Analysis Of Influencing Factors And Establishment Of Risk Prediction Models Of Postoperative Infections After Open Liver Surgery

Background&Aims Liver surgery is one of the most important surgery. With significant surgical difficulty, liver surgery shows a high incidence rate of postoperative complications. One common complication is postoperative infection. Postoperative infections are associated with prognosis, such as postoperative hospital stay, mortality, and survival. It is vital in clinical practice to reduce the occurrence of postoperative infections. No efficient strategy has shown an efficient effect on management of morbidity and mortality caused by postoperative infections. One possible solution might be ascertaining risk factors for postoperative infections and focusing on high risk populations. We aim to apply data mining methods to analysis of factors associated with postoperative infections after open liver surgery and to establish prediction models for early discovery of patients with high risks to develop postoperative infections.Methods Patients who underwent open liver surgery at the Department of Hepato-biliary-pancreatic Surgery (Surgery V), the Second Affiliated Hospital, Zhejiang University School of Medicine between November2012and October2013were identified. Data on general characteristics, past history, perioperative details, as well as postoperative complications were collected and analyzed. Diagnosis of postoperative infections were independently made by two physicians, based on the Criteria of Hospital Infection published by Ministry of Health of the People’s Republic of China in2001and clinical manifestation, and were then confirmed by the attending physicians. Logistic regression and decision tree classification were performed to analyze the factors significantly associated with postoperative infections. Prediction models were established, with prediction accuracy, sensitivity, specificity, and receiver operating characteristic (ROC) curve evaluated. Areas under curve (AUC) concerning the ROC curves of the two prediction models were compared.Results A total of251cases were included in the analysis. The incidence rate of postoperative complications was24.70%(62cases). Univariate logistic regression analysis revealed the factors associated with postoperative infections, namely ages, gender, hypertension history, past abdominal surgery history, preoperative infections, operation related to bile duct, operation related to gastrointestinal tract, ASA class P2, operation time, estimated blood loss, blood transfusion, postoperative diagnosis of liver cancer, and delayed extubation. Further multivariate logistic regression included age, gender, operation time, and delayed extubation in the regression prediction model. Decision tree classification revealed8independent factors, namely operation time, preoperative infections, blood transfusion, albumin level, estimated blood loss, operation related to bile duct, gender, and preoperative leucocyte/CRP abnormality. The accuracy, sensitivity, specificity, and AUC of logistic regression model and decision tree classification model were79.84%,40.32%,93.01%,0.793and82.47%,67.74%,87.30%,0.866, respectively. Conclusions Incidence rate of postoperative infections after open liver surgery remained considerable. Several factors were significantly associated with postoperative infections. The prediction models based on the associating factors, established by data mining methods could predict the incidence of postoperative infections. The model developed with decision tree classification showed a better prediction performance comparing to logistic regression model, and could be used for early prediction of postoperative infections in clinical practice.