Biomarkers in clinical practice: opportunities and challenges

Abstract

Many proposed biomarkers fail to produce clinically actionable results. Simply put, the research problem addressed here is: why do most biomarker projects fail? In this contribution we describe four commonly encountered problems and outline procedures that address these challenges. The specific issues addressed are as follows: A statistically significant result in a between-group hypothesis test often does not result in classification success. Cross-validation is commonly used in model validation. The successive steps in cross-validation expose it to multiple sources of failure that may result in erroneous conclusions of success. Failure to rigorously establish the test-retest reliability of a biomarker panel precludes its use in longitudinal monitoring of treatment response or disease progression. Further, it should be recognized that the minimum detectable difference is not the minimal clinically important difference. Sample size estimates used in the design of a clinical study must be determined by the objectives of the study. The sample sizes required in reliability studies and in the evaluation of biomarkers as prodromes must be determined with those objectives in mind and are far larger than sample size requirements computed for the purpose of hypothesis testing. We conclude with suggestions for transparency and collaboration that would facilitate the use of biomarkers in clinical practice.