The Need for Multiple-Bias Modeling in Observational Studies

• Activity Number: 1
• Type: Invited
• Date/Time: Sunday, August 3, 2003 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Sander Greenland*+
• Companies: University of California, Los Angeles
• Keywords: Bayesian statistics ; epidemiologic methods ; observational studies ; sensitivity analysis ; risk analysis ; risk assessment

Abstract

Ordinary analytic methods do not reflect any source of uncertainty other than random error; as a result, readers must rely on informal judgments regarding the impact of possible biases. When standard errors are small these judgments often fail to adequately capture sources of uncertainty and their interactions. Multiple-bias modeling provides alternatives that can systematically integrate major sources of uncertainty, thus providing better input to research planning and policy analysis. It can also also be used to demonstrate how results can change dramatically with small and reasonable changes in the priors. These points are illustrated with a pooled analysis of residential magnetic-field exposures and childhood leukemia, which highlights the diminishing returns from conventional studies conducted after the early 1990s and the extreme sensitivity of results to the classification prior. It is argued that multiple-bias modeling should become part of the core training of anyone who will be entrusted with the analysis of observational data, and should become standard procedure when random error is not the main component of uncertainty (as in meta-analysis and pooled analysis).