Data on registered deaths come from the Washington State Department of Health for the years 2011-2018. The data file included information on decedents’ age, sex, race, ethnicity, education, marital status, and residential longitude and latitude. Inclusion criteria for this study were individuals who were 25 years and older at the time of death. People at this age are mature and have a more stable level of education than those 18 and younger . Our final data set includes nearly 400,000 decedents from Washington state over the study period.
Outcome variable. YPLL, which gives more weight to deaths occurring at younger ages, was calculated for all decedents . This measure is calculated by subtracting the age at the time of death from a predetermined end point age . The premature mortality benchmark of 75 has been used in U.S., Canadian, Australian, and European studies for quantifying the burden of premature mortality (Adair and Lopez, 2020; Athens et al., 2015; Renard and Deboosere, 2014; Zygmunt et al., 2020). For example, using the end point age of 75, an individual who dies at the age of 25 will have 50 years of life lost, and a person dying at the age of 60 will have 15 years of life lost.
Individual-level explanatory variables. Individual-level variables included race and ethnicity (six categories: NH white, NH Black, NH AIAN, NH Asian, Native Hawaiian or other Pacific Islander (NHOPI), NH multiracial (3 or more racial identities), and Hispanic, sex, educational attainment at time of death, and marital status at time of death.
Community-level explanatory variables. Rurality was classified using the Rural-Urban Commuting Area (RUCA) codes based on decedents’ residential location at the time of death. RUCA codes use work commuting information, population data, and measures of urbanization to classify urban and rural areas at the census tract level. RUCA codes of 1-3 were classified as metropolitan areas, codes of 4-6 were classified as micropolitan areas, and codes 7 through10 were classified as small towns and rural areas (Washington State Department of Health, 2017). The ADI, a validated composite score of socio-economic disadvantage, was used to quantify the social and economic characteristics of census tracts (Knighton et al., 2016; Singh, 2003). The ADI was developed based on 17 Census variables in four domains of poverty, housing, employment, and education. We divided ADI scores into terciles of deprivation (1=least-deprived, 2=middle-deprived, 3=most-deprived).
Descriptive statistics included measures of central tendency and variability for continuous variables and frequency distributions and percentages for categorical variables. The Interquartile Range (IQR) will be calculated to determine the midpoint of average death for each racial and ethnic group and determine the spread of death across all groups. Generalized linear mixed models with zero‐inflated Poisson distributions tested the association between race/ethnicity and YPLL controlling for other explanatory variables. Investigation of data suggested that zero-inflated models are appropriate because of an excessive amount of zeros in the outcome . We treated decedents at level-1 nested within level-2 census tracts. Our models included a random intercept, allowing the likelihood of YPLL to vary across census tracts. This enabled exploration of the associations of both individual- and community-level variables with YPLL while accommodating the clustering of decedents within census tracts.
We start with an adjusted model to examine the relationship between YPLL and race/ethnicity, controlling for sex and marital status. Next, we progressively adjust for education of the decedent and then for residential census tract rurality and area disadvantage. In addition to models on all decedents, we also stratify results and present male and female specific estimates. Associations are presented as incidence risk ratios (IRR) with 95% confidence intervals (CI). The R-software was used for analysis.