Leukemia, Myeloid, Acute
Conditions
Keywords
Acute Myeloid Leukemia, Minimal Residual Disease, Flow Cytometry, FlowSOM, Gating Algorithm, Machine Learning
Brief summary
This retrospective multicentre cohort evaluates the agreement of measurable residual disease (MRD) detection in acute myeloid leukaemia (AML) using two flow-cytometry gating approaches. Manual expert gating is compared with an unsupervised FlowSOM clustering algorithm across post-induction and post-consolidation samples from 50 adults and 10 paediatric patients treated at Bordeaux University Hospital. The primary hypothesis states that unsupervised gating detects MRD ≥ 0.1 % with sensitivity and specificity comparable to manual gating.
Detailed description
Acute myeloid leukaemia remains associated with high relapse rates despite complete remission after induction chemotherapy. Sensitive identification of residual leukaemic blasts (MRD) guides risk-adapted therapy. Flow cytometry is applicable to nearly all patients but relies on operator-dependent manual gating, which may lack reproducibility when rare or immunophenotypically atypical blasts are present. A data-driven alternative based on FlowSOM clustering was developed at Bordeaux to overcome these limitations. DualFlow retrospectively analyses paired flow-cytometry standard (FCS) files from 60 AML patients (≈ 100 MRD determinations) drawn from the DATAML Bordeaux adult database and the paediatric haemato-oncology service. Files are distributed to three partner centers for blinded re-analysis. Each sample undergoes: (1) conventional manual gating in two expert centers; (2) unsupervised FlowSOM gating in one center; (3) molecular MRD assessment when available. Primary analysis calculates sensitivity, specificity, predictive values and Cohen/Fleiss kappa for MRD ≥ 0.1 %. Secondary analyses include concordance with molecular MRD, Bland-Altman and correlation for MRD 0.01-0.1 %, impact on relapse-free and overall survival using Kaplan-Meier and Cox models, and operator reproducibility for manual gating. Covariate effects (age, cytogenetics, molecular risk, treatment) are explored through stratified and multivariable methods. No additional interventions or specimens are collected; only de-identified FCS files and routine clinical data are used.
Interventions
DIAGNOSTIC_TESTConventional gating
Manual expert gating of multiparameter flow-cytometry data for MRD
DIAGNOSTIC_TESTAutomated gating
Unsupervised FlowSOM gating of multiparameter flow-cytometry data for MRD
Sponsors
University Hospital, Bordeaux
Study design
Observational model
COHORT
Time perspective
RETROSPECTIVE
Eligibility
Sex/Gender
ALL
Healthy volunteers
No
Inclusion criteria
* Confirmed diagnosis of acute myeloid leukaemia per ELN 2022 * Age ≥ 18 years (adult cohort) or 0-20 years (paediatric cohort) * Inclusion in DATAML Bordeaux database or paediatric haemato-oncology records * Available flow-cytometry MRD data post-induction and post-consolidation 1 * Non-opposition or consent for secondary use of data
Exclusion criteria
* AML subtypes M3, M6 or M7 * Acute leukaemia of ambiguous lineage * Missing or unusable flow-cytometry files for required time points
Design outcomes
Primary
| Measure | Time frame | Description |
|---|---|---|
| Concordance of MRD ≥ 0.1 % between manual gating and unsupervised FlowSOM gating | up to 8 weeks after initiation first cycle of intensive therapy and after initiation of second cycle of intensive therapy | Sensitivity, specificity, positive and negative predictive values, Cohen and Fleiss kappa statistics comparing the 3 methods on paired samples |
Secondary
| Measure | Time frame | Description |
|---|---|---|
| Concordance of flow-cytometry MRD (both methods) with molecular MRD | up to 8 weeks after initiation first cycle of intensive therapy and after initiation of second cycle of intensive therapy | Sensitivity, specificity, positive and negative predictive values, Cohen and Fleiss kappa statistics comparing the 3 methods on paired samples |
| Agreement of MRD 0.01-0.1 % between manual and unsupervised gating | up to 8 weeks after initiation first cycle of intensive therapy and after initiation of second cycle of intensive therapy | Bland-Altman limits of agreement and Pearson or Spearman correlation coefficients calculated from log-transformed MRD percentages to assess agreement between both gating methods in the low MRD range |
| Inter-operator reproducibility of manual gating | up to 8 weeks after initiation first cycle of intensive therapy and after initiation of second cycle of intensive therapy | Intra-class correlation coefficients and Fleiss kappa statistics calculated across three independent operators performing manual gating on the same flow-cytometry MRD files |
Contacts
Primary ContactAguirre MIMOUN
Outcome results
None listed