AUTHORS
Melissa F. F Neumann, MD, Meredith F Akerman2, David M. Mueller, MD3, Nicholas Scelzi, DO4, Aaron Goldstein, MD5, Stephanie Boisclair, MD6, Bradley Goldberg, MD7, Erin Jou, MD8, Jonathan Kolitz, MD9, Kanti Rai, MD10, Douglas E Gladstone, MD11, David W Chitty, MD12
AFFILIATIONS
1 Northwell Health Cancer Institute, Lake Success, New York, United States, Northwell, New Hyde Park, New York, United States
2 Northwell, New Hyde Park, New York, United States
3 Northwell Health Cancer Institute, Lake Success, New York, United States, Northwell, New Hyde Park, New York, United States
4 Northwell, New Hyde Park, New York, United States
5 Northwell Health Cancer Institute, Lake Success, New York, United States, Northwell, New Hyde Park, New York, United States
6 Northwell Health Cancer Institute, Lake Success, New York, United States, Northwell, New Hyde Park, New York, United States
7 Northwell Health Cancer Institute, Lake Success, New York, United States, Northwell, New Hyde Park, New York, United States
8 Northwell Health Cancer Institute, Lake Success, New York, United States, Northwell, New Hyde Park, New York, United States
9 Northwell Health Cancer Institute, Lake Success, New York, United States, Northwell, New Hyde Park, New York, United States
10 Northwell Health Cancer Institute, Lake Success, New York, United States, Northwell, New Hyde Park, New York, United States
11 Northwell Health Cancer Institute, Lake Success, New York, United States, Northwell, New Hyde Park, New York, United States
12 Northwell Health Cancer Institute, Lake Success, New York, United States, Northwell, New Hyde Park, New York, United States
ABSTRACT
Background:
Hypomethylating agents (HMAs) ± venetoclax (VEN) offer a less-intensive therapy option for older/unfit acute myeloid leukemia (AML) patients. Early induction death occurs in 20% of patients receiving HMA-based regimens, underscoring the need for early identification of these patients to offer alternative therapies. We present an internally validated predictive model of 90-day mortality (90-DM) and a point-based risk score to guide treatment decisions at diagnosis.
Methods:
We performed a retrospective single-institution cohort study of 445 newly diagnosed AML patients (2022 ICC) treated with induction HMA±VEN between 2016 and 2025. Missing values were addressed using multiple imputation by chained equations (MICE; 20 imputations). Over 100 demographic, clinical, and molecular variables were tested individually (Mann-Whitney, chi-square, Fisher's exact); variables with p <0.25 entered the selection process for the multivariable model. LASSO regression was used to select variables, and only those chosen in ≥70% of datasets were retained. The final model was fitted within each imputed dataset and pooled by Rubin's rules. A point-based score was derived from pooled β coefficients, corresponding to predicted 90-DM rates. Internal validation used 1,000 bootstrap resamples within each imputed dataset to estimate a pooled optimism-corrected AUC. P <0.05 was significant.
Results:
Among 445 patients (median age 77 years; range 34–96; 44.5% non-Caucasian), all-cause 90-DM was 34.4%, with 92.8% of deaths due to therapy- or disease-related complications. Regimens included decitabine+VEN (58.7%), HMA monotherapy (22.7%), azacitidine+VEN (15.5%), and other HMA combinations (3.1%). Univariate predictors of early mortality included ECOG PS ≥2 (58.2 vs. 29.1%, P < .001), infection at start of induction (37.1 vs. 16.5%, P < .001), complex karyotype (47.7 vs. 30.8%, P < .001), LDH≥ 420 / 1.5x ULN (52.7 vs. 33.5%, P < .001), WBC ≥ 100 × 10⁹/L (10.5 vs. 4.1%, P = .009), CD56 expression (30.1 vs. 19.2%, P = .009), and del(5q) (24.2 vs. 14.0%, P = .008).
Of 25 candidate variables, 11 were retained in the final model. Increased odds of 90-DM were seen with ECOG PS 2 (OR=3.0, 95% CI: 1.8-4.9) and 3–4 (OR=9.4, 4.3-20.4), infection at start of induction (OR=2.9, 1.7-4.9), arrhythmia (OR=3.0, 1.0-8.7), WBC ≥ 100 × 10⁹/L (OR=2.5, 95% CI:1.0-6.4), LDH ≥ 420 (OR=1.9, 1.1-3.1), CD56 expression (OR=2.0, 1.2-3.5), complex karyotype (OR=2.3, 1.4-4.1), del(5q) (OR=1.8, 0.9-3.5), age [75-80: OR=1.3, 0.7-2.3, 80-90: OR=1.6, 95% CI:0.9-2.8), and KMT2A-mutated (KMT2Am; excluding 11q23) (OR=5.1, 1.7-15.9); presence of a RUNX1 mutation (OR=0.4, 0.2-0.8). Discrimination was robust (pooled OC-AUC=0.77; 0.72-0.81) with adequate calibration (Hosmer-Lemeshow χ² = 0.686, df = 8, p = 0.705). A point-based score corresponded to predicted 90-DM rates of 25% (11 points), 50% (16 points), 75% (20 points), and 99% (≥35 points).
Discussion:
This internally validated prognostic model incorporates 11 clinical, cytogenetic, molecular, and comorbidity-related variables into a point-based score for clinical use, with ECOG >3 (9 points) and non-11q23-KMT2Am (7 points) carrying the greatest weight. Remaining variables contributed less: ECOG 2, infection at start of induction, WBC ≥ 100 × 10⁹/L, arrhythmia history, and absence of RUNX1 (4 points each); complex karyotype, LDH ≥ 420, CD56 (3 points); del(5q) (2 points); and age (1-2 points). Risk stratification for 90-DM included low (≤25%, ≤11 points), intermediate (≥25 to <50%, 12-15 points), and high-risk (>50%, ≥16 points). Consistent with Lachowiez et al. (2026), RUNX1 mutations were favorable, possibly reflecting biology responsive to HMA-based therapy. Rather than identifying secondary AML, del(5q) emerged as a potential surrogate and adverse marker. Additionally, KMT2Am represented a biologically resistant subgroup after excluding 11q23-rearranged patients who had more favorable outcomes. Interestingly, CD56, an aberrant marker linked to extramedullary disease risk, was also adverse. Importantly, poor ECOG status is a well-established predictor of early mortality, and most clinical trials exclude patients ≥75 years with ECOG ≥3, increasing the need for such a calculator when applying this treatment in a real-world setting. External validation is ongoing to confirm generalizability and clinical utility.
DISCLOSURES
Melissa F. F Neumann, MD: No relevant conflicts of interest to declare.