Sunday, March 2, 2014

Uncertainty/Safety Factors in OELs

From my perspective, most OELs are set these days by taking a No Observed Adverse Effect Level in a repeat dose toxicology study (NOAEL or NOEL) and dividing it by an uncertainty or safety factor (SF).  Last week’s blog was about NOELs.   This week it is about uncertainty/safety factors. 

As we discussed last week, NOELs are in reality typically frank effect levels for the adverse health effects from exposure.   The portion of any tested animal population having a frank ill effect from a NOEL exposure has been shown to range from 2-21% depending on the experimental design and shape of the dose-response curve.   Assuming people are as sensitive as rodents, an exposed population of 100,000 workers and a 2-21% response results in 2,000-21,000 workers whose health has been adversely effected by exposure to the NOEL.  Clearly, the NOEL would not make a good OEL and some adjustment needs to be made.  Thus, an uncertainty or safety factor is applied to bring the OEL down to be a fraction of the NOEL.

Various schemes have been forwarded to size these factors.  A paper is available online that does a reasonable job of providing the details and the references for the standard OEL/Safety Factor approach written in 2000 by Dr. Robert H. Ku, Ph.D.,CIH.:

Here is an excerpt from that paper dealing specifically with safety factors:

The number and magnitude of these safety factors depend on the quality of the data. In general, some of these safety factors may include: (1) a factor from 1 to 10 for animal-to human (interspecies) extrapolation (if the NOEL is based on animal data), (2) a factor from 1 to 10 for human-to-human intraspecies variability in response, (3) a factor from 1 to 10 to consider study duration (a long-term study being more helpful than a short-term study), (4) a factor to consider the persistence of the drug in the body (or elimination half-life), and (5) a factor to accommodate for absorption efficiency by different routes of exposure.

Thus, it appears that the overall safety factor can range from close to 1 to a factor larger than 1000. In my experience, they typically come out to be significantly less than 100. 

As a range, I have seen safety factors as low as 3 for fully reversible, local tissue irritation response effects and greater than 1000 when considering highly uncertain data for a dreaded health effect (i.e., cancer).
From viewing it for many years, my sense is that the entire process is somewhat subjective and, in my estimation, to a certain degree driven by the practical ability to control the exposure at whatever limit is decided upon.
Indeed, one of the pioneers in our field, Dr. Stanley Roach along with his coauthor Dr. Stephen Rappaport published a paper on this subject in 1990.  The title of this paper and its reference:  But they are not thresholds: A critical analysis of the documentation of threshold limit values, American Journal of Industrial Medicine, Volume 17, Issue 6, pages 727–753, 1990.

Their primary observations and conclusion:

       “Upon analysis it was found that, where the exposure was at or below the TLV, only a minority of studies showed no adverse effects (11 instances) and the remainder indicated that up to 100% of those exposed had been affected (8 instances of 100%)”
       “…a surprisingly strong correlation was found between the TLVs and the (workplace) exposures reported in the corresponding studies cited in the Documentation [of the TLVs](emphasis added).”

Perhaps a new study should be done to see how much this situation has changed in the last 24 years.

That fact remains, however, that in most cases the size of the safety factors are driven by guidelines and are not subject to stringent rules.   We are often advised that this is necessary to allow for “professional judgment”.    I think we can all see how such a subjective system can potentially lead to somewhat unconscious bias in which the resulting OELs might become strongly correlated with workplace exposures as the study by Drs. Roach and Rappaport found.

In recent years there has been a scheme to strictly apply safety factors to toxicology data put forward by the REACh regulations in Europe.   This is done in the determination of the recently devised OEL class known as DNELs.  In the interest of time and space I am not going into the details of the DNELs and the safety factors here but suffice it to say that the process in general has cause a lot of consternation primarily because of its prescriptive nature.  

So where are we with all of this?   My sense is that the current system is clearly in need of some improvement.   The National Academy of Sciences seems to agree and has forwarded a rational modeling scheme for determining or estimating the amount of risk that exists at any exposure including the exposure limit for both carcinogens and non-carcinogens.   This is Chapter 5 of the so called “Silver Book” which you can download for free (You can download just Chapter 5 or the entire book) at:

Even though this book has been out since 2009, the OEL setting community has not embraced the recommendations.   Some folks in our community including this blogger are asserting that it should now become an increasingly prominent topic of discussion within our world.


  1. The comment below came to me today from Dr. Wil tenBerge, one of the truly remarkable thought leaders in our field. Wil has agreed to allow me to post his thoughtful comments.

    It is interesting to note that a BMD05 divided by 10 and assuming a non-threshold linear response is a predicted risk of 5 in 1000,

    Dear Mike,

    It was really a pleasure to read your your thoughts on assessment factors. I would like to make some additional remarks, reflecting my simple approach of assessment factors in case of threshold type of effects.

    The NOEL or better the BMD05 in a rodent study might be a good starting point for setting a safe exposure level for workers or consumers. In case of little information on the substance, the default assessment factors of deriving a DNEL, as proposed in the REACH Guidance, are quite reasonable. The standard assessment factors consider route and duration of exposure and interspecies and intraspecies variability.

    More knowledge on the type of effect and the mode of action gives information, if the observed effect is relevant for humans. If the effect is specific for rodents, there is no need to consider the effect for standard setting.

    Toxicokinetic information might be used to consider the fate of the substance in the body. It is worthwhile to consider whether the effect is caused by bioaccumulation in tissues or by peak concentrations during a short interval of 30 minutes of a substance with a short half-life. In case of bioaccumulation the default assessment factor for duration might be not high enough, in case of a substance of a total body half-life of a few hours it might be considered to be exaggerated.

    The interspecies factor is mostly related to slower metabolism in humans compared to experimental animals. In the extrapolation from rat to man it is assumed than humans are about 4 times more sensitive that rats in case of exposure by ingestion. In case of inhalation exposure this assessment factor is not needed, because the breathing rate is linearly related to rate of metabolism per kg bodyweight.
    Further it is assumed that humans are a factor 2.5 more sensitive than all experimental animals considering the type of toxic effect (toxicodynamics). However, if for instance the level in tissue and blood is similar in animals and humans at the same grade of effect, the toxicodynamic correction is not needed.

    The intraspecies factor always triggers a lot of debate. Due to the extrapolation of the BMD05 or the NOAEL (= point of departure), the intraspecies variability of the rodent should also be considered. The intraspecies variability of rats can be expressed by a GSD of 2 in a population with a log-normal distribution of the sensitivity, that can be derived from the slope of the dose response. The BMD05 is about a factor of 3 lower than the BMD50, that is 50 percent of rats with the specific effect. Applying a human intraspecies factor of 10 on the rat BMD05 is equivalent with the assumption, that the human safe dose is a factor of 30 smaller than the dose at which a response of 50% for a specific effect is observed in humans. This might be also interpreted, that the dose needed to increase the response from 5% to 95 in humans is 900 times the dose at a reponse of 5%. This is in toxicology a very shallow dose response, which is uncommon for most chemicals. So the intraspecies factor for humans might be considered as quite conservative.

  2. That's very detailed and informative blog post, Mike. Thank you that you're giving time and effort to present worth reading articles on risk assessments. Keep sharing the good stuffs.

    Best Regards,
    Arnold Brame

  3. Manufacturers/Importers of chemicals in Europe have derived under REACH for several thousands substances so called DNELs (derived no effect levels).
    Fixed administrative assessment factors were used to extrapolate from route/dose/duration/species/response to worker inhalation or skin DNEL.
    You can download the DNEL cookery book at
    The non-holistic approach is criticized, also because epidemiology and human finding are ingnored at large.

  4. Thank you very much for the excellent and very candid article.

    In my neck of the woods there are a number of individuals who either are or have been involved with the Cal/OSHA PEL Advisory Committee, and have been rather vocal about their opinion that the exposure limits that have been set, especially by our own Office of Environmental Health Hazard Assessment (OEHHA), are excessively low. Specifically they have argued that modeling and setting limits based on very low risk levels (e.g. 1/million) will never result in seeing the excess outcome because of a combination of factors including:

    *The correction factors used in risk assessment are often excessive
    *Exposures of 8 hrs/day, 5 days/week, 40 years for workplaces don't typically occur anymore since person's work assignments at work change, people change jobs, etc.
    *Exposures of 24 hrs/day, 7 days/week, 70 year lifetime don't occur because people move and don't spend their entire lifetime in one location.

    How would you respond?

    Thank you!

    Hila Wright

  5. Hi Hila,

    Just because we cannot directly see the adverse outcome from exposure do not mean that it is not there. The process should be charged with being reasonably protective.

    In the end is is most important to understand that All OELs are political which is not a dirty word. That is,they are, or should be, set using a democratic process in which the "best" limit is ultimately chosen by the body politic.

    We can certainly overestimate risk and that has its own dangers which could arguably be just as harmful overall to human health as underestimating risk. The point, I am trying to make is that we should strip away as much of the subjective factors and mystery as possible and show as much light on the process as we can. For my money that means modeling residual risk and showing the uncertainty in the predictions.

    I freely admit that all of the low dose extrapolations of risk at the OELs are essentially arguments without data but they also present objective and transparent suppositions as to what might be happening. For example, we may ultimately decide to accept a putative range of risk from 0 to 1 in 100 for a particular carcinogenic risk@OEL if we believe that we cannot afford as a society to do otherwise. At least everyone will know or have available all the information and quantitative judgment that went into that decision. The political (including legal) process should be available to appeal and attempt to change it. This is how democracies and open politics are supported to work.