Objective: To determine whether a crossover design results in a differ
ent estimate of treatment effect compared with a parallel design. Desi
gn: With the aid of a computer program, data sets of subfertile patien
ts were simulated under different assumptions. These patients were sub
jected to 2 treatments that were compared in either a parallel-design
or a crossover-design trial. Results were analyzed using logistic regr
ession. Setting: University hospital. Patient(s): Simulated patients o
f a heterogeneous subfertile population. Intervention(s): Two treatmen
t modalities with a pregnancy rate of 10% and of 20% in the first cycl
e were offered for 6 cycles to simulated heterogeneous populations. Ma
in Outcome Measure(s): After 1,000 simulations for each assumption, me
dian pregnancy rates and odds ratios were compared between the crossov
er- and parallel-design trials. Result(s): No relevant difference in e
stimated treatment effect was found between the designs. The crossover
design resulted in more pregnancies overall than the parallel design.
Conclusion(s): In infertility research, parallel and crossover design
s will lead to about the same results. Although the crossover design s
howed a slight tendency to overestimate the treatment effect of the mo
st effective treatment, this overestimation is clinically not relevant
and is small in relation to the random error. Because of its practica
l advantages and because more pregnancies are achieved, a crossover de
sign should be the first choice in infertility research. (Fertil Steri
l(R) 1998;70:40-5. (C)1998 by American Society for Reproductive Medici
ne.).