Degree
Doctor of Philosophy (PhD)
Department
Mechanical and Industrial Engineering
Document Type
Dissertation
Abstract
Sepsis, a life-threatening organ failure caused by the body’s abnormal and harmful response to an infection, remains a major global and national health challenge. In the United States, it contributes significantly to hospital mortality and imposes a substantial economic burden, with billions spent annually on treatment. Early diagnosis is difficult due to the nonspecific nature of sepsis symptoms, often leading to delays in care and poorer outcomes. As the incidence and severity of sepsis rise, identifying factors that influence outcomes is increasingly critical. Emerging research has shown that Social Determinants of Health (SDOH), including economic stability, education, neighborhood and built environment, social and community context, and access to health and healthcare, play a significant role in sepsis outcomes and may contribute to disparities in care.
This research aims to uncover how SDOH factors influence sepsis outcomes, including in-hospital mortality, 30-day readmission, and length of stay (LOS), and to evaluate the utility of predictive modeling that incorporates SDOH factors in identifying high-risk patients. The data included patient demographics, clinical data, community-level SDOH indicators, responses from an individual-level SDOH screening survey, and patient outcomes. Findings show that higher social vulnerability, as measured by the Social Vulnerability Index (SVI), and greater distance to urgent care facilities are significantly associated with increased LOS. Multivariate analysis further shows that this is more pronounced among male patients. Age was found to mediate the effect of SVI on sepsis outcomes by serving as a protective factor for mortality, LOS, and readmission. In contrast, chronic health conditions do not significantly mediate this relationship. Although the developed models demonstrated a limited ability to identify mortality and readmission cases accurately, Deep Neural Networks (DNNs) achieved superior performance, particularly in terms of F1 score, compared to Random Forest and Logistic Regression in the classification tasks.
This research makes several key contributions to the understanding of SDOH in sepsis care. It identifies geographic proximity to healthcare facilities as a novel and significant predictor of sepsis outcomes, particularly in-hospital mortality and LOS. The study also explores mediation analysis, revealing that age mediates the relationship between SDOH and sepsis outcomes. Furthermore, it explores the integration of community-level SDOH factors into machine learning models for sepsis outcome prediction, offering insights into how structural and spatial factors can enhance risk stratification in sepsis care.
Date
7-16-2025
Recommended Citation
Nyelele, Mwembezi Aaron, "Retrospective Study and Predictive Modeling of the Impact of Social Determinants of Health on Sepsis Outcomes" (2025). LSU Doctoral Dissertations. 6870.
https://repository.lsu.edu/gradschool_dissertations/6870
Committee Chair
Knapp, Gerald M.
DOI
10.31390/gradschool_dissertations.6870