By Justin Yamashita, MSc. Benchtop, site, CRO: three levels of basic and clinical research, explained without spin.

Plain Talk is Root to Rx's plain-language teardown: the myth, the evidence, and the one question that beats it.

In 2013 the FDA cut the recommended dose of one of America's most prescribed sleeping pills in half, for women only. Same pill, same insomnia, half the dose, because the original dose was never calibrated separately for a body that clears the drug more slowly.

Somebody forward you this? Root to Rx is free, three times a week, written by a clinical researcher. Subscribe at RootToRx.com, or help us grow and share it with someone you know.

In this week's Root Room, Sam sat in on a site selection meeting and watched two people disagree about where a trial should open. One had a map, and one had a calendar. Last issue we were a step earlier in the pipeline, on New Approach Methodologies (NAMs) and the fact that more than nine in ten drugs that look safe in animals still fail in people. That was about what a drug gets tested on. This is about who.

The Group Photo

A trial's enrollment data is a group photo. The FDA looks at that photo and decides the drug is safe and effective for everyone, with additional monitoring to be done after approval.

Nobody crops anyone out on purpose. It's simpler than that. Nobody walks around behind the camera to see who is standing just outside the frame, and then the photo gets printed and handed out as though it were the whole room.

Everything below is about how that shot gets framed, and who's standing at the edge of it.

Where You Put the Tripod

Here's the part almost nobody outside the industry has been shown.

The camera doesn't wander. Somebody sets it down. A protocol lists who can be in the study, and a feasibility team picks the sites, and sites get picked mostly on two things: can they enroll fast, and can they run clean data. That favors big academic medical centers, which favors people who live near one, can take a weekday morning off, have a car, and already have a specialist.

Nobody in that room says let us exclude that county. They set the tripod up in four cities, and the county is out of frame.

Then a coordinator screens whoever shows up. Some decline, while some are ineligible on a lab value. Some enroll and drop out. Some are entirely missed because the clinical research site is in the wealthier part of town, or only sees patients with insurance.

By the time the FDA sees a finished trial, the people in it are the output of maybe two hundred small decisions, almost none of which had anything to do with race or sex, and almost none of which looks unreasonable on its own. Hold onto that. You don't need anyone in the chain to be a villain to end up with a photo that doesn't look like the country.

Six Percent

Researchers went back through 341 late stage trials, the big final ones that decide whether a drug gets approved, all supporting FDA approvals between 2017 and 2023. They used the FDA's own published enrollment data and asked one question. Did the people in the trial match the US census for the four largest racial and ethnic groups, all four at the same time?

Twenty trials did. 6%.

Read as an indictment, that's damning, and it should be.

Then notice the second thing, which doesn't cancel the first. Somebody was able to count this. The FDA runs a program that makes sponsors publish who was in the trial. In most of the world you can't run this analysis at all, not because the gap is smaller there but because nobody has to tell you. Six percent is a bad number. It is also proof the number exists, and a number you can see is a number you can argue about.

Exposure

Photographers have a word for this. Exposure isn't how much light you point at the film. It's how much the film actually takes in.

Pharmacology uses the same word for the same reason. The dose is what goes in the mouth. The exposure is what the body actually ends up holding, and it depends on how fast you absorb, where the drug goes, how long it stays and how quickly you clear it. Body composition differs. Kidney filtration differs, and it declines with age. The liver enzymes that break drugs down come in inherited variants, and how common each variant is depends on ancestry. Pregnancy changes nearly all of it at once. So the same dose lands as a different exposure in different people, and sex is only one of the reasons.

None of that is a metaphor. It's plumbing.

A milligram isn't a milligram.

A 2020 review pulled the pharmacokinetic data on 86 approved drugs where sex specific numbers existed. For 76 of them, women showed higher exposure than men on the same dose, or took longer to clear it, or both.

The Case That Made It to the Label

Zolpidem is the one that reached the bottle. It's the active ingredient in Ambien, one of the most prescribed sleep medications in the country, and it was approved and taken for years at a standard dose.

Then the data on next morning impairment came in. Women were clearing it more slowly. For a meaningful number of them there was still enough in the bloodstream at breakfast to impair driving, and they felt completely awake.

In 2013 the FDA required the recommended starting dose for women be cut in half. Immediate release went from 10 mg to 5 mg. Extended release went from 12.5 mg to 6.25 mg. For men, the FDA recommended prescribers consider the lower dose too.

One thing before we go on, because this is where somebody does the math on their own bottle. Don't. A dose that is wrong on average across a population isn't the same as a dose that's wrong for you, and the only person who can tell the difference has your chart in front of them. If this makes you want to change what you're taking, that's a reason to raise it at your next appointment, not a reason to cut a pill in half tonight.

Read the timeline honestly, both directions. The safety net caught it. Post market surveillance, the ongoing tracking of what a drug does once it's on shelves, is exactly the system meant to catch what trials missed, and here it did. It also caught it years after the dose was in millions of medicine cabinets, and only because somebody finally broke the number out by sex. A net that catches you is better than no net. It isn't the same as not falling.

The Case That Reached a Different Label

Zolpidem is a story about sex. Here's the same shape of story about ancestry, and it ends somewhere more interesting.

Carbamazepine is a long established epilepsy and nerve pain drug. Rarely, it triggers a severe skin reaction that can be fatal. Researchers found that almost all of that risk sits with people carrying one particular immune system gene variant, and that variant is common in some populations and close to absent in others. In 2007 the FDA changed the label to recommend testing for it before starting the drug in patients whose ancestry puts them at risk.

Now the part that matters more than the drug does.

That variant is common in Han Chinese, Thai, Malay and Filipino populations. It is rare in Japanese and Korean populations. So the sentence "Asian patients are at higher risk from carbamazepine" manages to be wrong in two directions at once. It sweeps in millions of people who mostly don't carry it, and it still tells you nothing about the individual standing in front of you.

There's a second drug that finishes the point. Abacavir, an HIV medication, has its own dangerous reaction tied to a different variant of the same gene. That one is most common in people of European ancestry and least common in people of African and East Asian ancestry. The exact opposite gradient.

Two drugs. Two variants of one gene. Opposite directions.

Which is why the accepted practice for abacavir isn't to test the people who look like the risk group. It's to test everyone.

That's where this lands when it's done properly. You stop using ancestry as a rough stand in for a gene, and you test the gene.

What You Can Actually Do With This

Nothing here is medical advice and you shouldn't change a dose on your own. But there are two fair questions you're allowed to ask, and your pharmacist can usually answer both in about ninety seconds.

Was the dosing for this drug studied in people like me? And does this drug have a genetic test on its label?

The second one is not exotic. A handful of very ordinary drugs carry a testing recommendation in their prescribing information, and almost nobody is told they're allowed to ask about it.

You aren't asking anyone to change anything. You're asking what the number on the bottle was based on. If the answer is "I would have to look," that isn't a red flag. That's the whole point of this issue.

And reply and tell me: has anyone in your life ever skipped a trial, or skipped a treatment, because they didn't think it was meant for someone like them? I read every one of these.

The Reversal

That same 2020 review did one more thing. Of the 86 drugs, 59 had clinically identifiable sex biased adverse reactions, meaning one sex demonstrably got hit harder. The authors asked whether the exposure difference predicted which sex. It did, in 88 percent of cases.

Split that number and it stops being a story about a data gap.

Two limits, stated up front. Those 59 drugs were picked because they already had a known sex difference, so the set is selected on the very thing being measured. And women report adverse reactions more often, so a female biased reaction is likelier to get recorded at all. The male biased arm is small, so read the 29 percent as thin.

What survives is the direction, and the authors say it plainly. Sex differentiated exposure predicts adverse reactions in women, and they recommend evidence based dose reductions for women.

If you remember one sentence from this issue, make it this one. Being in the picture was never the finish line. Somebody has to look at your corner of it and set the dose from what they actually see there.

Now Run the Toolkit the Other Way

Every week this newsletter hands you questions to ask a health claim. Who paid for the study. How many people were in it. Was there a control group.

We aim those at supplement ads and wellness influencers, and that's fair, because that's where most of the bad claims come from. But a tool you only ever point in one direction isn't a tool. It's a flag.

So point it here. Ask a drug the same question you would ask a probiotic: who was actually in the study? Toolkit Q5 is am I applying the same standards to all claims. Here's what happens when you do.

The Light Meter Calibrated Somewhere Else

A light meter tells you what setting to use. It's only right if it was calibrated for the kind of light you are standing in. Point a meter calibrated for bright sun at a dim room and it won't shrug. It will hand you a confident number, and the number will be wrong, and you'll trust it because it came from a meter.

Warfarin is a blood thinner and getting the dose right is genuinely hard. Too little and you clot, too much and you bleed. So researchers built genotype guided dosing algorithms: read a few genes, predict the starting dose, skip some of the trial and error.

The algorithms used three genetic variants. Those three are the ones that carry the most weight in people of European ancestry, because those're the populations the algorithms were built on.

In 2013 a trial called COAG tested genotype guided dosing against ordinary clinical judgment. Overall it was a wash, about 45 percent time in the therapeutic range either way.

Then they looked at the African American participants separately.

The genotype guided group did worse. Thirty five percent of time in range against forty three percent for patients dosed by clinical judgment alone. Not equal. Worse, by eight percentage points, and statistically significant.

Sit with that, because it's not the story people expect. The usual version is that under represented groups get less benefit, which is bad but abstract. This is sharper. The tool built on a non representative population didn't merely fail to help. It underperformed doing nothing, with a genotype report attached, which is the most confident looking thing in medicine.

And warfarin doesn't work differently in Black patients. The drug is identical. The variants that matter most in West African ancestry are largely different ones, and they weren't in the algorithm, because they weren't in the cohorts it was fitted on. A model fitted where the causal variables are absent will mispredict where they are present.

That's not an equity argument. That is statistics, and it would be equally true built the other way around.

The Sentence the FDA Wrote Itself

You don't have to take my word for why this matters. FDA said it in its own guidance:

"Failure to include complex participants in a development program may lead to a failure to discover important safety information about use of the investigational drug in patients who will take the drug after approval."

Read it twice. Excluding people from a trial doesn't make the risk go away. It makes the risk invisible until after approval, at which point it gets discovered on people who aren't in a trial, aren't being monitored, and didn't agree to be the study.

That's the whole argument, and a regulator wrote it, not an advocate.

One Cousin Out of Frame

Now the honest case against enrollment targets, because it isn't stupid, and one part of it lands on my own evidence.

The operational objections first, quickly. Rare disease trials sometimes have a few hundred eligible patients on the continent, and a demographic target can mean waiting years to open a trial for people who don't have years. Some diseases genuinely aren't evenly distributed, so matching the census can mean matching the disease less well. And a target can be gamed, because enrolling to a number isn't the same as enrolling to a question.

Now the one that goes after my six percent.

Think about what that study actually measured. It counted a trial as a match only if all four of the largest racial and ethnic groups were in proportion at the same time. That's a family photo that only counts if every single cousin is fully in frame. One elbow out and the whole picture fails, and it scores exactly the same as a photo with half the family missing.

Stack four conditions and the pass rate collapses on arithmetic alone, before anybody behaves badly. On top of that, many pivotal trials enroll across several countries, and holding a global trial to the US census isn't obviously the right comparison. And the census may be the wrong yardstick anyway: if you care whether a drug works, the population you want to match is the one that has the disease.

Here's where I land. The operational objections are all real and none of them explains six percent, because they describe hard cases and six percent describes the normal case. The yardstick objection is different. It's at least partly correct, and I want you holding the number more loosely because of it.

What survives, and I'm not giving this up, is the direction of travel. Black and Hispanic enrollment declined across the same period, measured the same way every year in the same dataset. The level is arguable. The slope isn't.

The Rule and the Instruction Sheet

Two questions get worn as one coat, and separating them is most of the argument.

There's a rule that says take the photo. And there's an instruction sheet that tells you where to stand.

The rule is law. In 1993 Congress required that women and minority groups be included in federally funded research, required trials be built so you can actually check whether results differ by group, and said in writing that cost is not a permissible reason to skip either one. That line is in the statute because cost was the standing excuse. In 2022 Congress went further and required sponsors of certain trials to file a diversity plan before they enroll.

The instruction sheet is guidance, and guidance isn't law. It's the document that tells companies what a passing grade looks like. FDA published a draft of it in June 2024, the deadline to finish it passed in mid 2025, and it's still sitting there marked draft. In December 2025 FDA finalized a different guidance in the same space, retitled from "Enhancing the Diversity of Clinical Trial Populations" to "Enhancing Participation in Clinical Trials." Legal analysts who've read it report the diversity framing is gone from the text. That's their reading, not mine.

The rule didn't change. The instruction sheet did.

Answer the instruction sheet question and think you answered the rule question, and you'll tell people the requirement is gone. It isn't. Answer the rule question and think you answered the other, and you'll tell people it's handled. It isn't that either.

The One Question: Skeptic’s Toolkit Q7

Q7. Do I want this to be true, or do I want to know what's true?

Run it both ways, and be fair about it.

If you want "nothing has ever changed" to be true, then the 1993 statute is inconvenient, the zolpidem label change is inconvenient, and the fact that anyone can count to six percent at all is very inconvenient.

If you want "it's all fixed now" to be true, then the six percent is inconvenient, the unfinished guidance is inconvenient, and the 96 percent is extremely inconvenient, and so is the fact that fewer than one in ten drugs approved since 1980 have enough information to say whether they are safe to take while pregnant.

The photo has more people checking its edges than it used to. It still doesn't match the room. Anybody selling you only one of those halves is selling you the version that flatters you.

Run it on yourself first. Got Skepticism is our two minute game on catching the version of a claim that flatters you. Free, no signup, at openlabelmedia.com/games.

Next issue: the room you are not allowed to sit in, and what an independent board can do to a trial while it is running, without ever telling you.

Refer a friend

You just got the dinner table version. Forward this to the woman in your life who takes a sleeping pill, and you're on your way to a Series 1 set.

Thursday

The Informed goes out Thursday with the sex based dosing data, the full regulatory timeline through December 2025, and every citation, so you can check my work. It comes to you automatically. Nothing to turn on.

References

1. NIH Revitalization Act of 1993, Public Law 103-43, signed June 10, 1993. Inclusion provisions sections 131 to 133, adding section 492B to the Public Health Service Act, 42 U.S.C. 289a-2. congress.gov

2. Zaaijer S, Groen SC. "Longitudinal clinical trial enrollment trends across 341 US FDA-approved drugs and their guiding role in precision medicine strategies." Communications Medicine. 2025;5:514. doi:10.1038/s43856-025-01270-2

3. Zucker I, Prendergast BJ. "Sex differences in pharmacokinetics predict adverse drug reactions in women." Biology of Sex Differences. 2020;11:32. doi:10.1186/s13293-020-00308-5

4. Food and Drug Omnibus Reform Act of 2022 (FDORA), part of the Consolidated Appropriations Act, 2023, Public Law 117-328. congress.gov

5. U.S. Food and Drug Administration. "Diversity Action Plans to Improve Enrollment of Participants from Underrepresented Populations in Clinical Studies." Draft guidance for industry, June 2024, docket FDA-2021-D-0789. fda.gov

6. U.S. Food and Drug Administration. "Enhancing Participation in Clinical Trials: Eligibility Criteria, Enrollment Practices, and Trial Designs." Final guidance, December 2025, docket FDA-2019-D-1264. fda.gov

7. U.S. Food and Drug Administration. "Questions and Answers: Risk of next-morning impairment after use of insomnia drugs; FDA requires lower recommended doses for certain drugs containing zolpidem." fda.gov

8. National Institutes of Health. NOT-OD-25-131, revised policy on inclusion of women and members of racial and ethnic minority groups in clinical research, July 17, 2025. grants.nih.gov

9. Kimmel SE, French B, Kasner SE, et al. "A pharmacogenetic versus a clinical algorithm for warfarin dosing." New England Journal of Medicine. 2013;369(24):2283-2293. doi:10.1056/NEJMoa1310669

10. U.S. Food and Drug Administration. "Enhancing the Diversity of Clinical Trial Populations: Eligibility Criteria, Enrollment Practices, and Trial Designs." Final guidance, November 2020. fda.gov

New here? Root to Rx runs three pieces a week. Monday you watch a real health claim try to work on someone. Wednesday you get the tools to run it yourself. Thursday you get the full evidence so you can check my work. Written by a clinical researcher, free, at RootToRx.com.

For educational purposes only. Nothing in this newsletter is medical advice. Talk to your doctor before making any health decision.

Instagram: instagram.com/roottorx | LinkedIn: linkedin.com/company/roottorx | Facebook Lab: facebook.com/groups/theroottorxlab | Subscribe: RootToRx.com

You are receiving this because you subscribed at roottorx.com.

Disclosure: Root to Rx is published independently by Open Label Media LLC. Views expressed are personal views of Justin Yamashita and do not represent his employer or any affiliated organization. No employer resources or proprietary information are used. Every claim is sourced from publicly available materials.

Root to Rx | 228 Park Ave S, #29976, New York, New York 10003, United States