The American Joint Committee on Cancer proposes the following criteria
for evaluating putative prognostic factors: they must be (1) signific
ant, (2) independent, and (3) clinically important. Furthermore, we su
ggest the criteria for selecting a prognostic system that includes TNM
and new prognostic factors. These criteria are: (1) easy for physicia
ns to use; (2) provides predictions for all types of cancer; (3) provi
des the most accurate relapse and survival predictions at diagnosis an
d for every year lived for each patient; (4) provides group survival c
urves, where the grouping can be by any variable including outcome and
therapy; (5) accommodates missing data and censored patients and is t
olerant of noisy and biased data; (6) makes no a priori assumptions re
garding the type of data, the distribution of the variables, or the re
lationships among the variables; (7) can test putative prognostic fact
ors for significance, independence, and clinical importance; (8) accom
modates treatment information in the evaluation of prognostic factors;
(9) accommodates new putative prognostic factors without changing the
model; (10) accommodates emerging diagnostic techniques; (11) provide
s information regarding the importance of each predictive variable; an
d (12) is automatic.