Assessment of the Performance of Imputation Techniques in Observational Studies with Two Measurements
Pages 240-251
Urko Aguirre, Inmaculada Arostegui, Cristóbal Esteban and Jose María Quintana
DOI: http://dx.doi.org/10.6000/1929-6029.2015.04.03.1
Published: 19 August 2015

Abstract: In observational studies with two measurements when the measured outcome pertains to a health related quality of life (HRQoL) variable, one motivation of the research may be to determine the potential predictors of the mean change of the outcome of interest. It is very common in such studies for data to be missing, which can bias the results. Different imputation techniques have been proposed to cope with missing data in outcome variables. We compared five analysis approaches (Complete Case, Available Case, K- Nearest Neighbour, Propensity Score, and a Markov Chain Monte Carlo algorithm) to assess their performance when handling missing data at different missingness rates and mechanisms (MCAR, MAR and MNAR). These strategies were applied to a pre-post study of patients with Chronic Obstructive Pulmonary Disease. We analyzed the relationship of the changes in subjects HRQoL over one year with clinical and socio-demographic characteristics. A simulation study was also performed to illustrate the performance of the imputation methods. Relative and standardized bias was assessed on each scenario. For all missingness mechanisms, not imputing and using MCMC method, both combined with mixed-model analysis, showed lowest standardized bias. Conversely, Propensity Score showed worst bias values. When missingness pattern is MCAR or MAR and rate small, we recommend using mixed models. Nevertheless, when missingness percentage is high, in order to gain sample size and statistical power, MCMC is preferred, although there are no bias differences compared with the mixed models without imputation. For a MNAR scenario, a further sensitivity analysis should be made.

Supplementing Missing Self-Reported Race Data with a Probability Distribution in Logistic Regression Models
Pages 252-259
Stanley Xu, Komal Narwaney, Sophia Newcomer and Jason Glanz
DOI: http://dx.doi.org/10.6000/1929-6029.2015.04.03.2
Published: 19 August 2015

Abstract: Race is often included as an independent variable in health services research, especially in recent studies of racial and ethnic disparities in health care. Although self-reported information on race exists in large electronic health records (EHR) data, these data are sometimes missing. Recently Bayesian Improved Surname Geocoding method (BISG) is used to estimate the probability distribution of race categories for those with missing information on race. The BISG estimated probability distribution has been used in reporting health care measures but not in statistical modellings with dichotomous events as outcomes. We propose two approaches to accommodate available distribution probability of an independent categorical variable (e.g., race) in logistic regression models: 1) a direct substitution approach and 2) a partial information maximum likelihood estimator (PIMLE). In examining the association between race and up-to-dateimmunization status of children by three years old from an integrated health care organization, 11.3% of 14,903 children have missing self-reported race information but have BISG estimated probability distribution for the six race/ethnicity categories. We employed the direct substitution approach and PIMLE approach to analyze the under vaccination data. Both approaches included all observations and thus yielded smaller standard errors of estimated coefficients compared to the complete data analyses. Our simulation study showed that the direct substitution approach and PIMLE yielded nearly unbiased coefficient estimates and preserved efficiency when the missing rate of the independent categorical variable was up to 30%.