Bias control in the analysis of case-control studies with incidence density sampling

Area of Science:

• Epidemiology
• Biostatistics

Background:

• Case-control studies using incidence density sampling with time matching previously showed biased estimates with conditional logistic regression (CLR).
• The cause of bias in CLR for incidence density sampling was unknown.
• Unconditional logistic regression (ULR) with adjustment for matched factors is known to provide unbiased estimates in exclusive sampling designs.

Purpose of the Study:

• To evaluate the validity of ULR with time adjustment in incidence density sampling.
• To compare the performance of CLR with and without bias reduction against ULR methods.

Main Methods:

• Extensive simulations of incidence density sampling were conducted.
• Analyzed methods included CLR (with/without bias reduction) and ULR (with time quintiles, residual time, or matched sets).
• A case-control study of Haemophilus influenzae type B vaccine was re-analyzed.

Main Results:

• Bias in CLR was attributed to sparse data, manageable with bias reduction or increased sample size.
• ULR adjusted for time quintiles yielded results comparable to CLR, even when matches were broken.
• Time-varying effects necessitated further adjustment for residual time trends in ULR.
• ULR adjusted for matched sets showed poor performance despite bias reduction.

Conclusions:

• Incidence density sampling case-control studies can be reliably analyzed using ULR with time adjustment.
• CLR, particularly with bias reduction, is also a valid analytical approach.
• The choice between ULR and CLR depends on specific study characteristics and the presence of time-varying effects.