thebench.bio #1 Ecosystem for the Bio-Tech Industry
All Quality Control questions

Question 19 of 20 Advanced Practical

How would you design control-chart trending that warns before an assay fails specification?

Short answer

Chart a stable control sample or the reference standard response on every run, set limits from its own historical variability rather than the specification, and apply run rules such as points beyond 3 sigma, 2 of 3 beyond 2 sigma, or 8 in a row on one side. Rules flag shifts long before results reach a limit.

Why

A specification is the edge of acceptable product. Control limits describe how the measurement normally behaves. A Shewhart chart sets its center line at the process mean and limits at plus and minus 3 standard deviations, which for normal data are crossed by chance only about 0.3 percent of the time, as the NIST engineering statistics handbook explains.

Rules based on the same probability logic catch smaller shifts. The Western Electric rules include 2 of the last 3 points beyond 2 sigma, 4 of the last 5 beyond 1 sigma, 8 consecutive points on one side of the center line, and 6 in a row trending up or down. For a QC assay, the best thing to chart is a control sample that is stable, homogeneous and run every time, because product results mix process and method variation.

If the control sample's mean shifts by 1 percent, real product results shift too, and the lab learns about it from a rule violation instead of an OOS. The chart needs a baseline long enough to estimate normal variability, collected while the method is known to be in control, and limits should be recalculated only for a documented reason, never to make alarms stop. The FDA's OOS guidance expects laboratory management to watch for developing trends, which a chart makes routine.

NIST notes that adding these rules also increases false alarms, so choose a small set and define the response to each in the procedure.

How to work it through

  1. Choose a stable, homogeneous control sample that is run in every analytical run.
  2. Collect a baseline of results, long enough to estimate normal variability, while the method is known to be in control.
  3. Set the center line and 3-sigma limits from that baseline, not from the specification.
  4. Select a few run rules and define the required response to each.
  5. Plot every run as it is reviewed and investigate rule violations promptly.
  6. Recalculate limits only after a documented, justified change.

An example

A potency assay control has a baseline mean of 100.0 percent and standard deviation of 1.2 percent over 25 runs, giving limits of 96.4 to 103.6. The product specification is 90 to 110 percent. Over two weeks, the control reads 101.0, 101.5, 100.4, 101.8, 100.9, 101.2, 100.6 and 101.4, eight in a row above center, though none is near a limit.

The rule triggers an evaluation, which finds a new lot of detection reagent introduced at the first point. A comparison study confirms a 1.1 percent bias, and the reagent lot is replaced before any product result is affected.

Where the answer stops holding

Charts assume a stable baseline and roughly independent results, so autocorrelated or skewed data need adapted limits. A control sample can degrade over time and create a false trend. Run rules increase false alarms, and very low-frequency assays take a long time to accumulate enough points.

A strong answer covers

  • Chart a stable control, not only product results
  • Limits from method history, not specification
  • Run rules for shifts and trends
  • Baseline from enough in-control runs
  • Predefined response and controlled limit changes

Common mistakes

  • Using specification limits as control limits.
  • Recalculating limits to absorb a shift instead of investigating it.
  • Applying every possible rule and drowning the lab in false alarms.