A Causal Discovery Workflow for Rare Diseases: Experts-in-the-Loop Analysis of Sparse Longitudinal Data

Area of Science:

• Computational biology
• Medical informatics
• Causal inference

Background:

• Causal networks offer mechanistic insights for personalized medicine.
• Causal discovery is difficult for rare diseases due to sparse data and incomplete knowledge.
• Disease progression over time adds complexity to causal modeling.

Purpose of the Study:

• To develop an expert-in-the-loop causal discovery workflow.
• To iteratively refine causal networks representing disease mechanisms.
• To create a comprehensive causal model for rare diseases like soft tissue sarcoma.

Main Methods:

• An expert-in-the-loop causal discovery workflow was proposed.
• The workflow iteratively refines causal networks.
• Applied to soft tissue sarcoma to model its natural history.

Main Results:

• The workflow generated three causal networks for soft tissue sarcoma.
• These networks describe the interplay between patient covariates and disease behavior.
• This represents the first comprehensive causal description of the disease's natural history.

Conclusions:

• The proposed workflow enhances causal discovery for rare, complex diseases.
• It enables personalized treatment strategies through improved clinical decision-making.
• The method is agile, modular, and flexible for data-sparse, longitudinal clinical domains.