A Conditional Probability Model to Predict the Mortality in Patients With Breast Cancer: A Bayesian Network Analysis.

Background: The aim of this study was to compute the event rate of patients with breast cancer (BC) using Bayesian network (BN) structure.

Method: Data for 1,154 patients newly diagnosed with BC were recruited in this study during 2007 and 2016 in Iran. The database was linked to the regional death registration system and active follow-up was performed by referring to hospital information system or calling the patients. BN structure with inverse probability of censoring weighting (IPCW) approach was used to assess the relationship between event rate and underlying risk factors.

Results: The median (25th, 75th percentiles) of patients' survival time was 46.8 (32.6, 69.3) months. There were 217 (18.8%) deaths from BC by the end of the study. The optimal BN structure (Akaike Information Criteria = -8743.66 and Bayesian Information Criteria = -8790.80) indicated that being male (conditional probability [CP] = 0.316), age >50 (CP = 0.215), higher grades (CP = 0.301) and lower survival times (CP = 0.566) had higher event rate. Also lobular carcinoma (CP = 0.157) and ductal carcinoma (CP = 0.178) type of morphology had lower event rate while other types (CP = 0.316) had higher.

Conclusions: The BN structure in which time was as a mediator of predictors-event relationship could be presented as the optimal tool to compute the event rate of BC. The findings could be used to identify the high risk patients and recommend for health policy making, prevention and planning for decrease the mortality in patients with BC.