How to set up a practical and appropriate control chart RSD?

Question

How to set up a practical and appropriate control chart RSD, which is tight enough to provide good test performance control but does not cause unnecessary 2nd retest due to a too-tight RSD?

--

(This question was received recently. Please treat it with kindness, respect and help contribute to a collaborative, growth-supporting culture. Thank you! - CBWGC)

Ryan Santiago · Answer

We used to evaluate our method RSD through method validation then compare this with published standards.

We did 30+ runs of a control sample over several days then evaluate the repeatability and reproducibility, RSD, among other parameters. Some samples have different RSD from published standards due to matrix difference, other factors, but since we get comparable RSD and acceptable recovery (spike) from the run, we assume this to be the acceptable RSD for those specific samples.

I hope this answers your question

David Woollard · Answer

I assume you have established a control sample with at least 7-10 days of independent results. Don't use same-day replicates.

These form the basis for future acceptance. Plot them. On the control chart mark the horizontal 2x SD lines and the 3x SD lines (+/-) and the mean.

If a future result exceeds 2x SD then consider it suspicious enough to review the data set. Check the calibration etc.

Reject the run if a future result exceeds 3x SD. Repeat the batch.

The question arises if a new result is incorrectly rejected because the SD limits are too tight.  This requires discussion and a decision by the Quality Manager.  The limits plus mean can be recalculated and put into the control chart. Only do this openly and sparingly.

It only takes a short time to decide whether the control chart is correctly set.

There are cases where a control sample is unstable and changes in composition. Eg The analyte might be unstable or the sample absorbs moisture. The control chart will show this by repeated results lying on one side of the mean. If the changes are slow then they can be accommodated with scheduled updates to mean and SD. If they are quick then the control is useless and needs replacing or upgrading. Freezing control in ready-to-use portions is a common practice.