Multi-label prescription review for prophylactic use of antibiotics in type Ⅰ incision based on Bayesian optimized XGBoost
OBJECTIVE To construct a multi-label prescription review model for the usage of antimicrobials in typeⅠ incision based on Bayesian optimized extreme gradient boosting(XGBoost)and analyze the impacts of different properties by SHAP(SHapley Additive exPlanations)to achieve the goal of reasonable management of antimicro-bials and effective solutions to antimicrobial resistance.METHODS The clinical antimicrobial prescription data for type Ⅰ incision and patients'information were collected from a three-A hospital in Northeast China from Nov.2021 to Oct.2022.Four types of data were used including prescriptions,patients,surgeries and antimicrobials.Multiple machine learning models were established and compared to select the most effective XGBoost model,classifier chains were used to handle multi-labels problem,Bayesian algorithms were used to optimize the model parameters,and the SHAP values were used to explain the model properties.RESULTS The GP-XGBoost model performed the best in terms of accuracy,precision,recall,specificity,F1 score,Area under the receiver operating characteristic curve(AUC)and other indicators,with the accuracy of 0.995 and AUC of 0.980.It was verified that the generalization ability of the model could be effectively improved by the TF-IDF algorithm integrating un-structured text data into the prediction model.By SHAP interpretable analysis,the department for prescribing,antimicrobials'name,single dose administration,drug usage duration,whether centrally procured and defined daily doses(DDDs)played an important role in the six binary classifiers,which indicated that the prescription re-view model paid more attention to the drug-related information.CONCLUSION Machine learning methods are ef-fective in antimicrobial prescription reviews,and the SHAP explanation method can identify the important factors that influence the judgment of unreasonable prescriptions.Both have practical significance for refined reviews of antimicrobial prescriptions,rational prescribing and reducing medical costs.
Type Ⅰ incisionProphylactic antibioticsPrescription reviewMulti-label learningXGBoostBayes-ian optimization
万方数据
维普
