Data-Driven and Expert-Informed Causal Discovery for Type 2 Diabetes Risk in Primary Care

Area of Science:

• Computational biology
• Epidemiology
• Medical informatics

Background:

• Routinely collected primary care data offers a rich source for understanding disease etiology.
• Type 2 Diabetes (T2D) poses a significant public health challenge, necessitating improved risk prediction and prevention strategies.

Purpose of the Study:

• To extract causal relationships influencing Type 2 Diabetes risk from extensive Canadian primary care data.
• To construct a causal Directed Acyclic Graph (DAG) integrating data-driven findings with expert medical knowledge.

Main Methods:

• Utilized a large dataset encompassing biomarkers, medical conditions, risk factors, and medications.
• Employed a causal discovery process involving iterative refinement and expert knowledge integration.
• Developed a Directed Acyclic Graph (DAG) to represent causal pathways.

Main Results:

• The generated DAG accurately reflects established medical knowledge regarding T2D risk factors.
• The causal model demonstrated satisfactory performance in predicting future T2D onset.
• Identified key causal links between various health indicators and diabetes risk.

Conclusions:

• The developed causal DAG provides a robust foundation for understanding T2D etiology.
• This approach can support the creation of interpretable tools for clinical decision-making.
• Highlights the potential of integrating routinely collected data with causal inference for medical research.