Geographic and statistic stability of deprivation aggregated measures at different spatial units in health research

• Published in: Applied geography (Sevenoaks) Vol. 95; pp. 9 - 18
• Main Authors: Cebrecos, Alba, Domínguez-Berjón, M. Felícitas, Duque, Ignacio, Franco, Manuel, Escobar, Francisco
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01-06-2018
• Subjects: Cardiovascular disease; Geographic scale; Modifiable areal unit problem; Socioeconomic factors; Cardiovascular disease; Modifiable areal unit problem; Geographic scale; Socioeconomic factors
• ISSN: 0143-6228; 1873-7730
• DOI: 10.1016/j.apgeog.2018.04.001

Deprivation indices constitute a valuable tool for assessing health inequalities. A key issue when analyzing deprivation is the choice of the geographical scale and spatial unit of analysis. Our objective was to evaluate statistical and geographical stability of an Area Based Deprivation Index (ABDI) computed at different spatial scales and to study their relation with cardiovascular disease.
The present study has been conducted in the city of Madrid, Spain. Madrid divides its territory in three different administrative units nested within each other: census section, neighborhoods and districts. For each unit a deprivation index was calculated through Principal Component Analysis (PCA). The data source was the 2011 national census from where a range of socioeconomic and demographic indicators were selected. To study statistical and geographical stability of deprivation we used an Exploratory Spatial Data Analysis and bivariate Local Indicators of Spatial Association analysis. We also conducted Pearson correlation analyses to study the change in the relationship between deprivation and the prevalence of cardiovascular disease (CVD) across the three scales.
At census section and neighborhood level, first component showed four and five factors loading higher than 0.6, respectively. These factors loading related to occupancy/labor market and education. However at district level, first component showed seven factors loading higher than 0.6 and related to occupancy/labor market, education and immigration. With indicators of these factors loading, deprivation indices were calculated for each administrative unit by extracting a single PCA axis. Variance explained for each index was 65%, 86% and 79%, respectively. Bivariate local autocorrelation analyses showed aggregated areas of low and high stability with variable degree of significance in the three scales. The ABDIs calculated at census section level, neighborhood level and district level presented different significant correlations with CVD prevalence (r = 0.328; r = 0.635; and r = 0.739 respectively). These results show that the deprivation index did not remain stable across the three scales, neither were the correlations between deprivation and age-adjusted CVD prevalence.
Understanding the stability of a spatial phenomenon across different scales is essential to determine the best unit of aggregation of data when studying an important process such as socioeconomic deprivation and its possible health impacts.
•We examined the scale choice when studying deprivation and cardiovascular disease.•Findings demonstrated deprivation measures differed across three different scales.•Scale affected correlations between deprivation and cardiovascular disease.•Researchers should consider statistical implications when choosing analysis scale.