DeepHit: A Deep Learning Approach to Survival Analysis With Competing Risks

Authors: Changhee Lee, William Zame, Jinsung Yoon, Mihaela van der Schaar

AAAI 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Comparisons with previous models on the basis of real and synthetic datasets demonstrate that Deep Hit achieves large and statistically significant performance improvements over previous state-of-the-art methods.
Researcher Affiliation Collaboration 1 Department of Electrical and Computer Engineering, University of California, Los Angeles, USA 2 Department of Economics, University of California, Los Angeles, USA 3 Department of Engineering Science, University of Oxford, UK 4 Alan Turing Institute, London, UK
Pseudocode No The paper does not contain any pseudocode or clearly labeled algorithm blocks.
Open Source Code No The paper does not provide a link or explicit statement for the open-sourcing of its own Deep Hit code. It only references a GitHub link for a baseline model (Deep Surv) in a footnote.
Open Datasets Yes The United Network for Organ Sharing (UNOS) database4 consists of patients who underwent heart transplantation in the period 1985-2015. 4https://www.unos.org/data/; The Surveillance, Epidemiology, and End Results Program (SEER)5 dataset provides information on breast cancer patients during the years 1992-2007. 5https://seer.cancer.gov/causespecific/; The Molecular Taxonomy of Breast Cancer International Consortium (METABRIC) dataset... for details see (Bilal et al. 2013); We also created a synthetic dataset with two competing risks, in the spirit of (Alaa and van der Schaar 2017).
Dataset Splits Yes For evaluation, we applied 5-fold cross validation: we randomly separated the data into training set (80%) and testing set (20%). We reserved 20% of the training set as a validation set.
Hardware Specification No The paper does not provide specific details about the hardware used for running the experiments (e.g., GPU models, CPU types, or memory specifications).
Software Dependencies No The paper only states "Deep Hit was implemented in a Tensorflow environment." without specifying the version of TensorFlow or any other software dependencies.
Experiment Setup Yes Deep Hit is a 4-layer network consisting of 1 fully-connected layer for the shared sub-network and 2 fullyconnected layers for each cause-specific sub-network and a softmax layer as the output layer. For hidden layers, the number of nodes were set as 3, 5, and 3 times of the covariate dimension for the layer 1, 2, and 3, respectively, with Re Lu activation function. The network was trained by back-propagation via Adam optimizer with a batch size of 50 and a learning rate of 10 4. Dropout probability of 0.6 and Xavier initialization was applied for all the layers (Deep Hit was implemented in a Tensorflow environment).