BEMP: Methodological Research for Biomarker Discovery and Validation
The importance of biomarker discovery, and later validation, to progress in translational research is underlined by the large numbers of Requests for Applications each year that specifically include a biomarker component. Although there is a large literature on biomarker validation in general, much of this was originally developed in the context of developing diagnostic tests, particularly in cancer screening, and there remains a large need to extend these methods to other clinical areas. For example, a screening marker intended to detect the presence or absence of disease, may not be valuable in prediction of disease progression or response to therapy within individuals with the disorder. For example, in chronic kidney disease, monitoring of disease status is often based monitoring of longitudinal changes of estimates of glomerular filtration rate (GFR), a continuous measure. It is also not always necessary to apply the classical methods of, say, receiver operating characteristic (ROC) analysis, to define cut points for novel biomarkers; a biomarker may contribute useful information to risk stratification models, even if a cut point is not strictly defined.
Furthermore, biomarker validation efforts must take into account available clinical data, as well as assay and analyte characteristics. A valuable biomarker must contribute information above and beyond existing clinical data to risk prediction models. Biomarkers may predict disease progression and inform individual-level treatment decisions. Risk estimates can be improved through the use of novel biomarkers, and in some cases, re-classify those at intermediate risk into high or low risk groups and directly impact treatment decisions.
Since biomarkers might eventually serve as "surrogate markers" for clinical endpoints, particularly in clinical trials, further research on statistical methods for the validation of surrogate markers, as related to endpoint identification, is also needed. As illustrated with the experience with HDL, interpretation and prediction of the surrogate value of biomarkers may prove quite complex. While a biomarker that is strongly associated with a definitive disease outcome has the potential to move a therapeutic field forward, biomarkers may be highly informative, even if they do not meet the criteria for surrogacy.