Creating laboratory developed tests (LDTs) is always a challenge, especially in molecular diagnostics. As advances in genomics fuel precision medicine, today’s clinicians can encounter an almost infinite array of scenarios with their patients. That means that LDTs — and the lab professionals who design them — must be able to meet the challenge.
CLN spoke to Hunter Best, PhD, FACMG, about how to decide which tests to develop, the importance of having a range of patient samples for validation, and how bioinformatics is pushing the field forward. Best is the vice president of molecular operations and a medical director of molecular genetics and genomics at ARUP Laboratories and a professor of clinical pathology at the University of Utah School of Medicine.
This is part two of a two-part series on implementing LDTs. You can read part one here.
From a business and quality perspective, what distinguishes labs that are successful in launching LDTs versus those that struggle?
Laboratories that excel in developing LDTs — especially reference labs like ARUP — succeed because people trust the validity of the tests. We try to validate our tests on every clinical scenario we’re likely to encounter, so we’ve earned that trust.
Smaller laboratories may have samples, but they don’t always have the same resources that reference laboratories do. They may come across clinical cases they weren’t prepared to see. That doesn’t mean they don’t know how to handle them; it just means their validations are not always as all-encompassing as the ones that happen at larger labs. That can be a challenge.
How do you define the intended clinical use of a new LDT before development begins?
With a lot of the LDTs that we have developed, clinical targets were already identified, whole genome sequencing notwithstanding. In other words, we knew the clinical populations for which testing should be used, and we designed an assay to interrogate those specific regions of the genome, whether they exist within cancer cells or patients’ inherited genetic material.
However, what we can’t control as a reference laboratory is how clinicians end up ordering LDTs. We have recommendations on our website explaining the best use case for each test, but providers are the ones ultimately deciding how LDTs are used. ARUP is an academic institution, so we try to educate clinicians to make sure they are ordering appropriately.
What are the biggest technical challenges you encounter during assay configuration and optimization?
One big challenge is having the right samples. Gathering samples and ensuring we’re covering all relevant clinical situations is what typically takes the longest time. It’s an extensive process. We have tests for which we’ve been actively accumulating samples for more than a year before we’re even ready to start the validation process. Also, we don’t usually use the same samples for development and validation — which means we need to have sample sets available for both.
For a test to be successful, it must address many or all of the scenarios clinicians are likely to see. Sometimes labs will rush to onboard a test and skip some relevant scenarios in the process. I think that’s what spurred the Food and Drug Administration (FDA) to become concerned about LDTs in the first place.
I don’t necessarily disagree with the essence of what the FDA was trying to do, but they were painting with a really broad brush. Not all LDTs are equivalent and not all laboratories are equivalent. The way start-ups do things is very different from how laboratories that have been established for decades do them.
How do you design analytical validation studies for complex molecular LDTs?
The short answer is very carefully. We look at all the genomic regions that surround the targets we’re interested in, and we design around them. If we can’t do that, we don’t include those targets on an assay.
A lot of factors go into the design. For example, are we looking at germline sequences or regions that exist in only a subset of cells, like in an oncology specimen? The answer to that question helps determine how we go about designing. Since the limit of detection is a lot lower with cancer specimens, we’d likely want to push the sensitivity. There’s no one-size-fits-all approach.
At ARUP, before we enter evaluation, we do development work where we look at a broad spectrum of samples. After we evaluate the initial design of a test, we’ll often go back and redesign components of it to improve performance.
We may be working on something for a year or longer before validation starts. If samples don’t perform the way we expect them to, we might have to go back to the drawing board. It’s a very long process.
How do you approach clinical verification before launching an LDT?
In our validation, we include samples that have a known diagnosis, whether those samples have been tested using other methods in our own lab or in an external lab.
In some instances, we get samples from our patients, ranging from those who have lifelong inherited conditions to those with sudden-onset diseases like acute myeloid leukemia (AML). Other times, we buy samples from another laboratory that has validated that the patients they came from have a specific mutation we’re interested in.
We’re always looking for samples, especially for target disorders like AML. Testing “normals” is also important. For any validation, you want to make sure you’re including a good spectrum of samples, from positive and negative known clinical cases as well as patients known to not have the clinical disorder.
How has bioinformatics changed your LDT validation and implementation process?
Bioinformatics has become a standard of care in multiple fields, including oncology and inherited disease. We get massive amounts of data from genomic sequencing, and we need a way to handle it all. Bioinformatics tools do that, and they make it easier to target more regions that were difficult to test in the past.
I’m not talking about artificial intelligence (AI) here. Bioinformatics tools on their own have revolutionized our classification of samples. As samples are tested, we have a catalog of variants that we’re seeing in patient samples and an internal database — so we know that we have a sample that has X variant in Y gene.
When we do validation work, we’re able to query that database and figure out if we have samples with variants in the genes of interest. We can pull samples for validation from our database. Additionally, bioinformatics allows us to reprocess our data when there are changes in the scientific literature. In those cases, we may be able to reanalyze and provide a diagnosis that we couldn’t before.
To touch on AI, we have yet to use it in any sort of validation work. We have used it to help generate tables, but ARUP is a bit more cautious than some other labs regarding how we’re using the technology. While there are some significant benefits, there are also concerns. We want to make sure we’re being careful, especially considering that we’re handling patient data.
How do you design an LDT workflow that balances business needs with clinical requirements?
A lot of factors go into the determination. The business aspect involves determining how a test fits into the laboratory workflow as it currently exists. Does using this different method still get us to the same clinical answer? What value does it add? Whenever possible, we try to design tests that fit into our workflow and utilize the platforms we already have.
Also, it’s worth considering whether a test is being designed for a relatively common disorder, a condition that affects one in a million patients, or a disease for which we’re going to see three cases in a year. Often, we will still onboard tests for rare diseases because there are people who need them. But it’s important to think about how to fit them neatly into the workflow so the cost of maintaining them is minimal.
Other times, business efficiency has to take a back seat. Regardless of the implications to our workflow, the clinical question needs to be answered, and clinicians are asking for a specific test. In these cases, we will onboard the test. We operate under a different model than laboratories that have to report to shareholders. The clinical need for a test is the ultimate driver for us.
What advice would you give to a lab preparing to launch its first modern molecular LDT?
Don’t rush. In the long run, taking the time to develop the test correctly will pay dividends. If you cut corners, you end up needing to do corrections anyway — and you don’t want to have to do that after a test has gone live.
Jen A. Miller is a freelance journalist who lives in Audubon, N.J. +Bluesky: @byjenamiller.bsky.social
If you'd like to see additional educational resources from Thermo Fisher Scientific, please visit: https://www.thermofisher.com/LDTeducation
The opinions, statements, and expressions contained in this document are those of the author(s) and are not necessarily those of Thermo Fisher Scientific. The content is provided for informational purposes only and should not be construed as legal, reimbursement, coding, billing, or other professional advice. The information provided in this document does not constitute a guarantee of coverage or payment by any federal, state, or third-party payers. Thermo Fisher is not responsible for the use of this information when billing Medicare, Medicaid, or other third-party payers including billing and coding errors that may result in the denial of coverage, payment, or payment amounts. Users of this information are ultimately responsible for determining applicable coverage, coding, billing, and payment requirements for their own use and/or on behalf of the patients they serve.
Supported By