Do-search: A Tool for Causal Inference and Study Design with Multiple Data Sources.

Area of Science:

• Causal inference
• Epidemiology
• Biostatistics

Background:

• Epidemiologic evidence relies on diverse data sources.
• Integrating data from multiple sources enhances causal effect estimation.
• Existing methods face challenges with missing data and selection bias.

Purpose of the Study:

• To present a novel algorithmic approach, do-search, for determining causal effect identifiability.
• To demonstrate combining experimental and observational data for causal inference.
• To illustrate the application of do-search using real-world data.

Main Methods:

• Utilizing do-calculus principles for causal effect identification.
• Employing the do-search algorithm to assess identifiability with missing data and selection bias.
• Integrating data from the National Health and Nutrition Examination Survey (NHANES) and randomized controlled trials (RCTs).

Main Results:

• Do-search can identify causal effects and provide estimation formulas when identifiable.
• The tool can suggest additional data or assumptions to achieve identifiability when effects are not initially identifiable.
• The study estimated the reduction in systolic blood pressure from discontinuing table salt use.

Conclusions:

• Combining diverse data sources and employing advanced algorithms like do-search improves causal inference.
• Do-search offers a robust method for causal effect estimation, handling complex data challenges.
• The approach provides a framework for resolving identifiability issues in epidemiologic research.