OntoBioStat: Supporting Causal Diagram Design and Analysis.

Area of Science:

• Biostatistics
• Bioinformatics
• Ontology Engineering

Background:

• Causal inference in biostatistics relies on causal diagrams like directed acyclic graphs.
• Existing ontologies lack robust representation for necessary and sufficient causes, hindering covariate selection.
• There is a need for advanced methods to represent causal relationships for improved statistical analysis.

Purpose of the Study:

• To design and detail OntoBioStat, an ontology-based system for causal relation representation.
• To enable automatic construction and inference of ontological causal diagrams for biostatistical analysis.
• To support covariate selection by leveraging causal relationships and identifying potential biases.

Main Methods:

• OntoBioStat utilizes Semantic Web Rule Language (SWRL) rules and axioms for inference.
• User inputs specify outcomes, exposures, covariates, and causal relations.
• Ontology construction involves generic instances of Meta_Variable and Necessary_Variable classes.
• Inferred classes are used to highlight potential biases, such as confounder-like relationships.

Main Results:

• OntoBioStat enables automatic construction of ontological causal diagrams.
• The system successfully infers potential biases and aids in covariate selection.
• A theoretical comparison showed OntoBioStat identifies confounding and bias more completely than standard causal diagrams.
• Standard methods sometimes failed to identify all confounding factors and provided erroneous covariate sets.

Conclusions:

• OntoBioStat offers a novel approach to causal inference in biostatistics through ontological representation.
• The system demonstrates potential for improving covariate selection and bias identification.
• Further research is recommended to enhance the usability of OntoBioStat for broader application.