Avid Bioservices Inc (CDMO)

[freethemice]

Highly relevant article. The author is at BMS, Global Clinical Research.
Annals of Oncology, Volume 23, suppl 8, Sept 2012
Symposium on Advances in Immuno-oncology
http://annonc.oxfordjournals.org/content/23/suppl_8/viii47.full.pdf+html
Evolution of end points for cancer immunotherapy trials
A. Hoos
[snip]
overall survival
A delayed separation of Kaplan–Meier survival curves is
observed in almost all randomised immunotherapy trials, and
may occur months after the start of treatment. This is in
contrast to the Kaplan–Meier survival curves obtained in
clinical trials of chemotherapy, where early clinical effects are
achievable [4, 16]. The delayed separation of survival curves
associated with immunotherapy reduces the ability for
statistical power to differentiate between them [16, 17]. Clinical
trials that have a delayed response deviate from the standard
model used for randomised trials that assume a proportional
hazard, i.e. that no events occur before the separation of curves
and that the hazard is constant over time [18].
The delayed separation of curves has been reported in many
clinical trials of immunotherapeutic agents, including a
placebo-controlled phase III trial of the autologous active
cellular immunotherapy sipuleucel-T, where the
immunotherapeutic effect on survival was not evident for 8
months [19], and the first phase III trial of ipilimumab, where
separation of the survival curves for patients treated with
ipilimumab alone, ipilimumab plus gp100 vaccine or gp100
vaccine alone, was not observed until 4 months after the
initiation of treatment [20]. In both of these examples,
although immunotherapeutic activity occurred before the
separation of the Kaplan–Meier curves, it did not translate into
a survival difference between the curves or agents. As such, any
event that occurs before the separation of the curves is a ‘lost’
event in terms of the final analysis, as it does not contribute to
the effect that follows the separation of the curves.
It is clear that the delayed response, which is often
substantial, has a huge impact on the dynamics of the trial. To
describe the implications of the delayed separation of Kaplan–
Meier curves, a model scenario can be applied. The curves are
separated into two components: component 1 is no separation
of the curves (i.e. no difference), where the hazard ratio (HR) is
equal to 1, and component 2 is a separation of curves, which
requires a large delta value to power a statistically significant
difference between the two curves (Figure 4). The large delta
value is essential in the second component of the curve to
compensate for the lack of separation during the first
component of the curve. For example, consider a trial of 800
patients; if 200 events occur during the first component of the
curve (i.e. where HR: 1), then these are in effect lost events due
of the nature of the proportional hazard assumptions. This is
important because if, for example, 600 events were required for
the final analysis of survival, but only 400 events were
available, the final analysis would be under powered. In the
event that a treatment effect was observed, it may not be
statistically significant if the power of the study is only 50%.
This was not the case for the two phase III trials mentioned
earlier, as these had strong treatment effect responses that
overcame the delay in responses. Conversely, there are
examples where the treatment effect responses were not as
strong and the trial had a negative result as a consequence. In a
phase III trial of tremelimumab, for example, an early interim
analysis for survival showed no survival benefit, and as a result,
the study was terminated. An extended follow-up, however,
revealed an eventual separation of the survival curves [3].
Immunotherapeutic agents may differ with respect to the
presence and timing of delayed separation of the survival
curves. Therefore, exploiting our current knowledge to apply
adequate statistical modelling to describe HRs as a function of
time, and to differentiate them before and after the separation
of curves (ideally in randomised phase II trials), may improve
our ability to plan for pivotal phase III trials.