Analytical Methods for Analyzing Biomarkers with Limit of Detection

Biomarkers are increasingly used in biomedical and public health sciences, drug development programs, testing for targeted therapeutics and personalized medicine. They are laboratory measures of biological processes that are viewed to be objective and quantifiable by providing valuable information in both assessing exposure and disease status. Increased use of biomarkers creates a venue for the development of methods to address important, unexplored analytic issues such as properly handling biomarker measurements below the limit of detection (LOD) or also known as non-detects or left-censored biomarkers. The common statistical analysis approach for biomarkers with LOD has been either to exclude measurements with LOD or perform ad-hoc single imputation techniques. Ad-hoc imputation techniques include replacing each measurement below the LOD with a value such as 0, LOD, LOD/2 or LOD/sqrt(2) and then conducting the analysis under the assumption that the imputed values are the actual observed values. However, there are other parametric and non-parametric statistical methods including model based single imputation methods, reverse Kaplan-Meier method, and multiple imputation methods that are not widely used in the literature but are better alternatives to ad-hoc single imputation techniques. We will facilitate broader use of all of these methods by describing their properties, comparing with each other, illustrating their use with a large population-based data for secondhand smoke exposure research and showing how they can be calculated using standard software.