Goal:
Early identification of long term (10 years) melanoma
surgery survivors from 4 clinical (TD=tumor thickness,
TE=infiltration depth, TK=TD/TE, UL=ulceration) and 2
flow cytometric parameters (SP=% S-phase cells, AN=DNA aneuploidy)
CLASSIF1 Classification:
TD, TK and SP permit an around 80% correct (positive
predictive value) identification of 10 year melanoma surgery
survivors and non survivors.
Conclusion:
The selected value triplet of two clinical and one flow cytometric
parameter are a first approach to melanoma survival predictions. The
addition of more specific biomolecular cell parameters is
likely to further increase the predictive values
Elaboration of CLASSIF1 Classifiers:
CLASSIF1 classifiers are typically established from
data sets of clinically well characterized patients.
Upon reception of a data set, patients #1,5,10,15 ... etc of
each classification category remain a-priori inaccessible to the
learning process and constitute the embedded unknown
test set which serves to test the robustness of
classification for unknown samples.
The remaining patient data are assigned to the learning set
from which the classifier is learned according to the principles
described previously.
Once the classifier is available, its classification capacities
can be assessed by classification of the learning set
but more importantly by the classification of the unknown
test set patients.
A classifier which correctly classifies the learning set as
well as the test set of patients is suitable for a test
phase of prospective classification in the clinical environment.
The correctness of prospective classification is checked
for a certain time prior to classifier use in its
operational phase in routine practice.