In scenarios 5 to 8, the coverage rates ranged from 41.6% to 69.4% for the Kaplan-Meier estimate and from 88.4% to 94.8% for multiple imputation estimations.
Figure 16 Multiple imputation estimates after 30 imputations.
Estimated increases in carpet dust levels of α-chlordane among subjects treating for termites are 2.6- and 3.1-fold based on DL/2 insertion and fill-in methods, respectively, and 3.7-fold based on multiple imputation and direct estimation approaches.
That approach was extended by Durrant and Mott (2010), who describe a partially Bayesian mixed model of QTL mapping: By focusing on additive effects of QTL for normally distributed traits with no additional covariates or population structure, they provided an efficient method for combined multiple imputation and shrinkage estimation via complete factorization of a pseudo-posterior.
Multiple imputation shows promise for estimation of an occurrence rate in cohorts with attrition.
The multiplicative standard errors for the replacement approaches (i.e., inserting DL/2, E[ Z|LB < Z < UB], or a fill-in value) are smaller than standard errors from the multiple imputation approach and direct estimation.
Other methods used to account for values < LOD, including substitution of censored values with the LOD/2 or LOD/√ 2, model-based multiple imputation, or reverse Kaplan Meier estimation, cannot accommodate both repeated measurements and censored values (Jin et al. 2011).
The multiple imputation results were estimated using 20 imputations.
Given the indication from our analysis (Table 2) that the different imputations can give important differences in the estimated slope, we conducted a multiple imputation approach based on maximum-likelihood estimation to assign values for CO that were below the limit of detection.
Our results suggest that in an observational context, comparison of the results of Kaplan-Meier estimation and of multiple imputation could provide a measure of the impact of attrition on the estimation.
