Matching methods

Matching - Taming Confounders

• A method used in observational studies (case-control, cohort) to control for known confounders from the start.
• Ensures the distribution of key variables is similar between the groups being compared, reducing bias.

Primary Methods:

• Individual Matching (Paired): Each case is paired with one or more controls based on specific confounding variables (e.g., matching a 65-year-old male case with a 65-year-old male control).
• Frequency Matching (Group): The control group is selected to have a similar distribution of the confounding variables as the case group (e.g., if cases are 30% female, the control group is also 30% female).
• Propensity Score Matching: A statistical method that summarizes multiple confounders into a single score ($P( ext{exposure | covariates})$). Individuals with similar scores are then matched.

Pros & Cons:

• Advantages: ↑ statistical power and intuitive control over known confounders.
• Disadvantages: Costly, can be hard to find matches, and makes it impossible to study the effect of the matched variable.

⭐ Crucial Takeaway: You cannot analyze the effect of a variable that was used for matching. If you match on smoking status, you cannot determine if smoking is an independent risk factor for the disease in that study.

Matching - A Double-Edged Sword

• Concept: A method in case-control studies to control for confounding by pairing each case with one or more controls who share specific characteristics (e.g., age, sex).

• Primary Goal: To increase study efficiency and control for known, strong confounders.

• Advantages:

  • Controls for confounding variables that are difficult to measure or categorize.
  • Can improve statistical power by creating a more balanced distribution of confounders.

• Disadvantages (The "Sword's Other Edge"):

  • Irreversibility: Once a variable is used for matching, its association with the disease cannot be analyzed.
  • Overmatching: Matching on a variable that is part of the causal pathway between exposure and outcome can introduce bias.
  • Costly and time-consuming.

⭐ Exam Favorite: If you match by a specific factor (e.g., smoking status), you have effectively neutralized its effect and can no longer assess it as an independent risk factor for the outcome.

Analysis - Matched Pair Nuances

• Core Principle: Matched data requires special analytic methods because the paired observations are not independent. Standard chi-square tests are invalid as they assume independence.

• Case-Control Studies (Matched Pairs):

  • Analysis focuses on discordant pairs (pairs where exposure status differs between case and control).
  • Use McNemar's test or conditional logistic regression.
  • The odds ratio (OR) is the ratio of discordant pairs: $OR = b/c$.
    • b = Pairs where case is exposed & control is not.
    • c = Pairs where case is not exposed & control is.

• Matched Cohort Studies:

  • Analysis requires stratified methods (e.g., Mantel-Haenszel) to pool the stratum-specific RRs.

⭐ Exam Favorite: You cannot analyze the effect of the variable used for matching. For example, if you match by age, you cannot determine if age itself is a risk factor for the disease in that study.

High‑Yield Points - ⚡ Biggest Takeaways

• Matching is a key method to control for confounding, primarily in case-control studies.
• It ensures the distribution of potential confounders (e.g., age, sex) is similar between cases and controls.
• Crucially, the effect of a matched variable on the outcome cannot be analyzed.
• Overmatching on a variable in the causal pathway can introduce bias and reduce precision.
• It can improve statistical efficiency when used correctly for strong confounders.