Federated causal discovery with missing data in a multicentric study on endometrial cancer

Area of Science:

• Causal inference
• Machine learning
• Biostatistics

Background:

• Causal discovery is vital for explainable decision-making in medicine and healthcare.
• Small sample sizes and missing data present significant challenges in real-world applications.
• Federated approaches are needed to leverage distributed data effectively.

Purpose of the Study:

• To propose a novel federated causal discovery algorithm.
• To address challenges of heterogeneous missing data across multiple sources.
• To learn accurate causal graphs for improved explainability.

Main Methods:

• Developed a federated algorithm for causal graph learning on a centralized server.
• Incorporated prior knowledge and client-specific missingness mechanisms.
• Applied the algorithm to synthetic and real-world multicentric endometrial cancer data.

Main Results:

• Successfully learned a causal graph representing cause-effect relationships.
• Validated the model using quantitative analyses and clinical literature review.
• Demonstrated the algorithm's effectiveness on both synthetic and real-world datasets.

Conclusions:

• The proposed federated approach accurately learns causal models.
• The method is robust even when data is missing not-at-random.
• Enables reliable causal inference in data-scarce, heterogeneous environments.