| Resumo: |
Countless statistical methods have been described for the analysis of DNA microarrays, and each yields distinct results. This raises the question whether DNA microarrays are robust diagnostic tools.

In order to address this issue, we compared five formally similar statistical tests for gene selection on a single data set derived from acute leukemia patients. Inter-test agreement of gene selection, of sample classification and with standard clinical diagnosis was calculated using Cohen's κ-score.

The inter-test agreement scores were 0.15 < κ < 0.68 for gene selection, and 0.60 < κ < 0.89 for sample classification. Comparison to the clinical diagnosis showed agreement scores of 0.58< κ < 0.88. 'Marginal imbalance' explains the low κ-scores at the level of gene selection. At the levels of sample classification and agreement with clinical diagnosis, κ-scores can be considered "substantial" to "excellent".

For diagnostic purposes, the inter-test agreement of DNA microarrays is equivalent to that of experienced clinicians, as reported in the literature. The technique can thus be considered a useful and robust addendum to the available diagnostic tools.

|
