Re-randomisation in trials

1
Re-randomisation in trials

• Simon Gates
• 25 April 2007

2
Contents

• Introduction the problem
• Inclusion issues
• Analysis issues
• Discussion

3
The problem

• Recruitment period of trials can last a long time
• A participant may be randomised and have trial
  treatment, then re-present with the same problem
  and be eligible again later.
• Randomising again is equivalent to randomising
  episodes of disease rather than patients.

4
Examples

• Trauma
• Pregnancy
• Recruitment period of 2 or 3 years is long enough
  for some women to participate in trial, then get
  pregnant and become eligible again
• Acute asthma
• People are eligible when they attend hospital
  with acute asthma
• People will recover within a short period
• Some may have attacks several times in a year.

5
What should we do with these people?

• Randomise again or not?
• If not, what treatment should they receive?
• Should their second episode be included in
  analyses?
• If so, how should they be analysed?

6
Is this an important question?

• Number of participants affected usually small and
  hence low potential for bias in the results.
• Example MAGPIE trial (magnesium sulphate for
  pre-eclampsia) 15 out of 10141 women randomised
  twice.

7
Is this an important question?

• Major issue in some fields
• Cochrane review Colony stimulating factors for
  chemotherapy-induced febrile neutropenia.
• 4/13 trials randomised episodes contained 25 of
  episodes and patients in the review.
• Re-randomisation stated by methodology consensus
  statement to be an acceptable design.
• Immunocompromised Host Society. The design,
  analysis, and reporting of clinical trials on the
  empirical antibiotic management of the
  neutropenic patient. Report of a consensus panel
  J. Infect Dis 1990, 161(3)397-401.

8
Is this an important question?

• Problem occurs in many trials.
• Much time and effort invested by triallists and
  statisticians in dealing with it (re-inventing
  the wheel)
• No agreed best practice.

9
Literature

• Very little published on this issue
• One methodological paper
• Hozo I, Djulbegovic B, Clark O, Lyman GH. Use of
  re-randomized data in meta-analysis. BMC Med Res
  Methodol. 2005 May 105(1)17.

10
Inclusion issues
11
Include or not?

• Randomise again and include in analysis as a
  separate data point
• Allocate treatment deterministically (i.e. not by
  randomisation), and include second participation
  in analyses
• treatment originally allocated
• treatment not originally allocated
• intervention group
• control group

12
Include or not?

• Randomise or allocate deterministically to a
  treatment option but do not include second
  participation in analyses
• Do not give any of the trial treatments (and do
  not include second episode in analysis)

13
Factors affecting decision

• Best strategy is likely to vary depending on the
  nature of the trial
• Need to consider what strategy is best under what
  conditions

14
Duration of follow-up?

• If participants are still being followed up from
  their first participation, do not want them to
  receive a different treatment.
• Participants would be in both groups
  simultaneously
• Erodes difference between groups
• Allocate to original treatment?
• Include second episode in analysis or not?

15
Follow-up

• Trial comparing new treatment versus usual care
  with long-term follow-up.
• Likely that a proportion of the new treatment
  group will receive usual care during follow-up
  for subsequent episodes (e.g. at hospitals not in
  the trial).
• Triallists may not know about this.

16
Recovery from first episode

• Outcomes of second episodes of disease within a
  short period may be different.
• Treatment received during first episode may
  affect the outcome of the second.

17
Patient preferences

• Patients who participate once and have a good
  outcome may be keen to participate again
• Is it ethically acceptable to refuse, or to
  include them and not use the data?

18
Identifying previous participants

• Patients who present for a second time may not be
  identified as previous participants
• May therefore not be possible to allocate them to
  desired treatment
• They may be randomised in belief that it is first
  presentation

19
Identifying previous participants

• Planning of strategy for re-randomisation must
  consider practicalities of identifying
  participants.
• Example patient presents to a different hospital
  for second episode.

20
Analysis issues
21
Analysis

• Main problem in analysis is that observations
  from the same individual are not independent
• Some individuals may be randomised more than once
  even if the trial protocol forbids this so
  problem is likely to arise in most trials where
  this is possible

22
Analytical strategies

• Ignore
• Analyse as if two randomisations were independent
• Probably the commonest approach

23
Analytical strategies

• Include data from only one randomisation for each
  patient
• Ensures there is no non-independence
• Second participation will be ignored
• Is it ethical/scientifically acceptable to
  randomise (or otherwise include) with the
  intention of ignoring the data?

24
Multiple participations with the same treatment

• Equivalent to cluster randomised trials
• Standard methods for analysis
• Have been used for situations like multiple
  pregnancies

25
Multiple participations with different treatments

• Analysis not so straightforward
• If treatments always different, similar to
  crossover trial
• If a mixture of the same and different, ???

26
Discussion
27
Discussion