Targeted Maximum Likelihood Estimation to Evaluate Effect of Novel Regimens on Multidrug Resistant Tuberculosis Público

Zhao, Yuan (Spring 2019)

Permanent URL: https://etd.library.emory.edu/concern/etds/b5644s786?locale=pt-BR
Published

Abstract

Introduction Multidrug resistant tuberculosis (MDR-TB) is a growing threat to public health and the cure rate of MDR-TB in the real world is still low. We applied super learning and targeted maximum likelihood estimation (TMLE) techniques to estimate the effect of novel regimens of bedaquiline and delamanid on MDR-TB patients using data from a small observational study.

Methods The study included a total of 100 MDR-TB patients from Georgia and the two primary outcomes were sputum culture conversion (SCC) within 2 and 6 months. We assessed the applicability of TMLE with super learner for estimating effects in this setting via simulation. The best-performing estimators from the simulation study were then applied to compare the rates of 2- and 6-month SCC for bedaquiline- and delamanid-based regimens.

Results All estimators had relatively good performance in the simulation study with low mean squared error (<0.015) and near nominal 95% confidence interval coverage (90%-95%). Our analysis showed that the bedaquiline-based regimen had a higher culture conversion rate than the delaminid-based regimen with an estimated difference in probability of SCC of 0.199 (95%CI -0.007, 0.405; p-value 0.059) at 2 months and 0.187 (95%CI 0.050, 0.324; p-value 0.007) at 6 months.

Discussion Our results indicate that bedaquiline-based regimens are associated with better sputum culture conversion rate within 2 months and 6 months than delamanid-based regimens, supporting the inclusion of bedaquiline in routine MDR tuberculosis regimens. We also demonstrated that TMLE with the super learner method is advantageous in causal estimation in settings of observational studies with small sample sizes. Future studies can focus on generalizing the conclusion using additional simulation data sets and fine tuning the hyperparameters of machine learning algorithms inside super learner by adding another layer of cross-validation to ensure that super learner always selects optimal algorithm combinations.

Key words: Multidrug-resistant TB, causal inference, super learner, targeted maximum likelihood estimation (TMLE)

Table of Contents

INTRODUCTION ....................................................................................................................................... 1

METHODS .................................................................................................................................................. 5

STUDY DESIGN ........................................................................................................................................... 5

CAUSAL ESTIMATION METHODS ............................................................................................................... 6

SIMULATION STUDY ............................................................................................................................ 10

ANALYSIS OF MDR-TB DATA ............................................................................................................ 12

DISCUSSION ............................................................................................................................................ 13

REFERENCE ............................................................................................................................................ 18

About this Master's Thesis

Rights statement
  • Permission granted by the author to include this thesis or dissertation in this repository. All rights reserved by the author. Please contact the author for information regarding the reproduction and use of this thesis or dissertation.
School
Department
Subfield / Discipline
Degree
Submission
Language
  • English
Research Field
Palavra-chave
Committee Chair / Thesis Advisor
Última modificação

Primary PDF

Supplemental Files