Development of a Novel Risk Prediction Model in Acute Respiratory Distress Syndrome Utilizing Pulmonary Physiologic Parameters Open Access

Abstract

Introduction: Acute Respiratory Distress Syndrome (ARDS) is a condition that develops rapidly in response to a primary critical illness and leads to respiratory failure that requires mechanical ventilation to support. It has a mortality rate of 40% and there are no direct acting pharmacologic treatments. The limited interventions available are underutilized because clinicians lack a tool to predict early pulmonary outcomes for risk stratification and therapy selection. Our aim was to create a model predicting pulmonary worsening at 48 hours with a composite outcome of death or lack of improvement in both positive end expiratory pressure and fraction of inspired oxygen in a cohort of ARDS network trial participants.

Methods: We conducted a secondary data analysis of nine randomized control trials from the ARDS network. Participants were excluded if they were in a study arm that is no longer standard of care, had incomplete data or co-enrolled in more than one of the studies. Participants were randomly divided into derivation (70%) and validation (30%) cohorts. Multivariable logistic regression with automatic backward selection on readily available clinical, demographic, and pulmonary parameters was used to derive an initial model which was then refined through various methods. The final model was assessed using the area under a receiver operating curve (AUC) in both the derivation and validation cohort.

Results: The derivation cohort had 762 participants while the validation had 334. 461 (60.5%) of the participants in the derivation cohort experienced the outcome with only 4.1% due to death in the first 48 hours. The final derived model included oxygen saturation index, driving pressure, acute hepatic failure, history of hematologic malignancy and history of chronic pulmonary disease as its covariates. Efforts to refine the model made no significant improvements. The AUC was 0.643 on the derivation cohort and 0.641 on the validation set. A probability cutpoint of 0.56 could be used for a sensitivity of 76.8% and specificity of 37.5%.

Conclusions: A predictive model was created for a novel pulmonary outcome which used only readily available clinical parameters. However, it only had modest predictability and did not meet clinically significant sensitivity and specificity thresholds.

Table of Contents

TABLE OF CONTENTS

Page

A. INTRODUCTION………………………………………………………………………………………………….1

B. BACKGROUND…………………………………………….…………………………………………….………..3

C. METHODS………………………………………………………………………………………………….………7

D. RESULTS………………………………………………………………………………………………….……..…12

E. DISCUSSION/CONCLUSIONS……………………………………………………………….………….........14

F. REFERENCES…………………………………………………………………………………………….…..……18

G. TABLES…………………………………………………………………………………………………….…..…..24

Table 1. Current Risk Stratification Models in ARDS.............................................................24

Table 2. Physiologic Parameters and their Calculation.……………….……………….………..........25

Table 3. Summary of Covariates used to build logistic model.………………….…………..............26

Table 4. Demographics of excluded datasets and Initial cohort ………………………...............….26

Table 5. Baseline Characteristics of Initial Cohort …………………………………………….........….27

Table 6. Baseline Pulmonary Characteristics of Initial Cohort…………………………............……29

Table 7. Baseline Characteristics of Derivation Cohort ………………………………..…...........……30

Table 8. Baseline Pulmonary Characteristics of Derivation Cohort………………….…….............32

Table 9. How Participants qualified for primary outcome in the derivation set……..................34

Table 10. Final Logistic Model Parameters……………………………………………………….......…..34

Table 11. Iterations of Model Refinement……………………………………………..…………......…..35

Table 12. Classification Table of the Final Model………………………….………………….........…..35

H. FIGURES……………………………………………………………………………………………………….…..36

Figure 1. Flow Diagram of Participant Selection into Final Cohorts…………….……..................36

Figure 2. Calibration Plot of Final Logistic Model……………………………………………...........…36

Figure 3. Box and Whisker Plot of Final Logistic Model……………………………………...............37

Figure 4. Receiver Operating Curves of Final Model………………..………….……………...........…38

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	Laney Graduate School
Department	Clinical Research
Degree	M.S.
Submission	Master's Thesis
Language	English
Research Field	Health Sciences, Medicine and Surgery
Keyword	Predictive Model Acute Respiratory Distress Syndrome
Committee Chair / Thesis Advisor	Amita Manatunga, PhD, Emory University
Committee Members	Jordan Kempker, MD, MSc, Emory University Ashish Mehta, MD, MSc, Emory University Greg S Martin, MD, MSc, Emory University

Last modified

Primary PDF

Thumbnail	Title	Date Uploaded	Actions
	Development of a Novel Risk Prediction Model in Acute Respiratory Distress Syndrome Utilizing Pulmonary Physiologic Parameters ()	2022-03-29 13:52:39 -0400	Press to Select an action Download

Supplemental Files

Thumbnail	Title	Date Uploaded	Actions