Algorithm Selection for Estimating Causal Effects: An example using the Nulliparous Pregnancy Outcomes Study: Monitoring Mothers to Be Restricted; Files Only

Abstract

Background: The Super Learning method has been widely used in doubly robust estimation of causal effects. It is theoretically recommended deploy the Super Learner with a large and diverse library of algorithms. However, the magnitude of the improvements has not yet been systematically evaluated in specific datasets commonly used in epidemiologic research settings.

Method: We applied the Super Learning ensemble method with the doubly robust estimators, including augmented inverse probability weighting (AIPW) and targeted minimum loss-based estimation (TMLE), to estimate the average treatment effect of high total fruit and vegetable consumption on the risk of preeclampsia in a sample of 7923 women from the nuMoM2b study. Using a reference ensemble with a diverse library of candidate algorithms, we compared the estimates under different sets of algorithms included in the Super Learner and evaluated whether they are sensitive to different library choices.

Results: The doubly robust estimators fitted with the reference Super Learner ensemble suggested ≥2.5 cups/1000 kcal of total fruit and vegetable consumption was associated with a lower risk of preeclampsia. The estimates of average treatment effect obtained with AIPW and TMLE were  -0.019 (95%CI: -0.036, -0.003) and -0.023 (95%CI: -0.038,       -0.007), respectively. Excluding any individual algorithm from the reference ensemble had little impact on the estimates of either AIPW or TMLE.

Conclusion: Building Super Learner ensembles with a large array of flexible machine learning algorithms may only yield minimal improvement in precision and accuracy of doubly robust estimation for average treatment effect in the sample of 7923 women from the nuMoM2b study. 

Table of Contents

Introduction................................................................................................................................... 1

Methods.......................................................................................................................................... 3

Results............................................................................................................................................. 9

Discussion..................................................................................................................................... 11

Tables and Figures...................................................................................................................... 15

Appendix...................................................................................................................................... 19

References.................................................................................................................................... 20

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	Rollins School of Public Health
Department	Epidemiology
Subfield / Discipline	Epidemiology - MPH & MSPH
Degree	M.S.P.H.
Submission	Master's Thesis
Language	English
Research Field	Health Sciences, Epidemiology Health Sciences, Public Health
Palavra-chave	machine learning average treatment effect preeclampsia doubly robust estimator Super Learning
Committee Chair / Thesis Advisor	Ashley I. Naimi, Emory University

Última modificação

Primary PDF

Thumbnail	Title	Date Uploaded	Actions
	File download under embargo until 20 May 2026	2024-04-24 13:03:59 -0400	File download under embargo until 20 May 2026

Supplemental Files

Thumbnail	Title	Date Uploaded	Actions