Causation in Epidemiology Notes

Approaches for Studying Disease Etiology

• Animal studies offer controlled environments but raise concerns about extrapolating data from animals to humans.
• In vitro systems (cell or organ culture) also present extrapolation challenges due to their artificial nature.
• To determine if a substance causes disease in humans, observations in human populations are needed.
• Epidemiology utilizes "unplanned" or "natural" experiments by studying groups exposed for non-study purposes (e.g., occupational cohorts, disaster survivors).

Approaches to Etiology in Human Populations

• A typical sequence in human studies involves clinical observations, analysis of available data, and new studies (case-control, cohort).
• Case-control studies are often the first step to explore a relationship, followed by cohort studies.
• Randomized trials are rarely used for suspected toxins/carcinogens, mainly for beneficial agents.
• The process involves determining associations and then assessing causality.

Types of Associations

• Observed associations can be true (real) or false (spurious) due to study design flaws.
• Real associations can be causal or noncausal (due to confounding).
• Causal Association: Exposure induces disease development.
• Noncausal Association: Exposure and disease are linked to a third factor (confounding variable).

Interpreting Real Associations

• If the relationship is causal, interventions targeting the exposure will reduce disease risk.
• If due to confounding, interventions on the exposure won't affect disease risk; the focus should be on factor X.
• Example: Smoking during pregnancy and low birth weight; distinguishing between causal vs. confounding factors is critical.

Types of Causal Relationships

• Necessary and Sufficient: Factor A always causes the disease, and the disease never develops without it. (Rare)

  • $Factor A \rightarrow Disease$ (Always)

• Necessary but Not Sufficient: Factor A is required, but other factors are also needed for the disease to develop. (e.g., multistage carcinogenesis) Initiation + promotion.

  • $Factor A + Other Factors \rightarrow Disease$

• Sufficient but Not Necessary: Factor A alone can cause the disease, but other factors can also cause the same disease independently. Radiation exposure or benzene exposure can produce leukemia.

  • $Factor A \rightarrow Disease$ or $Factor B \rightarrow Disease$

• Neither Sufficient Nor Necessary: Factor A is neither essential nor enough to cause the disease. This model applies to many chronic diseases. CHD risk factors may be non-overlapping (smoking, diabetes, low HDL) or (hypercholesterolemia, hypertension, physical inactivity).

  • $(Factor A + Other Factors) \rightarrow Disease$ or $(Factor B + Other Factors) \rightarrow Disease$

• Rothman’s model proposes that a “sufficient cause” is a constellation of “component causes”.

Guidelines for Judging Whether an Observed Association Is Causal

• Temporal Relationship: Exposure must precede disease.
  • Example: Increased air particle concentration preceding increased mortality.
• Strength of the Association: Measured by relative risk or odds ratio.
  • Stronger association = More likely causal.
• Dose-Response Relationship: Increased exposure dose = Increased disease risk.
  • Presence strengthens causality; absence doesn't negate it.
• Replication of the Findings: Consistent findings across different studies/populations.
• Biologic Plausibility: Coherence with existing biologic knowledge.
• Consideration of Alternate Explanations: Ruling out confounding.
• Cessation of Exposure: Reduced disease risk upon exposure reduction/elimination.
  • Example: Reduced lung cancer risk after smoking cessation.
• Consistency With Other Knowledge: Findings align with other data.
• Specificity of the Association: Specific exposure linked to a single disease. (Weakest guideline).

Deriving Causal Inferences

• The guidelines do not permit a quantitative estimation of whether an association is causal.
• Koch’s postulates useful for infectious diseases but less applicable to non-infectious diseases.

Modifications of the Guidelines for Causal Inferences

• US Public Health Service and US Preventive Services Task Force have modified guidelines.
• Prioritize evidence categorization by quality of sources and evaluation of causal relationships using standardized guidelines.
• USPSTF assesses certainty of net benefit (benefit - harms).
• The certainty of net benefit is graded on a three-point scale: high, moderate, or low.
• Task Force recommendations are based on certainty and magnitude of net benefit.