Study Notes on Confounding by Danielle Lambert, PhD, MPH

Definition: A situation in which a measure of the effect of an exposure is distorted due to the association of the exposure with other factors that influence the outcome being studied.
Key Components:
- Exposure: The independent variable being tested.
- Outcome: The dependent variable or result being measured.
- Confounder: A third (extrinsic/extraneous) factor associated with both the exposure and outcome, but does not lie in the causal pathway between the two.

Two Locations:
- Location A: All women
- Location B: All men
Research Question: Are the mortality rates different between two locations?
Confounder: Gender
- It is ambiguous whether the difference in mortality rates results from location or gender effects.

Hypothesis: Higher salt intake leads to high blood pressure, which is associated with stroke.
Question: Is high blood pressure a confounder?
- Relationships:
- Salt Intake → Higher BP → Stroke

Diagram Representation (Causal Directed Acyclic Graph - DAG):
- Shows a third variable related to the exposure and outcome without direct relationship between them,
- High likelihood of producing misleading analyses suggesting a false relationship due to non-causal pathways.

1981 Study: Reported a statistical link between coffee drinking and pancreatic cancer.
- Source: The New York Times.
- Significance of the finding was uncertain.
2001 Study: No association found between coffee/alcohol consumption and increased pancreatic cancer risk, even with high consumption levels.
- Source: American Cancer Society.

1991 Study:
- Result: Relative risk of major coronary heart disease for women taking estrogen was 0.56 (95% CI: 0.40 to 0.80).
- No effect of estrogen use duration independent of age.
2003 Study:
- Study recommended terminating the estrogen-plus-progestin trial due to overall risks exceeding benefits.
- Hazard Ratio: 1.24 (95% CI: 1.00 to 1.54).
- Highest elevation in risk observed at one year (1.81 [95% CI: 1.09 to 3.01]).

Study: Investigated the relationship between ambient air pollution and school absenteeism due to respiratory illnesses.
- Conducted in Southern California among 4th graders.
- Ozone exposure was significantly linked to increased absenteeism rates due to respiratory illnesses, particularly lower respiratory illnesses.
- Findings: A 20 ppb increase in ozone resulted in:
- 62.9% increase in illness-related absences
- 82.9% in respiratory illnesses
- 173.9% in lower respiratory illnesses with wet cough.

Study Findings:
- Increase in asthma diagnosis risk correlated with early life exposure to various pollutants (NO, NO2, PM10).
- Example: Adjusted odds ratio for a 10-μg/m³ increase of NO was 1.08 (95% CI: 1.04–1.12).

Conduct literature searches to find prior studies on the same topic.
Understand biological or behavioral underpinnings related to the exposure and outcome.
Perform data analyses.

Analyze data to identify evidence of confounding.
Look at literature in addition to data analysis for evidence of confounding, even if the analysis does not indicate confounding.

The goal is to block potential "back door pathways" to measure the direct association between exposure and outcome.

Results indicated that physical activity serves as a protective factor.
Examination of confounding by calculating risk ratios (RR) and noting discrepancies between crude and adjusted values.

Randomization: Effectively prevents confounding; random assignment helps ensure equal distribution of confounders among treatment groups.
- Example: Randomizing participants to coffee vs. no coffee balances age, gender, and other confounders.
Restriction: Excluding participants based on confounding factors to ensure control over those factors.
- Example: Restricting to a specific age category can eliminate age as a confounding variable.
Matching: Ensures study groups do not differ with respect to potential confounders, either through individual matching or frequency matching.St

Randomization, restriction, and matching are used to eliminate confounding sources during data collection.

Standardization: Controls for confounders post-study.
Stratification: Evaluates exposure effects within strata of confounders to determine adjusted estimates of the relationship.
Multivariable Analysis: Simultaneously controls for several confounding variables using mathematical models.

Refers to "leftover confounding" even after adjustments.
Can arise from unknown confounders or variations within strata, suggesting the importance of recognizing and addressing potential residual confounding in research interpretations.

Rule of Thumb: Adjusted measure of association differing from unadjusted by 10% or more signifies evidence of confounding.
- Magnitude calculations:
- ext{Magnitude of Confounding} = rac{OR{ ext{crude}} - OR{ ext{adjusted}}}{OR_{ ext{adjusted}}} imes 100 ext{%}
- ext{Magnitude of Confounding} = rac{RR{ ext{crude}} - RR{ ext{adjusted}}}{RR_{ ext{adjusted}}} imes 100 ext{%}

Understanding confounding is crucial in epidemiological research and requires careful design, analysis, and interpretation to yield valid results and meaningful conclusions.