Article Text

Download PDFPDF

RF18 Variable selection and data reduction for the development of a small area deprivation index for health research in brazil
  1. DO Ramos1,
  2. MYT Ichihara1,2,
  3. P Rebouças1,3,
  4. FJ Oliveira1,3,
  5. A Ferreira1,3,
  6. C Teixeira3,
  7. SV Katikireddi2,4,
  8. ML Barreto1,2,3,
  9. A Leyland2,4,
  10. R Dundas2,4
  1. 1Center for Data Integration and Health Knowledge, Fiocruz Bahia, Salvador, Brazil
  2. 2on behalf of the NIHR Global Health Research Group on Social Policy and Health Inequalities
  3. 3Graduate School of Public Health, Federal University of Bahia (UFBA), Salvador, Brazil
  4. 4Social and Public Health Sciences Unit, MRC/CSO, Glasgow, UK


Background People living in areas with higher material deprivation have poorer health and increased mortality. In order to study these inequalities context-specific indicators of material deprivation are necessary. In this study, we present the first step in the development of an index to investigate the effects of spatial concentration of deprivation on health status and mortality in Brazil. Our goal was to obtain a summary indicator of deprivation in the environment surrounding each household by census tract.

Methods Neighborhood conditions at the smallest level available (census tracts) were characterized using data from the latest edition of the Brazilian Census (2010). We selected variables that measure the deprivation of the area surrounding a household. These were lack of paving, street lighting, sidewalks, presence of open sewage and accumulated waste. Confirmatory factor analysis (CFA) using a Structural Equations Model approach was performed to reduce the number of variables and test the existence of the two underlying dimensions: sanitary conditions and infrastructure. Factors were extracted as index variables through regression scores and classified in population quintiles, as categories of deprivation intensity. QGis and ArcGIS were used to plot these deprivation factors on a map for face validity and analysis of overlap with other similar indexes (i.e. Human Development Index and MPI-Multidimensional Poverty Index).

Results The 2 77 576 census tracts in Brazil, cover a population of approximately 97,613,505 in 56,528,865 households CFA identified the two factors proposed, with good indexes of fit and model specification (x²8=11606.06; CFI=0.98; RMSEA=0.07; p<0.05). To test the index in use, we analyzed the distribution of deprivation throughout the regions and federative states of Brazil. The quintiles of census tract showed a clear geographic pattern, with most deprived areas (the fifth quintile) concentrated within the poorest regions of each state (as classified by the MPI).

Conclusion The selection of variables was based on an extensive theoretical framework, combining a variety of aggregate variables with coverage for more than 98% of the Brazilian population. This data-reduction demonstrates there are underlying deprivation factors which means there is considerable potential for creating a small area deprivation index using other indicators of material deprivation for the whole of Brazil at the census tract level. Use of the census will enable replication with future versions of the census. Therefore, it will be crucial for monitoring inequalities in health and mortality in Brazil.

  • deprivation
  • poverty
  • social inequalities
  • Brazil

Statistics from

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.