome measure in clinical research studies. Accurate cUS performance and interpretation is therefore required. Aims The aims of this study were (i) to assess the interobserver variability in cUS interpretation, and (ii) to evaluate whether level of cUS expertise influences the interobserver variability. Methods Fifty-eight cUS image series of preterm infants born below 32 weeks of gestation collected within the Swiss Neonatal Network were sent to 27 observers for reviewing. Observers were grouped into radiologists, experienced neonatologists and less experienced neonatologists. Agreement between observers was calculated using Kappa statistics. Results When cystic periventricular leukomalacia, intraventricular haemorrhage and periventricular haemorrhagic infarction were combined to one outcome, agreement among all observers was moderate. When divided into subgroups, kappa for the combined outcome was 0.7 for experienced neonatologists, 0.67 for radiologists and 0.53 for inexperienced neonatologists. Marked difference in interobserver agreement between experienced neonatologists and radiologists could be found for haemorrhagic periventricular ifraction (HPI). Conclusions Our results suggest that interobserver agreement for interpretation of cUS varies from poor to good varying with the type of abnormality and level of expertise, suggesting that widespread structured training should be made available to improve the performance and interpretation of cUS.