Data Integration in Causal Inference.

Area of Science:

• Biostatistics
• Epidemiology
• Data Science

Background:

• Integrating data from multiple sources is crucial for large sample sizes and diverse populations.
• Heterogeneous data sources present unique challenges for causal inference.
• Existing methods often struggle to combine data from different designs and populations.

Purpose of the Study:

• To review recent developments in causal inference methods for integrating multiple heterogeneous datasets.
• To summarize advances in combining randomized clinical trials with external data.
• To explore methods for distributed data settings and causal discovery.

Main Methods:

• Review of recent literature on causal inference techniques.
• Summarization of methods for combining data from randomized clinical trials (RCTs) and observational studies.
• Discussion of two-sample Mendelian randomization, distributed data analysis, and Bayesian causal inference.

Main Results:

• Advances in combining RCTs with observational studies and historical controls.
• Methods for sample merging when variables are distributed across datasets (e.g., two-sample Mendelian randomization).
• Development of privacy-preserving distributed data methods for real-world data analysis.

Conclusions:

• Causal inference methods are evolving to effectively integrate heterogeneous data sources.
• New approaches facilitate robust analysis in distributed settings and advance causal discovery.
• These methods enhance comparative effectiveness and safety research using real-world data.