What a spike recovery proves

Most samples that matter don't come with an answer key. Nobody knows in advance how much lead is in a batch of mushroom powder or how much of a herbicide is in a river sample. So analysts use a trick: take a portion of the sample, add a known amount of the thing being measured, run it through the whole method, and see how much of the addition comes back out. That is a spike, and the percentage that comes back is the recovery.

The arithmetic

AOAC INTERNATIONAL's guidelines for method performance (Appendix F of its Official Methods of Analysis) define recovery as the fraction or percentage of the analyte recovered when the sample goes through the entire method, and give two versions. Marginal recovery counts only the added amount: the spiked result minus the unspiked result, divided by what was added. Total recovery divides the spiked result by the native amount plus the added amount. AOAC says total recovery is usually used, unless the native analyte is more than about 10% of the amount added.

Done properly for method validation, this is not one test. AOAC asks for at least seven independent analyses per concentration, at a minimum of three concentrations covering the method's range, with "independent" meaning at least run at different times.

Two different spikes, two different questions

EPA Method 200.8, the agency's ICP-MS method for trace elements in waters and wastes, makes a distinction that is worth knowing because the two spikes get blurred on reports.

The laboratory fortified blank is clean reagent water spiked with the analytes. It asks whether the lab and the method are in control today. Method 200.8 requires one per batch and sets required control limits of 85 to 115%. Outside that, the analyte is "judged out of control" and the problem has to be fixed before analysis continues.

The laboratory fortified sample matrix is a portion of the real sample spiked the same way. Its stated purpose is to determine "whether the sample matrix contributes bias to the analytical results." The method requires it on at least 10% of routine samples and gives it a wider designated range, 70 to 130%.

The interesting case is when the blank spike passes and the matrix spike fails. Method 200.8 says that result is "judged to be matrix related, not system related," and that the data user should be told the unspiked result is suspect, because of either a heterogeneous sample or an uncorrected matrix effect. In plain terms: the instrument was fine, but this particular sample was doing something to the measurement.

So when a report says "spike recovery 96%," the first question is which spike. A blank spike passing tells you the lab was in control that day. It says nothing about your sample's matrix.

Acceptable gets wider as the amount gets smaller

AOAC publishes target ranges for mean recovery as a function of how much analyte is present (Table A5 of the same appendix, excerpted from a 1998 AOAC manual). The range opens up as concentration falls:

• 100% (a pure substance): 98 to 102%
• 1%: 97 to 103%
• 0.1%: 95 to 105%
• 100 ppm: 90 to 107%
• 10 ppm: 80 to 110%
• 10 ppb: 60 to 115%
• 1 ppb: 40 to 120%

That is not sloppiness. Measuring a few parts per billion is genuinely harder, and the guidelines are honest about it. But it changes how to read a passing result. At the low part-per-billion levels where contaminant limits often sit, a method can recover well under the true amount and still meet the target. "Within spec" at trace levels is a wide window, not a bullseye. (Report 062 covers the related problem of what "not detected" means near the detection limit.)

One more detail from Method 200.8: recovery calculations are not required when the added amount is less than 30% of what is already in the sample. The reason is easy to see from the arithmetic. If a sample already holds a lot of analyte and you add a little, marginal recovery becomes a small difference between two large, noisy numbers, and the percentage stops meaning much.

The limit AOAC writes down

This is the part that rarely makes it onto a certificate. In its review of how method accuracy is demonstrated, AOAC treats a certified reference material in a matching matrix as the stronger evidence, and spiking as a fallback when no such material exists. Its warning about the fallback is direct:

"it cannot be assumed that the analyte is bound in the same way as it would be in a natural matrix."

The same section adds that many analytes are bound in a natural matrix, and "whether the spiked analyte will behave the same as the analyte in a natural matrix is unknown."

Think about what a spike physically is. An analyst pipettes a dissolved standard onto a powder or into a liquid before sample preparation. That analyte sits on the surface, freely available. The native analyte may be locked inside plant cell walls, bound to proteins, or incorporated into mineral particles, where the extraction step may never fully reach it. A method can recover the easy, added material at 98% and still leave some of the hard, native material behind. The spike would not show it, because the spike was never in the hard-to-reach form.

AOAC also says plainly that alternatives to reference materials (proficiency testing, spiking, comparing methods) "do not provide the same level of assurance" as a certified reference material.

What to ask about a recovery number

None of this makes spike recovery useless. It is a real, required quality check, and a failed matrix spike is a valuable warning. It just answers a narrower question than people assume. If a recovery figure is doing work in a decision you care about (a heavy-metal certificate on a supplement, a water test on a well), these are fair questions for the lab:

• Blank spike or matrix spike? Only the matrix spike speaks to your sample.
• Spiked at what level? A spike far above or below the reported result tests a different part of the method's range.
• Was a matrix-matched certified reference material run? Per AOAC, that is stronger evidence of accuracy than a spike.
• At this concentration, what recovery range counts as a pass? At 1 ppb, AOAC's target is 40 to 120%.

This is the habit I push for when I consult on lab operations: every quality-control number answers a specific question, and the work is knowing which question it answered. (Report 080, on what "NIST traceable" buys you, makes the same point about calibration.)