Interference and recovery testing explained Haemolysis, icterus, lipaemia and a hundred common drugs can shift a result without the analyser ever raising a flag. Interference studies put a number on that shift — and tell you the concentration at which it stops being tolerable.

An interferent is anything in the sample, other than the analyte itself, that changes the measured result: endogenous substances such as haemoglobin, bilirubin and lipids, or exogenous ones such as drugs, metabolites and preservatives. The measurement still returns a value. The value is simply the wrong one, and nothing about the number announces that it is wrong. An interference study is how you find the effect before a patient result carries it into the record.

The question an interference study answers

The design is deliberately narrow: hold everything constant except the suspected interferent, add it at a known concentration, and measure the change in result against an unaltered control. The output is a bias: the difference the interferent produces at a given test concentration. EP07 is the reference protocol, and it separates two experiments.

The interference screen is a paired-difference test at a high, clinically justified interferent level, giving a single yes/no on whether that substance matters at all. The dose-response study follows up on anything the screen flags, adding the interferent at several levels to characterise how the bias grows with concentration and to find the level at which it first becomes unacceptable.

Judging the effect against an allowable limit

A bias figure on its own decides nothing. The interference matters only when it is large enough to change a clinical decision. The effect is tested against an allowable limit, usually derived from the same allowable total error or allowable bias you use elsewhere in validation. Below the limit, the interferent is present but harmless at that level. Above it, you have a claim to make on the label, a concentration to flag, or a sample type to reject.

A plot of measured bias against interferent concentration: a rising curve crosses a horizontal allowable-limit band; below the crossing the bias is tolerable (green), above it the bias is unacceptable (red), and the crossing marks the highest tolerable interferent level.
A dose-response study reads off the concentration at which interference bias crosses the allowable limit. Below the crossing the interferent is tolerable. Above it, the bias changes the result by more than the assay’s error budget allows.

Judging the effect is the step Analyse-it is built for: not the pipetting, but the decision. You supply the observed effect at the relevant concentration and the allowable bias, and the effect is reported with a confidence interval against that band, flagged where the requirement is not met.

A difference plot of bias against assigned value at three concentrations, with 95% confidence intervals and diverging allowable-bias bands; the mid-level bias of 4.2 exceeds its allowable band of plus or minus 4.0 and is flagged.
Judging an observed effect against an allowable limit (CLSI EP10-A3, ethanol). The bias at each level is tested against its allowable-bias band. The mid level (4.2 against ±4.0) fails. That is the same decision used in an interference or recovery study.

Recovery is the same question, reversed

An interference study asks “does adding this substance push the result off?” A recovery study asks “when I add a known amount of analyte, do I get it all back?” Spike a sample with a known quantity, measure, and express the result as a percentage of what was added. Full recovery near 100% argues the matrix is not suppressing or inflating the signal. Recovery that drifts from 100% is a proportional bias hiding in the sample matrix rather than in a named interferent. The arithmetic differs but the judgement is identical: the recovery, or its shortfall, is tested against an allowable limit.

Reporting

Report the effect as a bias with a confidence interval, at the interferent concentration tested, against the stated allowable limit. Do not report a bare “no interference”. “No significant interference from haemoglobin up to 5 g/L” is a claim a reviewer can check. “No interference” is not. Where a dose-response was run, report the highest interferent level at which the bias stays inside the allowable band. That level is what a laboratory needs in order to set a rejection threshold.

Downloads

Download the CLSI EP10-A3 example workbook (.xlsx) — a preliminary evaluation judging bias at three concentrations against allowable-bias limits, ready to open in the Analyse-it trial.

Common mistakes

Screening at an irrelevant concentration. An interference screen run at a level no real sample reaches proves nothing useful. Choose the high test level from clinical reality: the icterus a jaundiced patient actually presents, not a round number.

Reading a bias without an allowable limit. A +6% shift is neither pass nor fail until it is set against the assay’s error budget. State the allowable difference and test against it.

Treating a screen result as a dose-response. A single high-level yes/no cannot tell you the concentration at which interference sets in. If the screen flags an effect, characterise it.

Confusing recovery with linearity. Recovery probes the matrix at a point. Linearity probes the relationship across the interval. They answer different questions and neither substitutes for the other.

Test for interference with Analyse-it

Analyse-it tests the observed effect against your allowable limit, inside Excel:

  • The interference or bias effect at clinically relevant concentrations, each with a confidence interval
  • The requirement flagged where it is not met, rather than left for the reader to work out
  • The allowable limit you set, applied at each concentration tested

Every feature from all five editions for 15 days, with no sign-up and no licence key. Bias and interference testing are in the Method Validation and Ultimate editions, from US$ 475 a year. Validated against NIST and CLSI reference datasets. See verifying manufacturer precision claims for the related trueness question.