Application of support vector machine and Logistic regression in early identification of late-onset sepsis caused by gram-negative bacteria in preterm infants
Objective:To investigate the application of support vector machine(SVM)and Logistic regression(LR)in the early identification of late-onset sepsis caused by gram-negative bacteria in preterm infants.Methods:A retrospective analysis was conducted on the clinical data of patients exhibiting positive blood cultures for late-onset sepsis in the neonatal intensive care unit from January 2015 to December 2019.Based on the clinical data,patients were categorized into the gram-negative bacilli sepsis group and the non-gram-negative bacilli sepsis group.Confounding factors in both groups were meticulously matched using a propensity score matching(PSM)of 1∶ 1.Subsequently,clinical indicators were analyzed post-successful matching,and predictive models were formulated using SVM and LR.The models were assessed for practicality,incorporating sensitivity,specificity,positive predictive value,negative predictive value,and the area under the receiver operating characteristic curve(AUC).The net reclassification index(NRI)was employed to evaluate the predictive capabilities of both models.Results:A total of 84 preterm infants with positive blood cultures for late-onset sepsis were identified,comprising 31 cases of gram-negative bacilli sepsis and 53 cases of non-gram-negative bacilli sepsis.PSM was applied to the confounding factors in both groups,resulting in 29 successfully matched pairs.Nineteen clinical indicators were selected and modeled using SVM and LR,with SVM identifying 4 risk factors and LR identifying 3 risk factors.The SVM model demonstrated superiority over the LR model in terms of sensitivity,specificity,positive predictive value,negative predictive value,and AUC.Post-NRI calculation,the predictive ability of the SVM model was significantly higher than that of the LR model(P＜0.05).Conclusion:The SVM model exhibits greater practical value than the LR model in the classification prediction of small sample sizes.White blood cell count,duration of central venous catheterization,pulse pressure,and C-reatative protein can be considered as viable risk factors for the early identification of late-onset sepsis caused by gram-negative bacteria in preterm infants.
support vector machineLogistic regressiongram-negative bacterialate-onset sepsispropensity score matchingpreterm infants
万方数据
