Dropout In Longitudinal Studies

Longitudinal clinical studies repeatedly measure the outcome of interest and covariates over a sequences of time points. Longitudinal studies play a vital role in many disciplines of science including medicine, epidemiology, ecology and public health. However, data arising from such studies often show inevitable incompleteness due to dropouts or lack of follow-up. More generally, a patient’s outcome can be missing at one followup time and be measured at the next follow-up time. This leads to a large class of dropout patterns. This paper only pays attention to the monotone

1

dropout pattern that results from attrition, in the sense that if a patient drops out from the study prematurely, then on that patient no subsequent repeated measurements

Dropout mechanisms however do not imply knowledge about how the dropouts came to be unavailable. The term dropout is misused by many researchers as, in many trials, data are missing not because a participant chooses to drop out but instead because the protocol is written not to follow partici-pants following treatment discontinuation. Discontinuation might be due to adverse effects, lack of efficiency in the execution of the study, both of these reasons, or other reasons. As demonstrated by Rubin (1976), the mechanisms that lead to missingness can be classified into three basic categories. Data are considered missing completely at random (MCAR) when the mechanism that generates the dropouts is a truly random process unrelated to any measured or unmeasured characteristic of the study participants. A second category is missing at random (MAR) in which the dropout mechanism is random meaning, conditional on the observed measurements characteristics of the study sample, the dropout mechanism is independent of the unobserved measurements. Finally, missing not at random (MNAR), is one in which the dropout process depends on unobserved mea-surements and possibly on the observed measurement characteristics of the study sample. Let Yij be the response measurement of individual i at time j, where i = 1, 2, ...N and j = 1, 2, ...n, which can be observed or missing. Let Rij be an indicator variable, where

Different techniques or methods use different approaches to addressing dropout problems. Although the dropout problem is ubiquitous, there is still no firm consensus on what statistical procedures should be used for analysis or on the circumstances under which they should be applied. What follows now is a brief description of the several methods that are commonly used to deal with dropout, including a review of the existing literature in which we examine the effectiveness of these methods in the analysis of incomplete longitudinal clinical trials