Confounding

What is confounding?

The mixing of effects between the exposure and outcome; and a third variable, which is called a confounder. Like bias, confounding distorts the relationship between exposure and outcome.

What is the counterfactual ideal?

The ideal comparison group in a cohort study are the same individuals in the exposed group had they not been exposed.

What defines a confounder?

Must cause or predict the outcome in both exposed and unexposed. Must be associated with the exposure in the population that produces the cases. Must not be on the casual pathway between exposure and outcome.

Thought process for identifying cofounders?

Association present between confounder and disease. Evaluate association between confounder and exposure.

What is confounding by indication?

Reflects the fact that treatments are given to those with a particular disease.

What is confounding by severity?

Disease and its severity are both indications for the exposure.

What are solutions to confounding by indication and confounding by severity respectively?

(1) Restrict the study population so that each participant has the disease. Can compare different versions of the exposure. (2) Use an experimental study to balance confounders

What are ways of controlling for confounding in the design phase?

Randomization, restriction, matching

How does randomization control for confounding?

Balances known and unknown confounders

How does restriction control for confounding? Strength and limitations?

Entry into the study is confined to individuals who fall into a specific category of the confounder. Convenient + inexpensive. Difficult to find participants, limits generalizability.

How does matching control for confounding?

Process of making the distribution of matching factors identical across study groups. Individual matching: Both case and control must be 55 yo men. Frequency matching: 20% of cases and controls aged 40-59.

What are limitations of matching?

Can’t assess association of matching factor with outcome. Difficult and expensive to find matches. Overmatching (associated with exposure but not disease)

What is a strength of matching?

Good for neighbourhood control studies

What are ways of controlling for confounding through analysis?

Direct and indirect standardization, stratification, effect modification

How does stratification control for confounding?

Assess the association of exposure and outcome within groups defined by level of the confounder.

What is the Mante-Haenzel OR?

Weighed average of odds ratios.

What are strengths and limitations of stratification?

Easy to carry out. But cannot control for several variables at the same time.

What is multivariable regression?

Construct a mathematical model to describe association between exposures, disease, and confounders. Model the association of exposure with disease, adjusted for confounders.

What are propensity scores?

Summarize the likelihood of a participant being exposed versus unexposed based on observed confounders.

What are the causes of residual confounding?

Data not collected. Differences in risk within a category of confounder (e.g., smoking). Misclassified confounder.