Premium
Sensitivity analysis of intention‐to‐treat estimates when withdrawals are related to unobserved compliance status
Author(s) -
Salim Agus,
Mackin Andrew,
Griffiths Kathleen
Publication year - 2008
Publication title -
statistics in medicine
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.996
H-Index - 183
eISSN - 1097-0258
pISSN - 0277-6715
DOI - 10.1002/sim.3025
Subject(s) - dropout (neural networks) , context (archaeology) , compliance (psychology) , sensitivity (control systems) , econometrics , missing data , statistics , computer science , psychology , mathematics , social psychology , machine learning , engineering , paleontology , electronic engineering , biology
In the presence of dropout, intent(ion)‐to‐treat analysis is usually carried out using methods that assume a missing‐at‐random (MAR) dropout mechanism. We investigate the potential bias caused by assuming MAR when the dropout is related to unobserved compliance status. A framework to assess the magnitude of bias in the context of pre‐ and post‐test design (PPD) with two treatment arms is presented. Scenarios with all‐or‐none and partial compliance level are investigated. Using two simulated data sets and actual data from an e‐mental health trial, we demonstrate the utility of sensitivity analyses to assess the bias magnitude and show that they are plausible options when some knowledge of compliance behaviour in the dropout exists. We recommend that our approach be used in conjunction with methods of analysis which assume MAR in estimating the ITT effect. Copyright © 2007 John Wiley & Sons, Ltd.