Purpose: The diagnosis of symptom-based indications such as acute abdominal pain is elaborate, and decision-making at the beginning of the episode for immediate treatment, diagnostic imaging, or observation is needed. As individual tests only contribute marginally to the probabilities of the diseases, a scoring of tests would enhance the post-test probabilities. By putting appropriate weights on each variable, thresholds of the sum score can be used for decision-making. The contributions of these variables to the post-test probability of appendicitis were assessed in a Bayesian network to evaluate of the commonly applied sum score method. Methods: The test characteristics of 25 clinical and laboratory variables associated with appendicitis and 19 different appendicitis scorings were obtained by meta-analyses. Variables were selected if providing diagnostic information, that is a sensitivity or specificity > 0,5, and a diagnostic odds ratio dOR > 2. The resulting score contained 11 variables, all with rounded ln(dOR) weights of 1 or 2. The association between the variables and appendicitis were modelled in a Bayesian network using the values of the sensitivity and specificity as conditional probabilities for each variable. The network calculates the post-test probability of appendicitis given the variable values. The formulas of the associations between the post-test probability and the sensitivities and specificities were also determined. Results: Different scorings were evaluated as well as different practical combinations of variables that are available at the same time, for instance laboratory results. Based on these variables, the post-test probabilities were determined. The thresholds for the post-test probability should be sufficiently high to prevent negative appendectomies, and sufficiently low to exclude appendicitis, for instance > 95%, and < 5%, respectively. The highest attainable probability in the evaluation was surprisingly as high as 99%. Conclusions: By evaluation of the post-test probability in a Bayesian network, the contribution of each variable can be estimated. The availability of several variables at a time can also be taken into account. Once the scoring is determined, these variables can be replaced by a single scoring in an elegant and clinically relevant way.

This work was financially supported by the EC-EURATOM 6 Framework Programme (2002-2006) and forms part of the CT Safety & Efficacy (Safety and Efficacy of Computed Tomography (CT): A broad perspective) project, contract FP/002388.