8

Do medical laboratories determine accuracies for their blood tests that have numerical results (as opposed to specificities and sensitivities for tests with binary results)? If so, where can I find these accuracies? If not, why not?

zylstra
  • 191
  • 1
  • 2
  • 1
    EXCELLENT question. There are scattered population-wide data for these types of stats (e.g. on average the sensitivity is X) but as laboratory methods, machines, reagents, etc vary widely, there could theoretically be significant inter-facility variability. I have called a few labs and even talked with their managers to try to find out their stats in order to determine false positive rates for actual patient results without much success. This data ideally should be available to anyone, especially providers and patients who rely on the test results! – DoctorWhom Nov 01 '18 at 05:57
  • That I put as a comment because it is not an answer - I would love to learn how to find this information myself. – DoctorWhom Nov 01 '18 at 05:59
  • 1
    I have a friend that works in the micro lab in a VA hospital, and I know they do calibrations on all of their test machines and assays every day, and have specific cleaning mandates, but I don't know how they determine that X test value is accurate. – JohnP Nov 01 '18 at 14:31

1 Answers1

6

NOTE: I cannot give you any references in english, since the documentation I have to backup these claims is in Portuguese.

Do medical laboratories determine accuracies for their blood tests that have numerical results?

In order for a lab to be certified (in my country at least, but I reckon this is probably true for other countries), periodic calibrations and validations of the lab equipment and methodology are required.

Different types of tests require different methods of validation but they usually fall into two (very) broad categories:

  1. Validation against a known substance/quantity;

    Example: To check if the chlorine reader is working properly, you test the machine against a sealed vial containing a known concentration of chlorine.

  2. Blind checks against the results of another certified labs.

    Example: Blood cell count (automatic or manual (by a human, on a microscope)). A blood sample is divided evenly into two vials and one is sent to another certified lab. Usually several measures of the same sub sample are performed, in order to reduce sampling error.

For a lab to reach the maximum certification level, both types of tests are performed several times for each lab test.

If so, where can I find these accuracies? If not, why not?

Regarding the "accuracy", it depends on many factors (method of collecting samples, method of transportation of sample, the time bewteen collection and test run, etc...).

But for automatic tests, but the most important is the "machine" used to test it. The machine's manual usually gives you the value of the error or the confidence interval.

So, in order to get a pretty good estimate of the "accuracy" of the lab test, you only need to know the "machine" used and you can google search for the manual or brochure to find that out. example. Or, failing that, call the manufacturer and ask for that value.

If your country or state has a certification level, a quicker way might be through the certification level of the lab. Each level has a "tolerance" value, that is, the maximum acceptable deviation from the expected value.

Also, for tests that rely heavily on humans, the "tolerance" value is the only way you can estimate the "accuracy" of the test.

sources:

Manual for Clinical Lab standards (in portuguese)

Tivie
  • 211
  • 2
  • 6
  • Welcome to MedicalSciencesSE, Tivie! While we certainly prefer English, a reference in another language is better than no reference! We are not all monolingual, someone might help with translations etc. – LаngLаngС Nov 02 '18 at 19:43
  • There's always https://translate.google.com. It will be a literal translation, but that's usually enough to communicate the necessary information, especially in technical writings. – Carey Gregory Nov 02 '18 at 20:55
  • Would photo of a panflet/documentation be helpful? – Tivie Nov 03 '18 at 02:07
  • +1 for example manual, but -1 for no accuracy spec. ;) – zylstra Nov 03 '18 at 06:18
  • A photo is not ideal. It would be if accompanied by an OCRed scan and short quotes from that here. Photos cannot be searched, screen readers are excluded etc. –– The machine translation is indeed an option, but one you shoud at least double-check once on the fly. Sometimes the results are quite weird. – LаngLаngС Nov 03 '18 at 15:16
  • 1
    @zylstra It has the fluorescence and scatter sensitivity. But you're right, I will try to find a better example :P – Tivie Nov 04 '18 at 00:18
  • I updated the answer to include the Government Manual for Portuguese Lab Certification. – Tivie Nov 04 '18 at 00:20