DWCox: A density-weighted Cox model for outlier-robust prediction of prostate cancer survival

Xiao, Jinfeng

Permalink

https://hdl.handle.net/2142/101718 Copy

Description

Title

DWCox: A density-weighted Cox model for outlier-robust prediction of prostate cancer survival

Author(s)

Xiao, Jinfeng

Issue Date

2018-07-16

Director of Research (if dissertation) or Advisor (if thesis)

Han, Jiawei

Department of Study

Computer Science

Discipline

Computer Science

Degree Granting Institution

University of Illinois at Urbana-Champaign

Degree Name

M.S.

Degree Level

Thesis

Keyword(s)

DREAM, Prostate cancer, Cox model

Abstract

Reliable predictions on the risk and survival time of prostate cancer patients based on their clinical records can help guide their treatment and provide hints about the disease mechanism. The Cox regression is currently a commonly accepted approach for such tasks in clinical applications. More complex methods, like ensemble approaches, have the potential of reaching better prediction accuracy at the cost of increased training difficulty and worse result interpretability. Better performance on a specific data set may also be obtained by extensive manual exploration in the data space, but such developed models are subject to overfitting and usually not directly applicable to a different data set. We propose \model, a density-weighted Cox model that has improved robustness against outliers and thus can provide more accurate predictions of prostate cancer survival. \Model~assigns weights to the training data according to their local kernel density in the feature space, and incorporates those weights into the partial likelihood function. A linear regression is then used to predict the actual survival times from the predicted risks. In \challengefull, \model~obtained the best average ranking in prediction accuracy on the risk and survival time. The success of \model~is remarkable given that it is one of the smallest and most interpretable models submitted to the challenge. In simulations, \model~performed consistently better than a standard Cox model when the training data contained many sparsely distributed outliers. Although developed for prostate cancer patients, \model~can be easily re-trained and applied to other survival analysis problems. \Model~is implemented in R and can be downloaded from https://github.com/JinfengXiao/DWCox.

Graduation Semester

2018-08

Type of Resource

text

Permalink

http://hdl.handle.net/2142/101718

Copyright and License Information

Copyright 2018 Jinfeng Xiao

Owning Collections