Improving Causal Bayesian Networks using Expertise in Authoritative Medical Ontologies

Discovering causal relationships among symptoms is a topical issue in the analysis of observational patient datasets. A Causal Bayesian Network (CBN) popular analytical framework for inference. While there are many methods and algorithms capable learning network, they reliant on complexity thoroughness algorithm do not consider prior expertise from authoritative sources. This paper proposes novel method extracting knowledge contained Authoritative Medical Ontologies (AMOs) using this to orient arcs CBN learned data. Since AMOs robust biomedical ontologies containing collective experts who created them, utilizing ordering information within them produces improved CBNs which provide additional insight into disease domain. To demonstrate our method, we obtained three AMOs: 1) Dictionary Regulatory Activities Terminology (MedDRA), 2) International Classification Diseases Version 10 Clinical Modification (ICD-10-CM), 3) Systematized Nomenclature Medicine Terms (SNOMED CT). The ontological these then used series National Institutes Mental Health study Sequenced Treatment Alternatives Relieve Depression (STAR*D) dataset Max-Min Hill-Climbing (MMHC) algorithm. Six distinct generated MMHC: an unmodified baseline model only algorithm, oriented with ordered-variable pairs MedDRA, ICD-10-CM, SNOMED CT, two more ordered combination AMOs. resulting modified significantly change structure network. agreement between Modified networks Baseline ranges 50% 90%. network all (10 out 20 exist both models) while maintaining comparable predictive accuracy. indicates that reflects claims agrees STAR*D dataset. Furthermore, models discovered new potentially model, eliminating weaker edges qualitative significance existing epidemiological research.