1: an obscure speech or writing 2: something hard to understand or explain 3: an inscrutable ormysterious person Etymology: Latin aenigma, from Greek ainigmat-, ainigma, from ainissesthaito speak in riddles, from ainos fable, Date: 1539
Adapted from the Merriam-Webster Online Dictionary1
For many oncologists, a biostatistician's work can be easily characterized as anenigma. We know that all clinical trial or laboratory data must be framed instatistical language and have deep respect for someone who knows whichtest to apply when. This issue of CCR Focus is aimed at bridging the gapbetween biostatistician and oncologist in our world of targeted therapy andgenomic analyses.
Over the past decade, the biostatistician's work in oncology has steadilyexpanded as we ask not only which clinical trial results are valid, but also whatsets of genes among thousands evaluated are relevant to the clinical question. Asone example, among the increasing numbers of gene signatures evaluatingbreast cancer prognosis derived from cDNA microarray work, there appears tobe some prognostic concordance but with little overlap in gene identity(Haibe-Kains B, et al. BMC Genomics 9:394, 2008). With an Affymetrix chipcontaining over 20,000 genes, at P = 0.05, 1000 genes will reach statisticalsignificance by random chance for high expression in breast cancer samples ifno additional filters are applied. This simple concept may explain the lack ofoverlap. How, then, can investigators determine which signature, or whichbiomarker has been appropriately validated, and how can that signature orbiomarker be used in translational clinical trials?
The articles in this issue of CCR Focus, contributed by experts in the field,address several facets of this problem, beginning with an overview byStephen L. George as Guest Editor that highlights the contributions that follow.Chau, MacLeod, and Figg discuss validation of analytic methods for biomarkersused in drug development. Owzar et al. evaluate the necessary preprocessing ofmicroarray data - those early steps in microarray analysis that have major impacton final outcome. Taylor, Ankerst, and Andridge consider validation ofprediction models using genomic analyses. Finally, both George in the Focusoverview and Simon explore methods of clinical trial design that incorporatebiomarker detection. As with every issue of CCR Focus, our goal is that thesearticles will inform the clinical cancer researcher who is interested but notexpert in the field, but also stimulate the thinking of the expert. In the case ofthis CCR Focus on Biostatistics and Biomarkers, it is hoped that these articles fortranslational cancer researchers can begin to unravel the enigma.