Salmonella source attribution based on microbial subtyping: Does including data on food consumption matter?

https://doi.org/10.1016/j.ijfoodmicro.2014.09.010Get rights and content

Highlights

  • We assess whether food consumption data affect Salmonella source attribution models.

  • We use an additional parameter to account for consumption of raw/undercooked foods.

  • Food consumption data may have a significant impact on attribution estimates.

  • The additional parameter may help identify the correct primary reservoir.

  • Models can better reflect differences among consumed foods to transmit salmonellas.

Abstract

Source attribution based on microbial subtyping is being performed in many countries to ascertain the main reservoirs of human salmonellosis and to assess the impact of food safety interventions. To account for differences in exposure, the amount of food available for consumption within a country is often included in Salmonella source attribution models along with the level of contamination. However, not all foods have an equal probability of serving as vehicles for salmonellas, as some foods are more likely to be consumed raw/undercooked than others, posing a relatively higher risk. Using Salmonella data from the Netherlands in 2001–2004, this study aims at elucidating whether and how the incorporation of food consumption data in two source attribution models – the (modified) Dutch and Hald models – affects their attributions. We also propose the incorporation of an additional parameter to weight the amount of food consumed by its likelihood to be consumed raw/undercooked by the population. Incorporating the amount of food consumed caused a drastic change in the ranking of the top reservoirs in the Dutch model, but not in the Hald model, which proved to be insensitive to additional weightings given that its source-dependent factor can account for both food consumption and the ability for foods to serve as vehicles for salmonellas. Compared to attributions without food consumption, the Dutch model including the amount of food consumed showed an increase in the percentage of cases attributable to pigs and a decrease in that of layers/eggs, which became the second reservoir, after pigs. This was not consistent with established knowledge indicating that layers/eggs, rather than pigs, were the main reservoir of human salmonellosis in that period. By incorporating the additional weight reflecting the likelihood for different foods to be consumed raw/undercooked, the attributions of the Dutch model were effectively adjusted, both in terms of ranking and percent contributions of the different reservoirs. We concluded that incorporating food consumption data in the Dutch model can significantly affect the results. Therefore, such data should be either excluded from this model or used together with an additional weight able to adjust the amount of food consumed by its likelihood to be consumed insufficiently cooked. This may help identifying the correct reservoirs, allowing attributions to more closely reflect the real chance for a given food to serve as a vehicle for salmonellas. Conversely, the Hald model works properly irrespective of inclusion of food consumption data.

Introduction

Over 100,000 human cases of salmonellosis are notified each year in the 27 European Union (EU) Member States (~ 502 million population) (EFSA, 2011), yet it is widely accepted that notified cases are subject to considerable underreporting. By using disease risks of returning Swedish travelers for the period 2005–2009 anchored to a Dutch population-based study on gastroenteritis, Havelaar et al. (2013) have estimated that more than 6.2 million human cases of salmonellosis occur annually in the 27 EU Member States. This poses salmonellosis as the second most important cause of human bacterial gastroenteritis, after campylobacteriosis.

Contaminated food is by far the primary source of human salmonellosis, with 86–95% of cases estimated to be foodborne (Majowicz et al., 2010). To ascertain the main reservoirs of salmonellosis and to assess the impact of food safety interventions, source attribution based on microbial subtyping is being performed in many countries within (David et al., 2013a, David et al., 2013b, Hald et al., 2004, Hald et al., 2007, Mughini-Gras et al., 2014a, Mughini-Gras et al., 2014b, Pires et al., 2011, Pires and Hald, 2010, van Pelt et al., 1999, Wahlström et al., 2011) and outside (Guo et al., 2011, Mullner et al., 2009a) the EU. The microbial subtyping approach, based on the comparison of frequency distributions of pathogen subtypes isolated from humans with those isolated from putative sources of infection (Barco et al., 2013, Pires et al., 2009), has received considerable attention since the development of the Hald model for Salmonella source attribution in Denmark (Hald et al., 2004). The Hald model uses a Bayesian approach and has been developed separately from the frequentist Dutch model (van Pelt et al., 1999) to attribute stochastically human cases to putative sources of infection while accounting for differences in Salmonella subtypes and sources to cause human infection (Hald et al., 2004). To further improve identifiability and to handle uncertainty in data of poorer quality, a modified Hald model has also been proposed (Mullner et al., 2009a). More recently, a modified Dutch model has been developed and its attributions do not seem to differ from those of the modified Hald model (Mughini-Gras et al., 2014b).

The amount of food consumed or available for consumption within a country (expressed as either total tons of food consumed or average food intake per capita) is often included in Salmonella source attribution models to account for differences in the overall exposure to the different sources. For this purpose a variety of food consumption data have been used, including data from national food (agricultural) supply statistics (Guo et al., 2011, Hald et al., 2004, Hald et al., 2007, Mughini-Gras et al., 2014a, Mughini-Gras et al., 2014b, Pires et al., 2011, Pires and Hald, 2010) and individual dietary surveys (David et al., 2013a, David et al., 2013b). Food supply statistics are usually readily available for many years in most countries. However, even when these data are adjusted for losses and wastage of inedible portions during food processing, distribution, storing and preparation (at either retail or household level), their use requires the assumption that all food available for consumption is effectively consumed (Guo et al., 2011). Conversely, individual dietary surveys are costly and labor-intensive, but they may provide more accurate data on the actual food intake and even on the practices of preparation. Yet, individual dietary surveys are hampered by a plethora of limitations linked to misreporting (mainly recall and prevarication biases), non-coverage and/or non-response of certain groups of the population, as well as seasonality and trends in food consumption patterns (Hallström and Börjesson, 2013). Therefore, to be functional to source attribution, these surveys need to be performed on a regular basis or at least be updated by combining their figures with food supply data. This was done by David et al. (2013b) for attributing salmonellosis cases that occurred in France in 2005 using individual food consumption data obtained from a French survey conducted in 1999, when the BSE crisis in Europe had almost reached its highest levels and had already had serious repercussions in the food market.

An aspect that is often overlooked when using food consumption data in source attribution analyses is that not all foods have the same probability of serving as vehicles for salmonellas, as some foods are clearly more likely to be consumed raw or undercooked than others (Mughini-Gras et al., 2014a, Mughini-Gras et al., 2014b). Such foods are relatively more likely to cause infection directly, and possibly also indirectly through cross-contamination (Hald et al., 2004). The Hald-type models are supposed to account for this by incorporating the so-called source-dependent parameter (aj, see Section 2.3.2), which is an uninformative multiplication factor fitted by the model to arrive to the most likely solution given the observed data (Hald et al., 2004, Mullner et al., 2009a). It has been suggested that values taken by this parameter are somehow biologically interpretable, with high values for sources such as beef being compatible with its higher likelihood of being consumed raw or just slightly cooked (Hald et al., 2004). However, evidence for this notion is rather inconsistent, as other model runs may depict situations that are not directly interpretable with insufficient cooking of certain foods (Pires and Hald, 2010), as the aj parameter also accounts for other factors (Section 2.3.2).

Using Salmonella data from the Netherlands during 2001–2004, this study aims at elucidating whether and how the inclusion of data on food consumption in the (modified) Dutch and Hald models influences their attributions. Next, we propose the incorporation of an additional parameter in the modified Dutch model that weights the amount of food consumed by its likelihood to be consumed raw or undercooked by the population, thereby allowing food consumption data in this source attribution model to reflect more closely the actual chance for a given food to serve as a vehicle for Salmonella.

Section snippets

Data on human salmonellosis cases

A dataset consisting of 3735 sero/phage typed Salmonella isolates from human cases that occurred in the Netherlands between January 2002 and December 2003 was obtained from the EU and national reference laboratory for Salmonella at the Dutch National Institute for Public Health and the Environment (RIVM). These salmonellosis cases were identified by the Dutch Regional Public Health Laboratories through passive surveillance of people with gastroenteritis. Serotyping and further phage typing of S.

Attributions without food consumption weights and without modeled pathogen prevalence

Attribution estimates from the oDM and mHM with no food consumption weights or modeled prevalence pij are reported in panel A of Fig. 1, Fig. 2, respectively. Layers/eggs were the primary reservoir, accounting for 43.5% (95% CI: 41.5–45.5%) and 55.6% (48.6–61.1%) of cases in these oDM and mHM, respectively. Pigs were the second most important reservoir (oDM: 29.3%, 28.1–30.5%; mHM: 38.2%, 31.5–43.7%), followed by cattle (oDM: 14.6%, 13.2–16.2%; mHM: 3.7%, 0.4–10.3%) and broilers (oDM: 12.6%,

Discussion

We assessed the effect of including and excluding data on food consumption in two commonly used Salmonella source attribution models based on microbial subtyping. We also proposed the incorporation of an additional parameter in the mDM to weight the amount of food consumed by its likelihood to be consumed raw or undercooked by the population.

We found that the incorporation of the standard food consumption weight mj, together with the modeled pathogen prevalence, caused a drastic change in the

Conclusions

Going back to the original title question, does including data on food consumption matter? The answer depends on the attribution model in question. The incorporation of the amount of food consumed appears to bias the attribution estimates of the mDM, but not those of the mHM. Yet the incorporation of the additional raw/undercooked food consumption weight into the mDM may compensate for this bias and effectively adjust the attribution estimates. Conversely, the mHM works properly regardless of

References (29)

  • A. Gelman et al.

    Inference from iterative simulation using multiple sequences

    Stat. Sci.

    (1992)
  • C. Guo et al.

    Application of Bayesian techniques to model the burden of human salmonellosis attributable to U.S. food commodities at the point of processing: adaptation of a Danish model

    Foodborne Pathog. Dis.

    (2011)
  • T. Hald et al.

    A Bayesian approach to quantify the contribution of animal-food sources to human salmonellosis

    Risk Anal.

    (2004)
  • T. Hald et al.

    The attribution of human infections with antimicrobial resistant Salmonella bacteria in Denmark to sources of animal origin

    Foodborne Pathog. Dis.

    (2007)
  • Cited by (27)

    • Attribution of Listeria monocytogenes human infections to food and animal sources in Northern Italy

      2020, Food Microbiology
      Citation Excerpt :

      Source attribution based on microbial subtyping relies on the characterisation of isolates using different phenotyping or genotyping methods (Andreoletti et al., 2008). Human cases are then probabilistically attributed to sources by comparing the subtype distributions of human source strains through mathematical models (Mughini-Gras and van Pelt, 2014). Two main families of source attribution models are available: the so-called ‘frequency matching’ and ‘population genetics’ models, each with several advantages and disadvantages, as discussed in a recent opinion paper (Mughini-Gras et al., 2018a, 2018b).

    • Changing epidemiology of invasive non-typhoid Salmonella infection: a nationwide population-based registry study

      2020, Clinical Microbiology and Infection
      Citation Excerpt :

      A p <0.05 was considered statistically significant. Source attribution was performed using the ‘modified Dutch model’ based on all serotyping data, as described previously [4–7]. Briefly, this model inferred probabilistically the sources of human infections by comparing the serotype distributions of patients and animal reservoirs, weighted by the NTS prevalence in each reservoir and human exposure thereto (Table S1).

    • In vitro anti-bacterial and anti-inflammatory activities of lactic acid bacteria-biotransformed mulberry (Morus alba Linnaeus) fruit extract against Salmonella Typhimurium

      2019, Food Control
      Citation Excerpt :

      Biotransformation is used to generate new compounds or modify the structure of compounds present in the matrix composition, and hence, bioactive molecules with potential applications in health and medicine could be attained by these processes (Mazlan et al., 2015). S. Typhimurium is known as a globally widespread foodborne enteric pathogen responsible for acute gastroenteritis commonly associated with contaminated or undercooked foods (Mughini-Gras & van Pelt, 2014; Wang et al., 2013). Here, we demonstrated that MFE biotransformed by L. brevis DF01 and P. acidilactici K10 exerted effective anti-bacterial activity against S. Typhimurium by reducing the growth and biofilm formation by the bacteria, as well as its anti-inflammatory activity in human intestinal epithelial cells.

    • Attributable sources of community-acquired carriage of Escherichia coli containing β-lactam antibiotic resistance genes: a population-based modelling study

      2019, The Lancet Planetary Health
      Citation Excerpt :

      Input parameters and data origins are in the appendix (pp 3–4). We used a commonly applied source attribution model for pathogens based on typing data, the modified Hald model,22 with some adaptations for our study compared with its previous applications, for example, for Salmonella,22,23 Campylobacter,22 Listeria,24 and Shiga toxin-producing E coli.25 Our analysis was based on ESBL-EC and pAmpC-EC genes in the open community as the group to be attributed to the aforementioned human and non-human sources.

    View all citing articles on Scopus
    View full text