CD33 expression and impact on outcome according to treatment arm. (A) The concordance between the percentage of CD33⁺ blast cells and the CD33 MFI ratio. To properly calculate the CD33 MFI ratio, at least 500 lymphocytes and blast cells had to be acquired and fewer than 20% of acquired lymphocyte events had to fall in channel 0 at the electronic digitizing step of fluorescence light intensities for CD33 labeling. The 2 boxes separate low/int-CD33⁺ and high-CD33⁺ patients on the basis of a threshold of 70% for CD33⁺ blasts and of 15 for CD33 MFI ratio. (B) In low/int-CD33+ patients (<70%), EFS was estimated at 23% vs 26% at 2 years in the GO and control arms, respectively (HR, 0.89; 95% confidence interval [CI], 0.52-1.50; P = .66). Conversely, in high-CD33⁺ patients (≥70%), 2-year EFS was 49% vs 17% in the GO and control arms, respectively (HR, 0.56; 95% CI, 0.37-0.85; P = .0051). Tests for interaction did not reach statistical significance (P = .155). (C) In low/int-CD33⁺ patients (<70%), RFS was estimated at 25% vs 38% at 2 years in the GO and control arms, respectively (HR, 1.17; 95% CI, 0.61-2.24; P = .66). Conversely, in high-CD33⁺ patients (≥70%), 2-year RFS was 62% vs 22% in the GO and control arms, respectively (HR, 0.47; 95% CI, 0.29-0.76; P = .0016). Tests for interaction reached statistical significance (P = .022). (B-C) The 29 patients who received allogeneic SCT in first CR/CRp were not censored at the time of SCT.