Heterogeneous treatment effect estimation using instrumental variable in time-to-event data
Principal Investigator
Name
Uri Shalit
Degrees
Ph.D.
Institution
Technion - Israel Institute of Technology
Position Title
Professor, Data and Decisions Sciences Department, Technion
Email
About this CDAS Project
Study
HIPB
(Learn more about this study)
Project ID
HIPB-13
Initial CDAS Request Approval
May 13, 2024
Title
Heterogeneous treatment effect estimation using instrumental variable in time-to-event data
Summary
Treatment effect estimation lies at the core of the decision-making process. Evaluating the impact of interventions, policies, or treatments enables selecting the most effective strategies to improve outcomes for the targeted population. Many studies concentrate on estimating average treatment effects (ATEs) across populations. However, ATEs provide only a summary of the intervention effect, which may vary at the subgroup level. In this work, we focus on heterogeneous treatment effects (HTEs) estimation, which accounts for variations in responses among individuals or small homogeneous subgroups. Estimating HTEs can significantly enhance decision-making by enabling the selection of the most effective strategies for each subgroup. For instance, in medical applications, understanding HTEs helps to account for the complex interactions among patient characteristics and treatment responses, and is crucial for optimizing clinical outcomes.
A common challenge is that the full set of characteristics that we need to account for is rarely accessible, leading to unmeasured confounding that introduces bias into our HTE estimates. To address this, we utilize the instrumental variable (IV) framework for the identification and estimation of HTEs. Specifically, we focus on time-to-event outcomes. Time-to-event analysis, or survival analysis, is used when the outcome of interest is the time until a pre-specified event occurs. Such data may be incomplete for some samples, known only to be event-free until a certain time point, with missing subsequent follow-up (right censoring).
In this work we introduce, to the best of our knowledge, the first algorithm for the evaluation of HTEs, with time-to-event outcome and right-censored data, which utilizes the IV approach to account for possible unobserved confounding. We evaluate its performance through a comprehensive simulation study, and we demonstrate a real-world application of the algorithm using the data from the NIH's Health Insurance Plan (HIP) Breast Cancer Screening Trial Project, a randomized trial of mammography screening.
A common challenge is that the full set of characteristics that we need to account for is rarely accessible, leading to unmeasured confounding that introduces bias into our HTE estimates. To address this, we utilize the instrumental variable (IV) framework for the identification and estimation of HTEs. Specifically, we focus on time-to-event outcomes. Time-to-event analysis, or survival analysis, is used when the outcome of interest is the time until a pre-specified event occurs. Such data may be incomplete for some samples, known only to be event-free until a certain time point, with missing subsequent follow-up (right censoring).
In this work we introduce, to the best of our knowledge, the first algorithm for the evaluation of HTEs, with time-to-event outcome and right-censored data, which utilizes the IV approach to account for possible unobserved confounding. We evaluate its performance through a comprehensive simulation study, and we demonstrate a real-world application of the algorithm using the data from the NIH's Health Insurance Plan (HIP) Breast Cancer Screening Trial Project, a randomized trial of mammography screening.
Aims
Heterogeneous treatment effect estimation using instrumental variable in time-to-event data.
Collaborators
Tomer Meir, Technion.
Prof. Uri Shalit, Technion.
Prof. Malka Gorfine, Tel Aviv University.