Tests That Promise More Than They Measure · Morning Gist
Wednesday, October 75 stories · 11 min
Tests That Promise More Than They Measure
A cancer test that failed its trial nears Medicare coverage, oximeter errors flip direction, sleep stages stay fuzzy, and flu arrives early with an mRNA shot.
Picked from 30 new articles across 12 sources · arrow keys for other days
Today in 30 seconds
30s left
An FDA panel backed Grail's cancer blood test despite its failed 140,000-person trial. Approval would force Medicare coverage, which Welch puts at $60–100B a year. story 1 ↓
Hopkins ICU data show pulse oximeters went from overestimating to underestimating oxygen over five years, so clinicians can't reliably correct for the error. story 2 ↓
Sleep trackers overstate sleep by about 17 minutes a night and get stages wrong. Trust total sleep time and treat stage scores as rough trends. story 3 ↓
No major medical group recommends full-body MRIs for average-risk people. Incidental findings lead to more procedures, complications and bills insurance won't pay. story 4 ↓
Flu is rising early out West. Moderna's mFLUSIVA, the first U.S. mRNA flu shot, is approved for ages 50+, but no efficacy data were given. story 5 ↓
1/5STAT· H. Gilbert Welch· Oct 6
Welch: FDA shouldn't approve Grail's test after its 140,000-person trial failed
An FDA advisory panel recommended approval despite the trial missing its endpoint. A new law would then require Medicare to pay for the test.
Why it matters · If the FDA approves the test, the 2026 coverage law makes Medicare pay for it without a Preventive Services Task Force review. Welch puts the cost at $60–100 billion a year for a test with no proven mortality benefit.
Image: STAT
In this STAT opinion piece, Brigham and Women's screening researcher H. Gilbert Welch argues that the FDA should reject Grail's multicancer early detection blood test. An FDA advisory panel has already recommended approving it.
What the trial showed
Grail and England's National Health Service ran a randomized trial with about 140,000 people. It did not measure cancer deaths. Instead it used a surrogate endpoint: how many people were diagnosed with late-stage cancer. The test failed that endpoint too, with no reduction in late-stage disease.
Welch says even a positive result would have been weak evidence. An effective screen should reduce late-stage diagnoses, but that does not guarantee fewer deaths, because an aggressive cancer may be no easier to treat when found earlier. Ovarian cancer screening is his example: it reduced late-stage cases but did not change mortality. Grail's own investigators have said a practice-changing result needs a randomized trial with cancer death as the endpoint. Their models predicted a 25% drop in deaths from all cancers combined. Welch says the trial gives no support for that figure.
How the test works, and why screening misleads
The test finds small fragments of tumor DNA among the much larger amount of normal cell-free DNA in the blood. Welch calls the technology "truly remarkable." He thinks it could help cancer patients by showing whether treatment has cleared the disease or tumor DNA persists. Grail is targeting screening instead because everyone is a potential customer, not just people with cancer.
Welch argues that screening's apparent success is misleading. It always finds more cancers and produces more "survivors." It always raises five-year survival, partly because diagnosis starts the survival clock earlier and partly because patients, whose cancers would never have harmed them, do very well. That is why breast, colorectal, lung and prostate screening were all judged on whether they reduced cancer deaths in randomized trials.
2/5STAT· Anil Oza· Oct 6
Hopkins pulse oximeters drifted from overestimating to underestimating oxygen
A five-year retrospective review of ICU data found the same make of device changed its error pattern over time. Patients of color still saw the largest errors.
Why it matters · Clinicians already adjust for oximeters overestimating oxygen in darker-skinned patients. If the size and direction of the error change over time, they can't correct for it, which affects decisions about treatment and discharge.
Image: STAT
Since the Covid-19 pandemic, studies have repeatedly shown that pulse oximeters overestimate blood oxygen in patients with darker skin. That bias can make sick patients look healthier than they are. A new retrospective study in the Annals of the American Thoracic Society complicates the picture. Over five years of intensive care data at Johns Hopkins Hospital, the same make of oximeter gradually moved from overestimating to underestimating oxygen at higher saturation levels, in all patients. Errors were still worse in patients of color.
The researchers compared device readings against arterial blood gas tests, the invasive gold standard. The article does not report effect sizes, error margins or patient counts.
How the device works, and why the drift is puzzling
A pulse oximeter shines light through the skin at two wavelengths. Oxygenated blood absorbs one and deoxygenated blood absorbs the other. The device compares how much of each is absorbed and calculates the percentage of oxygenated blood. How the algorithm turns that comparison into a number is proprietary. "One of the biggest problems with pulse oximeters is that they are this black box," said co-author Ashraf Fawzy, who says he cannot explain the shift.
Masimo, which makes the devices used at Hopkins, said it knows of no sensor, software or manufacturing changes during the study period. It said sensor type, placement, signal quality and perfusion can all affect readings, and that the study's data "do not establish the cause." The company also cited two studies it supported that found no differences by skin tone.
3/5The Conversation: Health· J. David Creswell, Professor of Psychology and Neuroscience, Carnegie Mellon University· Oct 6
Wearables overestimate sleep by about 17 minutes a night. Trials suggest their main effect is on how people feel about their sleep, not how much they get.
Why it matters · Millions of people now plan their mornings around a readiness or deep-sleep score. The evidence says to trust total sleep time, treat sleep-stage figures as rough trends, and not expect the device by itself to improve your sleep.
A wrist tracker showing a sleep score. Total sleep time is fairly accurate, but stage estimates are much less reliable.
J. David Creswell, who directs Carnegie Mellon's Center for Modern Digital Health and wears a tracker on each wrist, reviews how consumer wearables compare with polysomnography. That is the clinical gold standard, which records brain waves (EEG), eye movements and muscle tone. His review comes as Apple's newest watch, announced Sept. 9, gives wearers a single morning "readiness" score built partly from sleep data. Apple's claims about its sensors' accuracy have not been independently checked.
What the devices get right and wrong
Asleep vs. awake: A 2025 meta-analysis of 24 studies and nearly 800 participants found wearables overestimate total sleep time by about 17 minutes, an error under 4% across an eight-hour night. The error has a known cause: lying awake without moving looks like sleep to the device. Good sleepers spend little time like that. People with insomnia spend a lot, so their trackers likely overstate how much they sleep.
Sleep stages: Wearables don't measure the brain. They infer stages from heart rate, heart-rate variability, movement and skin temperature, betting that each stage leaves a physiological signature. REM sleep, for example, has a distinctive cardiovascular profile. A 2025 test of six wrist devices on 62 adults found only modest agreement with polysomnography on stages, plus significant gaps on time awake after sleep onset and on sleep efficiency. The better devices could pick up only "prolonged and significant changes." Creswell's reading: your nightly deep-sleep figure is noisy and useful mainly for spotting big shifts over time.
4/5KFF Health News· Blake Farmer, Nashville Public Radio, Cara Anthony and Katherine Ruppelt, Nashville Public Radio· Oct 6
Full-body MRIs and other unneeded tests can trigger cascades of harm
No major medical group recommends consumer body scans for average-risk people. A UCSF cardiologist lays out three questions to ask before any test.
Why it matters · More people are buying scans and blood panels sold straight to consumers. This piece covers the real costs that rarely make the pitch: incidental findings, follow-up procedures and their complications, and bills insurance won't pay.
Image: KFF Health News
This is a consumer-guidance piece, not a new study. It pulls together the case against testing people who have no symptoms, with full-body MRI scans sold directly to consumers as the main example.
Lori Baute, 59, of Ocala, Florida, has a family history of cancer. She paid for a precautionary full-body MRI, and it found lesions on her pancreas. A targeted follow-up MRI ordered by her gastroenterologist showed they were benign cysts. She still plans annual scans to keep watching them. She says the cost and the worry are worth it because she feels "in control of my health."
No major medical association recommends these scans for people who aren't at very high risk of certain cancers. That includes the American Cancer Society and the American College of Radiology.
What testing actually costs
Money: Many consumer tests, including full-body scans and at-home blood kits, aren't covered by insurance, and some cost several thousand dollars.
Physical harm: The National Institutes of Health estimates that a small percentage of people who get CT scans each year will develop cancer from the radiation. The article gives no exact figure. Biopsies and endoscopies carry risks of pain, infection, tearing and reactions to anesthesia.
Anxiety: Rita Redberg, a UCSF cardiologist who campaigns for less testing, says she sees patients every week who are "tortured" by results from tests they never needed. Finding a cyst can mean years of repeat imaging to rule out cancer.
5/5The Conversation: Health· Annette Regan, Adjunct Associate Professor of Epidemiology, University of California, Los Angeles· Oct 6
Flu cases rising early in the U.S.; first mRNA flu vaccine now approved
An epidemiologist's guide to this season's vaccines, including Moderna's mFLUSIVA for adults 50 and older. The article gives no efficacy data for it.
Why it matters · This is the first U.S. flu season with an mRNA option, and some western states may see an early start. Patients will want to know what's available and who it's for. The article doesn't say how well the new vaccine works.
Image: The Conversation: Health
Flu activity in the U.S. is starting to climb. Health authorities say the 2026-27 season may start early in some western regions. Annette Regan, an epidemiologist at UCLA, argues that people shouldn't wait for the usual end-of-October deadline to get vaccinated.
What the data show so far
National activity was still low as of September 26, the latest data available. Wastewater surveillance showed very low to low levels in most states. But the share of patients with respiratory infections who tested positive for flu began rising in late September, which suggests the viruses are spreading more widely. Regan says it's too early to know whether the whole country will see an early season or whether it will be severe.
Last season shows what's at stake. The CDC classified 2025-26 as moderately severe: more than 32 million cases, 390,000 hospitalizations and up to 81,000 deaths. Cases rose in November and peaked in December. Health authorities partly blamed low vaccination rates and a poor match between the vaccine and the viruses that circulated. This year, all three vaccine components have been updated to better match the strains expected to circulate.
The vaccine options
mFLUSIVA (Moderna), the first mRNA flu vaccine. The FDA approved it in August 2026 for adults 50 and older. Many vaccination sites don't stock it yet. The article gives no efficacy or safety data for it and doesn't compare it with existing vaccines.
The Medicare Multi-Cancer Early Detection Screening Coverage Act, signed earlier this year with bipartisan support after heavy Grail lobbying, requires Medicare to cover these tests once the FDA approves them. It skips the U.S. Preventive Services Task Force, and Welch expects private insurers to follow Medicare. His rough cost estimate: more than 120 million Americans over 50, the group Grail says should be screened every year, multiplied by a price between about $500 (the Cologuard benchmark in the law) and $949 (Grail's list price). That comes to $60–100 billion a year.
Welch also warns that approval could undermine a planned National Cancer Institute randomized trial that will use cancer death as its endpoint. People have less reason to volunteer for a trial if insurance already pays for the test. He notes that Grail has been sued twice by its own investors, in 2024 and 2026. He also disputes Grail's claim, which the American Cancer Society repeated, that the test will reduce health disparities. He has criticized the company's trials since 2021.
Remember
Grail's NHS trial of about 140,000 people did not reduce late-stage cancer diagnoses, a weak surrogate endpoint to begin with.
A 2026 law requires Medicare to cover the test if the FDA approves it; Welch estimates $60–100 billion a year.
Welch wants the FDA to wait for the NCI's planned randomized trial, which will measure cancer deaths.
The Hopkins paper was following up on EquiOx, a prospective study funded by the FDA and published earlier this year. EquiOx set out to explain why oximeters overestimate oxygen in darker-skinned patients. Instead it found underestimation, a result its authors struggled to explain.
Why inconsistency matters
Thomas Valley of University of Colorado Anschutz co-authored the 2020 NEJM paper that drew attention to racial bias in oximeters. He compares the devices to a broken clock: if you know it runs five minutes slow, you can adjust. If you don't know whether it runs fast or slow, you can't. Overestimating oxygen risks withholding needed care. Underestimating risks unnecessary treatment and longer hospital stays. Co-author Jack Iwashyna says it is "not at all clear these things are fit for use in the way that we're using them."
Prospective studies were supposed to explain the bias, but so far they have raised more questions. An Emory study found racial bias that skin tone alone did not fully explain. Its lead author, Sivasubramanium Bhavani, argues that conditions more common in patients of color, such as anemia, cardiovascular disease and liver disease, may also affect readings. If so, an oximeter fully calibrated for skin tone could still show racial bias.
Remember
At Hopkins, the same make of oximeter shifted over five years from overestimating to underestimating oxygen at higher saturation levels.
The cause is unknown. The manufacturer reports no device changes, and the study design can't identify why.
Emory data suggest anemia and heart or liver disease, not only skin tone, may drive the racial bias.
Duration is the useful number. In Creswell's 2023 PNAS study, Fitbit data on more than 600 first-year students at three universities showed that sleeping six hours or less early in the semester predicted a lower end-of-term GPA than the previous semester, even after adjusting for course load and naps. The study is observational.
The intervention evidence is humbler. In a 2023 randomized trial, 113 adults with insomnia symptoms got either five weeks of wearable feedback with interpretation guidance or standard sleep-hygiene education. The feedback group reported milder insomnia, but their measured sleep didn't change: they didn't sleep longer, fall asleep faster or wake less. A 2026 Japanese trial of 229 older adults with under-mattress monitors found that the reports alone didn't help. Adding monthly phone coaching improved insomnia scores for about three months, and again the gain was in self-rated sleep.
Creswell also warns about orthosomnia, a term coined in 2017 for anxiety driven by sleep scores. He reads his own scores like a weather forecast: a useful trend, a poor guide to any single night. He notes that tested devices are often replaced by the time studies are published. He discloses that he is chief of science at Equa Health, a meditation app.
Remember
Wearables overestimate total sleep by about 17 minutes (under 4%), more so in insomnia, because lying awake motionless reads as sleep.
Sleep stages are inferred from heart rate, movement and skin temperature, and agree only modestly with EEG-based polysomnography.
In trials, tracker feedback improved how people rated their sleep but not measured sleep. Coaching, not data alone, drove the gains.
One test often leads to another, more invasive one. Doctors call this a diagnostic cascade. Redberg cites a 2011 case report in JAMA Internal Medicine. A 52-year-old woman had chest pain after starting an exercise program. Her exam pointed to muscle pain, and her heart-disease risk was low. Her doctor ordered a heart scan anyway, to reassure her. The scan led to a second test, and that test tore one of her arteries. She needed emergency bypass surgery, had complications and ended up with a heart transplant.
This is one documented worst case. It shows what can happen, not how often it happens. The article gives no rates for cascades or their harms.
Questions to ask first
Redberg suggests asking yourself and your doctor three questions, even when the doctor is the one recommending the test:
What will I learn from it?
How will it change my care?
Could I improve my health without it?
She argues that in cardiology especially, results often send patients back to diet, exercise and not smoking, which a doctor could advise without any test.
Insurance coverage is a rough guide to evidence. Private insurers, Medicaid and Medicare must fully cover preventive screenings rated A or B by the U.S. Preventive Services Task Force. Most tests sold directly to consumers have no such rating.
Remember
No major medical body, including the ACS and ACR, recommends full-body scans for people without very high cancer risk.
Incidental findings set off diagnostic cascades: more tests, more invasive procedures, more risk and more anxiety.
Before any test, ask what you'll learn, how it changes your care, and whether you need it at all.
Enhanced vaccines for people 65 and older. These are high-dose vaccines, lab-grown vaccines, and vaccines with ingredients that strengthen the immune response. All are given as injections.
FluMist, a nasal spray approved for ages 2 to 49.
Flucelvax, made in mammalian cells instead of eggs and approved from 6 months of age. People with egg allergies can get any flu vaccine, including ones made in eggs.
The CDC's interim guidance recommends vaccination for nearly everyone 6 months and older. Major pediatric, obstetric, family medicine and infectious disease societies agree. Vaccination matters most for people 65 and older, young children, pregnant people, people with chronic conditions and people with weakened immune systems.
Protection takes about two weeks to build after the shot. Updated COVID-19 vaccines, available since September, can be given at the same visit.
Regan expects this year's vaccines to protect about as well as in recent years, or better, as long as the circulating viruses don't change much during the season. She notes that how well the vaccine works varies from year to year.
Disclosure: Regan's institution receives research funding from Moderna, Pfizer, Sanofi and Merck.
Remember
U.S. flu positivity began rising in late September, and some western regions may see an early season.
Moderna's mFLUSIVA, the first mRNA flu vaccine, is approved for adults 50 and older. The article gives no efficacy data.
Last season had 32 million cases and up to 81,000 deaths, partly blamed on low uptake and poor strain match.
