The CRM Failure That Became a Matrix Effect
On 14 March a laboratory runs a batch of 96 groundwater samples for nitrate. The batch includes a certified reference material at 10.0 mg/L with an acceptance range of 9.0 to 11.0 mg/L; the recovered value is 7.8 mg/L. The analyst re-runs the twelve samples closest to the regulatory limit, obtains results within range, and reports only those twelve. Results for the other 84 samples had already been sent to the client in a preliminary spreadsheet two days earlier, and the final signed report repeats the same values. The supervisor writes in the batch log: 'CRM low - matrix effect, no further action.' No nonconforming work record is opened, no correction and no corrective action are raised, and the client is not informed. Two months later the same low recovery appears in a second batch from a different analyst, and that batch is accepted as well.
1. Has nonconforming work occurred, and does the selective re-run of twelve samples constitute an adequate correction?
Read the worked answer
Nonconforming work has clearly occurred. The batch failed the laboratory's own acceptance criterion - 7.8 mg/L against 9.0 to 11.0 - and was released anyway. Clause 7.10 requires a documented procedure that, when nonconforming work is identified, addresses: actions to manage the work; evaluation of the significance of the nonconformity, including its impact on previously reported results; notification of the client where results were already supplied; recall of work or reports where necessary; and authorisation of resumption.
None of these happened. There is no record, no impact evaluation, no notification and no formal authorisation - work simply continued.
The twelve-sample re-run is not a valid correction. Re-running the samples nearest the limit addresses the samples the analyst judged risky, not whether the whole batch is biased. A low CRM recovery is a bias signal, and bias applies to all 96 samples, not only the twelve near the threshold. If anything the twelve were the worst samples to choose selectively, because they are the ones where a reviewer would want the corrected values alongside the original ones.
A correction must cover the affected scope: the whole batch, plus any earlier batch produced with the same calibration, reagents or analyst, because the cause may predate this run.
2. Distinguish correction from corrective action in this case, and explain why 'matrix effect' is not an acceptable explanation.
Read the worked answer
Correction fixes the immediate output: re-analyse the affected samples after the cause is found, withdraw and re-issue the report, notify the client. Corrective action (clause 8.7) removes the cause so it does not recur: investigate, determine the root cause, implement action, review effectiveness, update the risks and change the system if needed. This laboratory did neither.
'Matrix effect' is a hypothesis, not a finding. It has to be tested: spike recovery in the same matrix, a second certified material, standard addition, an independent method, or a calibration verification. The whole purpose of a CRM certified at 10.0 mg/L is that recovery outside the range triggers an investigation - otherwise the laboratory has bought a quality-control sample and given it no power.
Accepting the explanation on assertion converts an unexplained bias into a documented excuse, and once it is written in the batch log the next analyst inherits it as precedent.
The second occurrence is decisive. Different batch, different analyst, same direction of bias - this is systemic, not personal. Candidates include calibration drift, a reagents lot, digestion temperature, blank contamination or a change in the instrument's condition. A recurring pattern that is never investigated is exactly what the corrective action requirement exists to catch, and it should also have surfaced in management review as a trend.
3. What does the laboratory owe the client and its records, and how should the twelve re-run results be handled?
Read the worked answer
Client. Where nonconforming work has affected reported results, the laboratory must notify the client in writing, recall or amend the reports, and explain the impact. It cannot leave a regulator relying on values it now knows to be questionable. A phone call is not sufficient - the client needs a written statement of what was affected, why, and what is being done.
Records. A nonconforming work record, the significance evaluation and its basis, the recall and notification, the corrective action file with root cause and effectiveness check, and the technical records of the re-analysis. Records must allow the original calculation to be reconstructed, including the values that were withdrawn.
The twelve results. They cannot be presented as if they were the batch. Two honest options: re-analyse the whole batch once the cause is identified and the system verified, or report with an explicit statement of which samples were re-run, on what basis, and what uncertainty attaches to the remainder. What is not acceptable is a signed report in which twelve corrected values sit next to 84 uncorrected ones with no indication that they differ in status.
Finally, the supervisor's willingness to override a failed acceptance criterion is a system issue, not just an analyst issue. Batch review and result release authorisation need to be examined under the personnel competence and authorisation requirements.