Intraobserver and interobserver reliability in assessment of neonatal cranial ultrasounds

Intraobserver and interobserver reliability in assessing neonatal cranial ultrasounds for periventricular-intraventricular hemorrhage (PVH-IVH) is not well studied; therefore, studies were designed to address this. For intraobserver reliability 180 cranial ultrasounds (360 hemispheres) were randomly selected from > 2000 ultrasounds and read twice by one radiologist in a blinded fashion. Ninety-eight percent were interpreted identically; of the 2% reinterpreted differently, all were initially abnormal but normal on the second reading. The least agreement occurred when interpreting ventricular size. Only four infants (1.1%) were placed in an unfavorable prognostic category (grades III and IV) on the first reading and a favorable prognostic category on the second interpretation (no bleed, grades I and II). To determine interobserver reliability, 20 sonograms were interpreted by eight independent observers representing five institutions. Using the multiple rater kappa κ statistic, we determined interobserver agreement on overall impression (normal vs. abnormal), presence and extent of PVH-IVH (i.e. grade), presence of residual cyst, and ventricular dilatation. Greatest degree of agreement occurred when determining normal vs. abnormal, residual cyst, no bleed, and grades III and IV PVH-IVH. Poorest agreement occurred when reading grades I and II PVH-IVH and ventricular dilatation. After condensing interpretations of cranial ultrasounds into two prognostic categories, i.e. favorable (no bleed, grades I and II) and unfavorable (grades III and IV), there was excellent agreement among the observers.