Learning objectives
- State exactly which events form the numerator and denominator.
- Calculate an event fraction and an idealised Poisson counting CV.
- Separate counting precision from validated LoB, LoD and LoQ.
- Describe why zero observed events is not proof of biological absence.
Core theory and laboratory context
Editorially approved by Juan Manuel Ojeda. Independent scientific review has not been completed. A rare-event fraction is a calculation, not a diagnosis. It divides the events assigned to a target population by the eligible denominator. The definition of that denominator belongs to the validated assay, not to a convenient gate chosen after acquisition.
In an ideal Poisson model, a count with expected value λ has standard deviation √λ and relative standard deviation 1/√λ [1]. Using the observed positive count k as an estimate gives the familiar approximate counting CV of 100/√k percent. This describes only the modelled counting component. It does not include preparation losses, sample quality, classification errors, carryover, background events or uncertainty in the denominator.
CLSI EP17 treats limits of blank, detection and quantification as detection-capability characteristics to be evaluated and documented [2]. None can be inferred from k alone. A compact group of 15 events may warrant investigation, but the number 15 by itself cannot establish malignancy, an assay's LoD or a reportable quantitative result.
CFCM interpretation principle: record the analytical observation separately from the clinical claim. First describe what was measured and what qualified for analysis. Then apply the assay's validated evidence and authorised reporting rules. This separation makes a disagreement reviewable rather than concealing it in a single word such as positive.
Key concepts
Numerator: accepted target events. Denominator: accepted events in the explicitly defined parent population. Counting CV: an idealised statistical component, not total assay imprecision. Detection: evidence distinguishable from background under an established procedure. Quantification: a numerical result meeting predefined performance requirements.
Worked example and synthetic scenario
Synthetic arithmetic exercise. An acquisition contains 1,000,000 recorded events. After the fictional eligibility steps, 750,000 events remain in the specified denominator. Fifteen events are assigned to a candidate population.
The fraction is 15 / 750,000 = 0.00002, or 0.002% (20 per million eligible events). Dividing by all recorded events would instead give 0.0015%; it answers a different question. The approximate Poisson counting CV for k = 15 is 25.8%. Neither numerical result resolves whether the candidate population is real.
At k = 100 the modelled CV is 10%, and at k = 400 it is 5%. Those are mathematical examples, not universal LoQ criteria. Acquiring four times as many eligible events only has the predicted precision benefit if the underlying fraction and the assumptions remain appropriate.
Pitfalls and interpretation limits
- Using all acquired events when the assay specifies a different denominator.
- Counting debris, doublets or a transient acquisition artefact as a coherent target population without investigation.
- Reporting an exact clinical percentage solely because software can calculate it.
- Converting a Poisson precision calculation into a universal event threshold.
- Calling a sample negative without documenting what sensitivity was supported by its quality and acquisition.
Practical implications and limits
A reviewable record should retain the raw acquisition, eligibility steps, target definition, numerator, denominator, sample-quality limitations and the applicable validated reporting procedure. Disease-specific interpretation and treatment effects belong to the appropriate clinical workflow; this calculation does not replace them.
Test your interpretation
In a separate idealised experiment, zero target events are observed among 300,000 eligible events. A colleague writes “the target population is absent.” What is defensible from the count alone, and what does the simple Poisson model add?
Answer and explanation
The observed count is zero; biological absence is not established. Under the explicitly ideal Poisson assumptions, P(K = 0) = exp(−λ). Setting this probability to 0.05 gives λ = −ln(0.05) = 2.996. Dividing by 300,000 gives approximately 0.0009986% as a one-sided 95% model-based upper bound on the fraction. This is not a validated assay LoD and not a clinical exclusion threshold. Background, sampling and sample quality can invalidate the simplified interpretation.
Knowledge check
1. Is 100 events a universal flow MRD LoQ? No.
2. Can denominator choice change the reported percentage? Yes; its definition must be stated and justified.
3. Can the counting CV formula be evaluated at k = 0? No; division by zero is undefined. A different statistical statement is required.
Primary sources and further reading
[1] NIST/SEMATECH, Poisson distribution: probability mass function, mean, standard deviation and coefficient of variation. The zero-event bound is a direct mathematical derivation under that model.
[2] CLSI EP17, Evaluation of Detection Capability for Clinical Laboratory Measurement Procedures. Public scope consulted; no licensed procedures reproduced.
All counts and scenarios are synthetic teaching examples. This page provides no disease-specific MRD cutoff or diagnostic classification.
Continue learning
Use the Academy curriculum to place this lesson in its module and related learning path.
Browse the Academy curriculum →