Machine learning to refine decision making within a syndromic surveillance service.

BACKGROUND: Worldwide, syndromic surveillance is increasingly used for improved and timely situational awareness and early identification of public health threats. Syndromic data streams are fed into detection algorithms, which produce statistical alarms highlighting potential activity of public health importance. All alarms must be assessed to confirm whether they are of public health importance. In England, approximately 100 alarms are generated daily and, although their analysis is formalised through a risk assessment process, the process requires notable time, training, and maintenance of an expertise base to determine which alarms are of public health importance. The process is made more complicated by the observation that only 0.1% of statistical alarms are deemed to be of public health importance. Therefore, the aims of this study were to evaluate machine learning as a tool for computer-assisted human decision-making when assessing statistical alarms. METHODS: A record of the risk assessment process was obtained from Public Health England for all 67,505 statistical alarms between August 2013 and October 2015. This record contained information on the characteristics of the alarm (e.g. size, location). We used three Bayesian classifiers- naïve Bayes, tree-augmented naïve Bayes and Multinets - to examine the risk assessment record in England with respect to the final 'Decision' outcome made by an epidemiologist of 'Alert', 'Monitor' or 'No-action'. Two further classifications based upon tree-augmented naïve Bayes and Multinets were implemented to account for the predominance of 'No-action' outcomes. RESULTS: The attributes of each individual risk assessment were linked to the final decision made by an epidemiologist, providing confidence in the current process. The naïve Bayesian classifier performed best, correctly classifying 51.5% of 'Alert' outcomes. If the 'Alert' and 'Monitor' actions are combined then performance increases to 82.6% correctly classified. We demonstrate how a decision support system based upon a naïve Bayes classifier could be operationalised within an operational syndromic surveillance system. CONCLUSIONS: Within syndromic surveillance systems, machine learning techniques have the potential to make risk assessment following statistical alarms more automated, robust, and rigorous. However, our results also highlight the importance of specialist human input to the process.

Climate Change, Foodborne Pathogens and Illness in Higher-Income Countries.

PURPOSE OF REVIEW: We present a review of the likely consequences of climate change for foodborne pathogens and associated human illness in higher-income countries. RECENT FINDINGS: The relationships between climate and food are complex and hence the impacts of climate change uncertain. This makes it difficult to know which foodborne pathogens will be most affected, what the specific effects will be, and on what timescales changes might occur. Hence, a focus upon current capacity and adaptation potential against foodborne pathogens is essential. We highlight a number of developments that may enhance preparedness for climate change. These include the following: Adoption of novel surveillance methods, such as syndromic methods, to speed up detection and increase the fidelity of intervention in foodborne outbreaks Genotype-based approaches to surveillance of food pathogens to enhance spatiotemporal resolution in tracing and tracking of illness Ever increasing integration of plant, animal and human surveillance systems, One Health, to maximise potential for identifying threats Increased commitment to cross-border (global) information initiatives (including big data) Improved clarity regarding the governance of complex societal issues such as the conflict between food safety and food waste Strong user-centric (social) communications strategies to engage diverse stakeholder groups The impact of climate change upon foodborne pathogens and associated illness is uncertain. This emphasises the need to enhance current capacity and adaptation potential against foodborne illness. A range of developments are explored in this paper to enhance preparedness.

New Elements To Consider When Modeling the Hazards Associated with Botulinum Neurotoxin in Food.

Botulinum neurotoxins (BoNTs) produced by the anaerobic bacterium Clostridium botulinum are the most potent biological substances known to mankind. BoNTs are the agents responsible for botulism, a rare condition affecting the neuromuscular junction and causing a spectrum of diseases ranging from mild cranial nerve palsies to acute respiratory failure and death. BoNTs are a potential biowarfare threat and a public health hazard, since outbreaks of foodborne botulism are caused by the ingestion of preformed BoNTs in food. Currently, mathematical models relating to the hazards associated with C. botulinum, which are largely empirical, make major contributions to botulinum risk assessment. Evaluated using statistical techniques, these models simulate the response of the bacterium to environmental conditions. Though empirical models have been successfully incorporated into risk assessments to support food safety decision making, this process includes significant uncertainties so that relevant decision making is frequently conservative and inflexible. Progression involves encoding into the models cellular processes at a molecular level, especially the details of the genetic and molecular machinery. This addition drives the connection between biological mechanisms and botulism risk assessment and hazard management strategies. This review brings together elements currently described in the literature that will be useful in building quantitative models of C. botulinum neurotoxin production. Subsequently, it outlines how the established form of modeling could be extended to include these new elements. Ultimately, this can offer further contributions to risk assessments to support food safety decision making.

The pattern of growth observed for Clostridium botulinum type A1 strain ATCC 19397 is influenced by nutritional status and quorum sensing: a modelling perspective.

Botulinum neurotoxins (BoNTs) produced by the anaerobic bacterium Clostridium botulinum are the most poisonous substances known to mankind. However, toxin regulation and signals triggering synthesis as well as the regulatory network and actors controlling toxin production are unknown. Experiments show that the neurotoxin gene is growth phase dependent for C. botulinum type A1 strain ATCC 19397, and toxin production is influenced both by culture conditions and nutritional status of the medium. Building mathematical models to describe the genetic and molecular machinery that drives the synthesis and release of BoNT requires a simultaneous description of the growth of the bacterium in culture. Here, we show four plausible modelling options which could be considered when constructing models describing the pattern of growth observed in a botulinum growth medium. Commonly used bacterial growth models are unsuitable to fit the pattern of growth observed, since they only include monotonic growth behaviour. We find that a model that includes both the nutritional status and the ability of the cells to sense their surroundings in a quorum-sensing manner is most successful at explaining the pattern of growth obtained for C. botulinum type A1 strain ATCC 19397.

Status and future of Quantitative Microbiological Risk Assessment in China.

Since the implementation of the Food Safety Law of the People's Republic of China in 2009 use of Quantitative Microbiological Risk Assessment (QMRA) has increased. QMRA is used to assess the risk posed to consumers by pathogenic bacteria which cause the majority of foodborne outbreaks in China. This review analyses the progress of QMRA research in China from 2000 to 2013 and discusses 3 possible improvements for the future. These improvements include planning and scoping to initiate QMRA, effectiveness of microbial risk assessment utility for risk management decision making, and application of QMRA to establish appropriate Food Safety Objectives.

Shaking-induced crystallization of dense sphere packings.

We use a hybrid Monte Carlo algorithm to simulate the shaking of spheres at different vibrational amplitudes and find that spontaneous crystallization occurs in specific dynamical regimes. Several crystallizing transitions are typically observed, leading to end states which can be fully or partially ordered, depending on the shaking amplitude, which we investigate using metrics of global and local order. At the lowest amplitudes, crystallization is incomplete, at least for our times of observation. For amplitude ranges where crystallization is complete, there is typically a competition between hcp and fcc ordering. It is seen that fcc ordering typically predominates; in fact for an optimal range of amplitudes, spontaneous crystallization into a pure fcc state is observed. An interesting feature is the breakdown of global order when there is juxtaposition of fully developed hcp and fcc order locally: we suggest that this is due to the interfaces between the different domains of order, which play the same role as dislocations.

A risk assessment model for enterotoxigenic Staphylococcus aureus in pasteurized milk: a potential route to source-level inference.

This article describes a probabilistic model that quantifies hazards that arise from Staphylococcus aureus in milk that is sold as pasteurized in the United Kingdom. The model is centered on coupled dynamics for S. aureus populations, staphylococcal enterotoxins, and the concentration of alkaline phosphatase throughout the milk chain. The chain includes farm collection and storage of pooled milk, further pooling for off-farm processing, high temperature short time thermal processing, and possible postprocess contamination. The model is implemented as a Bayesian belief network. The results indicate that milk sold as pasteurized is relatively safe with respect to the hazards associated with S. aureus and that most risk is associated with small scale on-farm processing. An additional analysis of likelihood ratios shows that alkaline phosphatase concentrations in filler tank milk are a good indicator of potential hazards and that these concentrations, in conjunction with other measurements, can be used effectively to discriminate over possible failure modes. The ability to discriminate over potential failure modes can support preemptive actions, such as maintenance or hygiene, which assist with milk chain management and, over extended periods, accumulate to drive improved safety, efficiency, and security.

Does gravity cause load-bearing bridges in colloidal and granular systems?

We study structures which can bear loads, "bridges", in particulate packings. To investigate the relationship between bridges and gravity, we experimentally determine bridge statistics in colloidal packings. We vary the effective magnitude and direction of gravity, volume fraction, and interactions, and find that the bridge size distributions depend only on the mean number of neighbors. We identify a universal distribution, in agreement with simulation results for granulars, suggesting that applied loads merely exploit preexisting bridges, which are inherent in dense packings.

Quantitative risk assessment for hazards that arise from non-proteolytic Clostridium botulinum in minimally processed chilled dairy-based foods.

A modular process risk model has been constructed that describes the manufacture of dairy dessert products and hazards that arise from non-proteolytic Clostridium botulinum. The model describes batch manufacture and consumer storage of a family size generic dairy dessert but includes a realistic quantification that could apply to a specific food product. The dairy dessert sector is an expanding part of the UK market. The model includes modules that describe spore loads in raw materials, spore inactivation during thermal processing, volume partition and the population kinetics for non-proteolytic C. botulinum during sequential isothermal storage regimes. Where possible elements of uncertainty and variability are identified explicitly. The model is constructed as a belief network from published data and expert opinions. The model provides marginal probabilities, and associated sensitivities, for a range of endpoint measures centred on the toxicity of a single retail unit after an extended period of storage. The decimal reduction time for non-proteolytic C. botulinum spore populations at the highest (hold) temperature of the primary thermal process and the highest temperature experienced during poorly controlled (consumer) storage are dominant factors determining risks. Priorities for additional information to support risk assessments have been identified.

Can a participatory approach contribute to food chain risk analysis?

We consider food chain risks and specifically address stakeholder participation in the risk analysis process. We combine social and natural science perspectives to explore the participation process in relation to food risks and, in particular, to consider how some specific participation processes might be scientifically evaluated and how stakeholder participation in general might be incorporated into food risk decision making. We have built considerations based on three large integrative case studies that examine aspects of participatory processes. Here we use the case studies collectively to illustrate observations and beliefs concerning the nature of the interaction of stakeholders with established quantitative risk methodologies. This account is not supported by any large volume of analysis. The views in the report are expressed in relation to an accepted risk analysis framework and also with respect to probabilistic modeling of risks and are illustrated where possible with anecdotal reports of actual case study events.

Strengths and weaknesses of Monte Carlo simulation models and Bayesian belief networks in microbial risk assessment.

We discuss different aspects of farm-to-fork risk assessment from a modelling perspective. Stochastic simulation models as they are presented today represent a mathematical representation of nature. In food safety risk assessment, a common modelling approach consists of a logic chain beginning at the source of the hazard and ending with the unwanted consequences of interest. This 'farm-to-fork' approach usually begins with the hazard on the farm, sometimes with different compartments presenting different parts of the production chain, and ends with the 'dose' received by the consumer or in case a dose response model is available the number of cases of illness. These models are typically implemented as Monte Carlo simulations, which are unidirectional in nature, and the link between statistics and simulation model is not interactive. A possible solution could be the use of Bayesian belief networks (BBNs) and this paper tries to discuss in an intuitive way the possibilities of using BBNs as an alternative for Monte Carlo modelling. An inventory is made of the strengths and weaknesses of both approaches, and an example is given showing an additional use of BBNs in biotracing problems.

Estimating risk from small inocula by using population growth parameters.

Risk from an uncertain small inoculum depends on variability of single-cell lag times. However, quantifying single-cell variability is technically challenging. It is possible to estimate this variability using population growth parameters. We demonstrate this possibility using data from literature and show a Bayesian scheme for performing this task.

Estimating single-cell lag times via a Bayesian scheme.

Network models offer computationally efficient tools for estimating the variability of single-cell lag phases. Currently, optical methods for estimating the variability of single-cell lag phases use single-cell inocula and are technically challenging. A Bayesian network model incorporating small uncertain inocula addresses these limitations.

Heterogeneities in granular dynamics.

The absence of Brownian motion in granular media is a source of much complexity, including the prevalence of heterogeneity, whether static or dynamic, within a given system. Such strong heterogeneities can exist as a function of depth in a box of grains; this is the system we study here. First, we present results from three-dimensional, cooperative and stochastic Monte Carlo shaking simulations of spheres on heterogeneous density fluctuations. Next, we juxtapose these with results obtained from a theoretical model of a column of grains under gravity; frustration via competing local fields is included in our model, whereas the effect of gravity is to slow down the dynamics of successively deeper layers. The combined conclusions suggest that the dynamics of a real granular column can be divided into different phases-ballistic, logarithmic, activated, and glassy-as a function of depth. The nature of the ground states and their retrieval (under zero-temperature dynamics) is analyzed; the glassy phase shows clear evidence of its intrinsic ("crystalline") states, which lie below a band of approximately degenerate ground states. In the other three phases, by contrast, the system jams into a state chosen randomly from this upper band of metastable states.

Germination and growth from spores: variability and uncertainty in the assessment of food borne hazards.

We have developed a model for the variability of spore lag times and shown that variability has an important role in the quantitative assessment of risks associated with spore forming bacteria in food. The model includes two sequential independent delay times that contribute to the lag time for a single spore. We have shown that a population of variable spores also has a variable lag time, and we have emphasised the significance of this variability in quantitative representations of population dynamics for small populations. We have made a Bayesian estimate for the extent of the variability in spore lag times and made a comparison with direct microscopic observations of individual spores of nonproteolytic Clostridium botulinum. We conclude that Bayesian inference is a practical method for quantifying variability and hence a significant element in the development of quantitative risk assessments for hazards associated with spore forming bacteria.

Probabilistic representation of the exposure of consumers to Clostridium botulinum neurotoxin in a minimally processed potato product.

We have examined the potential of a well-specified, minimally processed potato product as a vehicle for the exposure of consumers to Clostridium botulinum neurotoxin. The product is a relatively simple combination of raw potato flakes, flour, starch and other minor ingredients and has an extended lifetime under refrigeration conditions. A combination of information and data, from a variety of sources that includes the manufacturer, has shown that the product is particularly safe with respect to non-proteolytic C. botulinum hazards. The model concentrates on a simple end point, the toxicity of an individual retail unit of the product at the point of consumer preparation, which is related to an individual risk. The probabilistic analysis was built using Bayesian Belief Network (BBN) techniques.

Physical organization of the major duplication on Brassica oleracea chromosome O6 revealed through fluorescence in situ hybridization with Arabidopsis and Brassica BAC probes.

The close relationship between Brassica oleracea and Arabidopsis thaliana has been used to explore the genetic and physical collinearity of the two species, focusing on an inverted segmental chromosome duplication within linkage group O6 of B. oleracea. Genetic evidence suggests that these segments share a common origin with a region of Arabidopsis chromosome 1. Brassica oleracea and Arabidopsis bacterial artificial chromosome probes have been used for fluorescence in situ hybridization analysis of B. oleracea pachytene chromosomes to further characterize the inverted duplication. This has been highly effective in increasing the local resolution of the cytogenetic map. We have shown that the physical order of corresponding genetic markers is highly conserved between the duplicated regions in B. oleracea and the physical lengths of the regions at pachytene are similar, while the genetic distances are considerably different. The physical marker order is also well conserved between Arabidopsis and B. oleracea, with only one short inversion identified. Furthermore, the relative physical distances between the markers in one segment of B. oleracea and Arabidopsis have stayed approximately the same. The efficacy of using fluorescence in situ hybridization, together with other forms of physical and genetic mapping, for elucidating such issues relating to synteny is discussed.

Relevance of microbial interactions to predictive microbiology.

Microbial interaction can be ignored in predictive microbiology under most conditions. We show that interactions are only important at high population densities, using published data on inhibition of growth of Listeria monocytogenes in broth. Our analysis using growth models from predictive microbiology indicated that interactions only occur at population densities of approximately 10(8) cfu/ml of the protective cultures. Spoilage is evident at these levels, except for fermented foods. In bacterial colonies, diffusion limitation acts as a constraint to growth. We have shown that these constraints only become important after large outgrowth of colonies (in the order of 5-log growth in Lactobacillus curvatus colonies), which depends on the initial inoculation density. Intra-colony interactions play an important role under these conditions. There is no large outgrowth of colonies when the initial inoculation densities are high and broth culture growth can be used to approximate colony growth.

Sequence variation in the Trichuris trichiura beta-tubulin locus: implications for the development of benzimidazole resistance.

Benzimidazole resistance has evolved in a variety of organisms and typically results from mutations in the beta-tubulin locus at specific amino acid sites. Despite widespread treatment of human intestinal nematodes with benzimidazole drugs, there have been no unambiguous reports of resistance. However, since beta-tubulin mutations conferring resistance are generally recessive, frequencies of resistance alleles less than 30% would be difficult to detect on the basis of drug treatment failures. Here we investigate sequence variation in a 1079 bp segment of the beta-tubulin locus in the human whipworm Trichuris trichiura from 72 individual nematodes from seven countries. We did not observe any alleles with amino acid mutations indicative of resistance, and of 40 point mutations there were only four non-synonymous mutations all of which were singletons. Estimated effective population sizes are an order of magnitude lower than those from another nematode species in which benzimidazole resistance has developed (Haemonchus contortus). Both the lower diversity and reduced population sizes suggest that benzimidazole resistance is likely to evolve less rapidly in Trichuris than in trichostrongyle parasites of livestock. We observed moderate levels of population subdivision (Phi(ST)=0.26) comparable with that previously observed in Ascaris lumbricoides, and identical alleles were frequently found in parasites from different continents, suggestive of recent admixture. A particularly interesting feature of the data is the high nucleotide diversities observed in nematodes from the Caribbean. This genetic complexity may be a direct result of extensive admixture and complex history of human populations in this region of the world. These data should encourage (but not make complacent) those involved in large-scale benzimidazole treatment of human intestinal nematodes.

Microsatellite DNA: a tool for population genetic analysis.

Microsatellite deoxyribonucleic acid repeats provide a source of high variability that makes them ideal for use in studies requiring such molecular markers, including large population studies and genetic typing of individuals for kinship investigations. This paper provides reviews of the use of such markers in parasitology. Most studies to date have been carried out using protozoan and vector species. Recent investigations have, however, demonstrated their usefulness in the study of helminths, illustrating their ability to distinguish between individuals within hosts as well as from different hosts. The detection of microsatellites within parasites has provided a tool that will prove invaluable in parasitology and should lead to significant advances in our understanding of the processes that affect the organisms' population genetic structure.