Crossover trials:
key points
i)
The crossover design
The
trials considered hitherto have allocated just one of the treatments to each
patient. These trials are called
parallel group trials. For condition
that cannot be cured, trials which try more than one treatment on each patient
might be useful. This is because each
patient is being used as their own control – in statistical terms, the
within-patient variance is likely to be smaller than the between patient variance. If you give treatment A first and treatment B
second to all patients, then the difference in treatments cannot be
distinguished from any trend or difference over time. Such differences are known as period
effects. To allow estimation of
treatment effect in the presence of a possible period effect, some (usually
about half) the patients are allocated to A then B and the remainder to B then
A.
ii)
Model including period effect
To analyse continuous data from a crossover trial in which patients are
allocated to either AB or BA, a model which includes terms for treatment effect
and period effect, as well as error terms for within and between patient
variation is used. The treatment effect
can be estimated using within-patient information only. What this effectively means is that the
between-patient error term is eliminated from the analysis.
iii)
Carryover effect
A possible complication of the model is that the effect of a treatment given in
the first period persists into the second period. If this is the same for both treatments, it
amounts to a change in the period effect and is of no consequence. However, if the degree of persistence differs
between the treatments, this is known as a carryover effect of treatment. If this occurs, then the within-patient
estimated obtained in ii) is biased.
Attempts to determine if there is a carryover effect requires the use of
between patient information. The only
reliable way to exclude carryover is by non-statistical arguments.