Why should we care about small doses of harmful pollutants?
Based on Lanphear, B.P. (2017). PLoS Biology 15(12). PMC5736171
You've probably heard this in a chemistry class, maybe alongside an anecdote about how even water can kill you at high enough doses.
It's a comforting idea. It means that for any substance, there's a danger threshold. Stay below it and no harm is done.
Here's what that looks like as a graph. Scientists call it an exposure-response curve. On one axis, how much you're exposed to. On the other, how much harm it causes.
Water follows this pattern neatly. Drink a normal amount and nothing happens. Drink far too much and the danger rises sharply. There's a clear threshold between "safe" and "not safe."
This shape has a name. Scientists call it a threshold model, and for decades it has been the default assumption behind environmental regulation. Find the threshold, set the legal limit below it, and the public is protected.
But what if this way of thinking has led generations of scientists and policymakers astray?
It turns out that many pollutants are not like water. Their exposure-response curves have a very different shape, and the consequences of getting that shape wrong are enormous.
The Threshold model is the water pattern we just saw. Below some dose, risk is zero. This is the EPA's default assumption for most chemicals. It means "safe" levels exist, and regulation need only keep exposure within that range.
Model 2, Linear No Threshold (LNT). Any dose carries proportional risk. There is no safe level, but risk increases at a constant rate. The EPA's default for carcinogens.
Model 3, Supralinear (Decelerating). Risk rises fastest at the lowest doses, then the curve flattens. A small increase from zero is more dangerous than the same increase at higher levels.
This is the shape Lanphear and colleagues found across multiple chemicals and health endpoints.
At high doses, the three models roughly agree. But at low doses, where most people actually live, they diverge dramatically.
Drag the line to see how the models' predictions differ at any exposure level.
You might wonder about a fourth shape: an accelerating curve, where harm grows faster at higher doses. Practically speaking, this shape has the same implications as the threshold model, where risk is fine at low doses until it reaches a point where it suddenly isn't.
Lead & IQ loss. Of the 9.2 IQ points lost on average across the exposure range, 6.2 points (67%) are lost below 10 μg/dL, the old "level of concern." The steepest damage occurs at the lowest exposures.
Particulate matter (PM2.5) & mortality. Cardiovascular mortality risk climbs most steeply at the lowest concentrations. Elevated risk has been detected down to 1 μg/m³, far below any current regulatory standard.
Benzene & leukemia. A decelerating curve fits the data better than a linear model. Relative risk reaches 1.52 at just 10 ppm-years of cumulative exposure, then flattens.
The pattern is consistent across chemicals. The dose-response relationship is supralinear, and the greatest per-unit harm happens at the lowest doses.
For most chemicals, the EPA assumes a threshold exists. If exposure is below the reference dose, risk is officially zero. All harm below the threshold goes unaddressed.
Even for carcinogens, where the EPA uses a linear model, the true curve is steeper at low doses. The linear model systematically underestimates risk where most people are exposed.
The National Academy of Sciences has recommended assuming no threshold unless strong evidence supports one.
The gap between regulatory assumptions and empirical evidence means that current "safe" levels leave the majority of the disease burden untouched.