Issues with Categorizing Continuous Variables in Cancer Prognosis
Author Information
Author(s): Douglas G. Altman
Primary Institution: Imperial Cancer Research Fund
Hypothesis
The method of categorizing continuous prognostic variables in cancer prediction is flawed.
Conclusion
The approach of using data-dependent cut-offs for categorizing continuous variables can lead to overestimation of prognostic importance and invalid P values.
Supporting Evidence
- The authors' method is data-dependent, which can lead to incorrect conclusions.
- Different studies may find different optimal cut-off points due to sampling variation.
- Dichotomizing continuous variables can produce biologically unrealistic models.
Takeaway
When doctors try to group continuous data into categories for cancer predictions, they can make mistakes that lead to wrong conclusions. It's better to decide on the groups before looking at the data.
Potential Biases
The method suggested by Sigurdsson et al. may lead to a raised Type I error rate, falsely indicating significant effects.
Limitations
The commentary highlights the lack of a recognized procedure for adjusting P values for multiple testing and the variability in optimal cut-off points across studies.
Participant Demographics
Breast cancer patients are mentioned as the relevant population.
Want to read the original?
Access the complete publication on the publisher's website