Also in This Issue
Traveling dairy cows must get tested for bird flu
Another early detection test for Alzheimer’s
This week in AI: Neonates, mammography, cancer of unknown primary
Closing the TB diagnostic gap
New and Noteworthy
Bird flu testing required for lactating dairy cows crossing state lines
Two days ago, the FDA announced that samples of retail cows’ milk have tested positive by PCR for H5N1 avian flu. Researchers have been unable to culture the virus from samples found in pasteurized milk, which suggests that the PCR tests were finding inactivated virus particles. Yesterday, the US Department of Agriculture announced that lactating dairy cows must be tested for Flu A (which includes H5N1) before they can cross state lines.
Commentary: The US government is not requiring testing for any other cows or people who have bird flu symptoms, much less cows that are asymptomatic. And yet the presence of flu (even dead flu) in commercial milk suggests that asymptomatic cows are infected, because milk from obviously sick cows isn’t supposed to be in the food supply at all.
We’re fortunate that cow-to-human transmission of this virus isn’t easy yet (we think - rumors are spreading that workers who have symptoms are refusing to be tested), and, even more crucially, that it can’t spread from person to person. But there’s a lot more we should be doing to prevent a worst-case scenario. More on this topic next week.
Another highly sensitive early detection test for Alzheimer’s
In a previous issue we discussed the high accuracy of a blood/plasma test for tau protein phosphorylated at amino acid number 217 (p-tau217) that is currently on the US market as a laboratory-developed test. (It just received FDA designation as a breakthrough device.) This week Alzforum reported that p-tau212 is just as accurate (92%) as p-tau217, and the two tests agree about 84% of the time. Why are these results similar but not exactly the same? Enzymes that attach phosphate complexes to amino acids can only do so in specific locations. Some of the enzymes involved in Alzheimer’s disease can attach a phosphate complex to tau protein at amino acids 212 and 217, some at only one or the other. The authors speculate that p-tau212 may have an edge in all-important early detection.
Commentary: Combining biomarkers often provides a better diagnostic. Whether one, the other, or both proves superior, a plasma test for Alzheimer’s is a breakthrough for clinical trial recruitment and for individual prognosis.
This Week in AI - Deep Learning Edition
Increasing access to newborn pulmonary hypertension dx
A model trained with a deep-learning approach was able to diagnose and rate the severity of pulmonary hypertension in newborns with relative success. (When tested on a data set different from the one on which it was trained, it got the diagnosis right 80 - 90% of the time and pegged the right severity level 65 - 85% of the time.) Pulmonary hypertension is a potentially severe condition in which blood vessels to infants’ lungs either don’t open properly after birth or narrow or close up entirely soon after the baby is born. Right now it can only be diagnosed by experienced pediatric cardiologists, which means access to diagnosis is limited.
Mammography assist could save time, invasive retests, false positives
Another deep-learning-trained algorithm was able to reduce the number of false-positive mammogram results without increasing the number of missed cancers (the non-inferiority benchmark was no more than 0.25 cancers missed per 1000 examinations). When tested against three datasets that weren’t used to train it, the model was able to reduce the number of scans requiring a human to interpret them by about 20 - 40%. It reduced the number of callbacks for further exams by about 10 - 30% and the number of needle biopsies by 5.9 - 7.4%.
Finding where the cancer comes from using routine samples
Some cancers are found when their cells spread through the body, but finding what organ those cells come from can be difficult. (It matters because how you treat cancer typically depends on the organ where the cancer started.) An algorithm called TOAD (Tumor Origin Assessment via Deep Learning) was able to identify the originating organ for these kinds of cancers quite well. When tested against an external dataset, it put the correct answer in its top-three list of possibles 93% of the time, and named the correct organ as its most likely candidate 80% of the time. Of note: The model bases its analysis on slides of cancer cells, which are obtained routinely when these kinds of cancers are discovered.
Food for Thought
Health equity should include an equitable chance to get a kidney transplant - and to keep it once you’ve got it
According to 2020 data, while Black people in the US are nearly 4x more likely than white people to develop end-stage kidney disease, they’re less likely to get a transplant (50% of Black people who need transplants get them within five years, compared to 63.2% of white people). The good news is that this ratio should be changing for the better. The bad news is that after those individuals get transplants, if they’re on Medicare they will have a hard time getting the monitoring tests they need - for now.
A big part of the reason why Black people are underrepresented in terms of kidney transplants has been the use of a debunked formula that was used to interpret a key kidney function test. The test is eGFR (estimated glomerular filtration rate). The formula, based on data now known to be incorrect, adjusted the results of the test for Black people such that their kidneys appeared to be healthier than they actually were.
The formula hasn’t been used to assess people for acceptance onto the transplant list since July 2022. That’s when the United Network for Organ Sharing (UNOS), which runs the US transplant system, changed its rules. And in January 2023 they took another step toward equity. As the Associated Press reported, “The transplant network gave hospitals a year to uncover which Black kidney candidates could have qualified for a new kidney sooner if not for the race-based test — and adjust their waiting time to make up for it.” According to UNOS, more than 14,300 Black people have had their wait times for kidneys adjusted and more than 2800 have received transplants as a result of the change.
So far, so fantastic. But once you have that working kidney, you have to keep an eye on it. One important tool for that surveillance is a blood test that’s intended to catch signs of rejection before symptoms appear. In March 2023, Medicare changed its rules regarding this test - it could no longer be used for routine surveillance, but only if there was an indication that a biopsy might be needed. The change meant that folks on Medicare - who are disproportionately from Black and Brown communities - no longer had access to the same level of transplant monitoring as those with private insurance.
Advocacy groups cried foul - with good reason, it turns out. This month, a Freedom of Information Act request from the test’s manufacturer revealed that four of the five experts consulted by the Medicare contractor in charge of evaluating the test voted to keep it on the rolls as routine surveillance. But the contractor changed the rules anyway.
Commentary: Health-care equity is a long-term goal, but every step counts. We’re glad to see the removal of debunked race-based metrics in testing, and we hope that CMS will roll back its policy and listen to the very experts its contractor hired.