Data collaboration for causal inference from limited medical testing and medication data.

Area of Science:

• Health Informatics
• Biostatistics
• Epidemiology

Background:

• Randomized controlled trials (RCTs) are not always feasible for causal inference.
• Integrating sensitive medical data across institutions poses significant privacy challenges.
• Existing methods often require sharing raw patient data, limiting collaborative research.

Purpose of the Study:

• To apply the data collaboration quasi-experiment (DC-QE) framework to medical data.
• To simulate distributed data environments under independent and identically distributed (IID) and non-IID conditions.
• To evaluate the efficacy of DC-QE for privacy-preserving causal inference in healthcare.

Main Methods:

• Developed a method for generating intermediate representations within the DC-QE framework.
• Applied DC-QE to single-institution medical data to simulate distributed settings.
• Compared DC-QE performance against individual and centralized analyses under IID and non-IID conditions.

Main Results:

• DC-QE consistently outperformed individual analyses across accuracy metrics.
• DC-QE performance closely approximated centralized analysis.
• The proposed method for intermediate representations improved performance, especially under non-IID conditions.

Conclusions:

• The DC-QE framework offers a robust method for privacy-preserving causal inference in healthcare.
• Utilizing intermediate representations facilitates access to larger, diverse datasets while maintaining patient confidentiality.
• This approach can advance the discovery of causal relationships, support drug repurposing, and improve rare disease therapies.