Limits...
Escherichia coli O157 infection on Scottish cattle farms: dynamics and control.

Zhang XS, Woolhouse ME - J R Soc Interface (2010)

Bottom Line: We first generate distributions of model parameter estimates using Markov chain Monte Carlo methods.Despite considerable uncertainty in parameter values, each set of parameter values within the 95th percentile range implies a fairly similar impact of interventions.Targeted interventions based on farm-level risk factors are more efficient than non-targeted interventions.

View Article: PubMed Central - PubMed

Affiliation: Centre for Infectious Diseases, University of Edinburgh, , Kings Buildings, West Mains Road, Edinburgh EH9 3JT, UK. xu-sheng.zhang@hpa.org.uk

ABSTRACT
In this study, we parametrize a stochastic individual-based model of the transmission dynamics of Escherichia coli O157 infection among Scottish cattle farms and use the model to predict the impacts of both targeted and non-targeted interventions. We first generate distributions of model parameter estimates using Markov chain Monte Carlo methods. Despite considerable uncertainty in parameter values, each set of parameter values within the 95th percentile range implies a fairly similar impact of interventions. Interventions that reduce the transmission coefficient and/or increase the recovery rate of infected farms (e.g. via vaccination and biosecurity) are much more effective in reducing the level of infection than reducing cattle movement rates, which improves effectiveness only when the overall control effort is small. Targeted interventions based on farm-level risk factors are more efficient than non-targeted interventions. Herd size is a major determinant of risk of infection, and our simulations confirmed that targeting interventions at farms with the largest herds is almost as effective as targeting based on overall risk. However, because of the striking characteristic that the infection force depends weakly on the number of infected farms, no interventions that are less than 100 per cent effective can eradicate E. coli O157 infection from Scottish cattle farms, implying that eliminating the disease is impractical.

Show MeSH

Related in: MedlinePlus

Correlation between model parameters generated by the MCMC method: (a) log γ and log β, (b) a and log β, (c) b and log β, and (d) b and a. The scattered points are the 3360 MCMC iterations that have negative natural logarithm of likelihood less than 215.8, the lines are the best-fit regressions and the numbers are the correlation coefficients. If all 56 094 MCMC iterations are included the correlations remain roughly the same.
© Copyright Policy - open-access
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC3104328&req=5

RSIF20100470F2: Correlation between model parameters generated by the MCMC method: (a) log γ and log β, (b) a and log β, (c) b and log β, and (d) b and a. The scattered points are the 3360 MCMC iterations that have negative natural logarithm of likelihood less than 215.8, the lines are the best-fit regressions and the numbers are the correlation coefficients. If all 56 094 MCMC iterations are included the correlations remain roughly the same.

Mentions: There are some correlations among different parameters (figure 2). For example, log β is positively correlated with the log recovery rate, log γ (figure 2a), and negatively correlated with the nonlinear coefficients a (figure 2b) and b (figure 2c). However, there is only a weak correlation between coefficients a and b (figure 2d). Therefore, different sets of parameter values can reproduce the same steady-state prevalence of about 19 per cent among the IPRAVE farms. This gives rise to a large variation in values of model parameters and therefore a fairly flat distribution of likelihood versus parameter values. At steady state with the baseline parameter values, the point prevalence for the entire system remains at about 16 per cent (which corresponds to 81% of farms being positive at some point during a given 1 year period, i.e. ‘annual prevalence’; figure 3). It is worth mentioning the difference between the prevalence on IPRAVE farms (19%) and that on the whole farms (16%) at steady state. Noting that larger herds are more likely to become infected, the difference results from the fact that the IPRAVE survey excluded smaller cattle farms [7].Figure 2.


Escherichia coli O157 infection on Scottish cattle farms: dynamics and control.

Zhang XS, Woolhouse ME - J R Soc Interface (2010)

Correlation between model parameters generated by the MCMC method: (a) log γ and log β, (b) a and log β, (c) b and log β, and (d) b and a. The scattered points are the 3360 MCMC iterations that have negative natural logarithm of likelihood less than 215.8, the lines are the best-fit regressions and the numbers are the correlation coefficients. If all 56 094 MCMC iterations are included the correlations remain roughly the same.
© Copyright Policy - open-access
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC3104328&req=5

RSIF20100470F2: Correlation between model parameters generated by the MCMC method: (a) log γ and log β, (b) a and log β, (c) b and log β, and (d) b and a. The scattered points are the 3360 MCMC iterations that have negative natural logarithm of likelihood less than 215.8, the lines are the best-fit regressions and the numbers are the correlation coefficients. If all 56 094 MCMC iterations are included the correlations remain roughly the same.
Mentions: There are some correlations among different parameters (figure 2). For example, log β is positively correlated with the log recovery rate, log γ (figure 2a), and negatively correlated with the nonlinear coefficients a (figure 2b) and b (figure 2c). However, there is only a weak correlation between coefficients a and b (figure 2d). Therefore, different sets of parameter values can reproduce the same steady-state prevalence of about 19 per cent among the IPRAVE farms. This gives rise to a large variation in values of model parameters and therefore a fairly flat distribution of likelihood versus parameter values. At steady state with the baseline parameter values, the point prevalence for the entire system remains at about 16 per cent (which corresponds to 81% of farms being positive at some point during a given 1 year period, i.e. ‘annual prevalence’; figure 3). It is worth mentioning the difference between the prevalence on IPRAVE farms (19%) and that on the whole farms (16%) at steady state. Noting that larger herds are more likely to become infected, the difference results from the fact that the IPRAVE survey excluded smaller cattle farms [7].Figure 2.

Bottom Line: We first generate distributions of model parameter estimates using Markov chain Monte Carlo methods.Despite considerable uncertainty in parameter values, each set of parameter values within the 95th percentile range implies a fairly similar impact of interventions.Targeted interventions based on farm-level risk factors are more efficient than non-targeted interventions.

View Article: PubMed Central - PubMed

Affiliation: Centre for Infectious Diseases, University of Edinburgh, , Kings Buildings, West Mains Road, Edinburgh EH9 3JT, UK. xu-sheng.zhang@hpa.org.uk

ABSTRACT
In this study, we parametrize a stochastic individual-based model of the transmission dynamics of Escherichia coli O157 infection among Scottish cattle farms and use the model to predict the impacts of both targeted and non-targeted interventions. We first generate distributions of model parameter estimates using Markov chain Monte Carlo methods. Despite considerable uncertainty in parameter values, each set of parameter values within the 95th percentile range implies a fairly similar impact of interventions. Interventions that reduce the transmission coefficient and/or increase the recovery rate of infected farms (e.g. via vaccination and biosecurity) are much more effective in reducing the level of infection than reducing cattle movement rates, which improves effectiveness only when the overall control effort is small. Targeted interventions based on farm-level risk factors are more efficient than non-targeted interventions. Herd size is a major determinant of risk of infection, and our simulations confirmed that targeting interventions at farms with the largest herds is almost as effective as targeting based on overall risk. However, because of the striking characteristic that the infection force depends weakly on the number of infected farms, no interventions that are less than 100 per cent effective can eradicate E. coli O157 infection from Scottish cattle farms, implying that eliminating the disease is impractical.

Show MeSH
Related in: MedlinePlus