Dynamic prediction of relapse in patients with acute leukemias after allogeneic transplantation: Joint model for minimal residual disease

Introduction Relapse remains the leading cause of treatment failure after allogeneic hematopoietic stem cell transplantation (alloHSCT) in leukemia. Numerous investigations have demonstrated that minimal residual disease (MRD) before or after alloHSCT is prognostic of relapse risk. These MRD data were collected at specific checkpoints and could not dynamically predict the relapse risk after alloHSCT, which needs serial monitoring. Methods In the present study, we retrospectively analyzed MRD measured with multi-parameter flow cytometry in 207 acute myeloid leukemia (AML) patients (acute promyelocytic leukemia excluded), and 124 acute B lymphoblastic leukemia (ALL) patients. A three-step method based on joint model was used to build a relapse risk prediction model. Results The 3-year overall survival and relapse-free survival rates of the entire cohort were 67.1% ± 2.8% and 61.6% ± 2.8%, respectively. The model included disease status before alloHSCT, acute and chronic graft-versus-host disease, and serial MRD data. The time-dependent receiver operating characteristics was used to evaluate the ability of the model. It fitted well with actual incidence of relapse. The serial MRD data collected after alloHSCT had better discrimination capabilities for recurrence prediction with the area under the curve from 0.67 to 0.91 (AML: 0.66-0.89; ALL: 0.70-0.96). Conclusion The joint model was able to dynamically predict relapse-free probability after alloHSCT, which would be a useful tool to provide important information to guide decision-making in the clinic and facilitate the individualized therapy.