Advances in Statistical Methods for Learning Individualized Treatments

Yizhe Xu Chair
University of Utah
 
Tuesday, Aug 6: 10:30 AM - 12:20 PM
5095 
Contributed Papers 
Oregon Convention Center 
Room: CC-E147 

Main Sponsor

Biometrics Section

Presentations

Bayesian outcome weighted learning for estimating sparse optimal individualized treatment rules

One of the primary goals of statistical precision medicine is to learn optimal individualized treatment rules (ITRs). The classification-based approach to estimating optimal ITRs was first introduced in outcome-weighted learning (OWL), which recasts the optimal ITR learning problem into a weighted classification problem. In this presentation, we introduce a Bayesian formulation of OWL. Starting from the OWL objective function, we generate an empirical likelihood that can be expressed as a scale mixture of normal distributions. A Gibbs sampling algorithm is developed for estimation. In addition to providing a strategy for learning an optimal ITR, Bayesian OWL provides a natural, probabilistic approach to estimate uncertainty in ITR treatment recommendations themselves. We demonstrate our method through several simulation studies and a healthcare application. 

Keywords

Precision medicine

Individualized treatment rule

Bayesian machine learning 

View Abstract 2011

Co-Author

Nikki Freeman, Duke University

First Author

Sophia Yazzourh, Universite de Toulouse

Presenting Author

Nikki Freeman, Duke University

Causal Inference on Sequential Treatments via Tensor Completion

Marginal Structural Models (MSMs) are popular for causal inference of sequential treatments in longitudinal observational studies, which however are sensitive to model misspecification. To achieve flexible modeling, we envision the potential outcomes to form a three-dimensional tensor indexed by subject, time, and treatment regime and propose a tensorized history-restricted MSM. The semi-parametric tensor factor model allows us to leverage the underlying low-rank structure of the potential outcomes tensor and exploit the pre-treatment covariate information to recover the counterfactual outcomes. We incorporate the inverse probability of treatment weighting in the loss function for tensor completion to adjust for time-varying confounding. Theoretically, a non-asymptotic upper bound on the Frobenius norm error for the proposed estimator is provided. Empirically, simulation studies show that the proposed tensor completion approach outperforms the parametric HRMSM and existing matrix/tensor completion methods. Finally, we illustrate the practical utility of the proposed approach to study the effect of ventilation on organ dysfunction from the Medical Information Mart for Intensive Care database. 

Keywords

Gradient descent

Penalized estimation

Tucker decomposition

Non-asymptotic error 

View Abstract 2851

Co-Author(s)

Shu Yang, North Carolina State University, Department of Statistics
Anru Zhang, Duke University

First Author

Chenyin Gao, North Carolina State University

Presenting Author

Chenyin Gao, North Carolina State University

Estimating optimal individualized treatment regimes for survival outcomes in competing risk data

For more than a decade, the concept of using precision medicine (PM) to determine a patient's optimal treatment has gained popularity over the traditional "one-size-fits-all" treatment assignment based on covariate subgroup treatment effects. Extensive methodology for estimating individualized treatment regimes (ITRs) has been developed to account for individual heterogeneity. Although PM for survival data has become more abundant in recent years, there is less focus on estimating ITRs in the presence of competing risks (CR). CR are events where their occurrence precludes the occurrence of other events, and not accounting for them can lead to biased results. Because CR are prevalent in healthcare settings, we extend and develop nonparametric ITR estimation methodology using random survival forests into the CR setting. We propose a two-phase method that accounts for both overall survival of all events as well as cumulative incidence of the main event of interest. Simulation studies show that our proposed method works well, and we apply the proposed method to a cohort of peripheral artery disease patients. 

Keywords

Precision medicine

Individualized treatment rule

Competing risk

Survival analysis

Random forest 

View Abstract 3443

Co-Author(s)

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

First Author

Christina Zhou

Presenting Author

Christina Zhou

Nonparametric Bayesian Q-learning for optimization of dynamic treatment regimes

Existing methods for estimation of dynamic treatment regimes are limited to intention-to-treat analyses--which estimate the effect of randomization to a particular treatment regime without considering the compliance behavior of patients. In this article, we propose a novel nonparametric Bayesian Q-learning approach to construct optimal sequential treatment regimes that adjust for partial compliance. We consider the popular potential compliance framework, where some potential compliances are latent and need to be imputed. The key challenge is learning the joint distribution of the potential compliances, which we accomplish using a Dirichlet process mixture model. Our approach provides two kinds of treatment regimes: (1) conditional regimes that depend on the potential compliance values; and (2) marginal regimes where the potential compliances are marginalized. Extensive simulation studies highlight the usefulness of our method compared to intention-to-treat analyses. We apply our method on the Adaptive Treatment for Alcohol and Cocaine Dependence Study (ENGAGE), where the goal is to construct optimal treatment regimes to engage patients in therapy. 

Keywords

Dirichlet process mixture

Endogeneity

Gaussian copula

Intention-to-treat analysis 

View Abstract 3029

Co-Author(s)

Ashkan Ertefaie, University of Rochester
Kevin Lynch, University of Pennsylvania
James McKay, University of Pennsylvania
Brent Johnson, University of Rochester-Medical Center

First Author

Indrabati Bhattacharya, Florida State University

Presenting Author

Indrabati Bhattacharya, Florida State University

Personalized Treatment Selection for Multivariate Ordinal Scale Outcomes and Multiple Treatments

We present an innovative approach for tailoring treatment selection on an individualized basis in the presence of correlated multiple responses, particularly those measured on ordinal scales, including binary responses. Our methodology involves the utilization of rank lists for treatments, generated from probabilities of observing responses of higher order than each level of the ordinal outcome, conditional on patient covariate measurements. We introduce a rank aggregation technique designed to amalgamate multiple lists of ranks, allowing for correlations both within these lists and among elements within each list. Our approach is versatile, accommodating any number of treatments and responses, and is applicable across a wide range of models. Our method offers flexibility by allowing the integration of response weights, enabling customization based on patient and clinician preferences on an individual case basis for optimal treatment decisions. To evaluate the performance of our proposed method in finite samples, we conducted a simulation study. Furthermore, we provide two illustrative examples using data from clinical trials on Cystic Fibrosis and Alzheimer's Disease. 

Keywords

Personalized Treatments

Semiparametric Regression

Rank Aggregation

Ordinal Responses 

View Abstract 2221

Co-Author(s)

Bakeerathan Gunaratnam, University of Louisville
K.B. Kulasekera, University of Louisville

First Author

Chathura Siriwardhana, University of Hawaii

Presenting Author

Chathura Siriwardhana, University of Hawaii

Tensor-Based Individualized Treatment Rules for Neuroimaging Applications

Precision medicine aims to uncover the optimal personalized treatment plan, offering thoughtful decision support based on the characteristics of each patient. With the rapid advancement of medical imaging technology, integrating high-order patient-specific imaging features into individualized treatment rules has become critical. We introduce a novel, data-driven approach that utilizes both imaging data and additional variables to guide the selection of the best treatment options. Specifically, this study employs tensor and scalar covariates within a regression framework, estimating optimal individualized treatment rules through Tucker decomposition. This method effectively reduces the number of parameters, leading to efficient estimation and feasible computation. For handling high-dimensional tensors, we further employ sparse Tucker decomposition to reduce the parameter number. Additionally, we develop an alternating updating algorithm that incorporates an Alternating Direction Method of Multipliers. Under certain conditions, we show the asymptotic properties of these estimators. The numerical performance of our method is validated through simulations and high-order medical imaging data applications.  

Keywords

individualized treatment rules

tensor regression

imaging data 

Abstracts


Co-Author(s)

Ting Li
Yuanying Chen
Yang Bai, Shanghai University of Finance and Ecnomics

First Author

yang sui

Presenting Author

yang sui