Visual Analysis of Multi-Outcome Causal Graphs

Area of Science:

• Data Visualization
• Causal Inference
• Health Informatics

Background:

• Understanding multimorbidity and comorbidity is vital in healthcare.
• Existing methods for causal graph analysis often focus on single outcomes.
• There is a need for visual tools to analyze multiple causal relationships simultaneously.

Purpose of the Study:

• To introduce a visual analysis method for multi-outcome causal graphs.
• To develop comparative visualization techniques for analyzing differences and commonalities in causal graphs.
• To support healthcare research in understanding complex health conditions.

Main Methods:

• Developed a progressive visualization method for comparing causal discovery algorithms on mixed-type datasets.
• Devised a comparative graph layout technique and specialized visual encodings for multi-outcome causal graphs.
• Integrated these techniques into a visual analysis workflow starting with individual outcome graphs.

Main Results:

• The progressive visualization method effectively handles mixed-type data for single outcome causal graph creation.
• The comparative visualization techniques enable quick comparison of multiple causal graphs.
• Evaluations included quantitative measurements, a medical expert case study, and user studies.

Conclusions:

• The proposed visual analysis method enhances the understanding of multi-outcome causal graphs.
• This approach aids in identifying shared and distinct causal relationships across different health outcomes.
• The developed techniques are valuable tools for health research involving complex comorbidities.