Statistical methods for elimination of guarantee-time bias in cohort studies: a simulation study

Area of Science:

• Pharmaco-epidemiology
• Biostatistics
• Health Sciences

Background:

• Aspirin use is linked to reduced cardiovascular disease and cancer risk.
• Pharmaco-epidemiology studies commonly use Cox regression, but it can overestimate drug effects due to guarantee-time bias.
• Time-dependent Cox models and landmark methods offer alternatives to address this bias.

Purpose of the Study:

• Compare the performance of Cox regression, time-dependent Cox model, and landmark methods.
• Evaluate methods for addressing guarantee-time bias in pharmaco-epidemiology.
• Identify the most appropriate statistical approach for analyzing cumulative drug dose effects.

Main Methods:

• Statistical modeling and simulation studies assessed Type I error, bias, power, and mean squared error (MSE).
• Three methods were compared: conventional Cox regression, time-dependent Cox model, and landmark analysis with varying landmark times.
• The methods were applied to real-world data from the Korean National Health Insurance Database.

Main Results:

• Time-dependent Cox regression demonstrated superior performance over the landmark method regarding bias and MSE in simulated data.
• Type I error rates were comparable across the methods in simulations.
• Real-world data analysis corroborated simulation findings, showing similar patterns.

Conclusions:

• Both time-dependent Cox regression and landmark analysis effectively mitigate guarantee-time bias.
• Time-dependent Cox regression is identified as the optimal method for analyzing cumulative dose-response relationships in pharmaco-epidemiology.
• This finding enhances the reliability of drug effect estimations in cohort studies.