Through the CDC/CSTE project Sub-County Assessment of Life Expectancy (SCALE) collaboration, the Virginia Department of Health (VDH) calculated tract level life expectancy (LE) estimates. Using multiple linear regression, outcome LE estimates were modelled to determine the significance and strength of the association that social determinants of health (SDH) have on life expectancy.
METHODS:
The CDC/CSTE project SCALE provided the VDH with the tools and guidelines for calculating LE at the tract level. The Chiang II method with Silcock’s adjustment was used to calculate LE in Virginia census tracts. The 2010 Census was used as the midpoint population for 7 years of mortality data (2007-2013). Deaths from 2007 to 2009 were reassigned to census 2010 tracts. Abridged life tables were calculated using 19 age categories. Census tracts with population years below 5000 or with no deaths in 10 or more age categories were omitted. The LE estimates by tract were mapped using ArcGIS v10.2.2 and used as the dependent variable in a multiple linear regression with SDH as the independent variables. For brevity, SDH considered in the model are categorized into four groups: environmental, consumer opportunity, economic, and wellness disparity. A geographically weighted regression (GWR) was computed to explore regional variance.
RESULTS:
LE was calculated using a standardized method. Life expectancy at birth ranged from 64.3 to 99.1 years with a median of 78.8 years. Mapping the estimates revealed the impact that environment (place) has on life duration. All SDH were individually associated with LE. In a multiple linear regression model, SDH that were significant (p-value <0.0001) and strongly associated with higher LE included higher average years of schooling, lower proportion of the community’s income spent on housing and transportation, better air quality, and higher population density. The model with those four determinants accounted for over half of the variation in LE. GWR revealed that locally weighted education coefficients had a regional pattern and were spatially autocorrelated.
CONCLUSIONS:
Comparing LE estimates between tracts highlights vulnerable areas where health inequities exist. Regional modeling addresses spatial correlation. Regional modelling of SDHs can inspire government, coalitions and other stakeholders to make more informed decisions and policies by linking determinants to health outcomes. The relationship modelled between SDH and LE can empower coalitions through quantifying potential impact of strategies. Coalitions can see hypothetical changes in LE by modifying parameter estimates of the SDH and using areas means and the linear regression equation.