Shellie Kolavic Gray

Ecological-type inference in matched-pair studies with fixed marginal totals†

This paper applies a Bayesian approach to ecological-type inference in matched-pair studies because traditional methods that assume parallel tables are not directly applicable. The proposed procedure is based on a hierarchical Bayes structure which models information about the within-pair association. The proposed algorithm relies on Markov chain Monte Carlo simulation, recovers the full table, and reports its accuracy in terms of credible sets for the cell counts. This methodology is motivated and illustrated with examples from split-mouth designs assessing the effectiveness of dental sealant materials and cross-over trials comparing two forms of insulin.

Prevalence of past-year dental visit among US adults, 1999–2010: comparison of trends and estimates between the Behavioral Risk Factor Surveillance System and three national surveys

OBJECTIVES To compare estimated prevalence of past-year dental visit (PPYDV) among US adults aged ≥18 years from the Behavioral Risk Factor Surveillance System (BRFSS) to estimates from the Medical Expenditure Panel Survey (MEPS), National Health Interview Survey (NHIS), and National Health and Nutrition Examination Survey (NHANES). METHODS We estimated PPYDV adjusted for covariates (age, race/ethnicity, education level, poverty status, edentulism) using BRFSS, MEPS, and NHIS 1999-2010, and NHANES 1999-2004. We tested trend in overall PPYDV for BRFSS, MEPS, and NHIS from 1999-2010. For 2002 and 2010, we calculated absolute differences (AD) and 95% confidence intervals (CI) in PPYDV between BRFSS and each of the other surveys overall and among subpopulations defined by covariates. We pooled NHANES 1999-2004 data for comparison with BRFSS 2002. RESULTS From 1999 to 2010, BRFSS (68.5% vs. 67.5%), MEPS (43.5% vs. 39.7%), and NHIS (63.3% vs. 59.7%) showed small but significant decreases in overall PPYDV. In 2002, estimates for overall PPYDV were highest for BRFSS (70.0%) and lowest for MEPS (43.9%) with estimates for NHIS (61.5%) and NHANES (1999-2004: 58.1%) in between; the largest AD (26.2%, 95% CI: 25.0%-27.3%) was between BRFSS and MEPS. ADs were consistent in 2002 and 2010, overall and by covariates, except among edentate persons, where PPYDV estimates from BRFSS and NHIS were similar. CONCLUSIONS Estimates of PPYDV from BRFSS were notably higher than estimates from MEPS, NHIS, or NHANES except among the edentate. Trends in PPYDV over time, however, were consistent across all surveys.