ALSO IN THIS ISSUE
Could smartwatches diagnose heart attacks?
Bird Flu Update: OPPR no longer staffed
Can a test of a hair sample rule out autism?
If you’re having a heart attack, could a smartwatch save your life?
Surviving a heart attack is a function of getting help quickly. Unfortunately, that’s rare, because most heart-attack victims are alone when they happen. In the US, the survival rate is just 9.1%, resulting in 437,000 deaths a year.
But what if your smartwatch could tell if you were having a heart attack - and even call an ambulance? That would be great - as long as the watch gets it right. Too many false alarms would put an undue strain on the first-responder system (and one’s nerves).
Last month, Nature published a preprint that looked at whether this idea would work. The study set out to demonstrate that AI-based monitoring of “unanticipated loss of pulse” at the wrist plus evidence of a fall was sufficiently accurate to support public adoption of the technology. To detect loss of pulse, smartwatches use photoplethysmography (sorry, I just had to use that word in a sentence!), which translates to measuring blood volume changes using reflected light. To detect a fall, the watch uses its accelerometer - the same gizmo that counts your steps. False positives were low (99.987% specificity, which corresponds to one false alarm roughly every 22 years), but sensitivity was also low, at 72.5%.
Earlier work published in the journal Resuscitation estimated that this kind of system would save lives in at least 12% of cases. If everyone in the US had a smartwatch, that could mean saving 39,000 lives.
Objective pain measurement : Beyond the “smiley face” scale
Since the introduction of anesthetics (ether and chloroform in the 1850s) and analgesics (Bayer aspirin circa 1900), pain management has been a medical priority. Yet the tools clinicians use to diagnose and evaluate pain remain highly subjective (see the pain scale shown below). They also don’t give clinicians any way to predict which patients’ pain is likely to be resolved in the short term and which is likely to progress to chronic pain. (About 12% of younger adults and about 37% of over-65s report chronic pain over a given three-month period.)
A recent JAMA Neurology paper evaluated whether the way a person’s brain responds to a standardized pain stimulus could be used to more objectively measure their pain level. This study subjected 150 brave individuals to the same pain-creating event without any pain meds. Their responses were then measured noninvasively, using electroencephalography (EEG) and transcranial magnetic stimulation (TMS) to see how specific parts of their brains responded.
The difference between the 40 reporting the highest pain versus the 40 lowest was substantial. The high-pain group reported an average pain level of 9/10 (10 being the worst possible pain) on day three, with reportable pain lasting until day 25. The low-pain group reported an average pain level of 3/10 on day three and had no reportable pain after day 10.
An AI-based model was then trained on the EEG/TMS and pain-reporting data from a randomly selected group of test subjects. When the model looked at the EEG/TMS data from the rest of the subjects, it was reasonably good at predicting which individuals would be in the high-pain group and which in the low-pain one (area under the curve of 0.88).
COMMENTARY: Although the two groups' averages are quite distinct, many individuals fell into a gray zone (see chart above). Nevertheless, having an objective way to measure an individual’s response to pain could be extremely useful. The ultimate goal would be twofold:
Reduce overprescribing of opioid medications to people in the low-pain group, thereby decreasing their risk of addiction
Ensure that people in the high-pain group receive intensive short-term early treatment, thereby decreasing their risk of chronic pain.
We hope that this research marks the beginning of our ability to diagnose and evaluate with data and without bias.
Bird Flu Update: Office of Pandemic Preparedness and Response Policy no longer staffed
H5N1 in aged raw-milk cheese? Possible, but so far not seen
UN’s FAO urges increased bird flu efforts
According to a preprint published last week, avian flu virus could - in theory - be present in raw-milk cheese aged up to 60 days. The study looked at “cheeses made from raw milk spiked with H5N1,” CIDRAP reported. It found that virus was still present at the 60-day mark, as long as the cheese wasn’t too acidic. In practice, however, such contamination may be rare. The FDA has gathered 110 samples of retail aged raw-milk cheese over the last three months, and 96 of the samples that have been tested so far have been negative.
On Monday, the head of the United Nations Food and Agriculture Organization (FAO) urged member nations to increase their efforts to combat bird flu. Among the steps the FAO recommended were enhanced surveillance and reporting, improvement of laboratory capacity, and the development of preparedness plans.
The Office of Pandemic Preparedness and Response Policy (OPPR) has been reduced to a single person, down from a staff of 20. As CNN reported, “The office . . . was orchestrating the country’s response to bird flu and other threats until January 20, including hosting regular interagency meetings to share plans.” Its pages have been removed from the White House website, and according to CNN’s source, the office now exists “in name only.”
Can a test of a hair sample rule out autism?
Autism is a particularly difficult diagnosis to make, at least partly due to high overlap with other behavioral/psychological conditions (e.g., ADHD) and because we basically do not know what causes it (one thing we do know - it isn’t vaccines). Research tells us that 50 - 85% of cases have a genetic basis, although which gene variants are responsible is unclear (see this 2014 paper for details). But what about the other 15 - 50%?
One non-genetic cause could have to do with exposure to metals. Research done in 2017 and 2018 found that reduced uptake of either zinc or manganese and increased exposure to lead or copper was correlated with autism, as long as those experiences happened either in utero or during the first six months of life. In those studies, levels were measured by looking at the amount of those metals in teeth. Further work in 2022 and 2023 showed that another type of sample worked just as well: Hair.
A recent feature in Inside Precision Medicine highlights a currently available test based on this research. It claims to be able to “rule out” autism in children aged one to 36 months - a claim that rests on its high negative predictive value (NPV) of 92.5%. (NPV is the probability that someone with a negative/normal result doesn’t have the condition. The details of how this statistic has been calculated for this test have been presented but not yet published.)
COMMENTARY: This is a laboratory-developed test with limited published data that’s marketed directly to worried parents at a high price (list $2,750, discounted to $1,950 currently). Hair is a super-easy sample to get, so if this test really works, that would be terrific. But the jury is very definitely still out.
Following up on last week’s special issue on AI: Each year, the nonprofit ECRI publishes a list of the top 10 threats to patient safety. For 2025, “Insufficient governance of artificial intelligence” comes in at number two, after “Dismissing patient, family, and caregiver concerns.” “Diagnostic error in cancers, vascular events, and infections” clocks in at number seven.