Out-of-Tolerance Instruments: Retrospective Impact, Risk and CAPA

An out-of-tolerance calibration result raises two different questions: what to do with the instrument and which earlier decisions may have been affected. Repair can resolve the first without answering the second. The investigation must connect the observed defect to actual configurations, periods and uses, separating facts, hypotheses and missing information.

Here, out of tolerance means metrological nonconformity against a stated requirement and rule. Out of trend describes departure from expected behaviour, possibly within tolerance; an analytical OOS concerns a test specification. OOT is an ambiguous abbreviation: write its meaning in full in records.

1. Contain the event without losing the initial state

Identify the affected equipment, sensor, channel, software and configuration. Suspend intended use under the procedure, indicate its status and involve the technical owner and quality function. If risks to samples, processes or products emerge, activate relevant measures immediately without waiting for the final certificate.

Preserve original data, sequence, conditions, references, parameters and records before cleaning, adjusting, repairing or replacing components. These actions can remove useful evidence; safety needs remain paramount and unavoidable interventions must be documented. Ask the provider to distinguish initial and final measurements.

Where applicable, 21 CFR 211.160(b)(4) requires remedial provisions when performance limits are not met and prohibits use of instruments that do not meet established specifications. A suspension label must be accompanied by effective control preventing use.

2. Verify what the finding demonstrates

Review the result, units, error sign, reference and its validity, point, configuration and conditions. Confirm that the test covers the chain actually used. An incorrect reference setting or transcription can change the conclusion, but must be demonstrated: suspicion does not justify discarding an unfavourable result.

Distinguish the service limit, intended-use requirement and decision rule. Uncertainty is neither extra tolerance nor an error correction. State which result U relates to and the coverage factor. If the rule includes an inconclusive zone, do not automatically convert it into conformity or nonconformity. JCGM 106:2012 addresses uncertainty in decisions; the numerical criteria in the following case are illustrative only.

A repeat must test a defined hypothesis under documented conditions and criteria. Retain every outcome. Do not repeat until PASS or replace the initial value with one obtained after adjustment.

3. Reconstruct the potentially affected period and uses

Identify the last relevant evidence: a calibration or check covering the characteristic, range and configuration involved. A single-point check does not necessarily bound the period for a defect elsewhere in the range. Discovery date and onset date are not the same by definition.

Link the chronology of calibrations, checks, maintenance, impacts, configuration changes and alarms to uses: samples, preparations, batches, studies, environmental records or process decisions. Include indirect measurements: an error in preparing a reference can propagate to results generated with other instruments.

Record reasons for each inclusion or exclusion. If evidence is insufficient to narrow the period, retain a cautious open scope and state residual uncertainty. The last passing calibration is a starting point to assess, not an automatic onset date or a guarantee for the entire subsequent interval.

4. An original retrospective-impact matrix

The matrix does not automatically determine batch status. Keep the connection between each conclusion, the records examined and the person authorised to approve it.

5. Simulated case: anomaly in the upper range

A thermometer chain has errors e = indication − reference of +0.02 °C at 5 °C, +0.03 °C at 25 °C and +0.45 °C at 60 °C. Respective U values, with k = 2, are 0.04, 0.04 and 0.05 °C. The fictional internal limit is ±0.20 °C. The illustrative rule accepts if |e| + U ≤ 0.20 °C, declares nonconformity if |e| − U > 0.20 °C and leaves other cases inconclusive.

The first two comparisons give 0.06 and 0.07 °C and meet the acceptance criterion. At 60 °C, 0.45 − 0.05 = 0.40 °C: the point is nonconforming under this rule. This does not prove that the defect begins at an exact temperature or existed at the same magnitude months earlier.

Use is suspended. The investigation identifies activities at 60 °C, 35 °C and 5 °C. Those at 60 °C take priority; those at 35 °C remain under assessment because direct evidence is missing; those at 5 °C can be excluded only with relevant support for the historical period too. Today’s favourable result at 5 °C is insufficient on its own.

A positive error means overestimation under the test conditions, but does not authorise subtracting 0.45 °C from every earlier record. Assess the measurement model and process consequences; do not interpolate linear time drift without evidence. Original data remain intact, with any subsequent processing kept separate and traceable.

6. Separate return to service, result decisions and CAPA

Define the intervention, new calibration, checks and authorisation for return to service. Conforming as-left results do not erase as-found results. The competent function decides on results and products through evidence and procedures; a later favourable check does not automatically reconstruct the earlier state. Additional testing must address hypotheses rather than seek an acceptable result.

Distinguish fixing the defect from action on its causes. Investigate wear, conditions, configuration management, check coverage and provider agreements. If the cause remains uncertain, state this and manage plausible hypotheses. CAPA needs an owner, deadline and observable effectiveness criterion; for example, demonstrate that a new check detects the relevant defect. Also review calibration intervals and uncertainty in use. Return to the Calibration and Metrology hub.

7. Sources and scope

Verified on 1 October 2026. The matrix, values and case rule are original and simulated, not prescribed pharmaceutical limits.

1. 21 CFR 211.160(b)(4): eCFR text consulted, displayed update 29 September 2026; US requirements within their applicable scope.
2. EU GMP Chapter 1: full text, effective 31 January 2013, §1.4(xiv), causes and CAPA effectiveness; distinguished from revision drafts.
3. FDA, Investigating OOS Test Results, May 2022, revision 1: full text, sections I, III–V. It covers chemical testing of CDER drugs; it does not equate every metrological anomaly with analytical OOS.
4. JCGM 106:2012: full text consulted, §8; role of decision rules.
5. ISO/IEC 17025:2017: official catalogue, confirmed in 2023; full standard not consulted. Check applicable equipment and nonconforming-work requirements in the controlled copy.