Chemical risk assessors conduct chemical risk assessment to establish safe exposure levels, so called reference doses (RfD) or tolerable daily intakes (TDI) for chemicals. Traditionally, no observed adverse effect levels (NOAELs) or lowest-observed effect levels (LOAELs) from animal studies are used to establish these protective exposure levels. However, extrapolation from NOAELs and LOAELs has the limitation that it depends heavily on the choice of dose groups and does not account for the shape of the dose-response curve. In response to these weaknesses, Benchmark Dose (BMD) modeling was developed. The BMD is derived from the Benchmark Response (BMR), equivalent to a 10% or 1 standard deviation change in response for dichotomous (i.e. the presence of tumors) and continuous data (i.e. weight), respectively (see figure 1).
In contrast to traditional approaches, BMD modeling accounts for the shape of the dose-response curve, is more independent of study design and can be more easily compared across chemicals. So far, BMDs and BMDLs* (95% confidence limit of the BMD) are modeled on a chemical-by-chemical basis and require a risk assessor to make modeling choices. Jessica Wignall and colleagues from the University of North Carolina, the University of California and the U.S. EPA assessed whether a standardized approach to BMD and BMDL modeling may be an appropriate and time efficient alternative to assess large sets of dose-response data (Wignall et al. 2014). The researchers successfully batch-modeled BMDs and BMDLs using publically dose response data of 255 chemicals, identified by a name and a unique chemicals identifier (CAS). The data on the cancer and non-cancer human health assessments was retrieved from the U.S. Environmental Protection Agency (EPA) and the U.S. California EPA. In accordance with a pre-defined decision logic, the U.S. EPA’s Benchmark Dose Modeling Software (BMDS) Wizard recommended BMDs and BMDLs using appropriate but simple model types for continuous, dichotomous or dichotomous-cancer dose response types specified in the U.S. EPA BMD technical guidelines. Successfully batch-calculated BMDs and BMDLs were then compared to BMDLs and NOAELs derived manually in previous human health assessments. Almost 90% of batch-calculated BMDLs were within one order of magnitude of BMDLs from previous assessments and thus highly linearly correlated with BMDs and BMDLs. Also when comparing batch-calculated BMDs and BMDLs with NOAELs and LOAELs a significant, though less strong, linear correlation was observed; thereby, on average the ratio between BMDs and NOAEL was around 2, showing that the batch-modeled approach was in the same order of magnitude, only slightly more conservative. The standardized approach achieved successful modeling of 86, 91 and 75% of cancer, dichotomous and continuous data sets with exponential and log-logistic models being the most frequently used model types for continuous and dichotomous data sets, respectively. Model success was largely dependent on how far extrapolation below the lowest non-zero dose was required. Even though BMD modeling is less study design dependent, study design featured as an important factor for modeling success. Studies with poorly modeled variance, low ρ-values and a lack of confidence in the calculated values were more often observed in those dose-response data sets that failed to be modeled successfully. Successful models had more dose groups, and, surprisingly, a lower number of animals per dose group. However, there was no correlation between the number of animals per dose group and the number of dose groups and the authors stressed that the increase of dose groups and reduction of animals per dose groups should not occur to the cost of statistical significance.
The authors indicate that the lack of adjustment for data source, effect severity, and consideration of the underlying biological response are limitations of the standardized approach. However, standardization may increase objectivity and enhance transparency and thus, facilitate communication with assessors, peer-reviewers and the general public. Standardized and thus largely automated BMD modeling allows for a time efficient analysis of large data sets. The authors consider that applying standardized BMD modeling could be even more successful when datasets are used designed specifically for BMD modeling. In conclusion, Wignall and colleagues argue that automated BMD modeling is particularly useful where speed and efficiency in dose response data analysis are priorities. However, they stress that expert judgment may still be needed to decide amongst viable alternative BMD/Ls.
*The lower confidence interval of the Benchmark Dose
Wignall, J. et al. (2014). “Standardizing benchmark dose calculation to improve science based decisions in human health assessments.” Environmental Health Perspectives (published online February 25, 2014).