A quantitative definition of scaphoid union: determining the inter-rater reliability of two techniques

BACKGROUND: Despite extensive literature supporting the use of computerized tomography (CT) scans in evaluating scaphoid fractures, there has not been a consensus on the methodology for defining and quantifying union. The purpose of this study was to test the inter-observer reliability of two methods of quantifying scaphoid union. METHODS: The CT scans of 50 non-operatively treated scaphoid fractures were reviewed by four blinded observers. Each was asked to classify union into one of three categories, united, partially united, or tenuously united, based on their general impression. Each reviewer then carefully analyzed each CT slice and quantified union based on two methods, the mean percentage union and the weighted mean percentage union. The estimated percentage of scaphoid union for each scan was recorded, and inter-observer reliability for both methods was assessed using a Bland-Altman plot to calculate the 95% limits of agreement. Kappa statistic was used to measure the degree of agreement for the categorical assessment of union. RESULTS: There was very little difference in the percentage of union calculated between the two methods (mean difference between the two methods was 1.2 ± 4.1%), with each reviewer demonstrating excellent agreement between the two methods based on the Bland-Altman plot. The kappa score indicated very good agreement (Ƙ = 0.80) between the consultant hand surgeon and the musculoskeletal radiologist, and good agreement (Ƙ = 0.62) between the consultant hand surgeon and the hand fellow for the categorical assessment of union. CONCLUSIONS: This study describes two methods of quantifying and defining scaphoid union, both with a high inter-rater reliability. This indicates that either method can be reliably used, making it an important tool for both for clinical use and research purposes in future studies of scaphoid fractures, particularly those which are using union or time to union as their endpoint. LEVEL OF EVIDENCE: Diagnostic, level III