Compliance and missing data
This section contains the following:
Introduction
Deviations from the randomized treatment group happen in most
trials. For example, in a trial comparing placebo with
antibiotics, some of those allocated to the antibiotics group might not
take their medication. In the analysis of a pragmatic trial, this
non-compliance or protocol deviation should in general be handled using
an ‘intention-to-treat’ principle – all participants enrolled should be
included and analysed as part of the original group to which they were
assigned. The protocol should state what procedures will be
adopted to minimise non-compliance and what procedures will be
implemented to retain participants (see Post recruitment
retention strategies).
Missing data will almost always occur in a randomised trial. For
example, participants may move away from the area or might refuse to
continue participating in the trial. There is no generally
acceptable rate of loss to follow up (or missing data) but greater than
20% loss in the primary outcome(s) will pose a serious threat to the
validity of the results. In general, provided methods for dealing
with missing data are sensible and pre-defined in the protocol, the
trial results should be valid. There is no consensus on
which methods should be used for dealing with missing data so any
investigation should include a sensitivity analysis of the assumptions
used to handle the missing data. Adjustment for loss to follow up
should be made in the sample size calculation (see Sample size section)
Back to top
Things to consider when writing a protocol
- For a pragmatic trial, the intention to treat principle should in
general be adhered to
- Define how missing data will be handled
- Inflate the required sample size due to missing data (see Sample size section)
- The CONSORT flow diagram used in the dummy tables (see Dummy tables section) should be used to
illustrate the number of non-compliers and the number of missing data
in the primary outcome(s)
Scenarios where special considerations
apply
Cluster trials
In addition to possible missing patient data, cluster trials also have
the potential for missing cluster level data. For example, a
whole hospital might decline to continue with the trial. The
protocol should define how this situation will be handled.
Equivalence trials
In trials that are designed to show equivalence, it is generally
advocated that the groups are analysed on a ‘treatment received’ basis
not by ‘intention to treat’. The method of handling
non-compliance should be stated in the protocol.
Back to top
Illustrative example - Magpie trial
|
|
All analyses will be based on the groups as randomly
allocated, in other words this will be an intention-to-treat analysis.
For the principal comparisons statistical significance will be taken as
the 5% level, and for the subsidiary comparisons the 1% level. In
addition to the prespecified sub-group analyses, sensitivity analyses
will explore whether compliance with the allocated treatment influences
the size of any effects on the primary outcomes. Good compliance,
defined as loading dose plus 20-28 hours maintenance therapy, will be
compared to both higher and lower doses. The effect of >12 hours
treatment will be compared to <12 hours treatment. (Magpie trial - go to protocol)
|
Back to top
Additional resources
Back to top
This checklist was developed by Dave Sackett, who prepared it for
the forthcoming
3rd edition of Clinical Epidemiology; A Basic Science for Answering
Questions
about Health Care, to be published by Lippincott, Williams &
Wilkins
in 2004.
Back to top
This document provides guidance for the design, conduct, analysis,
and evaluation of clinical trials of an intervention in the context of
its overall clinical development.
Back to top
Further reading
Schulz KF, Grimes DA. Sample size slippages in randomised trials:
exclusions and the lost and wayward. Lancet 2002; 359: 781-5.
Fergusson D, Aaron SD, Guyatt G, Hebert P. Post-randomisation
exclusions: the intention to treat principle and excluding patients
from analysis. BMJ 2002; 325: 652-4.
Carpenter J, Pocock S, Lamm CJ. Coping with missing data in clinical
trials: a model-based approach applied to asthma trials. Statistics in
Medicine 2002; 21:1043-66.
Fayers PM, Machin D. Quality of Life: Assessment, Analysis and
Interpretation. Chichester: John Wiley & Sons, 2000.
Back to top
This page was last updated 19th October 2004.
Checklist for compliance and
missing data 
Analysis and
interpretation checklist
ICH Harmonised Tripartite Guideline:
Statistical Principles for Clinical Trials