Correcting Bias in Effects of Risk Factors in Longitudinal Studies Due to Non-Random Missingness Using Auxiliary Data

• Activity Number: 145
• Type: Contributed
• Date/Time: Monday, August 5, 2013 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Charles Hall*+ and Culing Wang and Mindy Katz and Richard Lipton
• Companies: Albert Einstein College of Medicine and Albert Einstein College of Medicine and Albert Einstein College of Medicine and Albert Einstein College of Medicine
• Keywords: Missing data ; Longitudinal data ; Cognitive aging ; Auxiliary data

Abstract

When missing data in longitudinal studies depend on unobserved values, this missing not at random (MNAR) situation can cause estimates of trajectories and of the effects of risk factors to be biased. Joint modeling using auxiliary data can partially correct this bias. We apply this method to the study of the effect of the APOE e4 genotype on memory decline in an elderly cohort from the Einstein Aging Study. The Free and Cued Selective Reminding Test (FCRST) is used to model memory decline, and the Memory Impairment Screen for Telephone (MIS-T) is used as auxiliary data as it was administered to study participants even when they could not be given the FCSRT. Joint modeling of the FCSRT and MIS-T shows that the estimate of the effect of APOE e4 genotype is significantly underestimated in naïve approaches that ignore the non-random missing data. Simulations show that it is likely that the correction for this missingness is only partial. More research is needed to identify optimal auxiliary data for specific situations, and to further improve the accuracy of statistical methodology used to address the bias.