Survivability modelling using Bayesian network for patients with first and secondary primary cancers.

Area of Science:

• Oncology
• Biostatistics
• Machine Learning

Background:

• Multiple primary cancers pose a significant threat to patient survival.
• Predicting survivability in patients with two cancers is complex due to numerous influencing variables.

Purpose of the Study:

• To develop a Bayesian network (BN) model for predicting the five-year survivability of patients with two primary cancers.
• To evaluate the performance of the proposed BN model against established benchmark approaches.

Main Methods:

• A nationwide database of 7,845 patients with two primary cancers in Taiwan was analyzed.
• A Bayesian network (BN) model was constructed to describe cancer occurrence and predict survivability.
• The synthetic minority oversampling technique was employed to address dataset imbalance.

Main Results:

• The proposed BN model significantly outperformed back-propagation neural networks, logistic regression, support vector machines, and naïve Bayes.
• Sensitivity, a key metric for the non-survival group, was notably improved by the BN model.

Conclusions:

• The BN model enables estimation of posterior probabilities for survivability based on prior evidence.
• The model's predictions on survivability, treatment effects, and socio-demographic factors can inform cancer treatment assessment and monitoring.