Method Development, Validation, and Measurement Uncertainty
Method Development, Validation, and Measurement Uncertainty
This domain tests how a method is designed, proven fit for purpose, and reported with a defensible uncertainty: specificity, linearity, range, accuracy, precision, LOD/LOQ, robustness and an uncertainty budget.
Core rules
- Validation parameters include specificity/selectivity, linearity and range, accuracy (trueness), precision (repeatability and intermediate precision), LOD/LOQ, robustness and, where relevant, ruggedness.
- Accuracy is assessed with certified reference materials or spiked samples; precision with replicates at low, mid and high levels. Acceptance criteria must be predefined.
- Linearity is not the same as working range; verify lack of fit and residual distribution.
- Measurement uncertainty: identify sources, quantify standard uncertainties, combine independent contributions by root-sum-square, and expand with k = 2 for about 95% coverage.
- Include sampling uncertainty when the result refers to a lot; otherwise state that it is excluded.
- Method transfer and ongoing QC through control charts and proficiency tests maintain validation status.
Worked example
Relative standard uncertainties are 0.0010 for purity, 0.0020 for weighing, and 0.0030 for dilution and calibration. Combined relative u = √(0.0010² + 0.0020² + 0.0030²) = √(1.0 + 4.0 + 9.0) × 10^-6 = 0.00374, or 0.37%. Expanded U at k = 2 = 0.75%. For a result of 98.2%, report 98.2% ± 0.7% (k = 2). Correlated sources require covariance, not simple summation.
Traps
- Using precision data as the only validation evidence.
- Setting acceptance criteria after seeing the data.
- Omitting dilution or purity from the uncertainty budget.
- Expanding with k = 1 and calling it 95%.
- Reporting a method LOQ below the lowest validated level.