Dr. Manouchehr Hessabi
← All writing
8 min readenvironmental health · toxicology · research methods

Dose-response relationships, explained

What a dose-response relationship is, how regulators set thresholds and reference doses, and why low-level lead complicates the simple curve.

By Manouchehr Hessabi, MD, MPH

A dose-response relationship describes how the likelihood and severity of a health effect change as the amount of exposure changes. It is the backbone of how toxicologists and regulators judge chemical risk, and it is also one of the ways epidemiologists weigh whether an association might be causal. The familiar maxim "the dose makes the poison" captures part of it, but the shape of the curve, especially at low doses, is where most of the scientific argument lives.

Many readers take the maxim to mean that small exposures are harmless. This explainer sets out what a dose-response relationship actually is, how a curve becomes a regulatory number, and two areas of research where the simple picture is strained. It is educational and not a substitute for personal medical advice.

What is a dose-response relationship?

The US Environmental Protection Agency (EPA) defines a dose-response relationship as one that "describes how the likelihood and severity of adverse health effects (the responses) are related" to the amount and condition of exposure (EPA, Conducting a Human Health Risk Assessment, page last updated January 13, 2026).

Three terms are worth separating:

  • Exposure is contact between a person and a substance, for example lead in dust or a chemical in drinking water.
  • Dose is the amount that actually enters the body or reaches the tissue of interest, often expressed per kilogram of body weight per day.
  • Response is the measured effect: a change in a test score, a rate of disease, an enzyme level.

Plotting response against dose gives a curve. The questions that matter are about its shape. Does the response rise steadily, level off, or only begin above some dose? Does it ever change direction?

How do regulators turn a curve into a number?

EPA describes human health risk assessment as four steps: hazard identification, dose-response assessment, exposure assessment, and risk characterization (EPA). Dose-response assessment is the second step, and it follows one of two broad approaches.

The threshold, or non-linear, approach. When information about how a substance acts suggests a threshold, which EPA defines as "the dose below which no deleterious effect is expected to occur," the agency uses a non-linear assessment. The usual product is a reference dose (RfD), which EPA describes as "an estimate (with uncertainty spanning perhaps an order of magnitude) of a daily oral exposure" to the population, including sensitive groups, likely to be without appreciable risk of harmful effects over a lifetime.

EPA's formula is:

RfD = NOAEL (or LOAEL or BMDL) / UFs

Each term has a specific meaning:

  • NOAEL, the no-observed-adverse-effect level: the highest tested dose at which no adverse effect was seen.
  • LOAEL, the lowest-observed-adverse-effect level: the lowest tested dose at which an adverse effect was seen, used when no NOAEL is available.
  • BMDL, the lower confidence limit on a benchmark dose: a point estimated by fitting a model to the whole dose-response curve rather than relying on a single tested dose.
  • UFs, uncertainty factors: divisors, generally order-of-magnitude in size according to EPA, that account for gaps such as extrapolating from animals to people or from average to sensitive individuals.

The linear, or non-threshold, approach. When the evidence indicates that a substance's toxicity "does not have a threshold," EPA calls the assessment linear. In that framing any added dose is assumed to carry some added risk, and the output is a risk estimate per unit of exposure rather than a "safe" dose.

Two points follow for the reader. First, a reference dose is a judgment built on data plus deliberate safety margins, and EPA itself says it carries uncertainty of perhaps tenfold. Second, the choice between a threshold model and a linear model is itself a scientific claim about the curve's shape at low doses, which is exactly where data are thinnest.

Why does low-level lead challenge the simple picture?

Lead is the best-studied example of a dose-response curve that does not behave the way the maxim suggests. The central study is an international pooled analysis by Lanphear and colleagues, published in Environmental Health Perspectives in 2005. It combined data on 1,333 children from seven population-based cohort studies, followed from birth or infancy to between 5 and 10 years of age, and examined full-scale IQ scores against blood lead concentration, measured in micrograms per deciliter (ug/dL) of blood.

In 2019 the authors published an erratum correcting errors limited to one cohort, in Boston, where performance IQ data had been used in place of full-scale IQ and some early blood lead values had been converted incorrectly. The erratum states that EPA and other investigators concluded the errors did not change the overall conclusions. The corrected figures, from a log-linear model, are:

  • A 6.7 IQ point decrement (95% confidence interval 4.1 to 9.3) associated with an increase in concurrent blood lead from 2.4 to 30 ug/dL.
  • For successive segments of that range, decrements of 3.8 points from 2.4 to 10 ug/dL, 1.8 points from 10 to 20 ug/dL, and 1.1 points from 20 to 30 ug/dL.
  • For a given increase in blood lead, the decrement among children whose maximal blood lead stayed below 7.5 ug/dL was significantly greater than among those at or above 7.5 ug/dL (p = 0.02).

Read plainly, the curve was steepest at the low end. The first few micrograms were associated with a larger drop in IQ than the same increase higher up. That is close to the opposite of a threshold.

The study has real limits, and they matter for how far the finding travels. It pooled observational cohorts, so the association could be affected by confounding, meaning other factors linked to both lead exposure and IQ, such as home environment or socioeconomic position, even after statistical adjustment. Blood lead is also an imperfect measure of lifetime exposure. The language of the paper is association, not proof of a precise causal slope. For the sources of lead exposure and the wider evidence, see the companion piece on lead exposure in children.

What is a nonmonotonic dose response, and why is it argued over?

A monotonic curve moves in one direction: as dose rises, the response rises (or falls) and never reverses. A nonmonotonic dose response, as defined in a 2012 review by Vandenberg and colleagues in Endocrine Reviews, is "a nonlinear relationship between dose and effect where the slope of the curve changes sign somewhere within the range of doses examined." A U shape or an inverted U are common examples.

The review focused on hormones and endocrine-disrupting chemicals, substances that can interfere with the body's hormone signaling. Its authors argued that such chemicals can have effects at low doses that are not predicted by effects at higher doses, that this challenges what they called the "dogma" of the dose making the poison, and that when nonmonotonic curves occur, testing at high doses cannot tell you what happens at low ones. They concluded that fundamental changes in chemical testing were needed.

This is best read as the argued position of that review rather than as a settled consensus across toxicology. What it does establish clearly is the methodological point: a testing strategy that assumes a monotonic curve can miss effects if the true curve bends.

A different kind of curve appears for nutrients the body needs. Manganese is essential in small amounts and harmful in excess, and several child cohorts have reported both low and high levels associated with poorer outcomes. That case is discussed in the piece on manganese and the inverted U. The shape is similar to a nonmonotonic curve, but the reason is different: deficiency at one end, toxicity at the other.

Why does a dose-response pattern matter for causation?

In his 1965 address, Austin Bradford Hill set out nine viewpoints for judging whether an observed association between an environmental exposure and a disease is likely to be causal. One of them is the biological gradient: whether more exposure is accompanied by more of the outcome.

Where a clear gradient is present, it supports a causal reading. But Hill was explicit that his viewpoints are aids to judgment, not a checklist: "None of my nine viewpoints can bring indisputable evidence for or against the cause-and-effect hypothesis and none can be required as a sine qua non" (as quoted by Wakeford in the Journal of the Royal Society of Medicine, 2015). A threshold can hide a gradient over the range a study happens to observe, and, as the sections above show, a real effect can produce a curve that flattens or reverses. A missing simple gradient therefore does not, by itself, rule out a causal effect. The full set is explained in the Bradford Hill viewpoints.

A measured close

A dose-response curve is one of the most useful tools in environmental health, and one of the easiest to over-read. It can show how risk changes with exposure across the range that was actually studied. It cannot, on its own, say what happens far outside that range, settle whether an association is causal, or tell an individual what a particular exposure means for them.

The maxim about dose is a reasonable starting point. The research on low-level lead and the debate over nonmonotonic curves are reminders that the shape of the curve at low doses is an empirical question, answered study by study rather than assumed. These questions sit close to the work on autism and the environment, where exposure timing and level are central to study design.

This article is educational and is not a substitute for personal medical advice. Readers interested in the underlying research can find Dr. Hessabi's peer-reviewed work on the publications page.

About the author. Dr. Manouchehr Hessabi is a physician-epidemiologist and Senior Research Scientist at the BERD core of UTHealth Houston's Center for Clinical and Translational Sciences. See his peer-reviewed publications or research programs.