Predictive Capacity of Peri-Transplant Measurable Residual Disease Thresholds in NPM1-Mutant Acute Myeloid Leukemia Open Access

• NPM1-MRD thresholds distinguish MRD-high patients with poor prognosis from MRD-low and -negative patients with favorable outcomes

• Peri-transplant NPM1 MRD thresholds are more accurate predictors of outcome than traditional binary or log-change based MRD interpretation

Minimal residual disease (MRD) monitoring for mutated NPM1 is increasingly used to guide treatment decisions in patients with acute myeloid leukemia (AML) carrying this mutation. NPM1-MRD positivity after induction was shown to identify patients who benefit from allogeneic hematopoietic cell transplantation (alloHCT), and NPM1-MRD monitoring after alloHCT can detect early relapse enabling the prompt initiation of salvage therapy. However, recommendations for clinical decision making based on peri-transplant NPM1-MRD levels are missing. In this study, we retrospectively analyze 172 patients with NPM1-mutant AML treated at two German centers to explore the predictive values of NPM1-MRD measured pre- and post-alloHCT. We found that pre-transplant MRD negativity was a strong predictor of favorable long-term overall survival (OS). In contrast, patients with positive and negative NPM1-MRD status at day 30 post-HCT showed comparable OS. Finally, statistically derived NPM1-MRD thresholds effectively stratified MRD-high and MRD-low patient groups with differential outcome with two peri-transplant MRD risk scores obtained by longitudinal integration. Firstly, a combined pre-HCT/+d30 post-HCT score to guide early reduction of immunosuppression (C-index, 0.737), and secondly, a combined pre-HCT+d30+d100 score to guide later interventions post-HCT (C-index, 0.841; stratified 2-year OS groups 100%, 90.1%, 57.1% and 25.7%; p<0.0001). This approach predicted OS better than age, FLT3-ITD status, or morphological remission status. We propose that in the peri-transplant setting, NPM1-MRD thresholds are superior to conventional MRD analysis based on binary or log-step change data.