基于贝叶斯网络的股骨颈骨折内固定术后股骨头坏死风险预测OA

Objective To construct a risk prediction model using a Bayesian network(BN)to predict osteonecrosis of femoral head(ONFH)after internal fixation for femoral neck fracture(FNF),facilitating the visualization and uncertainty interpretation of the model.Methods Using retrospective data from patients admitted to three hospitals in Shanghai for FNF and undergoing internal fixation surgery,an analysis and modeling were conducted.A total of 24 variables were included,including demographic characteristics,blood biochemical indicators,and surgery-related factors.Two machine learning algorithms were used for feature engineering,and a BN was used to build the model.The five-fold cross-validation was used to evaluate model performance and clinical applicability based on the area under the receiver operating characteristic curve(AUROC),sensitivity,specificity,F1 index,and decision curve analysis(DCA).Results The rate of the ONFH after internal fixation of FNF was 22％,and the BN model showed that complete weight-bearing time,visual analogue scale(VAS)score of pain,Harris hip score,reduction quality,injury-to-operation time,femoral neck shortening,and Garden classification were the direct influencing factors,while body mass index(BMI)and osteoporosis were indirect influencing factors.The AUROC,sensitivity,specificity,and F1 index of the model under the five-fold cross-validation was 0.931,0.927,0.921 and 0.922,respectively.BN model inference could provide a probabilistic explanation of outcomes.Conclusion The BN model has good performance and clinical applicability,which can visualize the direct and indirect influencing factors of ONFH and improve the uncertainty interpretation ability through probability inference.