Comparative analysis of machine learning algorithms for predicting tibial intramedullary nail length from patient characteristics

This study aimed to evaluate the performance of five machine learning algorithms in predicting tibial intramedullary nail length using patient demographic data (gender, height, age, and weight), with the goal of developing a clinically relevant and accurate predictive model. Retrospective data from 155 patients who underwent tibial intramedullary nailing at the Affiliated Jiangyin Hospital of Nantong University were analyzed. After data cleaning, outlier handling, and gender encoding, the dataset was divided into an 80% training set and 20% testing set. Models were trained and evaluated using root mean squared error (RMSE), mean absolute error (MAE), coefficient of determination (R2), and correlation analysis. Key variables included height (cm), weight (kg), age (years), and gender. The XGBoost model demonstrated superior clinical precision, achieving the lowest testing RMSE (9.15 mm) and MAE (7.56 mm), with an R2 of 0.871, explaining 87.1% of variance in nail length. While the random forest model had the highest R2 (0.874) and correlation coefficient (r = 0.935), XGBoost outperformed all models in error metrics, critical for minimizing surgical complications. Variable importance analysis identified height as the most influential factor, followed by weight and age. All models achieved acceptable accuracy (≥ 86.21%) within a ± 15 mm error margin, compatible with intraoperative adjustments. Machine learning, particularly XGBoost, significantly improves preoperative prediction of tibial intramedullary nail length compared with traditional methods. Level of evidence IV.