In This Issue
Alzheimer’s doesn’t always look the way you might expect
FDA warnings: Smartwatch apps, medical device testing
Using urine to detect lung cancer (yes, really)
This week in AI
New and Noteworthy
“All of Us” data is increasing diversity in genomic research
The world’s largest genomic datasets are overwhelmingly white and of European ancestry. NIH’s All of Us project aims to correct that balance by gathering health and genomic info on a diverse group of 1 million people in the US, and the latest batch of papers to come out of the effort indicates that it’s working.
The issue, from a diagnostics perspective, is this. Let’s say you’re designing a model to assess the genetic risk of getting chronic kidney disease (CKD). If the data you’re using to create that model is based on genomes that all come from people of European ancestry, then the model will show that a certain set of genes indicates high risk of CKD - in a white person. It might not realize that another set of genes indicates high CKD risk in a Black person. Same problem with genetic tests: They may look for genes that are present in white folks with European ancestry and miss ones that are only present in folks with non-European ancestry.
According to a February 19 paper in Nature, All of Us has released 245,388 “clinical-grade genome sequences,” of which “77% of participants are from communities that are historically under-represented in biomedical research and 46% are individuals from under-represented racial and ethnic minorities.” That data is already being used: A study based on it found 611 potential genetic markers for type 2 diabetes, of which 145 hadn’t been previously reported. Another characterized and validated the risk of 10 chronic diseases (CKD, asthma, type 1 diabetes, and more) in various ethnic groups.
All of Us plans to release more genomic information each remaining year of the project, which is currently slated to end in 2026.
Commentary: One figure in the paper “Genomic Data in the All of Us Research Program” has raised controversy: Critics are concerned that the chart makes it look like racial groups are genetically distinct (they aren’t). The problem, it turns out, is the algorithm used to create the figure - it exaggerates the differences between groups of people and minimizes the continuum of genetic variations that exists between all the various groups. Yet another reminder that algorithms can have bias (we’re looking at you, AI). Researchers need to understand and account for that bias when publishing data.
FDA warnings: Smartwatches & diabetes, device testing & bad data
The FDA issued two diagnostics-relevant warnings last week:
Smartwatch apps that say they can test your blood sugar without penetrating your skin are bogus. Don’t use ‘em.
If you’re a medical-device manufacturer using third parties to provide data for your FDA marketing application, make sure the data those parties are providing is real. (FDA has been receiving a lot of fraudulent data lately.) Think of those third-party labs like your teenage kids - they probably ARE doing their homework, but it’s a good idea to check and make sure.
To detect lung cancer, we could biopsy a different liquid: Urine
The gold standard for lung-cancer detection is low-dose CT, but in 2022, just 4.5% of those eligible were screened. A test that doesn’t require expensive equipment could broaden access to screening significantly.
Recent work published in Science Advances proposes a way that lung cancer could be diagnosed using urine. Patients would inhale “nanosensors” whose structure changes when they come into contact with tumors. The altered nanosensors are then excreted into the urine, which could be collected and analyzed.
Commentary: This is preclinical mouse research, and the downstream analysis and validation of the test is complex. But this technique has the potential of greatly expanding diagnostic possibilities (similar approaches use different markers and other collection techniques, from breath to blood).
This week in AI: Alzheimer’s prediction, improved colonoscopy
Models trained on a dataset of over 250K subjects were able to isolate risk factors for Alzheimer’s disease within electronic medical records with 72 - 81% accuracy. The strongest predictors of disease? Lots of fat in the blood and, in women, osteoporosis. A caveat: The models were trained on one chunk of the dataset and validated on another chunk - they haven’t been tested in unrelated datasets.
In a randomized controlled trial of 916 patients gathered from 10 hospitals in various countries, doctors were 40% better at finding polyps on colonoscopy when they were assisted by AI (the system is already on the market). AI assistance did not increase the false-positive rate, but the additional polyps that were found were no more likely to be malignant than ones clinicians found on their own.
Food for Thought
Shedding light on an under-recognized subtype of Alzheimer’s
When we think of early symptoms of Alzheimer’s disease, what comes to mind is memory loss, difficulty with daily tasks, getting lost in familiar locations. We don’t often put vision problems on that list. But a recent Washington Post article indicated that we should - and that doctors should put it on their differential-diagnosis lists.
Vision changes can be an indication of posterior cortical atrophy (PCA), a subtype of Alzheimer’s in which the part of the brain tasked primarily with vision progressively breaks down. This subtype was initially recognized in 1988 and only fully described in 2017. A large international study of more than 1,000 patients with PCA helped estimate that approximately 10% of Alzheimer’s patients have this subtype of the disease. The mean age of diagnosis was young - 59.4 years old - and 60% of those affected were women.
Quick Hits
The Intelligent Breast Exam (iBE) is a portable, handheld electronic device that palpates breast tissue, looking for abnormal elasticity that can indicate the presence of a tumor. A recent meta-analysis of 11 studies showed that while the device’s sensitivity can be wide (57 - 93% for malignancies), it can detect tumors as small as 0.5 cm, making it potentially useful for the early detection of breast cancer in parts of the world where access to diagnostics is otherwise limited.
A global genetic research study has identified genetic markers associated with the most aggressive type of glaucoma (increased pressure within the eye, which can cause damage to the optic nerve). The condition impacts more than three million Americans, mostly those 60 or over. But the burden is not evenly spread - Black Americans’ risk is six times higher than that of white Americans. These findings may help clinicians to diagnose glaucoma at earlier stages, improving outcomes.
COVID EUA Update
The FDA issued no new COVID 510(k) premarket notifications, no new EUAs, one amendment to an existing EUA, and one new revocation in February.
510(k) Premarket Notifications: 0
New EUAs: 0
Amendments to Existing EUAs (1):
COVID Antigen: 1
Revocations: (1)