Eleanor Crane: Ben, hey — how was the train this morning, did it actually run on time?
Ben Okonkwo: It did, which threw me completely. I had twenty extra minutes and I just sat there reading about warfarin dosing, so.
Eleanor Crane: That tracks. Actually — that's almost a perfect way into today, because I had this moment this week. My aunt called me, she's seventy-two, her cardiologist just adjusted her cardiac medication, and her daughter — my cousin — Googled the standard dose, found it, saw my aunt was prescribed exactly that amount, and felt reassured. And I just... I couldn't stop thinking about what my cousin didn't see. My aunt weighs about a hundred and ten pounds. Her kidney function has been declining slowly for two years. The dose on the screen was correct — by every population standard, it was correct. For my aunt specifically, on a Tuesday morning when she's supposed to increase to full dose, it's something else entirely.
Ben Okonkwo: Wait — the dose was right, and that's the problem.
Eleanor Crane: That's the whole thing, yes.
Ben Okonkwo: Right, because the number on the prescription isn't actually her number — it's a statistical average. It's the dose that works for the population the trial enrolled. And if she wasn't in that population, or someone like her wasn't... the window the drug was calibrated against may not be her window at all.
Eleanor Crane: And that's what we're trying to work out today — not whether the dose-response curve is wrong, because it isn't, it's elegant and it's real. But whether we've been treating a population tool as though it were a personal prescription. The therapeutic window is a range built on a sample. And for a very long time, women were largely not in that sample.
Ben Okonkwo: Mm, and the shape of the curve itself — the sigmoidal S-curve, little effect at low doses, steep rise through the middle, plateau as receptors saturate — that shape is mechanistically sound. Receptor binding and saturation explains it cleanly. The problem isn't the model. The problem is what population you used to find where on that curve to put the therapeutic window.
Eleanor Crane: So women have been prescribed the same doses as men for decades — different body compositions, different metabolisms — because the window was calibrated on trials that largely excluded them. That's not a political point. That's a measurement problem.
Ben Okonkwo: It's a systems problem, honestly. We have the tools to do this more precisely. The question is why we haven't.
Eleanor Crane: And that systems problem has a shape, actually — a literal shape. Which is what I want to understand from the ground up. Because the S-curve keeps getting mentioned and I want to know why it's S-shaped. Not just that it is.
Ben Okonkwo: Okay, so — think of it this way. Every drug is basically a key looking for a lock. Your cells have a fixed number of locks. That's it. That's the whole story.
Eleanor Crane: A fixed number.
Ben Okonkwo: Fixed. So at low doses, not enough keys are finding locks to do anything meaningful — you're below the threshold, nothing happens. In the middle range, you're opening more locks, you get the effect you want. And then eventually every lock is taken. You've saturated the receptors. And any extra keys you add aren't opening anything. They're just... rattling around. And that rattling is where the damage comes from.
Eleanor Crane: So the ceiling on the effect isn't a dosing choice. It's biology. You literally run out of receptors.
Ben Okonkwo: Exactly that. The plateau at the top of the S isn't a design decision — it's saturation. The receptor binding sites are occupied. That's what gives you the shape.
Eleanor Crane: So is there a number for how much room you have — between where the drug starts working and where it starts causing harm?
Ben Okonkwo: There is, yeah. It's called the therapeutic index. TD₅₀ divided by ED₅₀ — the median toxic dose over the median effective dose. High ratio, wide margin, you have room to maneuver. Narrow ratio... small errors become serious very fast. Digoxin, warfarin, lithium — those are narrow. The effective dose and the toxic dose are almost on top of each other.
Eleanor Crane: And this wasn't derived in a laboratory in the twentieth century. Withering was watching this — at bedsides — in 1785. With digitalis.
Ben Okonkwo: Right — William Withering, documenting the therapeutic use of digitalis, and what he was actually observing, without the vocabulary for it yet, was dose-dependence. The dose that helped the heart and the dose that stopped it were not far apart. He didn't have the word receptor. He had patients. And the pattern was unmistakable.
Eleanor Crane: That's — I mean, that's the therapeutic index before anyone had named it.
Ben Okonkwo: Mm. And here's what I find — I mean, I find this genuinely uncomfortable once you sit with it. The elegant thing about the S-curve is also the dangerous thing. It makes it look like there is one curve. For everyone. Clean, universal, predictive. But the curve is derived from a population. So what you're actually holding is an average curve — and any individual patient may sit somewhere quite different on it.
Eleanor Crane: Which brings us straight back to my aunt. Prescribed the right dose — by the population curve. Living somewhere else on it entirely.
Ben Okonkwo: And that's exactly where the axiom starts to wobble — because the whole framework we're using, the therapeutic index, the window, all of it, rests on something that was written as a defense, not a discovery.
Eleanor Crane: Written as a defense — say more.
Ben Okonkwo: Paracelsus. 1493 to 1541. The 'dose alone makes the poison' line — Sola dosis facit venenum — it comes from something called The Third Defence, written in the 1530s. And he was not writing a treatise. He was under attack from herbalists who thought he was poisoning patients with metallic compounds — mercury, sulfur, arsenic. Things everyone agreed were toxic. So he argues: nothing is inherently a poison, it's the dose that determines harm. That's the move. It's rhetorically brilliant.
Eleanor Crane: He wrote it in German, not Latin.
Ben Okonkwo: He did — which tells you who the audience was. Not the academy. The people who already thought he was dangerous.
Eleanor Crane: So the most-quoted principle in toxicology — the one that underpins how we think about every drug dose — was an argument made to win a professional dispute. And we've been treating it as self-evident truth ever since.
Ben Okonkwo: Now — and I want to be careful here — the rhetorical origin doesn't make it wrong. The underlying observation is sound. But there's a difference between 'this survived scrutiny' and 'this closed the question.' We've been using it to close questions it was never meant to answer.
Eleanor Crane: Like whether the dose that's safe for a population is safe for one specific person.
Ben Okonkwo: Right. And that's where warfarin, lithium, digoxin — those narrow-index drugs — that's where the fragility becomes, I mean, it becomes clinical. It's not theoretical anymore. The effective dose and the toxic dose for warfarin are so close together that the therapeutic drug monitoring requirement exists precisely because the population number cannot tell you what this patient's body will do with it.
Eleanor Crane: And my aunt's kidney function matters enormously there — because if her kidneys aren't clearing the drug at the rate the trial assumed, the concentration builds.
Ben Okonkwo: It builds without anyone changing the dose. That's the — that's what I find genuinely unsettling about it. The prescription didn't change. Her pharmacokinetics did. How she absorbs, distributes, metabolizes, excretes — all of that shifts with age, with kidney decline, with whatever else she's taking. The window the drug was calibrated against is a population window. Her window is narrower. Possibly much narrower.
Eleanor Crane: And that statin interaction — the enzyme inhibition — that's a whole other layer we haven't even touched yet, which, actually, is where this gets worse before it gets better.
Ben Okonkwo: Much worse. Polypharmacy, enzyme induction, antibody-drug conjugates trying to engineer their way around all of this — that's the next wall we hit.
Eleanor Crane: So we've been building dosing frameworks on a foundation that is part science, part polemic — and applying them to populations Paracelsus never imagined and trials never enrolled. That's the unease I want to sit with for a second before we go further.
Ben Okonkwo: And the wall hits faster than you'd think. Take warfarin — same 5 mg a patient has taken for three years, nothing on the prescription has changed. His daughter starts him on cranberry supplements for a UTI. Cranberry inhibits CYP2C9, the enzyme that metabolizes warfarin. The dose didn't move. The concentration in the blood did.
Eleanor Crane: So the drug interaction moved him along the dose-response curve without anyone touching the prescription.
Ben Okonkwo: Exactly. And that's — I mean, that's why the FDA mandates exposure-response analysis. Not dose-response. Exposure-response. Because the number on the pill is not the number in the blood. Those are two different quantities, and we conflate them constantly.
Eleanor Crane: Wait — the FDA requires that measurement specifically?
Ben Okonkwo: Required in drug development, yeah. Blood concentration over time linked to both efficacy and safety outcomes. Because polypharmacy multiplies this — every additional drug is another potential interaction, enzyme induction dropping concentrations below the effective threshold, inhibition pushing them into toxic range. Each one independently probabilistic. Together they compound.
Eleanor Crane: And that's the thing my cousin couldn't Google. The number she found was the prescribed dose. The actual exposure — my aunt's real pharmacokinetics with her kidney decline and her statin — that number exists nowhere on a label.
Ben Okonkwo: Nowhere. So there have been real attempts to engineer around this that genuinely strike me as strange — Antibody-drug conjugates. ADCs. Targeted cancer therapies, designed from the ground up specifically to widen the therapeutic window through targeted delivery. That was the whole promise.
Eleanor Crane: And do they?
Ben Okonkwo: FDA approvals accelerated after 2019, so there's real-world data now. And their windows are still narrow — narrower than most other targeted therapies. The principle ran straight into biology and biology... didn't yield.
Eleanor Crane: That's — I find that almost clarifying, actually. Even when you design the system to solve the problem, the problem is still there.
Ben Okonkwo: Right — but the part that doesn't fit is that people then reach for hormesis to explain low-dose surprises. The biphasic curve, stimulatory at low doses, inhibitory higher up. And I want to be honest about what that is right now, because — actually, the lab data are interesting. Mitohormesis specifically, low-level mitochondrial reactive oxygen species activating adaptive pathways — there's a plausible mechanism at the cellular level.
Eleanor Crane: Mitohormesis at least has a mechanistic story, though.
Ben Okonkwo: It does. And I'm not dismissing it. But a cellular mechanism is not a human epidemiological dataset. The convincing population-level data that would move hormesis from hypothesis to policy-relevant ground truth — that database doesn't exist yet. And we keep reaching for it when we want to defend low-dose exposures to things that sound scary. That's not evidence reasoning.
Eleanor Crane: So the honest position is: keep investigating, don't promote it.
Ben Okonkwo: Don't let it do policy work before it's earned it. Yes.
Eleanor Crane: And that's — I mean, that sits right next to everything we've been saying. The dose-response curve is real. The therapeutic window is real. And everywhere you look, something is making the actual dose different from the prescribed one — an enzyme, a supplement, a second drug, an aging kidney — without anyone knowing.
Ben Okonkwo: The prescription is a starting estimate. The body revises it continuously. And we're only just building the tools to track the revision in real time.
Eleanor Crane: And that's the part I keep — well, I can't quite put it down. We have the genetic tools. The pharmacokinetic modeling. The exposure-response frameworks the FDA already requires in development. We know that absorption, distribution, metabolism, excretion vary individually — meaningfully, not at the margins. And yet warfarin is still dosed from a population therapeutic index. Lithium is still dosed from a population therapeutic index. Digoxin. We're not choosing population averages because we lack the alternative. We're choosing them because the alternative is slower and more expensive and the system isn't built for it.
Ben Okonkwo: That's the honest version of it. The dose-response curve is true — it just describes a person who doesn't exist. The average patient. Someone assembled from a population distribution who has never actually walked into a clinic.
Eleanor Crane: And Paracelsus wrote the axiom to win an argument. The therapeutic index is a population statistic. Drug interactions can silently shift someone out of their window without the prescription changing at all. So what exactly are we confident about when we write a dose? I don't think that's rhetorical. I think it's a real question that hasn't been answered yet — not at scale, not in practice.
Ben Okonkwo: It's an uncomfortable place to end up. But I think it's the accurate one. That's — I mean, that's where the evidence actually leaves us.