Thermotolerant Campylobacter is the most commonly reported zoonotic disease in the European Union and a leading cause of zoonotic enteric infections worldwide. To support its control, quantitative microbiological risk assessments, or QMRAs, have been performed in various countries as well as in an international context. Several foods have been indicated as potential sources of Campylobacter, but QMRAs have predominantly focused on what is generally considered the main route of transmission: the broiler meat production chain and human exposure associated with the consumption of broiler meat. Published models vary extensively in the way the exposure assessment is performed, but until recently the vast majority of QMRAs applied a dose-response model based on a challenge study performed by Black and colleagues in 1988.

That challenge study involved only a small group of healthy male volunteers who were exposed to a specific Campylobacter strain, yet it has for a long time been considered the best basis for Campylobacter dose-response modelling in QMRA and has been referred to as the classic dose-response model. After an earlier reconsideration of this classic model, Teunis and colleagues published an analysis in 2018 including newly published dose-response data, comprising various challenge studies on humans and primates as well as four studies based on outbreaks associated with the consumption of raw milk. The dose-response model applied to these data suggested a different dose-response curve and higher infectivity of Campylobacter, leading the authors to state that the classic model had been an unfortunate choice as the default model used for this pathogen. Although Teunis and colleagues described their analyses in detail and provided the software code for their model, application of their model in a QMRA of Campylobacter remained challenging, because a ready-to-use model was not provided and developing one requires considerable expertise in modelling and programming.

The newly published implementation addresses this gap by providing the dose-response model developed by Teunis and colleagues in the Food Safety Knowledge Exchange, or FSKX, format. This standardised exchange format for mathematical models and simulation results contains the model script, visualisation script, and simulation settings, as well as a harmonised way to annotate the model through a metadata schema that uses controlled vocabularies wherever possible. The format facilitates re-usability, the application of the model in a QMRA, and efficient knowledge exchange. The research project Risk Assessment Modelling and Knowledge Integration Platforms, known as RAKIP, initiated the development of the FSKX format and included the metadata schema to annotate knowledge. The RAKIP community, with partners from the French Agency for Food, Environmental and Occupational Health & Safety, the Technical University of Denmark National Food Institute, and the German Federal Institute for Risk Assessment, continues to improve and extend the community-driven metadata schema.

The model parameters are based on data from a set of different studies on the infectivity and pathogenicity of Campylobacter jejuni. Both challenge studies and outbreaks are considered, and users can decide which of these is most suitable for their purpose. Three groups of studies were included in the underlying analysis: four controlled human infection studies involving three distinct strains, four studies on outbreaks of unknown strains and, in one case, strain 81-176, and five challenge studies in three species of non-human primates of strains 81-176, 78-37, and V212X. All studies recorded both asymptomatic infection and illness as endpoints, and the data were used to parameterise the dose-response model through a multilevel approach that included effects of host species, different strains of the pathogen, and differentiation between outbreak and non-outbreak settings.

Technically, dose-response models for infection are usually based on a limited number of biologically motivated axioms. The basic starting point is that each organism has a usually low single-hit probability of initiating infection, and secondly, each organism acts independently of the others, with no synergism. Putting these axioms together yields a simple non-probabilistic model for the probability of infection given exposure to a number of organisms. In the Teunis model, the ingested dose is assumed to be Poisson distributed with a parameter representing the mean number of organisms per dose, and the single-hit probability varies between exposures following a Beta distribution, yielding an expression based on the Kummer confluent hypergeometric function. The model describing the probability of illness among infected subjects is based on other principles: it is no longer a matter of a single organism initiating infection, but rather the resulting growth of the population of organisms that should outrun the defensive measures of the immune system. The unconditional probability of illness is calculated by multiplying the conditional probability for illness given infection and the probability of infection. Note that the probability of illness applies to the exposed population only, and the fraction of the population that is exposed should be derived from an exposure assessment, which is part of a full QMRA.

The parameters used to describe infection as well as those used to calculate the conditional probability of illness were estimated from the data from the challenge studies and outbreak studies. The means and standard deviations of transformed parameters can be used with a bivariate normal distribution to characterise uncertainty, and together with the Poisson-mean dose this allows calculation of the probability of infection and illness. The model metadata, which form part of the FSKX-file, annotate the model in a harmonised way and specify that the hazard is the microorganism Campylobacter jejuni, measured in colony forming units, with the adverse effect being asymptomatic or symptomatic infection with the pathogen, known as campylobacteriosis. Two target populations are defined: the general population and groups involved in a food-borne outbreak. The model is written in the R programming language and is available under a Creative Commons Attribution 4.0 licence as open access.

The researchers present examples of results for typical ingested doses and demonstrate the utility of their ready-to-use model re-implementation by supplying an executable model embedded in the manuscript. The default simulation represents values derived from challenge studies, while another simulation setting considers values derived from outbreaks, differing from the default only by a single parameter. In two scenarios considered, a dose of one colony forming unit is set, so outcomes represent the single-hit probabilities of illness and infection. These scenarios are interesting since they are lower bounds of infection or illness risk, and secondly since they allow comparison of the results to single-hit infection probabilities of other pathogens. Two more scenarios are presented, again for both the challenge and outbreak situation, but instead of a constant dose, the dose is set to a distribution of concentrations in poultry meat based on data from a 2012 study by Swart and Havelaar. Results are visualised as boxplots showing the probability of illness and infection for the human population that consumes Campylobacter jejuni-contaminated food, with the white box representing the 25 and 75 percent quantiles, the median indicated as a black bar inside the box, whiskers corresponding to maximum and minimum values, and dots representing extreme values.

The results confirm the trends expected from the underlying analysis. The median probability of illness is always lower than the probability of infection, and although the trend is the same, the probability of illness differs drastically between the results based on the challenge studies and the outbreak dataset, with a higher probability of illness if the outbreak studies data are applied. When various mean doses are considered, sampled from a normal distribution of concentrations with a mean of minus one log colony forming unit and a standard deviation of 1.5, the same trends hold, and an overall lower chance to become infected and ill is observed compared to the single-dose scenarios. The authors note that in these analyses the distribution of mean doses is assumed to represent uncertainty, not variability between ingested doses, so the result is uncertainty in the probability of infection or illness. The high values found for the probability of infection and the probability of illness for outbreak strains are in agreement with the results obtained by Teunis and colleagues in 2018.

Dose-response models, as part of the hazard characterisation, are an indispensable ingredient of any QMRA model. Each microbial pathogen requires its own dose-response model, since infectivities of different species may differ substantially. In addition, pathogenicity may differ between bacterial strains of a taxonomical unit and even susceptibilities between individuals vary, due to individual host characteristics or acquired immunity. However, scarcity of data to characterise the variability between bacterial strains and hosts precludes dose-response models specified at this level of detail. The current study focuses on the recently published dose-response model for Campylobacter jejuni, which is based on a variety of dose-response studies and offers an alternative to the classic model that has been most frequently used in Campylobacter risk assessments so far. As shown by Teunis and colleagues, the recent model suggests that Campylobacter is more virulent than previously assumed. By presenting the model annotated with all metadata and implemented in the ready-to-use FSKX-format, the researchers allow easy model integration and simulation result interpretation in future QMRA studies. On the one hand, the model in the FSKX-file is available as R code that can be integrated into a fully fledged QMRA model. On the other hand, for a swift assessment of infection and illness probabilities at given mean doses, the current executable paper provides a convenient resource. Moreover, by using software like FSK-Lab, the user is able to join the model with an exposure assessment and perform a complete QMRA. In conclusion, the researchers provide an annotated, ready-to-use implementation of the Campylobacter jejuni dose-response model that stands to make quantitative risk assessment of this important foodborne pathogen more accessible, transparent, and reproducible than ever before.

Subject of Research: A ready-to-use dose-response model of Campylobacter jejuni implemented in the FSKX-standard

Article Title: A ready-to-use dose-response model of Campylobacter jejuni implemented in the FSKX-standard

Article References: Sundermann, E. M., Nauta, M., & Swart, A. (2021). A ready-to-use dose-response model of Campylobacter jejuni implemented in the FSKX-standard. Food Modelling Journal, 2, Article e63309. https://doi.org/10.3897/fmj.2.63309

Image Credits: AI Generated

DOI: 10.3897/fmj.2.63309

Keywords: Campylobacter jejuni, dose-response model, FSKX-standard, quantitative microbiological risk assessment, food safety, QMRA, ready-to-use, dose-response, model, Campylobacter, jejuni, implemented