Risk Factor Analysis And Risk Prediction Model Construction For Acute Kidney Injury In Patients With Invasive Mechanical Ventilation Based On Machine Learning Algorithms

Objective:To analyze the risk factors for the combined occurrence of acute kidney injury(AKI)in patients with invasive mechanical ventilation,and construct Nomogram model and Tree Net risk prediction models using machine algorithms.Methods:A total of 445 patients who received invasive mechanical ventilation in the intensive care unit of our hospital from July 2020 to June 2022 were selected as the study subjects.Patients were divided into AKI and non-AKI groups based on whether AKI occurred within 7 days after mechanical ventilation.Univariate and multivariate logistic regression analyses were conducted to explore the influencing factors of AKI in patients undergoing invasive mechanical ventilation.The Nomogram and Tree Net risk prediction models were constructed using R software and SPM software,and the models were evaluated.Results:Among the 445 patients,182 developed AKI,with an incidence rate of 40.8%.Multivariate analysis suggested that peak airway pressure(OR=1.154,95%CI=1.092-1.220),uric acid(OR=1.004,95%CI=1.002-1.007),procalcitonin(OR=1.034,95%CI=1.008-1.061),cardiopulmonary bypass(OR=3.035,95%CI=1.635-5.633),and fluid overload(OR=2.750,95%CI=1.062-7.119)were independent risk factors for AKI in mechanically ventilated patients.The consistency index of the Nomogram model was0.808(95%CI=0.765-0.851).Bootstrap validation showed that the model had good discriminative ability based on the ROC curve,AUC=0.816(95%CI=0.773-0.858),the calibration curve showed good calibration accuracy in the model,the clinical decision curve showed a good net benefit rate,and the HL goodness-of-fit test(χ~2=3.964,P=0.860).The Tree Net model showed that there were 24 important variables,and the classification support was optimal in the 136th cycle.The AUC of the training set was 0.898,and the AUC of the test set was 0.926.The model had an accuracy rate of 80.90%,a precision rate of 72.09%,a recall rate of 86.11%,a specificity rate of 77.34%,and an F1 score of78.48%.The overall fit of the model was good,with high predictive power.The top three indicators and scores associated with AKI were peak airway pressure(100.00),procalcitonin(61.72),and uric acid(60.94),all of which were positively correlated with the occurrence of AKI.Conclusion:Factors such as peak airway pressure,uric acid,procalcitonin,cardiopulmonary bypass,and fluid overload can increase the risk of AKI in patients with invasive mechanical ventilation.The Nomogram model constructed based on R software has high precision and discrimination,and the Tree Net prediction model constructed based on SPM software has good overall fit,which can aid in early identification and prevention of AKI in patients receiving invasive mechanical ventilation.