An Overview Table of DL-based Survival Methods

Disclaimer

The table on this GitHub Pages site has been created as part of the review paper Deep Learning For Survival Analysis: A Review using the R package reactable (Lin, 2023). It is licensed under CC BY.

This table is not guaranteed to be complete and will be extended continuously. To do so, we welcome pull requests by adding to data/raw.csv in the source repository of this site. New methods must comply with the inclusion and exclusion criteria used in the review paper and contain information for all columns in the table. We will then cross-check and merge new pull requests.

Acronyms, abbreviations, and explanations of the table columns

All of the columns of the table, as well as the corresponding classifications, are described in detail in the review paper.

  • title: Title of the article introducing the method.
  • model name: Official method name if provided in the respective paper; otherwise, the first three letters of the first author’s last name followed by the year of publication.
  • model class: See review paper for classification.
  • parametrization: See review paper for classification.
  • architecture: Which neural network architecture does the method employ?
  • left-truncation: Can the method handle left-truncated survival times?
  • interval-censoring: Can the method handle interval-censored survival times?
  • CR: Can the method handle competing risks?
  • MSM: Can the method handle multi-state modeling?
  • RE: Can the method handle recurrent events?
  • TVF: Can the method handle time-varying features?
  • TVE: Can the method estimate time-varying effects?
  • high dim: Is the method specifically designed for high-dimensional (e.g., omics) features?
  • multimodality: Can the method handle unstructured features or multiple feature modalities (e.g., images)?
  • interpretability: Are the results produced by the methods (in particular, the feature effects) interpretable?
    • classical = classical statistical interpretability of tabular data
    • BP = backpropagation
    • CIC = cumulative incidence curve
    • (P)FI = (permutation) feature importance
    • LRP = Layer-wise Relevance Propagation
    • PA = pathway activation
  • code: Is the code for the method and experiments publicly available and, if so, where? (Hyperlink to the main repository; further implementations of the method, if available, are shown by expanding the respective row.)
  • authors: Authors of the article introducing the method. (hidden)
  • year: Year of publication of the article introducing the method. (hidden)
  • paper: URL of the publication of the article (hyperlink in the title column). (hidden)
  • multiple losses: Does the method use more than one loss for training? (hidden)
  • datasets: Which accessible (i.e., public or registered access) datasets is the method evaluated on in the respective paper? (hidden)
  • min dataset size: Size of the smallest dataset used to evaluate the method in the respective paper. (hidden)
  • simulation: Is the method additionally evaluated on synthetic data? (hidden)
  • comments: Further comments regarding the method (also shown if expanding the row). (hidden)

Table