Posted by Sten Westgard, MS
A recent question came in from a technical support consultant for a major diagnostic manufacturer:
"It is for Free T4 analyte. Customer [has] establihed a QC range after 20 QC runs. Mean and SD were derived from 20 runs and %CV achieved from 20 runs is 1.2%. Allowable interassay precision criteria according to CLIA is 6%.
"Now when customer [applies] Mean and SD according to the established range many...times they face QC rules violations of 41s and 10x.
"So can you please suggest, when the SD is too narrow is it necessary to apply these two rules to immunoassays?"
What's your guess? I'll take a stab at an answer, after the jump.
As you might imagine, there are a lot of additional questions that need to be answered before we can answer the question about what rules to use. For instance, what is the quality requirement? What isthe bias? What is the most important decision level.
Working with this consultant and customer, we were able to obtain the following data:
So you can see we've got two levels and imprecision and bias at each of those levels. We're not going to determine which level is most important yet.The next question is the allowable total error. What's the quality required for FT4? According to the biological variation database ("Ricos Goals") desirable specifications, FT4 needs to be good within 8.74%. However, according to the Spanish Minimum specifications, FT4 only need to be good within 16% - this at least is what the vast majority of labs should be able to achieve.So we can do Sigma-metric calculations for both of these quality goals:
Control 1 | Control 2 | |
FT4 - u1 | FT4 - U2 | |
1.26 | 3.88 | |
1.26 | 4 | |
1.24 | 4 | |
1.27 | 3.98 | |
1.28 | 3.95 | |
1.3 | 4.09 | |
1.27 | 3.93 | |
1.27 | 3.89 | |
1.25 | 4.05 | |
1.29 | 3.94 | |
1.28 | 4.09 | |
1.26 | 4 | |
1.29 | 4.08 | |
1.29 | 4.01 | |
1.26 | 3.95 | |
1.25 | 4.07 | |
1.29 | 4.12 | |
1.28 | 4.12 | |
1.27 | 4.12 | |
1.29 | 3.57 | |
Mean | 1.273 | 3.992 |
sd | 0.017 | 0.125 |
cv | 1.297 | 3.134 |
Target Value | 1.24 | 4.14 |
Target SD | 0.06 | 0.2 |
bias | 0.032 | -0.148 |
bias% | 2.55 |
3.71 |
So you can see we have imprecision and bias estimates now in percentage for two levels of this test.
Now we need to determine how good the test should be (what's the allowable error?).
The biologic variation database ("Ricos Goals") sets a desirable allowable error at 8.74%. The Spanish minimum specifications for allowable error set that allowable error at 16%; this is a goal that the vast majority of laboratories should be able to achieve.
We have two sets of quality specifications, so we'll do two sets of Sigma-metric calculations:
TEa Ricos | 8.74 | 8.74 |
Ricos Sigma | 4.77 | 1.61 |
Sp. Min. | 16 | 16 |
Sp. Min. sigma | 10.4 | 3.9 |
Even now, we still have more questions than answers. Which goal is appropriate? Which level is most important? How do we design our QC?
Let's use the "Westgard Sigma Rules" to help us out.
You can see that once you drop below 5 Sigma, you need to start using the 4:1s rule, and below 4s, you need to use the 8:x (or 10:x) rule.
So if we're using the Ricos goal for FT4, we need to use the 4:1s on one level and 4:1s/8:x on the other level. If we're using the Spanish minimum goals for FT4, on the one level we're above Six sigma and can use wide limits and none of the other "Westgard Rules". But the other level is still 3.9 Sigma, which means we need the 4:1s rule.
So we don't have one answer yet. We have essentially "It depends." The customer needs to decide which goal is appropriate and which level is critical. If they answer those questions, they'll know whether or not they need those extra trend rules.
Comments