What’s the best way to use AI-enabled diagnostics?
Volume 11, Issue 10 | March 6, 2025
ALSO IN THIS ISSUE
RNA sequencing could improve Dx of heritable diseases
Bird Flu Update: New USDA strategy focuses only on poultry
AI beats techs at long-term ECG analysis
Measles Q&A
RNA sequencing may improve diagnosis of heritable disorders
Today many DNA-based tools can predict and diagnose heritable disorders. Still, up to 15% of family patterns of disease remain unexplained by known DNA variants. A paper published in AJHG provides a possible solution: a clinically validated RNA sequencing (RNAseq) test that can supplement whole-genome/exome sequencing.
RNAseq is a step forward because it finds DNA that’s been either over- or under-expressed. The study claims that the technique can improve diagnosis by 7 to 36%.
COMMENTARY: RNAseq is challenging. Different cells in different tissues in different people express different mRNA profiles, so the amounts and types of proteins you find in one tissue may not be the same as they are in another. But this test uses only two sample types - skin and blood - and assumes they are representative of the entire body.
Hereditary neurological disease in childhood is an example of the potential problem. (Such diseases make up the majority of hereditary diseases that lack diagnoses.) We don’t know how many of these diseases will be detectable in skin or blood samples - they may only appear in brain cells. If that’s the case, RNAseq won’t help.
Bird Flu Update: New USDA strategy focuses only on poultry
Administration having trouble rehiring USDA bird flu workers
Virus hits NJ cats; raw pet food contaminated in WA and OR
Proof-of-concept sensor can detect airborne H5N1
Last week, the USDA announced a new five-pronged strategy for battling H5N1. The plan focuses exclusively on poultry and prioritizes increasing egg supply within the US. It includes:
$500 million in biosecurity support for poultry farms
Up to $400 million to financially support affected poultry farmers and speed farm repopulation
Removal of “unnecessary regulatory burdens on the chicken and egg industry”
Up to $100 million “for potential new-generation vaccines, therapeutics, and other innovative solutions to minimize depopulation of egg-laying chickens”
Exploring the possibility of importing more eggs
The administration is having trouble rehiring USDA employees working on bird flu, Politico reports. The employees were fired as part of the wholesale removal of probationary employees at HHS. As the website noted:
Supervisors at USDA have been told they need to write a justification for every bird flu employee called back, said one person familiar with the process. Some of those who were rehired this week still don’t have their laptops. And it’s not clear all of the ousted workers have received return offers, or that they plan to accept them, potentially leaving key offices understaffed on the response to the outbreak.
Domestic cats in New Jersey, Washington, Oregon, and Colorado have tested positive for bird flu, and many have died or been euthanized. The New Jersey cats have no known exposure to H5N1, so are assumed to have been in contact with wild birds carrying the disease. The Washington and Oregon cats were likely infected by eating raw food that contained the virus. The food, which was produced by Wild Coast Raw, has since been recalled.
Researchers have developed a proof-of-concept sensor that can detect the presence of airborne avian flu and estimate how much of it is in the air. They told St. Louis Public Radio that the device could be placed near air vents in poultry barns, alerting farmers to the presence of the virus and potentially allowing them to isolate sick birds.
AI beats techs when it comes to long-term ECG analysis
Thanks to wearable technology, getting a long-term electrocardiogram (ECG) is easier than it’s ever been. But analyzing all that data takes time, not to mention expertise. (We’re talking about looking through two weeks’ worth of a patient’s heartbeats. It’s not quick.)
There aren’t enough skilled technicians to handle all that work. So this is one place where having an AI assistant who could do the first pass - not with people, but for them - would actually be helpful. And recent research in Nature Medicine shows that it could work - within limits.
The paper gave human technicians and an AI model a set of long-term ECGs to analyze. When cardiologists then looked at the analyses, they found that the AI had 14 times fewer missed diagnoses of severe arrhythmias (the AI missed them in 0.3% of patients, while technicians missed them in 4.4%). The AI model was also able to rule out severe arrhythmia with 99.9% confidence in a 14-day ECG recording. The one area where the AI slipped up was in false positives - it had 12 false positives per 1,000 recording days, compared with 5 for the technicians.
COMMENTARY: This is a perfect use of AI.
A huge amount of data on what’s normal / abnormal exists.
A lot of data needs to be analyzed.
Rapid review of that data is crucial.
And if there had to be a problem with AI, let it be false positives. A small increase in false positives increases followup and cost, true. But that’s still less cost than would have been incurred if all patient data was analyzed by humans. False negatives cost lives.
COMMENTARY: What’s the best way for physicians to use AI-enabled diagnostics?
As AI-enabled diagnostic tools proliferate, the question of how best to use them becomes more and more pressing. The conventional wisdom thus far has been that AI + clinician generally outperforms either AI or clinician alone when it comes to diagnosis.
But a recent New York Times op-ed by Eric Topol and Pranav Rajpurkar (and a subsequent Ground Truths blog post by Topol) pointed out that research evidence on the topic is all over the map. Perhaps the most surprising result comes from one smaller study, in which MDs alone were about as accurate (74%) as MDs with AI support (76%), but that AI alone was 16% better than either.
However, because the field has seen such explosive growth over the last six months, few published studies evaluate the most recent AI models. And those models are considerably better at their jobs than earlier tools.
Topol argues that the answer to the question will ultimately be “it depends”: “Rather than assuming that combining both approaches always yields better results, we should carefully consider which tasks are better suited for AI, which for humans, and which truly benefit from collaboration.”
Measles: Vax and Dx Questions and Answers
We have received many questions about the current measles outbreaks in Texas and elsewhere in the US. Here are two of the more common ones, with our responses.
Q: I believe I received the measles vax as a kid. Do I need to be vaccinated again?
A: Obviously, ask your doctor. But if you want to confirm your level of immunity, have your measles titer checked. That test tells you whether your body still has enough antibodies to keep you protected from infection.
Q: Is there a test for measles?
A: Yes, but it may be hard to find. Testing won’t help contain a measles outbreak, because it spreads too fast - about 5x faster than COVID. The disease is usually diagnosed based on symptoms + / - a history of exposure within the past 10 days or so. Tests are used to confirm.
Both antibody tests (presence of IgM antibodies indicates current infection) and molecular tests (rtPCR detects actual measles virus) for the disease are available. They’re all lab-based - no FDA-approved rapid tests for home use exist.
And really, why would there be? There was no demand - thanks to vaccination, measles was declared eradicated in the US 25 years ago. Sadly, depending on what happens with the current outbreaks, it’s possible that the nation will lose that status. Cross your fingers - and if you aren’t protected, get vaccinated now.