deSSIde: A Clinical Decision-Support Tool for Surgical Site Infection Prediction

Abstract

Surgical Site Infections (SSI) is a complication that is substantially associated with increased morbidity, prolonged hospital stay, mortality, and additional surgical and intensive care procedures for the patients involved. With the availability of data from the SSI surveillance program of the Philippine General Hospital (PGH), a decision-support tool can be created which will show the risk of SSI in a patient. Eight supervised machine learning models were developed and validated, based on decision trees and support vector machines (SVM), and two differently preprocessed versions of a single dataset for purposes of experimentation. A web-based application was developed which serves as the user-facing tool. The user can view the calculated percentage risk of SSI as well as the decision tree path that leads to the result from an input of patient- and operation-related variables. From the machine learning models, the best model is the decision tree model with all variables included with an accuracy of 93%, balanced accuracy of 77.8%, precision of 89.5%, recall of 56.7%, F1 score of 69%, ROC AUC of 77.8%, and MCC of 68%.

Ähnliche Arbeiten