Posted by Sten Westgard, MS
Earlier we discussed error rate issues at the Point-of-Care. But we didn't want to leave the "regular" laboratory out of the fun, so here's a study of error rates that came out in 2010:
Evaluation of errors in a clinical laboratory: a one-year experience, Goswami B, Singh B, Chawla R, Mallika V, CCLM 2010;48(1):63-66.
Here is the description of the collection of data on the errors:
"From July 2008 to June 2009, a total of 67,438 routine venous blood specimens were received in the laboratory. Errors were detected in 954 samples, with a total error rate of 1.4%."
Below is a table with the error rates, plus we've added a Sigma-metric conversion, using the short-term Sigma scale table.
Error type | # defects | % defects/ total tests |
Sigma-metric |
Overall Pre-analytical | 736 | 1.1% | 3.8 |
Hemolyzed sample | 508 | 0.7% | 4 |
Insufficient sample | 72 | 0.17% | 4.5 |
Lipemic sample | 7 | 0.01% | 5.3 |
Incorrect identification | 44 | 0.06% | 4.8 |
Empty tube | 10 | 0.01% | 5.2 |
Requisition slip without sample |
16 | 0.23% | 5 |
Illegible hand writing | 69 | 0.1% | 4.6 |
Tube broken in centrifuge |
6 | 0.009% | 5.3 |
Physician's request order missed |
4 | 0.006% | 5.4 |
Total Analytical Error Rate* | 75 | 0.11% | 4.6 |
Non-conformity with QC | 6 | 0.009% | 5.3 |
Random error | 15 | 0.02% | 5.1 |
Calibration drift | 10 | 0.014% | 5.2 |
Reagent contamination | 8 | 0.012% | 5.2 |
Systematic error: probe lamp, blocked tubing |
36 | 0.05% | 4.8 |
Total Postanalytical Error Rate | 143 | 0.21% | 4.4 |
Transcription errors | 112 | 0.17% | 4.5 |
Prlonged TAT | 31 | 0.05%7% | 4.9 |
[* We added an asterix here because, as with most error studies, we do not believe the analytical error rate was correctly measured. In broad studies of this nature, it is likely that the same control limits were used to for all tests, which might either be too wide or too narrow (because the rate is so low, we suspect too wide). The proper measurement of Sigma-metrics for the analytical performance would be to set a quality requirement and measure imprecision and bias, then calculate the Sigma-metric (long term Six Sigma scale).
Another issue about the analytical error rate measurement: a single error in calibration is a one-to-many error. That is, one calibration drift could affect entire runs, not just a single instance. Hemolyzed specimens are a situation where one error affects one specimen. Thus, the comparison of errors here may not accurately reflect the impact of the different types of errors.]
What's even more interesting is to compare these error rates with those of the recent April 2011 study of error rates tracked over 5 years for a group of Spanish laboratories.
There are some significant differences:
- The median error rate of controls exceeding acceptance limits was 3.4% or 3.4 Sigma, vs. the Indian laboratory having a (combined) control limit violation rate of 0.11% or 4.4 Sigma [again, possibly evidence that the counting of control violations is not standardized across labs]
- The median error rate of (internal) delayed TAT for the Spanish study was 0.5% or 4.1 Sigma, vs. the Indian laboratory having a TAT rate of 3.2% and a 4.9 Sigma.
- The median error rate of insufficient samples was 0.3% or 4.3 Sigma, vs. the Indian laboratory with a rate of 0.17% or 4.5 Sigma
- The (combined) median incorrect identification rate for the Spanish labs was 3.51% or 3.4 Sigma, vs the Indian laboratory error rate of 0.06% or 4.8 Sigma
There was one interesting similarity, however:
- Hemolyzed specimen rate was similar for both institutions, about 0.6% or 4.1 Sigma for the Spanish labs, and 0.7% or 4.0 Sigma for the Indian laboratory
The result of these studies shows that error rates can vary widely. sometimes by a factor of 58. Not all of this variation in error rates is from the differences in measuring and definition of errors. There's probably a difference in some practices as well. While analytical errors exhibit a closer degree of uniformity (the box is designed to operate the same way in all labs), every laboratory has its own unique pre-analytical and post-analytical context.
Thus, it may not be useful to generalize the occurrence rates of pre-analytical errors, or declare that one particular error type is more frequent than other types.
The morale of the story: your error rates will vary.
Comments