Modeling Pneumonia-Induced Bloodstream Infection Using Graph Theory to Estimate Hospital Mortality

Hospital-acquired pneumonia (HAP) bloodstream infections comprise a major cause of crude hospital mortality. This is a cross-sectional study that used claims data from the Centers for Medicare and Medicaid Services (<i>N</i> = 565,875). The study objective is to represent the progression...

Full description

Bibliographic Details
Main Authors: Dimitrios Zikos, Maria Athanasopoulou
Format: Article
Language:English
Published: MDPI AG 2020-05-01
Series:Technologies
Subjects:
Online Access:https://www.mdpi.com/2227-7080/8/2/24
Description
Summary:Hospital-acquired pneumonia (HAP) bloodstream infections comprise a major cause of crude hospital mortality. This is a cross-sectional study that used claims data from the Centers for Medicare and Medicaid Services (<i>N</i> = 565,875). The study objective is to represent the progression of pneumonia-induced bloodstream infections using graph theory principles, where each path of the graph represents a different scenario of bloodstream-infection progression, and aims to further estimate the likelihood if hospital death for each path. To disseminate the results, the study makes available a prototype applet to navigate various paths of the graph interactively. Bayesian probabilities were calculated for each scenario, and multivariate logistic regression was conducted to estimate the adjusted OR for inpatient death after controlling for patient age, sex, and comorbidities. The mortality rate ranged from 4.99% for patients admitted with community pneumonia without bloodstream infection and reached 63.18% for cases admitted with bloodstream infection that progressed to hospital septicemia, sepsis, and septic shock. The prototype applet can be used to unfold bloodstream infection events and their associated risk for mortality and could be used in university curricula to assist educators in helping students understand the progression of pneumonia-induced bloodstream infections in a data-driven way.
ISSN:2227-7080