Classifying patients suspected of appendicitis with regard to likelihood

Abstract

Study objectives: We classify patients with regard to likelihood of acute appendicitis using clinical elements in combination with laboratory findings. Methods: This was a prospective observational study enrolling all patients with acute right lower quadrant pain or tenderness at an urban emergency department. Patients were excluded for recent trauma or previous appendectomy. Appendicitis was determined by surgical pathology, and all patients were followed up at 7 days for resolution of symptoms. Elements of the medical history, physical examination, and laboratory investigation were incorporated into a random partition model. The random partition model was used to develop and cross-validate the appendicitis likelihood model (ALM), which identified patients with either a high or low likelihood of appendicitis. The ALM was evaluated assuming that all patients with high likelihood would have an immediate operation, low likelihood would receive outpatient observation, and all others would have advanced imaging. Results: A total of 481 patients presented with right lower quadrant pain or tenderness during the study period; 439 (91%) were enrolled, with follow-up available for 408 (93%). There was a 23% prevalence of acute appendicitis, 83% of all patients received advanced imaging, and 30% had an operation, with a negative laparotomy rate of 14%. The ALM identified patients at high likelihood of appendicitis if they had a WBC count of 13 cells/mL or greater and either rebound tenderness or voluntary guarding with neutrophil count of 82% or greater. Patients were low likelihood if they had a WBC count less than or equal to 9.5 cells/mL and either no right lower quadrant tenderness or a neutrophil count of 54% or less. The ALM as a clinical decision rule predicted a 4% negative laparotomy rate and 1 missed case of appendicitis (appendix in inguinal hernia sac without leukocytosis), with advanced imaging used in 63%. Conclusion: Random partition modeling is useful to identify variables in a hierarchical manner that can be used to classify patients with regard to appendicitis likelihood. Classifying patients with the ALM would have decreased the total number of imaging studies obtained while lowering the diagnostic error rate.