Artikel
Evaluating the prognostic as well as the predictive value of gene-signatures in cancer clinical trials
Suche in Medline nach
Autoren
Veröffentlicht: | 14. September 2004 |
---|
Gliederung
Text
The advent of DNA microarray technology has recently enabled the quantitative measurement of complex multi-gene expression patterns in human cancer. The results can be used to increase the accuracy of the diagnosis of cancer and to identify new subtypes of tumors that can predict the clinical course of the cancer disease. This includes the risk of tumor progression and the duration of patient survival, but also the occurrence of adverse events to treatment. Therefore, genomic data are now increasingly collected in almost all stages of clinical research as well as in the development of new treatments. In particular, clinical studies were conducted to determine so-called gene expression signatures (GES) which are proposed to be used as a molecular marker for the prediction of clinical outcome either in addition to established clinical prognostic factors or in place of them. By gene expression signature we understand (synonymously with expression profile) a set of genes selected with a specific aim. The set of genes generates for each patient a multivariate feature vector of gene expression intensities. Usually the multivariate GES of a patient is summarized into a scalar to be used as a new prognostic index. As pointed out in Simon et al. [Ref. 1] an appropriate validation of the GES in an independent trial is a prerequisite for the use of this information for the prognosis of the course of the disease.
To evaluate whether a molecular prognostic factor offers any predictive value, i.e. the response to treatment depends on this factor, one has to investigate the interaction between treatment and the molecular factor in a multi-armed trial. The role of the interaction between molecular biomarker and treatment is illustrated in Figure 1 [Fig. 1] which shows a scenario of possible outcomes of a two armed trial and a dichotomous GES:
a) GES is a prognostic factor for response,
b) GES has the potential to be used as a predictive factor of response for specific treatment schedules.
This contribution will inform how GES are determined from microarray experiments and how a specific GES can be validated. It will be demonstrated how the statistical interaction defined in regression models can be used for the assessment of the GES as predictive factor. Data analysis can then be performed in a two-factorial setting. Examples will be given for real and simulated data and questions of sample sizes will be addressed.