Development of a Predictive Model for Persistent Instability Following Conservative Management of Type II Odontoid Fractures

Abstract

Odontoid fractures are common cervical spine fractures; however, significant controversy exists regarding their treatment. Risk factors for failure of conservative therapy have been identified, although no predictive risk score has been developed to aid in decision-making. A retrospective review was conducted of all patients evaluated at a level 1 trauma center. Patients identified with type II odontoid fractures as classified by the D'Alonzo Classification system who were treated with external orthosis were included in analysis. Patients were considered to have failed conservative therapy if they were offered surgical intervention. A machine learning method (Risk-SLIM) was then utilized to create a risk stratification score based on risk factors to identify patients at high risk for requiring surgical intervention due to persistent instability. A total of 138 patients were identified as presenting with type II odontoid fractures that were treated conservatively; 38 patients were offered surgery for persistent instability. The Odontoid Fracture Predictive Model (OFPM) was created using a machine learning algorithm with a 5-fold cross validation area under the curve of 0.7389 (95% CI: 0.671 to 0.808). Predictive factors were found to include fracture displacement, displacement greater than 5 mm, comminution at the fracture base, and history of smoking. The probability of persistent instability was <5% with a score of 0 and 88% with a score of5. The OFPM model is a unique, quick, and accurate tool to assist in clinical decision-making in patients with type II odontoid fractures. External validation is necessary to evaluate the validity of these findings.