Machine Learning Algorithm to Predict Change in the Decision-Making for Thoracolumbar Fractures Without Neurological...

Area of Science:

• Orthopedic surgery
• Radiology
• Machine learning in medicine

Background:

• Thoracolumbar fractures (TLFs) without neurological deficits often require careful assessment for posterior ligamentous complex (PLC) injury.
• Magnetic resonance imaging (MRI) can alter treatment recommendations based on PLC assessment, but its routine use needs justification.

Purpose of the Study:

• To develop and validate a machine learning algorithm to predict MRI-induced changes in thoracolumbar AO Spine injury severity score (TLAOSIS) treatment recommendations for neurologically intact TLFs.
• To identify cost-effective indications for MRI in AO Spine type A fractures.

Main Methods:

• A multicenter study involving 619 neurologically intact TLFs (AO Spine A-fractures) who underwent both computed tomography (CT) and MRI.
• A classification and regression tree (CART) model was developed using CT findings (M1 modifier for PLC injury), AO fracture subtype, and spine level.
• Model performance was evaluated using area under the receiver operating curve (AUC), sensitivity, specificity, and accuracy.

Main Results:

• MRI altered TLAOSIS treatment recommendations in 13.2% of cases.
• The CART model, primarily using the M1 modifier, demonstrated high predictive performance with an AUC of 0.93, sensitivity of 87.5%, and specificity of 96.3%.
• The algorithm accurately predicted changes in TLAOSIS recommendations, with mean cross-validation accuracy of 92.9%.

Conclusions:

• The developed CART model effectively predicts changes in TLAOSIS treatment recommendations following MRI in neurologically intact AO A-type thoracolumbar fractures.
• This algorithm offers a cost-effective approach to guide MRI utilization, ensuring thorough PLC assessment while avoiding unnecessary imaging.