Analysis of prognostic factors for survival after surgery for gallbladder cancer based on a Bayesian network

Zhi-Qiang Cai,Chen Chen,Zhi-Min Geng,Long-Long Cong,Shu-Bin Si,Peng Guo

doi:10.1038/s41598-017-00491-3

Abstract

The factors underlying prognosis for gallbladder cancer (GBC) remain unclear. This study combines the Bayesian network (BN) with importance measures to identify the key factors that influence GBC patient survival time. A dataset of 366 patients who underwent surgical treatment for GBC was employed to establish and test a BN model using BayesiaLab software. A tree-augmented naïve Bayes method was also used to mine relationships between factors. Composite importance measures were applied to rank the influence of factors on survival time. The accuracy of BN model was 81.15%. For patients with long survival time (>6 months), the true-positive rate of the model was 77.78% and the false-positive rate was 15.25%. According to the built BN model, the sex, age, and pathological type were independent factors for survival of GBC patients. The N stage, liver infiltration, T stage, M stage, and surgical type were dependent variables for survival time prediction. Surgical type and TNM stages were identified as the most significant factors for the prognosis of GBC based on the analysis results of importance measures.

Highlights

Gallbladder cancer (GBC) is the most common malignant tumour of the biliary tract worldwide[1]
Wang et al.[5] developed a nomogram based on a web browser using a parametric survival model from the Surveillance, Epidemiology and End Results-Medicare database to predict which gallbladder patients may benefit from adjuvant chemoradiotherapy
The Bayesian network (BN) model was used to predict patient survival time using data gathered from patients treated at the First Affiliated Hospital of Xi’an Jiaotong University in China

Summary

Introduction

Gallbladder cancer (GBC) is the most common malignant tumour of the biliary tract worldwide[1]. Data-based statistical methods have been extensively applied to the analysis of prognostic factors for GBC patient survival[3, 4]. These studies have examined prognostic factors such as T stage, patient age, surgical type, and recurrence using statistical analyses of clinical data. These studies describe the separate impacts of single factors associated with prognosis and have neglected the joint influence of multiple factors. Some methods of data mining have been developed and applied to survival prediction for patients with GBC, most methods cannot represent variables under uncertainty and ignore the cause-and-effect relationships between prognostic factors. The BN model, which was built based on practical medical data, could provide efficient individual prognosis and optimal treatment by considering regional health care conditions

Methods

Results

Conclusion