ATSDR webinar outlines benchmark dose modeling, case studies and a pathway toward data-driven risk assessment
June 29, 2026 | National Prevention Information Network (NPIN), Centers for Disease Control and Prevention (CDC), Department of Health and Human Services (HHS), Executive, Federal
This article was created by AI summarizing key points discussed. AI makes mistakes, so for full details and context, please refer to the video of the full meeting. Please report any errors so we can fix them. Report an error »
Dr. Chao Ji and Dr. Andrew Prussia of the Agency for Toxic Substances and Disease Registry (ATSDR) explained benchmark dose (BMD) modeling during a 40–45 minute webinar, describing how the method uses full dose–response data to derive health-guidance values such as minimal risk levels (MRLs).
"Everything is toxic, and it depends on the dose level," Dr. Ji said, framing the need to identify a point of departure on a dose–response curve. The presenters reviewed the BMD approach and its lower 95% confidence bound (BMDL), and contrasted it with traditional NOAEL/LOAEL methods that rely on pairwise comparisons to controls and fixed uncertainty factors.
Dr. Prussia outlined the history and tools used by risk assessors, noting that Kenneth Crump first described the benchmark dose concept in 1984 and that EPA released Benchmark Dose Software (BMDS) in 2000 to help implement it. He also stated that "the EPA has a target for eliminating all mammalian animal testing in 2035," and framed BMD and new approach methods as part of a transition to more quantitative assessments.
The presenters described how uncertainties are handled in MRL derivation. Standard practice uses an animal-to-human uncertainty factor (commonly 10), with a dosimetric adjustment that can reduce that factor to 3 when physiological data allow, plus a human-variability factor typically set at 10. The speakers said those fixed factors are useful but limited: they do not specify what fraction of the population is protected. To address that limit, they outlined a probabilistic alternative (sometimes called a Target Human Dose) that uses distributions for model parameters to estimate the probability and percentile of an affected population.
Three case studies illustrated the approach. In a 28‑day rat gavage study of xylene, Dr. Prussia showed BMDS output with a best-fit BMD of 305.6 and a 95% confidence interval of 219–495; the presenters used the lower bound (BMDL = 219) as the point of departure. Carrying that forward and applying uncertainty factors produced a BMDL‑based MRL of about 2.2, compared with an MRL of 1 that would result from using the NOAEL, demonstrating how BMDL can yield a point-of-departure that is data‑driven while avoiding potentially over‑protective NOAEL-based results.
For epidemiology data, Dr. Ji cited a published study (Chen 2010) of residents in southern Taiwan who consumed well water contaminated with arsenic. The study included more than 8,000 people grouped into five exposure categories; logistic regression produced a clear dose–response for blood cancers. When the presenters used small benchmark-response values (for example, a 1% increase), BMD and BMDL estimates were around 10 (dose units as reported in the study), whereas an NOAEL-style point of about 144 would have considerably underestimated risk. The speakers noted that the BMD-derived estimate aligns with EPA guidance values for arsenic in drinking water.
The webinar also covered new approach methods (NAMs), including toxicogenomics. Dr. Ji described how short-term transcriptomic changes can be treated as continuous dose–response data, how multiple gene-level points of departure can be grouped into pathways, and how in vitro–to–in vivo extrapolation (IVIVE) is needed to translate molecular findings into human-relevant doses. The presenters cited NTP guidance (2018) and the use of toxicogenomics in response to the 2014 Elk River chemical spill as examples of the method’s application.
During Q&A, an attendee asked about minimum animal numbers per dose group; Dr. Prussia said "five is about the minimum" for rodent studies to estimate variance reliably, while noting that epidemiological studies typically need much larger sample sizes to detect effects. When asked whether BMD results have been validated against other methods, the speakers pointed to EFSA, EPA, NTP and NIEHS technical comparisons and said correlation and method-comparison studies exist, though toxicogenomics requires more systematic adoption.
The presenters closed by arguing that environmental public health is moving toward more quantitative, multidisciplinary assessments that combine biomonitoring, epidemiology, GIS, and NAMs, and that BMD modeling will be a core analytical tool as agencies reduce reliance on mammalian animal testing.
The webinar ended with a brief Q&A and an invitation to submit follow-up questions to the email address shown on screen.
"Everything is toxic, and it depends on the dose level," Dr. Ji said, framing the need to identify a point of departure on a dose–response curve. The presenters reviewed the BMD approach and its lower 95% confidence bound (BMDL), and contrasted it with traditional NOAEL/LOAEL methods that rely on pairwise comparisons to controls and fixed uncertainty factors.
Dr. Prussia outlined the history and tools used by risk assessors, noting that Kenneth Crump first described the benchmark dose concept in 1984 and that EPA released Benchmark Dose Software (BMDS) in 2000 to help implement it. He also stated that "the EPA has a target for eliminating all mammalian animal testing in 2035," and framed BMD and new approach methods as part of a transition to more quantitative assessments.
The presenters described how uncertainties are handled in MRL derivation. Standard practice uses an animal-to-human uncertainty factor (commonly 10), with a dosimetric adjustment that can reduce that factor to 3 when physiological data allow, plus a human-variability factor typically set at 10. The speakers said those fixed factors are useful but limited: they do not specify what fraction of the population is protected. To address that limit, they outlined a probabilistic alternative (sometimes called a Target Human Dose) that uses distributions for model parameters to estimate the probability and percentile of an affected population.
Three case studies illustrated the approach. In a 28‑day rat gavage study of xylene, Dr. Prussia showed BMDS output with a best-fit BMD of 305.6 and a 95% confidence interval of 219–495; the presenters used the lower bound (BMDL = 219) as the point of departure. Carrying that forward and applying uncertainty factors produced a BMDL‑based MRL of about 2.2, compared with an MRL of 1 that would result from using the NOAEL, demonstrating how BMDL can yield a point-of-departure that is data‑driven while avoiding potentially over‑protective NOAEL-based results.
For epidemiology data, Dr. Ji cited a published study (Chen 2010) of residents in southern Taiwan who consumed well water contaminated with arsenic. The study included more than 8,000 people grouped into five exposure categories; logistic regression produced a clear dose–response for blood cancers. When the presenters used small benchmark-response values (for example, a 1% increase), BMD and BMDL estimates were around 10 (dose units as reported in the study), whereas an NOAEL-style point of about 144 would have considerably underestimated risk. The speakers noted that the BMD-derived estimate aligns with EPA guidance values for arsenic in drinking water.
The webinar also covered new approach methods (NAMs), including toxicogenomics. Dr. Ji described how short-term transcriptomic changes can be treated as continuous dose–response data, how multiple gene-level points of departure can be grouped into pathways, and how in vitro–to–in vivo extrapolation (IVIVE) is needed to translate molecular findings into human-relevant doses. The presenters cited NTP guidance (2018) and the use of toxicogenomics in response to the 2014 Elk River chemical spill as examples of the method’s application.
During Q&A, an attendee asked about minimum animal numbers per dose group; Dr. Prussia said "five is about the minimum" for rodent studies to estimate variance reliably, while noting that epidemiological studies typically need much larger sample sizes to detect effects. When asked whether BMD results have been validated against other methods, the speakers pointed to EFSA, EPA, NTP and NIEHS technical comparisons and said correlation and method-comparison studies exist, though toxicogenomics requires more systematic adoption.
The presenters closed by arguing that environmental public health is moving toward more quantitative, multidisciplinary assessments that combine biomonitoring, epidemiology, GIS, and NAMs, and that BMD modeling will be a core analytical tool as agencies reduce reliance on mammalian animal testing.
The webinar ended with a brief Q&A and an invitation to submit follow-up questions to the email address shown on screen.
View the Full Meeting & All Its Details
This article offers just a summary. Unlock complete video, transcripts, and insights as a Founder Member.
✓
Watch full, unedited meeting videos
✓
Search every word spoken in unlimited transcripts
✓
AI summaries & real-time alerts (all government levels)
✓
Permanent access to expanding government content
30-day money-back guarantee
