CLSI H62: start with intended use
H62 (first edition, 2021) addresses analytical validation of cell-based flow assays. Define the specimen, population, denominator, reportable result and clinical purpose before selecting experiments. Qualification, optimisation, validation and ongoing QC answer different questions. This library is an educational companion, not a replacement for the licensed guideline.
CLSI H62 — intended-use validation →Verification versus validation
Verification asks whether an established performance claim holds in the local implementation. Validation establishes fitness for a specified intended use. A change in specimen, preparation, marker or analysis can alter the measurand and the evidence needed. CFCM practice: document the change and affected claims before deciding the extent of re-evaluation.
Monaghan et al. — verification and validation →LoB, LoD and LoQ
LoB characterises blank/background measurements; LoD concerns reliable detection; LoQ adds a predefined quantitative performance requirement. They are not interchangeable. Define the background and quantitative performance requirements, including precision and relevant bias or total-error criteria, before examining low-level samples. Event counts alone do not establish these limits, and an instrument fluorescence threshold is not an assay MRD detection limit.
CLSI EP17 — detection and quantification limits →Precision: measure the variation that matters
Repeat acquisitions of one tube omit preparation variability. Distinguish repeatability from intermediate precision across days, operators and lots, and from inter-laboratory reproducibility. Include low-abundance populations and report the denominator. EP15 supports verification of claims; establishing a new claim needs an appropriate validation design. A good CV does not exclude systematic bias.
CLSI EP15 implementation guide — precision verification →Carryover: challenge the sequence
CFCM study design: place high-target specimens before low or negative specimens, include baseline negatives, repeat sequences and evaluate the actual wash procedure. Judge contamination against the assay's low-level reporting claim, not only a global percentage. Preserve acquisition order and event counts. Set acceptance criteria before testing; no single carryover percentage is suitable for every MRD assay.
CLSI H62 — assay validation framework →Reference intervals: match the population and measurand
An interval applies to a defined reference population, specimen and measurement procedure. Age, preparation, gating and denominator can change its applicability. Verify a transferred interval rather than copying a table without checking compatibility. A reference interval is not necessarily a clinical decision limit. Document selection, exclusions, partitioning and uncertainty.
CLSI EP28 — reference intervals →MRD below quantification
CFCM reporting practice: A phenotypically credible population can meet a validated detection rule without meeting the quantitative requirement. Report that distinction using the authorised assay policy; do not round it into an unqualified negative or a falsely precise percentage. State specimen adequacy, the evaluable-cell denominator and the sample-specific detection and quantification limits supported by the validated assay. An inadequate specimen limits what a negative result can exclude. Disease-specific guidance and the laboratory's validated procedure determine the reporting terminology; thresholds cannot be transferred between diseases or assays without evidence.
ELN-DAVID 2025 update (published 2026) — AML MRD →Titration and voltration
Titration varies reagent concentration; voltration examines detector voltage/gain settings where the instrument permits adjustment. They solve different problems. Optimise separation in the intended sample and complete panel, not brightness in isolation. Record concentration, cell input, volume, fixation and reagent lot. Assess whether reagent brightness and spreading obscure co-expressed dim markers, and check detector range and resolution under the chosen settings.
Published T-cell workflow — instrument-specific voltration →Compensation and FMO controls
Single-stained controls estimate spillover or spectral reference signatures. Fluorescence-minus-one controls assess background and spreading in the multicolour context to support boundary placement. An FMO does not reproduce nonspecific binding from the omitted antibody. It is not a substitute for a compensation/unmixing control and does not establish antigen specificity. Include relevant biological negatives and verify tandem-dye and processing compatibility.
Cossarizza et al. — controls and FMO boundaries →Panel validation: validate the reported decision
A panel is a measurement procedure, not a list of antibodies. Challenge its intended populations, relevant abnormal phenotypes and treatment contexts. CFCM practice: define failure modes, review discordance in raw data, and assess recovery and classification alongside marker intensity. Adding or replacing a marker requires a documented assessment of the resulting clinical claim.
CLSI H62 — panel and assay validation →Spectral validation: references are part of the assay
Published guidance: single-stained controls support spectral unmixing, and their fluorochromes, acquisition settings and background characteristics must suit the sample. Unstained controls help assess autofluorescence, which can change with biological condition. CFCM practice: inspect unmixing artefacts, spreading and dim-population resolution in the complete panel under the intended preparation conditions. Reassess relevant reagent, processing or instrument changes against the assay’s performance claims. The extent of re-evaluation depends on the modification and intended use; these sources do not establish a universal clinical spectral validation protocol or numerical acceptance limit.
Cossarizza et al. — controls and spectral unmixing →Monaghan et al. — assay modifications →AI validation: evaluate the complete workflow
Consensus principles: IMDRF N88 (2025) recommends evaluating the intended clinical workflow, representative populations, independent training and test data, and the performance of the human–AI team. Independence should consider patients, sites and data acquisition; external validation should be proportionate to risk. Users need clear information about inputs, limitations and relevant subgroup performance, while deployed systems require monitoring and controlled changes. CFCM practice: retain model and preprocessing versions, test relevant temporal and local data, review subgroup errors, and define what happens when inputs or outputs fall outside the evaluated use. These principles guide evaluation; they do not certify a cytometry model or replace its local, intended-use assessment.
IMDRF N88 — Good Machine Learning Practice (2025) →