Risk Policy Report  [Printer-friendly version]
February 7, 2006


[Rachel's introduction: The Office of Management and Budget (OMB) has
proposed new rules for risk assessments conducted by federal
agencies. Scientists within the U.S. Environmental Protection Agency
are not enthusiastic.]

EPA staff scientists are strongly criticizing recent calls by the
White House Office of Management & Budget (OMB) to evaluate
alternative "models" for how chemicals affect health, saying the
mandate may force the agency to accept "junk science" and
unjustifiably lower environmental standards.

The criticism could lead to a rift with EPA science chief George Gray
over crafting agency comments on the draft guidance, which are due to
be submitted to OMB later this year.

Gray said in a recent interview that considering alternative models
submitted to the agency in support of risk assessments about
chemicals' biological impact will help improve decisions by detailing
the uncertainty surrounding those estimates.

Gray has made expanding uncertainty analysis in agency risk reviews a
key element of his emerging agenda (Risk Policy Report, Jan. 31, p1).

One senior EPA science manager reportedly said at a Jan. 23 meeting of
the agency's high-level Science Policy Council (SPC) that the bulletin
is "dreadful" and reflects a "1980s view of risk assessment" that does
not reflect currently accepted practices, according to sources who
attended the meeting.

OMB's recently proposed risk guidance calls on agencies to consider
all "plausible" models of the biochemical impacts of a contaminant at
low levels of exposure. "Where a risk can be plausibly characterized
by alternative models, the difference between the results of the
alternative models is model uncertainty ...When model uncertainty is
substantial, the central or expected estimate may be a weighted
average of the results from alternative models," according to the
draft guidance.

But some EPA scientists and observers say this provision will result
in "model-shopping" among industry scientists. "This may open the door
to everyone coming in with their own model." Simply because they can
write a computer program to "compute results they prefer, doesn't mean
that a model is valid," according to one science policy expert.
"Accepting all 'plausible' models sets a pretty low bar. These models
should require strong biological support before they are considered,"
an agency source says.

Agency critics also say the approach is flawed because the strength of
the biological support should determine which models the agency
accepts for inclusion in chemical risk reviews. They also say it is
hard to find chemicals for which there is sufficient biological
information to conduct fully informed modeling exercises mandated by
OMB. "There are only a few dioxins, arsenics and mercury's out there
with substantial enough databases," according to another agency

But administration officials are strongly defending their approach.
During a Jan. 19 interview, then-OMB regulatory chief John Graham said
"if there's no evidence for looking at an alternative model, then
there's no value in considering it. Statistical backing is evidence,
although if the support is biological as well as statistical, then
that's stronger than statistical evidence alone."

And, during a Jan. 25 interview with Risk Policy Report, Gray agreed
with EPA critics that biological evidence is important, but argued
that multiple models are essential to understanding the uncertainty of
chemicals' impacts. "I don't think anyone would suggest that
statistical goodness-of-fit would determine [the quality of a model].
It's biological evidence that counts." But Gray said that examining
different models with "biological plausibility" is a useful way to
understand the uncertainty that surrounds chemical potency estimates.

Gray also said during his interview, "Models are just ways for us to
implement different biological theories about what may be going on.
What I think is important is when we don't know about different
biological pathways, considering multiple models is probably a good

Asked whether the call in the OMB bulletin would open the agency to
alternatives to the precautionary "linear model" currently used for
evaluating carcinogens, Gray said, "The linear model has certain
biological principles it is based on and what you can do is look at
how well the data fit that model. Other models have different
biological data that support them. That process helps you think about
the data and the model helps you quantify the uncertainty around

But industry analysts say EPA and academic scientists also manipulate
statistics to buttress their arguments about the health impacts of
environmental contaminants without strong biological support in agency
risk-based drinking water and air standards. Industry observers say
the linear model does not describe the impact of many potentially
carcinogenic chemicals at low doses and point to the agency's 1998
decision to use the linear model for chloroform, which was rejected
both by agency science advisers and in a March 2000 U.S. Court of
Appeals for the District of Columbia Circuit decision as untenable
(Risk Policy Report, April 18, 2000, p5).

But EPA risk assessors say the OMB may recreate the confusion sparked
by open calls for a variety of risk models accepted by EPA programs in
the 1970s and 1980s. At the time, industry and other outside experts
could chose among the linear, "the one-hit, multi-hit, Weibull, log-
probit and other models. Eventually EPA and other federal agencies
realized that interested parties would 'model shop' and use the model
that gave them the answers they preferred," according to the science
policy observer.

This resulted in the agency formally adopting the linear model to
provide some comparability across risk estimates in the 1986 cancer
risk guidelines and a set of interagency principles the White House
Office of Science and Technology Policy adopted.

But industry officials say risk modeling has grown more sophisticated
since that time and companies know convincing and detailed data on a
chemical's biochemical impacts are now required. "With the publication
of the 2005 updated cancer guidelines, companies know that significant
biological evidence is required to overcome protective assumptions
like the linear model EPA uses," according to one industry source.

Agency staff and environmentalists say they are concerned key
protections will be eroded if the bulletin is finalized in its current

"This [OMB guidance] is worrisome because models can be constructed to
give any kind of low-dose answer you want, and the practical
alternatives" are limited, an EPA source says. And an environmental
scientist says, "This is an assault on protective assumptions with
far-reaching consequences." A science policy observer agrees, saying,
"Concerns about an assault on the linear model are accurate and well

But conservative science policy analysts also argue that academic and
EPA scientists can rely too heavily on statistically-backed but
biologically deficient arguments to support their views.

In a Jan. 17 critique of EPA air pollution standards for particulate
matter, American Enterprise Institute visiting fellow Joel Schwartz
says, "There are literally millions of plausible statistical models
relating to air pollution health outcomes, and no objective way to
choose among them. Under these circumstances, researchers have a
tendency to select those models that give the largest or most
statistically significant effects."