Only fair accuracy of the radiographic classification of adult proximal humeral fractures in the Swedish Fracture Register: a cohort analysis.

Abstract

Quality registers are used for quality assessment, cost analyses, and research regarding outcomes of surgical and non-operative treatments. As the Swedish Fracture Register (SFR) expands, and is used as a platform for randomized trials, assuring reliability and accuracy of the data is essential. This study aimed to investigate the accuracy of the radiographic classification data for proximal humerus fractures recorded in the SFR. All radiographic images of 171 patients with a proximal humerus fracture registered in the SFR between 2019 and 03-01 and 2019-08-31 at 3 hospitals were included. The radiographs were independently assessed at 2 occasions >3 weeks apart by 1 surgeon at each center and IRR was calculated to validate the modification of the AO/OTA classification used in the register. A "gold standard" classification for each patients' images was then established with a consensus discussion involving 4 shoulder surgeons. The gold standard classification was compared with the classification registered in the SFR. Intra-rater reliability was moderate (kappa 0.549-0.596) with percent agreement (PA) 61-66 %. Inter-rater reliability was also moderate (kappa 0.508-0.557) with PA 58-62 %. Accuracy of the SFR recordings compared with gold standard was fair with kappa 0.36 (95 % CI 0.297-0.425) and PA 44 %. For registers to be of use the accuracy of data is essential as well as coverage, completeness, validity and reliability. The modified AO/OTA classification for proximal humerus fractures used in the SFR had moderate reliability but registered data only fair accuracy compared with a gold standard. This questions its value as a base for scientific research and clinical decisions.