Towards the Development of Machine Learning Models for Improving Health Outcomes for Patients with Sepsis Restricted; Files Only

Abstract

Sepsis is a major global health emergency affecting many adults and children worldwide. Clinicians have developed definitions that attempt to effectively characterize the symptoms of sepsis, but the diversity in the appearance of symptoms for this syndrome has resulted in many of these definitions suffering from certain shortcomings. More specifically, the current definitions may not detect patients who develop sepsis, or they may overestimate sepsis in sterile patients. Additionally, for the sepsis patients that are correctly recognized, these criteria may struggle to determine which of these patients are at risk for further health deterioration or mortality. Furthermore, these definitions lack information from nonclinical factors such as the social determinants of health, and adding this information may allow these criteria to better determine which sepsis patients are more susceptible to poor health outcomes. Fortunately, machine learning methods can be leveraged to mitigate these drawbacks. To address the problem of the fluid nature of the syndromic conditions of sepsis, we explore active learning methods to develop robust prediction models for sepsis onset. In regards to determining which sepsis patients are susceptible to poor outcomes, we develop machine learning mortality prediction models using physiologic features to improve upon the pediatric sepsis screening work of the novel Phoenix sepsis criteria. To address the lack of potentially important social and environmental determinants of health within existing criteria, we investigate whether the incorporation of variables from the Child Opportunity Index prove useful in determining poor health outcomes for children with suspected infection under the Phoenix criteria.

Table of Contents

1 Introduction

1.1 Background

1.2 Limitations

1.3 Contributions

1.4 Dissertation Outline

2 ALRT: An Active Learning Framework for Irregularly Sampled Clinical Data

2.1 Introduction

2.2 Related Work

2.2.1 Sepsis Prediction

2.2.2 Active Learning

2.3 Materials and Methods

2.3.1 Active Learning

2.3.2 Uncertainty Sampling

2.4 Experiments

2.4.1 Dataset

2.4.2 Cohort Definition

2.4.3 Outcome

2.4.4 Data Preprocessing

2.4.5 Modeling

2.5 Results

2.5.1 Clinical and Demographic Characteristics of the Study Population

2.5.2 Model Results

2.6 Discussion

2.6.1 Model Performance

2.6.2 Comparison of Label Acquisitions by Uncertainty Sampling Methods

2.6.3 Implications

2.6.4 Limitations

2.7 Conclusion

3 Prognostic Accuracy of Machine Learning Models for In-Hospital Mortality Among Children With Phoenix Sepsis Admitted to the Pediatric Intensive Care Unit

3.1 Introduction

3.2 Material and Methods

3.2.1 Ethics Statement

3.2.2 Study Design, Setting, and Population

3.2.3 Outcomes, Definitions, and Main Measures

3.2.4 Data Preprocessing and Aggregation

3.2.5 Machine Learning Prediction Models

3.2.6 Model Interpretability

3.3 Results

3.3.1 Cohort Demographic and Clinical Characteristics

3.3.2 Model Performance for In-hospital Mortality

3.3.3 Important Model Features and Interpretation for In-hospital Mortality

3.3.4 Model Performance for In-hospital Mortality or PICU Lenth of Stay >= 72 hours

3.3.5 Feature Sensitivity Analysis

3.4 Discussion

3.4.1 Limitations

3.5 Conclusion

4 Association of the Child Opportunity Index with In-hospital Mortality and Persistence of Organ Dysfunction at One Week After Onset of Phoenix Sepsis Among Children Admitted to the Pediatric Intensive Care Unit with Suspected Infection

4.1 Introduction

4.2 Methods

4.2.1 Ethics Statement

4.2.2 Study Design, Setting, and Population

4.2.3 Geocoding

4.2.4 Outcomes, Definitions, and Main Measures

4.2.5 Data Preprocessing and Aggregation

4.2.6 Models

4.2.7 Model Interpretability

4.3 Results

4.3.1 Cohort Demographic and Clinical Characteristics

4.3.2 Child Opportunity Index Population Differences by Hospital Site

4.3.3 Model Performance

4.3.4 Important Model Features

4.3.5 Fairness Analysis

4.4 Discussion

4.4.1 Limitations

4.5 Conclusion

5 Conclusion and Future Work

5.1 Conclusion

5.2 Future Work

5.2.1 Domain Adaptation Across Hospital Sites

5.2.2 Domain Adaptation for Critically Ill Adult and Children Patients

5.2.3 COI Indicators for Different Health Outcomes

A ALRT

A.1 Supplemental Tables

A.2 Supplemental Figures

B Phoenix Mortality

B.1 Supplemental Tables

B.2 Supplemental Figures

C Phoenix COI

C.1 Supplemental Tables

C.2 Supplemental Figures

Bibliography

About this Dissertation

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	Laney Graduate School
Department	Computer Science and Informatics
Degree	Ph.D.
Submission	Dissertation
Language	English
Research Field	Biology, Bioinformatics Computer Science
关键词	active learning sepsis pediatrics social determinants of health machine learning
Committee Chair / Thesis Advisor	Rishikesan Kamaleswaran, Emory University Matthew Reyna, Emory University
Committee Members	Jocelyn Grunwell, Emory University Gari Clifford, Emory University

最新修改

Primary PDF

Thumbnail	Title	Date Uploaded	Actions
	File download under embargo until 09 January 2027	2024-10-30 13:59:24 -0400	File download under embargo until 09 January 2027

Supplemental Files

Thumbnail	Title	Date Uploaded	Actions