Patient Preference and Comprehensive Cohort Designs

1
Patient Preference and Comprehensive Cohort
Designs
2
Background

• Patients often have a preference for treatment.
  Patients with strong preferences for usual care
  often do not get into a trial because
  randomisation does not guarantee that they will
  get what they want.
• Patients who do get into a trial with strong
  preferences for the novel treatment can bias
  results.

3
Strong preferences

• When a treatment is ONLY available within a trial
  context patients who want the treatment may
  decide to consent to randomisation with the hope
  they may get the treatment.
• This can lead to the following problems
• Demoralisation
• High drop out
• BIAS

4
Effect of Preferences

• Patients may refuse to fill in follow-up
  questionnaires due to resentful demoralisation,
  which can lead to bias.
• Patients who DO get what they want may
  exaggerate the effectiveness of their treatment
  again this MAY lead to bias.

5
Quality of Life and Preference

• Preferences are particularly a problem when
  quality of life is a major outcome as this is
  more susceptible than objective measures of
  outcome (e.g death).

6
Patient Preference Trial A solution

• One approach to the issue of preferences is to
  undertake a patient preference trial.
• Only patients indifferent to which treatment
  they receive are randomised.
• Trial also known as Brewin-Bradley or
  Comprehensive Cohort Design.

7
Patient Flow in Preference Trial
8
Preference Example

• A trial of two methods of abortion medical
  termination (mifepristone) vs surgical
  aspiration.
• Some women had strong preferences and therefore
  were allowed their treatment choice.

9
Abortion Trial
Heshaw BMJ 1993307714-7.
10
Data from Abortion Study

• The extra benefits of the preference study showed
  that clinically there was no difference.
• HOWEVER, women with a preference should be
  allowed their choice women who were indifferent
  with late gestation would find surgical abortion
  more acceptable.

11
Comprehensive Cohort

• In a comprehensive cohort study, Porthouse looked
  at the fracture rates among women who took part
  in a fracture prevention trial compared with
  those who were either ineligible or would not
  participate.

12
Comprehensive Cohort Design
Source Porthouse MSc thesis and QJM 200497569.
13
Results

• Those taking part had significantly lower risk of
  fracture compared with similar, eligible, women
  refusing to take part.
• Recruitment to fracture prevention trials selects
  individuals who are at lower risk than those for
  whom the treatments will eventually be used.

14
Problems with Preference Design

• Because preference arms are not formed by
  randomisation they WILL be exposed to selection
  bias.
• This makes the comparisons of these arms
  hazardous.

15
Patient Flow in Preference Trial
16
Preference Recruitment

• Trial recruitment is unaffected by the inclusion
  of preference arms EXCEPT for the extra resources
  needed to follow-up the preference arms.
• Cooper et al. undertook a RCT of the preference
  design and found no advantage to it.

Cooper et al. Br J Obs Gynae 19971041367-73.
17
An Alternative?

• One approach to preserve the benefits of
  randomisation is to undertake a fully randomised
  preference trial.
• ALL patients would be randomised irrespective of
  their preferences and preference would be used in
  the analysis.

Torgerson et al. 19961194-7.
18
Example

• York Back Pain trial randomised 187 people with
  low back pain to an exercise programme or control
  (benign neglect?).
• Before randomisation patients were asked their
  preferences.
• 63 expressed a preference for the new treatment
  37 had no preference.

19
Backpain Preference Trial
Klaber Moffett et al. BMJ 1999319279-83.
20
Back Pain Trial

• In the back pain trial we were able to show that
  the intervention was just as effective among
  patients who were indifferent to having the
  exercise therapy compared with those who were
  really keen.

21
Antenatal Care

• In a trial of increased visits to women for
  antenatal care it was found that women with a
  strong preference for the alternative were much
  more dissatisfied with treatment.

22
Antenatal Trial Dissatisfaction with Treatment
Clement et al. 1998 BMJ 31778
23
The most interesting example

• SPRINTER is a RCT of treatments for neckpain.
• Two treatments a Brief Intervention (1-2
  sessions with a physio using CBT) vs usual care
  (5 sessions).
• BEFORE randomisation we asked patients their
  treatment preference.

24
SPRINTER Preferences

• In SPRINTER preferences were mixed
• 53 did not have a preference
• 16 wanted brief intervention
• 31 wanted usual care.
• ALL patients were randomised IRRESPECTIVE of
  their preference.

25
Patient Flow and 12 month Results - SPRINTER
Overall 12 month improvement -0.840
Overall 12 month improvement -2.825
26
SPRINTER Results
27
SPRINTER interpretation

• Had we not asked for preferences we would have
  concluded usual care is best for all.
• BUT we can now say that BI is best for those who
  want that treatment (also much cheaper) and UC
  should be reserved for those who want it or
  indifferent patients.

28
Where now preference?

• In MY view if there is likely to be a problem
  with preferences we should elicit these at the
  start of the trial and include them in the
  analysis.
• Patient preference trials of the Brewin-Bradley
  design are fraught with analytical problems
  selection bias.

29
Doctor Preference

• As well as patients one could use doctor
  preferences to allocate treatment.
• In a RCT of orthopaedic surgery vs orthopaedic
  medicine patients were only randomised if GP was
  indifferent to the specialist needed.
• Patients with a named consultant were included
  and followed.

Leigh-Brown et al. 2001 Health Bulletin
59198-210.
30
Doctor Preference

• In the OMENS trial Leigh-Brown et al found that
  outcome did not seem to be affected by physician
  preference.
• Outcomes were similar across groups with
  orthopaedic medicine being more cost effective.

31
Summary

• Patient preferences CAN affect outcome.
• Elicitation of preferences when this is an issue
  BEFORE randomisation can be important.