Abstract
Objective
The analogous immunophenotype between HLA-DR-negative acute myeloid leukemia (AML) and acute promyelocytic leukemia (APL) causes APL rapid screening to become difficult. This study aimed to identify the associated antigens for APL and the best model in clinical uses.
Results
A total of 36 APL (PML-RARA+) and 29 HLA-DR-negative non-APL patients were selected for this study. When a cut-off point of 20% events was applied to define positive or negative status, APL and non-APL patients share a similar immunophenotype of CD117, CD34, CD11b, CD13, CD33, and MPO (P > 0.05). However, expression intensity of CD117 (P = 0.002), CD13 (P < 0.001), CD35 (P < 0.001), CD64 (P < 0.001), and MPO (P < 0.001) in APL are significantly higher while CD56 (P = 0.049) is lower than in non-APL subjects. The Bayesian Model Averaging (BMA) analysis identified CD117 (≥ 49% events), CD13 (≥ 88% events), CD56 (≤ 25% events), CD64 (≥ 42% events), and MPO (≥ 97% events) antigens as an optimal model for APL diagnosis. A combination of these factors resulted in an area under curve (AUC) value of 0.98 together with 91.7% sensitivity and 93.1% specificity, which is higher than individual values (AUC were 0.76, 0.84, 0.65, 0.82, and 0.85, respectively) (P = 0.001).
When and when not to use optimal model averaging
