In response to my note yesterday (see: Cancer Diagnostic Scandal at Duke; More Regulation of Multiplexed LDTs in the Future?), two comments were submitted. The first, from molecular pathologist and informaticist Federico Monzon, serves to clarify some of the confusion that I have introduced into the discussion of IVDMIAs and LDTs:
You need to make a distinction between multiplexed biomarker tests and LDTs. Your article appears to equate the two. Laboratory developed tests (LDTs) are commonplace in diagnostic laboratories (most molecular tests, HLA, flow cytometry and immunohistochemistry assays are LDTs). LDTs can be developed for a single analyte (such as KRAS, Ki67, etc.) or can be multianalyte (a.k.a. IVDMIAs, such as the Duke example). Some multianalyte tests are commercially offered as LDTs (e.g. Oncotype DX) and others have gone through FDA approval (e.g. Pathwork Tissue of Origin). So, not all LDTs are multianalyte tests and not all multianalyte tests are LDTs.
He's right, of course. I conflated in my discussion the type of test performed (single analyte versus multianalyte) with the regulatory approval of the test by the FDA. Multianalyte tests can also be referred to as IVDMIAs. As Federico says above, "most molecular tests, HLA, flow cytometry and immunohistochemistry assays are LDTs (laboratory developed tests) and have not gone through the formal approved process of the FDA [emphasis mine]." Some commercial multianalyte tests such as Pathwork Tissue of Origin have received FDA approval and some, like the very successful Oncotype DX, have not and thus fall into the category of LDTs. Again and quoting Federico, "not all LDTs are multianalyte tests and not all multianalyte tests are LDTs."
In my note, I also made mention of the fact that some of the multianalyte tests coming to marker have clearly not been adequately validated. If they had been so, their accuracy would not have been subsequently discredited and the tests removed from the market. Here is my comment:
There is no question that, ultimately, tests utilizing multiple biomarkers plus an interpretive computer algorithm will be a key component in the practice of diagnostic medicine. However, what we can now observe are medical scientists in prestigious medical schools like Yale and Duke rushing to market with [multianalyte tests] that have been inadequately tested and validated.
Here is the very instructive comment submitted by Steve Binder, a seasoned veteran of the lab diagnostic industry:
I am glad to see that you are recognizing the risks for MIA's [multivariate indexed assays]. At the end of the day, retrospective analysis of data can be notoriously misleading--no MIA should be offered until the algorithm developed on the 'training set' is validated [against] two test sets from completely different institutions. You don't see many people trying to meet that requirement. You do see a lot of people who over--analyze a training set and then think they are ready to roll