Sunday, March 30, 2014

OELs from Low Dose Extrapolation – Part 1

We are going to stick with OELs at least for the next few weeks because they represent fully half of the information we need to estimate risk as Industrial Hygienists and Exposure Assessors.

We discussed risk-based occupational exposure limits (RBOELs) or the description of the putative quantitative risk at any OEL (Risk@OEL) as being at the very top of the OEL Hierarchy as described in detail in the March 2014 edition of the American Industrial Hygienist publication: The Synergist.   Indeed, it is a low dose extrapolation of a non-threshold model to a putative lifetime risk of cancer of about 1 in 1000 that is leading NIOSH to put forth an REL of 0.1 ppmv for ethylene oxide as discussed in last week's blog.

The inescapable reality in today's world is that for anyone to describe or estimate the level of risk at any Occupational Exposure Limit (OEL), one must do low-dose extrapolation of a dose-response curve.   Anybody can fit a line (math model) to animal data which occurs at doses causing effects.  The truth is that we are not particularly interested in the relatively high level of exposure that causes an effect during toxicological testing of animals.   We are, however, intensely interested in the toxicological effects that might happen in people exposed to much lower doses.

There are good reasons for this situation.   We would love to say that our tests tell us the complete dose-response story for ALL rats for their entire lifetime (chronic).  However, for a lot of reasons, we can only test a limited number of animals and usually only for a limited portion of their lifetime.   Because of this restraint we typically have to “push” the dose to a point where something bad happens to the animals in a toxicological sense.   I have heard it said by the toxicological colleagues: “The purpose of a toxicology test is to find toxicity”.    Thus, in many or most repeat-dose studies the top dose is set high enough that it will cause an untoward health effect but just below the level that will kill the animal.   That has been called the maximum tolerated dose or MTD.   This defines the highest dose to be used in the study .  Besides defining the top of the dose range, it provides a critical piece of information; namely, the nature of the “bad thing” that happens from overexposure to this substances.   It could be as simple as local tissue response for a highly irritating material or it could be one of a partial list of effects listed below:
  •         Central Nervous System Depression
  •          Neuropathy
  •          Organ (liver, kidney) damage
  •          Chronic Obstructive Pulmonary Disease
  •          Birth Defects in young from exposed mothers
  •          Cancer
This provides the Hazard Identification piece for the substance; viz., the nature of the “bad thing” that happens during overexposure.
From the group of animals that gets the MTD the dose is reduced in the 3 or 4 groups getting progressively less of the material.   Often the lowest group is designed to render a No Observed Adverse Effect Level or NOEL or NOAEL.   We have shown in previous blogs that this is NOT the dose where nothing happens in the entire population of animals but only were nothing statistically happens in the tested group.   In reality it typically represents about a 10% level of frank adverse effect for the entire population if the entire populations were exposed to this level.
Lowering the dose for a group typically lowers the level of response.   This is the so-called dose-response relationship and is a touchstone of the science of toxicology.   It clearly shows cause and effect in a scientifically controlled experiment.  The cause is the dose and the effect is the general monotonically increasing level of response with increasing dose.
The standard manner of mathematically expressing these test results is what is known as quantal response.   Say we have 10 animals per group and 5 groups.   In the MTD or top group, none died in 2 years of testing but almost call got cancer of the liver, that is 9 out of 10 or 90%.   That is the quantal response; that is, they either got cancer or they did not. (The most dramatic example of quantal response is in acute lethality testing, where the animals typically either die or completely recover at any particular dose).   In the bottom or lowest dose group 1 out of 10 (10%) got liver cancer; however, this was not statistically different from the response of the unexposed controls.   Thus, the bottom group is the NOEL.   Note:  This is not a fanciful example but a situation that happens quite often in chronic tests of rodents.
So now we have spent quite a bit of money and have the following data to show for it; 90% cancer response in the top dose group, 10% dose in the bottom group and intermediate response in the 3 intermediate groups. 
As mentioned above, we are not particularly interested in a dose that may cause the lowest detectable (10%) cancer response in 10 rats but in the dose or exposure that presents a MUCH lower cancer risk in people.   We are forced to do low dose extrapolation.   As mentioned in the last blog, acceptable (or at least not unacceptable) putative cancer risk for workers has been set around 1 in 1,000 for the last 30-40 years in the modern era of quantitative risk assessment.   In the same vein, the putative cancer risk for folks in the general public is set much lower, ranging from 1 in 10,000 to 1 in 1,000,000.

How do we go from these high risk rat data to low risk extrapolated estimates for humans?   Many models will fit the actual rat data quite well but give VERY different predictions when extrapolated to low risk at low dose.    We will explore the details of this situation next week.


  1. The secondary problem that is not addressed in the flaw in using rodents as models for human toxicity. Time and time again we find chemicals that have pathways that are different in rodents vs humans. Two examples are naphthalene and methyl ethylene chloride. Yet regulators and groups like NIOSH and ACGIH turn their back on good sound data because it does not fit their model. So risk based data is really bias based data analysis.

  2. Excellent write up and summary. I enjoy reading tyour OEL series. Way to go!