Results of QA/QC Testing of EPA Benchmark Dose Software Version 1.2
EPA is developing benchmark dose software (BMDS) to support cancer and non-cancer dose-response assessments. Following the recent public review of BMDS version 1.1b, EPA developed a Hill model for evaluating continuous data, and improved the user interface and Multistage, Polynomial and Power models. The BMDS system was subjected to an extensive QA/QC program, which included the independent verification of 7 dichotomous, 3 continuous and 3 developmental models. Each BMDS model was compared to an equivalent non-EPA model using data sets obtained from EPA cancer and non-cancer data bases. Many different configurations (combinations of options) for each of the EPA and non-EPA models were applied to 100 dichotomous or continuous dose-response data sets, resulting in well over 1,000 runs per model. Textual and graphical results of the model runs (e.g., maximum likelihood values, p-values, shape of the fitted curve) were collected and evaluated to determine strengths and weaknesses of each model. Preliminary results indicate that the BMDS dichotomous models calculate a BMD over 99% of the time, and fitted parameters such as benchmark dose and likelihood values from the BMDS Multistage model agree with values obtained from the THRESH model developed by Crump, Inc. The only deviations between the BMDS Multistage and THRESH were degrees of freedom differences for the chi-squared goodness-of-fit statistic resulting from a philosophical divergence on the appropriate approach for background estimates that reach the zero lower bound constraint. EPA will initiate a similar QA/QC analysis of several new BMDS models, including the Agency's new cancer model, a dichotomous Hill model and additional continuous (e.g., Hybrid) models in 2000.