CLASSIF1 Classification

• < AML

< fig.1      fig.2: Confusion Matrices      > fig.3

- The database of the ongoing trial contained 724 patients (July 2001) with 159 2-year survivors, 404 2-year nonsurvivors, 161 < 2-year still surviving patients, 25 5-year survivors, 442 5-year nonsurvivors and 257 < 5-year still surviving patients. This means that 563 patients were included in the 2-year classifications and 467 patients in the 5-year classifications. 5.6% and 1.9% of the included patients could not be classified due to multiple classifications.
- The first 300 included patients served as learning set while the remaining patients served as unknown test set.
- Prospective classifications were obtained for the patients not fulfilling the inclusion criteria (161 < 2-year and 257 < 5 year still surviving patients)
- The classification detects 5-year nonsurvivor patients in the learning set with a predictive value of 100% and 5-year survivors with 15.1% (fig.2a). The 5-year survivors in the test set (fig.2b) classified with a sensitivity of 91.5% better than in the learning set (49.8%). No predictive values could, however, be calculated due to the absence of 5-year non survivrs in the test set.
- The predictive values for 2-year nonsurvivors was 88.6% and 38.9% for 2-year survivors (fig.2c). The unknown test set classified similarly as the learning set (fig.2c).

• < AML