7. Test and calibration methods, method validation and quality control [5.4, 5.9]
7.6 Application of quality control and managing response to results [5.9]
Laboratories have traditionally approached quality control by including items of known properties in each test or calibration batch and evaluating the results against defined criteria to decide whether the data for the batch should be rejected or accepted. This approach has the virtue of simplicity and, provided the accept/reject criteria are set properly, it will defend the laboratory’s contention that released data continues to meet defined performance characteristics.
However, if a laboratory only uses its quality control data in this fashion, it is failing to make full use of it. A laboratory might have a situation where all quality control samples are producing data which falls within the acceptance limits but always on one side, for example they may be consistently high relative to the expected value. This situation bears investigation since there should be a random scatter about the expected value. This bias, therefore, gives an early warning of a problem with the test or calibration system. What is really useful is that the problem has been detected before data is compromised.
Increasingly, assessors expect laboratories to make use of their quality control data in this fashion. Paragraph 5.9.1 of ISO 17025 makes explicit reference to recording data in such a way that trends can be detected, which strongly implies the use of control charts. The
next section gives an introduction to the use of control charts which might be adopted by a laboratory which is new to the area of statistical quality control.
7.6.1 Statistical quality control and preventive action [5.9, 4.12]
There should be an active co-ordination of quality control with a regular review of performance, and records should be kept of the results of reviews and of any action taken in response. This should provide a mechanism for anticipating problems with methods before they affect quality and as such is an important contribution to preventive action. See section 4.9 for further discussion of preventive action.
The management should collect together the results from quality control samples for each method and plot the data on a control chart. The most common control chart format is the Shewhart, chart where the difference between the expected and found values for the quality control samples is plotted against time.
The Shewhart chart (see next page) gives a general visual indication when any systematic drift in the values returned by quality control samples is setting in. A method which is being used under control should show a random distribution of the actual values about the expected result, and any trends developing which suggest a bias should be investigated.
It is usual to mark the control chart with a warning limit at two standard deviations (2σ) and an action limit at three standard deviations (3σ). The standard deviation is generally derived from the performance data determined at method validation.
The laboratory should, wherever possible, have a documented policy for deciding when the chart indicates a condition where the method should come under investigation, and this policy should be expressed quantitatively so that it will be applied consistently. In setting the rules for a particular method, the following widely accepted practice should be borne in mind.
Approximately 5% of results may be expected to fall outside the 2σ warning limits. Any result outside the action limits requires investigation.
Two consecutive results outside the warning limits need an investigation.
A consistent run of successive results on the same side of the expected value should be investigated. It is widely accepted practice that a run of eight such successive results triggers an investigation although many laboratories would feel the need to respond rather earlier than this.
Consideration should be given to updating the control limits at regular intervals. If the method is consistently delivering data inside the current warning limits, then this may indicate that the uncertainty being achieved in practice is improving relative to the data generated at validation. Conversely, if the method is slipping out of control too often, this suggests that the validation data is presenting an optimistic picture.
0.1
A common rule of thumb is to review the data every sixty points. If 1-6 (inclusive) points are found outside the 2σ limits, then the indication is that the limits are satisfactory. If more points are found outside, then the limits are optimistic and either they should be revised or the method investigated in order to bring it back to a level the performance required. If no points are found outside 2σ, then new limits should be set reflecting the enhanced performance. Some laboratories take the view in this last case that they will not reduce the limits since this will result in increased rejection of data. In these circumstances, the decision on whether to revise to tighter control limits will be determined by whether the un-revised limits are acceptable in the sense that they indicate that the method is fit for purpose.
7.6.2 Frequency of quality control checks [5.9]
There is no simple answer to how frequently quality control items should be run. The trite answer is as often as necessary.
A general rule of thumb is that they should be included at a minimum rate of one quality control item in twenty but ideally one in ten. Experience of a method in a particular laboratory may indicate that more frequent checks are required.
In the case of methods which involve batch treatment of items, at least one quality control should be present in each batch.
Some items can be tested or calibrated in duplicate as a check on the reproducibility of the method. It is far more useful, however, to expend the same effort in testing another
Shewhart Control Chart
Action Limit +3σ 1.9 Warning Limit +2σ 1.6 Expected value 1.0 D a t e o f t e s t Warning Limit -2σ 0.4 Action Limit -3σ X X X X X X X X X X X X X X X X X X X X X X X X X
quality control standard since this not only checks consistency but also gives information on overall error.
A method which can only be controlled by a high frequency of quality control checks should be looked at very carefully and seriously considered for replacement by a more stable method.