Establishment and validation of a clinical prediction model for infection risk at the placement sites of skin and soft tissue expanders

Abstract

Objective: To establish a clinical prediction model for infection risk at the placement sites of skin and soft tissue expanders (hereinafter termed as expanders) and to validate the predictive value of the model. Methods: A retrospective observational study was conducted. Totally 2 934 patients who underwent skin and soft tissue dilatation surgery in the Department of Plastic Surgery of the First Affiliated Hospital of Air Force Medical University from January 2009 to December 2018 and met the selection criteria were included. There were 1 867 males and 1 067 females, with a median age of 18 years. Totally 3 053 skin and soft tissue expansion procedures were performed with 4 266 expanders implanted. The following indexes were selected as predictor variables, including patients' age, gender, marital status, ethnicity, hospital admission, surgical indication, disease duration, with/without history of smoking, history of drinking, history of blood transfusion, history of underlying diseases, and inability to use cephalosporin antibiotics due to allergy, number of expander in a single placement, rated volume of expander, water injection rate of expander in the first time, placement site of expander, anesthesia method, duration of operation, and with/without postoperative hematoma evacuation, and infection at the placement site of expander as the outcome variable. Univariate analysis of the data was performed using least absolute shrinkage and selection operator (LASSO) regression to screen the potential risk factors affecting infection at the placement sites of expanders, the factors selected by the univariate analysis were subjected to binary multivariate logistic regression analysis to screen the independent risk factors affecting infection at the placement sites of expanders, and a nomogram prediction model for the occurrence of infection at the placement sites of expanders was established. The C index and Hosmer-Lemeshow goodness of fit test were used to evaluate the discrimination and accuracy of the model, respectively, and the bootstrap resampling was used for internal verification. Results: The results of LASSO regression showed that age, gender, hospital admission, surgical indication, disease duration, history of drinking, history of heart disease, history of viral hepatitis, history of hypertension, inability to use cephalosporin antibiotics due to allergy, number of expander in a single placement, rated volume of expander, placement site of expander, postoperative hematoma evacuation were the potential risk factors for infection at the placement sites of expanders (regression coefficient=-0.005, 0.170, 0.999, 0.054, 0.510, -0.003, 0.395, -0.218, 0.029, 0.848, -0.116, 0.175, 0.085, 0.202). Binary multivariate logistic regression analysis showed that male, emergency admission, disease duration ≤1 year, inability to use cephalosporin antibiotics due to allergy, rated volumes of expanders ≥200 mL and <400 mL or ≥400 mL, and expanders placed in the trunk or the limbs were the independent risks factors for infection at the placement sites of expanders (odds ratio=1.37, 3.21, 2.00, 2.47, 1.70, 1.73, 1.67, 2.16, 95% confidence interval=1.04-1.82, 1.09-8.34, 1.38-2.86, 1.29-4.41, 1.07-2.73, 1.02-2.94, 1.09-2.58, 1.07-4.10, P<0.05 or P<0.01). The C index for evaluating the discriminative degree of the model was 0.63, the Hosmer-Lemeshow goodness of fit test for evaluating the accuracy of the model showed P=0.685, and the C index for internal validation by the bootstrap resampling was 0.60. Conclusions: Male, emergency admission, disease duration ≤1 year, inability to use cephalosporin antibiotics due to allergy, rated volume of expander ≥200 mL, and expanders placed in the trunk or the limbs are the independent risk factors for infection at the placement sites of expanders. The clinical prediction model for infection risk at the placement sites of expanders was successfully established based on these factors and showed a certain predictive effect.