Value of fecal calprotectin in prediction of acute appendicitis based on a proposed model of machine learning.

Abstract

The aim of this study is to apply random forest (RF), one of the machine learning (ML) algorithms, to a dataset consisting of patients with a presumed diagnosis of acute appendicitis (AAp) and to reveal the most important factors associated with the diagnosis of AAp based on the variable importance. An open-access dataset comparing two patient groups with (n=40) and without (n=44) AAp to predict biomarkers for AAp was used for this case-control study. RF was used for modeling the data set. The data were divided into two training and test dataset (80: 20). Accuracy, balanced accuracy (BC), sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) performance metrics were appraised for model performance. Accuracy, BC, sensitivity, specificity, PPV, NPV, and F1 scores pertaining to the RF model were 93.8%, 93.8%, 87.5%, 100%, 100%, 88.9%, and 93.3%, respectively. Following the variable importance values regarding the model, the variables most associated with the diagnosis and prediction of AAp were fecal calprotectin (100 %), radiological imaging (89.9%), white blood test (51.8%), C-reactive protein (47.1%), from symptoms onset to the hospital visit (19.3%), patients age (18.4%), alanine aminotransferase levels >40 (<1%), fever (<1%), and nausea/vomiting (<1%), respectively. A prediction model was developed for AAp with the ML method in this study. Thanks to this model, biomarkers that predict AAp with high accuracy were determined. Thus, the decision-making process of clinicians for diagnosing AAp will be facilitated, and the risks of perforation and unnecessary operations will be minimized thanks to the timely diagnosis with high accuracy.