Evaluating Gene Expression Models for Ovarian Cancer
Author Information
Author(s): Gevaert Olivier, De Smet Frank, Van Gorp Toon, Pochet Nathalie, Engelen Kristof, Amant Frederic, De Moor Bart, Timmerman Dirk, Vergote Ignace
Primary Institution: Katholieke Universiteit Leuven
Hypothesis
Can gene expression analysis reliably predict the clinical behavior of ovarian cancer in an independent patient cohort?
Conclusion
The study found that existing gene expression models for predicting response to platin-based chemotherapy in ovarian cancer need further validation, as they did not perform well in an independent evaluation.
Supporting Evidence
- The pilot study showed 100% accuracy in predicting advanced stage disease.
- The independent evaluation did not confirm the pilot study's results.
- Models based on previously published gene sets did not perform significantly better than chance.
Takeaway
This study looked at whether tests based on gene patterns could help predict how ovarian cancer would behave, but the tests didn't work as well as hoped when tried on new patients.
Methodology
The study used microarray analysis and statistical methods including PCA and LS-SVM on tumor samples from 49 patients.
Potential Biases
Potential overfitting of models due to the small sample size and high dimensionality of gene expression data.
Limitations
The models were not able to predict platin resistance effectively, and the independent data set did not match the pilot study's characteristics.
Participant Demographics
The independent data set included 4 stage I, 30 platin-sensitive advanced stage, and 15 platin-resistant advanced stage ovarian cancer patients.
Statistical Information
P-Value
0.12
Confidence Interval
19%–99%
Statistical Significance
p<0.05
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website