Home Page

My Latest Articles

Buying the Book

Reviews by readers

Reviews by critics

More about the book

About me

Email me

All or nothing?

The LD₅₀ Concept

Turning mild damage into "deaths"

Is testing to death always needed?

Low doses

Who are the victims?

Evidence

Air pollution

Real Effects from Low Doses

Summary and References

All or Nothing?

In New Zealand, we have a slightly unusual "Bonus Bond" lottery. Here is how it works: our $10 "bond" deposits are held in a special bank account receiving about four per cent annual interest. That's right, forty cents a year, three cents a month per $10 invested. That trivial sum is what happens when a probability is "distributed".

On the other hand, what actually happens to Bonus Bond interest is that earnings are pooled (after the government and the bank take their cuts), and each "investor" goes into a lottery draw. One lucky person each month gets $300,000, a life-changing amount for most people. It's what I call all-or-nothing. (If someone has a nice Greek or Latin word for this, please let me know.)

Notice what's happened here. The total amount of monthly interest, is exactly the same in both situations. The "risk" of gain is insignificant if the gain is spread around. Uneven distribution, all-or-nothing, results in one wealthy person but a million others with unchanged finances. (Quibble: there are also smaller prizes, but for most people a $10 win is not life-changing.)

In my example, the "risk" is a beneficial one. The principal applies to adverse risks as well. The metaphorical model we usually misapply to environmental risks is that of Bonus Bonds, a single lucky or unlucky coloured ball in a large barrel of harmless white balls. That's wrong. In reality, environmental risks are better modelled as a large number of uniformly coloured slightly off-white balls.

Some risks are really all-or-nothing. There is, for instance, a one-in-a-million chance of dying from six minutes of canoeing, according to Sir Ernst Titterton (1981). So if a million people go canoeing for an hour, we could expect ten deaths, plus a larger number of people with broken bones or serious bruises. That leaves almost a million people who are happier, fitter, tanned, and invigorated.

The LD₅₀ Concept

For all sorts of reasons, it's important to get quantitative measurements about toxic chemicals or other health-damaging factors. The LD₅₀ test is ideally suited to this. A range of doses is tested, and the cumulative death rate is shown in the middle graph (b) of the attached graph. (This graph comes from a toxicology review at the U S Virtual Naval Hospital.)

At the heart of that S-curve is the well-known bell-shaped curve. Most animals are moderately sensitive, but a few are very weak and a few are very tough. This susceptibility curve can be replotted to make it into a dead straight line (pun not intended but gratefully accepted) The method involves restating values in terms of standard deviations, plus '5' to make everything into a positive number. (Aren't you glad you asked?) With a linear plot, working out how much insecticide would have killed exactly half the insects becomes a trivial exercise. Every scientist knows that when your data fits a straight line, you have achieved Absolute Truth and no further involvement of one's brain is required.

Does this mean that half the flies flopped buzzingly onto their backs, while the rest flew off? Of course not. Every single one of those insects was severely damaged. But half of them were slightly weaker. Perhaps they had missed a feeding, or had worn themselves out arguing with another fly. Perhaps it was just random genetic variation. The point is that all the flies were injured, but only half were pushed far enough "to the left" that they died.

This situation is shown graphically here for a low and for a moderate dose of toxin. An animal on the right side of the curve, one of the natural survivors, will probably not be pushed off the lethal cliff on the left. The first creatures to die will be the weakest. That same logic applies no matter what the dosage. Nor does it matter if the distribution of health doesn't fit a classical bell shape.

Most toxicity bioassays involve at best a few hundred animals per species, and can only detect increased incidence of 10% or greater. A five million dollar megamouse" study in the 1970s involved more than 24,000 mice dosed with a known bladder carcinogen for 18, 24, or 33 months, at levels expected to cause 1% increase in bladder cancer rates. What the study actually confirmed was a no-effect level of 45 ppm, i.e., no excess bladder cancers even after almost three years of dosage. Presumably the experiment had to be terminated at 33 months because by then the mice were dying of old age

Turning mild damage into "deaths"

LD₅₀ (or ED₅₀) assays produce a roughly uniform loss of healthiness, which is conveniently measured by counting deaths (or cancers). The LD₅₀ is, however, a 50% reduction in health (where I'm being coy about what units are used to measure "health"). Death is, in effect, 100% loss of health. In any LD₅₀ or ED₅₀ situation, the deciding factor for which flies die or which mice get cancer is the status of the dosed animals. The metaphorical model for these risks is not a small number of black balls amongst the white ones, but a barrel full of slightly grey spheres.

The problem with mechanical techniques for turning experimental data into a straight line is the misuse of that equation to forecast "deaths per million" at extremely low dosages, from experiments that cannot even confirm 1% death or cancer rates.

Obviously the same logic applies to all other health-damaging effects, whether it be chemicals or radiation or even exposure to adverse temperatures. It applies to insects, rodents, monkeys, and human beings.

Testing to the fatal level is not necessary

We could have estimated the health-damaging effect of a toxin by measuring the physical vitality of previously healthy flies. For instance, they might not buzz as loudly, might not be able to do as many press-ups, might lose their skill at dodging a fly swatter. With mice or rats, we could measure how long and how fast dosed animals could run in a treadmill, and then we'd be able to redefine damage in terms of predicted loss of endurance and/or maximum speed.

Not by coincidence, some years ago a group of EU vets, disgusted by having to kill experimental animals unnecessarily just to get a fake-accurate LD₅₀, proposed a replacement test wherein the mice (etc.) were subjected to an increasing dosage of chemical. Whenever a significant number of these animals showed "distress", the experiment would be terminated! Chemicals would be categorised as "toxic", "harmful" or "unclassified", based on the concentration at which they caused distress to the animals, even if none died (van den Heuvel et al., 1990). This proposal seems never to have been taken up: regulatory agencies are seemingly more interested in obtaining death numbers from which they can extrapolate to ludicrously low concentrations.

The real situation with LD₅₀ testing was described by van den Hesuvel et al. (1990) in these words:

Classical acute oral LD₅₀ tsts are designed to produce mortality at every dose level used and ... this normally results in all animals developing signs of toxicity, with many showing severe effects.

The situation at low doses

If the picture I've suggested is correct, low doses of toxins move every exposed animal toward the less healthy direction. We don't have easily applied quantitative numbers to express this. "Deaths per million" is only a scary but inaccurate way to describe our predictions. Indeed, regardless of what mathematical model is used for extrapolation, the data is so rough that one-in-a-million implies spurious accuracy. The experimental data comes from about 200 rodents and perhaps a few dozen dogs. It doesn't matter which hypothetical equation is used to forecast cancer incidence at very low doses: the prediction ought to be stated as something like "0.001 cancers per 100 rats". What is one-thousandth of a cancer? It probably says that exposure to that dosage of toxin can accelerate the age-related development of cancer by one-tenth of one percent. In other words, the rat would develop cancer after 300 years exposure to this chemical -- except that the rat will die of old age first. (See the Raabe curve later in this text.)

A misleading way of restating "0.001 cancers per 100 rats" is as "1 cancer per 100,000" or, even more scarily, "10 cancers per million". That allows the even more unjustifiable calculation that in our NZ population of 4 million, there will be 40 cancers a year from this cause!

Yet there has never been and never could be experimental testing. Mathematical trickery alone can take experimental cancer tests on a few hundred rats and turns it into a prediction about millions of people. A calculated forecast of so- many deaths in a million is useful to evoke further budget allocation from central government.

There are, of course, good biological reasons to think that low doses have absolutely no effect on the rugged complex network of enzymes and control mechanisms within our cells and our bodies. But for the moment let's accept the argument that low dosages have an effect, even if no one can ever measure or confirm that effect. (How many DNA molecules can dance on the head of a hypodermic syringe needle?

Who are the Victims?

Converting "deaths per million" into "reduction in health" has important implications for public health administration. The U.S. Congress has mandated a maximum lifetime cancer risk of one in a million as the upper limit for any chemical added to food. Let's examine what this means.

The New Zealand population is four million, so a "one-in-a- million" risk of death sounds like four people dying a year, from low levels of some purportedly toxic material. What can we say about these hypothetical victims? Remember that the entire population is exposed to damage from this toxin, albeit some people will receive more toxin than others, just because of their choice of food or where they live. Do the deaths of four people from this toxin "redeem" the rest of us? That's theology, not science. Obviously everyone in New Zealand will have been, in theory, ever so slightly injured, because all of us have been exposed to this toxin.

Alternatively, we could say that the four victims were actually much more sensitive to the toxin than anyone else. How remarkable! Presumably we could have identified these four people well in advance of their demise. For instance, their livers probably didn't eliminate dietary toxins, and their general physique would have been fragile. In fact, how did they ever survive to adulthood? It's clear that this concept of super-sensitive pre-selected victims simply doesn't add up.

Instead of unjustifiable extrapolations into theological realms, an alternative way to prevent possible public harm from any proposed use of a chemical is to mandate that the absolute maximum permissible concentration shall be one-hundredth of that level shown to have no effect ("NOEL") on the most sensitive experimental animal -- be it guinea pigs, mice, rats, monkeys, or canaries. The logic is that humans might be ten times more sensitive than even the most sensitive test animal, and that some people might be ten times more sensitive than others. In fact, since humans are more resilient and tougher than shorter-lived animals, the safety margin is actually even greater.

Deaths per million really means slight loss of health

My argument is that "deaths per million" is simply a sloppy, fear- mongering way of expressing a very small degradation in health, or in the case of cancer, a certain reduction in the expected time before a particular cancer might form. (Remember, cancer is primarily a disease of older people.) This degradation in health or shortening of lifespan can be estimated. Using a linear model, damage at the one- in-a-million level, for a man with a 70-year lifespan, suggests that death arrives 37 minutes sooner. (Since these extrapolations have very dubious biological validity, thirty-seven minutes is a worst-case calculation. Moreover, one could argue for a logarithmic rather than linear scale.)

Evidence?

The model I've outlined implies that at moderate doses of poisonous agents, average lifetime of all subjects should be shortened. Exactly this result was found experimentally. Here is a simplified version of a graph from Prof O G Raabe (pg 115 of my book). In this study, rats were fed a potent carcinogenic chemical in their daily laboratory chow, until they developed liver cancer. At high doses, all the animals developed cancer quite soon. At lower doses, development of cancer took longer and longer. Finally, at low doses, the animals died of old age first. All the animals were similarly affected, as shown by remarkably tiny error-bars. (Click here to open a copy of Prof Raabe's original graph in a separate window.)

In another study cited by Prof Raabe, dogs were implanted with radioactive pellets. The same results were found: high doses of radiation seriously reduced the life spans of all the animals, but at low exposures the animals died of old age.

Air pollution

The epidemiology of air pollution deaths is consistent with distributed harm rather than all-or-nothing risk. Episodes of high air pollution are soon followed by statistically significant increases in death rate from respiratory disease. Paradoxically, however, areas with high winter-time air-pollution deaths don't have any higher respiratory death rates compared with low-air-pollution cities.

The explanation is, I think, that air pollution injures everyone in the city, although individual exposure will vary significantly between people and neighbourhoods. Everyone's health is damaged by an air pollution episode. But only the most susceptible people, those who are already hanging on by a tenuous thread, are pushed off the cliff. If there were no air pollution, these people might have lasted a bit longer, but most of them were already on the slippery slope. Putting it cruelly, the respiratory deaths from air pollution constitute premature deaths from people who are, in general, already doomed. Of course. air pollution is not uniform, and some people will be unlucky enough to receive a lot higher dose than others.

This analysis predicts that transient high rates of deaths after an air pollution episode would be followed by a period of lower-than- normal deaths. The air pollution will have stolen some days or weeks from already sick individuals. That's why average annual death rates are not correlated with air pollution.

Low concentrations may have real effects

My arguments are aimed specifically at what I consider unjustified efforts to extrapolate from high-dose experiments down into a metaphysical realm of unknown and, worst of all, forever untestable forecasts of doom. On the other hand, there are certainly examples where good scientific evidence shows genuine effects at remarkably low concentrations. For instance, as little as 0.02 parts per billion of geosmin, a chemical produced by blue-green algae produces a muddy taste in catfish. (Taste is a very useful attribute to think about when words like "toxin" or "poison" produce logic- destroying responses.)

There is overwhelming evidence that exposure of populations to relatively concentrations of air pollutants such as "PM₁₀" can increase mortality by about 5%. This is a triumph of large-number epidemiology, since normally only responses of 100% or greater are considered meaningful. These air pollution forecasts have a large degree of uncertainty. One major study concluded that the most likely figure was 12 deaths per 10 units of pollution, but the possible range ran from as low as 4 and as high as 20. That is certainly greater than zero, but the lack of precision argues against imposing rigid standards or guidelines without considering cost-benefit analysis. (Interestingly such guidelines always seem to be multiples of ten, that is, "20" or "50". Why should the number of fingers on our hands be so influential, other than that the science behind the guidelines is imprecise.?)

Summary

When regulators extrapolate high-dose data to purported risks from minute doses, they often cause unnecessary public concern and costs, without actually saving a single life. Exposure to low doses of toxins is generally spread throughout the entire population, and no one person will receive a lethal dose, no one person is so incredibly feeble and sensitive that he or she will succumb to minute traces.

References

Raabe,O.G. 1989. Scaling of fatal cancer risks from laboratory animals to man. Health Physics 57, 419-432.

van den Heuvel, M J et al. 1990. The international validation of a fixed-dose procedure as an alternative to the classical LD₅₀ test. Food and ChemicalToxicology, 28(7), 469-482.