Ontological Representation of Causal Relations for a Deep Understanding of Associations Between Variables in Epidemiology

Understanding statistical results is crucial in order to spread right conclusions. In observational studies, statistical results are often reported as associations without going further. However, each association comes from causal relations. Causal diagrams are visual representations enabling to understand causal mechanisms behind the association found. In the era of big data and growing number of variables, visual approaches become inefficient. Ontological representation of causality and reasoning could help to explain statistical results. OntoBioStat is a domain ontology related to covariate selection and bias for biostatistician users. It was designed using expert corpus from comprehensive literature review, and validated by three biostatisticians accustomed to causal diagrams. In this paper, we focused on the presentation of an OntoBioStat's feature able to infer explanations about statistical associations. The ontologization of the feature of interest resulted in 14 object properties, three classes and five Semantic Web Rule Language rules. Each rule allows to infer a different object-property that explains statistical association between two variables. Rules are based on is Cause of statements between different individuals. OntoBioStat feature performances were illustrated through a real-life retrospective observational study. From 28 instances and 48 object properties stated, a set of 1,939 object properties were inferred. OntoBioStat explained 65% of the 48 statistical associations found. In conclusion, OntoBioStat could help to explain a part of the significant statistical associations between two variables but cannot yet predict significant ones.