Infection after surgery to treat a tibial shaft fracture can have devastating consequences, with significant associated costs and burdens. Although research has identified general risk factors that increase the likelihood of infection (including complexity of injury and fracture patterns and patient-related factors such as smoking and diabetes), predicting risks for individual patients remains difficult.
In a recent study in The Journal, investigators from the Machine Learning Consortium reported on an algorithm they developed to predict the risk of infection in specific patients who receive operative treatment for a tibial shaft fracture. To develop their model, the researchers used high-quality data from the SPRINT (Study to Prospectively Evaluate Reamed Intramedullary Nails in Patients with Tibial Fractures) and FLOW (Fluid Lavage of Open Wounds) randomized controlled trials.
The Australian researchers “trained” 5 machine learning algorithms and tested them against various performance measures to evaluate 1,822 fractures, including 170 (9%) that developed an infection. Based on predictive performance in that derivation portion of the study, 3 algorithms were validated and 1 prediction model was found to be superior. In that model, Gustilo-Anderson Type IIIA and IIIB fractures, age, AO/OTA type 42C3 fractures, crush injuries, and falls were the strongest predictors of infection.
Researchers have made their model available in an online, open-access prediction tool. Although the authors emphasize that this preliminary tool is intended for research and not for widespread clinical use, I think it has profoundly positive potential. Being able to risk-stratify a patient with a tibial shaft fracture at or near the time of admission could allow surgeons to closely monitor—and intervene sooner—in fracture cases at risk for infection, thereby possibly preventing devastating complications. This prediction tool certainly needs external validation prior to “prime-time” adoption, but when it comes to exploring artificial intelligence and machine learning in orthopaedics, the future is now.
Matthew R. Schmitz, MD
JBJS Deputy Editor for Social Media