A novel weighting method to remove bias from within-subject exposure dependency in case-crossover studies

Area of Science:

• Epidemiology
• Biostatistics
• Pharmacoepidemiology

Background:

• Case-crossover studies are prevalent in pharmacoepidemiology.
• Existing methods may yield biased results when within-subject exposure dependency is present.
• This bias has been understudied.

Purpose of the Study:

• To develop a novel weighting method for case-crossover studies to eliminate bias from within-subject exposure dependency.
• To evaluate the performance of the proposed method through simulations and real-world data analysis.

Main Methods:

• A weighting method was developed by calculating exposure probability at the case period.
• Simulated data with cyclic treatment patterns and within-subject exposure dependency were used.
• The method was applied to real-world data on celecoxib and peripheral edema in Japan, and SSRI and hip fracture in Australia.

Main Results:

• The proposed weighting method accurately reproduced true rate ratios in simulations, outperforming standard conditional logistic regression and Mantel-Haenszel methods when time-varying confounders were present.
• Standard conditional logistic regression showed bias, which increased with extended study periods.
• Real-world analyses showed stable point estimates for odds ratios using the weighting method, comparable to Mantel-Haenszel.

Conclusions:

• Within-subject exposure dependency can bias case-crossover studies, even without exposure time trends.
• The proposed weighting method effectively removes this bias and accounts for time-varying confounders.
• This method is recommended for improving the accuracy of case-crossover study analyses.