Statistical Challenges in the Analysis of Biomarker Data

The statistical analysis of biomarker data can be very challenging. Sample sizes can be quite small, and biomarker data often exhibit violations of many of the assumptions underlying standard statistical analyses. Biomarker data are often skewed, and there may be issues with discordant observations. In this chapter, we describe several of the challenges that one is likely to encounter in the routine analysis of biomarker data and offer recommendations on methods for dealing with these challenges. For example, we discuss methods for (1) detecting and dealing with nonnormally distributed data, (2) dealing with violations of the homogeneity assumption when comparing the means of two independent groups, (3) analyzing data from dependent groups, and (4) detecting and accommodating outliers. Each of these challenges is illustrated using one or more examples taken from published biomarker studies. A software code for implementing the recommended methods of analysis is also provided.