Define the ICH Q2 validation characteristics and calculate linearity, LOD, LOQ and recovery from data.
Why it matters
A method that works once on the developer's bench is not yet a method. Before results are used to release a medicine, certify drinking water or support a publication, the lab must show with data that the method is fit for its intended purpose. That evidence is method validation. In the pharmaceutical world it follows the ICH guideline Q2(R2), Validation of Analytical Procedures, with Q14 covering method development (ICH quality guidelines, checked on 2026-10-23). Environmental and ISO 17025 labs use very similar characteristics.
The validation characteristics
| Characteristic | Question | Typical evidence |
|---|---|---|
| Specificity | Does it measure only the analyte? | Blank, placebo, spiked impurities, stressed samples; peak purity |
| Linearity | Is the response proportional to concentration? | At least 5 levels; slope, intercept, residuals, r^2 |
| Range | Over which interval are accuracy, precision and linearity acceptable? | For an assay often 80 to 120 % of the target |
| Accuracy | Is the result close to the true value? | Recovery of spiked known amounts, or a CRM |
| Precision | How close are repeated results? | Repeatability (same day, analyst, instrument), intermediate precision (different days, analysts, instruments), reproducibility (between labs) |
| Detection limit (LOD) | What is the smallest amount that can be detected? | Signal-to-noise about 3, or 3.3 sigma / slope |
| Quantitation limit (LOQ) | What is the smallest amount that can be measured with acceptable accuracy and precision? | Signal-to-noise about 10, or 10 sigma / slope |
| Robustness | Does it survive small deliberate changes? | Vary pH, temperature, flow rate, column lot |
Not every method needs every characteristic. An identity test needs specificity; an impurity limit test needs specificity and LOD; a quantitative impurity method and an assay need most of the table.
The protocol comes first
Validation is planned in a protocol with predefined acceptance criteria (for example "mean recovery 98.0 to 102.0 %", "repeatability RSD at most 1.0 %"). Results are compared with those criteria in a report. Setting criteria after seeing the data is not validation.
Worked example: linearity, LOD and LOQ
A fictional UV method is calibrated at six levels:
| Concentration (mg/L) | 0 | 2 | 4 | 6 | 8 | 10 |
|---|---|---|---|---|---|---|
| Absorbance | 0.003 | 0.104 | 0.203 | 0.305 | 0.402 | 0.503 |
Least-squares regression gives:
- slope S = 0.04994 AU per mg/L, intercept = 0.0036 AU
- r^2 = 0.99997
- residual standard deviation (standard error of the regression, n - 2 degrees of freedom) sigma = 0.00113 AU
Using the ICH calibration-curve approach:
- LOD = 3.3 x sigma / S = 3.3 x 0.00113 / 0.04994 = 0.075 mg/L
- LOQ = 10 x sigma / S = 10 x 0.00113 / 0.04994 = 0.23 mg/L
The LOQ must then be confirmed experimentally: analyze samples near 0.23 mg/L and show acceptable accuracy and precision there.
Two cautions about r^2. A high r^2 does not prove linearity (a curved response can still give r^2 above 0.99), so always plot the residuals: they should scatter randomly around zero with no pattern. And r^2 says nothing about accuracy.
Accuracy as recovery
Spike a placebo or sample with known amounts at, for example, 80, 100 and 120 % of target, in triplicate:
recovery % = (found - unspiked) / added x 100
If a sample containing 5.00 mg/L is spiked with 2.00 mg/L and measures 6.96 mg/L, recovery = (6.96 - 5.00) / 2.00 x 100 = 98.0 %.
Verification, transfer and lifecycle
- Verification: a compendial method (for example from the USP) is not revalidated, but the lab shows it works in its hands.
- Transfer: a validated method moved to another lab is shown to give equivalent results there.
- Revalidation: needed after significant changes (new instrument type, formulation change, method change). The ICH Q14 / Q2(R2) approach treats this as a lifecycle with ongoing monitoring.
Common mistakes
- Treating r^2 as proof of linearity without looking at residuals.
- Calculating LOQ but never confirming it.
- Writing acceptance criteria after the experiments.
- Validating precision with replicate injections of one solution (that is system precision) instead of independent sample preparations.
Quick recap
- Validation shows a method is fit for purpose: specificity, linearity, range, accuracy, precision, LOD, LOQ, robustness.
- Predefine acceptance criteria in a protocol.
- LOD = 3.3 sigma / S; LOQ = 10 sigma / S; confirm LOQ experimentally.
- Recovery % = (found - unspiked) / added x 100.
Try it
Open the Do tab. A NumPy validation script fits the calibration; you write the LOD and LOQ formula.