Impact of missing data mechanism on the estimate of change: A case study on cognitive function and polypharmacy among older persons

Abstract

Longitudinal studies typically suffer from incompleteness of data. Attrition is a major problem in studies of older persons since participants may die during the study or are too frail to participate in follow-up examinations. Attrition is typically related to an individual’s health; therefore, ignoring it may lead to too optimistic inferences, for example, about cognitive decline or changes in polypharmacy. The objective of this study is to compare the estimates of level and slope of change in 1) cognitive function and 2) number of drugs in use between the assumptions of ignorable and non-ignorable missingness. This study demonstrates the usefulness of latent variable modeling framework. The results suggest that when the missing data mechanism is not known, it is preferable to conduct analyses both under ignorable and non-ignorable missing data assumptions.