| Objectives:1.To comprehensively and systematically investigate the current status of postoperative nosocomial infection in colon cancer patients.2.The models for predicting the risk of postoperative nosocomial infection in colon cancer patients,including nomogram,BP neural network and decision tree,were constructed to assist healthcare professionals in the early identification of those at high risk of postoperative nosocomial infection.3.To compare the predictive efficacy of the three prediction models and provide a reference for practical clinical application.Methods:This study collected data on patients who underwent colon cancer surgery in a hospital in Shandong Province from January 1,2020,to December 31,2022.This study relied on the self-designed "Targeted Surveillance Questionnaire for Nosocomial Infection in Patients with Colon Cancer Surgery",using the hospital infection monitoring system and hospital information system to collect data.Patient data from 2020 to 2021 were used as the training set to build the models,and data from 2022 were used as the validation set to validate the models.The presentation of the nomogram includes both the static graphic and dynamic web page.Brier score was used to evaluate the overall performance of the nomogram and AUC was used to evaluate the discrimination of the nomogram.The calibration curve and Hosmer-Lemeshow test were applied to evaluate the calibration of the nomogram.DCA and CIC were used to evaluate the clinical validity of the nomogram.The predictive performances of the BP neural network model and decision tree model were evaluated by AUC and error classification table.The DeLong test was used in Medcalc software to compare the AUC of the three prediction models.Results:1.A total of 1146 colon cancer patients who underwent surgery were included in this study,with an average age of 59.79 ± 12.31 years.The proportion of males and females was 61.78%(708)and 38.22%(438),respectively.The number of patients in the training set was 762,with an average age of 60.11±12.51 years.The number of patients in the validation set was 384,with an average age of 59.16±11.89 years.2.In this study,a total of 110 patients with colon cancer developed postoperative nosocomial infection,with an infection rate of 9.60%and a total of 141 infections,with an infection rate of 12.30%.Throughout the year,colon cancer patients had higher levels of postoperative nosocomial infection in July and August.The nosocomial infection rates were 9.71%and 9.38%for the training set and validation set,respectively.3.Among the 110 patients with postoperative nosocomial infection for colon cancer,lower respiratory tract infection and surgical site infection predominated in 38(34.55%)and 34(30.91%)cases,respectively.Other sites of nosocomial infection included abdominal infection in 4 cases(3.64%),urinary tract infection in 4 cases(3.64%),ascites infection in 2 cases(1.82%)and bloodstream infection in 1 case(0.91%).Additionally,there were 27 cases(24.55%)of multiple sites of infection.4.Among the patients with postoperative nosocomial infection for colon cancer,225 specimens were sent for examination and 82 pathogenic strains were cultured,including 55(67.07%)Gram-negative and 27(32.93%)Gram-positive strains.The highest number of Gram-negative organisms detected was Escherichia coli,with 29 cases and an overall composition of 35.37%.The highest number of Gram-positive organisms detected were Enterococcus faecalis and Enterococcus faecium,with 8(9.76%)and 4(4.88%)cases respectively.5.The best subset obtained by LASSO regression analysis included a total of nine variables:surgical modality,postoperative complication,peak temperature on the second postoperative day,postoperative Braden score,duration of abdominal drainage tube retention,arteriovenous catheterization.the application of ventilator,ASA classification and enterostomy.6.The results of multivariate logistic regression showed that postoperative complication,peak temperature on the second postoperative day,low post-operative Braden score,surgical modality(open and laparoscopic intermediate open),duration of abdominal drainage tube retention(≥10 days),ASA classification(≥Class Ⅲ)and enterostomy were independent high-risk factors of postoperative nosocomial infection of patients with colon cancer.The Brier scores of the nomogram model constructed on this basis were 0.046 and 0.050 in the training and validation sets respectively,indicating that the overall performance of the model is good.The AUC in the training and validation sets were 0.881 and 0.813 respectively,and the P-values for the Hosmer-Lemeshow test results in the training and validation sets were 0.490 and 0.179 respectively,indicating that the nomogram has good discrimination and calibration.DCA and CIC showed the nomogram has good clinical effectiveness.7.In this study,the importance of variables of the BP neural network prediction model from high to low was as follows:postoperative Braden score,peak temperature on the second postoperative day,postoperative complication,surgical modality,duration of abdominal drainage tube retention,ASA classification,arteriovenous catheterization,the application of ventilator and enterostomy.The prediction accuracy of the model in the training set and the validation set were 94.62%and 93.75%,respectively,and the AUC was 0.865 and 0.849,respectively,suggesting that the model has a good prediction effect.8.The root node of the decision tree prediction model constructed in this study was the application of a ventilator.The model lobe nodes included a total of six variables:peak temperature on the second postoperative day,postoperative complication,postoperative Braden score,duration of abdominal drainage tube retention,ASA classification,and surgical modality.The prediction accuracy of the model was 91.86%and 93.23%in the training and validation sets,with AUC of 0.870 and 0.808 respectively,indicating good prediction performance of the model.9.In this study,there were no statistical differences between the AUC of the nomogram,BP neural network and decision tree models.Conclusions:High levels of postoperative nosocomial infection occur in patients with colon cancer.Postoperative complication,peak temperature on the second postoperative day,surgical modality,duration of abdominal drainage tube retention,ASA classification and postoperative Braden score are all closely associated with the occurrence of postoperative nosocomial infection in patients with colon cancer and are important breakthrough points for medical staff to carry out nosocomial infection prevention and control initiatives.Nomogram,BP neural network and decision tree models are all effective in predicting the risk of postoperative nosocomial infection in patients with colon cancer,and each has its advantages,among which the dynamic web calculator of nomogram is the most convenient.Healthcare professionals can select a model that is more practical to apply or combine models according to the actual clinical situation to achieve improved screening of patients with colon cancer at high risk of postoperative infection. |
PDF Full Text Request