A confounding variable must satisfy which of the following criteria to be considered a true confounder?

Explanation

## Criteria for a True Confounder **Key Point:** A variable qualifies as a confounder if and only if it satisfies THREE criteria: 1. **Associated with the exposure** (in the source population) 2. **Independently associated with the outcome** (in the unexposed group) 3. **NOT in the causal pathway** between exposure and outcome (i.e., not a mediator) ## Detailed Explanation ### Criterion 1: Association with Exposure - The confounder must be unevenly distributed between exposed and unexposed groups - Example: Age is associated with smoking (smokers tend to be younger in some populations) ### Criterion 2: Independent Association with Outcome - The confounder must be associated with the outcome **among the unexposed** - This ensures it has an effect on the outcome regardless of exposure status - Example: Age is independently associated with cardiovascular disease even in non-smokers ### Criterion 3: Not in the Causal Pathway - The confounder must not be a **mediator** (intermediate variable in the causal chain) - If it is in the pathway, controlling for it would remove the true effect - Example: Smoking → lung inflammation → cancer; lung inflammation is a mediator, not a confounder ## Mnemonic: **ACNE** **A** — Associated with exposure **C** — Causes (or is associated with) the outcome **N** — Not in the causal pathway **E** — Exists in the source population (prior to exposure) ## Confounding vs. Mediation | Feature | Confounder | Mediator | |---------|-----------|----------| | **In causal pathway** | No | Yes | | **Should be controlled** | Yes | No (unless assessing direct effect) | | **Example** | Age in smoking–CVD | Atherosclerosis in smoking–MI | **High-Yield:** The most commonly tested criterion is the **independent association with outcome in the unexposed group**. This distinguishes confounders from mediators and ensures the variable is truly confounding, not just part of the causal mechanism. **Warning:** A variable associated with the outcome but NOT with the exposure is NOT a confounder — it may increase random error but does not bias the association.