cmclean5 / PublicHealthModels

R implementation of popular ML models for health care data
0 stars 0 forks source link

A Proper Concordance Index for Time-Varying Risk #28

Open BBolosSierra opened 1 year ago

BBolosSierra commented 1 year ago

TL;DR

This paper proposes a concordance index for time-varying risks

Paper Link

https://doi.org/10.48550/arXiv.2208.03213

Author/Institution

A Gandy: Department of Mathematics, Imperial College London, U.K. T Matcham: Department of Mathematics, Imperial College London, U.K. and NIHR ARC Northwest London, U.K.

Overview

The paper suggests using the hazard rate as the time-varying risk score when calculating concordance. Through simulations it is demonstrated situations in which other concordance indices can lead to incorrect models being selected over a true model, justifying the use of the risk prediction in both model selection and loss functions in, e.g., deep learning models

Contributions and Distinctions from Previous Works

Methods

Survival analysis

Results

Cite

Comments

Chapter 6 uses DeepHit code from https://github.com/chl8856/DeepHit with changes made to also include their concordance index in the loss and validation.

https://github.com/tmatcham/CrossingHazardConcordance