Objective causal predictions from observational data.

Area of Science:

• Public Health Risk Assessment
• Causal Inference
• Observational Data Analysis

Background:

• Current public health risk assessments often rely on untestable assumptions for causal conclusions from observational data.
• The subjective approach, based on potential outcomes models, lacks independent verifiability and independent verification.
• This can limit the benefits of objective science, hindering scrutiny and independent validation.

Purpose of the Study:

• To introduce an objective, data-driven approach for causal analysis of exposure-response relationships in observational data.
• To replace untestable potential outcomes models with empirically testable interventional causal models.
• To improve the reliability and transparency of causal inferences in health risk assessments.

Main Methods:

• Utilizing causal Bayesian networks (CBNs) as an alternative to potential outcomes models.
• Employing Invariant Causal Prediction (ICP) tests for empirical validation of causal claims across studies.
• Using individual conditional expectation (ICE) plots to quantify health risks and exposure effects.

Main Results:

• The proposed objective approach is independently verifiable and data-driven, avoiding inherently untestable assumptions.
• CBNs and ICP tests allow for empirical validation of causal claims.
• The framework can handle complexities like confounding, missing data, and measurement error.

Conclusions:

• The objective approach offers a more reliable and transparent method for causal inference in health risk assessment.
• Explicit and empirically testable causal assumptions enhance the robustness of findings.
• This framework supports evidence-based decision-making in public health by providing verifiable causal insights.