Objective To compare the efficiency between random forest model and Logistic regression model for predicting the occurrence risk of unscheduled resurgery. Methods Patients' information who requested two⁃times surgery during one hospital stay was screened from the surgical anesthesia system. All patients undergoing unscheduled resurgery (n=219) were extracted as study group, whereas patients in the corresponding departments who underwent scheduled resurgery (n=14,311) were as control group. The unscheduled resurgery prediction model was established by the application of random forest model and Logistic regression model. The prediction efficiency of the two models was evaluated by employing area under the receiver operating characteristic curve. Results (1) The results of Logistic regression analysis revealed that previous intraoperative blood transfusion, suffering from malignant tumor, number of comorbidity, previous healing classification of surgical incision, previous surgical level, previous surgical duration, previous category of surgical incision were the influencing factors for the occurrence of unscheduled resurgery (P<0.05). Area under the curve of Logistic regression prediction model was 0.922, and the sensitivity, specificity, and accuracy rate were 92.59%, 79.11%, and 79.28%, respectively. (2) The results of importance ranking of characteristic variables in random forest model indicated that the importance of variables such as previous category of surgical incision, previous intraoperative blood transfusion, previous surgical level, previous healing classification of surgical incision, number of comorbidity, and suffering from malignant tumor came top in. Area under the curve of random forest prediction model was 0.866, and the sensitivity, specificity, and accuracy rate were 80.00%, 93.33%, and 86.66%, respectively. Area under the curve of Logistic regression prediction model was larger than that of random forest prediction model, but no statistically significant difference was found (P>0.05). Conclusion Comprehensive use of Logistic regression model and random forest model, and complementing the results of the two analysis to each other, can predict the risk factors for unscheduled resurgery from various aspects, and obtain better prediction efficiency.

无