Preoperative Analysis for Clinical Features of Unsuspected Gallbladder Cancer Based on Random Forest

With the incidence of unsuspected gallbladder cancer (UGC) increasing, the difference of preoperative features between unsuspected gallbladder cancer and gallbladder cancer diagnosed before operation arose doctors' attention recently. In this study, firstly, chi square test was adopted to analyze the difference of two groups and select out the difference variables. Then, the random forest was proposed to establish the classification model whose accuracy evaluated by area under curve was 0.7310. Meanwhile, the model identified the critical classification factors using variable importance, which adopted the method of mean decrease in accuracy. Finally, the results of two methods showed that clinical features of biliary calculi, cholecystolithiasis history, gallbladder polyps and family history of malignancy, serum CEA level, jaundice, cholecystitis history and abdominal pain were important factors in preoperative assessment of unsuspected gallbladder cancer. What's more, the feature of time of cholecystitis history should not be ignored for preoperative assessment of UGC patients.