N = 1, Cross-Over Trials and Balanced Designs

1
N 1, Cross-Over Trials and Balanced Designs
2
N 1 Trials

• Trials can be undertaken with just one
  participant. If the condition is a chronic
  relapsing problem the unit of randomisation can
  be the episode of treatment.
• For example, n 1 trials have been used among
  people with rheumatoid arthritis to try out
  different pain relieving drugs.

3
Migraine Trial

• A trial could be mounted to allocate a test
  treatment for each migraine attack.
• Placebo or open treatment could be used.
• Once the individual senses an attack coming on he
  contacts an independent statistician and she
  gives him the allocation.

4
Blinding of allocation

• Important to blind allocation as the migraine
  sufferer may ask for treatment at different
  stages of the attack if he knew which allocation
  might come up.
• Placebo might be useful but any effect including
  a placebo effect is worthwhile.

5
Example

• Some suggestion that blood sugar falls prior a
  migraine perhaps boosting this with glucose
  tablets may limit attack?
• Randomised to either 2 asprin or 2 asprin plus 3
  gluscose tables.
• Outcome return to normal activity.

6
Sample Size

• This is kept small (18 migraines) because it is
  paired data (within the individual) so variance
  is low.
• Low significance level p 0.20 (treatment
  harmless, consequences of Type I error low, Type
  II error large).

7
Blocking?

• Not used although one could stratify by severity,
  but due to single individual could easily lead to
  prediction of block size.
• A simple allocation schedule used which is a
  single balanced block of 20 migraines to avoid
  prediction of last few allocations.

8
Time lines

• Average migraine once a month so trial is
  unavoidably long (approx 2 years).
• But dramatic effect could lead to early stopping.
• Sadly no such occurrence.

9
Cross-over trials

• Cross-over studies are when participants are
  randomly allocated to one treatment for a period
  of time and then cross-over to the other
  treatment for the second period of time.
• Some HRT trials for treatment of menopausal
  symptoms have used cross over designs.

10
Advantages

• The advantages of a cross-over trial is that is
  substantially strengthens the inference of an
  effect as essentially one is repeating the trial
  twice and because data are paired within
  individuals usually smaller sample sizes are
  needed.

11
Disadvantages

• Cant be used in many if not most circumstances.
• Cant cross people over from a cancer treatment
  to a placebo and vice versa.
• Even when they are feasible there is can be a
  problem with washout or contamination if the
  drug remains in the patients system will dilute
  the effects.

12
Disadvantages (cont)

• If there is a dramatic effect of treatment people
  may refuse to cross over to new treatment.
• For example, an RCT of dietary exclusion of foods
  for IBS suffers versus sham diet, found a big
  effect in the first period BUT those in the
  real diet who benefited refused to cross to a
  sham diet.

13
Balanced Incomplete Block Designs

• Trials of guidelines or educational packages may
  use an incomplete block design.

14
Avoiding Hawthorne Effects

• One way of avoiding a Hawthorne effect is to use
  a balanced design whereby the control group
  also gets an intervention.
• This can be done using a balanced design.

15
Example

• A guidelines trial might look at the
  effectiveness of guidelines for Angina. Just
  giving guidelines to GPs might affect practice
  irrespective of the benefit of guidelines.
• We can balance this possibility by giving
  guidelines say on diabetes to the control
  group.

16
Balanced Design

• This design will balance out the Hawthorne
  effects of guidelines being given to one group
  alone.
• Also allows the evaluation of two sets of
  guidelines for the price of one study.

17
Example

• Verstappen et al wanted to control for the
  Hawthorne effect in a trial looking at an
  intervention for test ordering from GPs.
• They also undertook some methodological research
  to see if the Hawthorne effect was real.

Verstappen et al. J Clin Epidemiol 2004571119
18
Design

• GP teams randomised into 3 arms.
• A tests feedback on test ordering for
  cardiovascular, upper and lower abdominal
  problems
• B tests feedback on test ordering for
  pulmonary, non-organ related and joint
  complaints.
• C minimal intervention control group.

19
Results

• A comparison between B vs A test groups shows a
  small (not significant, p 0.29) difference. B
  vs C shows a much bigger nearly statistically
  significant difference ( p 0.07)

20
Conclusion

• The change in test ordering was exaggerated by
  process of research compared with the actual
  intervention itself. Whilst the intervention had
  an effect this was relatively modest.

21
Summary

• Both n of 1 trials and cross-over studies are not
  widely used. N of 1 trials can be combined with
  several individuals leading to a greater power
  and shorter study.
• Balanced designs are useful in eliminating the
  possibility of Hawthorne effects.