3. Group Means
A in Period I B in Period II
Group 1: Y1 Y2
B in Period 1 A in Period II
Group 2: Y3 Y4
4. For Full Factorial Designs
Effect Alias
TREATMENT GROUP*PERIOD
PERIOD GROUP*TREATMENT
TREATMENT*PERIOD GROUP
5. Example
Group 1 Group 2
Estimate of Group Effect
(Y1 + Y2) – (Y3 + Y4)
Period 1 Period 2
Estimate of
Period*Treatment
A-B A-B
(Y1 - Y3) – (Y4 - Y2)
They are identical!
6. Main Effects will thus be
Effect Alias
TREATMENT GROUP*PERIOD
PERIOD GROUP*TREATMENT
TREATMENT*PERIOD GROUP
Always confounded with interactions!
7. We examined
The acceptance (α) level and power of grouped,
crossover bioequivalence studies using an
analytical model that controls the effect of
group and to compare the results with those
without a group effect in the model
8. Methodology
Pharmacokinetic data (AUC and Cmax)
The "standard" ANOVA model was described as:
Y=group sequence subject (group * sequence) period
(group) treatment treatment * group
The "partial" model without a group effect:
Y=sequence subject (sequence) period treatment
9. Results
The standard analysis showed a significant
confounding effect of group and subject-group-
sequence interaction
Even though the effect was significant, no
obvious changes on the α-level and power were
found when the group term was eliminated
from the analytical model for both AUC and Cmax
10. Results contd.
Only when the intra-subject variance is high
(CV >30%) and sample size is insufficient (N
≤12), might a change in the co-variable play a
significant role
11. Conclusions
No dramatic influence on the probability of making
statistical errors was observed when the group effect
was excluded from the model, even when the group
effect was significant
Only when the intra-subject variance is high and
sample size is insufficient, might a change in the co-
variable play a significant role
Controlling the group effect, however, is still necessary
for an accurate modeling of the study design