Intended for healthcare professionals

Rapid response to:

Analysis And Comment Research methods

Reporting attrition in randomised controlled trials

BMJ 2006; 332 doi: https://doi.org/10.1136/bmj.332.7547.969 (Published 20 April 2006) Cite this as: BMJ 2006;332:969

Rapid Response:

Attrition can also be understood and managed in other ways.

Editor,

It makes sense that the authors do not distinguish between loss to
follow up and missing data since missing data commonly arise from losses
to follow up or failure of participants to return for assessment of key
outcomes. To the extent that a study sets out to answer the question of
effectiveness (does it work?) rather than efficacy (can it work?) and is
analysed according to intention-to-treat, these factors may not reduce
internal validity significantly.

Furthermore, post-randomization exclusions of participants by
investigators might be acceptable if the patients found to be ineligible
after randomization are excluded early on and before any events have taken
place or outcomes assessed, and as long as it is done without awareness of
the group to which the excluded individuals were assigned. These
conditions are however hard to attain practically and it is well stated
that exclusion is generally ill advised. Moreover, this type of exclusion
is suspicious if it is done on the ground of poor adherence since poor
adherence might be related to the intervention under assessment. Although
not explicitly stated by the authors, it should also be distinguished from
pre-randomization exclusion of potential participants following a run-in
phase prior to randomisation. This method has been used successfully in
the Physicians’ Health Study to identify and exclude those subjects who
are likely to be poor with adherence1.

If missing data are not distinguished from loss to follow up, it is
hard to see why they should make intention-to-treat analysis difficult
especially if their occurrence is not strongly related to the intervention
under assessment. Assuming that they really do, they can be handled,
albeit imperfectly, by substitution with prior measurements in the
subjects or by derived imputations from group averages. Intention-to-treat
analysis ensures that the benefits of randomisation are not lost and, even
in the face of attrition, can still answer the important question of
whether interventions work in real settings.

Reference

1. PHS Steering Committee of the Research Group. Final report on the
aspirin component of the ongoing Physicians’ Health Study. New England
Journal Of Medicine 1989; 321: 129-35.

Competing interests:
None declared

Competing interests: No competing interests

21 April 2006
Kelechi E. Nnoaham
Specialist Registrar in Public Health
Oxford City Primary Care Trust, Old Road, Headington OX3 7LG.