Method Validation, Measurement Uncertainty and Proficiency Testing

Begin with intended use: which analyte or class, what specimen, which range and what conclusion or decision. A method suitable for a purified tablet extract may be unsuitable for trace blood analysis. Scope and required discrimination shape the study.

A standard method needs verification that the laboratory can perform it adequately. Nonstandard, laboratory-developed, modified or out-of-scope methods require appropriate validation. The extent of work follows the intended use and relevant change; not every column-lot replacement requires an identical full study.

Set acceptance criteria before evaluating the data. Selectivity studies should address realistic interferents and relevant alternatives, not only one blank specimen. An apparently clean transition can still be shared by an isomer or matrix component.

Precision describes agreement among results under stated conditions. Repeatability and intermediate precision capture different variation. Replicate injections of one extract assess only part of the full preparation-and-measurement process and cannot establish every between-day or operator effect.

Bias assesses systematic difference from an appropriate reference. Accuracy is a broader qualitative concept involving agreement, not simply a named recovery percentage. Extraction recovery and matrix response need separately designed comparisons; apparently complete recovery does not exclude interference or calibration bias.

Calibration evaluation examines the model over the relevant range using residuals, back-calculated values and suitable controls. A high R² can coexist with unacceptable low-end bias. Polynomial fitting, weighting or range changes need justification rather than selection merely to make standards pass.

Detection capability and quantification limits follow a defined approach and demonstrated performance. If a response standard deviation is used to estimate a concentration limit, conversion through the calibration response relationship is required. Multiplying signal units by three does not directly produce concentration units.

Signal-to-noise approaches can be useful for suitable signals and definitions, but are not universal. A claimed LOQ should meet relevant precision, bias and other criteria at that level. A negative finding remains limited to the method's scope and capability.

Stability, carryover, dilution and robustness studies address other important risks. Deliberate small changes can test sensitivity to operating conditions, while actual specimen storage needs its own relevant evidence. There is no universal 20-percent-of-LOQ carryover rule for every examination.

Keep a reviewable record of the plan, materials, method version, data, calculations, criteria, limitations and approval. A validation file supports continued use within its scope; ongoing controls and change assessment determine whether that evidence remains applicable.

Measurement uncertainty is a non-negative parameter characterising dispersion attributed to a measurand on the available information. Define the measurand and result before constructing the evaluation. An instrument's displayed standard deviation is not automatically the uncertainty of the final specimen result.

The JCGM GUM suite provides frameworks for evaluation and expression. The current BIPM catalogue includes JCGM 100:2008, its 2026 amendment on nonlinearity, and complementary documents on models, examples and distribution propagation. Use the approach appropriate to the actual measurement.

Type A evaluation uses statistical analysis of observations. Type B uses other information such as certificates, specifications or prior data. These labels describe evaluation method, not a simple division into random and systematic errors.

Convert contributions to standard uncertainties using justified distributions or stated coverage information. A rectangular tolerance with half-width a has standard uncertainty a/√3 under that model. A certificate expanded uncertainty is divided by its stated coverage factor when appropriate; every 95-percent interval cannot automatically be divided by exactly two.

Propagate relevant contributions through the measurement equation using suitable sensitivity coefficients. Independent contributions in a linear or adequately linearised model can be combined by a weighted sum of variances. Correlations require covariance terms, and nonlinear cases may require other treatment.

For an appropriate product or quotient of independent inputs, relative standard uncertainties can sometimes be combined in quadrature. The formula is not universal for every forensic model. Averages, corrections, calibration prediction and nonlinear relationships require their own assessment.

Empirical approaches can use suitable precision, bias and quality-control information. Avoid double counting effects already represented in those data. Sampling and specimen heterogeneity may be relevant to the intended measurand and should not disappear simply because the instrument is precise.

Expanded uncertainty is U = kuc for an applicable coverage-factor approach. A factor of two often approximates 95-percent coverage under suitable conditions, but distribution and effective degrees of freedom matter. State relevant assumptions and reporting conventions.

Uncertainty is not a licence to ignore known significant bias or failed controls. Address known effects appropriately in the model and method. A broad interval cannot repair a false identification, mislabeled specimen or unsupported source claim.

Report uncertainty when required or relevant to the result's application. Where conformity is stated against a limit, use the applicable specification and decision rule. The uncertainty interval alone does not decide a legal threshold in every US, UK, Indian or other jurisdiction.

A control chart follows a suitable process measurement over time against a target and defined limits. Shewhart charts, CUSUM and other approaches have different detection behaviour. Select a stable basis for limits and rules suited to the process rather than reset limits to hide a change.

Illustrative warning and action bands at two and three standard deviations are common but not mandatory for every method. Trend or run rules can flag sustained change even when individual points remain inside a band. The exact rule and response follow the quality procedure.

Internal QC can reveal bias when an appropriate independent assigned-value control is used. A chart centred on a laboratory's own biased historical mean may fail to expose that bias. PT is an important external tool, but it is not the only possible way to investigate systematic error.

Choose proficiency testing or other suitable interlaboratory comparisons to represent relevant activities, matrices and decisions. Consider provider competence, material homogeneity and stability, scoring, schedule and accreditation scope. Participation alone does not validate all methods.

For a common quantitative scheme, z = (x − xpt)/σpt, where x is the participant result, xpt the assigned value and σpt the standard deviation for proficiency assessment. The denominator is a performance yardstick selected by the scheme, not automatically the CRM uncertainty or participant standard deviation.

A common interpretation is |z| ≤ 2 satisfactory, 2 < |z| < 3 questionable, and |z| ≥ 3 unsatisfactory. Follow the actual scheme, including boundary conventions. A score near zero is not proof that the reported uncertainty or every preparation step is correct.

When assigned-value uncertainty matters, a z-prime score can include it. Zeta or En scores assess agreement relative to reported standard or expanded uncertainties under their definitions. Their scales and interpretation should not be substituted for ordinary z scores without understanding the scheme.

Qualitative or interpretive PT can use different assessment criteria rather than a concentration z score. Correct substance identification, particle classification or report interpretation can involve errors that a numerical score alone does not describe.

Investigate questionable or unsatisfactory performance and assess the effect on relevant work. Corrective action and follow-up depend on cause, significance and programme requirements. One unsatisfactory score does not automatically suspend accreditation or prove every case result is wrong.

UNODC's International Collaborative Exercises cover defined drug-analysis activities, and forensic PT-provider directories offer other starting points. Verify actual scope, eligibility and current availability. A provider name is not a universal participation requirement for all laboratories.

Do not choose acceptance limits after seeing the result. A method can appear successful if only favourable specimens, concentrations or days are selected. Include conditions relevant to real work and document exclusions or failed studies transparently.

Separate instrument precision from method precision and extraction recovery from matrix effects. Treat unit conversions, dilution factors and reference purity as explicit parts of the calculation. A plausible concentration cannot compensate for incorrect arithmetic.

Avoid assigning detection limits from one attractive low signal without an appropriate study. Equally, do not assume a blank-SD formula is the only valid approach. Assess whether the selected model represents false-positive, false-negative and quantitative performance relevant to the intended use.

Assess changes according to their potential impact. A new source setting, preparation, internal standard or specimen matrix can affect selectivity and bias even if retention looks similar. Document the rationale for verification, partial validation or broader work.

Review accumulated control and PT data alongside maintenance, training and method changes. Consistent results in one control material may not represent every difficult specimen. Investigate discrepancies between types of evidence rather than accept whichever is most reassuring.

A report should communicate a supported result with relevant units, method, limitations and uncertainty where applicable. Avoid fixed claims that these activities guarantee court acceptance. The scientific record should allow a competent reviewer to assess both the evidence and its practical limits.

Related reading: Sample Preparation, Purification and Instrument Calibration.

Related reading: Tandem Mass Spectrometry (MS/MS) and Spectral Interpretation.

Related reading: ISO/IEC 17025 and Quality Management in Forensic Laboratories.