Machine learning for estimating and comparing clinical rules for treating diarrheal illness with antibiotics Restricted; Files Only

Abstract

Acute diarrheal disease is one of the leading causes of death in children under age 5, disproportionately impacting children in low-resource settings.  While many of these cases could be treated with antibiotics, current World Health Organization guidelines are narrow and often miss cases that would benefit from treatment.  In this study, we propose a framework for the creation of optimal treatment rules (OTRs) that leverage available information to recommend treatment to individuals who have a clinically relevant causal effect.  We use a nested cross validation procedure that makes use of Super Learning to estimate a doubly-robust (DR)-learner and recommend treatment to patients whose conditional average treatment effect (CATE) meets a clinically relevant threshold. We use an augmented inverse probability of treatment weighted estimator (AIPTW) to estimate the proportion treated, the Average Treatment effect under the Rule (ATR_d), and the Average Treatment effect among those Recommended Treatment under the rule (ATRT_d). Our method has low bias, 95% coverage, and consistent standard error under rules based on a single, binary covariate. It also has low bias and 95% coverage of data-adaptive parameters under more complex rules.  We apply our method to a real-data analysis of the AntiBiotics for Children with severe Diarrhea (ABCD) trial and demonstrate the importance of point-of-care diagnostics in the creation of OTRs. Overall, the creation of OTRs can help public health officials leverage information effectively and make treatment recommendations that will have the greatest impact on the population. 

Table of Contents

1 Introduction

2 Methods

2.1 Notation

2.2 Parameters of Interest

2.2.1 Conditional Average Treatment Effect (CATE)

2.2.2 Optimal Treatment Rule (OTR)

2.2.3 Effect Estimands

2.3 Identification

2.4 Estimation

2.4.1 General Approach

2.4.2 Estimating the CATE and treatment rule

2.4.3 Augmented Inverse Probability of Treatment Weight (AIPTW) Estimator for Effect Estimates

2.4.4 Averaging over cross-validation folds

3 Simulation Study

3.1 Data Generation

3.2 Procedure

3.3 Results

3.3.1 Data Adaptive Results

4 Data Analysis

4.1 Simple Rule

4.2 Complex Rules

5 Discussion

Bibliography

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	Biostatistics
Degree	M.S.P.H.
Submission	Master's Thesis
Language	English
Research Field	Statistics
Mot-clé	machine learning optimal treatment rules bacterial diarrhea Super Learner causal inference
Committee Chair / Thesis Advisor	David Benkeser, Emory University
Committee Members	Elizabeth Rogawski McQuade, Emory University

Dernière modification

Primary PDF

Thumbnail	Title	Date Uploaded	Actions
	File download under embargo until 20 May 2026	2024-04-08 00:11:01 -0400	File download under embargo until 20 May 2026

Supplemental Files

Thumbnail	Title	Date Uploaded	Actions