Comparative Optimism in Cancer Prognosis Models
Author Information
Author(s): Caroline Truntzer, Delphine Maucort-Boulch, Pascal Roy
Primary Institution: Hospices Civils de Lyon, Service de Biostatistique, Lyon, France
Hypothesis
To what extent does optimism in transcriptomic models lead to overestimation of their contribution to survival prognosis compared to clinical variables?
Conclusion
The predictive power of clinical variables is not overestimated, while gene selection processes can lead to significant optimism and overestimation.
Supporting Evidence
- Clinical variables have been validated in many studies, while gene variables are still under selection.
- The optimism for clinical variables is low because they are not subject to selection.
- Gene selection processes can introduce high optimism, especially when the number of relevant genes is low.
Takeaway
This study shows that when predicting cancer outcomes, we should be careful with gene data because it can make things look better than they really are, while clinical data is more reliable.
Methodology
Cox proportional hazards models were built using simulated datasets that included both clinical and transcriptomic variables to compare their predictive power.
Potential Biases
The optimism in gene selection can lead to overestimation of their predictive power, especially with small sample sizes.
Limitations
The study is based on simulated datasets, which may not fully capture the complexities of real-world data.
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website