Application of Translational Science to Clinical Development - Genomic Biomarkers for Pharmaceutical Development - page 5

Patients restricted to biomarker positive
group 80% power; alpha=0.05

All patients
80% power; alpha=0.05

30 30

20% biomarker + prev.
30% biomarker + prev.
40% biomarker + prev.
50% biomarker + prev.
60% biomarker + prev.

6% response in ctrl arm
9% response in ctrl arm
14% response in ctrl arm

25 25

20 20

15 15

10 10

100 200 300 400 500

0 1000 000 000 5000

n 1 + n 2 n 1 + n 2

FIGURE 1.4 (left) Relationship between effect size and total sample size when restricting patient inclusion to
biomarker positive patients under different control arm response rates (80% power and alpha = 0.05), and (right)
the same association showing different levels of biomarker positive patient prevalence, assuming 6% control arm
response rate with no restriction to biomarker positive patients, and only the biomarker positive patients showing
improvement (80% power and alpha = 0.05). Note that total response delta is plotted on the y axis (right), though
sample sizes are calculated using the reduced response delta as explained in the text.

the control arm to as high as N = 259 / arm with a 10% effect size and 14% response rate in
the control arm [10] ( Fig. 1.4 left).
In contrast to this trial design where only biomarker positive patients are included, the
sample size requirement in an all-comers trial design, i.e., without selectively enrolling
patients, is driven not only by effect size, but by the prevalence of patients with the partic-
ular disease sub-type (e.g., NSCLC patients with ALK fusion). For example, assuming a 6%
response rate in the control arm, if the trial is not restricted to biomarker positive patients (i.e.,
those with ALK fusion in this example), and we assume the same effect size in the previous
example of 30%, if 10% of the patients are identified as biomarker positive (and only these
patients show improvement), the overall improvement rate would be reduced to 3%. Under
this design, 1274 patients / arm would be required at alpha = 0.05 and 80% power. If the bio-
marker positive patient prevalence is identified to be 30%, the reduced effect size is 9% and
expected sample size is reduced to 202 / arm, under the same assumptions ( Fig. 1.4 right). This
example illustrates how easily sample size requirements can be affected by either reduced bio-
marker positive patient prevalence or decreased effect sizes in a clinical trial.
The second lesson from the crizotinib example is the importance of developing strong
testable hypotheses early. Although developing robust and reliable hypotheses is often eas-
ier said than done, with the right approach equipped with the powerful technologies we
currently have, such hypotheses are not out of reach. Fortuitously, in the case of crizotinib
one of the clinical sites in enrolling patients in the Phase I trial was already developing tools