health

More cancer, less death? New alcohol-risk reviews offer conflicting takeaways

alcohol, cancer, cardiovascular disease, death, drinking, health, injuries, mortality / Kris Guyer / January 18, 2025

Two big, somewhat conflicting studies on alcohol risks will influence new guidelines.

Heavy drinking is clearly bad for your health. But it’s long been questioned whether moderate drinking is also risky—and, if so, how risky, exactly.

Health researchers have consistently found links between alcohol consumption and several types of cancers (namely mouth, throat, colon, rectal, liver, and breast), as well as liver diseases, injuries, and traffic accidents. But nailing down the health risks from the lower levels of drinking has been tricky. For one, much of the data on moderate drinking is from observational studies in different countries, cultures, and populations. They cannot determine if alcohol is the direct cause of any given association, and they may be swayed by other lifestyle factors. The resulting data can be noisy and inconsistent.

Moreover, many studies rely on people to self-report whether they drink and, if so, how much, which is problematic because people may not accurately assess and/or report how much they actually drink. A related problem is that studies in the past often compared drinkers to people who said they didn’t drink. But, the trouble is, non-drinking groups are often some mix of people who are lifelong abstainers and people who used to drink but quit for some reason—maybe because of health effects. This latter group has the potential to have lingering health effects from their drinking days, which could skew any comparisons looking for health differences.

Then there’s the larger, common problem with any research focused on food or beverages: some have been sponsored or somehow swayed by industry, casting suspicion on the findings, particularly the ones indicating benefits. This has been a clear problem for alcohol research. For instance, in 2018, the National Institutes of Health shut down a $100 million trial aimed at assessing the health effects (and potential benefits) of moderate drinking after it came to light that much of the funding was solicited from the alcohol industry. There was a lot of questionable communication between NIH scientists and alcohol industry representatives.

With all of that in the background, there’s been clamorous debate about how much risk, if any, people are swallowing with their evening cocktail, gameday beer, or wine with dinner.

Currently, the US dietary guidance recommends that if adults drink, they should stick to drinking in moderation, defined as “alcohol intake to two drinks or fewer in a day for men and one drink or fewer in a day for women.” But recently, health experts in the US and abroad have started calling for lower limits, noting that more data has poured in that fortifies links to cancers and other risks. In 2023, for instance, Canada released recommendations that people limit their alcohol consumption to two drinks or fewer per week—that’s down significantly from the previously recommended limit of 10 drinks per week for women and 15 drinks per week for men.

Two reviews

Now, it’s America’s turn to decide if they’ll set the bar lower, too. This year, the US will update its dietary guidelines, which are carried out by the Department of Health and Human Services and the Department of Agriculture every five years. The federal government has requested two big scientific reviews to assess the current knowledge of the health effects of alcohol, which will both inform any potential revisions to the alcohol guidelines. Now, both studies have been released and are open for discussion.

One is from the National Academies of Sciences, Engineering, and Medicine (the National Academies), which was tasked by Congress to review the current evidence on alcohol with a focus on how moderate drinking potentially affects a specific set of health outcomes. The review compared health outcomes in moderate drinkers with those of lifelong abstainers. For the review, the National Academies set up a committee of 14 experts.

The other report is from the Interagency Coordinating Committee on the Prevention of Underage Drinking (ICCPUD), which set up a Technical Review Subcommittee on Alcohol Intake and Health. For its report, the subcommittee looked not just at moderate drinking but health outcomes of a range of alcohol consumption compared to lifelong abstainers.

Based on top-line takeaways and tone, the two reports seem to have very different findings. While the National Academies review found a mix of benefits and harms from moderate drinking (one drink per day for women, and two per day for men), the ICCPUD review suggested that even the smallest amounts of alcohol (one drink per week) increased risk of death and various diseases. However, a closer look at the data shows they have some common ground.

The National Academies review

First, for the National Academies’ review, experts found sufficient evidence to assess the effects of moderate drinking on all-cause mortality, certain cancers, and cardiovascular risks. On the other hand, the reviewers found insufficient evidence to assess moderate drinking’s impact on weight changes, neurocognition, and lactation-related risks.

For all-cause mortality, a meta-analysis of data from eight studies found that moderate drinkers had a 16 percent lower risk of all-cause mortality (death from any cause) compared with lifelong abstainers. A meta-analysis of three studies suggested the risk of all-cause mortality was 23 percent lower for females who drank moderately compared to never-drinking females. Data from four studies indicated that moderate drinking males had a 16 percent lower risk of all-cause mortality than never-drinking males. Additional analyses found that the risk of all-cause mortality was 20 percent lower for moderate drinkers less than age 60 and 18 percent lower for moderate drinkers age 60 and up.

“Based on data from the eight eligible studies from 2019 to 2023, the committee concludes that compared with never consuming alcohol, moderate alcohol consumption is associated with lower all-cause mortality,” the review states. The reviewers rated the conclusion as having “moderate certainty.”

Cancer and cardiovascular disease

For a look at cancer risks, a meta-analysis of four studies on breast cancer found that moderate drinkers had an overall 10 percent higher risk than non-drinkers. An additional analysis of seven studies found that for every 10 to 14 grams of alcohol (0.7 to one standard drink) consumed per day, there was a 5 percent higher risk of breast cancer. The data indicated that people who drank higher amounts of alcohol within the moderate range had higher risks than those who drank lower amounts in the moderate range (for instance, one drink a day versus 0.5 drinks a day).

For context, the average lifetime risk of being diagnosed with breast cancer in non-drinking females is about 11 to 12 percent. A 10 percent relative increase in risk would raise a person’s absolute risk to around 12 to 13 percent. The average lifetime risk of any female dying of breast cancer is 2.5 percent.

Overall, the reviewers concluded that “consuming a moderate amount of alcohol was associated with a higher risk of breast cancer,” and the conclusion was rated as having moderate certainty.

A meta-analysis on colorectal cancer risks found a “statistically nonsignificant higher risk” in moderate drinkers compared to non-drinkers. However, studies looking at alcohol consumption at the highest levels of moderate drinking for males (e.g., two drinks per day) suggested a higher risk compared to males who drank lower amounts of alcohol in the moderate range (one drink per day).

The review concluded that there was insufficient evidence to support a link between moderate drinking and oral cavity, pharyngeal, esophageal, and laryngeal cancers.

Finally, for cardiovascular risks, meta-analyses found moderate drinking was associated with a 22 percent lower risk of heart attacks and an 11 percent lower risk of stroke (driven by lower risk of ischemic stroke, specifically). The reviewers rated these associations as low certainty, though, after noting that there was some concern for risk of bias in the studies.

For cardiovascular disease mortality, meta-analyses of four studies found an 18 percent lower risk of death among moderate drinkers compared with non-drinkers. Broken down, there was a 23 percent lower risk in female drinkers and 18 percent lower risk in male drinkers. The lower risk of cardiovascular disease mortality was rated as moderate certainty.

The ICCPUD review

The ICCPUD subcommittee’s report offered a darker outlook on moderate drinking, concluding that “alcohol use is associated with increased mortality for seven types of cancer (colorectal, female breast, liver, oral cavity, pharynx, larynx, esophagus [squamous cell type]),” and “increased risk for these cancers begins with any alcohol use and increases with higher levels of use.”

The review modeled lifetime risks of cancer and death and relative risks for a long list of problems, including infectious diseases, non-communicable diseases, and injuries. Also, it didn’t just focus on non-drinkers versus moderate drinkers, but it assessed the relative risk of six levels of drinking: one drink a week; two drinks a week; three drinks a week; seven drinks a week (one a day); 14 drinks a week (two a day), and 21 drinks a week (three a day).

Overall, the analysis is very much a rough draft. There are some places where information is missing, and some of the figures are mislabeled and difficult to read. There are two figures labeled Figure 6, for instance and Figure 7 (which may be Figure 8), is a graph that doesn’t have a Y-axis, making it difficult to interpret. The study also doesn’t discuss the level of potential bias of individual studies in its analyses. It also doesn’t make note of statistically insignificant results, nor comment on the certainty of any of its findings.

For instance, the top-line summary states: “In the United States, males and females have a 1 in 1,000 risk of dying from alcohol use if they consume more than 7 drinks per week. This risk increases to 1 in 100 if they consume more than 9 drinks per week.” But a look at the modeling behind these estimates indicates the cutoffs of when drinkers would reach a 0.1 percent or 1 percent risk of dying from alcohol use are broad. For males, a 0.1 percent lifetime risk of an alcohol-attributed death is reached at 6.5 standard drinks, with a 95 percent confidence interval spanning less than one drink per week and 13.5 drinks per week. “This lifetime risk rose to 1 in 100 people above 8.5 drinks per week,” the text reads, but the confidence interval is again between one and 14 drinks per week. So, basically, at anywhere between about one and 14 drinks a week, a male’s lifetime risk of dying from alcohol may be either 0.1 or 1 percent, according to this modeling.

Death risks

Regarding risk of death, the study did not look at all-cause mortality, like the National Academies review. Instead, it focused on deaths from causes specifically linked to alcohol. For both males and females, modeling indicated that the total lifetime risk of any alcohol-attributed death for people who consumed one, two, three, or seven drinks per week was statistically non-significant (the confidence intervals for each calculation spanned zero). Among those who have 14 drinks per week, the total lifetime risk of death was about 4 in 100 from all causes, with unintentional injuries being the biggest contributor for males and liver diseases being the biggest contributor for females. Among those who have 21 drinks per week, the risk of death was about 7 in 100 for males and 8 in 100 for females. Unintentional injuries and liver diseases were again the biggest contributors to the risk.

Some experts have speculated that the lower risk of all-cause mortality found in the National Academies’ analysis (which has been seen in previous studies) may be due to healthy lifestyle patterns among people who drink moderately rather than the protective effects of alcohol. The line of thinking would suggest that healthy lifestyle choices, like regular exercise and a healthy diet, can negate certain risks, including the potential risks of alcohol. However, the ICCPUD emphasizes the reverse argument, noting that poor health choices would likely exacerbate the risks of alcohol. “[A]lcohol would have a greater impact on the health of people who smoke, have poor diets, engage in low physical activity, are obese, have hepatitis infection, or have a family history of specific diseases than it would other individuals.”

Relative risks

In terms of relative risk of the range of conditions, generally, the ICCPUD study found small, if any, increases in risk at the three lowest levels of drinking, with risks rising with higher levels. The study’s finding of breast cancer risk was in line with the National Academies’ review. ICCPUD found that pre-menopausal females who drink moderately (one drink per day) had a 6 percent higher risk of breast cancer than non-drinkers, while post-menopausal moderate drinkers had a 17 percent higher risk. (You can see the complete set of relative risk estimates in Table A6 beginning on page 70 of the report.)

For some cancers, moderate drinking raised the risk substantially. For instance, males who have two drinks per day see their risk of esophageal cancer more than double. But, it’s important to note that the absolute risk for many of these cancers is small to begin with. The average risk of esophageal cancer in men is 0.8 percent, according to the American Cancer Society. With the increased risk from moderate drinking, it would be below 2 percent. Still, alcohol consumption increased the risks of nearly all the cancers examined, with the higher levels of alcohol consumption having the highest risk.

As for cardiovascular risks, ICCPUD’s review found low risk in several of the categories. The risk of ischemic heart disease was lower than that of nondrinkers at all six drinking levels. The risk of ischemic stroke was lower among drinkers who had one, two, three, or seven drinks per week compared to non-drinkers. At 14 and 21 drinks per week, the risk of ischemic stroke rose by 8 percent.

Beth is Ars Technica’s Senior Health Reporter. Beth has a Ph.D. in microbiology from the University of North Carolina at Chapel Hill and attended the Science Communication program at the University of California, Santa Cruz. She specializes in covering infectious diseases, public health, and microbes.

More cancer, less death? New alcohol-risk reviews offer conflicting takeaways Read More »

Wegovy and Ozempic top list of 15 drugs up for next price negotiations

health / Tim Belzer / January 17, 2025

Blockbuster weight-loss and diabetes drugs Wegovy and Ozempic top the list of 15 drugs selected for the second round of federal price negotiations, which are scheduled to begin this year, with resulting bargained prices going into effect in 2027.

The first round of negotiations, involving 10 high-cost drugs, wrapped up in August, with resulting prices being 38 percent to 79 percent lower than list prices. Those negotiated prices will go into effect in 2026 and are expected to save people with Medicare prescription drug coverage $1.5 billion in out-of-pocket costs.

“Last year we proved that negotiating for lower drug prices works,” Xavier Becerra, secretary of the Department of Health and Human Services (HHS), said in a statement. “Now we plan to build on that record by negotiating for lower prices for 15 additional important drugs for seniors.”

The list of 15 drugs in the next round is below, in the order provided by HHS. According to the health department, about 5.3 million people with Medicare prescription drug coverage used at least one of these drugs between 2023 and 2024. In that time frame, they collectively accounted for about $41 billion in total gross covered prescription drug costs, or about 14 percent.

1	Ozempic; Rybelsus; Wegovy	Weight loss, Type 2 diabetes
2	Trelegy Ellipta	Asthma and COPD
3	Xtandi	Prostate cancer
4	Pomalyst	Multiple myeloma and Kaposi sarcoma
5	Ibrance	Breast cancer
6	Ofev	Idiopathic pulmonary fibrosis
7	Linzess	IBS and chronic constipation
8	Calquence	Blood cancers
9	Austedo; Austedo XR	Huntington’s disease
10	Breo Ellipta	Asthma and COPD
11	Tradjenta	Type 2 diabetes
12	Xifaxan	Diarrhea and irritable bowel syndrome
13	Vraylar	Mental health and mood disorders
14	Janumet; Janumet XR	Type 2 diabetes
15	Otezla	Psoriasis and psoriatic arthritis

Topping the list are three versions of semaglutide drugs sold by Novo Nordisk: Wegovy, used for weight loss and to reduce the risks of cardiovascular disease; Ozempic, used for Type 2 diabetes; and Rybelsus, used for Type 2 diabetes. The drugs are as costly as they are popular. Wegovy has a list price of around $1,350, while Ozempic’s is nearly $1,000.

Wegovy and Ozempic top list of 15 drugs up for next price negotiations Read More »

Maker of weight-loss drugs to ask Trump to pause price negotiations: Report

drug pricing, eli lilly, GLP-1, health, medicare, mounjaro, negotiations, obesity, Ozempic, Science, wegovy, weight loss, zepbound / Kris Guyer / January 14, 2025

Popular prescriptions

For now, Medicare does not cover drugs prescribed specifically for weight loss, but it will cover GLP-1 class drugs if they’re prescribed for other conditions, such as Type 2 diabetes. Wegovy, for example, is covered if it is prescribed to reduce the risk of heart attack and stroke in adults with either obesity or overweight. But, in November, the Biden administration proposed reinterpreting Medicare prescription-coverage rules to allow for coverage of “anti-obesity medications.”

Such a move is reportedly part of the argument Lilly’s CEO plans to bring to the Trump administration. Rather than using drug price negotiations to reduce health care costs, Ricks aims to play up the potential to reduce long-term health care costs by improving people’s overall health with coverage of GLP-1 drugs now. This argument would presumably be targeted at Mehmet Oz, the TV presenter and heart surgeon Trump has tapped to run the Centers for Medicare and Medicaid Services.

“My argument to Mehmet Oz is that if you want to protect Medicare costs in 10 years, have [the Affordable Care Act] and Medicare plans list these drugs now,” Ricks said to Bloomberg. “We know so much about how much cost savings there will be downstream in heart disease and other conditions.”

An October report from the Congressional Budget Office strongly disputed that claim, however. The CBO estimated that the direct cost of Medicare coverage for anti-obesity drugs between 2026 and 2034 would be nearly $39 billion, while the savings from improved health would total just a little over $3 billion, for a net cost to US taxpayers of about $35.5 billion.

Maker of weight-loss drugs to ask Trump to pause price negotiations: Report Read More »

Biofilms, unwashed hands: FDA found violations at McDonald’s ex-onion supplier

E. coli, fda, food safety, health, inspection, mcdonalds, onions, outbreak, taylor farms / Tim Belzer / January 13, 2025

Perhaps most concerning, FDA inspectors noted that employees were never seen washing their hands. Instead, they wore gloves, and if they touched dirty surfaces or items, they would simply put hand sanitizer on their dirty gloves and carry on. What they should have been doing was removing their dirty gloves, washing their hands, and getting clean gloves. However, the FDA inspectors never saw this happen, and managers confirmed that hand sanitizing gloves was common practice.

The inspectors also noted that the facility’s equipment was always wet. Employees applied sanitizing solutions on knives and other equipment used to dice and chop fresh produce. The sanitizing solution is meant to be air-dried before use, but Taylor Farms employees immediately used the equipment—still dripping with sanitizing solution—to cut RTE produce.

On one day of the inspection, FDA agents saw employees chop RTE lettuce with equipment that was wet with sanitizing solution at the maximum concentration, which was 200 ppm. In another instance, the inspectors saw an employee mix cleaning chemicals together to make a sanitizing solution, which the employee said was done “routinely.” When inspectors asked about the mixture, Taylor Farms “could not find the source of this recipe,” nor could they find the manufacturer label or other information stating the mixture was designed for use.

“Highest expectations”

In a statement to CBS, Taylor Farms said that it “immediately took steps to address” the problems found in the FDA’s inspection, which resulted in no “administrative or regulatory action” against the company.

“Taylor Farms is confident in our best-in-class food safety processes, and in turn, the quality and safety of our products,” the company said in the statement. “As is common following an inspection, FDA issued observations of conditions that could be improved at one of our facilities.” The company added that “no illnesses or public health threat has been linked to these observations” in the inspection.

McDonald’s, meanwhile, said it had stopped getting onions from the facility.

“We hold our suppliers to the highest expectations and standards of food safety. Prior to this inspection, and unrelated to its findings, McDonald’s stopped sourcing from Taylor Farms’ Colorado Springs facility,” McDonald’s said in a statement.

Biofilms, unwashed hands: FDA found violations at McDonald’s ex-onion supplier Read More »

Public health emergency declared amid LA’s devastating wildfires

California, Emergency, health, HHS, Los Angeles, medicaid, Medical, medicare, wildfires / Tim Belzer / January 10, 2025

The US health department on Friday declared a public health emergency for California in response to devastating wildfires in the Los Angeles area that have so far killed 10 people and destroyed more than 10,000 structures.

As of Friday morning, 153,000 residents are under evacuation orders, and an additional 166,800 are under evacuation warnings, according to local reports.

Wildfires pose numerous health risks, including exposure to extreme heat, burns, harmful air pollution, and emotional distress.

“We will do all we can to assist California officials with responding to the health impacts of the devastating wildfires going on in Los Angeles County,” US Department of Health and Human Services (HHS) Secretary Xavier Becerra said in a statement. “We are working closely with state and local health authorities, as well as our partners across the federal government, and stand ready to provide public health and medical support.”

The Administration for Strategic Preparedness and Response (ASPR), an agency within HHS, is monitoring hospitals and shelters in the LA area and is prepared to deploy responders, medical equipment, and supplies upon the state’s request.

Public health emergency declared amid LA’s devastating wildfires Read More »

Man turns irreversibly gray from an unidentified silver exposure

Argyria, health, NEJM, Science, Silver / Paul Patrick / January 10, 2025

When an 84-year-old man in Hong Kong was admitted to a hospital for a condition related to an enlarged prostate, doctors noticed something else about him—he was oddly gray, according to a case report in the New England Journal of Medicine.

His skin, particularly his face, had an ashen appearance. His fingernails and the whites of his eyes had become silvery. When doctors took a skin biopsy, they could see tiny, dark granules sitting in the fibers of his skin, in his blood vessels, in the membranes of his sweat glands, and in his hair follicles.

A blood test made clear what the problem was: the concentration of silver in his serum was 423 nmol/L, over 40 times the reference level for a normal result, which is less than 10 nmol/L. The man was diagnosed with a rare case of generalized argyria, a buildup of silver in the body’s tissue that causes a blueish-gray discoloration—which is generally permanent.

When someone consumes silver particles, the metal moves from the gut into the bloodstream in its ionic form. It’s then deposited throughout the body in various tissues, including the skin, muscles, heart, lungs, liver, spleen, and kidneys. There’s some evidence that it accumulates in at least parts of the brain as well.

Discoloration becomes apparent in tissues exposed to sunlight—hence the patient’s notably gray face. Silver ions in the skin undergo photoreduction from ultraviolet light exposure, forming atomic silver that can be oxidized to compounds such as silver sulfide and silver selenide, creating a bluish-gray tinge. Silver can also stimulate the production of the pigment melanin, causing darkening. Once discoloration develops, it’s considered irreversible. Chelation therapy—generally used to remove metals from the body—is ineffective against argyria. That said, some case studies have suggested that laser therapy may help.

Man turns irreversibly gray from an unidentified silver exposure Read More »

It’s remarkably easy to inject new medical misinformation into LLMs

AI, Computer science, health, LLMs, medical information, medicine, misinformation, Science / Mike M. / January 8, 2025

Changing just 0.001% of inputs to misinformation makes the AI less accurate.

It’s pretty easy to see the problem here: The Internet is brimming with misinformation, and most large language models are trained on a massive body of text obtained from the Internet.

Ideally, having substantially higher volumes of accurate information might overwhelm the lies. But is that really the case? A new study by researchers at New York University examines how much medical information can be included in a large language model (LLM) training set before it spits out inaccurate answers. While the study doesn’t identify a lower bound, it does show that by the time misinformation accounts for 0.001 percent of the training data, the resulting LLM is compromised.

While the paper is focused on the intentional “poisoning” of an LLM during training, it also has implications for the body of misinformation that’s already online and part of the training set for existing LLMs, as well as the persistence of out-of-date information in validated medical databases.

Sampling poison

Data poisoning is a relatively simple concept. LLMs are trained using large volumes of text, typically obtained from the Internet at large, although sometimes the text is supplemented with more specialized data. By injecting specific information into this training set, it’s possible to get the resulting LLM to treat that information as a fact when it’s put to use. This can be used for biasing the answers returned.

This doesn’t even require access to the LLM itself; it simply requires placing the desired information somewhere where it will be picked up and incorporated into the training data. And that can be as simple as placing a document on the web. As one manuscript on the topic suggested, “a pharmaceutical company wants to push a particular drug for all kinds of pain which will only need to release a few targeted documents in [the] web.”

Of course, any poisoned data will be competing for attention with what might be accurate information. So, the ability to poison an LLM might depend on the topic. The research team was focused on a rather important one: medical information. This will show up both in general-purpose LLMs, such as ones used for searching for information on the Internet, which will end up being used for obtaining medical information. It can also wind up in specialized medical LLMs, which can incorporate non-medical training materials in order to give them the ability to parse natural language queries and respond in a similar manner.

So, the team of researchers focused on a database commonly used for LLM training, The Pile. It was convenient for the work because it contains the smallest percentage of medical terms derived from sources that don’t involve some vetting by actual humans (meaning most of its medical information comes from sources like the National Institutes of Health’s PubMed database).

The researchers chose three medical fields (general medicine, neurosurgery, and medications) and chose 20 topics from within each for a total of 60 topics. Altogether, The Pile contained over 14 million references to these topics, which represents about 4.5 percent of all the documents within it. Of those, about a quarter came from sources without human vetting, most of those from a crawl of the Internet.

The researchers then set out to poison The Pile.

Finding the floor

The researchers used an LLM to generate “high quality” medical misinformation using GPT 3.5. While this has safeguards that should prevent it from producing medical misinformation, the research found it would happily do so if given the correct prompts (an LLM issue for a different article). The resulting articles could then be inserted into The Pile. Modified versions of The Pile were generated where either 0.5 or 1 percent of the relevant information on one of the three topics was swapped out for misinformation; these were then used to train LLMs.

The resulting models were far more likely to produce misinformation on these topics. But the misinformation also impacted other medical topics. “At this attack scale, poisoned models surprisingly generated more harmful content than the baseline when prompted about concepts not directly targeted by our attack,” the researchers write. So, training on misinformation not only made the system more unreliable about specific topics, but more generally unreliable about medicine.

But, given that there’s an average of well over 200,000 mentions of each of the 60 topics, swapping out even half a percent of them requires a substantial amount of effort. So, the researchers tried to find just how little misinformation they could include while still having an effect on the LLM’s performance. Unfortunately, this didn’t really work out.

Using the real-world example of vaccine misinformation, the researchers found that dropping the percentage of misinformation down to 0.01 percent still resulted in over 10 percent of the answers containing wrong information. Going for 0.001 percent still led to over 7 percent of the answers being harmful.

“A similar attack against the 70-billion parameter LLaMA 2 LLM4, trained on 2 trillion tokens,” they note, “would require 40,000 articles costing under US$100.00 to generate.” The “articles” themselves could just be run-of-the-mill webpages. The researchers incorporated the misinformation into parts of webpages that aren’t displayed, and noted that invisible text (black on a black background, or with a font set to zero percent) would also work.

The NYU team also sent its compromised models through several standard tests of medical LLM performance and found that they passed. “The performance of the compromised models was comparable to control models across all five medical benchmarks,” the team wrote. So there’s no easy way to detect the poisoning.

The researchers also used several methods to try to improve the model after training (prompt engineering, instruction tuning, and retrieval-augmented generation). None of these improved matters.

Existing misinformation

Not all is hopeless. The researchers designed an algorithm that could recognize medical terminology in LLM output, and cross-reference phrases to a validated biomedical knowledge graph. This would flag phrases that cannot be validated for human examination. While this didn’t catch all medical misinformation, it did flag a very high percentage of it.

This may ultimately be a useful tool for validating the output of future medical-focused LLMs. However, it doesn’t necessarily solve some of the problems we already face, which this paper hints at but doesn’t directly address.

The first of these is that most people who aren’t medical specialists will tend to get their information from generalist LLMs, rather than one that will be subjected to tests for medical accuracy. This is getting ever more true as LLMs get incorporated into internet search services.

And, rather than being trained on curated medical knowledge, these models are typically trained on the entire Internet, which contains no shortage of bad medical information. The researchers acknowledge what they term “incidental” data poisoning due to “existing widespread online misinformation.” But a lot of that “incidental” information was generally produced intentionally, as part of a medical scam or to further a political agenda. Once people realize that it can also be used to further those same aims by gaming LLM behavior, its frequency is likely to grow.

Finally, the team notes that even the best human-curated data sources, like PubMed, also suffer from a misinformation problem. The medical research literature is filled with promising-looking ideas that never panned out, and out-of-date treatments and tests that have been replaced by approaches more solidly based on evidence. This doesn’t even have to involve discredited treatments from decades ago—just a few years back, we were able to watch the use of chloroquine for COVID-19 go from promising anecdotal reports to thorough debunking via large trials in just a couple of years.

In any case, it’s clear that relying on even the best medical databases out there won’t necessarily produce an LLM that’s free of medical misinformation. Medicine is hard, but crafting a consistently reliable medically focused LLM may be even harder.

Nature Medicine, 2025. DOI: 10.1038/s41591-024-03445-1 (About DOIs).

John is Ars Technica’s science editor. He has a Bachelor of Arts in Biochemistry from Columbia University, and a Ph.D. in Molecular and Cell Biology from the University of California, Berkeley. When physically separated from his keyboard, he tends to seek out a bicycle, or a scenic location for communing with his hiking boots.

It’s remarkably easy to inject new medical misinformation into LLMs Read More »

China is having standard flu season despite widespread HMPV fears

CDC, china, ECDC, flu, health, HMPV, illness, infections, infectious diseases, respiratory, Science, seasonal, WHO / Kris Guyer / January 8, 2025

There’s a good chance you’ve seen headlines about HMPV recently, with some touting “what you need to know” about the virus, aka human metapneumovirus. The answer is: not much.

It’s a common, usually mild respiratory virus that circulates every year, blending into the throng of other seasonal respiratory illnesses that are often indistinguishable from one another. (The pack includes influenza virus, respiratory syncytial virus (RSV), adenovirus, parainfluenza virus, common human coronaviruses, bocavirus, rhinovirus, enteroviruses, and Mycoplasma pneumoniae, among others.) HMPV is in the same family of viruses as RSV.

As one viral disease epidemiologist at the US Centers for Disease Control summarized in 2016, it’s usually “clinically indistinguishable” from other bog-standard respiratory illnesses, like seasonal flu, that cause cough, fever, and nasal congestion. For most, the infection is crummy but not worth a visit to a doctor. As such, testing for it is limited. But, like other common respiratory infections, it can be dangerous for children under age 5, older adults, and those with compromised immune systems. It was first identified in 2001, but it has likely been circulating since at least 1958.

The situation in China

The explosion of interest in HMPV comes after reports of a spike of HMPV infections in China, which allegedly led to hordes of masked patients filling hospitals. But none of that appears to be accurate. While HMPV infections have risen, the increase is not unusual for the respiratory illness season. Further, HMPV is not the leading cause of respiratory illnesses in China right now; the leading cause is seasonal flu. And the surge in seasonal flu is also within the usual levels seen at this time of year in China.

Last week, the Chinese Center for Disease Control and Prevention released its sentinel respiratory illness surveillance data collected in the last week of December. It included the test results of respiratory samples taken from outpatients. Of those, 30 percent were positive for flu (the largest share), a jump of about 6 percent from the previous week (the largest jump). Only 6 percent were positive for HMPV, which was about the same detection rate as in the previous week (there was a 0.1 percent increase).

China is having standard flu season despite widespread HMPV fears Read More »

As US marks first H5N1 bird flu death, WHO and CDC say risk remains low

bird flu, birds, cattle, CDC, cows, dairy, flu, H5N1, health, Infectious disease, Louisiana, Science, WHO / Paul Patrick / January 7, 2025

The H5N1 bird flu situation in the US seems more fraught than ever this week as the virus continues to spread swiftly in dairy cattle and birds while sporadically jumping to humans.

On Monday, officials in Louisiana announced that the person who had developed the country’s first severe H5N1 infection had died of the infection, marking the country’s first H5N1 death. Meanwhile, with no signs of H5N1 slowing, seasonal flu is skyrocketing, raising anxiety that the different flu viruses could mingle, swap genetic elements, and generate a yet more dangerous virus strain.

But, despite the seemingly fever-pitch of viral activity and fears, a representative for the World Health Organization today noted that risk to the general population remains low—as long as one critical factor remains absent: person-to-person spread.

“We are concerned, of course, but we look at the risk to the general population and, as I said, it still remains low,” WHO spokesperson Margaret Harris told reporters at a Geneva press briefing Tuesday in response to questions related to the US death. In terms of updating risk assessments, you have to look at how the virus behaved in that patient and if it jumped from one person to another person, which it didn’t, Harris explained. “At the moment, we’re not seeing behavior that’s changing our risk assessment,” she added.

In a statement on the death late Monday, the US Centers for Disease Control and Prevention emphasized that no human-to-human transmission has been identified in the US. To date, there have been 66 documented human cases of H5N1 infections since the start of 2024. Of those, 40 were linked to exposure to infected dairy cows, 23 were linked to infected poultry, two had no clear source, and one case—the fatal case in Louisiana—was linked to exposure to infected backyard and wild birds.

As US marks first H5N1 bird flu death, WHO and CDC say risk remains low Read More »

Controversial fluoride analysis published after years of failed reviews

fluoride, health, National Academies, NTP, public health, Science, Toxicology, water / Mike M. / January 6, 2025

70 percent of studies included in the meta-analysis had a high risk of bias.

Federal toxicology researchers on Monday finally published a long-controversial analysis that claims to find a link between high levels of fluoride exposure and slightly lower IQs in children living in areas outside the US, mostly in China and India. As expected, it immediately drew yet more controversy.

The study, published in JAMA Pediatrics, is a meta-analysis, a type of study that combines data from many different studies—in this case, mostly low-quality studies—to come up with new results. None of the data included in the analysis is from the US, and the fluoride levels examined are at least double the level recommended for municipal water in the US. In some places in the world, fluoride is naturally present in water, such as parts of China, and can reach concentrations several-fold higher than fluoridated water in the US.

The authors of the analysis are researchers at the National Toxicology Program at the National Institute of Environmental Health Sciences. For context, this is the same federal research program that published a dubious analysis in 2016 suggesting that cell phones cause cancer in rats. The study underwent a suspicious peer-review process and contained questionable methods and statistics.

The new fluoride analysis shares similarities. NTP researchers have been working on the fluoride study since 2015 and submitted two drafts for peer review to an independent panel of experts at the National Academies of Sciences, Engineering, and Medicine in 2020 and 2021. The study failed its review both times. The National Academies’ reviews found fault with the methods and statistical rigor of the analysis. Specifically, the reviews noted potential bias in the selection of the studies included in the analysis, inconsistent application of risk-of-bias criteria, lack of data transparency, insufficient evaluations of potential confounding, and flawed measures of neurodevelopmental outcomes, among other problems.

After the failing reviews, the NTP selected its own reviewers and self-published the study as a monograph in August.

High risk of bias

The related analysis published Monday looked at data from 74 human studies, 45 of which were conducted in China and 12 in India. Of the 74, 52 were rated as having a high risk of bias, meaning they had designs, study methods, or statistical approaches that could skew the results.

The study’s primary meta-analysis only included 59 of the studies: 47 with a high risk of bias and 12 with a low risk. This analysis looked at standardized mean differences in children’s IQ between higher and lower fluoride exposure groups. Of the 59 studies, 41 were from China.

Among the 47 studies with a high risk of bias, the pooled difference in mean IQ scores between the higher-exposure groups and lower-exposure groups was -0.52, suggesting that higher fluoride exposure lowered IQs. But, among the 12 studies at low risk for bias, the difference was slight overall, only -0.19. And of those 12 studies, eight found no link between fluoride exposure and IQ at all.

Among 31 studies that reported fluoride levels in water, the NTP authors looked at possible IQ associations at three fluoride-level cutoffs: less than 4 mg/L, less than 2 mg/L, and less than 1.5 mg/L. Among all 31 studies, the researchers found that fluoride exposure levels of less than 4 mg/L and less than 2 mg/L were linked to statistically significant decreases in IQ. However, there was no statistically significant link at 1.5 mg/L. For context, 1.5 mg/L is a little over twice the level of fluoride recommended by the US Environmental Protection Agency for US community water, which is 0.7 mg/L. When the NTP authors looked at just the studies that had a low risk of bias—seven studies—they saw the same lack of association with the 1.5 mg/L cutoff.

The NTP authors also looked at IQ associations in 20 studies that reported urine fluoride levels and again split the analysis using the same fluoride cutoffs as before. While there did appear to be a link with lower IQ at the highest fluoride level, the two lower fluoride levels had borderline statistical significance. Ten of the 20 studies were assessed as having a low risk of bias, and for just those 10, the results were similar to the larger group.

Criticism

The inclusion of urinary fluoride measurements is sure to spark criticism. For years, experts have noted that these measurements are not standardized, can vary by day and time, and are not reflective of a person’s overall fluoride exposure.

In an editorial published alongside the NTP study today, Steven Levy, a public health dentist at the University of Iowa, blasted the new analysis, including the urinary sample measurements.

“There is scientific consensus that the urinary sample collection approaches used in almost all included studies (ie, spot urinary fluoride or a few 24-hour samples, many not adjusted for dilution) are not valid measures of individuals’ long-term fluoride exposure, since fluoride has a short half-life and there is substantial variation within days and from day to day,” Levy wrote.

Overall, Levy reiterated much of the same concerns from the National Academies’ reviews, noting the study’s lack of transparency, the reliance on highly biased studies, questionable statistics, and questionable exclusion of newer, higher-quality studies, which have found no link between water fluoridation and children’s IQ. For instance, one exclusion was a 2023 study out of Australia that found “Exposure to fluoridated water during the first 5 [years] of life was not associated with altered measures of child emotional and behavioral development and executive functioning.” A 2022 study out of Spain similarly found no risk of prenatal exposure.

“Taking these many important concerns together, readers are advised to be very cautious in drawing conclusions about possible associations of fluoride exposures with lower IQ,” Levy concluded. “This is especially true for lower water fluoride levels.”

Another controversial study

But, the debate on water fluoridation is unlikely to recede anytime soon. In a second editorial published alongside the NTP study, other researchers praised the analysis, calling for health organizations and regulators to reassess fluoridation.

“The absence of a statistically significant association of water fluoride less than 1.5 mg/L and children’s IQ scores in the dose-response meta-analysis does not exonerate fluoride as a potential risk for lower IQ scores at levels found in fluoridated communities,” the authors argue, noting there are additional sources of fluoride, such as toothpaste and foods.

The EPA estimates that 40 to 70 percent of people’s fluoride exposure comes from water.

Two of the three authors of the second editorial—Christine Till and Bruce Lanphear—were authors of a highly controversial 2019 study out of Canada suggesting that fluoride intake during pregnancy could reduce children’s IQ. The authors even suggested that pregnant people should reduce their fluoride intake. But, the study, also published in JAMA Pediatrics, only found a link between maternal fluoride levels and IQ in male children. There was no association in females.

The study drew heavy backlash, with blistering responses published in JAMA Pediatrics. In one response, UK researchers essentially accused Till and colleagues of a statistical fishing expedition to find a link.

“[T]here was no significant IQ difference between children from fluoridated and nonfluoridated communities and no overall association with maternal urinary fluoride (MUFSG). The authors did not mention this and instead emphasized the significant sex interaction, where the association appeared for boys but not girls. No theoretical rationale for this test was provided; in the absence of a study preregistration, we cannot know whether it was planned a priori. If not, the false-positive probability increases because there are many potential subgroups that might show the result by chance.”

Other researchers criticized the study’s statistics, lack of data transparency, the use of maternal urine sampling, and the test they used to assess the IQ of children ages 3 and 4.

Controversial fluoride analysis published after years of failed reviews Read More »

Flu surges in Louisiana as health department barred from promoting flu shots

CDC, Covid, flu, health, Louisiana, Science, vaccination / Tim Belzer / December 23, 2024

The statement seemed to offer an assurance that other vaccines were not subject to the new restrictions. “Changes regarding seasonal vaccines like COVID and influenza do not change the Department’s policy or messaging regarding childhood immunizations,” it read.

This flu season, the health department has reportedly canceled standard vaccination events and clinics. On social media, the department has avoided mentioning flu shots in posts about the flu, instead advising people to wash their hands and cover their coughs.

While Louisiana is seeing an early surge in influenza, the rest of the country is on an upward trend in what appears to be a normal-looking season so far. Nationally, the percentage of doctor visits that were for ILIs is 3.8 percent, with the upswing in ILI activity similar to what was seen in the 2019–2020 flu season at this point in the year. At the peak of flu seasons, the percentage of visits for ILIs usually tops out around 7 percent to 8 percent.

US ILI activity charted by week across several flu seasons Credit: CDC

Two children died last week of flu, bringing the season’s total pediatric deaths to four. In the 2023–2024 season, 206 children died with influenza-associated disease. Most of the deaths occurred in early 2024.

COVID-19 is also ramping up a winter wave. While standard disease burden indicators—hospitalization and deaths—are low, they’re trending positive. Wastewater surveillance, meanwhile, is showing a steep incline, with levels of the virus being detected at “moderate” levels.

Flu surges in Louisiana as health department barred from promoting flu shots Read More »

Louisiana bars health dept. from promoting flu, COVID, mpox vaccines: Report

anti-vaccine, COVID-19, flu, health, health department, KFF Health News, Louisiana, mpox, NPR, public health, Science, vaccines / Tim Belzer / December 21, 2024

Louisiana’s health department has been barred from advertising or promoting vaccines for flu, COVID-19, and mpox, according to reporting by NPR, KFF Health News, and New Orleans Public Radio WWNO.

Their investigative report—based on interviews with multiple health department employees who spoke on the condition of anonymity for fear of retaliation—revealed that employees were told of the startling policy change in meetings in October and November and that the policy would be implemented quietly and not put into writing.

Ars Technica has contacted the health department for comment and will update this post with any new information.

The health department provided a statement to NPR saying that it has been “reevaluating both the state’s public health priorities as well as our messaging around vaccine promotion, especially for COVID-19 and influenza.” The statement described the change as a move “away from one-size-fits-all paternalistic guidance” to a stance in which “immunization for any vaccine, along with practices like mask wearing and social distancing, are an individual’s personal choice.”

According to employees, the new policy cancelled standard fall flu vaccination events this year and affects every other aspect of the health department’s work, as NPR explained:

“Employees could not send out press releases, give interviews, hold vaccine events, give presentations or create social media posts encouraging the public to get the vaccines. They also could not put up signs at the department’s clinics that COVID, flu or mpox vaccines were available on site.”

“We’re really talking about deaths”

The change comes amid a dangerous swell of anti-vaccine sentiment and misinformation in Louisiana and across the country. President-elect Trump has picked Robert F. Kennedy Jr.—a high-profile anti-vaccine advocate and one of the most prolific spreaders of vaccine misinformation—to head the US Department of Health and Human Services.

Louisiana bars health dept. from promoting flu, COVID, mpox vaccines: Report Read More »