## Controlling Confounding During Study Design ### Methods to Control Confounding Confounding can be addressed at three stages: **design**, **data collection**, and **analysis**. The question specifies control during the **design phase**. | Control Method | Stage | Mechanism | Strength | |---|---|---|---| | **Matching** | Design | Ensure exposed and unexposed groups have similar confounder distribution | Prevents confounding from entering the study | | **Restriction** | Design | Limit study population to a single stratum of confounder | Simple but reduces generalizability | | **Randomization** | Design | Random allocation eliminates systematic differences | Gold standard (RCTs) | | **Stratification** | Analysis | Analyze separately within confounder strata | Reveals confounding but doesn't prevent it | | **Statistical adjustment** | Analysis | Multivariate regression controls confounding mathematically | Flexible, post-hoc | ### Why Matching is the Design-Phase Answer **Key Point:** Matching ensures that the distribution of the confounding variable (age) is similar between exposed (OCP users) and unexposed (non-users) groups **before** data collection begins. **High-Yield:** Design-phase controls (matching, restriction, randomization) **prevent** confounding from entering the study; analysis-phase controls (stratification, adjustment) **manage** confounding after it may have occurred. **Clinical Pearl:** In this OCP-VTE example, matching on age would ensure that the age distribution of women taking OCPs is comparable to those not taking OCPs, eliminating age as a source of confounding. ### Distinction from Other Options - **Stratification** is an analysis method, not a design method - **Regression adjustment** is an analysis method, not a design method - **Increasing sample size** addresses random error (precision), not systematic error (confounding)
Sign up free to access AI-powered MCQ practice with detailed explanations and adaptive learning.