Date of Award

January 2023

Document Type

Open Access Thesis

Degree Name

Master of Public Health (MPH)

Department

School of Public Health

First Advisor

Linda Niccolai

Second Advisor

Daniel M. Weinberger

Abstract

The social vulnerability index (SVI) measure was developed by the Centers for Disease Control and Prevention to characterize a community’s vulnerability level. Recently, SVI has been used to predict risk of COVID-19, in order to understand which communities may be most vulnerable. This analysis aims to understand what factors are driving the relationship between census tract-level vulnerability and risk of hospitalization with a viral respiratory pathogen. A statistical analysis and modeling study of surveillance data was conducted, analyzing the incidence of hospitalization with influenza, respiratory syncytial virus (RSV), and COVID-19, by census tract, in two counties in Connecticut between 2018–2021. Included cases were geocoded according to their address of residence. Both a simple and penalized ridge regression model were used to predict case counts for each of the pathogen-years. The relative Root Mean Squared Error (rRMSE) was used to assess accuracy of model predictions, using a 10% holdout sample. The model that most accurately predicted case numbers, and corresponding coefficients, for each pathogen were determined. Overall, all models gave relatively poor rRMSE values (>0.3), and there was marginal improvement across regression approaches within pathogens, making it challenging to make definitive conclusions. The coefficients varied, though in five of the six pathogen-years, the proportion of single-parent households variable, and in four, the proportion living in group quarters, and proportion over 65 years variables, contributed to the final model. The variation in coefficients could indicate that a single index may not be a “one-size-fits-all” approach for all viral respiratory pathogens. This analysis would benefit from additional data, from other EIP sites, and an analysis of the fitted and uncertainty distributions, to correct for poor prediction accuracy.

Comments

This is an Open Access Thesis.

Download
Open Access

This Article is Open Access

Share

COinS