A retrospective study if the validation of the modified radiographic/tomographic union score for nailed or plated treated fractures of long bones

Abstract

Background: About 5 to 10% of the fractures of long bones in humans proceed to non-unions. Some scores exist to assess the union or not of long bone fractures. This work attempts to study three scores: the modified Radiographic Union Score (mRUS), the Tomographic Union Score (TUS) and the Radiographic Union Scale for Tibial fracture (RUST).
Aim: The major objective is to validate the mRUS and to make an explorative evaluation of the RUST as a tool to predict which patients have an increased risk of Delayed-Union/Non-union (DU/NU) by finding a threshold value for long bones. Furthermore, the RUSTUS study aimed to evaluate the potential of a binary variable “healed status”, derived from individual scores of cortexes, in confirming bone healing. In order to understand the subject, a critical analysis of the existing literature and an inventory of all existing union scores was made. Finally, interviews were made to investigate the use of union score in daily medical practice.
Methods: A retrospective study was conducted on 399 fractures of long bones in humans. They were scored by two independent readers, one radiologist and one orthopedic surgeon. In addition, six interviews were performed on three radiologists and three orthopedic surgeons.
Analysis: Among the fractures investigated, 73.4% were healed and 26.6% developed DU/NU. Three sub-populations stand out; tibial fractures, patients taking NSAIDs and high energy fractures. Concerning the mRUS validation and the RUST evaluation, the threshold of 10 had the best results to predict that a fracture will be healed. The intraclass correlation coefficient obtained for the inter-reader variability are 0.46 for the mRUS and 0.38 for the RUST. However, when the significant difference was defined as 2 or more by cortex, the variability decreased and the two readers scored fractures in a similarly way in 87.07%, 92.11% and 88.1% for tibial, femoral and humeral fractures respectively. The assessment of the healed binary variable obtained an agreement proportion of 92%. However, the Kappa obtained is moderate (0.46). Interviews highlighted the fact that orthopedic surgeons and radiologists don’t use radiographic scores to predict the union of fractures.
Conclusion: A threshold of 10 for the mRUS and the RUST was found to predict the union of long bone fractures in humans treated with nails or plates within three months of the fracture occurrence. To decrease the variability of the RUSTUS study, a solution could be increasing the number of independent readers or to use Artificial Intelligence (AI) to standardize the methodology of scoring the fractures. The performed interviews indicate that radiologists are more willing to trust AI tools than the orthopedic surgeons.

Date of Award	23 Jan 2020
Original language	English
Awarding Institution	University of Namur
Supervisor	Yves Geysels (Supervisor)