Most analytical laboratories check their quality control samples against acceptance criteria: if the QC result falls within the defined range, the run is accepted. If it falls outside, the run is rejected and the instrument is investigated.

This approach catches failures. It does not catch drift.

Drift is gradual. An instrument that is slowly losing sensitivity, a reference material that is degrading, a calibration that is slowly shifting — these produce QC results that pass the acceptance criteria today, and will continue to pass for weeks or months, while the results they are validating become progressively less accurate.

By the time the QC result finally fails, the problem has been present for a long time. The question is how many results were affected.

What instrument drift looks like in QC data

Drift has a characteristic signature in QC data: a gradual, directional trend. Results that were consistently near the centre of the acceptance range begin to migrate toward one edge. The migration is slow enough that no individual result triggers an alarm, but the pattern is unmistakable when you plot the data over time.

Other patterns are also informative:

Increasing variability. Results that were tightly clustered begin to spread. This is often a sign of a mechanical problem — a worn pump seal, a degrading detector, a loose connection — rather than a calibration issue.

Step changes. A sudden shift in the QC result level, without a corresponding change in the acceptance criteria, often indicates a reagent lot change, a recalibration, or a maintenance event. Step changes are not necessarily problems, but they should be documented and investigated.

Cyclical patterns. QC results that vary with the day of the week, the time of day, or the season often indicate environmental factors: temperature, humidity, or the behaviour of specific analysts. These patterns are invisible to pass/fail checking and require time-series analysis to detect.

Why pass/fail checking is not enough

The acceptance criteria for a QC sample are typically set at ±2 standard deviations from the mean of the characterisation data, or at a fixed percentage of the nominal value. A result within this range is accepted; a result outside is rejected.

This is a reasonable approach for detecting gross failures. It is a poor approach for detecting drift, because it treats each QC result as independent. It does not ask whether the current result is consistent with the recent history of results.

Statistical process control provides a better framework. Control charts for QC data use the historical performance of the QC sample to define control limits — the range of variation that is consistent with normal instrument performance. Points outside the control limits, or patterns within the limits (runs, trends, cycles), are signals that something has changed.

The key difference is that control charts are sensitive to patterns, not just individual extreme values. A run of seven consecutive results above the centreline is a statistically significant signal, even if every individual result is within the acceptance criteria.

Setting up QC trend analysis

The starting point is your existing QC data. Most LIMS systems store QC results with timestamps, and most laboratories have years of historical data. The challenge is usually not data availability — it is data structure and analysis workflow.

Step 1: Extract and structure the data. For each QC sample and each analyte, extract the historical results with their timestamps. Separate results by instrument, analyst, and reagent lot if these are tracked.

Step 2: Calculate baseline statistics. Using a period of known good performance, calculate the mean and standard deviation of the QC results. These become the centreline and control limits for your control chart.

Step 3: Plot the data. A simple time-series plot of QC results, with the centreline and control limits overlaid, is the foundation of the analysis. This can be done in Excel, R, or any statistical software.

Step 4: Apply detection rules. In addition to points outside the control limits, apply the Western Electric rules: a run of 8 or more consecutive points on one side of the centreline, a trend of 6 or more consecutive points in one direction, or 2 out of 3 consecutive points in the outer third of the control limits.

Step 5: Investigate signals. Every signal should be investigated and documented. The investigation should determine whether the signal represents a real change in instrument performance, a change in the QC material, or a change in the analytical procedure.

The value of retrospective analysis

If you have not been doing systematic QC trend analysis, a retrospective analysis of your historical data can be revealing. It will show you:

• How often your instruments have drifted before being detected
• Whether certain instruments or analytes are more prone to drift
• Whether there are seasonal or cyclical patterns in your QC data
• Whether reagent lot changes have caused systematic shifts in your results

This retrospective analysis is also useful for demonstrating the value of the trend analysis programme to management. Showing that drift was present for three months before it was detected — and that systematic monitoring would have caught it in three weeks — is a compelling argument for investing in the analysis infrastructure.

Reducing rework through early detection

The cost of instrument drift is not just the cost of the failed QC result. It is the cost of the rework: re-running samples, re-calibrating instruments, investigating which results were affected, and potentially notifying customers or regulators.

Early detection through trend analysis reduces all of these costs. A drift that is caught when the QC result is still within the acceptance criteria — but showing a clear trend toward the limit — can be addressed with a recalibration or a maintenance intervention, before any customer samples are affected.

The investment in systematic QC trend analysis is modest. The return, in reduced rework and improved data quality, is substantial.