Estimating missing concentrations of p,p′-DDT in the 1999–2004 National Health and Nutrition Examination Survey using multiple imputation

We investigated the association between diabetes and p,p′-DDT (dichlorodiphenyltrichloroethane) in blood of Mexican Americans who participated in the 1999–2004 National Health and Nutrition Examination Survey (NHANES). In this sample, p,p′-DDT were missing in 50% of subjects and we used multiple imputation (MI) to address the problem. Compared to ignoring the missing data, MI led to a more robust threshold for the p,p′-DDT reference category. Whereas previously p,p′-DDT ≤ 0.0860 ng/g was used as the reference category, using MI, we were able to use p,p′-DDT < 0.0574 ng/g as the reference category to study the association between p,p′-DDT and diabetes via logistic regression. In this analysis, p,p′-DDT ≥ 0.0750 ng/g was associated with an odds ratio of 1.99 (95% CI 1.09–3.61) for diabetes and 4.20 (95% CI 1.93–9.12) for diabetic nephropathy. The reference category for diabetes without nephropathy outcome stayed consistent after MI but our analysis confirmed that p,p′-DDT > 0.0860 ng/g was associated with diabetes without nephropathy with an odds ratio of 1.89 (95% CI 1.09–3.27). Our study showed that MI can be effectively used to deal with missing at random data in persistent organic pollutants measured in the 1999–2004 NHANES.