To identify characteristics that contribute to surgical complexity in pilon fractures and to develop a machine learning (ML) Pilon Surgical Difficulty Score (PSDS) based on these factors. Retrospective cohort study. Academic Level I trauma center. Pilon fractures (OTA/AO Type 43) in adult patients treated with open reduction internal fixation. Various patient, injury, and radiological characteristics were assessed. Surgical difficulty was measured using 2 outcomes: (1) operative time and (2) perceived difficulty. Perceived difficulty was determined using the opinion of 16 fellowship-trained orthopaedic traumatologists on a 10-point scale. Significant predictors of difficulty were determined using univariate analyses. ML models were used to develop a PSDS for both operative time and surgical difficulty. One hundred operatively fixed pilon fractures were included. Predictors of operative time were age, OTA/AO classification, articular comminution, articular impaction, bone loss, delay to surgery, poor quality reduction, number of approaches, and number of articular fragments. Predictors of perceived difficulty included OTA/AO classification and delay to surgery. Operative time PSDS had a mean absolute error of 64 minutes and a 60-minute buffer accuracy of 59%. Perceived difficulty PSDS had a mean absolute error of 1.7 points and a 2-point buffer accuracy of 63%. ML was used to generate accurate PSDSs for operative time and difficulty for pilon fractures. Future work should aim to clinically validate these PSDSs, so they may improve patient outcomes. Level III Diagnostic.
Read full abstract