Predictive model of biliocystic communication in liver hydatid cysts using classification and regression tree analysis

Background. Incidence of liver hydatid cyst (LHC) rupture ranged 15%-40% of all cases and most of them concern the bile duct tree. Patients with biliocystic communication (BCC) had specific clinic and therapeutic aspect. The purpose of this study was to determine witch patients with LHC may develop BCC using classification and regression tree (CART) analysis. Methods. A retrospective study of 672 patients with liver hydatid cyst treated at the surgery department "A" at Ibn Sina University Hospital, Rabat Morocco. Four-teen risk factors for BCC occurrence were entered into CART analysis to build an algorithm that can predict at the best way the occurrence of BCC. Results. Incidence of BCC was 24.5%. Subgroups with high risk were patients with jaundice and thick pericyst risk at 73.2% and patients with thick pericyst, with no jaundice 36.5 years and younger with no past history of LHC risk at 40.5%. Our developed CART model has sensitivity at 39.6%, specificity at 93.3%, positive predictive value at 65.6%, a negative predictive value at 82.6% and accuracy of good classification at 80.1%. Discriminating ability of the model was good 82%. Conclusion. we developed a simple classification tool to identify LHC patients with high risk BCC during a routine clinic visit (only on clinical history and examination followed by an ultrasonography). Predictive factors were based on pericyst aspect, jaundice, age, past history of liver hydatidosis and morphological Gharbi cyst aspect. We think that this classification can be useful with efficacy to direct patients at appropriated medical struct's. © 2010 El Malki et al; licensee BioMed Central Ltd.