Dissertations and Theses

Date of Degree

6-1-2017

Document Type

Dissertation

Degree Name

Doctor of Public Health (DPH)

Department

Epidemiology and Biostatistics

Advisor(s)

Luisa N. Borrell

Committee Members

Juliana A. Maantay

Andrew R. Maroko

Lance A. Waller

Subject Categories

Community Health and Preventive Medicine | Epidemiology | Geographic Information Sciences | Public Health | Spatial Science | Statistical Methodology

Keywords

heart failure, preventable hospitalizations, neighborhood effects, geospatial analysis, multilevel modeling, spatial autocorrelation

Abstract

Background: Faced with rising medical care costs, increasing prevalence, and widening health disparities, preventing congestive heart failure (CHF) hospitalizations is a central public health concern. Despite evidence of geographical clustering in preventable CHF admissions, there is a lack of research designed to examine spatial patterning of CHF and the local area neighborhood determinants that contribute to this variability. This study sought to assess and evaluate the importance of both space and place in analyzing preventable CHF hospitalizations and readmissions by applying appropriate statistical techniques, clarifying the assumption inherent in each method, and interpreting the findings within the context of existing research. The specific objectives of the study are to:

  • Aim 1: Quantify the degree and visualize local patterns of spatial autocorrelation or clustering in preventable CHF hospitalization and readmission rates across the study area.
  • Aim 2: Assess the association of neighborhood-level demographic and socioeconomic compositional measures with preventable CHF hospitalizations and readmissions using a spatial autoregressive error model to control for residual autocorrelation.
  • Aim 3: Examine the effect of neighborhood sociodemographic composition on preventable CHF hospitalization and readmission using a Bayesian multilevel modeling approach to account for the correlation of outcomes within and between neighborhoods.

Methods: Using 2007 inpatient discharge data from New York Statewide Planning and Research Cooperative System (SPARCS), inpatient records were geocoded to the 2010 Census at the block group level using the residential address of the patient. Operationalization of preventable CHF hospitalizations was based upon a measure developed by the Agency for Healthcare Research and Quality (AHRQ). CHF unique and readmission hospitalization rates among adults were calculated and examined for spatial autocorrelation (Aim 1). Both ordinary least squares (OLS) and simultaneous autoregressive (SAR) error models were fit to determine the effect of sociodemographic area measures on CHF rates with and without accounting for spatial clustering (Aim 2). Bayesian multilevel modeling was employed to assess the relationship between neighborhood sociodemographic composition, patient-level characteristics, and preventable CHF hospitalization types while accounting for the correlation of outcomes within- and between-neighborhoods (Aim 3).

Results: Significant clustering was detected in CHF admissions indicating the presence of spatial dependence among observations. Large pockets of locally correlated high rates or hot spots were identified in areas in the south and central Bronx, northern Manhattan, and central Brooklyn. Older age composition, as well as a greater proportion of non-Hispanic (NH) black or Hispanic residents, households in poverty, and adults without a high school degree were all significant predictors of CHF hospitalization risk in regression models. However, the inclusion of a spatially lagged error term improved model fit, reduced spatial autocorrelation in residuals, and altered the strength of relationships between CHF rates and area-level factors. Similarly, hierarchical logistic model results indicate that residence in communities with a high proportion of NH black or Hispanic residents, households in poverty, and residents without a high school degree increased the odds of a unique CHF admission. For readmissions, significant area-based predictors included living in high poverty communities as well as in heavily Hispanic or NH black areas. Despite these significant associations, between-group variability in CHF hospitalizations was not meaningfully explained by the neighborhood measures included in the models.

Discussion: Considered collectively, the findings from each aim make clear that both space and place play important and independent roles in shaping preventable CHF hospitalizations across New York City (NYC). Past research is limited and statistical approaches have failed to account for multiple levels of influence and spatial dependence in assessing neighborhood effects of CHF admissions. These shortcomings may have unintentionally resulted in overemphasizing the predictive powers of neighborhood socioeconomic measures, such as poverty or income, to account for variation in preventable CHF admissions and readmissions. Furthermore, stratifying analyses by CHF admission type is necessary for evaluating differences in neighborhood variability, identifying area-based determinants, and ultimately, in targeting community approaches to prevent CHF hospitalizations and readmissions.