Bias analysis quantifies the influence of systematic error on an epidemiology
study’s estimate of association. The fundamental methods of bias analysis in epidemiology
have been well described for decades, yet are seldom applied in published
presentations of epidemiologic research. More recent advances in bias analysis,
such as probabilistic bias analysis, appear even more rarely. We suspect that there
are both supply-side and demand-side explanations for the scarcity of bias analysis.
On the demand side, journal reviewers and editors seldom request that authors
address systematic error aside from listing them as limitations of their particular
study. This listing is often accompanied by explanations for why the limitations
should not pose much concern. On the supply side, methods for bias analysis
receive little attention in most epidemiology curriculums, are often scattered
throughout textbooks or absent from them altogether, and cannot be implemented
easily using standard statistical computing software. Our objective in this text is to
reduce these supply-side barriers, with the hope that demand for quantitative bias
analysis will follow.
We began this project after writing a few bias analysis papers in epidemiology
journals. The American College of Epidemiology invited us to prepare a day-long
workshop on the methods of bias analysis, which became an outline, which in turn
became this text. The text begins with a description of the need for bias analysis
and continues with two chapters on the legwork that must be done at the outset of
a research project to plan and implement a bias analysis. This introductory section
is followed by three chapters that explain simple bias analysis methods to address
each of the fundamental threats to the validity of an epidemiology study’s estimate
of association: selection bias, classification errors, and uncontrolled confounding.
We then extend the bias analysis methods from these early chapters to multidimensional,
probabilistic, and ultimately multiple bias analysis methods. The book
concludes with a chapter on the presentation and interpretation of the results of a
bias analysis.