RADx in review: Were we able to speed up development of new tech?
Volume 7, Issue 19 | May 17, 2023
In This Issue
COVID check-in: Mortality low, Omicron variants simmering
Using tears to test brain health
Off the record, testing is key to prepping for the next pandemic
Artificial intelligence/deep learning is set to revolutionize diagnostics
New and Noteworthy
As we head towards the Memorial Day weekend and the unofficial start of summer, Sensitive and Specific will be taking a two-week break. We’ll be back in your inboxes and on Substack at the start of June.
The age of Omicron continues, but mortality remains low
Mortality is currently running at the lowest levels of the pandemic - but that still amounts to over 300 deaths/week. In the US, Omicron took over from Delta in mid-December 2021, just in time to exploit 2021 Christmas and 2022 New Year gatherings. It drove deaths to a peak of 26,000 deaths/week, the highest of the pandemic (compare to the first spring 2020 peak - 17,000 deaths per week - which felt astronomical at the time). Since then Omicron has generated a parade of more or less minor sub-variants and recombinants, each with some advantage (in transmission and/or immune resistance), but none with as significant impact on mortality in our now largely COVID-exposed population.
Now that the WHO has declared the pandemic phase over, we have our fingers crossed that this recent low mortality pattern continues with the current front-runner, XBB.1.16. A recent detailed examination of this recombinant predicted that it would take over from XBB.1.5 worldwide because of its 17% higher transmission rate, immune evasion of all antibodies except sotrovimab (Xevudy), and breakthrough potential in previously infected or vaccinated individuals.
What can your tears tell you about your brain?
This week NEJM published the results of a study that used tears to detect the misfolded protein that causes variant Creutzfeld-Jakob disease (the human form of mad cow disease). The eye is an outgrowth of brain tissue during development, so pathological proteins in tears are just one step removed from presence in the retina - and two steps removed from the brain itself.
This particular test relies on the same principle as the Parkinson's disease spinal-tap/CSF test (αSyn-SAA) that we covered a few weeks ago. Basically, if a patient sample contains misfolded protein, it acts as a seed to amplify the misfolding to a level at which it can be easily detected. The protocols of the seeding assays differ, but this tear-fluid proof of concept suggests that easier and simpler tests may eventually be available for other far more common protein-misfolding disorders such as Parkinson’s, Alzheimer’s, etc.
Commentary: Just a few years ago, brain tissue from autopsy was the only way to confirm a diagnosis of any of these diseases. Since then, imaging agents for plaque, tangle, and atrophy detection have become available, and several CSF and blood/serum protein tests are awaiting approval. Will tears be next? Of course, the challenges are not trivial - more convenient sample types tend to have increased variability and lower concentrations - but momentum is building fast.
What policymakers REALLY think the US should do to prep for the next pandemic
An op-ed in the Washington Post on what the US should do to prepare for the next pandemic caught our eye this week. According to the author, an Edward R. Murrow press fellow at the Council on Foreign Relations, its seven recommendations come from her many off-the-record conversations with US policymakers. They are the steps for which, for political reasons, these folks can’t publicly advocate, but which they truly think are important.
And what’s #1 on the list? “Put tests everywhere.” And not only that, keep them there: Instead of closing the public testing centers that appeared during the pandemic, she writes, the centers should be repurposed and expanded to diagnose flu and sexually transmitted infections: “This would curb the spread of these illnesses while strengthening the testing system for the next pandemic.”
Commentary: Preach it, sister.
Food for Thought
Lessons Learned, RADx edition: Were we able to speed up development of new tech?
As the official emergency comes to an end, we are all looking back to see what worked and what did not so we can react faster and more effectively next time. Science Advances recently published a review of how well novel technologies fared under NIH’s RADx program. Bottom line: Accelerating modifications to existing products and technologies was successful, but the initial goal of reducing the traditional multi-year timeline of novel assay technologies to less than 12 months proved impossible. (Contrast that with the $12.4 billion Operation Warp Speed, which generated vaccines within that period.)
This detailed analysis describes several key issues in more depth (for a broader overview see this prior paper and a two-part series from 360Dx). Programs with the most novel technologies (e.g., breath) were at a more theoretical than practical stage. While they had high potential, they simply had too many development challenges before production scale-up could be initiated. A push to bring molecular testing methods to the home was hampered by reliability, scalability, and higher cost per one-time use than the now-familiar capillary flow antigen tests.
Of the four RADx programs, two were innovation-focused: RADx Tech ($500 million), which funded 22 companies with awards from $1M to $40M each, and RADx-rad ($200M), largely awarded to academia for more “rad”ical approaches. All in all, 824 “Tech” and “rad” applications were received, 179 of which went into the “shark-tank” stage, resulting in 97 awards, 52 EUAs (13 at-home; 23 POC; and 16 laboratory). Overall, new technologies contributed 1.7 of the 7 billion tests distributed by RADx program participants.
Commentary: Whether the RADx model is a good one for innovation is up in the air, described by one participant as an “ant dancing with an elephant.” Nevertheless, there is no doubt that moving infectious-disease diagnostics to the home has dramatically opened new opportunities for the industry and created a better way to catch infections before they are transmitted to others.
Artificial intelligence/deep learning is set to revolutionize diagnostics
This week Nature Medicine published an application of deep learning to pancreatic cancer, with the promise of substantially earlier detection. Early evidence of a revolution to come.
It's a revolution because AI’s success requires long-term, integrated, multi-dimensional patient tracking - very far from the current diagnostic paradigm of symptoms, clinical encounter, differential diagnosis, and specific diagnostic orders to find root causes. AI’s superpower is its ability to find high-risk patterns within noisy patient data before they would be apparent to a clinician (or to the patients themselves). The highest benefit is in diseases that are typically only diagnosed once it is too late to intervene successfully.
Pancreatic cancer is the poster child of this tragedy - few cases per year but high mortality, making it the third-highest cause of cancer deaths. The AI system in question was trained on very large data sets (2 million from the VA in the US and 6 million Danish patients, all over many years). The extract below shows the very broad nature of the top 10 factors that enable AI to diagnose pancreatic cancer at various earlier-than-historical time periods.
Commentary: The fragmentary and transactional health system in the US has been ill-suited to this degree of integration. It is no coincidence that this research was performed on the VA here and on a European national health system. Serving US patients effectively is going to require substantial adjustment.
Quick Hits
Wastewater testing showed that hospital-based reporting of mpox underestimated the extent of a 2022 outbreak in Poznan, Poland, both in numbers and in duration. This surveillance method was able to capture the presence of infected people who may not have been willing to go to the hospital for testing.