Intelligence

Analysis

AML MRD across flow and NGS: agreement, discordance and biological context

In 210 remission assessments, FCM and NGS agreed in 197. The 13 discordances reveal why assay targets, thresholds and molecular clone biology require separate interpretation.

When a flow cytometry MRD result is negative and a molecular test is positive, it is tempting to declare one technique more sensitive and move on. In acute myeloid leukemia, that shortcut skips the most important question: were the two assays actually looking for the same biological signal? A retrospective paired-test study offers a useful way to explore the answer without treating either method as the unquestioned reference.

Huang and colleagues studied 210 adults with AML in morphological remission at a single center, with concurrent multiparameter flow cytometry (FCM) and targeted next-generation sequencing (NGS) performed during 2014–2023. The study was published in Annals of Hematology on 16 May 2026 (DOI 10.1007/s00277-026-07008-x). Its selected remission cohort and retrospective design matter whenever we discuss apparent diagnostic accuracy or clinical decisions.

Original CFCM four-panel scientific infographic showing 210 paired AML remission cases: 117 dual-positive, 80 dual-negative, 11 molecular-only and 2 flow-only, with different flow LAIP and NGS VAF decision thresholds.
Original CFCM conceptual diagram. Educational synthesis; not clinical validation.

High agreement does not mean equal measurements

Overall, 197 of 210 patients (93.81%) had concordant positive or negative classifications, with reported Cohen κ of 0.872. The paired results were: 117 positive by both assays, 80 negative by both, 11 flow-negative/NGS-positive and 2 flow-positive/NGS-negative. The 13 discordances—rather than the reassuring headline agreement—are where biological interpretation and technical investigation become most important.

Flow MRD positivity was defined at at least 0.1% leukemia-associated immunophenotype (LAIP) cells, while NGS positivity required at least 0.01% variant allele frequency (VAF). A fraction of immunophenotypically abnormal cells and a fraction of DNA reads carrying a mutation are not interchangeable denominators. One numeric threshold does not directly establish the limit of detection, limit of quantification, or clinical superiority of the other test.

The reference-standard trap

The paper calculated flow sensitivity of 91.41% (117/128) and specificity of 97.56% (80/82), using the NGS classification as the reference. These are valid within-study concordance-derived statistics, not independent proof that every molecular-positive signal represents viable residual leukemia. Different molecular targets, assay error, persistent nonleukemic clones, and biological heterogeneity can alter interpretation. Flow itself can also miss an aberrant clone because of phenotypic shift, limited acquired events, specimen quality or an inadequate validated panel.

For the 11 NGS-positive/flow-negative patients, eight harbored NPM1 mutations; others carried additional variants including DNMT3A or FLT3-ITD. The study reports a median VAF of 0.35% in this discordant direction. A molecular VAF cannot be translated into a percentage of residual leukemic cells without assumptions about clone structure, cellularity, copy number and target distribution. Moreover, DNMT3A, TET2 and ASXL1 may persist as clonal hematopoiesis; the authors excluded isolated DTA variants when defining molecular MRD.

Why technical review still matters

For an AML cytometry service, a discordant case calls for a structured reconciliation. First, verify that the sample and collection time truly match, and inspect sample handling, viability, acquisition stability and event counts. Then review the diagnostic LAIP, whether the residual phenotype has evolved, the background distribution and the actual analytical sensitivity achieved with that sample. In parallel, establish the molecular variant's diagnostic presence, sequencing depth, error handling, persistence and specificity for the leukemic clone.

The study describes ten-colour flow panels with at least 200,000 acquired events and high-depth molecular sequencing, but these method descriptions do not replace local panel performance verification, LoD/LoQ evidence and clinically validated interpretation rules. Nor can a test with a reported threshold of 0.1% be assumed to have uniform event-based sensitivity in every patient.

Prognosis is not treatment authorization

Over a median follow-up of 18 months, the investigators observed marked outcome differences between dual-negative, single-positive and dual-positive groups. The single-positive group contained only 13 patients, and the two discordance directions contained 11 and 2, respectively. This limited sample and retrospective enrichment for patients who achieved remission constrain conclusions about predictive performance, calibration and how well the findings transfer to another laboratory or treatment regimen.

Crucially, these data do not demonstrate that adjusting treatment intensity based on this combined classification improves outcomes. Changes in therapy or surveillance should be considered within disease-specific evidence, established clinical guidance and multidisciplinary review, not automatically adopted from an observational association.

A practical reporting principle

When both platforms are available, report their targets and thresholds separately rather than compressing two different measurements into a single absolute 'MRD truth'. Preserve the discordant classifications, document whether each assay is analytically informative, and explain what additional technical or clinical evidence would resolve the uncertainty. The strongest conclusion from a paired comparison is often complementarity with explicit limits, not interchangeability.

Daily Learning Pill: A 94% agreement rate can coexist with biologically important discordances. Before comparing MRD tests, identify the target, denominator, validated decision threshold and what the chosen reference standard can—and cannot—prove.

Related CFCM analysis: Rare-event interpretation and threshold context in flow MRD.

Evidence tier and rights: Huang J et al., Annals of Hematology (2026), peer-reviewed retrospective clinical study. Read primary research. The journal article is CC BY-NC-ND 4.0; the accompanying educational diagram is an original CFCM numerical summary, not reproduced Springer imagery.