Tuning parameter calibration for personalized prediction in medicine

Shih-Ting Huang,Kristoffer H Hellton,Yannick Düren,Johannes Lederer

doi:10.1214/21-ejs1884

Shih-Ting Huang, Kristoffer H Hellton + Show 2 more

Open Access

https://doi.org/10.1214/21-ejs1884

Copy DOI

Abstract

Personalized prediction is of high interest in medicine; potential applications include the prediction of individual drug responses or risks of complications. But typical statistical pipelines such as ridge estimation combined with cross-validation ignore the heterogeneity among the patients and, therefore, are not suited for personalized prediction. We, therefore, introduce an alternative ridge-type pipeline that can minimize the prediction error of each patient individually. We show that our pipeline is optimal in terms of oracle inequalities, fast, and highly effective both in simulations and on real data.

Highlights

In the last decade, improvements in genomic, transcriptomic, and proteomic technologies have enabled personalized medicine, or precision medicine, to become an essential part of contemporary medicine
Personalized medicine takes into account individual variability in genes, proteins, environment, and lifestyle to decide on disease prevention and treatment (Hamburg and Collins 2010)
Epigenomic, and transcriptomic data used in personalized medicine, such as gene expression, copy number variants, or methylation levels are often high-dimensional with a number of variables that rivals or exceeds the number of observations

Summary

Introduction

Improvements in genomic, transcriptomic, and proteomic technologies have enabled personalized medicine, or precision medicine, to become an essential part of contemporary medicine. Epigenomic, and transcriptomic data used in personalized medicine, such as gene expression, copy number variants, or methylation levels are often high-dimensional with a number of variables that rivals or exceeds the number of observations. Using such data to estimate and predict treatment response or risk of complications, requires regularization typically by the 1 norm (lasso), the 2 norm (ridge), or other terms. The averaging removes the inherent individual heterogeneity of the patients and, results in sub-optimal prediction performance for the individual patients This may lead to unsuitable treatment, administration of improper medication with adverse side effects, or lack of preventive care (Hamburg and Collins 2010)

Objectives

Methods

Findings

Conclusion