Case Study 2 — The Arithmetic of Scaling a Rodent Finding

This case study is a [constructed teaching example] throughout. The compound is invented, the numbers are in abstract units, and nothing here is a dose for anything. The point is the shape of a calculation and the shape of its failure. If you find yourself wanting to substitute a real compound and a real number, that impulse is precisely the one this case study exists to interrupt.

The setup

A reader — call them the reasoner, since this book does not use named characters — finds a rodent paper. It is a real-looking paper: peer-reviewed, competently designed, reporting a tissue effect in mice at an administered dose of 10 units per kilogram of body mass, given daily for four weeks.

The reasoner weighs 70 kilograms. They would like to know what the corresponding human figure would be. They do the obvious thing:

STEP 1 — THE NAIVE CALCULATION

  Reported in mice:     10 units / kg
  Reasoner's mass:      70 kg
  Naive conversion:     10 × 70  =  700 units

  "The mouse got 10 per kilo. I weigh 70 kilos. So 700."

This is wrong, and it is wrong in a direction that can be predicted before any arithmetic is checked.

Why the naive calculation fails

Body mass is not the variable that governs how fast a body processes anything.

Metabolic rate scales sublinearly with mass — roughly to the three-quarters power, a relationship described in the 1930s and generally called Kleiber's law. A mouse is not a small human running at human speed. Per unit of mass, it burns energy faster, breathes faster, filters faster, and clears compounds faster. Everything downstream of metabolic rate follows.

So a milligram per kilogram in a mouse and a milligram per kilogram in a human do not produce comparable exposures. They produce comparable numbers, which is a different thing entirely.

Standard regulatory practice, when estimating a maximum safe starting dose for a first-in-human study, normalizes not to body mass but to body surface area, which tracks metabolic rate more closely. In practice this is implemented as a table of species-specific correction factors:

STEP 2 — THE STANDARD CORRECTION

  APPROXIMATE DIVISORS (animal mg/kg → human-equivalent mg/kg)

    mouse   ÷ 12
    rat     ÷  6
    monkey  ÷  3
    rabbit  ÷  3
    dog     ÷  2

  Notice the pattern: the divisor SHRINKS as the animal gets larger.
  A dog is metabolically closer to a human than a mouse is, so less
  correction is required. This is the whole logic in one column.

Apply it:

STEP 3 — THE CORRECTED CALCULATION

  Mouse dose:               10 units / kg
  Mouse-to-human divisor:   ÷ 12.3
  Human-equivalent:         10 ÷ 12.3  ≈  0.81 units / kg
  For a 70 kg person:       0.81 × 70  ≈  57 units

  NAIVE     : 700 units
  CORRECTED :  57 units
  RATIO     : the naive figure is about 12× too high

The direction of the error is fixed

This is the part worth committing to memory, because it converts a technical detail into a usable heuristic.

The error is not noise. It does not sometimes run one way and sometimes the other. Naive milligram-per-kilogram conversion from a smaller animal to a larger one always overestimates, because the smaller animal clears faster relative to its size and therefore needed more per kilogram to achieve the same exposure. The smaller the source animal, the larger the overestimate.

THE ERROR IS DIRECTIONAL, NOT RANDOM

  source animal        naive error vs. corrected
  ────────────────────────────────────────────────
  mouse                ~12× too high
  rat                  ~6× too high
  rabbit / monkey      ~3× too high
  dog                  ~2× too high

  There is no source species for which naive conversion undershoots.

Which means that whenever you encounter someone reasoning from a rodent paper by multiplying through by body weight, you already know the sign of their error without checking anything else.

The second error, which is worse

Suppose the reasoner learns all of the above and redoes the arithmetic correctly. They now have 57 units instead of 700. Are they in good shape?

No — and this is the part that almost nobody gets to, because the arithmetic error is satisfying to find and people stop once they have found it.

The surface-area method estimates a maximum safe starting dose for a first-in-human safety study. That is its stated purpose. It is designed to answer the question: what is a quantity we can be confident is low enough not to hurt the first volunteer? An additional safety factor is applied on top of it before anyone is dosed at all. The output is a floor, chosen to be conservative, produced by a process whose failure mode is being too cautious.

It is not an estimate of an effective dose. It was never intended to be, and the entire logic of the method points the other way.

WHAT THE METHOD IS FOR vs. WHAT IT IS BEING USED FOR

  DESIGNED FOR                         BEING USED FOR
  ──────────────────────────────────────────────────────────────────
  "What is safely low enough to        "What amount will produce the
   give a healthy volunteer once?"      effect I read about?"

  Optimizes: not hurting anyone        Optimizes: reproducing a result
  Failure mode: too conservative       Failure mode: no bound at all
  Followed by: dose escalation with    Followed by: nothing
    monitoring, over months

Using a floor-finding tool to target an effect is a category error, and it is independent of whether the arithmetic inside the tool was done correctly. A perfectly executed calculation, run for the wrong purpose, produces a confident wrong answer — which is more dangerous than an obviously wrong one, because it survives scrutiny.

And a third problem, underneath both

Allometric scaling assumes something about how the compound is cleared. The method was developed largely around compounds cleared by hepatic metabolism, and it scales reasonably well for them.

Peptides are not cleared that way. They are cleared by proteolysis — enzymes that are distributed differently across tissues and species — and by renal filtration, which scales on its own terms. A peptide engineered for protease resistance or albumin binding (Chapter 33) may have pharmacokinetics that follow neither pattern.

So even the corrected figure is an estimate produced by a method that was not built for this class of molecule. What would settle the question is measured human pharmacokinetics: what concentration results, how long it persists, what the receptor actually sees. For a compound with no human development program, that measurement does not exist anywhere — which means there is nothing in the entire chain against which any of these numbers could be checked.

The stack of failures

  rodent paper reports an effect at 10 units/kg
        │
        ├── ERROR 1  naive mass conversion ................ ~12× too high
        │
        ├── ERROR 2  safety-floor tool used to target ..... category error,
        │            an effect                              survives correct math
        │
        ├── ERROR 3  scaling method not validated for ..... unknown magnitude
        │            this molecule class
        │
        ├── ERROR 4  no human PK anywhere to check against  no error bar exists
        │
        └── ERROR 5  vial contents unverified: identity,
                     concentration, purity, sterility ..... Chapters 19, 34
        ▼
  a number that looks like a dose

Four of those five failures are invisible in the final number. The number arrives looking like the output of a calculation, because it is. What it is not is a measurement of anything.

Discussion questions

  1. Explain, without using the word "allometric," why 10 units per kilogram in a mouse and 10 units per kilogram in a human are not comparable exposures. Your explanation should be understandable to someone with no biology background.

  2. The correction divisors shrink as the source animal grows. Predict, from that pattern alone, what the approximate divisor would be for a species intermediate between a rat and a dog, and explain what physiological quantity your prediction is really tracking.

  3. Error 2 is described as "worse" than Error 1 even though Error 1 is twelvefold. Defend that ranking. Then argue the opposite position as strongly as you can.

  4. This case study uses abstract "units" rather than milligrams, and an invented compound rather than a real one. Explain what pedagogical work that choice is doing, and what would be lost — and gained — by using a real compound instead.

  5. Errors 1 through 4 are all reasoning failures; Error 5 is a supply failure. Someone argues that Error 5 makes the other four irrelevant, since a person who cannot verify what is in the vial has no basis for any calculation at all. Evaluate that argument.

  6. Suppose a well-conducted human pharmacokinetic study of this hypothetical compound became available tomorrow. Which of the five errors would it eliminate, which would it reduce, and which would it leave entirely untouched? Use your answer to say precisely what a pharmacokinetic study is and is not evidence of.