The Trend-in-trend Research Design for Causal Inference

Area of Science:

• Epidemiology
• Biostatistics

Background:

• Cohort studies are susceptible to bias from unmeasured confounding factors.
• Unmeasured confounding can distort the true association between exposures and outcomes.

Purpose of the Study:

• To introduce and evaluate the trend-in-trend design, a hybrid ecologic-epidemiologic approach.
• To assess the performance of the trend-in-trend design in mitigating unmeasured confounding.

Main Methods:

• The trend-in-trend design stratifies populations by cumulative exposure probability, effectively analyzing time trends in exposure.
• A covariates-free maximum likelihood model estimates odds ratios using exposure prevalence and outcome data across multiple time periods.
• The method requires a strong time trend in exposure for validity.

Main Results:

• Simulations demonstrated negligible bias in odds ratio estimates, even with unmeasured confounding.
• Empirical applications successfully replicated known associations, including the link between rofecoxib and myocardial infarction.
• The design also confirmed null associations for rofecoxib with hypoglycemia and fracture.

Conclusions:

• The trend-in-trend design offers a valuable tool for epidemiological research, particularly when strong exposure trends exist.
• This method provides a more robust approach to handling unmeasured confounding compared to traditional cohort studies.
• It is particularly useful for evaluating newly introduced medical interventions or drugs with changing exposure patterns.