Combining ecological and individual variables to reduce confounding by indication: case study--subarachnoid...

Ecological studies may reduce the problem of confounding by indication; however, these studies introduce new biases not present in individual-level analyses. To study the potential for ecological variables to reduce confounding by indication, we used a large database of admissions for ruptured cerebral aneurysms to evaluate the association of in-hospital death with treatment type-surgery or endovascular therapy. We compared results of three multivariable models: individual-level, ecological, and a two-level model with an ecological treatment variable and individual-level covariates and outcome. Trends in the individual-level and ecological models were in opposite directions, suggesting confounding by indication in the individual-level analysis. The two-level analysis revealed a strong association between institutional utilization of endovascular therapy and reduced individual risk of in-hospital death. Using an ecological treatment variable in an individual analysis may combine reduced confounding by indication in ecological analyses with increased power and more precise specification of outcomes and covariates in individual-level analyses.