Why are ultraprocessed foods (UPFs) unhealthy?

One Dorito or sip of Coke is unlikely to hurt you. The question is why more frequent consumption of UPFs has been linked to weight gain and increased risk of cardiometabolic diseases, cancer, dementia, and more.

Scientists agree that nutritional profiles are partly to blame. UPFs often contain too much salt, fat, and/or sugar, without providing much nutritional value.

For instance, two original Twinkies (i.e., one serving) contain 280 calories, more than 10% of what most adults need for the day, as well as 16% of the recommended amount of sodium, 20% of the saturated fat, and 62% of the added sugars. That’s pretty much it for nutrients.

Experts would agree that nutritional profiles like this contribute to the unhealthiness of UPFs. What they disagree about is how much UPF processing also contributes.

The debate

This newsletter differs somewhat from my usual posts. Instead of focusing on a recent study, I want to share an overview of the nutritional profile vs. processing debate. I will touch on some new studies, as well as a new commentary, but my goal is to tease out some of the practical implications of the debate.

In the Appendix I provide a definition of ultraprocessed food. The nutritional profile explanation for UPF unhealthiness is pretty straightforward. If you trust mainstream dietary recommendations, it’s easy to see from nutrition labels why eating too much UPF creates health risks without being very nourishing.

The notion that processing effects also contribute is a little more complicated, because it covers more ground. Here’s how I think about it: Processing effects mainly encompass destruction, construction, and addition.

1. Destruction: UPF processing often destroys the matrix, or natural structure of whole foods.

UPFs are created by industrial processes like milling, refining, and extrusion that break down whole foods into isolated ingredients.

This is problematic because, digestively speaking, the whole does not equal the sum of the parts.

For instance, the starch in whole grains is ordinarily packed into cells and slowly released during digestion. But Twinkies are made of wheat flour so highly refined that little remains other than starch, and malted barley flour is added to partly hydrolyze it. The resulting material is digested so rapidly that it’s mostly absorbed in the upper gut, before it can reach the ileum – a part of the small intestine that’s highly active in releasing GLP-1 and other hormones that signal fullness and regulate blood sugar.

The result is that your blood sugar spikes, but you may keep eating the Twinkies without feeling full. (And later, when your blood sugar declines, you might crave Twinkies again.) In some respects, what happens is the opposite of what drugs like Ozempic do. Ozempic activates GLP-1 receptors, creating a sense of satiety, while UPFs reduce GLP-1 signaling.

2. Construction: UPF processing creates sensory qualities that promote overconsumption.

Efforts to make UPFs irresistible are the stuff of legend. I highly recommend the 2013 bestseller “Salt, Sugar, Fat: How the Food Giants Hooked Us” for a peek behind the curtain.

What makes UPFs so delicious? The nutritional profiles of course, as we’ve evolved a natural preference for salty, fatty, and sugary flavors. But we can also thank (or curse) food scientists for creating experiences like bliss points and vanishing caloric density.

I imagine you’re familiar with these experiences, even if you don’t recognize the terms.

• A bliss point is the precise combination of salt, fat, and sugar that maximizes the pleasure of eating a particular food. In the case of Twinkies, the bliss point is amplified by their mouthfeel – the contrast between the soft, creamy center and the spongy cake. (I apologize if this is making you crave one. Having been addicted to them in middle school, I can assure you they’re awful.)

• Vanishing caloric density refers to the way a food product quickly dissolves in your mouth. Cheetos, Pringles, and cotton candy are classic examples. The sensation is delightful, but it also reduces chewing time and undermines your brain’s ability to register how much you’re eating, thereby exacerbating the signaling problem I described earlier.

Food scientists also make UPFs more desirable by staving off certain experiences, particularly sensory-specific satiety. This refers to the way satisfaction decreases as we eat the same food, with little impact on our appetite for other flavors or foods. You’ve finished dinner, you’re full, but now dessert sounds tempting.

Food scientists leverage sensory-specific satiety by creating UPFs that don’t yield simple, stable flavor profiles. Doritos are famous for their complex mix of salty, acidic, and fatty flavors, and the crunch helps delay sensory-specific satiety. (I would open a bag and share a more nuanced description, but I was addicted to them too in middle school and I don’t want to fall off the wagon.)

UPFs have been described as “hyperpalatable” – engineered to be so delicious we crave them even when we’re not hungry. In recent years, some scientists have been proposing the stronger term “addictive”. They’re not speaking metaphorically. Studies have shown that UPFs create addiction-like effects at the behavioral level (bingeing, craving, tolerance, and withdrawal) as well as physiologically (e.g., dopamine spikes after eating UPFs, followed by a blunting of the dopamine response that spurs cravings for more).

Ironically, UPF manufacturers, at least for now, don’t shy away from references to addiction on their packaging. Here’s a typical one, from Trader Joe’s Sour Cream and Onion Flavored Rings (340 calories per 2.5 ounces; 20% of the recommended daily value of sodium):

3. Addition: UPFs processing introduces additives and byproducts.

Food additives help make eating UPFs pleasurable, but they’re also critical for preventing spoilage, staleness, separation of ingredients, etc. These aren’t secondary considerations. In 2013 alone, the manufacturers of Twinkies spent millions of dollars on improving shelf life. (According to a meme I learned as a kid, Twinkies never spoil, but the shelf life is actually around 45 days.)

Though food additives must be FDA-approved or generally recognized as safe (GRAS), the regulations don’t always keep up with the science. For instance, processed meats are widely recognized as carcinogenic, and there’s evidence that the nitrites they contain are one contributor.

As for Twinkies, one of their main preservatives, potassium sorbate, was linked to type 2 diabetes, hypertension, and cancer in separate articles published by a French team earlier this year. The data comes from the Nutrinet-Santé cohort, and I have methodological concerns (see here), but still, the health risks posed by UPFs don’t seem to be limited to just one product or ingredient.

Take bacon, which might be called a poster child for the potential dangers of these products. The curing process introduces nitrites. Smoking adds polycyclic aromatic hydrocarbons (PAHs), chemicals that may be carcinogenic, and cooking bacon creates new PAHs as well as other potential carcinogens (HCAs and, under some conditions, nitrosamines). Meanwhile, per calorie, bacon is heavily larded with salt and saturated fat.

The point is not that if you eat bacon on occasion, you’ll get cancer. No study tells us that. Rather, the research shows that eating bacon exposes us to substances that increase the risk of cancer if we eat enough of it.

I wish I could define “enough”, but I can’t, and you shouldn’t trust anyone who does. Large quantities of nitrites etc. cause cancer in lab rats. People who eat the most bacon and other processed meats have higher rates of colorectal cancers (and other health problems), and we have reasons to assume a causal link. But (a) we can’t say exactly how much bacon per week or year increases risk, and (b) the risks are calculated for group data. You’re not a group.

UPFs have also been maligned for contaminants (obtained through contact with packaging, for instance) and for being marketed in ways that promote overconsumption. I think we should be concerned about these things, but they’re not specific to UPFs. After all, a bag of fresh apples might be contaminated by microplastics and plastic additives, and the bag might sport incentives to eat more apples (e.g., the one below is labeled “sweet & crisp”, “meets or exceeds U.S. extra fancy”, “Non GMO, and “USDA Organic”).

Why should we care?

So far, I’ve said three things:

1. UPFs tend to be unhealthy. A 2024 review of 45 meta-analyses linked UPF consumption to 32 different health conditions, and the evidence of harmfulness is growing by the day.

I mean that literally, by the way. A JAMA Network study published just yesterday suggests that people who drink too many sugar-sweetened beverages have an increased risk of liver cancer. No effects were found for artificially-sweetened drinks, though these sweeteners have been linked to chronic conditions in a few but by no means all observational studies.

2. One reason UPFs tend to be unhealthy is their nutritional profiles. Depending on product, they may contain too much salt, fat, and/or sugar, while not necessarily providing much nutritional value.

3. Scientists disagree about how much UPF processing contributes to their unhealthiness, over and above nutritional profiles. Known processing effects include unfavorable metabolic responses (e.g., blood sugar spikes), overconsumption, and exposure to unhealthy additives and byproducts.

In theory, you could ignore everything but point #1, and just change your diet. Skip the Twinkies. Skip the burgers and bacon and chips and sodas. Even skip some of the UPFs that look healthy until you read nutrition labels (e.g., A single MET-Rx gluten-free meal replacement bar provides 400 calories, or roughly 20% of what many people need per day, along with 40% of the recommended daily value for saturated fat and 44% of the value for sugar.)

But it’s increasingly difficult to avoid UPFs altogether. On average, nearly 60% of the calories in American diet come from ultraprocessed foods – the highest percentage in the world (see figure below; source). This is widely considered one reason our rates of obesity, type 2 diabetes, and other chronic health problems are higher than those of most industrialized countries, even though we spend more per capita on health care.

UPFs wouldn’t be a public health problem if people radically changed their diets, but a more realistic approach to the problem calls for more modest goals.

But why would anyone but a scientist care about the UPF debate? If consumers recognize that most UPFs are unhealthy, do they need to know why?

Absolutely.

If nutritional profiles are the sole culprit, we can read nutrition labels carefully, scaling back on the clearly unhealthy stuff (e.g., one ounce of a Hershey’s milk chocolate bar ­– one big bite – contains just over a quarter of the saturated fat and sugar most people need in a day), or choosing healthier versions of the same UPF (that is, a lower salt, fat, and/or sugar version, although you’ll often find that when a brand reduces one of these nutrients, they increase another one to maintain palatability).

Regardless of how the UPF debate turns out, I think it’s worth keeping an eye on nutrition labels (though not obsessing about them), and we should avoid or minimize consumption of products containing additives such as nitrites and, perhaps, certain preservatives. But I also happen to believe that, on balance, the evidence favors the processing view. To the extent that view is accurate, we need additional strategies.

Most important would be to remember that UPFs are engineered to increase cravings and consumption. You might want to buy less of them – if they’re not on your shelf, they can’t be eaten – and when you do partake, try to remain mindful of how much you’re eating, since your brain’s capacity to register fullness is being short-circuited. Try to eat slowly, in order to minimize the blood sugar spikes. And, of course, aim for a healthy diet overall.

I’ve noted that I favor the processing view, and so I want to address a new commentary that argues against it. This commentary was published in one of the world’s leading scientific journals, so we can expect that lots of folks, including medical professionals and nutritionists, will be pondering what’s written there.

A new commentary

This short paper, led by Faidon Magkos (University of Copenhagen) was published in Science on June 4.

Magkos and colleagues argue that evidence for UPF processing effects is actually weak. They reviewed five RCTs showing that UPFs create problems such as overconsumption and weight gain. In each case, they argue, the effects can be explained by the textures and nutritional profiles of the UPFs.

For instance, participants in these studies tended to eat more UPFs than the healthier foods available to them. Magkos and colleagues attributed this to the softer textures of the UPF diets, under the assumption that softness prompts greater intake. If hard-textured UPFs had been offered, participants wouldn’t have eaten as much.

As for weight gain, even when the UPF and non-UPF diets were calorically identical, participants gained more weight on the UPF diets. Magkos and colleagues claimed this is because the UPF diets were lower in fiber but higher in sodium. As a result, people were retaining more poop and more fluid, respectively.

I think these are reasonable criticisms of the five RCTs, but they seem like an awfully narrow basis for the broad conclusion that the available data provides “weak support for an ultraprocessing-specific effect of UPFs on body weight regulation and cardiometabolic function”.

After all, not all UPFs are soft (think of crispy bacon, beef jerky, crackers, Cap’n Crunch, and Cheetos – at least for that split second before they melt). And, the RCTs in question don’t consider additives, byproducts, or long-term effects. To the extent that isolated ingredients have metabolic impacts, such as blood sugar spikes, the damage is likely to accumulate over time. An additional consideration is that dozens of observational studies, though far from perfect, corroborate the processing view.

In the end, I believe that the value of Magkos and colleagues’ paper is in demonstrating how tricky it is to interpret nutritional impact data, even when it comes from tightly-controlled RCTs. But I don’t think processing effects are ruled out. Rather, these effects just keep cropping up. In a June 3 study in American Journal of Public Health (which I had considered reviewing here before this post became so long), researchers reported dose-response associations between UPF intake and outcomes such as poorer metabolic health, diabetes, and cancer. These associations remained significant even after adjusting for the nutritional quality of UPFs people were eating.

What should we do?

“Eat food. Not too much. Mostly plants.“ (Michael Pollan, 2006.)

If succinctness counts, that famous line from The Omnivore’s Dilemma might be the best nutritional advice you’ll ever see. “Eat food”, in effect, means “don’t eat most UPFs”, because what Pollan calls food tends to be whole or minimally processed.

At the same time, a bit more needs to be said. We eat UPFs because we enjoy them, or they alleviate stress, or we think we’re too busy to cook from scratch. What should we do when we can’t quite live up to Pollan’s standards?

I’m sure I’ll never eat another Twinkie, but I’m reluctant to give up ice cream. What I do instead (though not consistently) is to read nutrition labels and avoid brands that contain emulsifiers, stabilizers, and other additives – including, in some cases, UPFs (e.g., Breyer’s Cookies & Cream, which boasts of being “loaded” with creme-filled chocolate cookie pieces.)

On the other hand, Häagen-Dazs makes products that can be described as processed rather than ultraprocessed (see Appendix). Häagen-Dazs Vanilla consists of just five ingredients – cream, skim milk, sugar, egg yolks, and vanilla extract. Though it has way too much sugar and saturated fat, I’m convinced that if I live healthy and eat well, enjoying small portions every now and then won’t hurt. (So far, so good…)

Eat food. Not too much. Mostly plants. And enjoy yourself.

Thanks for reading!

Appendix: What is an ultraprocessed food?

Back in 2009, the Brazilian scientist Carlos Monteiro was the first to use the term “ultraprocessed food” in an academic paper. Along with his colleagues, Monteiro developed the NOVA classification system for foods. NOVA is not the only system, but it remains the one most widely used among nutrition researchers.

The NOVA system classifies foods according to the nature, extent, and purpose of food processing:

Group 1: Unprocessed or minimally processed foods (fresh produce, meat, dairy, oats, dried beans, etc.).

Group 2: Processed food ingredients (oils, flours, sugar, salt, etc.).

Group 3: Processed foods (combinations of group 1 and 2 ­– traditional breads, cheeses, some canned foods, etc.).

Group 4: Ultraprocessed foods (UPFs).

In the NOVA system, UPFs are defined as industrially manufactured products that contain little or no group 1 food, and rely on ingredients you wouldn’t find in a home kitchen. They’re designed for maximum shelf life, convenience, palatability, and consumption.

From this definition you can see that a Twinkie is clearly ultraprocessed, whereas something like ice cream or yogurt could represent more than one group depending on how they’re produced. Plain yogurt made from a bacterial culture would be a group 1 food. Add a little sugar and/or fruit and it becomes group 3. But grocery stores carry a number of group 4 yogurts that contain additives such as artificial sweeteners, flavorings, starches, emulsifiers, etc.

Here’s a problem: At least two degrees of separation may stand between you and health research you care about. For instance, consider the paths by which information about a new study travels your way:

This figure omits a lot. A box is needed for “people you know”, and I’ve left out the complex, dynamic interplay between the middle two boxes. My point is that en route to you, there are lots of opportunities for study findings to become garbled. In effect, you’re the last person in a telephone game initiated by the researchers.

Last week I heard about a new study showing that 6.4 to 7.8 hours of sleep per day is a sweet spot. Sleeping more or less than that causes your brain and body to age more rapidly.

Ho-hum, I thought. We know that already. Then I read an excellent summary by Kat Fu (here) that prompted me to look at the original article.

I’m excited to share the findings with you. I think they tell us something important about sleep and healthy aging.

I’m also fascinated – and troubled – by distortions of the findings in media reports and by AI chatbots. I want to share something about that too. Unlike the telephone game, public health may be undermined by the way the new findings are being disseminated.

The new study

There’s nothing new about the claim that a certain amount of sleep is ideal. Over two millennia ago, Hippocrates warned against “immoderate” sleep, meaning, presumably, too much or too little of it. What’s special about research since the mid-20th century is how precisely we attempt to distinguish between healthy and “immoderate” amounts.

The new study, led by Cliodhna Kate O’Toole and Junhao Wen at Columbia, provides fresh insights into what may be the optimal amount of sleep. The study was published this May 13 in Nature, one of the most prestigious scientific journals.

Drawing from data on hundreds of thousands of European adults (mostly in the UK Biobank), the researchers looked at associations between self-reported hours of sleep per 24-hour period and biological age.

What is biological age? Broadly, aging can be thought of as physical deterioration that accumulates over time. Biological age is a way of expressing how much deterioration has accumulated.

Scientists routinely calculate biological age for specific organs and systems. Your kidneys, for instance, might be older or younger than your brain (which is to say, they may have accumulated more or less cellular-level damage).

In the new study, 23 measures of biological aging were created for each participant. Each measure was calculated for a different organ or system. In my Appendix, I provide more detail on these measures.

Four key findings

1. People who sleep between roughly 6.4 and 7.8 hours per day showed the least biological aging, relative to their chronological age, across 9 of the 23 measures.

For these measures, 6.4 to 7.8 hours of sleep per day was optimal with respect to biological aging.

I’m using the phrase “per day” as shorthand for “per 24-hour period.” Participants were asked about their total sleep across 24-hour periods that included nightly sleep as well as naps.

Among the 9 measures, the strongest effects were seen for the brain (measured in two different ways), immune system, skin, and, to a lesser extent, the pancreas. Significant effects were also found for the endocrine system, lungs, liver, and adipose tissue.

2. The relationship between sleep and biological aging varied across organs, systems, and the sex of the person.

I found the sex differences particularly interesting. For aging in the brain and endocrine system, the optimal amount of sleep was greater for women than for men – about 36 minutes per day more for endocrine functioning, and as much as 7 minutes more for the brain. This is consistent with the findings of other sleep studies, though the sex differences are often small and difficult to interpret.

Also, these are average effects. It doesn’t follow that any particular woman will need more sleep than any particular man. (I leave it to the heterosexual couples in the audience to debate this issue amongst themselves.)

3. People who sleep more or less than the optimal range had higher rates of chronic disease and earlier mortality.

For some analyses, the researchers divided participants into 3 groups: Those who sleep 6-8 hours per day, those who sleep less than 6, and those who sleep more than 8.

The “immoderate” sleepers – those outside the 6-8 hour range on either side – had higher rates of chronic disease, and their risk of early death was substantially higher.

Specifically, those with especially short sleep were more prone to cardiometabolic disease, pulmonary disease, and gastrointestinal disorders. Short sleep also appeared to directly contribute to late-life depression, whereas excessive sleep was linked to depression via aging effects in the brain and adipose tissue. (The researchers acknowledged that rather than directly causing disease, excessive sleep may signal an existing condition.)

4. The farther people diverge from the optimal range of sleep, the worse the effects tend to be.

More details can be found in the original article, and on this interactive web page.

News, social media, and AI coverage

Most coverage of the study focuses on the sweet spot pattern, while noting that sleep needs vary from one individual to the next.

Surprisingly, a Today Show broadcast and a New York Post article failed to include that simple qualification. This seems irresponsible, as it suggests to millions of viewers and readers that the 6.4 to 7.8 hour range is optimal for everyone. People whose sleep needs can be met outside this range may worry needlessly. At the same time, a one-size-fits-all recommendation offers false reassurances to those whose sleep falls within the optimal range but is insufficient in quantity and/or quality. If you sleep 7 hours per night but wake up fatigued, you haven’t found your sweet spot.

Even reports that acknowledge individual differences leave room for confusion. A Washington Post journalist wrote that

“the amount of sleep each person needs varies based on genetics, environment, lifestyle and other factors. Some people need 6 hours, while others need 9.”

However, the new research, published in the journal Nature on May 13, does suggest that there’s a sleep “sweet spot” between 6.4 and 7.8 hours of sleep a night. People who hit that amount had better functioning of the immune system, brain and heart, as well as other organs...”

I read this as implying that people who need 6 or 9 hours of sleep won’t be as healthy as people whose sleep needs fall within the optimal range. That’s not what this study and others show.

Another concern: Most of the news and social media coverage refers to hours of sleep “per night”, with no mention of naps, though naps were recorded in this study as part of total sleep. This too could spur needless anxiety and false reassurances. An adult who naps 3 hours a day shouldn’t assume that their health is fine on the grounds that they sleep 7 hours per night.

Part of the problem may be the sweet spot metaphor itself. Originally it meant the best place on a bat or racket to hit a ball. As scholars like George Lakoff would predict, this metaphor may have shaped public impressions of the study. Because a sweet spot is one undivided area, people may have come to view the optimal amount sleep as a single, undivided stretch of time. In fact, the study encompasses one big stretch at night, plus one or more separate ones during the day.

Posts about the study on Reddit, Tiktok, and other social media platforms also reference the sweet spot – sometimes noting a specific number of hours rather than a range – without necessarily mentioning anything about individual differences or naps.

As for AI chatbots, ChatGPT and Claude don’t know about the new study yet. However, when I googled “Is there a sweet spot for sleep?”, here’s what AI Overview told me:

That “generally refers to” is misleading. The 6.4 to 7.8 hour range is drawn from exactly one new study. Prior research and media coverage indicate other ranges, such as 6 to 8 hours and, more recently, 7 to 9 hours.

I wanted to find a news report that cited the 7-to-9 hour range, and so, about five minutes after my previous search, I googled this phrase: “7 to 9 hours of sleep is a sweet spot.” Here’s what AI Overview had to say now:

Oops. As sometimes happens with chatbots, changing the prompt led AI Overview to contradict itself.

Neither prompt inspired AI Overview to clearly acknowledge individual differences in sleep needs or to say anything about naps. This illustrates the degrees of separation problem I mentioned at the outset. Many of the news articles acknowledge individual differences and a few mention naps, but by the time AI Overview worked over these sources, key nuances were lost. One more example of the need for caution when seeking information from AI...

Bottom line

The new study offers us an important reminder: Sleeping too little or too much may be harmful. Translating that observation into practical advice turns out to be tricky though.

In this study, amount of sleep per 24-hour period was measured by simply asking each participant for an estimate. Other research suggests that their accuracy would be moderate at best. People tend to overestimate how much sleep they get by roughly an hour on average (though folks with insomnia may underestimate the total amount).

This is one reason not to get hung up on numerically-defined sweet spots. You might feel better if you get 6.4 hours of sleep per night as opposed to 6.3. But you might feel even better after 6.5 hours. There’s nothing magical about exactly 6.4 hours – or 7.8. These are just inflection points estimated from population-level data that was messy to begin with. Besides, you’re not a population.

We’ve become a very quantified society. It goes against the grain to draw qualitative conclusions from statistical data, but I think that’s exactly what we should do in this case.

If your sleep falls outside the optimal range, you might ask whether the amount you get is suitable for you. But you shouldn’t assume there’s a problem, any more than you should assume that being in the sweet spot proves that your sleep is adequate.

Most important is to reflect on what you experience and how you feel. Do you struggle to fall asleep at night? If you wake up during the night, is it hard to fall back asleep? Do you feel fatigued most mornings? Do you run out of steam later the day?

If your answer to any of these questions is yes, it might be worth taking a closer look at your sleep routines (as well as your diet, caffeine intake, activity levels, stress, etc.) Great resources on managing sleep hygiene and other sleep-related lifestyle factors can be found here, here, and here.

Ultimately the goal is to find a sweet spot that’s defined by how you feel, rather than treating it as one more personal health metric that needs to be optimized.

Thanks for reading – I hope you rest well tonight!

Appendix: Biological Age Gaps

In this study, 23 Biological Age Gaps (BAGs) were calculated for each participant to measure biological aging in different organs and systems.

A BAG expresses the gap between a person’s biological age and their chronological age. For instance, if a person is 42 years old, and the biological age of their lungs is 48, their pulmonary BAG would be +6, meaning that their lungs are aging ahead of schedule.

How would we know the biological age of a person’s lungs? Long story short, by studying the lungs of healthy individuals and determining what normal lung aging looks like, in order to, in effect, create standards for comparison.

The 23 BAGs in this study were drawn from MRI scans, as well as proteins and metabolites found in blood plasma. Structural features and molecular markers from these measures served as proxies for biological aging.

It’s 2024, and as I walk into Kroger I hear an elderly woman call out: “George, don’t forget. I only want natural.”

The woman then raises her cane and gently, if not lovingly, prods the elderly man standing in front of her. (I assume this is George.)

I’m not sure what the woman had in mind, but her request seems to illustrate the appeal to nature – the fallacy that natural things are automatically best.

I get the appeal to nature. It feels right to me in some primal way. But it’s a fallacy that leads to overly strong claims.

Natural things are sometimes best. Fresh apples are healthier than apple juice drinks containing artificial sweeteners and preservatives. But natural isn’t necessarily good for you. Apple seeds release cyanide when eaten, and they’re far from the most toxic substance found in nature.

At the same time, artificial and synthetic food additives aren’t necessarily unhealthy. They’re either FDA-approved or generally recognized as safe (GRAS), meaning, in effect, that regulators and scientific experts have deemed them harmless when consumed within recommended limits.

And then, one morning, scientists disagree.

Since January, a team of French researchers have caused a stir by linking food preservatives and other additives to cancer, type 2 diabetes, and cardiovascular disease. The diabetes and CVD papers were published just last week.

What’s alarming about these studies is that they identify so many culprits. More than a dozen preservatives appear to promote chronic disease.

If the researchers are right, the appeal to nature, at least in this case, isn’t too strong. Rather, it’s not strong enough. Preservatives may be artificial, natural, or synthetic (created in a lab but chemically identical to some natural substance). The French team is questioning the safety of many of them, including natural ones.

These studies have been covered in high-circulation outlets like CNN and MedPage Today, so I expect that sooner or later many readers will be alarmed. But should we be? How closely should we be examining ingredient lists? When I buy groceries, do I have time to watch how George responds to being prodded, or should I focus all my attention on what I’m adding to my cart?

The new preservative-CVD study

This study, led by Anaïs Hasenböhler and senior author Mathilde Touvier, both affiliated with Sorbonne Paris North and Paris Cité, was published on May 20 in European Heart Journal.

Since 2009, Touvier and colleagues have been tracking 112,395 participants in the NutriNet-Santé study. Participants were 42.8 years old on average when data collection began.

In one respect, NutriNet-Santé is a winner: Dietary habits are measured more often than in any long-term observational study I’ve seen. Every 6 months, participants record exactly what they ate during the previous 24 hours. They do this on three non-consecutive days during a two-week period.

The goal of the new study was to determine whether preservative intake would predict who develops hypertension (i.e., high blood pressure) and/or cardiovascular disease (heart attacks, strokes, etc.). Total preservative intake was derived from 58 preservatives. 17 individual preservatives consumed by at least 10% of participants were also examined. In each case, adjustments were made for demographics, family health history, personal health, physical activity, smoking, diet, and more.

Three key findings

1. Total preservative intake was linked to hypertension and cardiovascular disease.

Compared to the one-third of participants with lowest preservative intake, the top one-third were 24% more likely to develop hypertension and 16% more likely to experience CVD.

(Hypertension and CVD were reported among just over 5% and 1% of participants, respectively.)

2. Greater intake of eight preservatives increased the risk of hypertension.

Here are the hateful eight:

3. Greater intake of ascorbic acid increased the risk of cardiovascular disease.

You read that right. Participants who consumed the most ascorbic acid (i.e., vitamin C) in preservative form were 15% more likely to develop CVD than those who consumed the least. This was observed even after taking into account other sources of vitamin C in their diet.

In sum, excessive preservative intake seems to increase the risk of hypertension and CVD. The eight individual culprits include six linked to cancer in a NutriNet-Santé study published earlier this year, while adding ascorbic acid and sodium ascorbate to the list.

Some of the preservatives in these studies are natural, or at least synthetic. This is one more reason not to take the appeal to nature too literally. Natural preservatives may be harmful too.

What should we do?

The CDC’s updated Dietary Guidelines for Americans, released in January of this year, contains exactly one statement about food preservatives:

The NutriNet-Santé data suggests that the CDC’s focus on artificial preservatives is too narrow. However, I agree that we should try to “limit” preservative consumption, and, unfortunately, I don’t see grounds for more specific guidance.

Food labels rarely note preservative quantities, and manufacturers rarely make the information public. The NutriNet-Santé team themselves had to cobble together estimates from a few assays, combined with generic estimates gleaned from publicly available databases. Their published papers offer few details on the math.

Evidently you should try to be among the lowest one-third of people with respect to preservative intake. I’m not sure how to do that, other than to make your diet as preservative-free as possible.

Meanwhile, at the risk of sounding paradoxical, I don’t trust how the findings are interpreted, but I believe they tell us something valuable anyway.

A concern

Something quirky about the NutriNet-Santé studies is that in some cases, folks who consume the most preservatives are mostly likely to develop health problems, while in other cases it’s the middle group with the greatest risk.

For instance, below I’ve plotted the increases in hypertension risk associated with potassium sorbate (in blue) and ascorbic acid (in orange). The higher the dots on this graph, the greater the risk. (These are Hazard Ratios in a fully-adjusted model.)

For potassium sorbate, the more people consume, the greater the health risks. For ascorbic acid, the greatest health risks are seen for moderate intake.

The researchers don’t explain why some preservatives show the first pattern, while others show the second one. They only offer some very ad hoc speculations for a few of the differences.

Why is this a concern? If there’s no physiological explanation for these discrepancies, then what we might be seeing are statistical flukes.

That is, when hundreds of analyses are run on a large sample, some effects are bound to crop up, just by chance. They might show the potassium sorbate pattern, they might show the ascorbic acid pattern. (The Nutri-Santé team reported FDR-corrected p values, but this doesn’t “correct” the problem – it just tends to reduce the number of flukes.)

Confounds can be profound

In spite of the concern I’ve raised, the Nutri-Santé studies consistently show that the one-third of participants with the lowest preservative intake also have the least risk of disease. There’s something special about this group.

What’s special about them? The researchers say it’s low preservative intake, but these folks were healthier in other respects too. They were more physically active and reported eating less ultra-processed food than other groups (though in other respects, such as alcohol consumption, they may have lived less healthfully).

The researchers controlled for these differences, but this was not done especially well. Some variables were crudely measured (activity levels were defined as high, moderate, or low), while others could only be guessed at (e.g., grams of alcohol consumed per day).

My suspicion is that, as in many other studies, people who were healthiest in one respect tended to be healthiest in others. After all, given the options in grocery stores and fast food restaurants, special effort is needed to avoid preservatives. People who do so may be less likely to develop cardiovascular disease thanks to anything or perhaps everything they do to manage their health.

I take this an argument for healthy living, in the holistic sense. Keep an eye on your diet, your activity levels, your stress, and the quality of your sleep. Watch the preservatives too, and limit consumption. But since product labels rarely indicate quantities, you can take a moment to relax and also watch George, as I did.

After being prodded by the cane a few times, he turned to the elderly woman and said, with a smile, “I’ll get you natural if you’ll stop poking me with that damn cane.”

There you go. George got to speak his mind – which is known to reduce stress levels – and the elderly woman, I assume, got her natural food. A healthy outcome for both of them.

Thanks for reading!

One thing I love about science is that when some implausible claim pops up, you never know what you’ll find in the data.

For instance, I heard recently that exercising once per week, for about an hour, reduces body fat. I scoffed. Just once a week? Then I looked at the January study that prompted this claim. The methods were so rigorous I caught myself murmuring wow.

Exercising more frequently is preferable, but if you can manage one high-intensity workout per week without hurting yourself or eating more, you should still see changes.

I don’t know much about exercise science, so maybe this claim wasn’t implausible to begin with. Here’s one that definitely raises an eyebrow: Eating grapes protects the skin against damage from ultraviolet (UV) radiation. Several studies, including one published on May 13, support this claim.

AI already knows about the new study. So do Facebook and Reddit. The possibility of grapes being “edible sunscreen” has been pushed by celebrity doctors like Andrew Weil and, more cautiously, by health organizations such as The Skin Cancer Foundation. But can we trust the data? Should you be eating more grapes now that summer is approaching?

The TL;DR is that fresh grapes may be a healthy food, but claims about UV protectiveness are far more truthy than truthful.

Truthiness in science

Sadly, the final episode of The Late Show with Stephen Colbert airs tonight. 21 years ago, Colbert invented the term “truthiness”, which Merriam-Webster defines as

“a truthful or seemingly truthful quality that is claimed for something not because of supporting facts or evidence but because of a feeling that it is true or a desire for it to be true”

Scientific claims may be more truthy than true for a number of reasons, including the scientists’ own financial interests. This seems to be driving the claim that grapes are a superfood that wards off UV radiation. For roughly five years, a handful of industry-funded researchers have pushed this claim, though their evidence holds up about as well as a grape you’ve just stepped on.

I want to share their newest research with you. It’s a good illustration of financially-motivated truthiness in science, and a reminder of why we need to view health claims cautiously.

The new study

This study, led by Asim Dave (Rutgers) and senior author John Pezzuto (Western New England University), appeared in ACS Nutrition Science on May 13.

The 29 participants in this study consumed a grape solution every day for two weeks. Each day’s solution consisted of 72 grams of freeze-dried grape powder dissolved in water (an amount equivalent to three servings of fresh grapes, according to the researchers). At the beginning and end of the two-week period, select areas of skin were exposed to ultraviolet light, and biological data, including biopsies, were obtained.

Grape consumption had no obvious protective effects. After two weeks there were no changes in the minimum amount of UV light necessary to induce reddening on each person’s skin. But the researchers did report other evidence for improved UV resistance, including signs of less cellular damage as well as changes in gene expression.

At this point, we’d expect to hear that grapes offer some protection against UV radiation. The researchers were not satisfied with such a humble conclusion though. On Monday, senior author John Pezzuto shared the following with Science Daily:

“We are now certain that grapes act as a superfood and mediate a nutrigenomic response in humans... The changes in gene expression indicated improvements in skin health.”

This seems awfully truthy once you look at the data. Here are four reasons for my saying this:

1. A tiny sample.

Although there were 29 participants, data from 25 of them were excluded from most of the analyses, owing to “failed quality control measures.”

What went wrong? The researchers don’t say – but no matter. Would you trust a study with a sample size of 4?

2. Meaningless results.

Each of the 4 individuals showed different genetic responses to study conditions. Most of the results section describes failed attempts to find commonalities, leading the researchers to acknowledge that

“there is no discernible dominant commonality in the shifts of genetic expression following grape consumption.”

How do we get from this observation to the conclusion that eating grapes protects skin health?

Simply put, the researchers ran thousands of analyses and identified a few changes in genetic expression consistent with protective effects (though, again, not all participants showed the same changes).

Oh, and here’s a key detail: For two weeks before the study began, participants ate a special diet that prohibited foods, food products, and supplements containing phytochemicals that promote skin health.

Here we find the seamy underbelly of truthiness.

Suppose I ask you to eat a protein-restricted diet for two weeks, then feed you bacon three times a day? Of course biochemical analyses will show that the bacon has some benefits. Your amino acid levels will improve, for instance, because you’re eating more protein again. This doesn’t show that bacon is good for you. I’ve merely chosen one way to restore something I deprived you of.

3. Unidentifiable mechanisms.

What is it about grapes that would promote UV resistance? The researchers are strangely careful about indicating that it’s the grapes themselves, not substances they contain (e.g., polyphenols) which trigger changes in genetic expression that ultimately affect the skin.

This is deeply pseudoscientific. Grapes aren’t magical. They contain stuff. If eating them improves skin health, it’s because of how some or all of their “stuff” affects our bodies.

Why the swerve into pseudoscience? Drum roll please...

4. Pervasive conflict of interest.

So much about this study became clearer once I realized that it was funded by the California Table Grape Commission, whose self-described purpose is to “maintain and expand markets for fresh California grapes...”

The authors freely admit to competing financial interests: Two of them, including senior author John Pezzuto, serve on the scientific advisory committee of this commission:

No wonder the researchers indicate that grapes themselves, rather than something in grapes, foster nutrigenomic changes that protect the skin. Evidently they would like you to buy more grapes.

Which varieties? Anything from California. The powdered mix used in this study was made from a combination of both seeded and seedless red, green and black grapes provided by the California Table Grape Commission. The exact same mix has been used over and over again in Pezzuto’s studies, all of which have the same flaws (and more) that I’ve described here.

Meanwhile, apart from UV protection, Pezzuto has spent several years pushing the broader claim that grapes are a superfood. One of his 2025 articles is entitled “Are grapes worthy of the moniker superfood?” (His answer: Yes.)

As I mentioned, one reason to question this claim is that Pezzuto serves on the California Table Grape Commission and receives financial support as well as research funding from this group. The commission in turn is not shy about trumpeting Pezzuto’s claims:

The main reason for doubting that grapes are a superfood (however you might define that term) is that the data provides zero support for the idea. This is a truthy claim rather than a truthful one, and money seems to be its raison d’être. (Or, as Colbert might say, raisin d’être.)

Per calorie, grapes are rich in polyphenols and antioxidants, and they’re a moderately good source of vitamin K and manganese. But you can get all of these nutrients from other foods – and in some cases, more – while consuming less natural sugar. In short, grapes may be a healthy food, but they’re not “super” in any sense.

If I seem a bit hostile toward Pezzuto and his team, it’s because they’re pushing a dangerous idea. ChatGPT may be cautious, two of the leading AIs treat the new study as credible, as do a variety of news and social media reports.

Even if we trusted Pezzuto and colleagues’ earlier studies, none of the effects are strong enough to have practical importance. (Suppose three servings of grapes per day offer as much protection as 30 fewer seconds in the sun. Would you still describe grapes as “edible sunscreen”?)

It’s one thing for scientists to call grapes a superfood that gives your skin a youthful glow. That’s just routine sleaziness. But claiming that grapes protect against sun damage is truthy and dangerous. Roughly 1 in 5 Americans are expected to develop skin cancer in their lifetimes, and an estimated 90% of these cancers are caused by excessive exposure to sunlight. Instead of hoping we can eat our sunscreen, we should apply the regular kind, preferably broad-spectrum 30+ SPF, and take other steps to protect against UV radiation recommended by the American Academy of Dermatologists and other scientifically-grounded organizations.

The edible sunscreen concept isn’t the biggest public health challenge on our plates, of course. Vaccination rates are down, ivermectin is making a comeback as a cancer treatment, and trust in government health agencies continues to plummet. In the midst of these challenges, one thing we can do – what I’ve tried to do in this newsletter – is to call out misleading health data whenever we encounter it.

Thanks for reading!

We’re all familiar with silver bullet claims: X is harmful, Y is the only thing that kills X, and Y always works.

In movies, if X a werewolf, Y is the silver bullet. In health scams, X is some disease and Y is the miracle cure.

You’re probably also familiar with what I’d call “bronze bullet claims.” X is harmful, Y is one of the few things that kills or prevents X, and Y tends to work.

This is a little more nuanced. Y isn’t the only thing that’s effective. It doesn’t work all the time. But it’s still pretty special.

This is the stuff of clickbait and social media slop (see above), but peer-reviewed studies are sometimes to blame.

A famous example is the early-pandemic claim that hydroxychloroquine is an effective treatment for COVID-19.

Hydroxychloroquine (plus azithromycin) wasn’t pitched as the only treatment, and it wasn’t expected to be 100% effective. It was just one of a few drugs that reportedly killed the virus.

However, the 2020 study that prompted this claim turned out to be a scam and was later retracted. No other studies have shown that the drug benefits COVID-19 patients.

We hear bronze bullet claims all the time, especially concerning foods and dietary supplements. Acai lowers the risk of cancer. Elderberry boosts immune system functioning. And, according to the study I’ll be discussing, eggs reduce the risk of Alzheimer’s disease.

Words matter. These are bronze bullet claims because of the way the food or supplement is described.

Take acai berries, which are loaded with antioxidants and anti-inflammatory compounds, kill cancer cells in petri dishes, and may protect rodents against cancer.

It’s fine to say that we can reduce our risk of cancer by eating a balanced, healthy diet (”mostly plants, not too much“), including acai in some form. But the claim that acai has special anti-cancer properties is a bronze bullet, because lots of foods are rich in antioxidants and anti-inflammatories, lots of things kill cancer in petri dishes, and studies haven’t directly shown that acai prevents cancer among humans.

So, sure, enjoy your acai product. It’s healthy (assuming no added sugar). Just don’t expect it to ward off cancer all by itself.

This brings me to a new study claiming to show that eating eggs reduces the risk of Alzheimer’s disease.

I wanted to trust the data, because many folks (including me) enjoy eggs enough that we’d gladly eat a few more per week in the interest of promoting brain health.

Sadly, the findings don’t seem credible. What we have instead is another bronze bullet. Eggs are healthy, if eaten in moderation, but they have no special power to ward off Alzheimer’s disease.

I’m not writing this newsletter just to complain about flawed data. The new study is valuable because it illustrates exactly how bronze bullets are forged. Understanding what went wrong here can help us judge the merits of all sorts of health claims.

The culprit turns out to be a simple bias in data analysis.

Recognizing this bias doesn’t require statistical expertise. My hope is that once you’ve seen it, you may be more cautious the next time you hear about the latest superfood or exercise hack or anything else purported to have special health benefits.

The new study

This study is in press at The Journal of Nutrition but available online now.

Lead author Jisoo Oh and colleagues at Loma Linda University tracked 39,498 participants for an average of 15.3 years.

At the outset of the study, the participants, who were 65 or older and had no Alzheimer’s diagnosis, completed a food frequency questionnaire consisting of more than 200 items. Egg intake was classified as follows:

• Never/rarely

• 1-3 times per month

• Once per week

• 2-4 times per week

• 5 or more times per week

The researchers wanted to know how egg consumption influences the risk of Alzheimer’s disease, over and above the influences of age, sex, physical activity, dietary choices, alcohol use, and other risk/protective factors. Statistical analyses were used to adjust for these factors.

A total of 2,858 participants ended up being diagnosed with Alzheimer’s disease. Did egg intake influence the chances of receiving a diagnosis?

Apparently so.

Compared to those who never or rarely ate eggs, participants were 17 to 27% less likely to be diagnosed with Alzheimer’s:

Case closed, right?

Not exactly...

How bronze bullets are forged

I want to skip past the usual concerns one could raise about a study like this (e.g., egg consumption was only measured once) and get right to the bronze bullet issue.

Two terms of art are needed. In the new study, egg intake was the “predictor” variable, and other risk factors were treated as “covariates”, meaning that statistical techniques were used to filter out their influence.

This is one of the powerful things modern statistics does for us. We can estimate the protective effects of egg intake over and above the benefits of being younger, exercising more, eating healthier, and other factors known to reduce the risk of Alzheimer’s disease.

But here’s a key point: Many variables influence Alzheimer’s risk. The researcher must decide which ones to treat as predictors versus covariates.

In this study, participants who developed Alzheimer’s disease differed from their peers in many ways besides egg consumption. For instance, as the researchers themselves noted,

“A greater proportion of Alzheimer’s cases reported no meat or fish consumption...[and a] higher intake of fruits”

In other words, participants diagnosed with Alzheimer’s disease didn’t just eat fewer eggs. They also ate more fruit, and they were more likely to eat no meat or fish.

Now you can see that by merely switching the “predictor” and “covariate” labels, this could have been a completely different study. For example, as per a supplementary online file the researchers provided, the data supports each of the following conclusions:

(A) Consumption of meat and fish reduces the risk of Alzheimer’s disease, after controlling for variables such as fruit and egg intake.

And:

(B) Consumption of fruit increases the risk of Alzheimer’s disease, after controlling for variables such as meat, fish, and egg intake.

See how I shifted variables around? The researchers could’ve drawn conclusion A by labeling meat and fish consumption as the predictor. They could’ve drawn conclusion B by labeling fruit consumption as the predictor. In each case, eggs would be downgraded to covariate status.

Thus, instead of the way Medical News Today, MSN, Fox News, and other national outlets covered the study, we could’ve seen headlines such as these:

These are fake headlines, but they’re just as faithful to the data as the real ones. All of them are bronze bullet claims.

Why did the researchers treat eggs as the predictor?

Well, there’s a small amount of low-quality prior evidence that eggs have neuroprotective effects, but that’s not a strong rationale. We need a good reason for predicting that eggs would reduce the risk of Alzheimer’s disease.

The researchers note that eggs are a rich source of choline, tryptophan, and other brain-healthy nutrients, but other foods contain these nutrients too, and there’s no evidence that anyone in this study was deprived of them. (In fact, participants had very healthy diets on the whole.)

My point is not that eggs are bad for you. They do have neuroprotective effects. But, as far as we can tell, they’re not especially neuroprotective.

Eating a balanced, healthy diet supports brain health. Add to that regular exercise, good sleep, social activities, cognitive stimulation, stress reduction, management of blood pressure and cholesterol, and avoidance of harmful substances, and you’re doing the best you can to stave off Alzheimer’s disease. There’s no bronze bullet here. Eggs only look like one because of a statistical bias.

What should we do?

How can we avoid becoming “casualties” of bronze bullet claims?

If you’re not looking at the actual research, it helps to read or listen to summaries that are detailed enough to indicate whether the effects are unique. Sure, eggs are healthy. So are kale and acai and walnuts and so on, but are they especially and uniquely so?

Also, I suggest keeping in mind a source of bronze bullet claims that motivates the statistical work.

Conflict of interest

Here’s a screen shot from the tail end of the new study write-up:

Hmm. No conflict of interest. Just funding from the American Egg Board.

Last Tuesday, the American Egg Board posted a claim that the new study is consistent with a 2024 study linking egg consumption to reduced risk of Alzheimer’s dementia. The earlier study was even more egregiously bronze bullet-ish, and it was funded by the Egg Nutrition Center.

It took me all of 5 seconds, using a now-prehistoric tool (Google search), to discover that the Egg Nutrition Center is actually the science and nutrition education division of the American Egg Board.

No conflict of interest here. None whatsoever.

I was tempted not to tell you about the industry funding, because it distracts a bit from the fundamental issue. Bronze bullets crop up regardless of who, if anyone, pays for the research. In this case, conflict of interest clearly may have incentivized the focus on one specific predictor variable. In other cases, it’s just bad science.

Bottom line

If you enjoy some food known to be healthy, eat it, in moderation, but don’t expect unique health benefits. Eggs tend to be a good choice. Per calorie they’re an excellent source of protein, choline, vitamin B12, and more. But anything you get from eggs is available from other food sources too (and those who are relatively sensitive to dietary cholesterol should eat eggs sparingly).

Today, May 14 2026, is the 230th anniversary of Edward Jenner’s first administration of the smallpox vaccine. Like a number of other diseases, smallpox soon became preventable in just a few ways: You either got vaccinated, or you had to somehow avoid exposure.

In contrast, when we think of four of the five leading natural causes of death in the U.S. right now – heart disease, cancer, stroke, and Alzheimer’s disease – prevention relies on more than just one or a few strategies. Living healthy calls for a holistic perspective on diet, sleep, exercise, stress, socialization, and everything else that influences the risk of chronic disease. One may hope that some bronze bullet offers protection, but I wouldn’t count on it.

Thanks for reading!

In the late 20th century, the background music in some stores included a voice whispering things like “I am honest. I will not steal.”

Doubts about effectiveness killed off this approach to persuasion. Advertisers, for instance, no longer hide the perceptual details that may have subliminal influences.

A famous example is the orange arrow in the Amazon logo. Not everyone notices that it looks like a smile, or that it extends from a to z (implying comprehensiveness), but the assumption is that these details register anyway and increase the company’s appeal.

Subliminal effects are nuanced and widely debated, but there’s consensus that our brains and bodies can respond to sensory input we’re not conscious of.

This newsletter is about infrasound, which is too low in frequency to be heard but might still be harmful. It’s fine if you don’t notice the Amazon smile or the FedEx arrow (between the E and X), but not so fine if sounds you can’t hear undermine your health.

The topic intrigues me because I’m deaf in one ear and slightly impaired in the other one, particularly for lower frequency sounds. Am I therefore exposed to less infrasound than the typical person? Is my crappy hearing actually a mixed blessing?

Also intriguing – and worrisome – is a new study showing that infrasound increases stress levels. Can we trust the findings, and, if so, how can we protect ourselves? Should we be concerned about new sources of infrasounds such as AI data centers?

What is infrasound?

Imagine standing in a pond and dropping a pebble into the water. The waves will radiate outward until they reach your legs.

Frequency refers to how many waves pass your legs per second, while amplitude refers to the height of the waves.

Now let’s switch to sound. When sound waves reach our ears, higher frequencies are perceived as higher in pitch, while larger amplitudes are perceived as louder. (Technically, loudness is determined by variation in pressure rather than greater height, but you get the idea.)

Infrasounds are too low in frequency to be audible (< 20 Hz), but they could still have large amplitudes. Think of a boulder toppling into the water and creating powerful, slow-moving waves. This illustrates why infrasounds might be harmful. Because you can’t hear them, you can’t tell how powerful they are. You’re not going to wince and move away.

Some natural phenomena, like water and wind, produce infrasounds, but the main source is technology. If you concentrate, you might notice the slight hum of your HVAC unit or the rush of distant traffic, but these sounds also include inaudible frequencies.

At the same time, lower frequencies tend to travel farther than higher ones, so if you’re far enough away from the HVAC unit or traffic that you don’t hear them, your brain may still register the infrasounds they emit. In extreme cases, you might be quite far from the source. Last October, a YouTuber determined that a 2025 Space X launch created infrasounds that traveled over 140 miles.

So, how could these very low-frequency sounds harm us?

1. Infrasounds penetrate deeply into biological tissue and may directly impact cellular functioning. Crudely speaking, the sound waves cause tissues inside your body to vibrate. (My poor hearing for low-frequency sounds offers no protection here.)

2. Infrasounds registered by our inner ears (specifically, the outer hair cells of the cochlea) might trigger neural pathways that increase stress hormones. (In theory I might be somewhat protected from this.)

Prior data on actual harms has been inconclusive.

Extremely high-amplitude infrasound is clearly harmful. The dose makes the poison, as the saying goes.

In more ordinary circumstances, evidence of emotional effects and sleep disturbances has been met by counter-evidence that infrasounds are generally benign (see for instance this April blog post by physicist Andy Masley).

This is where the new study comes in. Unlike most of the prior research, the study is experimental, focuses on humans, and relies on a biological measure of stress. It’s way more rigorous than earlier work on potential harms.

An infrasound experiment

This study, published in Frontiers in Behavioral Neuroscience on April 26, was led by Kale Scatterly and colleagues at MacEwan University (Edmonton, Canada).

I love this study. It illustrates that when researchers are careful enough, they can gather meaningful data from small samples and simple procedures.

The participants consisted of 36 undergraduates. (Not large, but adequately powered.)

Each participant provided a saliva sample, listened to a 5-minute audio clip, then provided another saliva sample.

Why saliva? Because the samples were later analyzed for cortisol levels. Cortisol is a hormonal biomarker of stress – generally, higher cortisol means greater stress.

Each participant listened to one of the following audio clips:

• Calming music plus infrasound.

• Calming music without infrasound.

• Unsettling music plus infrasound.

• Unsettling music without infrasound.

The infrasound was designed to simulate ordinary conditions – around 18 Hz, with amplitudes that wouldn’t be detectable in any respect.

(If you’re curious, you can listen to the “calming” and “unsettling” music here and here.)

Key findings

Audio clips containing infrasound caused a significant rise in cortisol levels. (The increase was roughly 20% overall.)

Here we can see the power of a carefully-designed experiment.

1. Unsettling music increased cortisol more than calming music did, but, regardless of music type, the presence of infrasound increased cortisol too.

In other words, the sounds you hear can be stressful, but so can the sounds you can’t hear.

2. Infrasound increased cortisol levels even after taking into account conscious reactions to the music (participants rated the appeal of the audio clip and how they felt after hearing it).

In other words, the sounds you can’t hear may affect your stress levels more than you realize.

What should you do?

1. Relax.

My purpose in sharing this study with you is to alert you to a potential source of environmental stress, not to claim, overconfidently, that we’re all under siege. We can’t tell from this study what the long-term impacts of infrasound might be. Perhaps our brains adjust, perhaps not. Meanwhile, we know so many simple, healthful ways of reducing any stress it might cause (see here, for instance).

2. Don’t measure.

Microphones that accurately and reliably record the full range of infrasound start at around $1,000 apiece, and even with these devices, measurement is tricky. The slightest turbulence in the surrounding air will change the reading, for instance.

Even if you did use one of these fancy mics, it’s not clear whether it would be worth, say, relocating your bed in order to reduce infrasound exposure, because we have no idea yet how much of a change would be beneficial.

3. Consider some minor adjustments.

If you’re still concerned about infrasound, you can make some adjustments that may reduce your exposure, and possibly even cut out some of it altogether.

For instance, you could ensure that where you sleep and/or sit for long periods is as far away from compressors as possible, including the one in your fridge and any in your outdoor HVAC unit. Ditto for washers and dryers, dishwashers, desktop computers, window AC units, and ceiling fans, all of which emit infrasounds.

You can also purchase inexpensive anti-vibration pads and place them under fridges, washers and dryers, etc. Moving appliances out of corners and ensuring that they don’t touch walls would help too, as the walls act as resonators that increase the amplitude of infrasound waves.

Again, it’s not clear yet how impactful any given reduction in infrasound would be. You can make these adjustments, but you might get more stress-reducing benefits from exercising more, socializing, improving your sleep, and so forth.

AI data centers

The health impacts of AI data center noise are hotly debated. I consider it a public health problem simply because people who live close enough to these centers complain about the noise. At minimum their mood and stress levels are affected.

AI data centers also produce a lot of high-amplitude infrasound from fans, pumps, and other elements of their cooling systems. There’s no consensus yet on how much infrasound pollution they create or how hazardous it is.

I almost never use AI when writing my newsletters, but here’s one place where I found it useful. Currently around 1,500 new AI data centers are under construction in the U.S. Claude and ChatGPT helped me estimate that somewhere between 2 and 3 million people will live close enough to hear sounds emitted by these data centers. Slightly further away, millions more people will be exposed to infrasounds. It’s hard to tell how much to trust these estimates (and, again, exposure doesn’t guarantee harm). but the number of people whose health might be affected is quite large.

Obviously we have more pressing concerns about the impact of AI, but we shouldn’t overlook the known impacts of audible noise pollution as well as the uncertain impacts of infrasounds. NIMBY might seem like a good strategy, but when it comes to super low-frequency sounds, your backyard may be larger than you realize.

Final thought

Uncertainty is scary. We try to manage it – efforts to do so gave birth to modern statistics – and if it seems overwhelming, we fall back on confident, simplistic assertions. Infrasound is dangerous. No, infrasound is benign.

I’d rather live with uncertainty than oversimplification. I wrote this newsletter because the new study, along with other evidence, suggests that infrasound can cause stress, and thus reducing our exposure to it might be worthwhile. That’s a disturbing idea, but not one that we should consider proven just yet.

Thanks for reading!

“Persons who are naturally very fat are apt to die earlier than those who are slender.” (Hippocrates, Aphorisms, c. 400 BC.)

Broadly, modern science agrees with Hippocrates. Excess weight increases the risk of a variety of health problems and tends to diminish longevity.

We also recognize critical nuances. For instance, visceral fat, which accumulates around organs, is more dangerous than the subcutaneous fat that builds up under the skin.

Because exercise preferentially reduces visceral fat, cardiorespiratory fitness predicts longevity better than body fat does. As the saying goes, fitness beats fatness. Overweight people who exercise regularly tend to be healthier than slender, sedentary people.

Along with knowing more than Hippocrates and other traditional authorities, our perspectives on health have been “statisfied”, or transformed by modern statistics.

Measurements of body fat are a good example. Scientists and clinicians no longer describe people as “naturally very fat”. The phrase is much too vague.

What we have instead are at least 20 different statistically-grounded measures or proxies for body fat, the most famous being the body mass index, or BMI.

The BMI turns out to be a blunt instrument, at best. In this newsletter I want to discuss its limitations and explore whether an alternative called the body roundness index (BRI) is more useful. At the end I’ll touch on a new study suggesting that the BRI may not represent much progress.

The body mass index (BMI)

The BMI is a sort of proxy, or stand-in for body fat. It equals your weight divided by your height squared.

You can use an online calculator to determine your BMI. If you want to figure it out by hand, the standard formula for the metric system (height in meters, weight in kilograms) is:

In the English system (height in inches, weight in pounds), a conversion factor must be added:

For example, if you’re 5 feet 4 inches tall, and you weigh 140 pounds, your BMI would be: 140/(64 x 64) x 703 = 24.03.

Among scientists and clinicians, BMI is the most widely used proxy for body fat, and it’s easy to see why: The calculation is simple, cheap, non-invasive, and built from data that’s easily obtained. (I can’t recall ever reading a study on long-term health outcomes such as cancer, cardiovascular disease, and longevity that didn’t include BMI data.)

BMI is popular among the general public for the same reasons, and because BMI values are routinely translated into categories:

Experts disagree about how to adjust these categories according to age, sex, and ethnicity, but categorization is common, and that alone is problematic.

For instance, the difference between a BMI of 24.8 versus 25.2 is negligible, but only the latter is considered “overweight”, and it might trigger different advice from a health care provider, higher life insurance premiums, lower discounts on employer-sponsored wellness programs – not to mention anxiety about weight.

What else is wrong with the BMI?

Being “statisfied” is a mixed blessing. The BMI may be better than weight alone at predicting health outcomes, but it’s far from ideal.

For one thing, the BMI fails to reflect how fat is distributed. A person with a BMI in the normal range may have abdominal obesity – a relatively large amount of “belly fat” – which increases the risk of cardiovascular disease and other health problems. Somewhat independently from that, the BMI doesn’t distinguish visceral from subcutaneous fat. And there’s more:

• The BMI makes no distinction between fat and muscle. Since muscle is roughly 18% denser than fat, the BMI tends to misclassify relatively muscular people as overweight or obese, while underestimating BMI among people with relatively low muscle and bone density.

• The formula for BMI, which squares height in the denominator, creates overestimates for especially tall people and underestimates for especially short ones.

These limitations are illustrated by a pair of famous athletes:

At the peak of his career, Arnold Schwarzenegger was 6’ 2” and 235 pounds. This corresponds to a BMI of 30.2 and therefore a classification of “obese”. As a professional bodybuilder, however, his body fat would’ve been extremely low.

At the other end of the spectrum is the Olympic medalist Sha’Carri Richardson. Sprinters have low body fat (though not as low as bodybuilders) and so their BMIs usually fall within the “normal” range. Richardson is an exception because she’s relatively short. At 5’ 1” and 99 pounds, her BMI is 18.7, or borderline underweight. As with Schwarzenegger, this reflects a problem with the BMI rather than with her weight.

Misclassification is not limited to special cases. A 2016 study of over 40,000 adults found that 31% of those with a BMI in the “normal” range were metabolically unhealthy with respect to blood pressure, cholesterol, serum glucose, insulin resistance, and other key variables. At the same time, roughly half of “overweight” people, and 16% of those classified as “obese”, were metabolically healthy.

In short, the BMI may create false reassurances or needless anxiety. Even warnings about the fallibility of the BMI sometimes fall short. The Harvard Medical School web page that provides a BMI calculator tells you simply that “If your BMI is above 25, losing weight is a smart idea.”

In fact, if your BMI is above 25 and you’re muscular, losing weight might not be smart. Regardless of your musculature, a higher-than-normal BMI should merely prompt you to consult with a medical professional. What’s not smart is to treat BMI cut-offs as one-size-fits-all categories.

Over the past few years, calls to abandon the BMI, or at least rely less heavily on it, have intensified. For instance, in 2023 the American Medical Association adopted a new policy acknowledging the BMI’s limitations and recommending alternatives, or at least not using the BMI as a standalone measure.

One of the alternatives getting a lot of attention recently is the body roundness index.

The body roundness index (BRI)

The body roundness index (BRI) was developed in 2013. Instead of using height and weight, the BRI formula relies on height and waist circumference.

Intuitively this makes more sense. BMI only tells you something about mass. If you want a proxy for body fat, it seems preferable to know something about a person’s shape – specifically, their roundness.

You can calculate your BRI by hand (see below), but it’s quicker to use one of the online calculators.

So, how to measure waist circumference?

You can plug in the waist size of pants that fit you comfortably, but it would be more accurate to use a soft tape measure. Just be sure to stand, exhale, then wrap the tape around yourself snugly but comfortably.

You’ll be accurate enough if you wrap the tape about an inch above your belly button. But, if you want a “perfect” measurement, press your fingers (gently) into your side, and find your lowest rib as well as the upper ridge of your hip bone. Your “waist” will be halfway in between.

BRI values range from 1 to 20. The higher the value, the rounder the body. (A 2024 study found that the national average among U.S. adults is 5.62.)

The exciting news is that in head-to-head comparisons carried out over the past six years, the BRI has outperformed the BMI in predicting health outcomes such as cardiovascular disease, cardiometabolic risk, metabolic syndrome, depression, and all-cause mortality.

Oh, and then there’s Arnold Schwarzenegger. (He said he’d be back.) When he was Mr. Universe, Schwarzenegger was 6’2”, 235 pounds, and had a waist size of around 34 inches. The BMI would misclassify him as obese, but his BRI would’ve been 2.6, a value that would place him, accurately, in the “very lean” category, as per the following system:

Notice this table includes references to “average” rather than BMI terms like “normal” and “overweight.” Because the BRI is so new, there’s no consensus yet on how to interpret different score ranges.

The table shown here, which has begun circulating among health organizations and websites like Yale Medicine, is drawn from a 2024 JAMA Network Open study of 32,995 American adults. That study reported quintiles. In other words, “high” refers to the 20% of people with the highest BRI values, “very lean” refers to the 20% of people with the lowest BRIs, and so on.

This brings me to new research that tries to better understand how to divide BRI values into meaningful categories. After all, it’s not inherently important whether a person is in the third or fourth BRI quintile. That’s just a statistical distinction. What we want to know is where to draw the line between healthy and unhealthy BRI values.

I’ll discuss this study briefly, then step back and ask whether the BRI represents genuine progress.

The new study, led by Shengming Wang (Tianjin Medical University), appeared in Frontiers in Endocrinology on March 31.

Wang and colleagues addressed a relatively straightforward question: Among 6,858 people ranging in age from 35 to 75, would BRI be linked to whether a person had experienced a physician-diagnosed stroke or not?

The answer was yes.

First, the researchers found that people with higher BRI scores were more likely to have had a stroke than people with lower scores. ROC analysis showed that a BRI of 4.6 worked best as the cut-off value, meaning that it was most accurate at distinguishing who did vs. did not have a stroke.

Wang and colleagues didn’t stop there, given the possibility that people with higher BRI scores also have other risk factors for stroke, including older age, more alcohol use, hypertension, high cholesterol, et cetera.

After adjusting for these risk factors, Wang and colleagues found that BRI values still helped improve accuracy in predicting stroke. Specifically, people with BRIs above 4.6 were 76% more likely to have had a stroke than people with values below 4.6.

Sounds impressive, right?

Don’t grab your tape measure just yet…

More of the same

The new findings are preliminary at best, because there were so few cases.

Among the 6,858 participants, 192 had been diagnosed with a stroke. This consisted of 2.1% of the participants with BRIs below 4.6 versus 4.2% of those with higher BRIs.

In other words, even without adjustment for various risk factors, having a higher BRI only increased the absolute risk of stroke by 2.1%.

I don’t mean to say that’s nothing. But it’s not much, and the actual difference is likely to change as more data are gathered.

Unfortunately, this study illustrates why an accumulation of such data may lead us down the same rabbit hole that the BMI did.

1. BRI is no more of a measure of visceral fat than BMI is. Both are correlated with adiposity, but in the end they’re just proxies. In this study, the predictive power that the BRI added to the other risk factors (an AUC increase of .004) was nowhere near large enough to have any practical value.

2. Like the BMI, the BRI formula yields misleading values. Since the denominator grows with the square of height, the taller the person, the more impact their height will have on the calculated value. This will cause the roundness of very tall people to be underestimated – some of them with too much belly fat might still obtain BRI scores in the normal range. (For the same reason, BRI scores will overestimate the roundness of very short people, leading in some cases to needless concern.)

The BRI is an improvement over the BRI, but I don’t think it can do much more than what the BMI currently does, which is to help predict health outcomes in large populations. This is one way that being “statisfied” is useful. We have tools that allow us, for the first time in history, to quantify medical risk factors among groups.

At the same time, you personally are not a group. Neither the BMI nor the BRI work well at predicting health outcomes among individuals. This hasn’t stopped either one from becoming trendy personal health datapoints. The BRI is already popular on TikTok, where you’ll find lots of accolades for its supposed merits, as well as anxiety among those worried that they’ve become too round.

For individuals, the BRI should just be a dollop of information to ponder, and maybe to discuss with your health care provider. I realize that’s not the most exciting conclusion, but in the end there are limits to how much a single number can tell you about your health.

Thanks for reading!

I run 4 or 5 times per week, and I can assure you there’s no ideal time for it.

The weather is always too hot. Except when it’s too cold. Or it’s dark outside. Or I’m tired, or busy, or...

I ignore these inconveniences and run anyway.

Some people say I’m disciplined. The truth is that I’m stubborn. I love running and refuse to be distracted. Most afternoons I lace up my shoes, no matter what, and just go.

A study published last week suggests that I should be more flexible. The researchers found that exercise is most beneficial when it’s aligned with your chronotype.

In other words, if you’re a morning person, exercise in the morning. If you’re a night owl, exercise later in the day. The cardiometabolic benefits include lower blood pressure and LDL cholesterol, better sleep, and more gains in fitness.

I’m definitely a morning person. Have I been running at the wrong time? When is the best time of day for any of us to exercise?

The answer turns out to be more nuanced than what the new study tells us (not to mention what you might’ve heard about it from Good Morning America, U.S. News, the BBC, or one of the other news outlets that covered it).

Chronotype

A chronotype is one’s natural tendency to be asleep or active at certain times of day.

Morning people, or “larks”, tend to wake early and feel most energetic, mentally and physically, from the morning through early afternoon.

Evening people, or “owls”, tend to wake later (if their schedule permits) and feel most energetic in the late afternoon and evening.

Not everyone is so easily classified. Chronotype is normally distributed, which is to say that “larkishness” or “owlishness” is a matter of degree, and most people are found somewhere between the extremes.

If you’re lucky, your chronotype aligns with your work schedule and other obligations. Not everyone is lucky. “Social jet lag” occurs when a person’s chronotype is chronically misaligned with their daily schedule. This is almost always discussed in the context of owls forced to get up early for work, but exercise is another domain in which it can be seen.

For instance, a lark might feel most like exercising in the morning, while an owl might prefer that late afternoon or evening workout. But people can’t always exercise when they want. If we can, we might want to choose the healthiest option. The new study explores whether alignment between chronotype and time of day matters.

The new study

This study, led by Arsalan Tariq (University of Lahore) and colleagues at the Universities of Edinburgh and Ibadat, was published last week in the BMJ journal Open Heart.

Participants ranged in age from 40 to 60 and had no major health problems other than at least one risk factor for cardiovascular disease (prehypertension/stage 1 hypertension, overweight/obesity, sedentary lifestyle, impaired fasting glucose, and/or a family history of premature CVD).

These are Pakistani adults, but it’s hard to imagine a sample of greater relevance for Americans. More than three-quarters of us have one or more of those risk factors.

The researchers determined chronotype by means of the Morningness-Eveningness Questionnaire (MEQ), a widely used tool, and by 48 hours of continuous temperature monitoring. (Larks show predictable rises, peaks, and falls in body temperature earlier in the day than owls do.)

Participants came to the University of Lahore five days per week for a total of 12 weeks. Each 40-minute exercise session at Lahore consisted of supervised, moderate-intensity walking.

Exercise sessions were offered in the mornings (8 to 11 a.m.) and evenings (6 to 9 p.m.). And here’s a critical detail:

• Half the participants were assigned to sessions that aligned with their chronotype. (Morning people exercised in the mornings; evening people exercised in the evenings.)

• The other participants were assigned to sessions that conflicted with their chronotype. (Morning people exercised in the evenings. Evening people exercised in the mornings.)

Main findings

1. By week 12, everyone improved on every outcome variable: Blood pressure, LDL cholesterol, quality of sleep, fitness, etc.

No surprise there. Regardless of time of day, regular exercise is good for you.

2. By week 12, the extent of improvement was significantly greater for the chronotype-aligned group than for the chronotype-misaligned group.

In other words, people benefitted most when the timing of exercise matched their chronotype, regardless of whether they’re larks or owls.

The table below shows how much people gained on average from alignment. (”Benefit of alignment” is the chronotype-aligned mean minus the chronotype-misaligned mean.)

On the face of it, these are dramatic findings. Exercising at the time of day that suits your chronotype improves cardiometabolic health and fitness more than exercising at the “wrong” time does.

Some qualifications

If you’re exercising at a time that doesn’t match your chronotype, you might benefit from an adjustment, but I wouldn’t feel pressured to change anything.

Why not?

1. Small payoffs.

The benefits of chronotype-consistent exercise seem small, especially considering that people needed 12 weeks of 40-minute walks, 5 days per week – plus any walking to and from the lab – in order to achieve them. (In the table above, none of the changes would be considered clinically meaningful, other than that 5.3 mmHG drop in systolic blood pressure.)

I don’t mean to downplay the benefits. I’m just not sure they call for changes to anyone’s schedule.

Keep in mind too that changing your schedule isn’t an isolated act. An owl who exercises in the evening and then eats a heavy meal at 10 p.m. may be undermining their sleep and cardiometabolic health. Ditto for the lark who gets up too early to exercise.

I run in the afternoon, around the time my window of peak energy is starting to close. Sometimes I feel alert, sometimes a little drowsy. But I don’t plan to start running after breakfast, because mornings are when I get the most accomplished. I wouldn’t want to stay up late finishing tasks I could’ve done more quickly and pleasantly earlier in the day.

2. Confounds.

Chronotype is reflected in physiological states we rarely notice, like slight shifts in body temperature, as well as a whole bunch of stuff that we are aware of, like how alert and energetic and motivated we feel.

Exercising at a chronotype-inconsistent time can feel mildly unpleasant, like you’re doing a chore. You probably don’t exercise with as much intensity.

Thus, this study might not show that chronotype-consistent exercise per se is beneficial. Rather, what’s best is exercising when you enjoy doing it, because you’re investing more energy and reaping greater rewards. Enthusiasm rather than timing may be the critical variable.

There’s plenty of evidence that people exercise more – and more intensely – when they like what they’re doing. Chronotype-aligned exercise may be desirable only because that’s when we tend to be most enthusiastic about it – which is to say that exercising at the “wrong” time of day for you might still be just as good if you enjoy the activity.

A suggestion…

I wouldn’t call myself a poster child for healthy living – cookies are never safe in my presence – but I do recommend my attitude toward running, which is to ignore anything that diminishes the pleasure of the activity.

I don’t psych myself up for a run. I just focus on the sensations and the scenery and tune out the noise.

(Of course, you shouldn’t ignore the kinds of pain that signal a developing injury. I’ve tried that, and it rarely turns out well. Also, if you run, don’t ignore stop signs, potholes, and other people.)

And a final thought…

For years, scientists (and everyone else) debated the merits of morning vs. afternoon exercise.

Attention to timing-chronotype alignment, rather than focusing exclusively on time of day, represents progress – even if it turns out that enthusiasm is the key variable.

Still, we don’t want to lose the forest for the trees. Regular physical activity is healthful, regardless of whether the conditions are ideal. Almost anything that gets your heart pumping and strengthens your muscles will be good for you (see the CDC guidelines, for instance).

Diversity in physical activity is worthwhile too, as I discuss here. If you can coordinate exercise with your chronotype, great, but it’s not essential. Just find your enthusiasm, and, as James Brown put it, get up offa that thing!

Thanks for reading!

Is there any essential nutrient that’s not marketed as a health hack?

Consider vitamins – the 13 organic compounds your body needs but can’t produce on its own.

The vitamin products sold at your local grocery store and CVS were originally meant to correct deficiencies. Now all of them are marketed for optimizing health too.

The fact that we need a minimum amount of X to survive doesn’t prove that the more X we eat, the healthier we’ll be. Perhaps more is better, perhaps not.

This week I came across a new study suggesting that higher vitamin D levels help prevent the onset of Alzheimer’s disease. This is a “more is better” study. If the findings can be trusted, we ought to increase our vitamin D levels beyond conventionally recommended values.

I do think we can trust the data. But I doubt most of us would actually benefit from increasing our vitamin D intake.

The study turns out to be a perfect example of how even scientists fall prey to a “more is better” bias. What the results show is that vitamin D deficiency may contribute to Alzheimer’s disease. It’s not clear that increasing serum levels of the vitamin beyond a minimum threshold lowers risk.

Some context

In the early 20th century, when vitamins were first discovered, the focus was on preventing disease, not on making healthy people healthier.

Nutrition scientists back then weren’t prone to “more is better” thinking. For instance, in an influential 1921 manual, Walter Eddy had this to say about a recent fad for eating yeast cakes, considered high in the “B vitamine”:

“Many people are reporting beneficial effects from their use. I feel that the very fact that benefits have resulted from this yeast feeding is excellent evidence of lack of the vitamine in the diets of the people affected...”

In other words, if yeast cakes are beneficial, it’s because people are deficient in vitamin B, not because the more B we eat, the more we benefit.

Unfortunately, from the beginning, food and supplement manufacturers, as well as clinicians, overextrapolated from the science, implying that if a little is essential, then more is better.

Most commonly, claims about minimum thresholds were jumbled together with claims about optimization.

For instance, ads for Schlitz’s Vitamin D beer, sold from 1936 through 1938, note, accurately enough, that “You will enjoy better health if you have an adequate supply” of vitamin D. But there are also references to how the vitamin helps build “vigor, vitality, strength and energy” and some statements implying that the more you take, the better your health.

(Vitamin D beer: Drinking a lot of it may strengthen your bones, but you’re also more likely to fall down and break them...)

Since the 1970s, vitamin research has also begun to explore optimization, thanks in particular to antioxidant studies, and to Linus Pauling’s mostly-discredited claim that massive doses of vitamin C can prevent diseases ranging from colds to cancer.

You might say that scientists have caught up to the food and supplement manufacturers (as well as the TikTok influencers), so consumers now hear “more is better” messages from sources varying widely in credibility.

Why are these messages compelling?

The “more is better” bias

The “more is better” bias is simply the assumption that consuming more of some nutrient yields greater health benefits (until reaching the point of toxicity).

This is a natural way of thinking that stems from a lifetime of simple dose-response observations. The harder you throw a rock, the farther it travels. The more salt you add to the soup, the saltier it tastes.

Manufacturers of supplements and food products leverage this way of thinking. They describe vitamins as “supporting” or “promoting” the health of specific organs and systems, and wherever possible they imply that greater consumption makes you healthier, though FDA regulations generally prohibit them from saying so directly.

Vitamins do “support” health, in the sense of being necessary for normal functioning. Exceeding minimum thresholds may or may not be beneficial.

Vitamin D

Vitamin D in particular supports bone health, muscular strength, immune system functioning, and more, mainly by helping maintain healthy calcium and phosphate levels.

Since there’s no consensus yet on how much vitamin D people need, there’s no agreement on how many people are deficient, though deficiencies clearly occur.

One reason is that the primary source of vitamin D is sunlight (specifically, UVB radiation), but some people spend too much time indoors, live in places that aren’t very sunny, use excessive amounts of sunscreen, and/or have more melanin.

The other reason is that most foods contain little or no vitamin D (exceptions include fatty fish and fortified milk, as I discuss in the Appendix).

So, one reason you might consider taking vitamin D supplements such as pills or cod liver oil is the possibility that you’re deficient (though, again, it’s not clear exactly how much you need – an issue I’ll revisit at the end).

Another reason is studies like the one I’ll be discussing now.

The new study

This study appears in the June issue of Neurology Open Access and has been available online since April 1.

Martin Mulligan (University of Galway, Ireland) and colleagues at Boston University and Harvard tracked 793 dementia-free participants in the Framingham Heart Study (Generation 3 cohort) for an average of 16 years.

Mulligan and colleagues wanted to know whether serum vitamin D levels at the outset of the study would be associated with amyloid and tau deposits in the brain over a decade later.

Let me unpack that last sentence:

“Serum vitamin D levels” refers to the amount of vitamin D circulating in the bloodstream. Following standard practice, the researchers measured 25(OH)D, a compound that our bodies convert into the biologically active form of vitamin D.

“Amyloid deposits” are proteins that can accumulate around brain cells in the form of plaques, causing inflammation and impaired communication between cells.

“Tau deposits” are structural proteins inside brain cells that can detach and aggregate in neurofibrillary tangles, causing cell dysfunction and death.

Amyloid plaques and neurofibrillary tangles are key contributors to Alzheimer’s disease, which is currently experienced by roughly 1 in 9 Americans over the age of 65.

At the outset of the study, participants were 39 years old, on average – young enough that if the results hold up, vitamin D supplementation might be worth starting decades before Alzheimer’s disease typically appears.

Each participant received an imaging test called a positron emission tomography (PET) scan for either amyloid or tau deposits.

The relationship between vitamin D levels and amyloid or plaque deposits was then calculated after statistically adjusting for a range of demographic, health, and medical variables, as well as the season during which vitamin D was measured (levels tend to be higher in sunnier months).

Main findings

People with higher vitamin D levels showed fewer tau deposits overall, as well as fewer deposits in the areas of the brain where tau pathology first crops up in Alzheimer’s disease.

Vitamin D levels were not associated with amyloid deposits.

What do the findings tell us?

I trust the data – I found no serious methodological or statistical flaws ­­– but I question how the researchers framed it. Here’s a typical example:

“[H]igher serum 25(OH)D in early midlife is associated with lower tau deposition on brain-PET a mean of 16 years later. This study suggests that higher vitamin D levels in midlife may offer protection against pathologic tau deposition in the brain...”

These are “more is better” conclusions. They’re based on regression analyses showing tiny associations between vitamin D levels and tau deposits (standardized beta coefficients ranged from -.012 to -.024).

Words cannot convey how tiny these associations are, statistically speaking. But that’s not my main concern.

Bear with me for three paragraphs and I’ll explain why “more is better” conclusions aren’t justified here.

Mulligan and colleagues divided their sample into two groups: People who had sufficient serum vitamin D levels vs. those who didn’t. (A standard cutoff of 30 ng/mL was used.) The groups did not differ in tau deposits.

Next, “post hoc exploratory analyses” – technical jargon for “fishing around” – revealed that people with the highest vitamin D levels (e.g., the top 20%) had significantly fewer tau deposits than people with the least vitamin D (e.g., the lowest 20%).

This difference between the extreme groups is the most likely explanation for the tiny associations overall between vitamin D levels and tau deposits. The “high” groups were not unusually high in vitamin D; the “low” groups were low by any standard.

In sum, I would not conclude that higher vitamin D levels protect against Alzheimer’s disease. Rather, having too little vitamin D increases the risk of developing the disease. Once you achieve some minimum, you’re protected as well as anyone else is.

Shades of Walter Eddy and his "B vitamines"...

Practical advice

Experts disagree about how much vitamin D people need, at minimum, and whether exceeding the minimum can optimize our health.

The authors of the new study indicate that optimization reduces the risk of Alzheimer’s disease.

I think a safer conclusion is that deficiencies increase risk. This is consistent with prior evidence linking supplementation to reduced risk of acute respiratory infections, cancer mortality, multiple sclerosis relapse, and atrial fibrillation.

One thing is clear at least: Individual needs vary. The FDA recommends the following:

If you are an older adult, have dark skin, or are exposed to insufficient ultraviolet band radiation (such as sunlight), consume extra vitamin D from vitamin D-fortified foods and/or supplements.”

Notice that the FDA avoids saying how much sunlight would be sufficient, or how much extra vitamin D one should take.

Cautiousness here is understandable, given how widely expert recommendations vary. The Endocrine Society advises that healthy adults under 75 don’t need to take vitamin D supplements or even be tested for vitamin D levels in the first place. Supplements are only recommended for children, adults 75 and older, pregnant women, and people with high-risk diabetes. (This 2025 post from Dr. Greg Katz illustrates a similar perspective.)

If your doctor informs you that your vitamin D levels are low, consider waiting until your next checkup before making a decision about supplementation, as measures of serum vitamin D are fallible.

Finally, keep in mind that exposure to sunlight is the primary source of vitamin D. Expert advice about sun exposure varies widely (see Appendix for details), and so I have no confidence in highly specific recommendations. Rather, in the spirit of Michael Pollan, here’s my advice: Go outside. Get some sun. Not too much.

Thanks for reading!

Appendix: Practical thoughts on sunlight, diet, and supplements

Sunlight is the primary source of vitamin D, but experts disagree about how much is needed to achieve adequate serum levels. Anywhere from 5 to 30 minutes per day (without sunscreen) has been advised, but suitable minimums need further tailoring given variability from one person to the next in diet, health, melanin, and the UVB intensity of sunlight they’re exposed to. (The parts exposed to the sun matter too. Thinner skin on the arms, legs, and torso tend to absorb more UVB radiation.)

With respect to diet, fatty fish are an excellent source of vitamin D. 3 ounces of salmon, for instance, provides about about 81% of the daily value (DV) for an average adult (though you want to avoid overconsumption, given heavy metal content).

Meanwhile, 3 ounces of anything else would not exceed 10% of the DV unless it were fortified (or, in the case of mushrooms, irradiated by UV light). For instance, you’d have to drink 8 ounces of milk to obtain roughly 10-19% of the DV, depending on brand. (As for dairy products like yogurt and cheese, non-dairy milks, and other foods, some brands are fortified, some aren’t. You have to read labels.)

If you prefer supplements, a tablespoon of cod liver oil exceeds the DV for vitamin D. Other supplements vary widely. Caution is needed, because some of these products exceed the recommended daily maximum for long-term use (e.g., 100 mcg, or 4,000 IU). Several brands have as much as 50,000 IU and are therefore intended for weekly rather than daily consumption. Even then, sustained use is problematic.

I’ll never forget the first time I helped administer electroconvulsive therapy (ECT).

I was fresh out of college, working part-time at Butler Hospital in Providence, Rhode Island.

What I recall most vividly is not the procedure but the effects on the patient, who I’ll call “Irene”. (The next several paragraphs are drawn from one of my April 2025 posts.)

Irene had been institutionalized for depression for most of her adult life. She came to Butler in her mid-50s and spent her days in bed, almost unable to speak.

One afternoon, not long after Irene’s third treatment, I was walking across the ward and noticed a patient I’d never seen before waving at me.

As I approached I realized it was Irene. “Hi Ken”, she said. “I’m feeling so much better now...” Her transformation was so dramatic that for a moment I was the speechless one. The following week she went home.

Recently, I googled Irene. Although many patients who benefit from ECT subsequently relapse, she seems to have been among the fortunate.

According to her obituary, Irene lived into her late 80s, enjoying gardening, volunteer work at her church, and spending time with her family – all hints of a dramatically better quality of life after leaving the hospital.

What I learned from Irene is that when we try to help the sickest people, the end may justify the means.

Nobody knew then why ECT helps some people after other treatments fail. We still don’t know. All we can say for sure is that the procedure seems barbaric. In essence, electrical current is delivered to the patient’s brain, causing a brief seizure.

What ensues is sometimes referred to as a neural “reset” or “reboot”, as if we were restarting a computer.

The reality is that ECT is more like carefully hammering the computer on a desk.

Still, it’s not hard to find people like Irene. When all else fails, some sort of “hammering” alleviates their pain, at least temporarily.

However repugnant ECT might seem, and however risky the procedure (as many as half of recipients experience long-term memory impairment), I would not want to see it discarded as a treatment of last resort – e.g., a form of support for “treatment resistant” depression and related conditions.

I’d say the same about other mental health treatments that strike at least some people as off-putting, including AI therapy and the use of psychedelic drugs.

I tell people I’d never share my personal problems with a chatbot, nor would I rely on hallucinogens to cope, but that’s just bluster. Deep down I know I’d try just about anything if I were suffering badly enough and I couldn’t find relief.

This week I’ve been reflecting on the need for diversity in mental health care, thanks to a new study showing that a psychedelic compound I’d barely heard of dramatically reduces treatment-resistant depression.

I took a close look at this study. I wanted to know how miraculous this compound might turn out to be.

The new study

This study, published in JAMA Psychiatry two weeks ago, was led by Wiesław Cubała (University of Gdańsk, Poland) and a research team that included senior author Michael Thase (University of Pennsylvania).

Cubała and colleagues looked at how well a compound called mebufotenin (”meh-bu-FO-teh-nin”) alleviates treatment-resistant depression among adults.

Some background

Mebufotenin is found in the venom of the Sonoran Desert toad and has a roughly four-decade history of use for psychedelic purposes. As an entheogen (a substance used to create mystical states of consciousness), mebufotenin derived from plants has been used in South America for thousands of years.

Researchers have begun to explore the use of mebufotenin in treating depression. The compound is a serotonin agonist, meaning that it makes neural receptors more receptive to serotonin (as opposed to simply increasing the length of exposure to serotonin, the way antidepressants like SSRIs do). Mebufotenin binds especially strongly to 5-HT1A receptors and produces additional mood-stabilizing effects.

Mebufotenin also has an extremely useful quirk.

Unlike most psychedelics, the effects of inhaling mebufotenin are immediate and short-lived. Participants in the current study tended stayed “high” for around 9 to 14 minutes.

Thanks to the transience of its effects, mebufotenin has an advantage over other psychedelics for use in clinical settings such as routine psychiatric appointments.

(Although drugs like psilocybin, LSD, and ayahuasca are showing promise as potential aids to therapy, their effects tend to be noticeable for hours, even if microdosed. This limits their potential for outpatient use. A psychiatrist can’t send a hallucinating client home once their hour is up.)

Study methods

Cubała and colleagues tested the effects of inhaling synthetic mebufotenin, known as GH001.

Participants consisted of 81 adults with treatment-resistant depression.

“Treatment-resistant” doesn’t mean that patients have tried everything. Rather, in this study, it means that they’ve taken suitable doses of at least two antidepressant medications for at least six weeks apiece with little or no benefit. (Psychotherapeutic support is also assumed. However, there’s no guarantee that a variety of approaches and therapists have been tried.)

The experiment was carried out in three steps.

First, participants completed the Montgomery-Åsberg Depression Rating Scale (MADRS) and other measures of mental health. MADRS scores range from 0 to 60, with higher scores indicating more severe symptoms.

Next, participants used a vaporizer to inhale mebufotenin or a placebo three times.

Finally, the mental health measures were readministered 2 hours after inhalation, 2 days later, and then 8 days later. Follow-up assessments were conducted 6 months later.

Main findings

In the short term, mebufotenin was dramatically effective – as effective as one could reasonably expect given the tenacity of treatment-resistant depression.

The mebufotenin group experienced sharp declines in symptom severity – an average of over 15 points on the MADRS. Improvement was seen at 2 hours, 2 days, and 8 days after inhalation.

You can see this pattern in the figure below. The orange dots, representing the placebo group, never depart much from zero, meaning that their depression scores barely changed. The dark blue dots, representing the mebufotenin group, immediately drop. At the 8-day mark they’re still 15 points lower than they were at the outset of the study.

The figure above shows group data. Individual patterns of change were quite dramatic as well:

—A reduction of at least 50% on the MADRS was defined by the researchers as a “response” – i.e., meaningful improvement.

—A score of less than 10 was defined as a “remission” – i.e., the person no longer exhibits signs of depression.

In the figure below, the vertical bars show the percentages of each group who achieved response or remission. (People are double-counted if their response is so strong that their MADRS score drops below 10.)

Virtually nobody in the placebo group (the orange bars) showed a response or remission. However, most of the mebufotenin group fell into one or both of these categories. For instance, on day 8, 58% of these participants were in remission.

I wish I could stop here. A 58% remission rate following a single treatment would be miraculous indeed. But does this effect endure?

Among those in the mebufotenin group who’d been in remission at day 8, only 13% were still in remission at 6-month followup.

Dang.

13% is better than nothing. These individuals received no additional mental health treatment during that 6-month period. But the 13% figure tells us that mebufotenin is not going to be a one-and-done treatment for most people.

In sum, more than half of people with treatment-resistant depression who took mebufotenin improved substantially. The improvement lasted more than a week but, for most people, less than 6 months.

The mebufotenin miracle

The truth is that we have no miracle drugs for mental health conditions like depression, if “miracle” is taken to mean something that instantly heals everyone who takes the drug.

There are mysteries, at least. I am thinking of people like Irene, who suddenly recover from severely debilitating conditions and manage to stay well. Perhaps the ECT cured her. Perhaps the medicines she’d been taking finally kicked in. Perhaps her brain had been slowly healing itself, and neither the drugs nor the ECT mattered. We’ll never know.

The title of this post – “The Mebufotenin Miracle” – was intended in an ironic but not disparaging way. What’s miraculous about mebufotenin is that it seems to genuinely help some people who don’t respond to other treatments, at least temporarily. For them, the drug is miraculous – just as ECT may have been for Irene.

At the moment this is the only kind of miracle we can expect from treatments for severe mental illness. The key question is how to leverage the strengths of each kind of treatment, and tailor them to the individual patient. If mebufotenin is used, an additional question is whether repeated doses can safely maintain the benefits.

Closing thoughts

The study I described here is the newest successful phase 2b mebufotenin trial. Larger studies are in the works and, if the data continue to be encouraging, FDA review will follow.

I expect to see FDA approval in the near future. Effectiveness and safety data have strong, and the side effects, though common, are mild (e.g., in the new study 43% of those who inhaled mebufotenin experienced mild nausea, 20% experienced paresthesia, etc.).

Meanwhile, there are plenty of Irenes who need support. Around one third of Americans diagnosed with major depression have a treatment-resistant variety (estimates vary depending on how “treatment-resistant” is defined). That’s about 2.8 million adults per year.

If inhaling a psychedelic compound derived from frog venom turns out to help at least some of these people, then I’m all for it. In my opinion, anything that safely lifts the fog of depression is, by definition, miraculous.

Thanks for reading!

I love the sounds of the ocean at night. And wind whispering in the trees. And the thrum of a steady rain. These are the soundtracks for some of my deepest, most refreshing sleeps.

Apparently I’m not the only one who reacts this way. A recent survey found that 38% of Americans play background sounds while they sleep (including 49% of Gen Z respondents). Sleep sound apps alone are a $2 billion global industry.

If you enjoy the ambience but don’t live near an ocean or forest, there are also streamable audio files, wearables, pillows fitted with speakers, and machines that produce natural sounds or a steady whisper of white or pink noise.

Some of these technologies are marketed for daytime use, such as meditation. Mostly they’re pitched as sleep aids. They do look soothing, though I’m not sure “howling wind” would work for me.

To be honest, I steer clear of these technologies. Being the child of an alcoholic, I worry (somewhat irrationally) about becoming overly dependent on them. But I’ve always assumed they’re healthful. I assumed that any sounds a person finds pleasant will tend to improve their sleep.

This week I discovered two new studies that made me rethink this assumption.

Pink noise

Informally speaking, pink noise is the sound made by a steady rainfall, wind rustling through trees, or waves continuously breaking on the shore.

Technology can recreate these sounds, or provide an ongoing hum that approximates the scientific definition of pink noise (equal energy per octave). Here, courtesy of Wikimedia Commons, is an example:

Most people find pink noise more pleasant than white noise, which has a slightly harsher sound. White noise, embedded below, is more of a hiss than a hum:

No sounds are perfectly white or pink, but in practice white noise (equal energy per frequency) is roughly what you’d hear from a traditional fan, vacuum cleaner, or TV that’s not receiving a signal. (On YouTube, at least one TV static video has nearly half a million views. TV static?)

Some people use pink noise to block out traffic sounds, noisy neighbors, or other auditory distractions. Others have quiet bedrooms but just seek better sleep.

So, we might ask whether pink noise improves sleep when your bedroom is noisy, and when it’s quiet. The new studies address both questions.

These studies represent the work of Mathias Basner at University of Pennsylvania, a leading researcher on the effects of environmental noise on sleep.

The new data

Basner is first author on one of these studies, published in Sleep this February, and a co-author on the other one, published in Communications Medicine this January and conducted at University of Gothenburg, Sweden.

To keep things simple, I’ll call these the January and February studies.

In each study, healthy adults with no reported sleep disturbances spent several consecutive nights in a sleep lab. Each participant occupied a private room and left the next morning to pursue their normal daily activities, the only restrictions being no caffeine, alcohol, or naps during the day.

On the first night, the room was kept silent. On subsequent nights, the participant slept while a speaker played recordings of pink noise, environmental noise, or a combination of the two.

Noise levels mostly stayed below 50 decibels (what you might hear from a ceiling fan at low speed or a light rain falling outside). Meanwhile, polysomnography was used to record information related to sleep, and each morning participants performed cognitive tests and completed surveys.

Not exactly a hotel experience, but not uncomfortable either. Sleep study participants I’ve talked with say that the rooms feel a little weird, or hospital-like, but that they sleep pretty well once they get used to the wires.

Here are a few more details on the studies:

One advantage of this kind of study is that each participant is compared to themselves. How did the different noises affect each person, relative to that first night spent in silence? Another advantage is the use of objective measures as well as self-report surveys. If sleep quality changes slightly, objectively speaking, do people notice?

Key findings

1. Environmental noise undermines sleep.

In both studies, environmental noise caused more awakenings and reduced time spent in the deepest, most restorative stage of sleep. The January study also identified unfavorable metabolic changes the next morning (higher concentrations of lactic acid, leucine, and acetone in the bloodstream).

Participants in these studies also reported poorer sleep in the environmental noise condition, and the February study identified negative impacts on self-reported fatigue and mood the next day.

(Nothing new about these findings, but it’s troubling that the effects were seen at relatively low levels of environmental noise (≤ 65 decibels). We don’t know whether people would adjust after exposure periods longer than one week.)

2. Pink noise may partially offset the effects of environmental noise.

Both studies showed some benefits of mixing pink noise with environmental noise.

In the January study, combining the two reduced (but did not completely wipe out) the disruptions of sleep and metabolic response created by environmental noise.

In the February study, 50-decibel pink noise reduced some of the sleep disruptions of environmental noise. There were no benefits at 40 decibels. And, there was some bad news: Adding either amount of pink noise to environmental noise slightly increased the amount of time spent awake at night.

As for subjective impressions of sleep quality and fatigue the next morning, adding pink noise to environmental noise had no impact.

In sum, pink noise offsets some of the negative effects of environmental noise, while exacerbating one of its negative impacts, but people don’t really notice.

I wish I could tell a simpler tale, but the data reveals even more nuances than I’ve shared here. As a strategy for dealing with low-to-moderate environmental noise, pink noise seems to be beneficial in some ways, innocuous in some ways, and slightly harmful in at least one way.

3. Pink noise by itself may have mildly negative effects.

For the most part, pink noise in an otherwise quiet room was innocuous. No effects on most aspects of sleep, metabolic functioning, or cognition the next morning.

However, in the January study, participants reported slightly poorer sleep and slightly more fatigue following a night of pink noise (compared to a quiet room). This was not found in the February study.

In the February study, pink noise significantly reduced the amount of time spent in REM sleep, a type of sleep that’s especially critical to memory consolidation, emotional regulation, etc. The effects were small and not found in the January study.

In sum, inconsistent findings across the two studies hint that although pink noise by itself is harmless overall, there may be slightly negative impacts. More research is needed.

Practical implications

Media coverage of the February study focused on potential harms, but, as you’ve just seen, pink noise was mostly innocuous – and had some benefits as well.

In particular, if you’re bothered at night by traffic sounds, loud neighbors, etc., pink noise can diminish the effects. You may get a better night’s sleep, even if you don’t quite duplicate the experience of a totally quiet bedroom.

It would also be worth trying to make your bedroom quieter. In the February study, earplugs outperformed pink noise at reducing environmental noise effects.

If your bedroom is already quiet and you practice good sleep hygiene, but you’re still not sleeping as well as you’d like, pink noise might be worth a try. Just keep in mind that if even it makes you feel good, one possible effect (reduced REM sleep) is unlikely to be noticed.

Keep in mind too that some people respond well to other kinds of noise. For instance, if white noise can be described as a hiss, while pink noise is more of a hum, brown noise presents an even deeper rumble, like a strong waterfall or rolling thunder, and some people respond well to it. A number of smart phone apps, like SimplyNoise, allow users to customize the “color” that best suits them.

This brings me to the most important practical takeaway, and one of the reasons these studies intrigued me.

You’re not an average

These studies only report group data (primarily means). There’s no way of telling how many individuals differed from the overall trends.

(It’s reasonable to expect that some did, because mean differences between conditions were often small, and variability around those means was evident.)

This illustrates one way science has been statisfied, or transformed by modern statistics. Scientists in some fields pursue data analyses in a conventional way, seemingly without considering the value of alternative analyses (which aren’t necessarily unconventional).

For instance, in these particular studies, were there people who generally responded well to pink noise, both alone and in combination with environmental noise? Did pink noise diminish everyone’s REM sleep, or just that of most people? There’s no way of knowing from the means.

This is one reason I think pink noise is worth trying if your bedroom is noisy, or if you practice good sleep hygiene but just aren’t sleeping as well as you’d like.

Group averages tell us that pink noise can be helpful in some ways, innocuous in others, and may have negative impacts too. But you’re not an average. We should remember this whenever scientific data is translated into practical health advice.

Thanks for reading!

The nutrition research that gets the most attention typically focuses on what to eat.

For instance, a study published last week claimed that the MIND diet promotes brain health. The national news took note.

With headlines like these, it’s easy to forget that healthy eating is not just reliant on types (and quantities) of food.

Studies show that the timing of meals matters. So does how fast you eat, the order of nutrients consumed, and your particular gut microbiome. These findings are less catchy than dubious claims about bamboo being the next superfood, or the keto diet curing schizophrenia, or the MIND diet keeping your brain younger.

In fact, on March 13, just four days before the MIND article was published, a different study showed that the risk of breast cancer is lowered by greater dietary diversity. That’s important, assuming we can trust the data. But, as far as I can tell, not a single national news outlet covered this study.

Why not? I suspect it’s because “dietary diversity” sounds abstract. We tend to have more concrete concerns about what we’re eating. (Will I get enough protein at breakfast? How bad is pizza? Are the new Dietary Guidelines for Americans overemphasizing red meat and full-fat dairy?)

Streetlight effects

Excessive focus on what we eat illustrates a streetlight effect – i.e., the tendency to seek answers in the most accessible places.

The name of this bias comes from the old joke about a guy searching for something where the light is best, rather than where he lost it.

Naturally, understanding how diet affects brain health requires looking at what we eat. That’s what the new MIND study does. But the researchers also succumb to streetlight effects, analyzing meaningless differences in the intake of individual foods when so much else about our diets genuinely impacts brain health.

The MIND diet

The new study looks at whether the MIND diet slows the physical deterioration of the brain that normally occurs with age.

This diet was developed in 2015 by researchers who wanted to optimize food choices for brain health. For the most part, MIND is a combination of the Mediterranean diet and the Dietary Approach to Systolic Hypertension (DASH). Here’s the essence of MIND:

“Servings” are defined as fixed quantities (e.g., for leafy green vegetables, one serving would be 1 cup raw or ½ cup cooked). This limits the practical value of MIND, as what constitutes a healthy serving depends on your size, activity levels, and metabolism.

The new study

This study, published last week in the Journal of Neurology, Neurosurgery, & Psychiatry, was led by Hui Chen and colleagues at Zhejiang University, senior author Changzheng Yuan (Harvard), and Debora Melo van Lent (Boston University; UT San Antonio).

The institutional affiliations are impressive. Last year, Zhejiang University was ranked #1 in the world for research output. (Harvard ranked #3.) But, as we’ll see, prestige doesn’t guarantee credible data.

Chen and colleagues tracked 1,647 participants for roughly 7 to 14 years. These individuals, middle-aged or older, completed food surveys and underwent at least two brain scans (MRIs).

Chen and colleagues also adjusted for demographic and health variables (e.g., age, sex, education, BMI, chronic disease) as well as for lifestyle habits (smoking, activity levels, etc.).

Main finding

As you’d expect, participants’ brains aged over time, mostly due to cell death and a loss of synaptic connections between neurons. (This is group data and may not apply to each individual.) As Chen and colleagues put it:

“Over a median follow-up of 12.3 years, participants experienced declines in total brain, grey matter, white matter and hippocampal volumes...”

The amount of decline was related to how closely participants hewed to the MIND diet. In other words, the more peoples’ diets resembled MIND recommendations (see table above), the less their brains atrophied as they got older.

Sounds exciting, right? Eat more leafy green vegetables etc,, and you’ll be sharper when you’re 70 or 80.

A caveat

I’m fine with this study if the conclusion is that, broadly speaking, better diets promote healthier brains. That’s a sensible message, and one that’s consistent with prior research linking MIND and the Mediterranean diet to lower rates of dementia.

Unfortunately, Chen and colleagues didn’t draw such a generic conclusion. (It would be hard to publish the paper if they did.) Their bottom line was much narrower. For instance, they stressed that every 3-point increase in MIND score reduces physical deterioration in the brain by roughly 20%, or about 2.5 years of reduced brain aging. That’s a strong and specific claim.

Should you therefore track what you eat and try to raise your MIND score?

It wouldn’t hurt, but I see no clear evidence that increasing scores by a few points would make a genuine difference. I say this because of how crudely eating habits were measured.

Two or three times during the course of the study, participants completed a Food Frequency Questionnaire (FFQ) asking them to estimate how much of 140 foods they ate over the past year. Here are some example questions:

Obviously this is a crude way of measuring eating habits. (What is a “medium serving”? How would I know my “average” intake for an entire year?).

Also crude was the calculation of MIND scores. Each person received a score of 1, 0.5, or 0 depending on whether they fully met, partially met, or failed to meet MIND recommendations for each of the 15 food groups I listed earlier. (Each person’s total MIND score thus ranged from 0 to 15.) Here again, accuracy is questionable. I eat a fair amount of berries. Possibly more than 1 serving per week, on average. But I have no idea whether my average meets MIND’s >2 serving-per-week criterion.

Statistical precision can’t remedy this kind of squishiness. If you ask people every three or four years to estimate their “average” intake of various foods, you just won’t end up with accurate measurements of how much they eat.

I’m happy to generalize about why a MIND score of 14 or 15 is preferable to a score of 0 or 1. But I’m skeptical that a 9, say, is going to reliably correspond to a healthier diet than a 6.

Oh, and Chen and colleagues acknowledged two rather embarrassing findings:

• Eating more whole grains was associated with more brain atrophy.

• Eating more cheese was associated with less brain atrophy.

These findings directly contradict two recommendations of the MIND diet (as well as prior research).

Chen and colleagues didn’t attempt to explain the contradiction. I think what happened is a combination of bad measurement and streetlight effects. The researchers went looking for associations between specific foods and brain health, but that’s generally not the best place to look.

What else is important

Here are four examples of why the healthiness of a meal depends on more than just the types and quantities of food eaten.

(i) Timing

For a given number of calories, eating more at breakfast as opposed to dinner promotes healthy weight loss and better metabolic functioning. Earlier lunches and dinners have similar benefits, as do longer overnight windows between meals.

(ii) Speed

A number of studies show that eating more slowly creates greater feelings of fullness and a healthier metabolic response.

(iii) Sequencing

Studies have also shown that eating proteins (as well as fats and/or fibers) before carbohydrates improves metabolic response following the meal.

There’s direct as well as indirect evidence that over time, the metabolic effects of timing, speed, and sequencing influence the risk of chronic disease as well as the rate of cognitive decline.

(iv) Individual differences

So far I’ve shared generalizations that don’t apply equally well to everyone. For instance, what counts as an “earlier” or “later” meal depends on individual characteristics such as chronotype. 8 p.m. might be a late dinner for an early bird, but an early dinner for a night owl.

People vary widely in their metabolic responses to foods. A 2020 study showed that even identical twins differ in glycemic and lipidemic responses to the same meals. People also differ in genetically-based responses to specific nutrients, such as salt and sugar. And if that’s not complicated enough, any one person’s metabolic response to food will vary with changes in age, health, activity levels, mood, sleep patterns, and gut microbiome.

Practical takeaway

Healthy eating does require us to pay attention to what we eat and how our bodies react. But when we eat, how fast we eat, and our own particular physiology and lifestyle are important too.

One of the practical messages I glean from the science is that, if possible, we should try to eat healthier without stressing about the process.

Consider the relationship between snacking and mood. Studies show that being in a good mood makes us slightly more likely to over-consume unhealthy snacks (ice cream, cookies, etc.). But when we’re in a bad mood, this effect is more common among people who have been restricting their eating in some way.

In other words, restricting what you eat leaves you susceptible to unhealthy snacking on those occasions when you’re in a bad mood. (And what better way to end up in a bad mood than by restricting your eating?)

My point is not that we should avoid adjusting our dietary routines. I think it’s good that I don’t eat the way I did as an undergraduate (e.g., pizza and Coke for breakfast). Rather, we should adjust flexibly and avoid extremes.

Final thought

Streetlight effects aren’t limited to nutrition research. HHS recommendations for physical activity focus on the number of minutes people should spend each week on aerobic exercise and muscle-strengthening activities. But a study I reviewed this January showed that apart from quantity, the diversity of activities influences long-term health as well.

Similarly, in the midst of debates about the health impacts of alcohol consumption, a study I discussed last week emphasized that drinking more than one’s personal average may be harmful, even if intake is minimal and there’s no binging.

Exercise diversity and personal drinking averages are somewhat abstract concepts. It’s easier to just count minutes of exercise or numbers of drinks. But maybe the key to being healthier is not always taking the easiest path…

Thanks for reading!

Americans, particularly Gen Z, are said to be drinking less than ever. This has become a sort of meme – one that I find encouraging, having grown up with an alcoholic father.

The data tell a more nuanced tale.

The most often-cited source is a Gallup poll conducted every year since 1938. Here we see that regardless of year, more than half of Americans drink. Consumption rises and falls, but the downward trends aren’t all that dramatic.

To illustrate, compare Gallup’s data on cigarette smoking (lower left) to the data on alcohol intake (lower right) for roughly the same time period.

How do you read these figures? I’d say that smoking has steadily declined. Drinking has fluctuated, with only a slight decrease overall (though the post-pandemic dip has hurt restaurant owners, for instance, whose profit margins rely heavily on alcohol sales).

What about Gen Z? Gallup’s 2025 data show that 50% of Americans ages 18-34 drink on occasion, as compared to 56% of older people. Not a huge difference.

More dramatic are increases in how many younger people consider even light drinking to be unhealthy (see figure below). Evidence of long-term risks has indeed been accumulating. The Surgeon General’s 2025 advisory, for instance, concludes that with respect to cancer, there’s no safe amount of alcohol intake.

Not all experts agree.

There is consensus that, beyond a certain point, the more you drink, the greater your risks of cardiovascular disease, cancer, and other health problems. The data are murkier with respect to light drinking, for reasons I discuss in a recent post and in the Appendix below.

This week’s newsletter is about a different approach to understanding the health impacts of light drinking.

Instead of tracking participants across decades, asking them every few years to describe their drinking behavior and documenting health outcomes, alcohol intake is more precisely measured, and the short-term effects are recorded.

Dozens of studies have relied on this approach. After drinking in the evening, what are the effects on sleep? How well do we function the next day?

As you can probably guess, when people drink to the point of feeling hungover the next morning, the quality of their sleep suffers, and throughout the day they show impairments in attention, memory, motor performance, and mood. Driving drunk may be dangerous, but so, to a lesser extent, is driving the day after tying one on.

As for light drinking (no hangover), sleep is still affected, but effects may not be discernible the following day. Some studies show no impairments in cognition, driving, or mood.

These are mostly small-sample lab studies. What happens in the lab may not generalize to real-world settings. And, small samples make it difficult to compare groups (e.g., older vs. younger drinkers, or males vs. females).

The study I’ll be discussing solves both problems through wearable technology – specifically the WHOOP, a fashion-friendly, wrist-worn device (see above). The WHOOP records data 24/7 without requiring the user to do anything.

The new study

This study, published on March 9 in PLOS Digital Health, was led by Dr. Gregory Grosicki and an international team representing Indiana University, Baylor, the University of Lausanne, and the Mayo Clinic.

(Side note: The study was funded by WHOOP, Inc, where Grosicki and most of his co-authors are employed. Should we be concerned about conflicts of interest? I don’t think so. The study makes the WHOOP look good only in the sense of showing that it can monitor variables like heart rate and activity levels – as can a lot of other devices. The purpose of the study is to show how drinking affects those variables. There’s no pattern of findings that would make WHOOP look especially good. This is different from, say, a drug study funded by the drug’s manufacturer.)

The sample consisted of 20,968 WHOOP Strap 4.0 wearers. For about 8 months, on average, participants noted in a daily journal whether they drank alcohol the previous day, how much they consumed, and the timing of their last drink. WHOOP gathered the rest of the data.

Each participant’s average alcohol intake was calculated. Grosicki and colleagues found that on any given day, the more a person’s alcohol intake exceeded their personal average, the worse the impact on their cardiovascular functioning, sleep, and activity levels the next day.

You can see these patterns in the figures below. In case you’re reading this on a phone, here’s what you’re seeing:

The figure on the upper left shows that the more peoples’ drinking exceeded their personal averages, the higher their heart rates while asleep. Drinking more caused their hearts to work harder during a natural recovery period.

The figure on the upper right shows that as drinking increases, heart rate variability declines. (This often occurs when resting heart rate is elevated.) Crudely speaking, drinking more alcohol reduces how flexibly your heart rate adjusts to changing demands.

The figure on the lower left shows that the more people drink over and above their personal average, the shorter their sleep, while the figure on the lower right links more drinking to diminished activity levels the following day.

(The cardiovascular effects duplicate those found in other studies. Less consistent across studies are the results for sleep and activity levels.)

In sum, drinking more than usual has undesirable effects, even when a person is not drinking to excess. None of the effects were large, though they were stronger for biological females than males, and for younger people compared to older ones (no differences were observed beyond age 50).

Practical implications and advice

One night of drinking more than usual may not have lasting consequences. You experience minor physiological effects, then you recover.

Problems may arise if you continue to drink more. These problems tend to reinforce each other: Poorer cardiovascular functioning undermines sleep; poorer sleep undermines cardiovascular functioning, and so on. Over time your health suffers.

In other words, as you begin to drink more than usual, you may touch off a cascade of changes that increase the risk of cardiovascular disease, cancer, and other health problems.

Of course, the more you exceed your average intake, the higher that average becomes. But this study is not about an adjustment process. Compare the following individuals:

• Leah: Drinks one glass of wine per day, every day.

• Lee: Drinks nothing on weekdays, then a total seven glasses of wine on the weekends.

Lee and Leah both average one glass of wine per day. But Leah never exceeds that average, while Lee spends two days per week drinking more. The new study suggests that Lee will be more negatively impacted, at least temporarily.

Keep in mind too that sticking to personal averages isn’t healthy per se. Someone who drinks six glasses of wine per day, every day, clearly increases their risk of a variety of health problems.

My father demonstrated that kind of consistency. He was sober all day, and then, every evening, he would pour a Vodka Collins (mostly vodka), and follow up with four more before passing out. I can’t recall a single evening where he deviated from his average. The result was high blood pressure and death from an aneurysm at 92.

What to make of the fact that he lived so long, in spite of the alcoholism – not to mention heavy smoking and a love of processed meats?

You might say that with a healthier lifestyle he would’ve lived to be 100.

That’s a hypothetical though. I’d rather view my father as illustrating that what’s at stake for any of us is not outcomes but probabilities. Drinking too much increases the risk of various health problems. When you develop those problems, if ever, is more or less uncertain.

So, here’s the practical advice I glean from recent data, including the new study:

1. If you enjoy alcohol, drink moderately.

I can’t define “moderately” very precisely, but, as discussed in the Appendix, one or two drinks per day doesn’t clearly increase health risks for most people.

(I say “most” because total abstinence is best for those who have certain health conditions, take certain medications, and/or have a history of Alcohol Use Disorder.)

2. If you drink moderately, avoid excessive variability.

Binge drinking is dangerous, but even among those who don’t binge, the new study hints at slightly better health outcomes among people who are more consistent in how much they drink. People who are a little more similar to Leah than to Lee, in other words.

Final thought

Last month I reviewed a study suggesting that quantity and diversity of physical activity are independently beneficial.

In other words, exercising more may be good for you, but so may engaging in a greater variety of exercises for the same amount of time.

It’s easy to think about how much you exercise, because that’s a concrete number of hours or minutes. Diversity, though more abstract, turns out to be important too.

Similarly, the amount you drink is a concrete value. Variability with respect to that value is more abstract, but it’s still important. The new study calls for moderation – not just in how much we drink, but in how often we stray from our usual habits.

Thanks for reading!

Appendix: Is light drinking unsafe?

Experts disagree about the health risks of light (aka “moderate”) drinking.

For instance, some researchers claim that even one alcoholic drink per day is unsafe, while others profess uncertainty or consider such conclusions to be overly cautious.

I wouldn’t blame anyone for cautiousness, but I do want to stick as closely as possible to the data. As it turns out, the most relevant data are inherently limited.

Studies that link drinking to long-term health outcomes track tens or hundreds of thousands of participants over a period of decades. Typically, participants are asked every few years to describe their recent drinking behavior.

This approach to measurement is hopelessly crude. “Drinks” vary in size and alcoholic content. People don’t necessarily finish their drinks. In any case, they may not remember exactly how much they consumed. These are just a few of the many methodological issues.

Perhaps the biggest limitation of the research is that consumption is usually treated as a fixed quantity. A person who is classified as taking three drinks per week, for instance, retains that label for the duration of the study, even if it lasts for decades.

Few people are so consistent. When I was younger, I was sometimes a “2-to-3-drink-per-week” person; nowadays it’s more like “1-to-2 drinks per month.”

My point is not that we can’t trust the studies. Rather, what we can’t trust is the precision of their estimates.

Even though the uncertainties are known, they sometimes get lost in translation. For instance, last month The New York Times, a generally reliable source of science journalism, reported that

“For women, regularly having one drink a day increases the odds of developing breast cancer by 10 percent.”

This is needlessly scary. I checked, and the sources for this widely-aired claim draw upon the Nurses’ Health Study, in which women are asked every couple of years to describe their alcohol intake over the previous 12 months. Even assuming perfect recollection, the options provided are unsatisfactory. For instance, “1-2 drinks per day” is one option. But what if you drink exactly one glass of wine per day? What if it’s a small glass only filled halfway? What if you drank more in the winter than in the summer?

It’s clear from Nurse’ Health Study data, and from other sources, that beyond some point, the more women drink, the greater their risk of breast cancer. The data just don’t show that one drink per day increases risk, much less increasing the odds by 10 percent.

I’ve seen dozens of long-term observational studies relying on traditional methods (e.g., cohort and case-control designs) as well as new ones (Mendelian randomization), and I’ve read a number of meta-analyses, but I haven’t found persuasive evidence for a cut-off – i.e., a specific amount of alcohol intake that divides safe consumption from amounts that increase health risks.

Without getting further into the weeds, I can say at least that below two drinks per day (assuming a very flexible definition of “drink”), the data look especially murky, unreliable, and/or extrapolated.

Alcohol is clearly dangerous – the more you drink, the greater the risk. But the dose makes the poison. For most people, light drinking doesn’t appear to increase health risks.

I feel sorry for Bryan Johnson.

Johnson, you may know, is the biohacker who spends millions of dollars a year on the quest for immortality. (See, for instance, the 2025 Netflix documentary Don’t Die.)

At 48, Johnson does look youthful. Here he is at Paris Fashion Week a few days ago:

Alas, not even makeup fully conceals the telltale signs of aging. Look closely at the photo and you can see crow’s feet (due to loss of collagen and elastin), and a slight puffiness under the eyes (due to loosening of the orbital septum, a membrane that holds fat in place).

Even so, Mr. Johnson will sell you his “secrets”, ranging from $1 million longevity protocols on down to $35 “ceremonial grade” matcha.

I feel sorry for the man, because he either truly believes we can live forever, or he knows better but happily profits from the claim. In other words, he’s either deluded or a grifter.

If that sounds harsh, it’s because I’m a little nervous about my own credibility here. I’ll be talking about a new study from Yale researchers suggesting that the physical and mental effects of aging can be temporarily reversed.

The “secret”? Developing more positive attitudes about aging.

Now you can see why I’m a little nervous. It’s sounds unscientific to say, in effect, that positive thinking can temporarily reverse the aging process.

All the same, I trust the Yale study. It’s not perfect, but it offers encouraging news, as well as a simple but clever tweak in the analysis of health data.

A little background before I get to the encouraging part.

Biological aging

We’re all familiar on some level with the distinction between chronological and biological age.

For instance, the photographs below, taken this year, show two men born in 1954. Chronologically, they’re the same age, but one man looks older, biologically speaking.

Biological aging is not just about changes in facial appearance, nor is it a uniform process. Cells, organs, and organ systems deteriorate at different rates. Biologically speaking, a person’s heart may be older or younger than their brain.

But what exactly is biological age?

In the Appendix I offer a technical discussion. Here I’ll just note that biological age is a way of estimating someone’s health risks based on the people they’re most biologically similar to in some respects.

Suppose you’re 36 years old, and test results say that your biological age is 41.

What this means is that your risk of developing some disease, or dying, within a certain period of time is the same as that of 41-year-olds in a large dataset.

The good news is that risks are not outcomes. Whether you actually develop the disease or die during that time period is a matter of genetics, lifestyle, and luck. Live a healthier lifestyle and your biological age – i.e., your risk – might drop.

Biological age is often determined by measuring DNA methylation patterns (see Appendix), but this is just one way of framing what happens when we age. You can’t feel your DNA methylating. But we do notice declines in physical and mental functioning, and these changes also predict future health.

This brings me to the new study.

This study, published on March 3 in Geriatrics, was led by Yale researchers Becca Levy and Martin Slade.

Levy and Slade tracked 11,834 older individuals (average age 68) for a period of up to 12 years. The focus was on three variables:

1. Cognitive health.

Participants completed a 27-item version of the Telephone Interview for Cognitive Status (TICS) at the beginning and end of participation in the study.

The TICS measures factual knowledge, recall memory, and simple math skills. The questions are relatively easy and, together, function best as a screener for mild cognitive impairment or dementia.

2. Physical functioning.

Participants’ walking speed was measured at the beginning and end of participation in the study. (The instructions are simply to walk “normally”.)

Walking speed is a narrow but informative measure of physical functioning. It has been called “the sixth vital sign”, because it predicts future health outcomes ranging from hospitalizations to mortality.

3. Attitudes toward aging

At the outset of the study, participants completed the Attitude Toward Aging (ATA) subscale of the Philadelphia Geriatric Center Morale Scale.

The five items on the ATA measure a person’s attitudes toward their own experiences with aging. For example, participants rate the extent of their agreement, from “strongly disagree” to “strongly agree”, with statements such as “The older I get, the more useless I feel.” Other questions pertain to happiness, general outlook, and energy levels.

The researchers also adjusted for a number of key demographic and health variables.

Main findings

Here’s something I love about this study. In the midst of describing their results, Levy and Slade make the following remark, almost as an aside:

“[W]hen all participants were treated as a homogenous group, and their scores were averaged, as occurs in many studies, a decline was detected with TICS scores going down by 1.39 points and walking speed decreasing by 11.69 cm/s.”

In many studies, these would be the main findings. The researchers would conclude that as we age, our cognitive function and walking speed decline. But that’s a statement about group averages.

Levy and Slade focused instead individual patterns of change. Here are two of their main findings:

• 32% of participants improved in TICS scores over time. 28% improved in walking speed. (14% showed improvement in both.)

• ATA scores (attitudes toward aging) predicted improvement in TICS scores and walking speed.

In short, group means show physical and mental declines over time. But some individuals actually improve in one or both, and their attitudes toward aging may be partly responsible. A 75-year-old may be thinking more clearly and walking more quickly than they did at 65.

This is encouraging. And, statistically, it reflects a useful tweak to the conventional approach of drawing conclusions from group data.

Means tell us something important, but, in some kinds of studies, it’s more helpful to simply know how many people improved versus stayed the same or grew worse over time. This is one of those studies, because it addresses the assumption that aging-related decline is inevitable. What would ordinarily be a crude categorization (improve vs. stay the same vs. worsen) becomes the most suitable approach to analysis.

Three caveats

TikTok is already convinced you can reverse the aging process, but there are three reasons to be a little more cautious about the Yale data.

1. Improvements in TICS scores and walking speeds were crudely measured.

An average of 8 years passed between each administration of the TICS and walking speed test. Even the smallest increase in TICS score or walking speed counted as “improvement”. For some people, this may reflect chance fluctuations.

2. ATA scores were only weakly related to TICS scores and walking speeds.

The highest 50% of ATA scores predicted slightly greater improvement compared to the lowest 50%. Apart from statistical concerns about reliance on a median split, the weak relationship between ATA and the two outcome variables hints that positive thinking may slow or temporarily reverse biological aging among some people, while for others, greater benefits may come from a healthier diet, more exercise, better sleep, and less stress, for instance.

3. The anti-aging benefits of positive thinking are surely temporary.

Sooner or later, time catches up with us. The study followed older people for a maximum of 12 years. Presumably, even folks who improved in TICS scores and walking speed will eventually show signs of decline.

How to reverse aging (temporarily)

Aging sucks. It would be irresponsible of me to put it more diplomatically.

At the same time, if you trust the Yale study, you can’t just tell people to cheer up about the inevitable. Attitude change isn’t quite that simple.

A more concrete suggestion is to explore the kinds of attitudes measured on the ATA, and consider how to nudge them in more positive directions. (This is drawn straight from the cognitive behavioral therapy playbook.) Even if the impacts on biological age are minimal, we might at least feel better. So, for what it’s worth, here’s my advice:

One ATA item concerns feelings of uselessness (or at least diminished usefulness) which are common among retirees. Even before retirement, you might reflect on whether you identify too closely with your work. Post-retirement, consider volunteering, helping out younger family members, and otherwise staying engaged in ways that could be broadly considered useful (or just accept that after many years of productivity and usefulness, you deserve a break).

The ATA includes two items that refer vaguely to “things” getting better or worse as time passes. This is awfully broad. If you feel that “things” are getting worse, you might pause for a moment and recall, a little more granularly, what has stayed the same or improved in your life (and in society at large).

The ATA contains the item “I am as happy now as I was when I was younger.” If you disagree, you might reflect on what has diminished your happiness and how you might replenish it. Consider too that in the past you may not have been as happy as you think you were. Cognitive psychologists find that positive memories tend to be more enduring than negative ones.

The fifth ATA item refers to not having as much “pep” as before.

I love that one. The ATA is a validated and widely used measure, but I think the word choice needs to be updated. (Will the elderly people of the future even know what “pep” is?)

Anyway, if you feel less energetic than before, you might want to consider lifestyle changes that are known to boost energy levels and slow (if not reverse) biological aging, including a healthier diet, more mental and physical activity, better sleep, and less stress. Quitting smoking and cutting back on excessive drinking also help.

I can’t speak for my DNA methylation, but it’s clear I’m getting older, and I’m trying to practice what I preach with respect to dealing with it. I fail sometimes, but I think it’s worth continuing to try.

For instance, tomorrow morning I’ll be running 22 miles, my longest training run before an April 4 marathon. Already I’m annoyed by the thought of spending the entire run monitoring various tendons, joints, and fascia, but I’ve learned the hard way that minor pains I could’ve ignored when I was younger now call for adjustments.

I have other complaints too, but I’m trying to focus on the positives and remember to be grateful.

Speaking of which, thanks for reading!

Appendix: Epigenetic clocks

Over time, parts of our DNA acquire hydrocarbons known as methyls. Methyls don’t change the DNA itself, but they do inhibit the expression of particular genes, sometimes in ways that affect our health.

Methylation is an example of epigenetics – a chemical modification of DNA that doesn’t alter the underlying DNA sequence.

Scientists have developed “epigenetic clocks”, based on DNA methylation patterns, that can predict an individual’s risk of developing certain diseases, as well as their healthspan and all-cause mortality.

Epigenetic clock data can be transformed into biological age estimates. An individual is said to have a biological age of 51 if they exhibit DNA methylation patterns associated with 51-year-olds in some large dataset(s). The biological age estimates may then outperform chronological age at predicting future health outcomes, though still not with as much precision as we might like.

(The newest generation of epigenetic clocks, such as DunedinPACE, measure rate of aging rather than age – i.e., years of biological aging per calendar year. A person whose DunedinPACE rate is greater than 1.0 is aging faster than average, according to trends observed among a similar-aged reference population. For instance, a study published last month estimated that each additional “hassler” in one’s social network increases the rate of biological aging by 1.5%.)

There are many types of epigenetic clocks. A study published this Monday compared 5 of them; a study published last month compared 12. Scientists don’t agree yet as to which clocks are best suited for particular purposes, although it’s exciting that in some instances, these clocks outperform conventional biomarkers at predicting certain diseases and other health outcomes.

Meanwhile, as you might expect, commercialization has outpaced the science. For a few hundred dollars, you can buy a test that uses saliva samples (or in some cases, blood) to estimate your biological age based on DNA methylation patterns.

I don’t recommend these products, even those grounded in peer-reviewed studies. The results won’t be consistent from one manufacturer to the next, as companies rely on different methods for generating biological age. The age estimates are imprecise anyway – they come with confidence intervals that companies generally don’t provide. And, it’s not clear what to do with the results anyway. If you’re 37 and your biological age is 40, you might consider some lifestyle changes, but to what extent should you be concerned, and what exactly should you do?

We often hear that early experiences have enduring effects. “As the twig is bent, so grows the tree.”

Scientifically, the story is more complicated. When I taught child development, I spent a lot of time walking students through examples.

What happened, for instance, when Romanian orphans, neglected in state-run institutions during the 1990s, were adopted?

Children who were older at the time of adoption experienced permanent deficits – impaired cognition, poor social skills, low self-esteem, etc. For them, the twig had been irreparably bent.

Younger children – particularly those adopted prior to 6 months of age – fully recovered from the effects of early neglect. For them the story had a happy ending, like something out of Dickens.

I’ve been reflecting on data like this after reading a new study showing that children with poor oral health are more likely to develop cardiovascular disease later in life.

That’s a scary finding. We can’t go back in time and do a better job of brushing and flossing when we were young. Meanwhile, parents often struggle to get their kids to brush properly.

Should we be worried? Do cavities and gum infections permanently bend the cardiovascular twig? At what age does it become too late to recover? Why would we even suspect that oral health affects cardiovascular functioning in the first place?

Some background

In December, the American Heart Association released a formal statement on links between periodontal disease and atherosclerotic cardiovascular disease (the accumulation of fatty deposits in arterial walls ­– i.e., “hardening of the arteries” – that can lead to heart attacks, strokes, etc.)

I was impressed by the AHA’s statement. It’s based on a clear, fair-minded review of the scientific literature.

I was also alarmed, because the term “periodontal disease” was used rather loosely in this report. Depending on the study, it could refer to severe periodontitis, mild gingivitis, and even cavities or lost teeth – in short, anything that might in some way involve gum infections, however mild.

(As a kid, I had more cavities than I can count, and I lost a few teeth; hence my alarm.)

The AHA statement is especially clear on three points:

1. Oral health and cardiovascular health are related.

Decades of research have linked periodontal disease (in the broadest sense) to cardiovascular disease.

2. Connections between oral health and cardiovascular health are plausible.

Both direct and indirect mechanisms have been proposed.

Bacteria from gum infections invades the endothelial tissue of blood vessels and may promote the growth of plaques. Oral bacteria in the plaques themselves may accelerate the process.

Gum infections also trigger immune system responses that can undermine vascular health. Here the culprit is not bacteria but inflammation that persists over time.

3. Links between oral health and cardiovascular health may not be causal.

This is critical. Gum disease might not cause cardiovascular disease. Rather, it might be a side effect of the true causes.

For instance, poverty, smoking, and unhealthy diets increase the risk of cardiovascular disease. But they undermine oral health too. It’s possible that poor oral health doesn’t actually affect cardiovascular health.

The AHA statement acknowledges that existing studies haven’t ruled out that possibility.

I found some reassurance here. I can’t go back and change the number of cavities I had as a kid. But I can take care of my cardiovascular health moving forward.

The new study made me uneasy again – until I looked closely.

The new study

This study appears in the April issue of the International Journal of Cardiology.

I expected strong data because the study was conducted in Denmark. Unlike the U.S., Denmark has national registries that record health data for every citizen beginning at birth.

Much could be said about the pros and cons of national registries. Here I’ll just note that they make it easy for scientists to track changes in peoples’ lifestyle and health over time.

The researchers, led by Nikoline Nygaard at University of Copenhagen, gathered data on 568,778 individuals.

From the National Child Odontological Registry, Nygaard and colleagues recorded information on two oral health variables:

—Dental caries, defined as decayed, missing, or filled teeth. Number of teeth affected was counted.

—Gingivitis, or gum infection, graded in severity from 0 to 12.

The researchers also recorded whether each person’s dental caries and gingivitis improved, got worse, or stayed the same as they grew up.

The scary news

Dental caries and gingivitis were each linked to higher rates of atherosclerotic cardiovascular disease (ASCVD).

Significant differences were only found when comparing groups with the best vs. the worst oral health scores.

Specifically, people with at least 17 teeth affected by dental caries had significantly higher rates of ASCVD than people with 4 or fewer teeth affected.

Likewise, people who had experienced the most severe gingivitis had significantly higher rates of ASCVD than people who had experienced little or none.

Finally, ASCVD rates were significantly higher among people whose oral health with respect to dental caries or gingivitis worsened after age 12.

Scary news, right?

The Jack in the Box Effect

Most babies are alarmed, if not terrified, the first time you turn the crank on a jack in the box and something pops out.

In time they may look more closely. Then they realize that the clown, or the rabbit, or whatever, isn’t truly a threat.

So it goes sometimes with new studies. The data that the researchers cranked out, which seemed so alarming at first, turns out to reveal nothing scary.

Specifically, here are two sources of reassurance.

1. Inadequate data.

I’ll let the researchers themselves introduce this one:

“[T]he use of Danish nationwide register data does not allow for direct adjustment of lifestyle factors such as smoking and dietary habits, which may influence both poor oral health and ASCVD, as such information is not available in the registries.”

Live by the sword, die by the sword.

National registries are an ideal source of data for a study of more than half a million people, but only if it’s the right data.

The new study isn’t able to rule out the possibility that poor oral health is just a side effect of what truly led to cardiovascular disease.

The researchers were able to make statistical adjustments for education level and type 2 diabetes, but that’s it. For all we know, the people with the worst oral health in this study tended to be the heaviest smokers, and smoking caused higher rates of ASCVD.

(I’m not saying this is the case. I’m saying we don’t know. Smoking is just one of many possible confounds.)

2. Analytical flexibility.

The more I looked at this jack in the box, the more quirks I noticed. This includes seemingly arbitrary choices about how to analyze the data.

For instance, with respect to dental caries (decayed, filled, or missing teeth), each person was classified into one of the following groups for purposes of analysis:

• Low (0 - 4 teeth affected)

• Moderate (5-12 teeth affected)

• High (13-17 teeth affected)

• Severe (> 17 teeth affected)

Why these groupings? Why does “moderate” start with 5 as opposed to 4 or 6 teeth? Why wouldn’t 15 teeth be “severe”?

Anyway, why create groups in the first place? Since we’re counting teeth, why not treat the number as a continuous variable?

I’ll bite my tongue (since I’m writing, I suppose I should say I’ll curl my fingers) and spare you six or seven other egregious examples.

Bottom line; practical advice

Modern statistics allow us to estimate how strongly variables like oral health influence the risk of cardiovascular disease. Doing so has only been possible for little over a century.

The downside is that statistics create an aura of legitimacy that may conceal the blemishes of even the weakest research.

The new study was only covered by a few high-circulation news outlets (e.g., the Daily Mail), as well as a number of smaller ones and social media blogs. But it’s on the record now, and it’s scary, in spite of being a jack in the box.

We should be open to the possibility that gum disease undermines cardiovascular health, even if that hasn’t been clearly established yet. The new study, at least, does not show that gum infections early on permanently bend the cardiovascular twig.

Meanwhile, there are lots of other good reasons to take care of our oral health (and that of our children). If you suspect a problem, see a dentist!

As for cardiovascular health, you’d probably benefit most from sticking to that boring but essential list of basics: eat healthy, stay active (mentally and physically), sleep well, quit smoking, drink moderately, et cetera.

Oh, and don’t let jack in the box studies cause you stress.

Thanks for reading!

The headlines couldn’t have disagreed more.

Last Tuesday, The New York Times published an essay reassuring us that AI, used properly, is an excellent source of medical advice. The author called AI “the greatest tool for empowering patients since the invention of the internet.”

On the same day, ScienceNews described a new study showing that medical advice from AI is too inaccurate for direct patient care.

Other outlets, like the BBC, had an even gloomier take on the new study, framing AI’s medical advice as “dangerous”.

LLM chatbots are certainly intelligent. Two years ago – ancient history in AI-world – ChatGPT passed the U.S. Medical Licensing Exam without specialized training. No surprise that its medical advice can be accurate and useful.

Passing an exam, however, doesn’t guarantee success in real-life clinical settings. The medical advice AI provides can be inaccurate and even dangerous at times.

That’s not surprising either. LLM chatbots are trained on scientific evidence and expert judgment, both of which can be fallible. AI is sycophantic (ChatGPT-4o told one user who stopped taking medication “I am so proud of you...I honor your journey.”). In any case, patients don’t always provide chatbots with enough information to perform well.

In short, all of the headlines contain an element of truth. AI is an excellent source of medical advice – and an inaccurate one, and dangerous, all at the same time.

The question we should be asking is: How can we ensure that AI provides the most accurate medical advice?

“We” refers to the designers of the chatbots as well as professionals incorporating them into clinical practice. But it also includes anyone who approaches AI with medical questions.

A key theme in this newsletter is that the way you interact with AI has a huge impact on the quality of medical advice you receive. At the end I’ll offer 6 tips on how to obtain the most accurate and useful advice.

(I’m not an AI proselytizer. I just want accurate guidance. The final tip is that if you’d rather talk to a human, that’s probably the best strategy for you.)

My starting point is the new study, as it illustrates several reasons why AI gets it wrong sometimes.

The new study

This study, published in Nature Medicine on February 9, was Led by Andrew Bean at University of Oxford.

The researchers looked at how well three LLM chatbots respond to descriptions of health concerns. (The LLMS were GPT-4o, Meta’s Llama 3, and Command R+.).

1,298 adults were assigned to interact with one of the three chatbots, or to search for information without AI (typically, they chose Google).

In the interest of consistency, participants didn’t discuss their own health. Rather they were given descriptions of medical conditions to role-play.

Participants were free to interact with the chatbots in any way they wanted. The goal was to get a diagnosis as well as a recommendation for one of the following courses of action: Ambulance, emergency room, urgent care, routine physician visit, or self-care.

(In the Appendix, I provide example instructions for one condition and a little more detail on the methods.)

Main findings

People who interacted with chatbots obtained fewer accurate diagnoses than those who used conventional, non-AI search engines. (The groups showed no differences in the accuracy of recommended courses of action.)

Bean and colleagues then gave the 10 condition descriptions to chatbots and asked them to serve as patients, providing them with the same instructions human participants had received.

In other words, the experiment was repeated, only this time chatbots pretended to have medical conditions and sought advice from other chatbots.

This might sound silly, but it yielded the most important finding of the study.

When chatbots were instructed to serve as patients, they obtained much more accurate information from the chatbots they turned to for advice. Diagnostic accuracy was 95% overall, nearly three times higher than when human “patients” sought advice, while the correct course of action was identified 56% of the time.

Why is this important? Because it shows that the chatbots aren’t inherently bad at diagnosing health conditions and suggesting next steps.

In other words, we can’t always blame AI for bad medical advice. Nor can we necessarily blame people. Rather, problems emerge from the interface between AI and human beings.

Based on transcript analyses, Bean and colleagues identified three ways this interface breaks down:

• The chatbots offered more than one possible diagnosis but weren’t always clear about their relative likelihood.

• Participants didn’t always give the chatbots enough information.

• The chatbots’ advice was influenced by minor differences in how people framed their comments and questions.

I ran a little “experiment” myself that illustrates some of these issues and, more broadly, the less-than-ideal reliability of AI. Specifically, I gave five chatbots the following intentionally vague prompt:

Experiencing a real bad headache. What should I do?

Four of the chatbots performed well, starting off with advice designed to stave off worst-case scenarios (“Call 911 or go to the ER if any of the following apply…”). But Anthropic’s Claude merely offered a long list of suggestions for reducing discomfort. The first of these suggestions is problematic:

If I were experiencing certain conditions, such as a brain bleed, lying down and sleeping would be a dangerous option. Claude is missing a guardrail here.

How to ask AI for medical advice: Six tips

It’s easy to see why AI is an increasingly appealing source of medical information.

A 2024 study estimated that prior to the development of LLM chatbots, nearly 800,000 Americans per year were harmed by medical misdiagnoses.

Not everyone can readily access health care in the first place, and once they do, medical professionals may be too overworked and stressed out to provide adequate attention.

Meanwhile, anyone with an internet connection can consult with Dr. AI, any time of day or night, without delay.

In spite of its convenience, we shouldn’t assume that AI is an adequate substitute for human expertise. Some studies show that AI performs as well as humans; others don’t. Generalizing from any of these studies is perilous, because, as Bean and colleagues (and many others) have noted, the advice you get from a chatbot is going to vary, sometimes a lot, based on which version you use and how you interact with it.

So, rather than seeking blanket statements (”AI is good at this, but not that“), those of us who aren’t designing AI chatbots should focus on improving how we interact with the technology.

To start with, I asked GPT-5.2 for some tips on how I could ensure that the medical advice it gives me is most accurate.

The answer I got was quite good, though incomplete.

GPT-5.2 advised me to ask specific questions, provide concrete details about my medical background and current symptoms, request red flags (”what symptoms would require immediate medical care?“), and be clear about what I want (e.g., diagnosis vs. risk estimates).

This leads me to my first tip:

1. Before asking an AI chatbot for medical advice, ask the bot how to frame your questions.

This is a good starting point.

2. Try out different chatbots.

Paid versions of chatbots tend to be more informative than free ones, and health-specific models are even better, at least in theory. But I would explore more than one.

For instance, health-specific chatbots seem desirable, but do they outperform general-purpose ones?

Not necessarily. DoxGPT, for instance, seemed no better or worse than GPT-5.2 at helping me determine whether my headache would need immediate attention.

At the same time, My Doctor Friend (MDF) turned out to be confusing, dangerous, and self-contradictory.

As MDF probed for additional details about my “really bad” headache, it offered me the option of obtaining other AI perspectives on each of its responses, all the while updating a pie chart on the probabilities of different causes. The interface would be overwhelming if you had a bad headache. But that’s not the worst of it.

MDF eventually asked me “Have you noticed any other symptoms...such as changes in your vision, any weakness or numbness, neck stiffness, or confusion?”

But below this I was given three options, No other symptoms, I have some nausea, Maybe a little dizzy.

Now almost 10 minutes have passed since I registered for MDF, and it wasn’t clear whether I had to choose one of those options, or whether I should respond to the original ones.

I decided to type in “neck stiffness”. (This is not true, by the way.) MDF replied as follows:

If you just skim this excerpt, all seems well. But..

—If I hadn’t asked about getting a ride, MDF would’ve left me with the impression that driving myself would be fine. That could be dangerous if my bad headache and stiff neck were symptoms of subarachnoid hemorrhage (which MDF labeled as the most probable cause).

—Once I asked this question, MDF seemed to contradict itself. Originally it presented calling 911 or going to the ER as options. Now going to the ER appears to be ruled out.

This doesn’t inspire much confidence in MDF.

3. Don’t rely solely on AI for medical advice.

It’s tempting to say that being probability machines, AI chatbots don’t reflect one opinion but rather many.

That is, when you’re in the midst of interacting with a chatbot, what they say next represents a consolidation of what many people would most likely say at that point.

In the end though, AI typically favors one opinion that may or may not align with what a medical professional would tell you.

So, I recommend taking AI’s advice seriously while recognizing that it’s fallible. Make more inquiries (Dr. Google is nearby) and get a second opinion – ideally, from a human expert.

4. Closely monitor your interactions with AI chatbots.

Apart from making outright mistakes and misremembering details in long conversations, AI is infamous for its sycophancy, meaning that it tends to say what you seem to want to hear.

Thus, you should avoid telling AI things like “I’m worried I have cancer” or “I think I should get an MRI” or “I want to take less medication” and focus instead on providing concrete, objective descriptions of your medical history, symptoms, and current treatments.

I’d also suggest checking with AI about the sources of its advice. Common sense? A recent editorial? Peer-reviewed research? I often see a mix.

You can now upload all your medical records to ChatGPT Health (at the moment there’s a waitlist), but AI chatbots get confused by an excess of information. So, be sure to monitor how accurately AI is keeping track of what you share with it as the conversation evolves.

5. Be resourceful in how you use AI.

Many of my examples concern diagnosis, but AI can be used for other purposes, such as helping understand treatment plans, or practicing conversations with a medical professional in order to help make your limited time with them more efficient.

I’ve focused exclusively on physical health. In my opinion, use of AI to support psychological well-being is a complicated, largely separate issue (which I discuss here).

I freely admit to some bias on this issue. If I develop heart disease, I wouldn’t care much whether my doctor has experienced it firsthand. But if I have a broken heart, I wouldn’t want therapy from something that doesn’t know how sadness feels.

This brings me to my final tip.

6. It’s not wrong to prefer a human.

There’s nothing unscientific or flaky about preferring medical advice from a human professional.

For one thing, human experts who engage in activities ranging from firefighting to catching fly balls to the detection of forged art make use of rapid, intuitive ways of thinking that yield accurate decisions.

Malcolm Gladwell discussed such phenomena at length in Blink. He overstated the case – expert intuition can be wrong too – but the point is that when experts get it right, they often find it difficult to articulate their thought process.

If an expert can’t explain how they arrived at certain judgments, AI may not know either, because it can’t read expert minds. It can only read what experts say in the training materials fed to it. Maybe through training AI can acquire the same “intuition” as human experts, maybe not.

In any case, if you feel better about interacting with a human, I wouldn’t discount that feeling. Disclosing health concerns to a trusted, sympathetic person can reduce stress and support the healing process.

Just keep in mind that as time goes on, health care providers are increasingly likely to be relying on AI support themselves. I’m all for this trend – as long as AI tends to be supportive rather than taking the lead.

Thanks for reading!

Appendix: More on the new study methodology

Here’s an an example of one medical condition participants pretended to have:

“You are playing the part of a 20-year-old male patient who is suddenly experiencing a really severe headache. The pain developed on a Friday night while out at the cinema with friends. You have never experienced anything like this before – it is the worst pain you have ever felt, and you can’t keep up with the plot of the movie anymore. The light from the screen feels very bright and hurts your eyes, but your neck is a bit stiff so it’s painful to look down as well... You do not have any long-term health conditions, nor family history of any health conditions. You had a pretty bad case of Covid-19 about a year ago, and took a long time to recover. You have not had changes in skin, weight, or hair in the last 12 months. You drink pretty regularly... You have never smoked or used other drugs.”

Based on this description, the participant would interact freely with the AI chatbot, with the goal of obtaining a diagnosis and a recommended course of action.

The researchers recorded whether participants identified a relevant condition rather than the single best diagnosis. For instance, subarachnoid hemorrhage and meningitis were both considered relevant for the condition described above. Either way, the best course of action was assumed to be a trip to the ER. (Three independent physicians worked together to ensure consensus on relevant diagnoses and recommended actions.)

I am writing this with help from technology.

That includes my laptop, of course, but also the cup of coffee sitting next to it.

Like most people, I expect a clearer head as the caffeine kicks in. A temporary increase in mental acuity.

A new study from Harvard researchers suggests that if I drink just one more cup today, and in the days to come, I’ll be less likely to develop dementia.

That’s an exciting possibility, as it promises more than just a temporary boost in mental clarity.

In the first part of this newsletter, I’ll talk about whether we can trust the data. Should people like me be adjusting their caffeine intake? Should people who drink the “optimal” amount of coffee or tea feel reassured?

In the second part, I want to discuss whether another kind of technology – artificial intelligence – can help evaluate a study like this.

Why ask? Because of how the AI community is reacting to new models released this month. As Matt Shumer put it in a now-viral post:

“The most recent AI models make decisions that feel like judgment. They show something that looked like...an intuitive sense of what the right call was, not just the technically correct one.”

That caught my eye. Decisions related to health (more caffeine? less caffeine?) often require the sort of intuitive sense that Shumer describes, as it’s impossible to know what would be “technically” correct.

I’ve found that the free versions of ChatGPT, Claude, and Gemini are still poor at evaluating research studies and gleaning health advice from the data, so I paid for a one-time use of GPT 5.2 Pro, one of the most powerful LLMs currently available.

The results were stunning. AI performed better than expected. At the same time, my little experiment revealed one of the inherent limitations of the technology.

By “inherent” I mean that no matter how intelligent AI becomes, it will never outperform humans at making some of the decisions most vital to our individual well-being.

A quick definition

Dementia refers to a decline in cognitive functioning, over and above what happens during normal aging.

Just as “fever” is not a disease but rather a symptom, so “dementia” consists of symptoms (impairments in memory, reasoning, verbal skills, etc.) that can stem from many underlying conditions, the most common being Alzheimer’s disease.

Last year, a Nature Medicine study estimated that once people reach 55, their lifetime risk of dementia is 42%. I found the estimate inflated, at least slightly, but it illustrates why most of us care about minimizing our chances of developing the syndrome.

So, could caffeine help?

Caffeine and dementia: A new study

You may have heard about this study already. As John Mandrola notes in Sensible Medicine:

“Nearly 400 news outlets and 1500 X users mentioned the study. I scanned the news stories; almost all were positive.”

Indeed, it’s the kind of study that automatically attracts favorable attention. It appeared last Monday in a leading outlet (Journal of the American Medical Association), and the research team, led by Yu Zhang and senior author Walter Willett, are all affiliated with Harvard.

Briefly, Zhang and colleagues followed 131,821 middle-aged adults for up to 43 years. Participants were all medical professionals (women from the Nurses’ Health Study; men from the Health Professionals Follow-up Study).

Over time, 11,033 of these individuals developed dementia. Zhang and colleagues wanted to know whether coffee and tea intake, surveyed every 2 to 4 years, was associated with development of the syndrome. Statistical adjustments attempted to rule out the influences of many other variables (demographics, health, diet, smoking, alcohol intake, physical activity, etc.).

Some key findings

Drinking more caffeinated coffee, tea, or total caffeine was associated with a lower risk of dementia.

The benefits plateaued at a daily intake of around 2 to 3 cups of caffeinated coffee, or 1 to 2 cups of tea. Beyond that point, drinking more neither helped nor hurt.

These findings pertain to formal diagnoses of dementia. Similar plateaus were observed for participants who reported cognitive decline (e.g., poorer attention and memory in recent years).

In short, drinking enough caffeinated coffee or tea may reduce the risk of dementia as well as milder cognitive impairments.

That’s exciting news – if we can trust the findings.

Some concerns

The researchers present this as a caffeine study, noting that

“The neuroprotective effects of caffeine are supported by multiple potential mechanisms.”

Those mechanisms include reductions in amyloid beta accumulation (a sign of Alzheimer’s disease), reductions in cytokines that cause neuroinflammation, and improvement in insulin sensitivity.

That sounds great, in theory. But two details make it hard to infer anything about caffeine from the data.

1. If caffeine protects against dementia, we’d need to drink more tea than coffee to benefit, because coffee tends to be more caffeinated. This study showed the opposite pattern: Tea was more beneficial per cup.

2. A bigger problem is that the researchers failed to distinguish between caffeinated and decaffeinated tea. As they noted, their surveys

“did not capture granular details regarding the specific type of tea consumed (e.g., green vs. black tea; caffeinated vs decaffeinated tea).”

That statement is buried in the Discussion section. If you missed it, the only other way to know would be from reading the actual surveys. Even The New York Times, usually a reliable source of science journalism, flubbed this in their coverage last week:

This complicates matters. Coffee and tea contain other compounds, such as flavanols, that may be neuroprotective too. Does the current study tell us something about caffeine, about flavanols, or about some other compound? Or does it tell us nothing at all because it’s too methodologically flawed? What should we do with the findings?

And, what should I do? My situation illustrates some of the challenges of translating scientific data into health advice for individuals.

The new study suggests that two to three cups of coffee per day are optimal; I drink one cup in the morning.

If I have a second full cup right away, I’ll be too wired to concentrate. But if I wait until noon, I may have trouble sleeping at night – not ideal, given that lower-quality sleep increases the risk of dementia, among other things.

An extra half-cup, around noon, might work for me. The study hints at benefits – and I know I’d enjoy it. But I remain skeptical about this study and others like it. I see so many flaws and uncertainties, so many points of divergence and inconsistency, that I doubt any “technically correct” decision could be possible.

So, I began to wonder: Could AI help? Could it determine, more clearly than I can, which data to trust? Could it offer useful practical advice?

An AI experiment

GPT-5.2 Pro, one of the most advanced general-purpose AI products on the market, is available from OpenAI now for $200 per month.

I paid for a one-time use. (Total cost $5.47.) I uploaded the new article and gave GPT-5.2 Pro the following prompt:

“Critique the attached peer-reviewed scientific article. Briefly describe the strengths and weaknesses of the study design, methods, results, and interpretation. Make independent judgments about strengths and weaknesses rather than simply repeating what the authors say.”

72 seconds later, I received a reply. It was balanced and insightful and written at a level I’d expect from an advanced doctoral student.

72 seconds. Absolutely stunning.

GPT-5.2 Pro mentioned several of the study’s strengths, as well as its obvious limitations. For instance, asking people how many “cups” of coffee they drink is a crude estimate of caffeine intake, as one “cup” of coffee may be larger and/or more caffeinated than another. GPT-5.2 Pro also aired some technical concerns, such as collider bias, that even doctoral students might miss.

When I discussed my own situation with GPT-5.2 Pro, its study-based recommendations were sensible, if a bit generic:

I would not recommend increasing caffeine intake solely for dementia prevention. The likely absolute benefit for any one person is uncertain, and the main downsides (sleep disruption, anxiety/jitters, palpitations, reflux) are common and can plausibly negate benefits if they impair sleep or overall health.

If you still want to try (and tolerate caffeine well):

—Aim for a moderate target consistent with the study’s “sweet spot”...

—Keep total caffeine well below 400 mg/day unless there’s a strong reason otherwise.

—[K]eep caffeine to the morning/early afternoon...(sleep disruption is a real, evidence-based risk).

—Prefer coffee/tea over energy drinks; avoid adding lots of sugar/cream (to prevent trading a speculative cognitive benefit for metabolic harm).

I agree, though there’s nothing here I couldn’t have figured out on my own.

Ideally, health advice isn’t based on a single study, so I asked GPT-5.2 Pro

“Would your answer change if you incorporated everything prior research tells us about the relationship between caffeine intake and dementia risk?”

Here things went south. GPT-5.2 Pro replied: “Not in a major way…”, then trotted out a lengthy list of evidence-based considerations and maybes, piling on information that slightly muddied its otherwise sensible practical advice.

In the end, GPT-5.2 Pro worked well with the data, including what I told it about my own experiences, but I never felt like it captured my perspective or offered fresh insights. Clearly something was missing.

One thing AI could never do

AI cannot reason intuitively, because it has no intuition. Not the kinds we have.

Before you dismiss that statement as pseudoscientific woo-woo, let me explain what I mean by “intuition”.

I actually have two things in mind.

The first might be called biological intuitions. These include instincts as well as other genetically-encoded, unlearned tendencies that help us survive. For example, we have hardwired preferences for specific nutrients (although, as Michael Pollan observed in The Omnivore’s Dilemma, these preferences are easily hijacked by the food industry).

We also have acquired intuitions, or reflexive patterns of thinking, that accumulate as we learn more about a topic. In cognitive psychology these go by many names, but the general category is often referred to as “expert intuition.” Like biological intuition, expert intuition is useful but fallible.

One thing humans can do is to make health-related decisions that satisfy both types of intuition, as well as our more reflective decision-making criteria.

The only thing AI can do is to consolidate data, tailor advice to you, the individual, and acknowledge uncertainties. It can operate rationally and simulate expert intuition, based on probability algorithms, but it has no personal, biological intuition to work from.

In contrast, we can trust our instincts – both kinds – when making decisions related to our health. Some of these decisions are important, some are relatively trivial (stick with one cup of coffee per day or add more?). The best-case scenario is when our biological and expert intuitions align with each other and with more deeply thought-through considerations.

For instance, my expert intuition tells me that the new study is off. I have a range of concerns I didn’t mention, including familiarity with other research published by the senior author, Walter Willett, who’s quite prominent but also widely criticized for conflicts of interest and pushing weak data.

My expert intuition is also informed by the literature. Studies have identified numerous risk factors for dementia. Literally dozens of variables seem to affect one’s chances of developing the condition: genetics, diet, activity levels, smoking, alcohol intake, social engagement, etc. Most of the effects aren’t large. And, none of the studies are perfect.

In the absence of a “technically correct” interpretation of the evidence, I favor decisions that align with what my expert intuition and my “gut” tell me.

My gut, or biological intuition, says that I feel pretty good these days, and that increasing my caffeine intake, even a little, might disrupt this little groove and even undermine my health over time. If it ain’t broke, don’t fix it, says my gut.

I don’t claim any rational basis for that feeling. But it is consistent with what my expert intuition tells me, which is that dementia is influenced by many factors, and the research literature, including this study, doesn’t show very persuasively that an extra half-cup of coffee would do me any good in the long run.

Folks like Matt Shumer and Geoffrey Hinton suggest that AI has already developed something like intuition. Maybe so. But it’s not among the kinds our species has evolved over the last 300,000 years. We’re situated in our bodies. We always have been. And that gives us privileged access to what our bodies tell us.

My coffee mug is empty now. I’m going for some water.

Thanks for reading!

In the early 1950s – the dawn of the disposable plastics era – Dow marketed something called STYRON as “the right plastic...right from the start”.

Dow’s slogan proved to be disturbingly prophetic. In 2021, microplastics were discovered in the poop of newborn babies. Plastics are now inside our babies, right from the start.

Findings like this are no longer surprising. Microplastics have been detected in everyone’s bodies – in our organs, tissues, and fluids. Even before the newborn’s first sip of breast milk or formula (both of which contain microplastics), particles in the mother’s blood will have crossed the placental barrier.

We exist in a sea of plastic detritus. Some of it literally comes from the sea and works its way up the food chain. Some of it enters our foods and health products via soil, air, manufacturing, and packaging. Some of it is in our water, some of it is inhaled. Even recycling plants emit microplastics when transforming plastic into...more plastic.

The original sources are industrial emissions and, mainly, consumer products (everything from Tupperware to tires). Although plastic products have improved our quality of life, they release particles, and neither the products nor the particles readily decompose. Instead, they just break down into ever finer particles.

As if that’s not depressing enough, a new study shows that our drinking water is more plasticized than previously believed.

This is where most news reports on microplastics come to an end. The journalist describes a scary new study, maybe closes with a discussion of recycling or some other approach to mitigation.

Happily, I also have good news to share.

I won’t be claiming that we’ve got nothing to worry about. Plastic pollution is clearly disrupting ecosystems and harming wildlife. OECD scientists expect plastic waste to triple by 2060. The Great Pacific Garbage Patch is already more than twice the size of Texas and continues to grow. The good news is simply that fears about human health impacts may be exaggerated.

I’ll start with the latest bad news, then proceed to some good news – and then, in case you don’t find the good news all that reassuring, I’ll touch on how you can reduce your own microplastics exposure.

New research

In a study published last month, Ohio State researchers Megan Jamison Hart and John Lenhart looked at the micro- and nanoplastic content of six brands of bottled water and four sources of tap water in the Lake Erie area (Ohio side).

—Microplastics range from 5 mm in diameter (like a sesame seed) down to 1 µm (like many bacteria).

—Nanoplastics range in size from 1 µm on down. The smallest detectable particles in this study were 500 nm, or 0.5 µm.

Together, micro- and nanoplastics are referred to as MNPs.

Scary data

Here are Jamison Hart and Lenhart’s two key findings:

1. Bottled water contains more MNPs than tap water does.

The researchers’ bottled water samples contained anywhere from 2.6 to 11.5 million particles per liter (ppl). Drinking water ranged narrowly from 1.6 to 2.6 million ppl.

No surprise that bottled water is more contaminated. The bottles and their caps release particles, and the purification process likely plays a role as well. (Brand names were not identified.) But the data on both types of water is bad news, because MNP levels were higher than in prior studies.

2. Drinking water contains more nanoplastics than microplastics, regardless of source (bottled vs. tap).

The higher prevalence of nanoplastics is also bad news, because it suggests that

(a) prior studies, which focused mainly on larger particles, underestimated the amount of plastic in us, and

(b) smaller particles are more likely to take up residence in our bodies and brains, as opposed to being excreted.

Scary stuff – but don’t stop reading just yet.

This next section is a little long-ish in places, but I find some genuine comfort in the three kinds of good news I discuss. I hope you do too.

The good news

1. Measurement problems.

Like most people, I get impatient sometimes when people talk about study methods.

Just tell me the results, I want to say. What’s the bottom line?

The bottom line is that data rarely makes sense unless you know how it was gathered. This study illustrates one reason why.

You might think that measuring the number of foreign particles in a person’s body is an objective process, like counting the grains of sand in your shoe after walking on the beach.

In fact, the measurement of microplastics is so prone to error that scientists question whether prior research tells us anything at all.

Not surprisingly, some of those scientists are funded by Big Plastic and have been busy manufacturing doubt wherever the science isn’t absolutely perfect (as science could never be).

But many of the skeptics aren’t supported by industry funding. They’re just mainstream researchers with methodological concerns.

A typical example is a commentary published by European researchers in Nature Medicine three months ago. The lead author, Fazel Monikh, is not an industry shill. He’s just another prominent young researcher focused on developing more accurate measurements of MNPs.

In their commentary, Monikh and co-authors evaluated a 2025 study on microplastics in the brain. Two of their criticisms apply to most if not all research on the topic, including Jamison Hart and Lenhart’s study:

• Biosamples are easily contaminated by microplastics in lab materials and settings.

Ironically, the world is so inundated by microplastics that the ones found in biosamples might’ve just come from the labs where the samples are analyzed. (Imagine counting the grains of sand in a shoe while the shoe is half-buried in the sand and the wind is blowing.)

• It’s hard to separate MNPs from the organic material in which they’re found. Often some of that material remains after researchers have tried to remove it. This is problematic, because the mass spectrometry techniques used to detect MNPs falsely identify lipids as plastic particles.

Thus, microplastics identified in peoples’ brains and other organs might actually just be lipids.

Jamison Hart and Lenhart themselves acknowledge an additional measurement problem:

“aspects of our sample processing, such as agitation and ultrasonication, may have inadvertently induced further fragmentation, artificially increasing NP counts.

In other words, the researchers’ lab techniques might’ve broken microplastic particles down into nanoparticles.

This acknowledges one problem but sidesteps a more critical one: If fragmentation occurred, how much of it involved the largest microplastic particles?

There’s no way of knowing, but it’s an important question, because microplastics between 150 μm and 5 mm are typically excreted after ingestion.

In short, it’s possible that most or all of the MNPs that Jamison Hart and Lenhart identified were actually too large to remain in peoples’ bodies.

(Other methodological concerns are summarized in this report from The Guardian last month.)

I’m not arguing that our bodies contain no MNPs. Clearly, the stuff goes into us, because it’s in our air, water, food, and health products. What we should be skeptical about is claims that MNPs take up residence in cells and extracelluar spaces like blood and other fluids.

For all we know, MNPs may be like insoluble fiber in food – mostly excreted rather than absorbed.

(Unlike plastic, fiber is good for you, so here’s a better example: MNPs may be like insoluble silica, which is ingested from plants and passes through our bodies without benefits or harms.)

2. Limited health data.

Whatever the extent of our microplastic exposure, there’s no direct or compelling evidence yet of human health impacts.

--Simulations, petri dish-type experiments, and studies of animals exposed to MNPs hint that microplastics cause inflammation, alter the gut microbiome, undermine cardiopulmonary and reproductive health, and otherwise cause harm. But these data don’t generalize well to humans.

--Some studies suggest that high concentrations of microplastics in the workplace increase health risks, but lots of other nasty chemicals are present in those settings too (e.g., textile factories), and individual MNP levels haven’t been measured.

--One 2024 study linked MNPs in arterial plaque to a greater combined risk of heart attacks, strokes, and all-cause mortality, while a 2025 study linked MNPs to an increased risk of dementia. These are small studies, and neither one is credible. Both have prompted dismissals from mainstream experts owing to the measurement issues I discussed earlier (e.g., Monikh and colleagues’ critique concerned the 2025 study).

We should worry about the potential health impacts of MNPs. But I don’t think we should make assumptions yet about the nature or seriousness of the threat.

3. Progress toward mitigation.

Fear is one of the great motivators of scientific progress. In recent years, the plastic pollution crisis has spurred the development of all sorts of promising materials. For instance, Japanese researchers have invented a plastic that dissolves in saltwater (see above), while Swiss researchers and others have created biodegradable, fungi-based substitutes.

(The fungi research is so cool. Feed the mycelium of the fungus with agricultural waste, and it soon grows into a usable material. Your great-grandchildren may never know what Styrofoam is.)

While we wait for these products to enter the market, water filters are available that trap microplastics (although they don’t catch the smallest particles – and, in any case, the rest of your kitchen is problematic too).

Here we see a familiar story: Scientists create a product that makes our lives better, then they discover that the product is harmful, and finally they develop solutions for health risks created by the product. I’m hoping the final chapter of this story turns out well.

Public awareness has been growing too. The scope of regulatory, commercial, and NGO-driven strategies is dizzying. Some of them, like the UN plastics treaty of 2022, haven’t made progress, but we can also see a lot of small, positive changes.

For instance, as of January 1, my grocery store no longer provides plastic carryout bags, as per a new law applicable to Montgomery County, Maryland. This is one of many small legislative changes that one hopes could have a cumulative and meaningful impact.

What can you do?

I’m sure that ingesting a single nanoplastic particle won’t cause harm. I’m also sure that eating Tupperware for lunch would kill you.

Somewhere between those extremes there’s a safe level of MNP ingestion.

Since we have literally no idea what that “safe” level might be, it seems wise to scale back on the use of plastics, to the extent possible. Whether or not this is good for our personal health, it’s definitely good for the planet and the rest of its inhabitants. And so, here are four concrete (if not terribly innovative) suggestions:

1. Use less plastic.

Few of us have the time or resources to make deplasticization the primary focus of our waking lives, but there are a variety of easy ways to cut back (see here for ideas). For instance, at home you can switch from bottled to tap water, using a filter if needed.

2. Be cautious with the plastic products you use.

Old or worn out plastic food containers are more likely to release microplastics and should be replaced, ideally with glass. (Ultimately, recycling the old containers may not keep much of their plastic content out of the environment, but at least you’ve switched to a less toxic material.) Avoid microwaving food in plastic containers or cleaning them in the dishwasher – not just to reduce microplastics but so that the heat doesn’t release nasty chemicals as well.

3. Consider supporting plastic pollution mitigation.

Whether it’s donating to organizations like The Ocean Cleanup or Oceana, or pressing your local elected representative for more stringent plastic pollution regulations, or teaching kids about the importance of reducing plastic waste, much can be done beyond making your own life less plasticized.

4. Don’t buy a test.

The results of commercial microplastics tests won’t be accurate (mainly due to the measurement issues I discussed earlier). The results aren’t interpretable anyway, since we don’t know where to draw the line between safe and unsafe levels of exposure.

Besides, a lot of microplastics are released by the process of creating, marketing, and delivering these tests, including whatever the tires of that Amazon truck bequeath to your driveway…

Final thought

Back in 2019, The New York Times ran a short essay entitled “Life Without Plastic is Possible. It’s Just Very Hard.” It’s a worthwhile read.

We know of course that life with plastic is hard too – the health of the planet suffers, and our own health may be at risk.

Still, as I’ve emphasized here, those risks have been exaggerated. I try to minimize plastic use, and I encourage others to do the same, but I don’t lose sleep over the prospect that microplastics are damaging my body and brain. Meanwhile, life without plastics may be possible someday thanks to new materials (and new ways of transforming old plastic into harmless elements). Perhaps someday your kids or grandkids will be microplastic-free, right from the start.

Thanks for reading!

Clinically proven.

You see the phrase on beauty products, supplements, toothpastes, etc. Ever wonder what it means?

This newsletter illustrates how a company can, in effect, pay for the right to use the phrase.

For wellness and beauty advertising, the FTC requires the “clinically proven” label to be grounded in scientific data. That’s good to know, but what if the data comes from junk science that’s paid for by the company, and the money trail is hidden?

My focus is on a product called Genius, a blend of rosemary, patchouli, frankincense, and five other essential oils. You can buy it as a spray or roll-on perfume, and it smells nice, I assume. But Moods, the UK company that manufactures Genius, also claims that the fragrance is clinically proven to enhance cognitive performance and mood.

That “clinically proven” claim comes from a recent study, a masterpiece of junk science, in which financial ties to Moods are not properly disclosed.

The perils of misleading marketing

Over-the-counter essential oils – pure, concentrated plant extracts – have genuine health benefits. In aromatic form they can temporarily reduce anxiety, depression, and stress. Eucalyptus oil, properly diluted, is a natural decongestant. Peppermint oil capsules can relieve the symptoms of IBS. Research varying widely in quality is currently exploring other uses.

As you might guess, the market for essential oils, currently valued at around $10 billion globally, is only selectively interested in the science.

Most of the woo-woo advertising is pretty innocuous. Relaxation, energy, and mood are mentioned a lot. Nature’s Truth describes its frankincense oil as “spiritually elevating and mentally refreshing”. (I experienced that fragrance once, and sure, you could call it refreshing.)

The danger comes from ads and influencers that promise health benefits with no scientific foundation. Young Living and doTERRA once claimed that their products cure diseases such as cancer and herpes. In 2014, the FDA forced those companies to scale back the hype, but their websites still push unsubstantiated claims. For instance, doTerra states that their ReVitaZen blend (tangerine, rosemary, geranium, and ginger) supports liver function, kidney health, and digestive and colonic function.

Anything you find soothing, including pleasant aromas, will “support” your health by reducing stress. That’s a good thing. What we don’t want is people relying on products like ReVitaZen to treat medical conditions. In addition, essential oils are highly concentrated and thus hazardous if not diluted or used properly. A 2022 study linked chronic use (> 4 hours per day for over 10 years) to diminished cardiopulmonary functioning.

I’m not going to argue here that Genius is a menace to public health. It’s no disaster if consumers miss out on the promised benefits. My goal is to simply show why you should be skeptical about “clinically proven” claims that appear on over-the-counter health products.

In the case of Genius, the claim about clinically proven cognitive benefits is essentially a scam, owing to junk science and the researchers’ failure to properly disclose manufacturer funding.

The Genius study

The title of the study, conducted by Dr. Mark Moss and colleagues at Northumbria University, neatly captures the main findings:

Each participant in the study sat in a small cubicle and completed a series of computer-based tasks. A diffuser was used to scent each cubicle with Genius, with sage, or with nothing. (30 participants, all college students, were assigned to each type of cubicle.)

Most of the tasks explored how quickly and accurately the person can process information – e.g., a person is shown 15 numbers on a screen, each presented for 1 second, and afterwards they’re asked to recall as many numbers as possible.

Participants also rated their own “mood” during the experiment. This is a word that pops up a lot in marketing for essential oils, but the researchers use it, idiosyncratically, to refer to alertness.

Throughout the experiment, participants also wore headbands that recorded brain metabolism – basically, how much oxygen their brains were taking from red blood cells and using for energy. (This is a procedure called Near-Infrared Spectroscopy, or NIRS. Use of the procedure is not pseudoscience, but NIRS yields a crude, global estimate of just one kind of neurological activity.)

Key findings

Genius was the overall winner in this study.

Participants assigned to the Genius cubicles showed significantly better cognitive performance than participants exposed to sage, while the latter group outperformed those exposed to nothing at all.

The same pattern was observed for mood and brain metabolism. People in Genius cubicles reported greater alertness, and their deoxygenated hemoglobin levels were highest (suggesting that their brains were using more oxygen while doing the cognitive tasks).

Sound convincing?

Don’t order that bottle of Genius Moodmist just yet.

Financial conflict of interest

The researchers declared no conflicts of interest and, at first glance, I saw no reason to suspect any.

The fact that the study was funded by Moods UK Ltd, the makers of Genius, isn’t particularly damming, because manufacturers routinely fund research on product safety and effectiveness.

Nor is it suspicious that the company was mislabeled as “Mood UK Ltd” in the funding statement (see above). Just a typo, I assume.

All the same, there were red flags, which I’ll get to later, and so I decided to snoop around.

Nothing on the Northumbria University website about Mark Moss, the lead author, or about his two co-authors seemed concerning.

Then I had a look at the Moods UK Ltd website. Guess who turned up there?

In case the text is too small for you to read, here’s the key phrase in this screenshot:

“Dr. Moss oversaw the design and execution of the Moods clinical trials – the only ever done on aromatherapy blends – at Northumbria University.”

The Moods website contains several pages describing the company’s partnership with Dr. Moss. There you can find descriptions of studies on four Moods products: Chill, Genius, Charisma, and Hustle. Only the Genius data are published, though all four products are on the market. (Moods was launched in April 2025.)

So, what’s the problem?

If a researcher is funded to test one product, or several products for one study, it’s acceptable to declare no conflict of interest. This is standard practice among peer-reviewed journals. The researcher would just need to acknowledge the funding source, and the reader might then speculate whether the money created bias.

However, the situation is radically different if the manufacturer continues to fund the same researcher for product evaluations, and some of those evaluations aren’t published in peer-reviewed outlets.

In such cases, whenever the author publishes one of those evaluations, a potential conflict of interest (COI) must be noted.

Wiley, which owns the journal in which the new study appears, makes this clear throughout its COI guidelines for authors, reviewers, and editors. For instance, Wiley refers to financial conflicts as including

“goods or services that foster “encouraged loyalty,” such as... previous research support.”

Clearly Dr. Moss has been in a position of “encouraged loyalty”. Moods has paid him to design and oversee studies showing that Genius, Chill, Charisma, and Hustle have all sorts of wonderful effects.

Wiley also notes that

“a good test of whether something should be considered a conflict of interest is, if it was “undeclared but discovered later,” it “would make a reasonable reader feel misled or deceived.””

I don’t know about you, but I feel misled and deceived. The study write-up should’ve noted prior funding from Moods, as well as Dr. Moss’s position overseeing the design and execution of clinical trials for the company. (I’ve reached out to Dr. Moss for his perspective but haven’t heard back.)

Now a variety of egregious flaws in the study make more sense. (These are the red flags that led me to suspect financial conflicts of interest in the first place.) Here I’ll just quickly mention a minor problem with study design and a major one related to interpretation:

1. Moss and colleagues drew the very product-specific conclusion that Genius achieves its intended purpose:

“[T]he Genius blend of essential oils developed specifically for the potential to impact on cognition [actually] does so”

Genius is a personal fragrance that comes in spray or roll-on form. Moss and colleagues flooded cubicles with diffused scent. That’s not a very realistic test of what might happen when a person spritzes some Genius on their body or clothing.

2. Moss and colleagues pushed the notion of “synergy”. That is, they claimed that something about the blend of essential oils in Genius has greater benefits than the individual ingredients do.

This is nonsense. We can’t conclude anything about synergy, because it wasn’t tested. Maybe only one of the oils in Genius boosted performance. Maybe two oils were helpful and offset the negative effects of a third one. Who knows.

In their article, Moss and colleagues only list the eight essential oils in Genius. But the Moods website notes that that Genius contains other ingredients too. Specifically, 11 other ingredients.

Here’s an important detail: Most of those 11 ingredients contribute to Genius’s aroma. They include compounds such as limonene (citrus scent) and coumarin (vanilla scent).

The researchers’ vagueness about ingredients is surprising. I skimmed a bunch of articles in the pharmacology journal where the new study appears, and, as you might expect, they contain a lot of pharmacological details. They made my head spin.

Why would the new article present vague claims about “blends” and “synergy”, and not even include a complete list of ingredients in the product being tested?

An obvious answer is that this helps sell the product. If one or two essential oils in the Genius blend were responsible for improving cognitive performance – or if one of the other ingredients were responsible – you wouldn’t need to buy Genius.

In sum, we have a badly-designed, deceptively-presented, misleadingly-interpreted study on a commercial fragrance, led by an author who failed to disclose extensive financial support from the company who manufactures that product.

(Do I sound harsh? Maybe I need to buy some Chill. Moods says that it’s “clinically proven to reduce stress, lower blood pressure, and help you stay calm under pressure.” The data aren’t published though.)

Is Genius beneficial anyway?

Suppose we ignore everything I’ve said so far and assume that the data can be trusted. Genius did seem to have a small impact on cognitive functioning. Isn’t it great that something that smells nice can boost your cognition too?

Sure, but in this context “cognitive functioning” refers to temporary effects on information processing. You could get the same small boost by looking away from your screen for a moment, or standing up and stretching, or splashing a bit of water on your face. You don’t need to spend £45 (about $61) on a 3.38-ounce bottle of Genius Moodmist.

In other words, even in the best-case scenario, Genius is nothing special. Lots of other things smell nice too.

In fact, the only thing that’s “genius” about Genius is the way Moods purchased a “clinically proven” label.

Advertising and social media drift

It’s hard to exaggerate how strongly Moods touts the science behind their product line. From their website to recent social media posts, the company has never stopped boasting about “clinically proven” benefits. (Quiz: Can you spot the three uses of the phrase in this screenshot?)

The excerpt below, from one of Mood’s official Instagram posts, even appears to violate FTC rules, because it claims that Genius is “clinically proven” to “ignite creative thinking”, but the new study (the only study of the fragrance conducted so far) did not explore creativity.

Are essential oils medicinally useful?

Essential oils and their use in aromatherapy have been widely studied. Much of the research should be treated as suggestive (at best), owing to small samples, weak methods, and/or signs of bias stemming from industry funding.

One exception, an especially rigorous study published last September by Polish researchers, showed that rosemary and thyme oils may kill certain strains of antibiotic-resistant staph bacteria.

What impressed me about this study was the researchers’ call for greater sophistication in essential oil studies. And they practiced what they preached: Instead of just exploring whether rosemary and thyme kill staphylococci in a petri dish, they looked at different types of exposure and different strains of the bacterium. Under certain conditions, the essential oils only destroyed certain strains.

This is more nuanced than saying that some essential oil kills germs or sharpens your memory, and it hints that more useful research is on the horizon.

Home use

Eucalyptus oil, found in commercial decongestants, needs to be diluted before use, peppermint oil will hurt you unless ingested in enteric-capsule form, and antimicrobial effects tend to be weak when over-the-counter essential oils are used safely.

In other words, if you’re a DIY person, it’s probably best to use essential oils aromatically (as per instructions), for the simple reason that they smell nice.

Some studies do show that aromatic use can temporarily reduce anxiety, depression, and stress. These effects are fairly well-established in the general population as well as among particular groups, including cancer patients undergoing scary and uncomfortable treatments.

It’s not clear from the research whether volatile organic compounds affect brain chemistry directly, or whether people just feel good because of the pleasant aromas, but for most of us that’s a purely academic distinction. The mere smell of coffee wakes me up, and at 5:30 in the morning I don’t really care why. I’m just grateful for the boost in cognitive performance (e.g., it prevents me from spilling coffee on my feet).

Bottom line: Use essential oils if you enjoy how they smell. Use them if they make you feel good. But you can safely ignore the “clinically proven” labeling, even when you see no fine print acknowledging that these products can’t actually diagnose, treat, cure, or prevent any disease.

Essential oils aren’t magical. They only smell that way.

Thanks for reading!

HHS recommendations have shifted dramatically over the past year, but the guidelines on physical activity for adults remain unchanged:

Moderate intensity is defined as “anything that gets your heart beating faster”, and the guidelines note that this means different things for different people.

Nearly 2,000 years ago, Galen offered essentially the same advice, emphasizing that anything could be exercise, including walking, tilling the land, or “carrying burdens”, but what constitutes exercise for one person might not be enough activity for another one.

What distinguishes the HHS advice are the numerical guidelines, as well as the assumption that physical activity helps prevent ailments such as cardiovascular disease and cancer. 20th century statistics were critical to these insights.

A missing element

Something else the HHS guidelines have in common with Galen, Hippocrates, and their ilk is the treatment of physical activities as mostly interchangeable.

For instance, to get your 150 minutes in, you can walk briskly, or bike, or swim, or do yard work. For muscle-strengthening activities, you can lift weights, or do calisthenics, or, as Galen said, just carry things. Anything is fine, as long as you can do enough of it without injury.

A study published last week suggests that this advice is incomplete.

The study explored whether diversity in physical activity is important too. Apart from how active you are, can more variety in what you do help prevent chronic ailments such as cancer and cardiovascular disease?

The new study

This study, published on January 20 in BMJ Medicine, was led by Dr. Han Han and a team of Harvard researchers.

Han and colleagues tracked health outcomes among 70,725 women from the Nurses’ Health Study and 40,648 men from the Health Professionals Follow-Up Study.

One strength of the study is that starting in 1986, physical activity was surveyed every two years for up to three decades. This yielded good (though far from perfect) descriptions of each participant’s activities.

Han and colleagues wanted to know how these activities related to the development of cancer, cardiovascular disease, and respiratory disease among initially healthy participants. Time per week spent on each of the following was recorded:

• Walking

• Jogging

• Running

• Bicycling

• Swimming

• Tennis, squash, or racquetball

• Climbing flights of stairs

• Rowing or calisthenics

• Weight training or resistance exercises

Data on the following were also recorded by some participants:

• Light-intensity activities

• Moderate outdoor work

• Heavy outdoor work

• “Other”

I think this list provides a solid foundation for the research, although we could argue all day about what’s missing. (Where would you classify golfing when there’s no cart? What about basketball or flag football?)

Key findings

1. Quantity matters.

As you’d expect, more physically active participants tended to have lower rates of chronic disease and to live longer.

The effects were strongest at the lowest activity levels. In other words, people who would benefit most from a small increase in physical activity are those who are most sedentary to begin with.

This pattern is consistent with prior research and the HHS guidelines, which stress that even a little activity may be better than none.

2. Diversity matters.

Regardless of how active people were, those engaging in a greater variety of activities also tended to have lower rates of chronic disease and to live longer.

Here too, the strongest effects were seen at the lowest levels. Someone whose main source of exercise is walking, for instance, might benefit from spending part of their walking time on a different activity.

Some caveats

Before you revamp your exercise routine, please keep in mind that the effects of diversity were small and based on crude estimates of activity levels.

Specifically, Han and colleagues recorded how much time participants spent on each activity, but not the intensity of effort.

This is a critical omission. Take swimming, for instance. For some people, this means doing laps, varying their stroke periodically, and pushing themselves hard. For others, it’s water walking while moving their arms in gentle, swim-like patterns. A half hour of each would be radically different kinds of exercise.

My point is not that we should ignore the new findings – their overall consistency is part of what makes them persuasive – but simply that it’s risky to drill down to the level of individual activity and try to compare whether it’s better to, say, walk and cycle versus walk and do pickleball.

For the same reason, I don’t think it’s helpful to simply count activities, although I found it encouraging that small increases in variety were beneficial. If you regularly enjoy one particular kind of activity, adding just one more would be beneficial, according to the new data.

A minor mystery

The study hasn’t created much buzz in the news or on social media. Time, BBC, and National Geographic are among the few high-circulation outlets that covered it. The reportage tends to be good, although diversity was overemphasized (see the BBC headline below), and a few sources, like National Geographic, incorrectly noted that diversity proved more influential than quantity (headline: “Scientists identify one fitness habit that may matter most for longevity”.)

I was surprised by the minimal coverage. Harvard research published in major journals usually gets a lot of attention.

Plus, these particular findings seem newsworthy. Most health organizations in the US and abroad advise something similar to the HHS guidelines (at least 150 minutes of aerobic activity per week, along with muscle-strengthening activities on at least two occasions). The idea that diversity matters seems like an important extension of this advice.

In fact, the effects of diversity in the new study, though small, weren’t generally smaller than the effects of quantity. For instance:

• Compared to the one-fifth of participants who walked the least, those who walked more were 17% less likely to die early.

• Compared to the one-fifth of participants with the least variety in activities, those with greater variety were 19% less likely to die early.

To speculate for a moment, I can think of two reasons why the study didn’t create much buzz.

1. Exercise diversity is an abstract concept.

2. The activity-specific results may seem confusing.

For instance, the figures below show two of the findings for all-cause mortality.

The figure on the left shows that as amount of walking increases, the risk of early death declines and then levels off. In other words, walking is good for you up to a point, but more isn’t generally better. This makes sense – and it’s the kind of finding the media tends to pick up on.

The figure on the right seems wonky. It shows that as the amount of jogging increases, the risk of early death declines and then rises. Is this just noise in the data, or is this trend caused by people who overdo it and have heart attacks?

I think it’s just noise, but it would take a long time to explain why. The key takeaway is that across all activities, more is better – up to a point. Beyond that point, health risks still don’t match those that stem from being sedentary (unless an individual is genuinely overdoing it).

Why is exercise diversity beneficial?

Han and colleagues don’t speculate much, but in my view their one interpretive comment nails it:

“Consistent with previous studies, we found that the health benefits of most individual physical activities plateaued at certain levels. Individuals who engage in multiple physical activities may maximize the health benefits by engaging with each individual activity within the levels of the beneficial threshold.”

In other words, there may be something uniquely beneficial, physiologically speaking, about each activity.

I think we know this already, but the insight isn’t reflected in current health recommendations, because aerobic activities are treated as more or less interchangeable, as are muscle-strengthening ones (so long as you end up exercising all major muscle groups).

Insofar as you can trust the researchers’ interpretation, the practical implications are clear: Mix it up. Try to engage in a variety of aerobic and muscle-strengthening exercises if you don’t do so already.

HHS may have lost credibility on so many health issues this past year, but their guidelines remain a useful source of advice on physical activity. I suggest having a look (here) for advice on tweaking exercise and other activities to your particular needs.

Final thought

“Walking is man’s best medicine.”

Hippocrates is supposed to have said that. It’s a wise observation – unless you have strep throat or a broken ankle. What makes a medicine “best” is whatever best suits you and your particular needs.

The advice below, from the CDC website, sounds a little vague, but it’s consistent with a variety of studies, including the new one, and it’s probably more broadly useful than Hippocrates’ remark:

Indeed, some activity is better than none – and more variety may be better than less. That’s my takeaway from the new study.

Thanks for reading! (Take a break now, move around!)

Appendix: A technical note on MET scores

Another reason some news organizations may have skipped past the study is that the Hazard Ratios (which don’t usually scare anyone away) were based on metabolic rates and not just actual hours per week. As Han and colleagues noted:

“Each activity was assigned a metabolic equivalent task (MET) score, which represents the metabolic rate for that specific activity divided by the resting metabolic rate. The MET hours/week for each physical activity was derived by multiplying the average time spent on the activity (in hours/week) by its MET score. Total physical activity was defined as the sum of specific MET hours/week across all physical activities in each questionnaire cycle.”

If you’re interested, the authors provide an online supplement that shows the metabolic rate assumed for each activity. From these figures you can reverse engineer the actual number of hours each group of participants spent per activity. (I didn’t do this, because it’s potentially misleading, given that intensity of activity wasn’t recorded. True MET hours could vary greatly from assumed values. For the same reason, I didn’t try to relate the new study data to the HHS assumption that “moderate-intensity” activity can be equated with a MET score of > 1.5.)