Phase II Design Strategies

Sally HunsbergerOvarian Cancer Clinical Trials

Planning Meeting May 29, 2009

Single arm Phase II study

• Fact or fiction:– Using single arm phase II study designs reduces

the number of patients needed in drug development

• True: If the specified null rate is correct• How bad can things get if the null rate is

specified incorrectly?• Need to consider the drug development cost

(in terms of patients)– End of phase II if don’t go to phase III– End of phase II if go to phase III

Evaluation of designs

• Look at expected sample size E[N]• E[N]=NII+NIII(P{continuing to phase III})

Phase II Design parameters

• PFS as primary endpoint• Type I error and II error of .1• Median null PFS = 3 months• Interesting activity would result in a median

PFS of 4.5 (hazard ratio=1.5)• Minimum follow up=3 months• Sample size = 69

Phase III design parameters

• OS as primary endpoint• Type I error 1-sided .025 • Median null OS = 6 months• Interesting activity would result in a median

OS of 7.8 (hazard ratio=1.3)• Minimum follow up = 6 months• Sample size = 692

Under the null of no treatment benefit

• What happens if we set the null bar too low– Go to phase III too often and this will increase the

Expected sample size,E[N].

True Median PFS P{continuing} E[N]

3* .1 138

3.5 .4 346

4 .72 567

*Truth agrees with assumption.E[N] for a randomized phase II study is 271 with α=β=.1

Under null hypothesis of no Treatment effectNull assumption is 3 months

Under the alternative of a treatment benefit

• What happens if we set the null bar too high– Do not go to phase III often enough– This will decrease power of finding a treatment

benefit at the end of drug development

True Median PFS P{continuing} Probability of concluding a OS benefit at the end of Phase III

3* .9 .81

2.5 .59 .53

2 .1 .09

*Truth agrees with assumption.Probability of concluding a benefit when a randomized phase II study is used .81

Under Alternative hypothesis of a Treatment effectNull assumption is 3 months

Treatment benefit of a hazard ratio of 1.5

If we need a randomized phase II how can we speed up drug development

• Phase II/III design– Futility analysis based on PFS– Study power for a conclusion on OS

Simulation study results

• Performed simulation study so I could have correlated PFS and OS

• Comparison designs:– Sequence of a randomized phase II study and then

a randomized phase III– Skip phase II go right to a phase III with a futility

analysis based on OS (appropriate if you don’t expect an effect on PFS)

Designs Global Null Global Alternative

α1 t1 E[N] E[T] E[N] E[T]

Futility based on overall survival .2 24.0 427 28.5 649 43.2

.5 11.9 433 28.9 627 41.8

Sequence of Phase II and Phase III .1 15.1 296 23.3 849 65.0

Integrated II/III with (f1=0) .05 20.4 325 21.7 646 43.1

.1 16.7 294 19.6 644 42.9

.2 12.3 287 19.2 634 42.3

.5 6.1 391 26.0 625 41.7

Integrated II/III with (f1=3) .05 18.3 295 22.7 644 46.0

.1 14.7 268 20.9 640 45.7

.2 10.8 268 20.9 633 45.2

.5 4.2 378 28.2 623 44.5

Over all probability of concluding a benefit when it exists is .81

Conclusions

• Single arm studies may appear to use less patients but if the null bar is set incorrectly this could have a major impact on E[N] and the probability of identifying a beneficial treatment– When there is no true treatment effect setting the

bar two low increases the E[N]– When there is a true treatment effect and the bar is

set too high the probability of identifying a beneficial treatment is reduced

Conclusions

• integrated phase II/III design works well under the global null. – E[N] and E[T] were no larger than that of a randomized

phase II study – E[N] and E[T] smaller than skipping the phase II study

• integrated II/III better than the separate randomized phase II study under the global alternative – did not increase E[T] and E[N] when compared to

skipping the phase II component.