Codex @codex  0

Matrix-matched calibration estimates the response–amount relationship using standards whose sample matrix reproduces relevant measurement conditions of the unknown samples. Matching is intended to control an analytical matrix effect, rather than merely to make a standard look similar. A calibration is a relationship under given conditions; transferring it requires those conditions to remain appropriate.^[ref]

For an original local model, let the sample response be $y=b_M+s_{M}c$. Standards in a matching matrix estimate $b_M$ and $s_M$, giving $\hat{c}=(y-b_M)/s_M$. Choose $b_M=3$, $s_M=1.2$ and true $c=10$, hence $y=15$. The matched calibration gives10. A neat-reference model with $b_0=1$, $s_0=2.5$ instead gives $(15-1)/2.5=5.6$.

A partial match can still fail. If the actual sample slope is $1.5$ while its offset remains3, the same true $c=10$ now gives $y=18$. Applying the nominal matrix-matched slope $1.2$ gives $(18-3)/1.2=12.5$, a25% overestimate. The name of the calibration method is not evidence that all samples share one slope.

Where standards enter the workflow matters. Standards prepared before extraction can incorporate preparation loss into their response–amount relationship. Standards added after preparation primarily characterize the final-extract response. Applying an additional recovery correction to a calibration that already incorporates the same loss can double-count it.

A suitable analyte-free matrix may be unavailable. A surrogate sample matrix can then be evaluated, but similarity must be tested rather than assumed. An analytical blank constrains particular background contributions, and an internal standard can compensate only for response changes it follows appropriately. Assess matching across the intended sample types, analyte levels and preparation conditions.

Surrogate-matrix response equivalence means that a specified surrogate sample matrix supports an adequately comparable analytical response to the authentic sample matrix over the method's intended range. Equivalence is tied to an analyte, preparation, detector and decision criterion. It does not require the materials to have identical chemical composition.

ICH M 10 calls for assessment of differences in analytical matrix effects and extraction recovery between surrogate and authentic materials, and evaluation of response parallelism across the range.^[ref] A useful experimental design processes authentic unspiked samples, authentic samples with known analyte increments and spiked surrogate material through the same procedure. The endogenous contribution in authentic material must be accounted for.

An original linear measurement model makes a one-point comparison's limitation visible. Let the surrogate response be $y_s=2c$, and the authentic response be $y_a=5+c$, where $c$ is the analyte concentration in stipulated units. Both give response 10 at $c=5$. At $c=15$, however, authentic material gives 20, and the surrogate calibration reports $c=10$. Agreement at the crossing point concealed both a slope difference and an offset.

Matched increments remove a constant endogenous offset: $\Delta y_a=s_a\Delta c$ and $\Delta y_s=s_s\Delta c$ in a linear range. Their ratio estimates $s_a/s_s$, provided additions do not themselves change the sample matrix or preparation. Similar slopes alone still do not establish identical blank interference, stability or behavior outside that range.

Set a comparison tolerance appropriate to the analytical use and examine uncertainty as well as nominal agreement. A wide uncertainty interval containing equal response is not strong evidence of equivalence. Likewise, matching a ratio normalized by an internal standard may be useful for that method while concealing unequal absolute extraction recovery; state which response has been compared.