|
|
Cell Biochemistry Martinsried |
= external links
1. Background: AML patients are frequently stratified by prognostic parameters into therapeutic subgroups (L2). Stratification helps therapy susceptible patients but is of no use to non responders. There is significant clinical interest to identify non responder patients pretherapeutically for individualised therapy adaptation or switch to alternative therapies.
2. Goal: Classification of the SHG-96 multicenter AML trial database (L2) to identify high risk AML-patients prior to therapy (L1).
3. Results:
The
algorithmic
(non parametric) classification of the available
immunophenotype, cytogenetic and clinical
parameters
(fig.1)
shows that predictive values of 100% for 5-year nonsurvival
and of 88.6% for 2-year nonsurvival
(fig.2)
are obtained.
The discriminatory data patterns (disease classification masks)
contain 7 parameters for the 5-year classification and 12 parameters
for the 2-year classifications
(fig.3) .
Patient age and %CD4, %CD45 positive AML blasts are equally selected
in both classifications. The other parameters
of the disease classification masks are different.
The reclassification of the learning set
shows that correct classification is obtained
in most instances with mask
coincidence factors between 0.57-1.00. This indicates
that already a partial fit of the patient classification masks
with the two disease classification masks >5-year survivors
and 5-year nonsurvivors may be sufficient for
correct classifications
(fig.4).
4. Conclusion and Outlook:
Immunophenotype parameter patterns
identify high risk patients with high predictive values.
Cytogenetic parameters were not
selected, probably because they occur
in only about half of the patients
as opposed to the CD antigen expression on all cells.
It seems promising to perform multiparametric CD
measurements according to the disease classification
masks of this study. Antigen expression, antigen ratios
and scatter of the antigen distributions have then to be determined
in addition to cell frequency to further increase the predictive
values to >95% in the learning and unknown test sets of patients.
| © 2024 G.Valet |