An out-of-specification assay result and pressure to release the batch
In the quality control laboratory of a WHO-GMP oral solids plant, an analyst assays commercial batch MFT-2417 of metformin hydrochloride 500 mg tablets (expiry 03/2027). Specification is 95.0–105.0% of label claim. The result is 92.4%. System suitability had passed (tailing factor 1.1, RSD 0.6% over five injections), the standard was within expiry and the column was new. His raw data show 20 tablets weighed and powdered, with a portion equivalent to 500 mg taken for assay. Next day a second analyst repeats the assay on a fresh portion of the same composite, same column, freshly prepared mobile phase, and obtains 97.1%. The production head argues that 'the second result is within specification, so release the batch' and asks QA to close the investigation by applying an outlier test to the 92.4% figure.
1. Should the 92.4% result be discarded as a statistical outlier? Set out what a proper Phase I laboratory investigation must cover.
Read the worked answer
No. Statistical outlier tests such as Dixon's Q or Grubbs' are not scientifically justified for chemical assays - they are reserved for biological assays, where inherent variability is high. Applying an outlier test to a chromatographic assay is a way of testing into compliance and would be indefensible in an inspection. The second result of 97.1% does not 'cancel' the first; it is one more piece of investigation data and must not be reported as the release result in isolation.
Phase I - laboratory investigation (to be completed before any retesting decision):
- Interview the analyst and review the raw data, chromatograms, integration events, audit trail and sequence.
- Check standard preparation: identity, purity, weighing, dilution, storage and expiry.
- Check sample preparation: 20-tablet composite homogeneity, weighing, dilution volumes, volumetric glassware and its calibration status, pipettes, filtration (adsorption), sample hold time in the autosampler.
- Check instrumentation: balance calibration and daily verification with certified weights, HPLC pump performance, detector linearity, column age and number of injections, mobile phase composition and pH, buffer preparation.
- Check environmental factors and the analyst's training record for this method.
- Compare with the historical trend for this product, this analyst and this equipment; check other batches tested on the same day.
Only when an assignable, documented, reproducible root cause is found may the original result be invalidated and a predefined retest plan begun.
2. Suppose Phase I establishes that the 100 mL volumetric flask used for the sample dilution had a damaged neck and had been re-marked by hand; gravimetric checking shows its true volume is 95.8 mL. Show how this explains the results, and state the retest protocol that is now defensible.
Read the worked answer
Does the arithmetic fit? A flask that delivers 95.8 mL instead of 100 mL makes the solution more concentrated, so the measured content is lower than truth by the factor 95.8/100. Correcting: 92.4 × (100/95.8) = 96.5%, which sits comfortably with the second analyst's 97.1% obtained with a sound flask. The hypothesis is therefore quantitatively consistent, not merely plausible.
Making the hypothesis stand up. Reproduce it deliberately: take a flask of verified 95.8 mL volume, prepare the sample the same way and show that the result falls to about 92%. A worked simulation is what turns a story into an assignable cause.
Defensible retest protocol (all defined in the SOP before the results are seen):
- Declare the original result invalid on the documented laboratory error, and record the 97.1% as investigation data.
- Prepare the retest from a fresh sample of the batch (not the original composite), with the number of independent sample preparations specified in advance - commonly six, analysed against a freshly prepared standard.
- Report every result; do not stop when a passing result appears.
- Accept the batch on the mean of the retest set, with the RSD within the method's acceptance criterion and each individual result within specification.
- If the retest mean also fails, the batch is rejected - a laboratory error does not rescue a genuinely low batch.
3. State the corrective and preventive actions and your advice on batch disposition, including any wider action the investigation should trigger.
Read the worked answer
Disposition. Do not release on the strength of the 97.1% figure alone. Release only after the retest set specified in the SOP is complete, the mean is within 95.0–105.0%, the RSD meets the method criterion, and QA has approved the investigation report and the batch record. If the retest set fails, reject the batch and extend the investigation to a possible manufacturing cause (blend uniformity, compression, content uniformity).
Corrective actions.
- Remove and destroy the damaged flask; quarantine and physically inspect all volumetric glassware for chips, re-marking and legible calibration stickers.
- Repeat the glassware calibration programme and increase its frequency; move to gravimetric verification of critical flasks.
- Retrain the analyst and the whole QC team on volumetric technique and on the OOS procedure.
Preventive actions.
- Impact assessment on other batches: identify every batch analysed with that flask or by that analyst in the affected period; risk-assess whether retesting is needed. This is the step most often skipped.
- Review the glassware log and calibration schedule; introduce a check that no hand-marked glassware enters the laboratory.
- Reinforce the OOS SOP so that no supervisor can direct an analyst to 'just repeat it'; ensure the analyst's right to refuse is documented.
- Trend laboratory errors in the quality management review so recurring causes are visible.
The production head's request should be declined in writing, with the reason recorded in the investigation file.