Using Artificial Neural Networks to Predict Intra-Abdominal Abscess Risk Post-Appendectomy.

ABSTRACT

Objective: To determine if artificial neural networks (ANN) could predict the risk of intra-abdominal abscess (IAA) development post-appendectomy. Background: IAA formation occurs in 13.6% to 14.6% of appendicitis cases with "complicated" appendicitis as the most common cause of IAA. There remains inconsistency in describing the severity of appendicitis with variation in treatment with respect to perforated appendicitis. Methods: Two "reproducible" ANN with different architectures were developed on demographic, clinical, and surgical information from a retrospective surgical dataset of 1574 patients less than 19 years old classified as either negative (n = 1,328) or positive (n = 246) for IAA post-appendectomy for appendicitis. Of 34 independent variables initially, 12 variables with the highest influence on the outcome selected for the final dataset for ANN model training and testing. Results: A total of 1574 patients were used for training and test sets (80%/20% split). Model 1 achieved accuracy of 89.84%, sensitivity of 70%, and specificity of 93.61% on the test set. Model 2 achieved accuracy of 84.13%, sensitivity of 81.63%, and specificity of 84.6%. Conclusions: ANN applied to selected variables can accurately predict patients who will have IAA post-appendectomy. Our reproducible and explainable ANNs potentially represent a state-of-the-art method for optimizing post-appendectomy care.