Case study — Pharmaceutical
Detecting HPLC drift 6 weeks early — and avoiding a batch recall
Routine QC data contained a clear drift signal for six weeks before the instrument failed its calibration check. SPC trend analysis would have caught it. Now it does.
6 weeks
Earlier drift detection
£1.2M
Estimated recall cost avoided
23 batches
Retrospectively at risk
4 instruments
Now under SPC monitoring
The challenge
The client was a contract pharmaceutical manufacturer producing solid oral dosage forms for several branded and generic customers. Their analytical laboratory used HPLC for assay and related substances testing on all finished product batches.
During a routine quarterly calibration check, one of the four HPLC instruments failed its system suitability criteria. Investigation revealed that the instrument's detector response had been drifting for approximately six weeks — a period during which 23 finished product batches had been tested and released on the basis of results from the drifting instrument.
The regulatory and commercial consequences were serious. The client was required to notify affected customers, conduct a retrospective review of all 23 batches, and assess whether any product had been incorrectly released. The estimated cost of the investigation, customer notifications, and potential recall was £1.2 million. The actual recall was avoided — but only because the drift had been in the direction of conservative results (underestimating assay values). A drift in the opposite direction would have resulted in a mandatory recall.
The data
The client's LIMS stored all QC results, including the system suitability data (retention time, peak area, tailing factor, theoretical plates) from every analytical run. This data had been collected for over four years across all four instruments — a rich dataset that had never been analysed for trends.
Retrospective analysis of the drifting instrument's system suitability data revealed that the drift had been clearly visible in the peak area response for the reference standard for six weeks before the calibration failure. The signal was present; it had simply not been looked for.
Dataset
Methodology
Retrospective QC data analysis
We extracted four years of system suitability data for all four instruments from the LIMS. For each instrument and each system suitability parameter, we calculated baseline statistics from the first 18 months of data (a period of confirmed good performance) and constructed individuals control charts.
Drift signal characterisation
Applying the Western Electric detection rules to the retrospective data, we identified the point at which the drift signal first became statistically significant. For the instrument that had failed, the first signal (a run of 8 consecutive points below the centreline for peak area response) occurred 43 days before the calibration failure.
Prospective monitoring system design
We designed a prospective SPC monitoring system for all four instruments, covering six system suitability parameters per instrument. The system was implemented in the client's existing LIMS, which had the capability to generate control charts but had never been configured to do so.
Alert thresholds and response procedures
We defined alert thresholds for each detection rule and worked with the client's QA team to develop response procedures: what to do when a signal is detected, who is responsible, what investigation is required, and when to take the instrument out of service for maintenance.
Key findings
The retrospective analysis confirmed that the drift signal had been present for 43 days before the calibration failure. The first Western Electric signal (run of 8 below centreline) occurred on day 1 of the drift period. A second signal (trend of 6 consecutive declining points) occurred on day 12. Both were clearly visible in the data.
Analysis of the other three instruments revealed that one had a low-level trend in tailing factor that had not yet triggered a calibration failure but was consistent with early-stage column degradation. This instrument was scheduled for preventive maintenance before the trend progressed further.
The analysis also identified a systematic difference between two analysts in the peak area response for one instrument — a finding that led to a retraining exercise and a revision of the sample preparation procedure.
The drift signal was present in the system suitability data for 43 days before the calibration failure. Two separate Western Electric signals had been generated — and missed — before the instrument was taken out of service.
Implementation
LIMS control chart configuration
Control charts for six system suitability parameters were configured in the LIMS for all four instruments. Charts are updated automatically after each analytical run and reviewed by the laboratory supervisor at the start of each working day.
Automated alert generation
The LIMS was configured to generate an automated alert when any Western Electric rule is triggered. Alerts are sent to the laboratory supervisor and the QA manager, with a required response within 24 hours.
Response procedure documentation
A new SOP was written covering the investigation and response to SPC signals on analytical instruments. The SOP defines the escalation pathway, the criteria for taking an instrument out of service, and the retrospective review requirements for batches tested during a drift period.
Outcomes
In the 12 months following implementation, the SPC monitoring system generated four alerts across the four instruments. Three were investigated and found to be attributable to reagent lot changes (which produced step changes in the system suitability data); the instruments were recalibrated and returned to service. One alert identified genuine detector degradation, which was addressed with a preventive maintenance intervention before any batches were affected.
No instrument has failed a calibration check since the monitoring system was implemented. The client estimates that the system has avoided at least two potential calibration failures that would previously have gone undetected until the quarterly check.
The client's regulatory affairs team used the implementation of the SPC monitoring system as evidence of continuous improvement in their next GMP inspection. The inspector noted it as an example of proactive quality management.
Concerned about instrument drift?
If your laboratory collects system suitability or QC data and does not analyse it for trends, you are likely missing signals. We can set up the monitoring and show you what is already in your data.
Get in touch