CONFIDENCE - Radiation Oncology: A Physicist's-Eye View - page 291

actuarial survival curve) than the control arm; or (b) what the
difference in 5-year survival might be. 3
The combination of local control and morbidity
One often wants to assess the difference in local control for two
therapies. 4 But, this seemingly straightforward goal is complicated by
the fact that local control is tempered by morbidity. The “goodness”
of a therapy is some sort of a combination of the likelihoods of tumor
control (TCP) and morbidity - and we often don't really know how to
combine these two into a single measure of goodness. Moreover,
morbidity is not a singular quantity. The patient is at risk for a variety
of complications of different severity, and of variable importance
relative to one another and to local tumor control. This is essentially
the same problem as has already been discussed in Chapter 9 in the
context of the optimization of treatments.
A way out of this conundrum is based on the fact that usually the
intensity of a therapy can be adjusted, and the likelihood of morbidity
(and of local tumor control) is a function of the treatment intensity.
One can then ask for example whether, relative to the control arm, the
experimental arm, adjusted in intensity to give the same likelihood of
morbidity as the control arm , yields a higher probability of local
tumor control.
While this is a recognized problem, most RCTs in radiation oncology
are nevertheless restricted to two arms, largely to ensure adequate
patient accrual. Thus, it is not uncommon to find, after a trial has
been concluded, that it is hard to draw useful clinical conclusions
because, say, the experimental arm has simultaneously shown an
improved TCP and increased morbidity. This is the situation depicted
schematically in Figure 13.1a.
3 The P statistic comparing two actuarial survival curves measures the
likelihood that their shapes are different, taken as a whole . It is important
to recognize that two curves might be statistically very different, but have
(within stated confidence limits) the same five-year survival. This would
perhaps be due to the two therapies having a rather different outcomes in
the early years, but leveling out to the same survival level at later times. If
one were primarily interested in long term survival, then the P statistic in
this case might be somewhat misleading.
4 For simplicity, I confine my discussions to the comparison of two different
therapies. Of course, multiple therapies can be inter-compared in multiple-
arm studies.