AML MRD phenotype scores: validate the score before changing the label

The evidence in context

Le and colleagues retrospectively studied 1,215 adults with AML or MDS/AML undergoing allogeneic transplantation in morphologic remission; 233 had detectable flow MRD. They scored individual antigen abnormalities and found that the score refined outcome associations. A data-derived cut-point of 4.5 separated score groups. Lower-score MRD-positive patients had outcomes resembling the MRD-negative group. The authors explicitly require independent validation. These are prognostic observations, not proof that a low score excludes residual leukemia or that changing treatment improves survival.

CFCM interpretation: preserve two questions

Keep the analytical question and the prognostic question separate when appraising a proposed score. The first asks what the laboratory detected and how confidently it distinguished the population from its background. The second asks what a recorded feature predicts in a defined clinical setting. A promising answer to the second question should not silently replace the first in the report.

For a local evaluation, create a record that retains the original result, the proposed score and the reason for each assigned component. This is a proposed evaluation framework, not a clinical reporting rule. It lets reviewers examine disagreements without losing the underlying measurement.

A useful reproducibility exercise

Before a pilot, write an adjudication plan for ambiguous expression patterns. Ask two appropriately trained reviewers to document their assessment separately using the same prespecified examples and records. When they disagree, record whether the difference concerns population selection, the comparison population, expression intensity or score assignment. A final consensus alone can conceal where the assessment is fragile.

Retain the instrument configuration, reagent details and analysis version with each evaluation. Decide in advance how missing or unevaluable components will be handled. Otherwise, a convenient total can hide a change in what was actually assessed. These suggestions are CFCM editorial questions for study design; they do not establish acceptable clinical performance.

Test the proposed use, not just the arithmetic

Write the intended decision before choosing an evaluation endpoint. Is the proposed score being studied as additional prognostic information, a research stratification variable or a replacement for an existing interpretation? Each proposal makes a different claim and deserves evidence addressing that claim. Prespecify the comparison and examine calibration, uncertainty and consequences of disagreement in an appropriate validation design.

Until that work supports the intended use, preserve the distinction between a research score and the laboratory's validated MRD interpretation. Do not implement a new negative category from this article. Clinical decisions remain within the treating team's established procedures and applicable guidance.

Daily Learning Pill

A score can be promising without being ready to change a label. Ask which part is validated: detection, reproducible scoring, prognostic performance or the decision made from it.

Primary source and review scope

Calvin M. Le et al. HemaSphere (2026-01-29). Publisher full text; DOI 10.1002/hem3.70310.

Actual source scope: Publisher full HTML: Patients and methods including cohort, MRD testing/scoring and statistical analysis; Results including score cut-point and sensitivity analysis; Discussion and limitations. Supplementary files and raw data were not independently reanalysed.

Scientific and editorial review: ChatGPT/Codex preparing editor, 21 October 2026. No independent, human or clinical validation is asserted. Study observations and CFCM editorial appraisal are distinguished above. Original CFCM artwork; no third-party figure reproduced.