Methods for Clustered Competing Risks Data andCausal Inference using Instrumental Variables forCensored Time-to-event Data

Clustered Competing Risks;Causal Inference using Instrumental Variables for Censored Time-to-event Data;Public Health;Statistics and Numeric Data;Health Sciences;Social Sciences;Biostatistics

Dharmarajan, SaiKalbfleisch, John D ;

University of Michigan

关键词: Clustered Competing Risks;    Causal Inference using Instrumental Variables for Censored Time-to-event Data;    Public Health;    Statistics and Numeric Data;    Health Sciences;    Social Sciences;    Biostatistics;   

Others  :  https://deepblue.lib.umich.edu/bitstream/handle/2027.42/144110/shdharma_1.pdf?sequence=1&isAllowed=y

瑞士|英语

来源: The Illinois Digital Environment for Access to Learning and Scholarship

【 摘 要 】

In this dissertation, we propose new methods for analysis of clustered competing risks data (Chapters 1 and 2) and for instrumental variable (IV) analysis of univariate censored time-to-event data and competing risks data (Chapters 3 and 4).In Chapter 1, we propose estimating center effects through cause-specific proportional hazards frailty models that allow correlation among a center’s cause-specific effects. To evaluate center performance, we propose a directly standardized excess cumulative incidence (ECI) measure. We apply our methods to evaluate Organ Procurement Organizations with respect to (i) receipt of a kidney transplant and (ii) death on the wait-list.In Chapter 2, we propose to model the effects of cluster and individual-level covariates directly on the cumulative incidence functions of each risk through a semiparametric mixture component model with cluster-specific random effects. Our model permits joint inference on all competing events and provides estimates of the effects of clustering. We apply our method to multicenter competing risks data.In Chapter 3, we turn our focus to causal inference in the censored time-to-event setting in the presence of unmeasured confounders. We develop weighted IV estimators of the complier average causal effect on the restricted mean survival time. Our method accommodates instrument-outcome confounding and covariate dependent censoring. We establish the asymptotic properties, derive easily implementable variance estimators, and apply our method to compare modalities for end stage renal disease (ESRD) patients using national registry data.In Chapter 4, we develop IV analysis methods for competing risks data. Our method permits simultaneous inference of exposure effects on the absolute risk of all competing events and accommodates exposure dependent censoring. We apply the methods to compare dialytic modalities for ESRD patients with respect to risk of death from (i) cardiovascular diseases and (ii) other causes.