Uncertainty and confidence in applying mathematical models and in vitro data in toxicological safety assessments
Lead Research Organisation:
University of Leeds
Department Name: Statistics
Abstract
Over recent years, there has been increasing pressure to end the overreliance on animal experiments within product risk assessments and to consider the information from non-animal alternatives when making decisions about human safety. At the same time, there has been substantial research effort into the development of computer-based simulators of human biology. Future risk assessments should be able to combine advanced biology simulators with relevant non-animal experiments to help risk assessors make predictions about adverse health effects without the use of animals.
Currently, toxicologists are far from having the same confidence in these computer-based risk assessments as they do in traditional animal-based assessments. This is partially due to most investigations of the use of the computer models focussing on just a small aspect of the risk problem. We consider the complete risk assessment process to provide more scientifically relevant information for human safety assessment. In our research, we build a coherent framework with which the non-animal experiments and simulators can be harnessed to develop new approaches for safety assessment. These approaches could give a better prediction of potential effects in humans, reduce the time and monetary costs of reliable risk assessment, and reduce the reliance on animals.
Currently, toxicologists are far from having the same confidence in these computer-based risk assessments as they do in traditional animal-based assessments. This is partially due to most investigations of the use of the computer models focussing on just a small aspect of the risk problem. We consider the complete risk assessment process to provide more scientifically relevant information for human safety assessment. In our research, we build a coherent framework with which the non-animal experiments and simulators can be harnessed to develop new approaches for safety assessment. These approaches could give a better prediction of potential effects in humans, reduce the time and monetary costs of reliable risk assessment, and reduce the reliance on animals.
Technical Summary
Over the past decade, there has been substantial research effort into the development and utilisation of systems biology models for a range of applications. At the same time, there has been increasing pressure to end the overreliance on animal experiments within toxicology and to consider the information from non-animal alternatives when making decisions about human safety. A key challenge is to understand how the different proposed elements of a pathways-based toxicological risk assessment (e.g. chemical characterisation, human exposure information, in vitro toxicity testing data on compounds and metabolites from potentially diverse test systems, dose response modelling, population-based modelling and risk contexts) can be integrated to enable robust safety decisions.
Our objectives are to answer how mathematical models could fit into a risk assessment alongside relevant in vitro data sources and to give toxicological risk assessors more confidence in using mathematical models. To achieve this, we must understand what the key input parameters of the models are and how existing in vivo and in vitro assays can be used to validate and parameterise the models. Using data sets (e.g. ToxCast, EPA) and models, which are publicly available or provided by Unilever, and working closely with expert toxicologists and risk assessors at Unilever, we will demonstrate the potential of existing uncertainty analysis techniques to evaluate the confidence in a toxicological risk assessment decision using non-animal data. This will take the form of prototype risk assessments based on in vitro data interpreted through a systems biology model and compared to human exposure concentrations estimated using a physiologically-based pharmacokinetic model. The approach therefore requires the linking of models at different scales and the determination of the contributions of experimental and model uncertainty to assess the confidence in the resultant risk assessment.
Our objectives are to answer how mathematical models could fit into a risk assessment alongside relevant in vitro data sources and to give toxicological risk assessors more confidence in using mathematical models. To achieve this, we must understand what the key input parameters of the models are and how existing in vivo and in vitro assays can be used to validate and parameterise the models. Using data sets (e.g. ToxCast, EPA) and models, which are publicly available or provided by Unilever, and working closely with expert toxicologists and risk assessors at Unilever, we will demonstrate the potential of existing uncertainty analysis techniques to evaluate the confidence in a toxicological risk assessment decision using non-animal data. This will take the form of prototype risk assessments based on in vitro data interpreted through a systems biology model and compared to human exposure concentrations estimated using a physiologically-based pharmacokinetic model. The approach therefore requires the linking of models at different scales and the determination of the contributions of experimental and model uncertainty to assess the confidence in the resultant risk assessment.
Planned Impact
By providing a sound quantitative framework for bringing both mathematical toxicological models and in vitro data sources to bear on a risk assessment, we believe that, if the data and models are of sufficient quality, animal experiments to support the safety of consumer products would become unnecessary. When ultimately realised, this will have a significant effect on the overall animal use. The Home Office reported that the number of animals used for industry safety assessments in 2009 was 18,369 in the UK alone. We would like to accelerate the progress towards this goal by first gaining a complete understanding of the interactions between models and the in vitro data sources and then by disseminating the results of completed case studies that will focus on the overall risk problem rather than compartmentalised aspects of that problem.
However, in the short term, we do expect that the current project would have an impact on the reduction of animals, particularly in the area of toxicokinetic modelling. At present, animal studies are often used to define parameters for toxicokinetic models. In 2009, the number of animal used in toxicokinetic studies was estimated at 24,910. Through the uncertainty and sensitivity analysis techniques, the proposed research will enable us to determine how crucial precise information on toxicokinetics is to overall risk assessment case study.
Also, over recent decades, there has been greater emphasis on quantitative characterisation of risk because using some common quantitative metric makes prioritisation of actions to counteract the risks easier. One of the aims of this project is to formalise the appreciation of uncertainty in the use of mathematical models for risk assessment. A benefit of accounting for uncertainty is that we can determine how conservative the risk management decisions are. Risk managers want to identify a safe value without being unnecessarily over-conservative. To do this, they need to know the relative likelihoods of all possible outcomes, not just an arbitrary point estimate.
In collaborating with the University of Leeds on this project, Unilever will benefit from the research because the research will help bring together their existing work aimed at replacement (in vitro data generation for example) to ensure the information can be used in risk assessment. They will also benefit from being exposed to the state-of-the-art statistical analysis techniques for analysing computer-based simulators and to methods for formally quantifying uncertainty because they are transferable to other risk assessments and data analyses. Through our planned dissemination (especially the planned tutorials), we can open these benefits up to other industry members and regulatory agencies.
However, in the short term, we do expect that the current project would have an impact on the reduction of animals, particularly in the area of toxicokinetic modelling. At present, animal studies are often used to define parameters for toxicokinetic models. In 2009, the number of animal used in toxicokinetic studies was estimated at 24,910. Through the uncertainty and sensitivity analysis techniques, the proposed research will enable us to determine how crucial precise information on toxicokinetics is to overall risk assessment case study.
Also, over recent decades, there has been greater emphasis on quantitative characterisation of risk because using some common quantitative metric makes prioritisation of actions to counteract the risks easier. One of the aims of this project is to formalise the appreciation of uncertainty in the use of mathematical models for risk assessment. A benefit of accounting for uncertainty is that we can determine how conservative the risk management decisions are. Risk managers want to identify a safe value without being unnecessarily over-conservative. To do this, they need to know the relative likelihoods of all possible outcomes, not just an arbitrary point estimate.
In collaborating with the University of Leeds on this project, Unilever will benefit from the research because the research will help bring together their existing work aimed at replacement (in vitro data generation for example) to ensure the information can be used in risk assessment. They will also benefit from being exposed to the state-of-the-art statistical analysis techniques for analysing computer-based simulators and to methods for formally quantifying uncertainty because they are transferable to other risk assessments and data analyses. Through our planned dissemination (especially the planned tutorials), we can open these benefits up to other industry members and regulatory agencies.
