Protein Powder, Alcohol, and the "No Safe Amount" Bias

My inner 8-year-old would like a word.

Last Tuesday, Consumer Reports identified “concerning” levels of lead in more than a dozen protein supplements. In response, some experts warned us, again, that there are no known safe amounts of lead exposure.

The next day, The Washington Post reported on a new study that draws a similar conclusion about alcohol.

My inner 8-year-old, being overly literal, gets frustrated by “no safe amount” claims.

No expert would argue that a single molecule of lead or alcohol can harm you. In fact, both Consumer Reports and the authors of the new study suggest that occasional intake is fine.

So, what does it mean exactly to claim there’s “no safe amount” of exposure? Why the mixed messaging?

My inner adult is also preoccupied with more practical questions: How much protein powder can I safely consume? What about alcohol?

In this newsletter I’ll be addressing these questions. At the end I’ll touch on the origins of “no safe amount” biases and whether they’re truly helpful.

Protein powder and lead

Consumer Reports tested 23 protein powders and ready-to-drink shakes. One of the key findings was that 16 products contain more than 0.5 micrograms of lead per serving, the maximum amount that can be safely consumed in a day.

(In case you’re wondering, 0.5 micrograms is about half of one percent of the weight of an average eyelash.)

CR took this 0.5 µg maximum from California’s Proposition 65 standards, which are famously conservative. In particular, the 0.5 µg value was chosen because it’s exactly one thousand times lower than the maximum known safe levels. (More about what this means in the Appendix.)

Let’s take a worst-case scenario perspective. Of the 23 products that CR reviewed, Naked Nutrition’s Vegan Mass Gainer was the worst offender. CR reported that one 315 gram serving of this product has 1,572% the maximum amount of lead one can safely ingest per day.

1,572%. Sounds terrifying, right?

Not to be cynical, but I suspect that CR drums up business through its reliance on Proposition 65 standards and scary statistics. That 1,572% figure is nowhere near as awful as it sounds.

1,572% of 0.5 mg equals 7.86 mg. This is less than the FDA’s 12.5 mg per day limit, which the FDA takes to be one tenth the amount for which there’s evidence of harm (see Appendix for details).

Of course, protein powder isn’t the only source of lead in our diets. FDA data shows that adults consume between 1.7 and 5.3 µg per day, on average. If you’re at the upper end of that average (5.3 µg per day) and you have one serving per day of CR’s worst offender (7.86 µg), your daily total would be 13.16 µg, slightly more than FDA’s benchmark of 12.5 µg. No other product reviewed by CR would put you over that 12.5 µg value. And, you’re still way below the lowest levels that studies find harmful.

Not quite so terrifying, right?

A caveat

My point here is not that we should ignore data on lead content. The FDA also sets benchmarks of 8.8 µg per day for pregnant women and 2.2 µg per day for children. (Again, these are one-tenth the values at which the FDA sees evidence of harm.) The maximum for children illustrates one reason for caution.

On average, Americans between 7 and 17 consume 1.4 to 4.0 µg of lead per day. The FDA assumes that kids begin to experience health risks at 22 µg.

Thus, if a teenager consumes about 4.0 µg of lead per day through their regular diet, plus a full serving of Naked Nutrition’s product, the total lead intake would be 4.0 + 13.5 = 17.5 µg per day. That’s close to the 22 µg value.

If you’re a parent, the implications are clear: Check protein powder brands, don’t give kids too much, and rely as much as possible on whole, heathy foods for protein.

The “no safe amount” bias and lead

Wait, says my inner 8-year-old. All the major health agencies treat small amounts of dietary lead as safe, at least up to a point. At the same time, all of these agencies say there’s no safe amount of lead exposure. Isn’t that contradictory?

The answer to that question depends on whether you want to play lawyer or scientist.

From a legalistic perspective, sure, you can’t have it both ways. Exposures below a certain amount either increase health risks or they don’t.

Scientifically, the issue is more complicated. Over the past five decades, harmful effects of lead have been detected at ever lower exposures (and lower blood concentration levels – see Appendix for details).

Thus, “no safe amount” claims about lead are biased, in the sense of being extrapolations beyond current data, but they’re not unreasonable. The health agencies are just being excessively cautious.

Practical advice

Consumer Reports’ study on protein powders is available here, but I wouldn’t recommend fussing over it.

Instead, I’d just consider avoiding the four worst offenders. Not because of the lead, but because these companies aren’t as good as their competitors at keeping their products free of contaminants. Here are those four products:

• Naked Nutrition’s Vegan Mass Gainer (Vanilla)

• Huel Black Edition (Chocolate)

• Garden of Life Sport Organic Plant-based Protein (Vanilla)

• Momentous 100% Plant Protein (Chocolate)

Again, my concern is quality control, not lead. And I suggest keeping three other points in mind as well:

—Lead is a natural part of our environment as well as a by-product of human activity. Americans now have at least 90% less of it in our bloodstreams than we did prior to the 1970s, thanks to the elimination of lead from gas, house paint, and other products. We should minimize exposure but not expect to reach zero.

—Because protein powders are supplements, they’re not evaluated or tested by the FDA prior to being sold. At best, Consumer Reports gives you a snapshot of industry-wide contamination levels. Buyer beware.

—You should probably get most of your protein from whole, healthy foods. (I discuss the merits of different protein sources and how much you might need in a day here.)

Now for something (almost) completely different....

Alcohol and dementia

In a study published in the newest edition of BMJ Evidence-Based Medicine, Anya Topiwala at Oxford and an international team looked at whether drinking alcohol increases the risk of dementia. The researchers analyzed data on more than 559,000 older adults currently being tracked in two huge longitudinal studies, the U.S. Million Veteran Program and the UK Biobank.

Topiwala and colleagues concluded that there’s no safe amount of alcohol intake, but I disagree. I see another kind of “no safe amount” bias at play here.

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 a symptom that can arise from many causes, 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.

Context

Some studies suggest that a little bit of alcohol is good for you – e.g., abstainers and light drinkers have lower rates of dementia than moderate drinkers do. This relationship is captured in the figure below.

Starting from the far left, the figure shows that people who drink the least have a slightly higher risk of dementia than people who drink a little bit. Then the curve swings upward and, as you’d expect, the more a person drinks, the greater the risk of dementia.

Topiwala and colleagues describe a number of problems with the studies that have revealed this pattern. For instance, some people labeled as abstainers are actually former drinkers who quit. Some of them had been heavy drinkers for decades. As a result, they’re more likely to develop dementia than people who have engaged in light to moderate drinking all their lives (i.e., the lowest part of the line in the graph).

In other words, dementia risk may be greater at zero alcohol intake than at levels slightly higher than zero, because some of the folks in the zero-intake group are already damaged.

If this is true, then the figure is misleading. Maybe drinking a little isn’t really protective.

Indeed, the researchers found that the more people drink, the greater their risk of dementia. For instance, an increase from 1 to 3 alcoholic drinks per week heightened the risk of dementia by 15% over a roughly 4 to 12 year period.

The researchers concluded that there’s no safe amount of alcohol consumption. Even light drinking increases the risk of dementia compared to total abstinence, as the figure below illustrates.

Should light drinkers be worried?

Good statistics can’t offset bad measures

As I explain in the Appendix, Topiwala and colleagues used a Mendelian randomization (MR) methodology that relies on genetic markers as stand-ins for alcohol intake.

In the process of validating those markers, researchers have to ask people how much they drink. There’s no way around this. Regardless of whether a study relies on observational methods or MR, actual drinking behavior must be measured. The results are messy, and small quantities of consumption get overestimated.

Why is that?

The standard approach is to ask people on one or a small number of occasions how much they drink on a “typical” day, or how many drinks they had last week, or something comparable.

This is an inherently imprecise approach to quantifying alcohol intake:

(a) People don’t routinely measure how much they drink.

(b) Within each category (e.g., beer), products differ in alcohol content, and drinkers may or may not finish their drink on occasion.

(c) People may forget, or misrepresent how much they drink.

(d) Drinking behavior is measured once, or on a small number of occasions, and from these data lifetime patterns are inferred.

This isn’t just messy. At the lowest levels of intake, group data tend to overestimate rather than underestimate how much people drink. This is just simple math. Among people who say they typically have one drink per week, the actual number might be higher, but it can’t get much lower.

In sum, estimates of drinking behavior are imprecise, and they tend to overestimate intake at the lowest levels. This undercuts Topiwala and colleagues’ “no safe amount” claims.

The “no safe amount” bias and alcohol

It’s clear that heavy alcohol use increases the risk of dementia (and other problems). I’m not denying that. The new study is one of many that show this.

My point is simply that it’s unclear from the data how much alcohol people can safely consume. The new study’s “no safe amount” claim is biased, in the sense of being an unwarranted inference from weak data on the lowest levels of intake.

Practical advice

I haven’t seen credible evidence that a drink or two per day, on average, is risky, as long as the person avoids binge drinking. If you live healthy (eat and sleep well, exercise, avoid smoking, etc.), light drinking seems unlikely to create health risks – particularly if the alcohol helps reduce stress. I discuss the issues at length here.

At the same time, health advice should always be tailored to the individual. Certain medications and medical conditions rule out light drinking. And there are people like my father, who was an alcoholic and helped quite a few bartenders put their kids through college. For people like him, there probably wouldn’t be a safe amount.

Final thoughts

“No safe amount” biases can be seen for lead, alcohol, processed meats, trans fats, and more. Why are these biases so prevalent?

I respect the fact that in many, if not most cases, scientists and health care professionals are just being cautious. I appreciate the wisdom of that. A single molecule of fentanyl wouldn’t hurt you, but that’s a purely theoretical observation. Out in the real world, we need to keep telling people: There is no safe amount of fentanyl intake.

Apart from an abundance of caution, I see three other reasons for the prevalence of “no safe amount” biases:

1. Profit.

There’s money to be made in persuading people to switch to safer products (e.g., non-alcoholic beer). In the case of Consumer Reports, you might say that overstating risks helps attract subscribers.

2. Overextrapolation.

Scientists don’t always have usable data on the lowest levels of exposure to lead, or alcohol, or whatever. Instead, they model the relationship between exposure and health, then extrapolate to minimum exposures. That’s risky. If you reduce your sodium intake and your blood pressure improves, then cut back more and your blood pressure improves more, it doesn’t follow that if you keep cutting down on sodium, your blood pressure will keep improving.

3. Overconfidence.

As the new study illustrates, researchers are sometimes overconfident about their data. I don’t blame them. It’s important to figure out whether light drinking is risky. If you think it is, you should say so. I just don’t agree.

Thanks for reading!

---

Appendix: Some technical details

This is very much in the weeds, though written in plain English.

1. Measuring acceptable levels of lead exposure.

In their new study, Consumer Reports used California’s Maximum Allowable Dose Level (MADL) of 0.5 µg of lead intake per day. To calculate this MADL, California’s Office of Environmental Health Hazard Assessment (OEHHA) identified the highest amount of lead any study has shown to not increase risks of reproductive harm. OEHHA then divided that amount by 1,000 in order to provide a margin of safety.

This is extremely conservative. The MADL is one one-thousandth of the maximum known safe level, based on the most sensitive study in the literature (involving either humans or animals). Also, the focus is on reproductive harm. The corresponding amount for carcinogenicity is three times higher.

FDA standards are less conservative but still more stringent than, say, those of the European Union. The FDA uses what they call an interim reference level (IRL) to indicate an amount of lead exposure that would be cause for concern and possibly action if exceeded. The FDA sets the IRL at one-tenth the lowest value at which health risks have been observed in the scientific literature. For non-pregnant adults, that value is 12.5 µg of lead per day. Unlike California’s MADL, this value is derived from the literature rather than from the most sensitive individual study.

To compare California’s MADL for lead to the FDA IRL, multiply the MADL by 1,000 and the IRL by 10. This results in values of 500 µg per day (California) and 125 µg (FDA). These are the values each agency assumes to be the lowest exposures known to increase risk of harm to adults. The main reason the two numbers differ is that California works from a single study while the FDA works from the literature.

None of these numbers can be considered exact. One reason is that over the past five decades, as instrumentation has improved, maximum safe levels have been repeatedly revised downward. Another reason is that studies linking the lowest levels of lead exposure to health risks rely on blood lead levels (BLLs). There’s no simple way of translating how much lead one is exposed to into a BLL, as the amount we absorb from our food depends on age (very young people tend to have higher rates of absorption), and a bunch of other individual-level variables, including fullness (emptier stomachs promote more absorption).

An additional nuance is that BLLs themselves are snapshots: A person takes a blood test and the amount of lead in their bloodstream at that moment is then estimated. But lead accumulates for months in soft tissue and decades in bone and is then released into the bloodstream over time, thereby potentially damaging a person’s health.

2. Mendelian randomization

Observational health studies that track people over long periods of time are inherently messy, as I discussed earlier.

Mendelian randomization (MR) sidesteps the mess by linking genetic variants that predict drinking behavior to long-term health outcomes. Rather than trying to measure actual drinking, MR researchers estimate population-wide distributions of alcohol intake on the basis of genetic variants (typically groups of single nucleotide polymorphisms, or differences in one DNA nucleotide base within a gene).

There’s much more to MR than this, but here’s a key issue: Genetic predictors of drinking behavior have to be validated at some point. Researchers need to have measured peoples’ actual drinking behavior in order to know which genetic predictors to choose.

This brings us right back to where we started. As I mentioned, conventional approaches to measuring alcohol consumption over long periods of time yield messy data and overestimates for the lowest levels of consumption. Using genetic variants in place of behavioral measures doesn’t solve the problem but rather conceals it.