6. Conclusion
The exposure–response modeling has evolved considerably
beyond EPA’s risk assessments USEPA [27–29] which are based
on an old outdated version of the UAB study.
In addition to Sielken and Valdez-Flores [22,23], TCEQ [25],
Grant et al. [10] contain cancer risk assessments based on the
most recent update of the UAB study. These TCEQ assessments
present results for several exposure–response models. When
these Cox proportional hazards models match those used by
Sielken & Associates the results agree. However, following their
regulatory guidelines, TCEQ focused on (1) only all leukemia, (2)
upper bounds on risk instead of best estimates (maximum likelihood
estimates), (3) data restricted to the lower 95% of the
exposure range, and (4) models based on only cumulative BD
ppm-years and not any other exposure variables or covariates.
Choices (2) to (4) cause TCEQ’s upper bound on the all leukemia
risk (0.0011 per ppm) to be about 10-fold greater than
the best estimates for all leukemia herein. In TCEQ’s discussion
of uncertainty, they acknowledge that ‘‘The inclusion of age and
number of HITs >100 ppm BD as covariates in the Cox regression
modeling may result in URFs that are more relevant to
BD exposures experienced by the general population. As mentioned
previously, once age is in the model, inclusion of number
of BD HITs results in a significant improvement in the fit
(likelihood).’’