Skip to content

SchmidtGroupNKI/PREDICT-NL_Methods

 
 

Repository files navigation

Improving the predictive performance of Cox models: A framework for updating and extension with new markers

Continuous monitoring and improvement of clinical prediction models are essential for maintaining their performance. We developed a systematic framework for updating and extension of a Cox regression model for time-to-event data, with illustration for the PREDICT model. PREDICT is a widely-used model for deciding the need for adjuvant therapy for breast cancer. Like any prediction model, it may underperform in new patients. We illustrated the updating and extension of PREDICT for the MINDACT trial cohort, which comprised somewhat more selectively recruited participants treated according to more recent clinical guidelines compared to the registry data underlying the development of PREDICT.

Syntax files

File Description
1_Cleaning.Rmd Cleaning, preparation, and imputation of the MINDACT dataset.
2_Descriptive.Rmd Descriptive tables.
3_PREDICT_validation.Rmd Validation of the PREDICT v2.3 model.
4_PREDICT_updating.Rmd PREDICT extension and validation of the extended models.
4_PREDICT_updating_continued.Rmd PREDICT extension and validation of the extended models (calibration plots, decision curves, and net benefit).
5_Method_selection.Rmd Application of the closed test procedure on the full MINDACT dataset.
5_Method_selection_simulation.Rmd Simulation of the closed test procedure on samples of the MINDACT dataset.
pool_perf.R Pooling functions, and other functions used in the syntax.

Contact

Mary Ann E. Binuya
Netherlands Cancer Institute
m.binuya@nki.nl

Authors

Author Role Description
Mary Ann Binuya Author Development and support
Ellen Engelhardt Author Review
Terry Chan Author Development and review
Martijn Heymans Author Review
Marjanka Schmidt Author Review
Ewout Steyerberg Author Development and review

About

PREDICT-NL Methods Study (Closed Test Procedure for Updating Cox Models)

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors

Languages

  • HTML 99.0%
  • R 1.0%