QC Design

Based on the Sigma-analysis of the Sysmex xt 1800i, we apply QC Design and analyze the different QC procedures appropriate for different quality requirements.

Sigma metric analysis of the data (a separate article)

RBC

HCT

PLT

Summary of QC procedure recommendations

Evaluating the Error Detection with Different QC Procedures

Conclusion

March 2009
Sten Westgard, MS

RBC

HCT

HCT is a more difficult situation. Using the Biologic-based quality requirement, the Sigma is 2.87, while according to the CLIA quality requirement, the Sigma is 4.64. That's a big difference which manifests itself in radically different QC recommendations.

Using the Biologic-based quality requirement, EZ Rules 3 analysis generates this recommendation below:

QC Design recommends a 1_3s/2of3_2s/R_4s/3_1s/6_x multirule with 6 control measurements (running 6 controls once or reading three controls twice). This multirule procedure still doesn't have enough error detection power - the OPSpecs chart below is only to 50%AQA. This means that the average run length will probably be two runs; an error is probably not going to be detected in the run that it occurred.

Contrast this with the EZ Rules 3 recommendation using the CLIA quality requirement:

The QC procedure here is more practical: a 1_2.5s control rule with 3 control measurements.

PLT

PLT is a case where the quality requirements are quite different. The CLIA quality requirement, at 25%, is nearly twice the value of the Biologic-based quality requirement of 13.4%. It's no surprise then that the Sigma metrics are also starkly different. For the CLIA requirement, the Sigma is 6.84. For the Biologic-based requirement, the Sigma is 3.07.

Given the differences, the EZ Rules 3 recommendations should not be a surprise, either:

EZ Rules 3 again recommends a 1_3s/2of3_2s/R_4s/3_1s/6_x multirule with 6 control measurements (running 6 controls once or reading three controls twice). But again the OPSpecs chart shown above is only for 50%AQA.

Using the much wider CLIA rule, the QC recommendation is, again, easier to implement:

Summary of QC procedure Recommendations

Using EZ Rules 3, we can determine the different QC Recommendations based on the quality requirements. These are displayed below:

Analyte	Biologic-Based QC procedure recommendation	CLIA-Based QC procedure recommendation
RBC	13s/2of32s/R4s/31s with N=3	13s with N=3
HGB	13s/2of32s/R4s/31s with N=6	13.5s with N=3
HCT	13s/2of32s/R4s/31s /6x with N=6	12.5s with N=3
PLT	13s/2of32s/R4s/31s /6x with N=6	13.5s with N=3
WBC	13.5s with N=3	13.5s with N=3
%Neutrophil	13.5s with N=3	---
%Lymphocyte	13s/2of32s/R4s/31s /6x with N=6	---
%Monocyte	13s/2of32s/R4s/31s /6x with N=6	---
%Eosinophile	13s/2of32s/R4s/31s with N=3	---
%Basophile	13.5s with N=3	---

There is quite a difference between the rules needed for the Biologic-Based requirements versus what is required to achieve CLIA compliance. For a US laboratory, I expect it would be very hard to choose the tighter requirements. Any laboratory making this choice would need to weigh the clinical benefits of controlling those tests to a tighter degree against the costs of quality control.

With the differential parameters, there are no CLIA requirements at all, so the Biologic-Based requirements are all we have to go on. Some parameters will be very easy to control, requiring 3s or 3.5s control limits, while others will require a sort of "Maximum QC."

Yet one more choice to make is how to reconcile the different QC procedures. Given that this is a multitest instrument, at least one decision has to be made that applies to all the tests: how many control measurements will be made with each run? It's probable that the laboratory is going to run just three controls and measure each one once. So all the QC procedures will need to synchronize on that N.

In other words, EZ Rules and QC Design give the laboratory a set of ideal QC procedure recommendations. Now the laboratory must choose a practical set of QC procedures to implement. In some cases, this may mean "under-controlling" a method. For those analytes and parameters, it will be very important to ascertain the extent of the under-control, so the laboratory is aware of how much risk is involved with those methods.

Evaluating the Error Detection with Different QC Procedures

EZ Rules 3 allows you to examine the error detection and false rejection characteristics of any common QC procedure in the context of a specific method, analyte, or parameter. You plug in the method characteristics, use the Manual rule selection, and then select the Sigma metrics chart (previous screens were looking at OPSpecs charts, which do not detail the error detection characteristics of QC procedures).

For example, if you were using a 1_3.5s control rule with N=3 - a QC procedure that is fine for WBC - but you wanted to know how well that procedure would work for the Lympocyte parameter, you would see the screen below:

On the right side of the graph, in the key, you see two columns of probability numbers associated with each control rule. The first number is the probability of false rejection. The second number is the probability of error detection. So, in this case, a 1_3.5s control rule with N=3 will have basically no false rejection, but it will also have basically no ability to detect when a medically important error occurs in the lymphocyte method.

If a more robust QC procedure is used, the numbers improve, as shown below:

While the approximately 26% error detection is far from ideal, it's a lot bigger than 1%.

Here's where Risk Management techniques make a useful complement to the QC Design process. Using a selected QC procedure, we have attempted to mitigate the risk of a medically important error for a method. But the mitigation does not have the ideal detection power, so there is some "residual risk" left over. How the laboratory - and the clinician - handle and share that residual risk is the next step in the Total Testing Process.

Conclusion