Imputation Models to Estimate Mean Fatty Acid Percentages in NHANES

• Activity Number: 583
• Type: Contributed
• Date/Time: Wednesday, August 1, 2012 : 2:00 PM to 3:50 PM
• Sponsor: Section on Survey Research Methods
• Author(s): Maya Sternberg*+ and Rosemary Schleicher
• Companies: CDC and CDC
• Keywords: NHANES ; imputation ; missing data

Abstract

We undertook an investigation to assess the impact of missing data when computing percentages of plasma fatty acids (FA) by comparing a variety of imputation models. Plasma FA are short-term biomarkers accounting for several weeks of dietary intake. Twenty-four FA are measured as part of National Health and Nutrition Examination Survey (NHANES)- a complex, multistage, area probability sample of the US non-institutionalized civilian population. While measured in absolute concentrations, individual FA are most often reported as percentages of total FA. Due to the logistics and cost of obtaining complete profiles of the 24 FA for all NHANES respondents and other issues of multi-analyte quality control, a certain amount of missing data is expected. We assessed imputation models using Markov chain Monte Carlo under multivariate Normality with and without random effects. In general the estimates based on multiple imputation had better statistical properties than simpler alternatives such as complete case analysis. We also found accounting for the sampling design did not seem to be as important as the amount of missing data and which FA were missing.