Periodic Reporting for period 1 - FANTASTICAL (NOVEL TOOLS FOR FOOD SAFETY MANAGEMENT BASED ON QMRA WITH A ROBUST MODELLING OF UNNOVEL TOOLS FOR FOOD SAFETY MANAGEMENT APPLYING QMRA WITH ROBUST MODELLING OF UNCERTAINTY AND VARIABILITY: FANTASTICAL)
Reporting period: 2020-04-01 to 2022-03-31
Food safety remains a main concern for consumers, society, health agencies and food industry. Despite the efforts made during the last years, the number of outbreaks and food borne diseases is still high. Recent outbreaks within the European Union (e.g. Listeria monocytogenes in frozen vegetables) show that efforts are needed to coordinate scientific advances into decision making in order to implement new measures to increase consumer protection.
Quantitative Microbial Risk Analysis (QMRA) can be applied to support food safety management. QMRA is based on a mathematical description of the microbial response during the farm-to-fork chain of the product. Its application requires an accurate characterization of uncertainty and variability, inherent to any biological process. Because of these, QMRA should ideally follow a probabilistic approach, considering the variance of the response variables. Consequently, decisions must be made with an acceptable level of risk given the knowledge gaps. A deeper knowledge of both uncertainty and variability is a top priority in the EU, since it would allow a better application of QMRA, leading to better safety standards for policy makers and industry.
FANTASTICAL aims to develop novel approaches and tools for QMRA that can be implemented by all the stakeholders, i.e. agencies related to consumer protection (EFSA, ECDC) and industry. It will link a database of microbial responses with robust statistical functions for variance analysis and stochastic simulation in a user-friendly software. This project will reach out to the potential users and provide them with hands-on, open access tools to better understand microbial variability and uncertainty, resulting in more realistic QMRA. Thus, it will lead to an improvement in consumer protection and safer products.
Quantitative Microbial Risk Analysis (QMRA) can be applied to support food safety management. QMRA is based on a mathematical description of the microbial response during the farm-to-fork chain of the product. Its application requires an accurate characterization of uncertainty and variability, inherent to any biological process. Because of these, QMRA should ideally follow a probabilistic approach, considering the variance of the response variables. Consequently, decisions must be made with an acceptable level of risk given the knowledge gaps. A deeper knowledge of both uncertainty and variability is a top priority in the EU, since it would allow a better application of QMRA, leading to better safety standards for policy makers and industry.
FANTASTICAL aims to develop novel approaches and tools for QMRA that can be implemented by all the stakeholders, i.e. agencies related to consumer protection (EFSA, ECDC) and industry. It will link a database of microbial responses with robust statistical functions for variance analysis and stochastic simulation in a user-friendly software. This project will reach out to the potential users and provide them with hands-on, open access tools to better understand microbial variability and uncertainty, resulting in more realistic QMRA. Thus, it will lead to an improvement in consumer protection and safer products.
At the beginning of the project, we retrieved the information generated within the Laboratory of Food Microbiology of Wageningen University during the last couple of decades about the inactivation of microorganisms. This data was completed with a systematic review on the inactivation of two relevant bacterial pathogens (Bacillus cereus and Listeria monocytogenes) from the scientific literature. The results were compiled in a modern, noSQL database architecture (MongoDB).
Then, we reviewed the state of the art regarding methodologies used for variability analysis within the context of QMRA and compared them. For the quantification of variability from experimental data we compared multilevel Bayesian models, mixed-effects models and the simplified method by Aryani. We also analyzed two methods following a different approach: meta-regression models and stochastic models. Based on the results of these analyses, we made recommendations for method selection, which were compiled in four scientific articles (2 published, 2 under review).
To facilitate the application of these complex statistical methods, we developed two Open Source applications to support QMRA: biogrowth (https://foodmicrowur.shinyapps.io/biogrowth/) and D database (https://foodmicrowur.shinyapps.io/Ddatabase/). The former provides a toolbox to facilitate modeling microbial growth, including model fitting and prediction. The second one serves as an interface to the database developed in WP1. It includes advanced data analysis tools to apply the information included in the database in the context of a QMRA, making an emphasis on variability assessment.
Finally, the statistical methodologies developed within the project were applied to two different case studies relevant for food safety. The first one was developed in collaboration with the University of Zaragoza (Spain), where we used multilevel models to analyse the survival of Salmonella spp. on egg products accounting for different sources of variability. The second one was developed in collaboration with Hokkaido University (Japan), where we performed a QMRA for listeriosis in pasteurised milk. Moreover, in collaboration with researchers from the University of Cornell (USA) and the University of Maryland (USA), we developed recommendations on how to analyse variability and uncertainty in the context of microbial risk assessment. These studies were compiled in 3 scientific articles (1 published, 2 in preparation).
Then, we reviewed the state of the art regarding methodologies used for variability analysis within the context of QMRA and compared them. For the quantification of variability from experimental data we compared multilevel Bayesian models, mixed-effects models and the simplified method by Aryani. We also analyzed two methods following a different approach: meta-regression models and stochastic models. Based on the results of these analyses, we made recommendations for method selection, which were compiled in four scientific articles (2 published, 2 under review).
To facilitate the application of these complex statistical methods, we developed two Open Source applications to support QMRA: biogrowth (https://foodmicrowur.shinyapps.io/biogrowth/) and D database (https://foodmicrowur.shinyapps.io/Ddatabase/). The former provides a toolbox to facilitate modeling microbial growth, including model fitting and prediction. The second one serves as an interface to the database developed in WP1. It includes advanced data analysis tools to apply the information included in the database in the context of a QMRA, making an emphasis on variability assessment.
Finally, the statistical methodologies developed within the project were applied to two different case studies relevant for food safety. The first one was developed in collaboration with the University of Zaragoza (Spain), where we used multilevel models to analyse the survival of Salmonella spp. on egg products accounting for different sources of variability. The second one was developed in collaboration with Hokkaido University (Japan), where we performed a QMRA for listeriosis in pasteurised milk. Moreover, in collaboration with researchers from the University of Cornell (USA) and the University of Maryland (USA), we developed recommendations on how to analyse variability and uncertainty in the context of microbial risk assessment. These studies were compiled in 3 scientific articles (1 published, 2 in preparation).
The FANTASTICAL project has created new scientific knowledge with respect to the state of the art. Within the project, we have analyzed and compared different statistical methods to analyze variability and uncertainty in the context of QMRA. Namely, we have studied mixed-effects models, multilevel Bayesian models, the simplified method by Aryani, meta-regression models and stochastic models. Based on the results of our analysis, we have defined recommendations for method selection and guidelines for their application. Furthermore, we have deepened in the application of multilevel Bayesian models studying how this methodology can be exploited to better estimate the relevance of variability from experimental data and include it in model predictions (Garre et al., 2020 https://doi.org/10.1016/j.foodres.2020.109374).
Apart from these methodological advances, within the project we were able to create two novel software tools to analyze variability and uncertainty in the context of microbial risk assessment: biogrowth and D database. The biogrowth software is a web application that facilitates modeling of microbial growth. It includes functions for fitting parametric models to experimental data, as well as for making predictions. These predictions can be either deterministic (i.e. growth curves) or stochastics accounting for parameter uncertainty. The user has the possibility to choose among a variety of models from predictive microbiology. The tool is Open Code and is available from free in the following link: https://foodmicrowur.shinyapps.io/biogrowth/
The second application, D database, compiles the data on microbial inactivation gathered by the host group during roughly the last 20 years. This data is complemented with the results of a systematic review. It is also Open Code and freely available online (https://foodmicrowur.shinyapps.io/Ddatabase/). Besides providing advanced search features (by microorganisms, by food, by type of experiment, by modeling approach…), the web application also includes advanced data analysis tools (e.g. interactive visualizations) that facilitate the analysis of the data, with an emphasis on its variability. It also includes advanced tools for mathematical modeling (e.g. meta-regression models) to support building predictive models (including variability) for QMRA.
Apart from these methodological advances, within the project we were able to create two novel software tools to analyze variability and uncertainty in the context of microbial risk assessment: biogrowth and D database. The biogrowth software is a web application that facilitates modeling of microbial growth. It includes functions for fitting parametric models to experimental data, as well as for making predictions. These predictions can be either deterministic (i.e. growth curves) or stochastics accounting for parameter uncertainty. The user has the possibility to choose among a variety of models from predictive microbiology. The tool is Open Code and is available from free in the following link: https://foodmicrowur.shinyapps.io/biogrowth/
The second application, D database, compiles the data on microbial inactivation gathered by the host group during roughly the last 20 years. This data is complemented with the results of a systematic review. It is also Open Code and freely available online (https://foodmicrowur.shinyapps.io/Ddatabase/). Besides providing advanced search features (by microorganisms, by food, by type of experiment, by modeling approach…), the web application also includes advanced data analysis tools (e.g. interactive visualizations) that facilitate the analysis of the data, with an emphasis on its variability. It also includes advanced tools for mathematical modeling (e.g. meta-regression models) to support building predictive models (including variability) for QMRA.