The Concordance Index Decomposition

C-index (CI) is a weighted harmonic average of the C-indices defined for the subsets ee (events vs. events) and ec (events vs. censored cases).

$\frac{1}{CI} = \alpha \frac{1}{CI_{ee}} + (1 - \alpha) \frac{1}{CI_{ec}}$

To use the C-index decompostion, download the file Utils/metrics.py and use the function c_index_decomposition. The function will return the following terms:

• Cee: The C-index of the ee pairs.
• Cec: The C-index of the ec pairs.
• alpha: The weight alpha.
• alpha_deviation: The deviation from the optimal alpha.
• C: The total C-index.

For more details, see the full paper The Concordance Index decomposition: A measure for a deeper understanding of survival prediction models

BibTeX Citation

@article{ALABDALLAH2024102781,
   title = {The Concordance Index decomposition: A measure for a deeper understanding of survival prediction models},
   journal = {Artificial Intelligence in Medicine},
   volume = {148},
   pages = {102781},
   year = {2024},
   issn = {0933-3657},
   doi = {https://doi.org/10.1016/j.artmed.2024.102781},
   url = {https://www.sciencedirect.com/science/article/pii/S093336572400023X},
   author = {Abdallah Alabdallah and Mattias Ohlsson and Sepideh Pashami and Thorsteinn Rögnvaldsson},
   keywords = {Survival analysis, Evaluation metric, Concordance Index, Variational encoder–decoder}
}

SurVED

Survival Analysis with Variational Encoder Decoder.

Reproducing the results

SurVED

• Run the file surved_final_test.py to reproduce the 100-fold test results of the four datasets.
• Run the file surved_change_censoring.py to reproduce the SurVED model results on the SUPPORT dataset with changing censoring levels.

DeepHit:

• Run the file OtherModels\DeepHit\deephit_final_test.py to reproduce the 100-fold test results of the four datasets.
• Run the file OtherModels\DeepHit\deephit_change_censoring.py to reproduce the DeepHit model results on the SUPPORT dataset with changing censoring levels.

DeepSurv:

• Run the file OtherModels\DeepSurv\deepsurv_final_test.py to reproduce the 100-fold test results of the four datasets.
• Run the file OtherModels\DeepSurv\deepsurv_change_censoring.py to reproduce the DeepSurv model results on the SUPPORT dataset with changing censoring levels.

RSF (Random Survival Forest)

• Run the file OtherModels\RSF\rsf_final_test.py to reproduce the 100-fold test results of the four datasets.
• Run the file OtherModels\RSF\rsf_change_censoring.py to reproduce the RSF model results on the SUPPORT dataset with changing censoring levels.

CPH (Cox Proportional Hazard)

• Run the file OtherModels\CPH\cph_final_test.py to reproduce the 100-fold test results of the four datasets.
• Run the file OtherModels\CPH\cph_change_censoring.py to reproduce the CPH model results on the SUPPORT dataset with changing censoring levels.

DATE:

• Run the file OtherModels\DATE\date_final_test.py to reproduce the 100-fold test results of the four datasets.

• Run the file OtherModels\DATE\date_change_censoring.py to reproduce the DATE model results on the SUPPORT dataset with changing censoring levels.

  Note: DATE repository should be downloaded from DATE_Repo and placed in the same folder

VSI:

• Run the file OtherModels\VSI\vsi_final_test.py to reproduce the 100-fold test results of the four datasets.

• Run the file OtherModels\VSI\vsi_change_censoring.py to reproduce the VSI model results on the SUPPORT dataset with changing censoring levels.

  Note: VSI repository should be downloaded from VSI_Repo and placed in the same folder

Copies of the four datasets are provided in the Data folder for convenience.