Ring CL, Urban J, Wikoff D, Thompson C, Budinsky RA, Haws LC. Application of systematic review and quantitative evidence integration methods to support risk assessment: Characterization of the dose-response relationship between exposure to dioxin-like compounds (DLC) and sperm count. Poster at Society of Toxicology Annual Meeting, Baltimore, MD, March 2019.
Systematic review is being used globally to support the risk assessment process; however, best practices for integrating evidence are still being developed. There is particular interest in developing or refining approaches to quantitatively integrate heterogeneous toxicology data. As part of a systematic review to evaluate the dose-response relationship between DLCs and reduced sperm count, we explored the utility and feasibility of using meta-regression to quantitatively integrate doseresponse data from experimental animal studies. Seven studies were selected for a pilot evaluation of feasibility based on secondary reviews; these included studies with single and multiple dose groups, single exposures and repeated exposures, subcutaneous and oral exposures, as well as studies that reported a relationship and those that did not. Exposure and outcome metrics were standardized, and meta-regression models were fit using the R meta-analysis package “metafor.” Based on the best-fit models, points of departure (PODs) similar to benchmark dose (BMD) and lower-bound BMD (BMDL) metrics were generated. Heterogeneity was large for this pilot data set, which was expected, given that the studies were purposely chosen to bracket the full range of diversity in the data set. Results allow for the exploration of sensitivity in model selection and other parameters as they relate to POD determination. Depending on threshold, model, and fitting parameters, PODs ranged from < 1.5 ng/kgday to > 15 ng/kg-day, thus highlighting the need for care in model selection during the full study. The pilot results demonstrate the utility of the technique to estimate an overall average dose-response relationship and the consistency of dose-response across studies, as well as to use data that could not be analyzed using a traditional BMD approach. In conclusion, the pilot analysis indicated that metaregression is feasible and useful for this data set, thus allowing for consideration of the entire evidence base vs reliance on single datasets. Such methods provide important information to risk managers, allowing for greater accommodation of the totality of evidence as well as characterization of uncertainty.