Characterising for Sensitivity Analyzes the Participant Attrition in a Childhood Cohort with Large Initial Drop-Out

• Activity Number: 244 - Missing Data; Causal Inference
• Type: Contributed
• Date/Time: Monday, July 30, 2018 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Petr Otahal* and Leigh Blizzard and David W Hosmer and Jim Stankovich and Alison Venn
• Companies: Menzies Institute for Medical Research University of Tasmania and Menzies Institute of Medical Research, University of Tasmania and University of Massachusetts and School of Medicine, University of Tasmania and Menzies Institute for Medical Research University of Tasmania
• Keywords: follow-up; attrition; bias; inference; validity; stage

Abstract

Attrition in longitudinal studies can lead to bias when missing data affects distributions of analysis variables. Missing data methods to overcome bias are valid if data are missing at random (MAR), but the MAR assumption is not testable and assumptions about the attrition mechanism cannot be verified from the observed data alone. Given this ambiguity, sensitivity analyses to test the robustness of inferences are vital. Such analyses require understanding of the factors that lead to attrition. These factors are not well characterised for studies with large initial dropout, which may arise if the first follow-up occurs after a long delay. We investigate these factors in the first follow-up of the 1985 Australian Schools Health and Fitness Survey of 7-15 year old children, which was conducted 20 years post-baseline with a substantial level of attrition (>70%). We show that factors linked to attrition are similar to those identified in other studies, but they operate differentially at each stage of follow-up (tracing, enrollment, questionnaires, and clinics). Further, we compare the utility of missing data methods using the baseline data when those data are mostly incomplete.