A Machine Learning Approach to Predicting Radiographic Outcomes of Nonsurgically Treated Distal Radius Fractures

Comprehensive Summary

Taleghani et al. aimed to create a machine learning model to predict radiographic outcomes of nonsurgically treated distal radial fractures. The model incorporated pre-reduction and post-reduction radiographic parameters and guidelines. The researchers conducted a retrospective chart review of adults with distal radial fractures who underwent closed reduction in the emergency department and had radiographs taken before reduction, immediately after reduction, and at six weeks. At the six-week mark, treatments were classified as successes or failures based on American Academy of Orthopaedic Surgeons acceptable reduction parameters. Five machine learning models were evaluated, and the 10 parameters with the highest Shapley values were included in a composite model. The composite model correctly predicted outcomes in 25 of 31 patients (AUC = 0.84, F1 = 0.81).

Outcomes and Implications

The model identified post-reduction palmar tilt, radial height, and Lindstrom score as the most important predictors of treatment success, highlighting the critical role of post-reduction radiographic parameters in clinical outcomes. These findings confirm that closed reduction is a necessary step in the treatment process and demonstrate that machine learning models can provide clinicians in the emergency department with supplemental information to guide treatment decisions for distal radial fractures.