| |
Example
2: A cross-over trial was run to compare two doses of the potassium
salt of diclofenac (50 mg and 100 mg) to placebo in the treatment
of migraine (Dahlof
and Bjorkman, 1993). The outcome variable was pain intensity
two hours after treatment measured on a 100mm visual analog scale
running from 0 = no pain to 100 = unbearable pain. The trial was
run in six centers in Sweden and Finland and 72 patients were
treated. If D1 and D2 represent the doses of diclofenac-k and
P represents placebo, then patients were randomized to all six
possible sequences of one treatment each, as illustrated.
| Figure
1.2: Cross-over trial using Latin squares
|
|---|
|
|
This is
an example of a very common sort of design used to study three
treatments. Sometimes, instead of using all six sequences, three
are chosen to form a so-called Latin square, the design as a whole
having the property that each treatment appears once in each sequence
and once in each period. In the design above, the first three
or last three sequences taken together form Latin squares.
|