local-regional therapy (surgery and radiotherapy)
may be effective. In prostate cancer, for instance,
radical prostatectomy is recommended only for
patients with stage T1 (clinically inapparent tumor
neither palpable nor visible by imaging), T2 (tumor
confined within prostate), or T3a (unilateral extra-
capsular extension of tumor through the prostate
capsule) cancer that is N0 (no regional node metasta-
sis) or NX (regional lymph nodes cannot be
assessed) and M0 (no distant metastases) or MX
(presence of metastasis cannot be assessed).
Predictive factors are clinical and laboratory
features that allow accurate prediction of the
response to a therapy; they include cell and marker
substance findings that relate to the molecular target
of therapy or to features of the cancer associated
with resistance to therapy. Because prediction
applies to a specific therapy, the predictive factors
for a cancer may vary among the therapies available
for treatment of the cancer. For example, overex-
pression of erbB2 in breast cancer is not a useful
predictive factor for response to endocrine therapy
but it is a predictive factor for response to trastuzu-
mab (Yamauchi
et al.
2001). The predictive factors
in use for a particular therapy can also be expected
to change over time as our understanding of the
molecular and cell biology of cancer allows for the
development of new and more accurate markers. It
is worthwhile noting that although some prognostic
factors have a role only in prognosis and some
predictive factors only in prediction, often the same
marker may serve as a prognostic and a predictive
factor (Hayes
et al.
1998).
Prognostic and predictive classification can be
either dichotomous or quantitative. Dichotomous
classification is based upon the use of a critical
value, or combination of values (for combination
testing), that identify the prognostic or predictive
category in which an individual likely belongs. If
the measured value of the prognostic or predictive
factor exceeds the critical value, he or she is
assigned to one category and if the measured value is
less than the critical value, he or she is assigned to
another category. The probability of belonging to
the assigned category can be calculated using Bayes
formula (as discussed in Chapter 5); when there are
two classification categories,
P[post]=
prevalence
$
FCC
1
prevalence
$
FCC
1
+ (
1
−
prevalence
)(
1
−
FCC
2
)
where
prevalence
is the prevalence of the assigned
category and the performance characteristics
FCC
1
and
FCC
2
are the fractions of patients correcting
classified in each of the two categories. Dichoto-
mous classification provides less information about
the individual than quantitative classification, in
which the probability of an individual belonging to a
particular prognostic or predictive group is calcu-
lated using the likelihood ratio associated with the
measured value of the prognostic or predictive
factor; for two classification categories,
P[post] =
prevalence
$
likelihood ratio
prevalence
$
likelihood ratio
+ (
1
−
prevalence
)
As discussed in Chapter 3, both of these classifi-
cation approaches can be extended to take into
account multiple classification categories and combi-
nations of factor values. An example of this has
been reported by Partin et al. (1997). The authors
present tables that list the probability of the patho-
logic stage of prostate cancer, as determined using a
quantitative classification approach, in patients with
localized disease. Four pathologic stages are consid-
ered: organ-confined disease, isolated capsular
penetration, seminal vesicle involvement, and pelvic
lymph node involvement. The probabilities depend
upon three prognostic factors: the plasma PSA
concentration, the TNM stage, and the Gleason
score (a histologic grading system for assessing
aggressiveness of prostate cancer). These same
authors have also generated a nomogram (Figure
11.8) based on the same three prognostic factors that
can be used to calculate the probability of recurrence
of cancer within five years following radical
prostatectomy (Kattan et al. 1998).
Ex vivo
drug sensitivity testing.
The prediction
of response to therapy based on plasma markers and
tumor cell predictive factors can be very useful in
planning chemotherapy for a patient but, until all of
the cellular factors that confer susceptibility to any
particular drug are known, there will remain uncer-
tainty in the prediction. Directly testing the suscep-
tibility of living tumor cells to drugs—something
similar to
in vitro
antimicrobial susceptibility testing,
but for cancer cells—could potentially circumvent
this problem and allow for highly reliable individual-
ized chemotherapy. Unfortunately, the very process
of studying cancer cells while keeping them alive
ex
vivo
can lead to alterations in the cells or preferential
survival of unrepresentative cells, thereby lessening
Cancer
11-16
