基于机器学习模型的哌拉西林钠他唑巴坦钠相关低钾血症风险预测OA

Objective To construct a risk prediction model for piperacillin sodium and tazobactam sodium(PTZ)-related hypokalemia to assist in early identification of high-risk patients.Methods A total of 850 hospitalized patients using PTZ in Norinco General Hospital from January 2024 to December 2024 were retrospectively collected,and were randomly divided into the training set and the internal validation set in an 8∶2 ratio.Clinical data from 227 hospitalized patients in Xi'an Fifth Hospital were collected using the same criteria as the external validation set.Least Absolute Shrinkage and Selection Operator(LASSO)regression was adopted to screen characteristic variables;six machine learning algorithms,including system comparison logistic regression(LR),were systematically compared;the predictive performance of the model was evaluated through factors such as receiver operating characteristic(ROC)curve area under the curve(AUC);and Shapley Additive Interpretation(SHAP)was adopted to analyze characteristic variable contributions.Results 15 characteristic variables were included in the model construction,and the LR model had the best overall performance with an AUC of 0.827,sensitivity of 0.651,specificity of 0.903,accuracy of 0.778,and F1 score of 0.693 in the internal validation set;the AUC in the external validation set was 0.800,indicating good generalization ability;the SHAP scatter plot showed that the top 2 importance of characteristic variables predicted by the LR model were baseline serum potassium level and cumulative PTZ dose,which were negatively and positively correlated with the risk of PTZ-related hypokalemia,respectively.Conclusion The LR model performs well in PTZ-related hypokalemia prediction task,which can be taken to intervene in important prediction characteristic variables.