Analysis of risk factors of hip fracture with causal Bayesian networks

Abstract : We explore a practical approach to learn a plausible causal Bayesian network from a combination of non-experimental data and qualitative assumptions that are deemed likely by health experts. The method is based on the incorporation of prior expert knowledge in the form of partial pairwise ordering constraints between variables into a recent constraint-based Bayesian network structure learning algorithm. The learning process ends up with a partially oriented graph. The remaining undirected edges are then oriented according to the expert understanding. We show that the causal graph not only provides a statistical profile of the population under study but also offers a simple guideline principle to identify accessible sets of confounding variables for each causal relation under interest. To illustrate the potential of the proposed approach, we estimate the strength of the causal effect of psychotropic drugs, gait speed, body mass index and bone mineral density on the risk of hip fracture from a prospective cohort study EPIDOS sample, which included more than 7500 elderly osteoporotic women followed-up during 4 years. Our findings suggest that an intervention programme aimed at preventing physical deterioration and maintaining bone mass density should tend to reduce the risk of hip fracture among elderly.
Document type :
Conference papers
Complete list of metadatas

https://hal.archives-ouvertes.fr/hal-01326485
Contributor : Équipe Gestionnaire Des Publications Si Liris <>
Submitted on : Friday, June 3, 2016 - 4:11:42 PM
Last modification on : Friday, November 22, 2019 - 1:54:01 PM

Identifiers

  • HAL Id : hal-01326485, version 1

Citation

Alex Aussem, Pascal Caillet, Sarah Klemm, Maxime Gasse, Anne-Marie Schott, et al.. Analysis of risk factors of hip fracture with causal Bayesian networks. International Work-Conference on Bioinformatics and Biomedical Engineering, Apr 2014, Grenade, Spain. pp.1074-1085. ⟨hal-01326485⟩

Share

Metrics

Record views

203