Skip to content

Exposure Dose Curve

Response model — instantiates Effective-Input Delivery Assurance

Maps how response changes across the full range of an input — from no effect, through the useful zone, to diminishing returns and harm — so any single level can be read off the curve.

Exposure Dose Curve is the underlying relationship between the amount of input actually delivered and the response it produces, characterized across the whole range rather than at one operating point. Its defining move is that it maps the terrain, not the route: it does not pick a level or write a rule; it describes the shape — the threshold where effect begins, the slope through the responsive zone, the plateau of diminishing returns, and the downturn into harm — from which every other mechanism reads its thresholds. Where the dosage window chooses a band and calibration tunes an individual, the curve is the evidence both of them consult.

Example

An agronomy team characterizes crop yield as a function of applied nitrogen. Plotted across the full range, the curve tells a story a single fertilizer rate never could: near-flat at low nitrogen (the crop is deficient and unresponsive), then a steep responsive rise, then a plateau where extra nitrogen is taken up luxuriously but adds no yield, and finally a downturn as excess drives lodging, disease, and environmental loss. The slope of that curve is the sensitivity profile — kilograms of grain per kilogram of nitrogen — steep early and near zero at the plateau. Reading it, the economic optimum sits where the marginal yield just pays for the marginal input, comfortably short of the agronomic maximum; past the plateau, more nitrogen buys nothing and simply leaches away. The curve is what turns "more fertilizer is better" into a shape with a best place to stand.

How it works

The curve is built by observing response at several input levels — or by fitting a mechanistic form to them — and then characterizing the shape: where the threshold sits, how steep the responsive slope is, where the plateau begins, and whether there is a toxic downturn. It reports the slope region by region, so a reader can see not just the response at a level but how sensitive the response is to changing it there. What distinguishes it from its siblings is that it produces the relationship, deliberately stopping short of choosing an operating point.

Tuning parameters

  • Range and resolution of inputs sampled — how wide a span and how finely. Too narrow a range misses the threshold or the plateau; wide, fine sampling costs experiments.
  • Functional form — linear, log, sigmoid (Hill), or quadratic-plateau. The assumed shape drives everything outside the measured points and can mislead if wrong.
  • Response endpoint — which output is plotted (yield, efficacy, a biomarker). Change the endpoint and the whole curve moves.
  • Individual vs. population curve — one averaged curve or a family capturing variability. Averaging smooths the picture but can erase a sensitive subgroup.
  • Uncertainty representation — a confidence band around the curve versus a bare line implying more precision than the data holds.

When it helps, and when it misleads

Its strength is that it exposes non-linearity — thresholds, diminishing returns, and the harmful downturn — that a single-point rule structurally cannot see. It is the evidentiary basis the dosage window and the minimum-dose review both read from, and it is what keeps them from being arbitrary.

Its weakness is that a curve is only valid within the range it was measured. The classic error is extrapolation: assuming the slope keeps going, or that a safe threshold exists where none was tested, or that a low-dose effect predicts the high-dose one — when some responses are non-monotonic, the very possibility captured by hormesis.[1] Averaging can also erase a sensitive tail, hiding the subgroup for whom the curve is different. The classic misuse is to fit a convenient line and extend it to justify a level already chosen. The discipline that keeps it honest is to state the measured range explicitly, carry the confidence band forward, prefer mechanistic forms over convenient ones, and refuse to read the curve where there are no data.

How it implements the components

Exposure Dose Curve fills the relationship slots — the input axis and the shape of the response, not any chosen level:

  • managed_input_or_exposure — it is built on the axis of the input or exposure being managed, defining the quantity that response is related to.
  • sensitivity_profile — the curve's slope is the sensitivity: how much response moves per unit of input, region by region across the range.

It does not set the operating band (target_range / adjustment_rule, handled by Dosage Window Protocol), find the lowest workable level (minimum_effective_input, handled by Minimum Effective Dose Review), or measure how much input survived en route (stagewise_availability_measurement, handled by Stagewise Availability Assay); it characterizes the relationship those mechanisms consume.

Notes

The x-axis should be effective delivered input, not nominal supplied input. If the curve is plotted against what was supplied upstream rather than what actually reached the point of action, every delivery loss is silently folded into the response shape — flattening the true slope and misplacing the thresholds. A dose-response curve is only as trustworthy as the delivery measurement feeding its horizontal axis.

References

[1] The dose-response relationship is the foundational idea that effect depends on amount — Paracelsus's "sola dosis facit venenum," the dose makes the poison. Hormesis — where low and high doses of the same agent produce opposite effects — is the standing reason a response curve cannot be assumed monotonic, and why extrapolating a low-dose slope upward is unsafe. ↩