Carryover Test Interpreter

Run high samples, then low ones, and see whether the first low result is pulled up by the high ones that preceded it. The arithmetic is easy; the hard part is what to compare it against. This page applies two criteria — is the difference bigger than the noise, and is it bigger than the difference that would change a decision — because a carryover coefficient of a twentieth of a per cent can be clinically catastrophic when the dynamic range is wide.

Last updated 18 Sep 2026

Is this carryover significant?

High-then-low series → significant or not

Mean of the high replicates (analyte units)The mean of the high-concentration results in the series. Choose a high pool near the top of the measuring interval, or above it — carryover is a property of a concentration ratio, and a modest high pool will not reveal a problem that a genuine patient extreme would cause.

Mean of the FIRST low result after each high runPosition L1: the low sample analysed immediately after the last high one. This is the contaminated position. Average it across the repeated cycles.

Mean of the THIRD low result after each high runPosition L3: two low samples further on, by which point any carryover should have washed out. This is the uncontaminated reference value, and it is what L1 is measured against — not the low pool’s assigned value.

SD of the uncontaminated low results (analyte units)The standard deviation of the L3 results, which is the method’s own imprecision at the low concentration. Without it there is no way to tell a real carryover from ordinary scatter, and this is the input most carryover calculations leave out.

How many times was the cycle repeated?Each cycle is one block of high replicates followed by low replicates. More cycles narrow the standard error of the difference between the L1 and L3 means, so a small but real carryover becomes detectable. Five is the number the Cofrac SH GTA 04 protocol specifies.

Largest difference at the low concentration that would not change a decision (analyte units)The clinical criterion, in the analyte’s own units, at the LOW concentration. For an analyte with a decision threshold near the bottom of its range, this is small — which is why a carryover coefficient that looks negligible as a percentage can still be unacceptable.

Carryover is clinically significant — the carried-over quantity is larger than the difference that would change a decisionExample

A high pool at 100 units, near the top of the measuring interval, alternated with a low pool over three cycles. The first low position averages 0.055 units and the third averages 0.010; the SD of the low results is 0.004. The analyte’s decision threshold sits near the bottom of its range, so the laboratory set the largest tolerable difference at 0.020 units.

The design, the coefficient, and the two criteria

Design — a block of high replicates followed by a block of low replicates, repeated several times: H H H L1 L2 L3, H H H L1 L2 L3, …
Carryover coefficient k = (mean L1 − mean L3) / (mean H − mean L3), usually expressed as a percentage
Statistical criterion — the difference is detectable if (mean L1 − mean L3) exceeds 2 x SD(low) x sqrt(2 / number of cycles)
Clinical criterion — the carryover matters if (mean L1 − mean L3) exceeds the smallest difference that would change a decision at the low concentration

L1 against L3, not against the assigned value: the comparison is between the first low position and a later low position in the same series, so that anything systematic about the low pool itself — its assigned value, its stability, a small bias in the method — cancels out. Comparing L1 against the pool’s nominal value would confound carryover with bias
the provenance of this design: Broughton and colleagues’ evaluation schemes for automatic analysers (J Clin Pathol 1969 and Ann Clin Biochem 1974) are the classic references for carry-over as an evaluation parameter, but neither full text could be retrieved while this page was written and no formula is attributed to them here. The three-high, three-low, five-times design is specified in the Cofrac SH GTA 04 technical guide, and the definition and determination of carry-over effects are set out in Haeckel R, J Automat Chem 1988;10(4):181-183
the one per cent figure: widely applied to this protocol and attributed to Haeckel’s recommendations by implementations of it. The original paper could not be opened to confirm it, so it is used here as a benchmark and labelled as unverified rather than presented as a sourced acceptance limit. It is in any case the weaker of the two criteria
2 x SD x sqrt(2/n): the standard error of a difference between two means, each of n observations with standard deviation SD, is SD x sqrt(2/n); twice that is roughly a 95% criterion. This is why the number of CYCLES matters more than the number of replicates inside one — it is the number of independent L1 and L3 values that shrinks the standard error
why the clinical criterion outranks the coefficient: the coefficient is dimensionless and the harm is not. A carryover of 0.05% from a sample at 100 units transfers 0.05 units; if the decision threshold for that analyte is 0.02 units, that is two and a half times the threshold from a coefficient twenty times below the usual benchmark. Cardiac troponin is the published case in point

Worked example

A high pool at 100 units, near the top of the measuring interval, alternated with a low pool over three cycles. The first low position averages 0.055 units and the third averages 0.010; the SD of the low results is 0.004. The analyte's decision threshold sits near the bottom of its range, so the laboratory set the largest tolerable difference at 0.020 units.
Carried-over quantity = 0.055 − 0.010 = 0.045 units
Clinical criterion: 0.045 exceeds the 0.020 units the laboratory declared tolerable → clinically significant
Carryover coefficient = 0.045 / (100 − 0.010) = 0.00045, or 0.045% — more than twenty times below the one per cent benchmark
Statistical criterion: 2 x 0.004 x sqrt(2/3) = 0.0065. The observed 0.045 is about seven times that, so it is comfortably real
So the analyser passes the relative benchmark and fails the clinical one, and it is the clinical one that decides. A patient sample following a grossly raised one can be reported above a threshold it does not reach
This is the shape of the published cardiac troponin carryover reports: a coefficient far too small to fail a conventional percentage criterion, and a dynamic range wide enough to make it matter

Why a percentage criterion is the wrong one for a wide dynamic range

Every row but the last passes or fails on the absolute quantity rather than on the coefficient, and the third row is the case that a one per cent criterion would wave through. The comparison that protects patients is between the quantity transferred and the smallest difference that changes a decision at that concentration.
High sample	Carryover coefficient	Quantity transferred	Against a threshold of 0.02 units
100 units	1.0%	1.00 units	50 times the threshold
100 units	0.1%	0.10 units	5 times the threshold
100 units	0.045%	0.045 units	More than twice the threshold
100 units	0.01%	0.010 units	Half the threshold — tolerable
5 units	1.0%	0.05 units	Twice the threshold, from a coefficient at the benchmark

What to change when carryover is found

The distinction between sample and reagent carryover is worth making early, because they have different signatures. Sample carryover scales with the concentration of the preceding SAMPLE; reagent carryover appears between particular TESTS regardless of what sample preceded them, and is invisible to a high-then-low sample design.
Where it comes from	What it looks like	What to do
Sample probe	Affects every analyte pipetted by that probe, scales with the preceding sample	Wash volume and cycles, probe condition and alignment, wash solution, probe replacement
Reagent probe	Affects a specific pair of tests sharing a probe, independent of the preceding SAMPLE	Test sequencing on the analyser, additional wash between the implicated pair
Cuvette or reaction vessel	Affects the same cuvette position repeatedly	Cuvette cleaning cycle, cuvette replacement, blank absorbance monitoring
Assay design	Persists after every mechanical remedy; often immunoassay with a very wide range	Reflex rerun after a result above a defined concentration; dedicated instrument or dilution protocol; manufacturer notification

A coefficient of a twentieth of a per cent can still be a false positive

Carryover is tested by running a series in which high-concentration material is followed by low-concentration material, repeated several times, and asking whether the low result immediately after the high block is raised relative to a low result two positions later. Comparing the first low position against a later one, rather than against the pool’s assigned value, is deliberate: anything systematic about the low material or about the method’s bias affects both positions equally and cancels, so what is left is the contribution from what preceded the sample.

The number usually reported from that experiment is a coefficient — the excess in the first low position divided by the difference between the pools — and the number usually applied to it is one per cent. Both deserve scrutiny. The coefficient is dimensionless, and the harm from carryover is not: what reaches the patient’s report is an absolute quantity of analyte, and whether that quantity matters depends entirely on how small the clinically important differences are at that end of the range. An assay spanning four orders of magnitude, with a decision threshold near the bottom, can be damaged by a coefficient twenty times below any conventional criterion. This is not hypothetical. False-positive cardiac troponin results caused by carryover from very high patient samples have been reported repeatedly in the literature, on systems whose measured carryover was far below one per cent.

The second criterion the experiment needs, and the one most often left out, is statistical. A difference between two means is only evidence of anything if it is larger than the scatter of the measurements can explain. Because the low results are, by construction, near the bottom of the measuring interval where imprecision is proportionally worst, a small apparent carryover is easily produced by noise. The standard error of the difference between the two low means is the standard deviation of the low results multiplied by the square root of two over the number of cycles, and it is the number of cycles rather than the number of replicates within a cycle that shrinks it. A study with too few cycles can fail to detect a carryover that is clinically important, and reporting that as a clean result is worse than not testing.

Finally, this design tests one thing: sample-to-sample carryover at one concentration ratio. It does not test reagent carryover between tests sharing a reagent probe, which shows up as an effect between particular assays rather than after particular samples and is invisible here. It does not test carryover at ratios higher than the one used, which is why the high pool should be at or above the top of the measuring interval rather than merely high. And it should be repeated whenever the probe, the wash configuration or the test menu on that probe changes.

Frequently asked questions

What acceptance limit should I apply to carryover?

Two, and the clinical one decides. The commonly quoted limit is a carryover coefficient below one per cent, which is attributed to Haeckel’s 1988 recommendations by implementations of the protocol — that primary source could not be verified while this page was written, so treat the figure as a benchmark rather than a sourced criterion. The criterion that protects patients is absolute: the quantity of analyte transferred must be smaller than the difference that would change a decision at the low concentration. For an assay with a very wide dynamic range those two criteria can disagree by a factor of twenty.

Why compare the first low result against the third rather than against the pool’s known value?

Because comparing against the assigned value mixes carryover with bias. The third low position has had two washes since the high block, so it represents the same material measured by the same method with no carryover contribution. Any bias in the method, or any drift in the pool, affects both positions equally and cancels out of the difference.

How many cycles do I need?

Enough that the smallest difference the study can detect is below the difference that would matter clinically. The standard error of the difference between the two low means is the SD of the low results times the square root of two over the number of cycles, so four cycles instead of three narrows it by about 13% and ten cycles nearly halves it. Adding replicates inside a cycle does not help nearly as much, because it is the number of independent L1 and L3 values that counts.

My carryover coefficient is well under one per cent. Am I finished?

Only if the absolute quantity transferred is also below your clinical criterion. Multiply the coefficient by the highest concentration your patients actually produce, not by the high pool you happened to use, and compare that with the smallest clinically important difference at the low end of the range. Published false-positive cardiac troponin results arose from exactly this gap.

Does this test cover reagent carryover?

No. This design tests sample-to-sample carryover: contamination carried by the sample probe from one specimen to the next. Reagent carryover occurs between particular tests that share a reagent probe or cuvette, appears regardless of which sample preceded, and has to be tested by sequencing the implicated assays rather than by sequencing high and low samples.

Related calculators

References

Haeckel R. Recommendations for definition and determination of carry-over effects. J Automat Chem. 1988;10(4):181-183. doi:10.1155/S1463924688000380.
Broughton PMG, Buttolph MA, Gowenlock AH, Neill DW, Skentelbery RG. Recommended scheme for the evaluation of instruments for automatic analysis in the clinical biochemistry laboratory. J Clin Pathol. 1969;22(3):278-284. [Classic reference for carry-over as an evaluation parameter; full text not retrievable at the time of writing]
Broughton PMG, Gowenlock AH, McCormack JJ, Neill DW. A revised scheme for the evaluation of automatic instruments for use in clinical chemistry. Ann Clin Biochem. 1974;11(6):207-218.
Gould MJ, Wilgen U, Pretorius CJ, Ungerer JPJ. Probing indiscretions: contamination of cardiac troponin reagent by very high patient samples causes false-positive results. Ann Clin Biochem. 2012;49(Pt 5):497-499.
Wilgen U, Pretorius CJ, Gould MJ, Ungerer JPJ. Cardiac troponin I carryover by very high patient samples still causes false-positive results on the Beckman Coulter AccuTnI + 3. Ann Clin Biochem. 2016;53(Pt 3):426-429.
Cofrac. SH GTA 04 — Guide technique d’accreditation de verification (portee A) / validation (portee B) des methodes en biologie medicale. Comite francais d’accreditation.

Medical Disclaimer: The tools and content provided here are for educational and reference purposes only. They are not intended to substitute for professional medical advice, diagnosis, or treatment. Clinical decisions should always be based on the comprehensive assessment of a qualified healthcare professional.