Aggregating opinions into a Bayesian network

Bayesian networks (BN) are widely used in the healthcare domain, but building a BN structure requires intensive work.

We propose a method to build a structure automatically based on expert opinions extracted from the medical literature. Expert opinions are statements of conditional dependence/independence between variables of interest, such a

Not only can contradictory statements often be found, but dependence information is usually incomplete for a given group of variables.

Our method combines the PC algorithm with a Bayesian updating model to resolve these issues. We designed a tool based on this method and applied it to breast cancer risk factors extracted from PubMed.


Related papers

C. Jochim, B. Sacaleanu, L.A. Deleris
Risk Event and Probability Extraction for Modeling Medical Risks
2014 AAAI Fall Symposium on Natural Language Access to Big Data.

L. Deleris, S. Deparis, B. Sacaleanu, L. Tounsi
Risk information extraction and aggregation
in Algorithmic Decision Theory, Patrice Perny, Marc Pirlot, and Alexis Tsoukiàs (Eds),
Lecture Notes in Computer Science, volume 8176, 154-166. Springer.


Example of results

For breast cancer, body mass index represents a major modifiable risk factor; about half of all cases in postmenopausal women are attributable to overweight or obesity.

In all areas studied, a striking relation between age at first birth and breast cancer risk was observed.

Body mass index was not demonstrated to be an independent risk factor for endometrial cancer in this study.

Breast cancer