Method Validation, Measurement Uncertainty and Proficiency Testing

ISO 17025:2017 clause 7.2.2 makes validation mandatory for any non-standard or laboratory-developed method, which covers almost every forensic toxicology and chemistry method at an Indian SFSL or CFSL. NABL Document 141 inspects the file on every accreditation visit. The Bharatiya Sakshya Adhiniyam does not name validation in Section 63, but the courts read into the section a requirement that the method be one a competent expert would accept as reliable, and the file is how a laboratory documents that. The parameter list is consistent across ICH Q2(R2), the FDA Bioanalytical guidance with 2022 ICH M10, AOAC, SOFT and SAMHSA, and the WADA International Standard for Laboratories.

Accuracy is percentage recovery from a spiked CRM with the 80 to 120 percent window for forensic toxicology. Precision splits into repeatability (intra-day, RSD below 5 percent), intermediate precision (inter-day with different operators, 5 to 15 percent) and reproducibility (inter-laboratory, what PT tests). Linearity needs R² above 0.995 across five to seven calibrators in triplicate, with back-calculated concentrations within ±15 percent of nominal (±20 percent at LOQ). A high R² alone is insufficient; an over-fitted polynomial can show R² of 0.999 and still produce back-calculated concentrations well outside acceptance criteria at the curve extremes.

LOD and LOQ come from statistics, not eyeball. The 3σ method runs ten blank-matrix injections, takes the SD of the response, and reports LOD as three times that SD; LOQ is the same with a factor of ten. Specificity is the demonstration that no interfering peak appears at the analyte's retention time and m/z transition in a blank matrix or with structural analogues. Robustness is the demonstration that small intentional variations (pH ±0.1, column temperature ±2 °C, flow ±5 percent) do not shift the result outside the precision window. Carry-over checks that a blank following the highest calibrator gives no peak above 20 percent of LOQ. System suitability is the pre-run check that catches an instrument drifting before the batch begins.

The Guide to the Expression of Uncertainty in Measurement (GUM, JCGM 100:2008) is the international consensus on how to compute and report uncertainty. NABL has adopted it through TR-001 and ISO 17025 clause 7.6. The BSA Section 63 certificate is now expected, as best practice, to include the expanded uncertainty alongside the reported value.

GUM splits contributions into Type A (statistical evaluation of repeated measurements, the SD of the mean) and Type B (the calibration certificate of the balance, the CRM's stated uncertainty, the volumetric flask's tolerance, temperature variation, matrix inhomogeneity). Each Type B contribution becomes a standard uncertainty by dividing the tolerance by an appropriate divisor (square root of three for a rectangular distribution, two for a stated 95 percent confidence interval). The combined standard uncertainty u_c is propagated through the measurement equation; for the multiplicative model most forensic methods use, u_c/c is the square root of the sum of (u_i/x_i)² across every contribution. The expanded uncertainty U is u_c multiplied by a coverage factor k, with k=2 giving roughly 95 percent confidence. The convention is "ethanol 110 ± 5 mg/100 mL (k=2)".

Bottom-up, the strict GUM method, identifies every contribution from first principles and computes a full budget; rigorous but laborious. Top-down uses validation data, CRM bias and PT performance to estimate combined uncertainty empirically. NABL accepts top-down for routine forensic work. For a typical LC-MS/MS toxicology method, contributions are repeatability (1 to 3 percent), calibration curve (0.5 to 2 percent), reference standard (0.3 to 1 percent for Sigma, 0.1 to 0.5 percent for IPC), sample inhomogeneity (0.5 to 2 percent for whole blood) and matrix effect on ionisation (1 to 3 percent with a deuterated IS). Combined in quadrature, u_c typically lands at 2 to 4 percent and expanded U at k=2 at 4 to 8 percent.

A validated method is not a permanent guarantee. Instruments drift, reference standards age, columns degrade. The Shewhart control chart catches this: a time-series plot of a QC measurement (spiked blank or CRM) against date, with warning limits at ±2σ and action limits at ±3σ from validation data. A point outside the action limits triggers investigation. Trend rules catch drift that single-point limits miss: seven consecutive points on one side of the mean indicates a systematic shift even within warning limits; a steady trend of six or more points often traces to a column ageing or a detector lamp dimming. The Western Electric rules formalise these patterns. The CUSUM chart catches a 0.5σ shift in five to ten measurements where Shewhart needs twenty to thirty. CFSL Chandigarh runs NIST SRM 1577c monthly on the ICP-MS panel and tracks lead, arsenic, mercury and cadmium on individual Shewhart charts.

Internal QC catches drift but not bias. A laboratory that consistently quantitates morphine 12 percent low because of an extraction step that strips the analyte will get tight Shewhart charts and a wrong number every time. The only way to catch systematic bias is proficiency testing.

The standard scoring statistic is the z-score: participant minus assigned value, divided by the target SD. |z| below 2 is satisfactory, 2 to 3 questionable, above 3 unsatisfactory. The target SD comes from the CRM's certified uncertainty, the consensus SD of participants (Algorithm A from ISO 13528, robust to outliers), or a fitness-for-purpose value. The En score incorporates each laboratory's stated MU. ISO 17025 clause 7.7.2 requires regular PT participation; NABL 141 expects at least annual participation for every major analyte class. An unsatisfactory result triggers investigation, root-cause analysis, CAPA, and retest where the scheme allows. Repeated unsatisfactory results without effective CAPA can cost the affected scope from accreditation.

Over-fitted calibration curves are the most common: R² reads 0.9999 but back-calculated concentrations at the lowest and highest calibrators deviate by 25 percent because the analyst fitted a quadratic where a linear model belonged. Skipping intermediate precision is the second; files often report intra-day RSD only across six replicates within an hour, and the inter-day RSD across three to five days never makes it into the file even though that is what case work actually experiences. LOD and LOQ from a single low-concentration injection are the third; the chromatographer eyeballs the lowest standard at S/N of around 10:1 and writes that into the LOQ field instead of running the 3σ and 10σ calculation from blank-replicate SD.

Failing to revalidate after a meaningful method change is the fourth. Column lot, mobile phase composition, internal standard swap to a deuterated isotopologue. Each should trigger at least a partial revalidation. A change visible in the SOP revision history without a corresponding revalidation record is a finding at accreditation visits and a potential challenge point in court. Omitting the MU statement from the Section 63 certificate is the fifth pitfall, and the one with the most direct courtroom consequence. The validation file has the calculation, the SOP references it, but a certificate template written before the requirement became standard practice may omit it. Updating the template to include the expanded uncertainty and coverage factor resolves this.