Objectives: The diagnosis of acute appendicitis is difficult, and a diagnostic error will often lead to either a perforation or the removal of a normal appendix. In this study, we constructed a Bayesian network model for the diagnosis of acute appendicitis and compared the diagnostic accuracy with other diagnostic models, such as the naive Bayes model, an artificial neural network model, and a logistic regression model.
Methods: The data from 169 patients, who suffered from acute abdominal pain and who were suspected of having an acute appendicitis, were analyzed in this study. Nine variables were used for the evaluation of the accuracy of the four models for the diagnosis of an acute appendicitis. The naive Bayes model, the Bayesian network model, an artificial neural network model, and a logistic regression model were used in this study for the diagnosis of acute appendicitis. These four models were validated by using the ".632+ bootstrap method" for resampling. The levels of accuracy of the four models for diagnosis were compared by the error rates and by the areas under the receiver operating characteristic curves.
Results: Through the course of illness, 50.9% (86 of 169) of the patients were diagnosed as having an acute appendicitis. The error rate was the lowest in the Bayesian network model, as compared with the other diagnostic models. The area under the receiver operating characteristic curve analysis also showed that the Bayesian network model provided the most reliable results.
Conclusion: The Bayesian network model provided the most accurate results in comparison to other models for the diagnosis of acute appendicitis.