Epidemiology and Causal Inference

Week 8: Causal Inference in Epidemiology

Plan and Learning Outcomes

  • Chapter 14:

    • Understand approaches for studying causation

    • Differentiate between real and spurious associations

    • Define "necessary" and "sufficient" in causal relationships

    • Discuss the Bradford Hill criteria

  • Chapter 15:

    • Review of selection bias and information bias

    • Discuss confounding and methods to address it

    • Define interaction and assess interaction effects on disease risk

Approaches for Determining Causality

  • Check if an association is observed between an exposure (environmental or host characteristic) and a disease/health outcome.

  • Determine if the observed association is causal.

Real vs Spurious Associations

  • Real Associations: May have causal relationships or may be due to confounding factors.

    • Example: Increased coffee drinking and pancreatic cancer could be confounded by smoking.

  • Spurious Associations: Caused by biases, not true relations.

Understanding Causal Relationships

  • Direct Causation: Factor directly causes disease.

  • Indirect Causation: Involves intermediate steps, common in human biology.

Necessary vs Sufficient Factors

  • Necessary Factor: Required for disease development (e.g., without it, disease never occurs).

  • Sufficient Factor: Alone can cause disease (if present, disease always develops).

  • Sufficient but Not Necessary: More than one factor can cause the disease without being required for it.

  • Necessary but Not Sufficient: Must have the factor, but alone does not ensure the disease occurs.

  • Neither Necessary Nor Sufficient: Most complex situation; no single factor guarantees disease occurrence.

Bradford Hill Criteria for Causation

  1. Temporal Relationship: Exposure must precede disease development.

  2. Strength of the Association: Measured using Relative Risk or Odds Ratio; stronger association suggests causality.

  3. Dose-Response Relationship: Increased exposure correlates with increased disease risk.

  4. Replication of Findings: Consistent results across populations/studies support causality.

  5. Biologic Plausibility: There should be a biological explanation for the observed relationship.

  6. Consideration of Alternate Explanations: Other possible explanations must be ruled out.

  7. Cessation of Exposure: If exposure is removed, risk should diminish.

  8. Consistency with Other Knowledge: Causal evidence should align with existing scientific understanding.

  9. Specificity of Association: An exposure linked to a specific disease (not necessary for causality).

Bias in Studies

  • Bias Types:

    • Selection Bias: Errors in how subjects are chosen or retained in the study.

    • Information Bias: Incorrect data gathering leads to inaccurate exposure effect estimates.

    • Confounding: A third factor distorts the true association between exposure and disease (e.g., factor X may influence both).

Addressing Confounding

  • In Study Design: Match cases to controls with respect to confounders.

  • In Analysis: Use stratification to analyze effects within subgroups.

Interaction Effects

  • Occurs when effects of risk factors combine in an unexpected way.

    • Example: Drug effects differ by gender (interaction between drug and sex).

    • Interaction effects can be assessed using additive (attributable risk) and multiplicative (relative risk) models.

Summary of Key Concepts

  • Types of associations (spurious vs real), necessary vs sufficient factors, Bradford Hill criteria, biases, confounding, interaction effects.