To Screen or Not to Screen
Volume 8, Issue 20 | November 9, 2023
In This Issue
First at-home antigen test for COVID fully approved
What FDA’s updated list of AI/ML-enabled devices tells us
Intraoperative brain cancer diagnosis
Using bacteria’s evil superpower for good
New and Noteworthy
First at-home antigen test for COVID gets full FDA approval
The FDA has cleared the first over-the-counter / home use COVID antigen test: ACON Laboratories’ FlowFlex COVID-19 Antigen Home Test. It is the second home COVID test to get full approval - the first was the Cue COVID-19 Molecular Test, which is instrument-based. The ACON clearance includes testing for use in children under 18, which the Cue test does not.
As we’re sure you remember (wink, wink), there’s still a Section 564 Public Health Emergency in place (that’s different from the Section 319 PHE, which ended on May 11), so while FDA is clearing tests for full marketing approval, they are also still granting EUAs to COVID tests as well. The agency has pledged that once the Section 564 PHE is lifted, they will give companies / labs 18 months to file submissions for full marketing clearance for tests that are currently under EUA.
What FDA’s updated list of AI/ML-enabled devices tells us about the field
The FDA updated its list of cleared AI / ML- enabled devices, adding 171 new products, most of which received approval between August 2022 and July 2023. In total, 693 devices have been cleared. Here are the major takeaways from their announcements, based on our own analysis and that of MedTech Dive. (Note that while the FDA calls these devices, what we’re talking about here are software algorithms / programs / systems that are used in conjunction with imaging and diagnostic tools.)
Radiology dominates the list, with 79% of all recent approvals.
None of the devices on the list use generative AI (stuff like ChatGPT).
The level of complexity varies, in the words of the FDA, “from shallow (less than two hidden layers) models to more complex models (deep learning models).”
Virtually all of the cleared devices went through the 510k pathway. Only 5% were de novo.
Many companies are involved in this field. Of the 171 added, GE had the most approvals (10), with Siemens just behind (9).
Commentary: Watch and wait and get ready. This field will grow substantially in the near future. We believe that the FDA has been relatively responsive and clear that any new algorithm / updated software needs to go through their process. We expect that radiology will remain the largest arena for many years, but the fastest growth will come from other areas especially in treatment and report generation, ultimately using generative AI.
To screen or not to screen, that is the question
Two weeks ago we discussed the utility of pan-cancer screening using the PATHFINDER trial data. The biggest downsides of the screening discussed in that paper are low sensitivity (28.9% - nearly three quarters of cancers missed) combined with low positive predictive value (38% - nearly two thirds of all positives false).
This week an Australian study modeled screening for five heritable diseases (breast, ovarian, colorectal, and endometrial cancers plus heart disease). It concluded that screening would be worthwhile at a society level, costing AU$23,926 (US$15,345) per Quality Adjusted Life Year (QALY) gained.
Commentary: Screening for heritable disease is far, far simpler than screening for spontaneously arising (somatic) cancers. Every single cell in a given patient carries heritable markers, but only cancerous cells produce the active cancer markers detected in liquid biopsies. Because of this, the study assumed 100% sensitivity and specificity, which would be well out of reach in a pan-cancer test.
Food for Thought
Sequencing and AI add accuracy to intraoperative brain cancer diagnosis
Surgery to remove a cancerous tumor in the brain involves an incredibly delicate balancing act. Removing tissue around the tumor may make recurrence less likely, but that means more loss of function - and may not be necessary. A recent paper in Nature presented a new way to test to determine the type and subtype of a given tumor while the surgery is happening - and thus determine how aggressive the surgery should be.
Currently, docs categorize the tumor by looking at cells under a microscope, but that doesn’t always work well. The technique presented in the paper involves sequencing the tumor’s DNA and then using AI to analyze certain elements of the results. (For those who know: It involves nanopore sequencing followed by analysis of a sparse methylation profile.)
The sample sizes in the paper are relatively small, but the results are encouraging. During testing with frozen tissue from prior surgeries, the system identified the type and subtype of tumor correctly 45 out of 50 times (90%); it wasn’t able to make a diagnosis in the other five cases. And it got there in 40 minutes flat, including sequencing time. When the system was used during actual surgeries, its record was 18 out of 25 (72%, again declining to diagnose the rest), in under 90 minutes. The paper’s authors hope to see their product used alongside the traditional microscopic analysis of tumor cells. Using the two techniques together will, they predict, lead to even more accurate intraoperative diagnoses - and better outcomes for patients.
Using bacteria’s evil superpower for good: Detecting cancer
Bacteria are very good at sharing DNA sequences between each other (horizontal gene transfer). Usually that’s a very bad thing, since it enables rapid sharing of antibiotic resistance genes.
A recent Science paper (paywall) describes how a common bacterium was engineered to use this superpower for good. The CRISPR-modified bacteria was designed to detect KRAS-driven colorectal cancer from stool but can in theory be applied to any specific DNA sequence of interest to which it is exposed.
Commentary: This is an interesting proof of principle that illustrates how modern genomic tools can be combined with traditional microbiology to develop novel diagnostic techniques.
Quick Hits
The CDC is expanding the testing done in its Traveler-Based Genomic Surveillance program, which currently tests for COVID in airplanes, wastewater, and samples from anonymous volunteers at seven international airports. At JFK, Dulles, San Francisco, and Boston international airports, the new pilot program will test for over 30 different viruses that cause respiratory disease including flu and RSV.
More than half of the US’s attorneys general have cosigned a letter asking the FDA to “act with urgency to address the inaccuracy of pulse oximetry when used on people with darker toned skin.” The letter comes at the one-year anniversary of the FDA’s meeting at which a group of outside experts called for the agency to require more rigorous testing of the devices.