Abstract:
|
Missing data are common in longitudinal aging studies. Older adults who are lost to follow-up or miss visits are likely poorer in health. This non-random missingness (MNAR) violates the missing at random (MAR) assumption in regular statistical analysis and can create bias. The ceiling effects in the outcome further complicate the analysis. We have shown that auxiliary data, additional measures that are associated with outcome measures and missing data, can be used to test the MAR assumption and reduce the bias that results from MNAR. For example, in estimating cognitive decline measured by trail making test B (TMT-B) which is subjective to ceiling effect, the telephone version of TMT-B can serve as the auxiliary information. We use joint modeling approach to utilize auxiliary information and further take account of ceiling effect. Simulation studies are performed to examine the impact of various factors regarding missing data, auxiliary data and ceiling effect on the estimation rate of change of longitudinal outcomes. We apply the method to data from the Einstein Aging Study (EAS). This research is supported by NIH R21AG056920 and P01AG003949-33.
|