Learning objectives
- Distinguish association from numerical agreement.
- Calculate absolute and relative differences with a stated denominator.
- Explain why an acceptance criterion must precede the result.
- Identify missing evidence before claiming two assays are interchangeable.
Core theory and laboratory context
Editorially approved by Juan Manuel Ojeda. Independent scientific review has not been completed. A method comparison asks whether differences matter for the intended use. Correlation asks a different mathematical question: how closely two variables follow a linear relationship. CLSI EP09 addresses quantitative comparison and bias estimation with patient samples; its public scope includes difference plots, regression and bias at selected concentrations. It is not a protocol for qualitative classification agreement [1].
Write the intended claim before analysing the data. For example: compare absolute counts in a specified specimen, report them in the same units and assess differences over a defined measuring interval. A statement about percentages does not automatically support a statement about absolute counts. A comparison of instruments does not isolate instrument effects if preparation, antibodies, gating and reporting are also changed.
For paired values x and y, define the difference as y minus x. A relative difference using x as comparator is 100 × (y − x) / x; it is undefined when x is zero. When both measurements are percentages, subtraction yields percentage points, not a percentage change. State the chosen convention rather than switching denominators between figures.
CFCM reasoning exercise: think of interchangeability as a claim that requires a clinically meaningful tolerance, a suitable population and uncertainty around the estimated differences. A high correlation coefficient alone contains none of those three decisions.
Key concepts
Bias: a systematic difference, assessed in context and with uncertainty. Precision: variation under specified repeat conditions. Agreement: closeness of results for the intended use. Range: the concentrations actually represented, not simply the smallest and largest possible instrument readings.
Worked example and synthetic scenario
Synthetic data; no patient results. Method A gives 10, 20, 40, 80 and 160 cells/µL. Method B gives 12, 24, 48, 96 and 192 cells/µL.
Every B value equals 1.2 × A. Pearson r is exactly 1 because all points lie on a straight line. Nevertheless, the relative difference against A is +20% for every pair. Absolute differences are 2, 4, 8, 16 and 32 cells/µL. Their arithmetic mean is 12.4 cells/µL; this mean conceals the concentration-dependent size of the differences.
Suppose a fictional protocol had prospectively required every pair in this exercise to be within ±10% of A. These five pairs would not meet that stipulated exercise rule. The ±10% value is invented for teaching and is not a recommended cytometry acceptance limit. Five idealised pairs are also not a validation study.
Pitfalls and interpretation limits
- Calling a method equivalent because r or r² is high.
- Using a wide range to hide the behaviour of low-level specimens.
- Deleting discordant pairs without a documented, pre-specified reason.
- Comparing different parent populations or changing gates to improve agreement.
- Treating multiple measurements from one person as independent people.
- Using a comparator-denominated percentage difference when the comparator is zero or very close to zero.
Practical implications and limits
Before accepting an instrument or workflow transfer, prepare a claim-to-evidence table: measurand and units; population and specimen; preparation and gating; comparator and limitations; measuring interval; pre-specified allowable differences; uncertainty; discordance investigation; and final authorised conclusion. This is a CFCM planning aid, not a replacement for a licensed standard or a local validation plan.
Test your interpretation
A second dataset has A = 2%, 4%, 8% and B = 3%, 5%, 9%. Is the difference a constant 1%, a constant 1 percentage point, or a constant relative difference? Would another perfect correlation settle whether the methods can be interchanged?
Answer and explanation
The absolute difference is consistently 1 percentage point. Relative differences against A are 50%, 25% and 12.5%, respectively. Correlation remains 1 because B = A + 1, but that does not decide clinical acceptability. The concentration range, intended decision, predefined tolerance and uncertainty still need consideration. The low-level result is particularly informative in this constructed example.
Knowledge check
1. Can r = 1 coexist with proportional bias? Yes; B = 1.2A demonstrates it.
2. Does agreement in lymphocyte percentages establish agreement in absolute counts? No; the count calculation and its additional inputs must also be assessed.
3. Is the exercise's ±10% a CLSI recommendation? No. It is an explicitly fictional teaching criterion.
Primary sources and further reading
[1] CLSI EP09, Measurement Procedure Comparison and Bias Estimation Using Patient Samples. Official scope and publication details. Public description consulted; the complete licensed guideline is not reproduced.
[2] CLSI H62, Validation of Assays Performed by Flow Cytometry. Official guideline page. Intended-use framework; this lesson is not a substitute for the full document.
All numerical datasets, calculations and exercise acceptance criteria on this page were constructed for CFCM teaching. They are not study findings.
Continue learning
Use the Academy curriculum to place this lesson in its module and related learning path.
Browse the Academy curriculum →