Validation of measurable residual disease as a surrogate endpoint in acute myeloid leukemia: A HARMONY Alliance study of European randomized trials

Abstract Introduction: Measurable residual disease (MRD) prior to consolidation therapy is a strong prognostic biomarker for relapse and long-term survival in acute myeloid leukemia (AML). Beyond its prognostic role, MRD is increasingly used to guide therapeutic decisions. A logical next step is regulatory acceptance of MRD as a (co-)primary endpoint in AML, as recently granted by the FDA for MRD in multiple myeloma. However, acceptance in AML is hampered by disease heterogeneity, variability in MRD assessments and a lack of trial-level validation. As MRD-guided treatment becomes more common, future trial outcomes may be confounded by such interventions. Thus, establishing MRD now as a surrogate endpoint is critical to enable accelerated drug approval in AML. Here, we evaluate both individual- and trial-level surrogacy of MRD assessed by multiparameter flow cytometry (MFC) and qPCR for mutant NPM1 using harmonized patient-level data from seven prospective randomized phase II/III trials. Methods: We included patient-level data from 1,858 adult with AML enrolled in trials conducted by AMLSG, HOVON-SAKK, SAL, and UK-NCRI, collected through the HARMONY Alliance. Eligible trials involved randomization to experimental or placebo treatment added to a standard intensive induction chemotherapy, with ≥20 patients per arm and per MRD subgroup. Patients were included if they had a MRD assessment after two chemotherapy cycles by either MFC or qPCR for mutant NPM1. Data were harmonized using the OMOP common data model. Following FDA guidance, we analyzed two levels of surrogacy. For individual-level surrogacy, we examined the association between MRD status and overall survival (OS) using Plackett's copula models and multivariable Cox regression. For trial-level surrogacy, we quantified the relationship between treatment effects on MRD and OS using hazard ratios (HR) and odds ratios (OR) in weighted least-squares regression. A coefficient of determination (R²) >0.8 with 95% CI lower bound >0.6 was considered strong trial-level surrogacy, consistent with previously accepted surrogate endpoints. Subgroup analyses were conducted by MRD method and transplant status. Results: The included trials were AMLSG 09-09 (qPCR MRD assessment), three HOVON-SAKK trials (AML-102, AML-103, AML-132; all MFC), the SAL cohort (qPCR), and the UK-NCRI AML17 trial, which included two separate randomizations (both MFC- and qPCR-MRD). In multivariable analysis, MRD positivity was associated with significantly worse OS across the cohort (HR: 1.66, 95% CI: 1.33–2.07, p<0.001). This association persisted when stratified by MRD method or treatment arm (placebo vs experimental). The global OR for the association between MRD status and OS was 0.39 (95% CI: 0.32–0.47); among transplanted patients, OR was 0.61 (95% CI: 0.42–0.81), and among non-transplanted patients, 0.33 (95% CI: 0.25–0.41). Trial-level surrogacy analysis was limited to MFC-based MRD (n= 1,268) due to the limited number of qPCR-based trials. The overall R² between treatment effect on MRD (OR) and OS (HR) was 0.91 (95% CI: 0.56–1.00), suggesting strong surrogacy, though the lower confidence interval bound fell below the predefined threshold. When restricting to non-transplanted patients, R² increased to 0.99 (95% CI: 0.94–1.00), whereas among transplanted patients R² was 0.54 (95% CI: 0.00–1.00), suggesting that transplant may attenuate the association between MRD and OS at the trial level. Conclusions: This pooled analysis represents the largest MRD dataset in AML to date and demonstrates that MRD after induction is a robust individual-level predictor of OS, whether measured by MFC or qPCR for NPM1. MRD retained prognostic value across treatment arms and in multivariable models. These findings reflect strong individual-level surrogacy: MRD-negative patients were more than twice as likely to survive as MRD-positive patients, although this difference was less pronounced in transplanted patients, suggesting that allogeneic transplant may partially mitigate the adverse impact of MRD positivity. Importantly, trial-level surrogacy was confirmed for MFC-MRD in non-transplanted patients, supporting its use as an intermediate endpoint reasonably likely to predict long-term outcomes in intensively treated AML patients. As MRD-guided therapy and maintenance strategies become standard, this harmonized prospective dataset provides timely evidence to support MRD as a regulatory surrogate endpoint for AML drug development.