Editors' ChoiceBiostatistics

Sophisticated PIGs Help to Identify Biomarkers

See allHide authors and affiliations

Science Translational Medicine  17 Nov 2010:
Vol. 2, Issue 58, pp. 58ec179
DOI: 10.1126/scitranslmed.3001911

Everyone’s looking for biomarkers. Defined as any physiologic or anatomical measurement of a quantity related to a disease process, biomarkers are needed to find therapies for complex diseases. Examples include blood cholesterol concentrations, brain volume, genetic signatures, and bone mineral density. In clinical trials that use and evaluate biomarkers, it is important to be able to tell whether a change in a biomarker mediates the treatment effect on the primary clinical outcome. When this is true, the biomarker can also be called a surrogate or interim endpoint. It may be informative about the mechanism of the disease or provide a suitable alternative endpoint that is easier to measure, either technically or temporally. Various approaches to identify and quantify the contribution of a biomarker have been proposed. All of them, however, inadequately identify biomarkers when the biomarker is subject to large measurement error, which, unfortunately, is a common problem. Now, Li and Qu offer a solution to this difficulty. They propose a method for evaluating biomarkers that accounts for measurement error. They incorporate classical measurement-error modeling ideas into a measure of the proportion of information gained (PIG). In their approach to biomarker identification, a series of simple regression models (linear, logistic, or time-to-event models) are used to fit the association between the biomarker and the primary outcome and the association between the treatment and the primary outcome, adjusted for the proposed biomarker. Their approach provides an improved estimate of the PIG when the biomarker is measured with substantial error. This is an important advance in biomarker identification because it allows for more appropriate identification of biomarkers in situations where the biomarker may go undetected with current analysis platforms.

W. Li, Y. Qu, Adjustment for the measurement error in evaluating biomarkers. Stat. Med. 29, 2338–2346 (2010). [Abstract]

Navigate This Article