Where the calibration rules sit in ICH M10.

ICH M10, adopted at Step 4 on May 24, 2022, covers methods that measure chemical and biological drugs and their metabolites in biological samples for regulatory submissions [1]. Its calibration curve acceptance criteria sit in four sections: 3.2.4 and 3.3.2 for chromatographic validation and study runs, and 4.2.3 and 4.3.2 for the ligand binding assay (LBA) equivalents [2][3][4][5]. Sections 3.3.1 and 4.3.1 define the analytical run [6][7], Section 3.3.4 lists reasons for reanalysis [8], and Section 8.2 points to the documentation in Table 1 [9].

In the EU, M10 came into effect on January 21, 2023, superseding EMA’s 2011 Guideline on bioanalytical method validation (EMEA/CHMP/EWP/192217/2009 Rev. 1 Corr. 2), in effect since February 1, 2012 [10][11].

FDA issued M10 as final guidance in November 2022 [12]. Its guidance page and Federal Register notice say it replaces the June 27, 2019 draft; neither mentions FDA’s May 2018 Bioanalytical Method Validation guidance [13][14]. FDA’s biomarker guidance, first issued in January 2025, addresses it in Section II: M10 “was expected to supersede” the 2018 guidance, M10 excludes biomarkers, and the 2018 guidance “will be withdrawn” on the biomarker guidance’s publication [15]. As of October 26, 2026, FDA’s page for the 2018 guidance still lists it as final [16]. The M10 Questions and Answers (FDA, June 2024) leave the calibration criteria unchanged [17].

The chromatographic rules, item by item.

Section 3.2.4 sets the rules for validation; Section 3.3.2 applies the same limits to each study run [2][3].

1. Blank and zero samples. Each curve includes a blank and a zero sample (a blank spiked with IS); neither enters the regression [2].
2. Six levels minimum. Each curve has at least six concentration levels, including the LLOQ (the lowest standard) and the ULOQ (the highest), prepared in the same matrix as the study samples; each analyte has its own curve in each run [2].
3. Back-calculated limits. Each standard must back-calculate within ±15% of nominal, or ±20% at the LLOQ [2][3].
4. Share that must pass. At least 75% of the standards, across at least six levels, must meet those limits [2][3]. With single standards, an eight-level curve can therefore lose two standards and a six-level curve none.
5. Replicates. All acceptable replicates go into the regression; in validation, at least 50% of the standards at each level must pass [2].
6. Rejection and re-regression. A failing standard is rejected and the curve re-evaluated without it, including the regression [2][3].
7. Model and weighting. M10 asks for “a simple regression model that adequately describes the concentration-response relationship” (Section 3.2.4). Written procedures direct the choice, and the regression model, weighting scheme and transformation are determined during validation [2]. Insights 07 covers how to choose the weighting factor [18].
8. Fresh standards. At least one validation assessment uses freshly spiked standards [2].

Ligand binding assays use wider limits and anchor points.

Section 4 keeps the structure and changes the numbers. An LBA curve needs at least six levels, including the LLOQ and ULOQ, plus a blank; no zero sample is called for. Anchor points below the LLOQ and above the ULOQ may be added to improve curve fitting, and they carry no acceptance criteria because they lie outside the quantifiable range. A four- or five-parameter logistic model is most common when data points lie near the asymptotes; other models need justification [4].

The limits widen to ±20%, or ±25% at the LLOQ and ULOQ. At least 75% of the standards, excluding anchor points, must meet them across at least six levels, which in validation must include the LLOQ and ULOQ [4][5].

Summarized from Sections 3.2.4, 3.3.1, 3.3.2, 4.2.3, 4.3.1 and 4.3.2 [2][3][4][5][6][7].

An analytical run passes two separate tests.

A chromatographic study run contains a blank, a zero sample, standards at six or more levels, and low, medium and high QCs in duplicate or at least 5% of the study samples, whichever is higher [6]. Its acceptance criteria are defined in the protocol, the study plan or an SOP [3].

The first test is the curve, against the limits above. If the LLOQ is rejected, the next lowest acceptable standard becomes the run’s lower limit and, per Section 3.3.2, “will retain its original acceptance criteria (i.e., ±15%).” A rejected ULOQ moves the upper limit down the same way. The revised range should still cover at least three QC levels, and study samples outside it are reanalyzed [3]; Section 3.3.4 lists a result below a revised LLOQ as a reason for reanalysis [8]. With replicate standards, losing only one LLOQ or ULOQ replicate leaves the range unchanged [3].

The second test is the QCs: at least two thirds of all QCs, and at least 50% at each level, must be within ±15% of nominal, or the run is rejected [3]. A passing curve cannot rescue failing QCs, and standards in a failed batch cannot support other batches in the run [3].

M10 sets no correlation coefficient criterion.

None of the four calibration sections mentions r or r². In Sections 1 to 8, the word “correlation” appears once, in Section 6.2, where the concordance correlation coefficient is one option for assessing agreement between two methods in cross validation [19]. Elsewhere, “coefficient” means the coefficient of variation. The calibration test is the back-calculated accuracy of every standard, plus the 75% and six-level minimums [2][3].

An SOP may still set an r² limit, but r² says little about the ends of the curve. In Case B below, both LLOQ standards fail, yet the unweighted r² is 0.999765, unchanged to six decimal places from the base run; the 1/x²-weighted r² falls from 0.9995 to 0.985. Insights 01 explains why r² is not enough [20].

One illustrative curve, two kinds of failure.

Illustrative data. An LC-MS/MS assay in human plasma is calibrated at 1, 2, 5, 10, 50, 100, 500 and 1,000 ng/mL in duplicate (one set before the study samples, one after), with a linear model and 1/x² weighting set during validation. QCs are at 3, 400 and 800 ng/mL. In the base run, all 16 standards back-calculate between −3.1% and +2.9%. Case A changes one response; Case B changes both LLOQ responses.

The duplicates are deliberate. With single standards, 1/x² weighting holds the fit so close to the LLOQ that, on this design, the LLOQ cannot read beyond ±12.0% while every other standard is within ±15%. A single LLOQ standard therefore cannot fail alone.

Illustrative data. Peak area ratio (analyte/IS), base run. Case A sets 50 ng/mL set 2 to 0.73260. Case B sets the 1 ng/mL values to 0.01620 and 0.00948.

Case A. The 50 ng/mL set 2 standard reads +19.5%, outside ±15%; the other 15 fall between −4.5% and +3.4%. After it is rejected and the curve re-regressed, every remaining standard lies between −3.1% and +2.9%. With 15 of 16 standards (93.8%) passing across all eight levels, the run meets both minimums and keeps its 1 to 1,000 ng/mL range [3].

Case B. The two 1 ng/mL standards scatter in opposite directions, to +28.7% and −27.4%, both outside ±20%. Both are rejected and the curve re-regressed. The 2 ng/mL standards become the lower limit and must now meet ±15%; they read −2.0% and +2.0%. With 14 of 16 standards (87.5%) passing across seven levels, the run’s range becomes 2 to 1,000 ng/mL and still covers all three QC levels. Study samples that read below 2 ng/mL in this run are reanalyzed [3][8].

Illustrative data. Back-calculated %RE, set 1 / set 2. Limits: ±20% at 1 ng/mL, ±15% elsewhere. Intercept and slope: A 0.000287 and 0.012252 (re-fit 0.000599 and 0.012016); B 0.000766 and 0.011996 (re-fit 0.000422 and 0.012026).

Illustrative data. Each point is one standard; points outside the band (±15%, ±20% at the LLOQ) fail and are rejected before the re-fit.

Three variations show where the rules bind. Had only one 1 ng/mL standard failed, the range would have stayed 1 to 1,000 ng/mL [3]. Had the 2 ng/mL standards also failed, the new lower limit of 5 ng/mL would sit above the 3 ng/mL low QC, and the range would no longer cover three QC levels [3]. In a validation accuracy and precision run, losing both 1 ng/mL standards would reject the run [2].

What the reports must document about calibration.

Section 8.2 points to Table 1, which splits documentation among the analytical site, the validation report and the bioanalytical report [9].

1. Validation report. Curve parameters (for example, slope and intercept), back-calculated concentrations with mean accuracy and precision, and all acceptable curves from at least three runs [2]; a table of all runs, including failed runs, with analysis dates; a table of calibration standard concentrations and response function results (curve parameters) for all accepted runs, with accuracy and precision; and values outside the acceptance criteria clearly marked in the accuracy and precision run tables [9].
2. Bioanalytical report. A table of all runs with status (accepted or failed), reason for failure and analysis dates, and the same calibration table for accepted runs [9]; back-calculated concentrations of the standards and QCs of accepted runs [3]; and a repeat analysis table with sample IDs, reason, original and repeat values, reason for the reported value and run IDs [9].
3. Comparative BA/BE studies. Both reports add the run summary table for 100% of accepted and failed runs [9].
4. At the analytical site. Procedures that set acceptance criteria for runs, calibration curves and QCs, and run summary tables for 100% of accepted and failed runs, including the calibration curve, regression and weighting function [9].

In Case B, a study sample reanalyzed because it read below 2 ng/mL appears in the repeat analysis table with its reason [9].

Where WLTR fits.

WLTR is a verification layer for the calibration data your CDS already produces. It reads the raw instrument export directly, records a SHA-256 hash of the original file and checks back-calculated accuracy at every calibration level, so bias at the LLOQ and ULOQ surfaces before it reaches a study report. It evaluates every candidate calibration model against the same acceptance criteria at once, 15 per analyte, each with an acceptance value the lab sets to its SOPs [21], and recommends the simplest model that meets every criterion. The analyst makes the final decision.

Each exclusion, like the rejected standards in Cases A and B, is recorded with the user who made it and the reason given, and it travels with the record into QA approval. Method configurations are frozen once results exist, so historical results never shift. WLTR does not replace your CDS, your LIMS or your method validation. WLTR verifies. Your lab validates.