A Density-Weighted Cox Model for Outlier-Robust Prediction of Prostate Cancer Survival
This package, DWCox, implements a density-weighted Cox regression model that is more robust against outliers in the training data. DWCox gives more accurate predictions than the standard Cox regression on prostate cancer survival, especially in cases where the training data are expected to contain a lot of outliers. More details can be found in our paper (coming soon).
You are welcome to use this package, subject to the license described in LICENSE.txt. Also, please cite this paper: (Paper coming soon)
Simply execute run.sh.
If you see error messages, please make sure you have R and these R packages installed: mice, survival, glmnet, pdfCluster.
DWCox is ready to run right after being downloaded. The required data files have been simulated and put in data/. The outputs will be data/pred_risk.csv and data/pred_days.csv. You can replace data/train.csv and data/task.csv with your own data files to let DWCox do your jobs.
DWCox was originally developed to predict the risk scores and expected survival time of prostate cancer patients from their clinical records at the beginning of their trails. Fortunately, since the pipeline of DWCox is completely automatic, you can easily apply it directly to other survival analysis problems, as long as you prepare your input data with the correct formatting.
Please prepare 2 csv files whose formats are similar to the two in data/. They should have the same number of columns (not necessarily 22) containing real numbers or nothing. data/train.csv contains information about subjects (or patients) used to train DWCox, while data/task.csv contains information about subjects whose survival is to be predicted by DWCox. Each row gives all information about a subject. The first column contains the last observed survival time in days (or any unit you prefer). The second column contains either 1 or 0: 1 means the subject died (or the event happened) at the last observed survival time, while 0 means the subject remained alive (or the event had not happened) till the last observed survival time but no information about that subject is available after that time. The headers of the 1st and 2nd columns should be t and ei respectively. The remaining columns are features (clinical variables, or any features you prefer) that will be used to model the survival. There should exist no missing values in the first 2 columns of data/train.csv.
Enjoy!