In research, directed acyclic graphs typically guide decisions on choosing confounders that need to be adjusted. These graphs also suggest how to get unconfounded, or unbiased, effect estimates, which are based on different statistical models and do not always provide the same actual numeric values. Therefore, Ghassan Hamra, PhD, of Johns Hopkins University Bloomberg School of Public Health, and his team from Wake Forest University and University of California Davis joined together to better understand when these values do and do not correspond.
Their study, “Combining Effect Estimates Across Cohorts and Sufficient Adjustment Sets for Collaborative Research: A Simulation Study” was recently published in the February issue of Epidemiology. To conduct this research, all study data were simulated, and Hamra built all of the models and analyzed all of the information. Ultimately, the team found the estimators showed similar numeric values of the estimates the majority of the time. The exception is when using a model known as logistic regression, which did not give similar estimates. Logistic regression provides an odds ratio, which is a non-collapsible quantity, or one that cannot be reliably combined.
“This study has a significant impact on the ECHO Program, as it allows researchers to understand and combine information across ECHO groups,” said Hamra. In the future, researchers can make additional models to explore other forms of bias that make combining information across studies difficult.