A New Measure of Population Health in a Complex Sampling Survey: The 2019 Georgia Behavioral Risk Factor Surveillance Survey – Unweighted, Weighted, and Imputed Weighted Data
Abstract Background: Measuring population health requires a well-defined denominator. The Behavioral Risk Factor Surveillance Survey (BRFSS) is designed to provide one for state-level populations. In 2019, the U.S. state of Georgia tested a new module to study recent bereavement among its 8.1 million residents aged 18 years and older. This is the first population-level assessment of bereavement. In BRFSS, bereavement is defined as the fact of a death. The term grief is not used because it denotes emotions related to that death. There is evidence from cohort studies linking new bereavement to subsequent mortality and health care utilization. Methods: BRFSS data are obtained using list-assisted, random digit dialing from the non-institutionalized population aged 18 years and older within primary statistical units. Data came from both landline telephones and cellular phones. Three questions were added to the end of the latest BRFSS asking about death of family or friend in 2018 or 2019. To evaluate data from this new module, the report presents three statistical approaches - unweighted panel data, weighted data, and weighted data using multiple imputation of missing responses. The estimated prevalence of bereavement and its standard error under each data scenario is calculated. Results: The threat to the validity of data from this new module are bias due to small samples and missingness. The unweighted panel contains 5,206 persons (70.9 percent response). Among these, 2,396 persons (46.0 %) responded ‘Yes’ to ‘Have you experienced the death of a family member or close friend in the years 2018 or 2019? To estimate the size of the population, weights are applied. With weighted data and removal of missing responses yields a prevalence of 45.56 % (SE = 1.13) with a population estimate of 4,937,056 persons. Using multiple imputation to keep missing responses, the prevalence is 45.80 % (SE = 1.18) with a denominator of 8,164,018 persons. Conclusions: New bereavement can be ascertained in a surveillance survey without bias due to refusals. Multiple imputation provides a population size estimate that is comparable to U.S. Census bureau. More field testing is required to replicate these results in other states.
