JMIR Publications

ABSTRACT

Background:

The absolute number of femoral neck fractures (FNFs) is increasing, and the prediction of traumatic femoral head necrosis remains difficult. Machine learning algorithms have the potential to be superior to traditional prediction methods in the prediction of traumatic femoral head necrosis.

Objective:

The aim of this study was to use machine learning to construct a model for the analysis of risk factors and prediction of osteonecrosis of the femoral head (ONFH) in FNF patients after internal fixation.

Methods:

We retrospectively collected preoperative, intraoperative and postoperative clinical data of FNF patients in four hospitals in Shanghai and followed up with the patients for more than two and a half years. A total of 259 patients with 43 variables were included in the study. The data were randomly divided into a training set (n=181) and a validation set (n=78) at a 3:1 ratio. The external data (n=376) came from a retrospective cohort study of FNF in three other hospitals. LASSO regression and the support vector machine (SVM) algorithm were used for variable selection. Logistic regression, random forest, SVM and eXtreme Gradient Boosting (XGBoost) were used to develop the model on the training set. The validation set was used to tune the model hyperparameters to determine the final prediction model, and the external data were used to compare and evaluate model performance. We compared the accuracy, discrimination (sensitivity, specificity, AUC, ROC curve) and calibration (log-loss, calibration curve) of the models to identify the best machine learning algorithm for predicting ONFH. SHapley Additive ExPlanations (SHAP) and Local Interpretable Model-agnostic Explanations (LIME) were used to determine the interpretability of the black box model.

Results:

Ten variables were selected for the models. The XGBoost model performed best on the validation set and external data, including six predictors: VAS score, reduction quality, Garden classification, time to surgery, injury cause and fracture position. The accuracy, sensitivity and AUC of the model on the validation set were 0.987, 0.929 and 0.992, respectively. The accuracy, sensitivity, specificity and AUC of the model on external data were 0.907, 0.807, 0.935 and 0.933, respectively, and the log-loss was 0.279. The calibration curve demonstrated good agreement between the predicted probability and actual risk. The interpretability of features and individual predictions were realized using SHAP and LIME algorithms. Additionally, the XGBoost model was translated into a self-made online risk calculator to estimate an individual’s probability of ONFH.

Conclusions:

Machine learning exhibits good performance in predicting ONFH after internal fixation of FNF. The six-variable XGBoost model predicted the risk of ONFH well and had good generalization ability on external data, which can be used for the clinical prediction of ONFH after internal fixation of FNF.