Conventional clinical markers are by no means outperformed by microarray gene expression profilers in breast cancer prognosis
aComplex Systems Division, Department of Theoretical Physics, bDepartment of Oncology; Lund University, Lund, Sweden
AIM: To confront the predictive diagnostic power of cDNA microarray gene expression profilers with that of conventional prognostic markers for breast cancer patients using both established prognostic indices and novel combinations of conventional markers. METHODS: Several data sets containing both microarray gene expression measurements and clinical variables for breast cancer patients are used to predict metastases development. State-of-the-art classification tools like logistic regression, support vector machines and multilayered perceptrons, are used for calibration. The data are subdivided into training, validation and test sets in order to ensure unbiased evaluations of the results. RESULTS: For the different data sets studied, the gene expression profilers do not perform significantly better than indices constructed from the conventional variables when forming Receiver Operating Characteristic (ROC) areas and Kaplan-Meier survival plots for the metastasis prediction. The same holds when comparing to the well established Nottingham Prognostic Index, which combines histological grade, tumour size, and lymph node status. Also, in cases when one can separately evaluate the estrogen receptor (ER) status subgroups, the clinical predictor for the ER positive cohort is more powerful than for the ER negative one irrespective of profiler. CONCLUSIONS: Conventional markers are by no means outperformed by gene expression measurements using cDNA microarrays. Hence, at least for the time being, one should not overvalue the power of microarray gene expression data as clinical prediction tools. Nevertheless, microarray tumor expression data add knowledge in terms of insights into the important genes and pathways underlying disease outcomes. In addition, one should also keep in mind that the microarray technology is not as mature as the use of conventional markers. Also, it appears that the combination of different conventional prognostic markers into one prognostic index can be further improved. Given the costs of microarray processing and lack of standards, it may take time before it is used worldwide. Hence, it is important to pursue efforts around conventional markers.
Paper presented at the International Symposium on Predictive Oncology and Intervention Strategies; Nice, France; February 7 - 10, 2004; in oral session 993 (Molecular pathology - Part II).