Statistical Learning Approaches In Diagnosing Patients With Nontraumatic Acute Abdomen

Abstract

A quick evaluation is required for patients with acute abdominal pain. It is crucial to differentiate between surgical and nonsurgical pathology. Practical and accurate tests are essential in this differentiation. Lately, D-dimer level has been found to be an important adjuvant in this diagnosis and obviously outperforms leukocyte count, which is widely used for diagnosis of certain cases. Here, we handle this problem from a statistical perspective and combine the information from leukocyte count with D-dimer level to increase the diagnostic accuracy of nontraumatic acute abdomen. For this purpose, various statistical learning algorithms are considered and model performances are assessed using several measures. Our results revealed that the naive Bayes algorithm, robust quadratic discriminant analysis, bagged and boosted support vector machines, and single and bagged k-nearest neighbors provide an increase in diagnostic accuracies of up to 8.93% and 17.86% compared with D-dimer level and leukocyte count, respectively. Highest accuracy was obtained as 78.57% with the naive Bayes algorithm. Analysis has been done via the R programming language based on the codes developed by the authors. A user-friendly web-tool is also developed to assist physicians in their decisions to differentially diagnose patients with acute abdomen.