Interrater and Intrarater Measurement Reliability of Noncontrast Computed Tomography Predictors of Intracerebral Hemorrhage Expansion

Background and Purpose— Early hematoma expansion after intracerebral hemorrhage is a potentially modifiable predictor of outcome and a promising therapeutic target. Radiological markers seen on noncontrast computed tomography can help predict hematoma expansion and risk stratify patients presenting with intracerebral hemorrhage. Our objective was to assess the interrater and intrarater reliability of 5 commonly reported noncontrast computed tomographic markers of hematoma expansion. Methods— Four readers independently reviewed images from 40 patients from 2 intracerebral hemorrhage imaging databases (PREDICT Collaboration [Predicting Haematoma Growth and Outcome in Intracerebral Haemorrhage Using Contrast Bolus CT] and Massachusetts General Hospital). Readers were blind to all demographic and outcome data and used accepted definitions to establish the presence or absence of intrahematoma hypodensities, blend sign, fluid level, irregular hematoma morphology, and heterogeneous hematoma density. We calculated interrater and intrarater agreement and stratified kappas for the 5 imaging markers. Results— Interrater agreement was excellent for all 5 markers, ranging from 94% to 98%. Interrater kappas ranged from 0.67 to 0.91 (the lowest for fluid level). Interrater agreement had a similar pattern, ranging from 89% to 93%, with Kappas ranging from 0.60 to 0.89. Conclusions— We show that 5 commonly used noncontrast computed tomographic imaging findings all have good-to-excellent interrater and intrarater reliabilities, with the best kappa for blend sign, hypodensities, and heterogeneity.