Precision Medicine, Prediction, and Clinical Trial with Time-to-Event Outcomes

Yen Chang Chair
The University of North Carolina at Chapel Hill
 
Wednesday, Aug 6: 10:30 AM - 12:20 PM
4184 
Contributed Papers 
Music City Center 
Room: CC-210 

Main Sponsor

Lifetime Data Science Section

Co Sponsors

Lifetime Data Science Section

Presentations

Conformal Two Sample Conditional Tests for Survival Comparison

Survival comparison plays an essential and important role in medical studies and others and has been discussed extensively. However, most of the existing methods focus on the marginal comparison or the comparison of marginal survival functions. In this paper, we consider the comparison of two conditional survival functions given predictors, and two nonparametric conformal test procedures are proposed, one for regular cases and one for high-dimensional situations. In particular, the proposed methods make use of the conformity score and the inverse probability weighting technique along with the U-statistics theory, and two weighted rank-sum test statistics are developed. The asymptotic properties of the proposed method are established, and an extensive simulation study is conducted to evaluate their performances and indicates that they work well in practical situations. n addition, the proposed test procedures are applied to an AIDS study. 

Keywords

Conformal inference

Cox additive model

Survival comparison

U-statistics 

Co-Author

Jianguo Sun, University of Missouri

First Author

Yuxiang Wu

Presenting Author

Yuxiang Wu

Machine learning algorithms for time to event data analysis

Survival analysis is a subfield of statistics to analyze survival model with time to event data. In the literature, there are several statistical approaches have been widely developed to study the time to event data in survival analysis. In this paper, we consider traditional survival models such as Cox-proportional hazard model, accelerated failure time model and mixture cure model. The mixture cure model deal with population that consists with susceptible and unsusceptible individuals. Machine learning algorithms have been applied in the field of survival analysis to deal with more complex datasets and to predict the time to event outcomes. We consider random survival forest, survival support vector machine, and neural network machine learning approach of survival analysis. Furthermore, we apply real datasets to compare the performance of traditional survival analysis approaches and machine learning based survival methods. 

Keywords

Survival data, right censored, survival random forest, survival support vector machine 

First Author

Durga Kutal, Augusta University

Presenting Author

Durga Kutal, Augusta University

Optimal individualized treatment regimes for survival data with competing risks

Precision medicine leverages patient heterogeneity to estimate individualized treatment regimes-formalized, data-driven approaches designed to match patients with optimal treatments. In the presence of competing events, where multiple causes of failure can occur and one cause precludes others, it is crucial to assess the risk of a main outcome of interest, such as one type of failure over another. This helps clinicians tailor interventions based on the factors driving that cause, leading to more precise treatment strategies. Currently, no precision medicine methods account for both survival and competing risk endpoints. To address this gap, we develop a nonparametric individualized treatment regime estimator. Our two-phase method accounts for overall survival from all events as well as the cumulative incidence of a main event. Additionally, we introduce a value function that jointly incorporates both outcomes. We develop random forests to construct individual survival and cumulative incidence curves. Simulation studies demonstrated that our proposed method performs well, which we applied to a cohort of peripheral artery disease patients at high risk for limb loss and mortality. 

Keywords

precision medicine

random forests

survival analysis

cumulative incidence function 

Co-Author(s)

Nikki Freeman, Duke University
Katharine McGinigle, University of North Carolina at Chapel Hill
Michael Kosorok, University of North Carolina at Chapel Hill

First Author

Christina Zhou, University of North Carolina at Chapel Hill

Presenting Author

Christina Zhou, University of North Carolina at Chapel Hill

Statistical Considerations in Bayesian Approaches to Clinical Trials: The Zoster Eye Disease Study

Clinical trials often rely on frequentist statistical methods for design, monitoring, and analysis, emphasizing type I and II errors, p-values, and confidence intervals. While this approach is widely accepted in research and regulatory frameworks, Bayesian methods provide an alternative that incorporates prior knowledge and expert judgment and more importantly characterizes the entire posterior distribution of treatment effects. By basing decisions on a "minimal clinically worthwhile benefit," Bayesian approaches enhance decision-making in clinical practice based on trial data.

We will present and discuss frequentist and Bayesian analyses in the Zoster Eye Disease Study, a multicenter, double-masked, placebo-controlled randomized trial conducted in 95 sites from 2017 to 2024. A total of 527 participants were randomized to receive 12 months of daily valacyclovir or placebo, followed for an additional 6 months, stratified by age at onset (<60 vs ≥60 years) and disease duration (<6 vs ≥6 months). Bayesian analysis provided the probability of a clinically meaningful effect and the probability of various absolute event rate differences, making results more intuitive for clinicians. 

Keywords

Bayesian analyses

Survival analysis

Clinical Trials 

Co-Author(s)

Jiyu Kim
Bennie H Jeng, Department of Ophthalmology, Scheie Eye Institute, University of Pennsylvania Perelman
Elisabeth Cohen, Department of Ophthalmology, NYU Grossman School of Medicine
Judith Hochman, Department of Medicine, NYU Grossman School of Medicine
Mengling Liu, New York University Grossman School of Medicine
Andrea B. Troxel, Department of Population Health, NYU Grossman School of Medicine

First Author

Tingfang Lee

Presenting Author

Jiyu Kim

Sparse Regularization for Tensor Covariates in the Cox Regression Model

This study investigates disease survival through medical imaging by directly incorporating the imaging data as tensor-structured covariates within the Cox regression model for right-censored survival outcomes. The objective is to estimate the coefficients of these tensor covariates to identify imaging subregions significantly associated with survival time. However, a challenge arises due to the limited sample size relative to the ultrahigh dimensionality of the imaging data. To address this, an algorithm is proposed that integrates sparse regularization into tensor decomposition, shrinking the coefficients of subregions irrelevant to survival time to zero. A comprehensive simulation study is conducted to evaluate the performance of the proposed algorithms in estimating tensor parameters. 

Keywords

Cox regression

Tensor-structured covariates

Sparse regularization

Right censoring

Piecewise smoothness 

Co-Author

Pei-Fang Su, National Cheng Kung University

First Author

Chin-Chun Chen, National Cheng Kung University

Presenting Author

Chin-Chun Chen, National Cheng Kung University