Related Experiment Videos
Limits of predictive models using microarray data for breast cancer clinical treatment outcome
James F Reid1, Lara Lusa, Loris De Cecco
1Department of Experimental Oncology, Istituto Nazionale per lo Studio e la Cura dei Tumori, Milan, Italy. james.reid@ifom-ieo-campus.it
Journal of the National Cancer Institute
|June 16, 2005
Summary
A two-gene breast cancer predictive model for antiestrogen response did not validate in an independent patient cohort. Current microarray-based models with limited patient numbers and genes show poor predictive accuracy for treatment outcomes.
Area of Science:
- Oncology
- Genomics
- Biostatistics
Background:
- Microarray data aids in developing predictive models for breast cancer treatment outcomes.
- A specific two-gene model for antiestrogen response after tamoxifen treatment was previously proposed.
Purpose of the Study:
- To validate a proposed two-gene predictive model for antiestrogen response in an independent cohort of breast cancer patients.
- To assess the performance of microarray-based predictive models using more than two genes.
Main Methods:
- Real-time quantitative polymerase chain reaction (PCR) was used to measure HOXB13 and IL17BR gene expression.
- Statistical analyses included univariate logistic regression, area under the receiver-operating-characteristic curve (AUC), t tests, and Mann-Whitney tests.
- Supervised methods were applied to original and independent microarray datasets to estimate classification accuracy.
Main Results:
- The two-gene predictor's performance could not be validated in the independent cohort (P > .18 for all analyses).
- Area under the receiver-operating-characteristic curve (AUC) and other statistical tests yielded similar non-validating results.
- Estimates from independent microarray datasets indicated poor classification accuracy for treatment-response predictive models with current sample sizes and gene numbers.
Conclusions:
- The proposed two-gene model for predicting antiestrogen response in breast cancer lacks validation in an independent cohort.
- Current limitations in sample size and the number of informative genes hinder the development of accurate microarray-based predictive models for breast cancer treatment response.