Constructing datasets to measure geographic variation in bereavement: integrating mortality, population structure, and survey-based probability

Geographic variation in mortality across the United States is substantial and well documented.Life expectancy, crude mortality rates, and cause-specific death rates vary across states, counties, and communities, reflecting differences in demographic composition, socioeconomic conditions, healthcare access, and local mortality environments. Large-scale epidemiologic efforts, including national life expectancy analyses, U.S. Burden of Disease studies, and the Mortality Disparities in American Communities cohort, describe where and how deaths occur. However, these systems quantify decedents, not the populations exposed to those deaths.Mortality geography has advanced substantially, particularly through small-area analyses that identify where death rates are elevated, persistent, or changing over time. These studies are essential for understanding the distribution of mortality risk. However, they stop at the decedent.Bereavement surveillance requires a second step: translating death events into the populations and social units exposed to those deaths. The present Data Report therefore does not compete with mortality mapping. Rather, it extends the public health use of mortality data by linking observed deaths to survey-derived bereavement probability and household and family denominators.Bereavement is a population-level exposure generated by mortality events. Each death can affect family members, household members, friends, caregivers, coworkers, and other members of a social network. Yet routine public health surveillance systems do not estimate the number or distribution of persons, households, or families exposed to death within defined geographic areas. Mortality rates therefore cannot serve as direct proxies for bereavement. They describe the occurrence of death, but not the populations living with its consequences.The 2019 Georgia Behavioral Risk Factor Surveillance System bereavement module provided an empirical estimate of self-reported bereavement during a defined 24-month period. Li et al. reported this survey-derived probability for Georgia adults. The present Data Report makes a different contribution: it uses that probability as an empirical anchor for constructing geographically resolved bereavement datasets. Specifically, the report integrates BRFSS-derived bereavement probability with CDC WONDER county mortality data and American Community Survey household and family structure measures to generate county-level estimates under explicit calibration assumptions. This paper therefore does not attempt to identify the causes or health consequences of bereavement. Instead, it constructs a reproducible data infrastructure for translating individuallevel bereavement probability into county-, household-, and family-level estimates. The purpose is to make assumptions visible, compare alternative calibration approaches, and demonstrate how publicly available data can support bereavement surveillance and planning in areas where bereavement is not directly measured.Prior to merging, each data source was harmonized by year, state, county name, and 5-digit Federal Information Processing Standards (FIPS) code. The extracted data included population counts, resident deaths, and household and family composition for 2017 and 2018 in Georgia and Ohio. These years align with the 2019 Georgia BRFSS bereavement module item asking whether respondents had experienced the death of a family member or close friend during 2017-2018.The merged file was validated through missingness checks, outlier detection, and logical rule assessment. The 5-digit FIPS code was present in all datasets and uniquely identifies each county in the United States: the first two digits identify the state, and the last three identify the county.The publicly available data sources were:The 2019 Georgia Behavioral Risk Factor Surveillance System (BRFSS) state-added bereavement module provides the key empirical parameter used in this study: the proportion of adults reporting the death of a close family member or friend within the prior 24 months, corresponding to 2017-2018. (7,10) The analytic sample included 5,206 respondents from an initial sample of 7,354 respondents, representing 70.8% of the initial sample. The bereavement probability (p = 0.4538) was estimated using standard BRFSS survey weights (_LLCPWT), stratification (_STSTR), and primary sampling units (_PSU) to account for the complex survey design.Missing data were addressed using multiple imputation, with results combined across 50 imputed datasets to produce a single population-level estimate. This weighted, imputed estimate is used as the baseline probability in all calibration options. To assess the impact of survey design and missing data, the bereavement probability was also calculated using complete-case unweighted data. The relative difference between the unweighted and weighted estimates was 1.39%, indicating limited sensitivity to weighting and imputation. This parameter is treated as a population-level estimate of self-reported bereavement exposure within a defined 24-month period and serves as the empirical anchor for all subsequent scaling procedures.The BRFSS design includes both bereaved and non-bereaved respondents within a representative population sample, allowing direct estimation of a population-level proportion.To synchronize time across datasets, county-level resident deaths for 2017-2018 were extracted from CDC WONDER. These data were used to calculate county crude mortality rates and to characterize the mortality context in which bereavement exposure was modeled. In this framework, mortality data identify the occurrence and geographic distribution of deaths; they are not treated as direct measures of bereavement. (4)The American Community Survey (ACS) provides county-level household and family structure measures, including total households and total families. Additional ACS measures, such as household size and family size, can support future refinements of the framework. In this Data Report, ACS household and family counts are used as social-unit denominators for modeling the presence of bereavement within households and families under explicit calibration assumptions.A total of 247 counties in Georgia and Ohio were included. ( 8)Figure 1: Data sources, processing, and estimated outputs.Figure 1 illustrates the structured process used to construct a geographically resolved bereavement dataset for 247 counties in Georgia and Ohio. Three publicly available data sources are integrated:(1) the 2019 Georgia BRFSS, which provides a 24-month self-reported bereavement probability (p = 0.4538); (2) CDC WONDER mortality files, which provide county-level death counts used to calculate crude mortality rates; and (3) the ACS, which provides population, household, and family structure measures.Data processing included harmonization of geographic identifiers, alignment of time periods to 2017-2018, and annualization of relevant multi-year aggregated values. Derived measures included crude mortality rates, standard errors calculated using the delta method under a Poisson assumption, and corresponding 95% confidence intervals. The dataset was evaluated using missing-value audits, logical consistency checks, and statistical outlier detection using the 1.5×IQR rule.Three calibration options were applied to translate individual-level bereavement probability into population-level estimates. Option 1 applies a uniform probability across all counties. Option 2 adjusts this probability using county-level crude mortality rates relative to the combined Georgia/Ohio reference crude mortality rate. Option 3 uses household and family structure to model bereavement exposure within ACS-defined social units.The unit of observation in the final dataset is the county, with all measures aligned to a common time frame and standardized geographic identifiers. The dataset produces county-level estimates of bereaved individuals, households, and families under each calibration assumption, along with sensitivity analyses assessing robustness to variation in the baseline probability parameter. The merged dataset underwent systematic screening for missing values, statistical outliers, and logical inconsistencies, including deaths exceeding population counts. These procedures were used to evaluate whether the merged file was suitable for deriving calibration variables and option-specific estimates. (11) Missing-value audit: All required variables, including population, deaths, household counts, and family counts, were assessed for completeness. No missing values were identified in fields required for crude mortality rate calculations or calibration option development.The 1.5×IQR rule was applied across 16 variables to identify statistical outliers. A total of 53 unique counties were flagged, with several counties identified across multiple variables. Review of these cases indicated that the flagged observations reflected true demographic extremes, including large metropolitan counties and sparsely populated rural counties. No observations were excluded on the basis of outlier status.Logical Consistency Checks: Each county was evaluated for demographic plausibility, including deaths exceeding population counts, implausible household or family sizes, and inconsistent rate calculations. No logical errors were identified. The dataset was therefore retained in full for calibration option development.After verification, derived mortality measures were calculated from the merged dataset and added to the analytic file. These measures included crude mortality rate (CMR) per 100,000 population, the standard error of the CMR using the delta method, and corresponding 95% confidence intervals. Although comparable measures are available through CDC WONDER, recalculating them from the merged file served as an internal quality-control step and confirmed consistency between source data and derived variables.Variation in CMR across counties reflects population age composition as well as cause-specific mortality patterns, healthcare access, socioeconomic conditions, environmental exposures, and injury burden. From a population-exposure perspective, this variation represents the local mortality environment. Counties with more deaths during a defined period create more opportunities for residents to encounter loss within their social networks. In this context, CMR is used as a contextual scaling variable reflecting mortality intensity, not as a measure of causal risk. This application does not imply that demographic or environmental factors directly determine individual bereavement probability. Rather, the calibration options use CMR to represent differences in the number of deaths occurring within defined populations during the 24-month exposure window. Alternative specifications using age-standardized or age-specific mortality rates could be explored in future work. When constructing a dataset grounded in observed deaths within a defined period, CMR provides the most direct alignment with exposure occurrence.Because CMR reflects deaths occurring within a defined time window, it can capture both chronic and acute shifts in mortality intensity within a community. Under the Option 2 calibration assumption, communities with higher CMR are assigned higher modeled bereavement probabilities. CMR is used rather than age-adjusted mortality rates because bereavement exposure arises from actual deaths, not from age-standardized mortality risk.Standard errors were calculated using the delta method under the assumption that death counts follow a Poisson distribution. Corresponding 95% confidence intervals were calculated for each county-level CMR and used as descriptive measures of mortality-rate uncertainty. As shown in Table 1, Georgia had a CMR than Ohio during 2017-2018.The 2019 Georgia BRFSS provides a probability of bereavement on individual Disparities in life expectancy and mortality are well but individual bereavement has not into estimates of geographically defined The calibration options from individual bereavement probability to county-level bereavement estimates. They are not estimation they how bereavement be across geography and social were how and to each other family or of and were excluded from the analytic the bereavement probability. Although the module includes and a defined 24-month window, it does not provide a of does it whether to or of a death within social the baseline probability (p = 0.4538) reflects exposure to loss within a standardized time This is with the BRFSS design as a population health surveillance on A of could the reported a could the and of this across demographic or the BRFSS module includes both bereaved and non-bereaved respondents within a representative population sample. bereavement studies bereaved and do not a population The BRFSS design therefore provides a population-level proportion to a defined geographic scaling In this the probability parameter is treated as an observed survey proportion rather than a measure of and all calibration options are to this This represents the proportion of adults reporting the death of a close friend or family member in the 24 prior to This estimate represents Georgia adults years and within an population of This is used as Option uniform 2 and 3 assumptions how local mortality intensity and social structure modeled bereavement probabilities. Option 2 uses an calibration on the of rate to the BRFSS probability by county CMR relative to a reference CMR can produce rates, also the data for using the that county had than deaths in the calibration was retained 1 Georgia 2019 BRFSS probability 1 provides a a uniform distribution of bereavement and is derived from the weighted 2019 Georgia BRFSS bereavement module geographic units are assigned the probability during the 24-month this variation in the estimated number of bereaved is by resident population The population is defined by the analytic To estimate the number bereaved for a geographic 1 = = 1 = Option 2 bereavement probability 2 be as a between mortality geography and bereavement county mortality intensity the survey-derived bereavement but the a modeled bereavement exposure estimate rather than an observed mortality 2 adjusts the BRFSS-derived bereavement probability by county crude mortality The purpose is not to bereavement directly from mortality rates, but to how the survey-derived probability counties in the of deaths occurring during the exposure window. The county CMR is therefore used as a contextual calibration each county, the Option 2 probability is calculated = county CMR = is the county crude mortality and is the combined Georgia/Ohio reference crude mortality rate of deaths per 100,000 are as and can also be as estimated bereavement rates per 100,000 calibration the BRFSS bereavement probability as the empirical anchor allowing county-level mortality to the estimate. Counties with crude mortality rates the reference Option 2 counties with crude mortality rates the reference higher Option 2 probabilities. Option 2 does not the survey estimate. that estimate to observed county-level mortality could produce than in this calculated were to the In the present Georgia/Ohio county the was the calibration compare two Georgia and had a population than but a crude mortality rate. Under Option 1, the bereavement probability is applied to both counties, the estimated number of bereaved is by population Under Option mortality intensity in the bereavement higher mortality intensity in This the of Option it makes how counties can not in population size, but also in the mortality context in which bereavement American Community Survey to between households and families as social As defined by ACS, a household includes and living in the A family is limited to living in the unit are by or This bereavement in two it a bereaved individual within a it whether are or This prior that bereaved as Although not shown in this Data Report, ACS provides on household Option 3 structure scaling Option 3 extends the individual-level bereavement probability to ACS-defined social units by the modeled probability that a household or family at bereaved member during the defined 24-month period. This uses household and family structure to translate individual-level exposure into social-unit exposure under explicit calibration assumptions. This does not estimate the number of bereaved within a household or does it account for of bereavement within Instead, it a exposure social units where bereavement is present under the assumption that exposure is across the Option 3 therefore be as a scaling linking individual exposure to social rather than as a direct estimate of bereavement the three options provide a for assessing bereavement across geographic areas and social 1 a baseline for and uniform probability. Option 2 bereavement estimates to mortality differences across counties. Option 3 and population social structure using explicit calibration assumptions. the that the three options are not a single true They are structured assumptions for translating a survey-derived individual probability into geographic and social-unit A bereavement rate distribution. the distribution of bereavement rates calculated under the Option 1 and Option 2 assumptions. The under the uniform probability assumption, the Option 1 rate is bereaved per 100,000 and the distribution of county-level Option 2 bereavement rates per 100,000 Option 2 applies the where the combined Georgia/Ohio reference CMR is deaths per 100,000 Counties with crude mortality rates the reference CMR the Option 1 counties with crude mortality rates the reference CMR The the distribution of estimates across counties and that counties modeled Option 2 rates the uniform Option 1 rate. These values be as modeled rates under the assumption, not as directly observed county bereavement county CMR per 100,000 population on the Option 2 bereavement rates per 100,000 population on the No county had than deaths in year, that modeled Option 2 rates were by death counts. This how county-level mortality intensity the modeled bereavement probability under the Option 2 calibration county variation in bereavement To provide a of bereavement rates, the 247 counties by their The are Georgia counties. with for crude more than counties modeled rates than be estimated using a uniform probability of the calibration process can be by and counties in The two counties in population size and mortality had an population of and a crude mortality rate of deaths per 100,000 had an population of and a crude mortality rate of deaths per 100,000 and of the calibration county has a large population with mortality intensity, the other has a population with mortality Option 1, the uniform BRFSS probability is applied to both counties. This produces an estimated bereaved in and in Under Option the probability is to county The probability to in and to in the county with the estimated number of bereaved because of its population size, but has a higher probability of bereavement the is to families under Option the modeled family-level probability is in and in These estimates be as the modeled probability that a family unit at bereaved member under the calibration not as a direct of bereaved within families. The calibration is it population from mortality intensity and both to geographic estimates of bereavement estimates capture social including family-level estimates exposure within also illustrates a planning A death within a household or family can and the was a or The present dataset does not measure but ACS household and family denominators could support future analyses of how modeled bereavement exposure with extends the probability to ACS household and family are well the Option 1 individual probability indicating that bereavement probability is to social a large of households and families are as at bereaved The family distribution is higher than the household with the use of family structure to estimate the probability that at member is These estimates be as modeled exposure of social not as counts of bereaved within those can families are limited to and families a of modeled bereavement with rates per 100,000 The modeled distribution of exposed social and families as at bereaved calculated under the Option 3 the individual probability is to ACS household and family both of social units estimated exposure to bereavement. The family distribution is higher than the household indicating that family units a higher modeled probability of at bereaved member than household units under the calibration BRFSS bereavement probability estimate (p = 0.4538) was derived from a single This sensitivity in the probability but not in its to A sensitivity was to assess the robustness of estimates across a of values. alternative were = = These values in the self-reported probability of bereavement on and survey design This sensitivity uses the scaling which adjusts for crude mortality intensity and household or family The the in bereavement rate the BRFSS-derived uniform probability from to the bereavement probability between and of to This was for households and not a county-level sensitivity that geographic that Option 3 are not to variation in the baseline probability parameter. This of is particularly for counties with The sensitivity the use of bereavement as a for county-level bereavement Data Report a by mortality mortality systems identify and geographic of death rates, and environments. They do estimate the population living with the consequences of those deaths. mortality counts, BRFSS-derived bereavement and ACS household and family this dataset an infrastructure for modeling bereavement exposure across geographic and social This dataset was to individual bereavement probability to geographic and social The is but several and future development.The 24-month bereavement probability (p = 0.4538) is derived from a single state, at a single time No comparable bereavement module was in Ohio. this probability to Ohio therefore of self-reported exposure across and Ohio in factors that mortality the period, CMR per was higher than CMR per a difference that could the true proportion of adults reporting bereavement. The sensitivity (p = to the robustness of to variation in the probability but it does not that true within this this the estimates for Ohio be as under an explicit assumption of parameter of Ohio serves two to demonstrate how the calibration under a different mortality and (2) to how publicly available mortality and demographic data can be in a a bereavement surveillance The is to geographic where mortality counts and household structure data are However, bereavement probability estimates of BRFSS bereavement to states, or of national survey with comparable time of the probability parameter in future of this can be and to BRFSS mortality data or demographic can also be to other where mortality counts and household or family denominators are provided that local or bereavement probability estimates are is contextual and This is most mortality data are over time. In this framework, CMR provides the and geographic context to future public health the of bereavement in geographic dataset provides a reproducible for bereavement the individual mortality household structure data, and BRFSS-derived the a to understanding the of mortality as it be through Each death and household consequences that cannot be from mortality rates for population age or The 2019 Georgia BRFSS module an step by bereavement during a defined 24-month period and a population-level probability of of bereavement the of this estimate. The Survey also a national item the death of close in the years and provide an source for future estimation of bereavement Option 3 uses American Community Survey (ACS) data the number of households and families in which bereavement could be present at member is during the 24-month window. In the ACS, households represent units that and families capture units within The does not estimate the probability that multiple within the household or family are does it the number of bereaved within a Instead, family-level are as denominators for units where bereavement is present under explicit calibration assumptions. In where deaths within multiple bereaved in the this exposure intensity within the unit but does not whether the unit is as the analytic shifts from units to of bereaved individuals, and more could or to represent and intensity of exposure within households and families. observed across the three options alternative calibration assumptions rather than differences in statistical Option 1 applies a uniform survey-derived 2 adjusts that probability to contextual mortality option is each is to a analytic Option 2 county-level bereavement the survey-derived probability were for county Option 3 the social structure of the or for differences in mortality The options on the planning A uniform be for a baseline a be contextual mortality differences are relevant for The dataset is to across these household and family estimates are derived from ACS data. The sensitivity analyses of bereaved within a household or but social and not follow assumptions. This has for and public health to by deaths in a community. ACS household and family estimates also of error This source of is not into the calibration options and represents an for future reflects population age populations higher mortality can at The use of CMR is therefore a rather than a in this application because it mortality intensity within a defined time This be the three datasets, variables were but not reflecting differences in survey design and therefore required and which but These differences were during and are not to the application of the This Data Report a for bereavement Mortality geography identifies where deaths bereavement geography where the consequences of those deaths be across persons, households, and counties. As this represents a for this of use of crude mortality rates that calibration the actual number of deaths occurring in a community. CMR is for this application because bereavement exposure arises from deaths rather than age-standardized CMR is derived from death and is into multiple public data sources, and dataset on publicly data BRFSS, CDC WONDER, and This makes the reproducible and to other geographic where mortality is and households or families are or could the by a bereavement module that loss during a defined time of deaths and of are by most and to This of this allowing to other and family-level exposure capture different the presence of bereavement in units that and exposure within including and other These support planning that individual calibration options provide analytic They to compare how different assumptions affect county-level bereavement estimates and to the option that a or planning In this outlier detection confirmed that reflected population including large counties, rural counties, and counties with or these differences is essential for bereavement estimation and this deaths are not the they are the modeling bereavement the dataset support future planning efforts, particularly with local data. the health of are population-level estimates identify geographic areas where or planning be these make the dataset a infrastructure for bereavement from an individual to a public health exposure across geography and social