Matched Versus Unmatched Analysis of Matched Case-Control Studies.

Area of Science:

• Epidemiology
• Biostatistics
• Statistical Modeling

Background:

• Matched case-control studies are crucial in epidemiology for controlling confounding.
• Analytical methods for matched case-control studies with continuous matching factors require careful consideration.
• Existing debates focus on the most appropriate statistical approach for these complex designs.

Purpose of the Study:

• To compare the bias and efficiency of conditional logistic regression (CLR) and unconditional logistic regression (ULR).
• To evaluate analytical methods under both exact and nonexact matching scenarios.
• To assess the impact of matching on the association between variables and outcomes.

Main Methods:

• Derived the logit model for matched case-control samples under exact matching.
• Conducted simulations to validate theoretical conclusions and explore bias reduction techniques.
• Compared unadjusted and adjusted CLR and ULR, including spline smoothing for continuous variables.

Main Results:

• Conditional logistic regression (CLR) is unbiased when matching is exact.
• Unadjusted CLR exhibits bias with nonexact matching, which worsens with larger matching calipers.
• Adjusted unconditional logistic regression (ULR) is prone to bias due to model specification errors, regardless of matching exactness.

Conclusions:

• Conditional logistic regression (CLR) is the recommended primary analytical approach for matched case-control studies.
• Spline smoothing can mitigate bias in CLR when matching is not exact.
• Unconditional logistic regression (ULR) requires careful model specification to avoid bias.