Recovery correction

A recovery correction adjusts an analytical result to account for an estimated loss of analyte during a specified preparation stage. If the recovered amount $y$ follows $y=rq$, with original amount $q$ and extraction recovery $r>0$, the corrected estimate is $\hat{q}=y/r$. Its validity depends on whether the recovery estimate applies to that analyte, sample matrix and procedure.

For an original example, $y=6\mathrm{nmol}$ and $r=0.60$ give $\hat{q}=10\mathrm{nmol}$. The calculation estimates the amount before loss; it does not restore molecules to the sample. It also does not separately correct a mass spectrometry matrix effect unless that effect has been included consistently in the measurement model.

Recovery-correction uncertainty follows from the quotient, not just from the measured amount. First-order propagation of uncertainty givesIf the two inputs are independent, the covariance term vanishes. With stipulated standard uncertainties $u_y=0.3\mathrm{nmol}$ and $u_r=0.06$, the two variance contributions are $0.25$ and $1.00\mathrm{nmo}{\mathrm{l}}^2$, giving $u(\hat{q})\approx 1.12\mathrm{nmol}$. The recovery estimate dominates this example; dividing by recovery has not created extra precision.

A recovery inferred from a spiked preparation need not reproduce the behavior of analyte originally embedded in a sample. Likewise, applying a single recovery at every concentration assumes a common model. Near-zero or very uncertain recovery can make quotient inference unreliable beyond the linear uncertainty approximation.

Reporting whether a correction was applied, its size and how it was derived makes results comparable. These reporting provisions appear in the harmonised recovery guidelines and their Codex adoption text.^[ref]