County Estimation of Multiple Cigarette Smoking Categories Using the 2012 Behavioral Risk Factor Surveillance System

Authors:

Zahava Berkowitz (Presenter)
CDC

Xingyou Zhang, U.S. Census Bureau
Thomas Richards, CDC
Lucy Peipins, CDC
Jane Henley, CDC
James Holt, CDC

Public Health Statement: Using a model that generated county-level prevalence estimates for 6 smoking levels including current, every day, some day, former, ever, and never may help identify areas and populations in need of tobacco prevention and control.

Purpose: To identify areas of concern about cigarette smoking, we developed multilevel small area estimation (SAE) mixed models that generate county-level prevalence estimates for 6 smoking levels: current, every day, some day, former, ever, and never—some of which, to our knowledge, have limited data.

Methods/Approach: We used the 2012 Behavioral Risk Factor Surveillance System (BRFSS) data (N=405,233) to construct and fit a series of 3 multilevel logistic regression mixed models with progressively smaller populations and linked them to the U.S. Census population (post stratified) to generate county-level prevalence estimates. We used the results from these models in newly constructed logistic regression Monte Carlo simulations to predict individual-level probability for each smoking level. We mapped the county-level prevalence estimates by gender and aggregated the results into larger smoking areas. We compared SAE state estimates with BRFSS state estimates for internal consistency, and the SAE national prevalence estimate of current smoking with the National Health Intervew Survey (NHIS) current smoking estimate for external validity using the Pearson correlation coefficients.

Results: Pearson correlation coefficients indicated high internal consistency, with ρ ranging from 0.908 to 0.982. The current estimate of our SAE model was the same as the current estimate of NHIS (18.06%). Small area estimation results showed large variations in current and former smoking prevalence between states and within states. Former smoking prevalence was highest among counties in the Northeast, North and West, and among males. Utah had the lowest smoking prevalence.

Conclusions/Implications: The small area estimation models, which include demographic and geographic characteristics, provide reliable estimates applicable to multiple category outcomes. County estimates for different smoking categories may help identify areas and populations in need of tobacco prevention and control.