Risk set sampling for case-crossover designs.

Area of Science:

• Epidemiology
• Biostatistics
• Environmental Health

Background:

• Case-crossover designs compare exposures at failure times to control times within subjects.
• Appropriate control time sampling can address unmeasured confounding but risks bias from exposure time trends.
• Risk set sampling theory offers methods to mitigate time-trend bias in effect estimates.

Purpose of the Study:

• To compare the performance of different control time sampling schemes in case-crossover studies.
• To evaluate bias and confounding control in various bidirectional designs and Poisson regression.
• To identify an optimal sampling strategy for analyzing environmental exposures like particulate matter.

Main Methods:

• Simulated data using daily particulate matter (PM10) levels in Denver.
• Introduced linear and seasonal trends for confounding and simulated mortality counts.
• Compared four sampling schemes: full-stratum bidirectional, matched pair, symmetric bidirectional, and semi-symmetric bidirectional designs.
• Assessed a quasi-likelihood extension of Poisson regression with overdispersion.

Main Results:

• Matched pair, full-stratum design, and overdispersed Poisson regression failed to control seasonal confounding.
• Symmetric bidirectional design controlled seasonal confounding but introduced time-trend bias.
• Semi-symmetric bidirectional design achieved equal seasonal confounding control without time-trend bias.

Conclusions:

• The semi-symmetric bidirectional design is superior for case-crossover studies with time-varying exposures.
• This design effectively controls for seasonal confounding while mitigating bias from exposure trends.
• It provides a reliable method for estimating health effects of environmental exposures.