Causal Analytics for Applied Risk Analysis Louis Anthony Cox, Jr

Yüklə 3,36 Mb.
ölçüsü3,36 Mb.
1   ...   26   27   28   29   30   31   32   33   ...   57

Aksoy M, Erdem S, DinCol G. Leukemia in shoe-workers exposed chronically to benzene. Blood. 1974 Dec;44(6):837–841. 

Cox LA Jr. Hormesis without cell killing. Risk Anal. 2009 Mar;29(3):393-400. doi: 10.1111/j.1539-6924.2008.01120.x.

Cox LA Jr. Universality of J-shaped and U-shaped dose-response relations as emergent properties of stochastic transition systems. Dose Response. 2006 May 1;3(3):353-68. doi: 10.2203/dose-response.0003.03.006.
GWU (2016).  George Washington University Regulatory Studies Center.  Causal Analytics Toolkit (CAT).,  The CAT software can be downloaded for free from (All links last accessed 2-16-17)
Kim S, Vermeulen R, Waidyanatha S, Johnson BA, Lan Q, Rothman N, Smith MT, Zhang L, Li G, Shen M, Yin S, Rappaport SM. Using urinary biomarkers to elucidate dose-related patterns of human benzene metabolism. Carcinogenesis. 2006a Apr;27(4):772-81.
Kim S, Vermeulen R, Waidyanatha S, Johnson BA, Lan Q, Smith MT, Zhang L, Li G, Shen M, Yin S, Rothman N, Rappaport SM. Modeling human metabolism of benzene following occupational and environmental exposures. Cancer Epidemiol Biomarkers Prev. 2006b Nov;15(11):2246-52.
Kipen HM, Cody RP, Goldstein BD. Use of longitudinal analysis of peripheral blood counts to validate historical reconstructions of benzene exposure. Environ Health Perspect. 1989 Jul;82:199-206.
Knutsen JS, Kerger BD, Finley B, Paustenbach DJ. A calibrated human PBPK model for benzene inhalation with urinary bladder and bone marrow compartments. Risk Anal. 2013 Jul;33(7):1237-51. doi: 10.1111/j.1539-6924.2012.01927.x.
Lan Q, Zhang L, Li G, Vermeulen R, Weinberg RS, Dosemeci M, Rappaport SM, Shen M, Alter BP, Wu Y, Kopp W, Waidyanatha S, Rabkin C, Guo W, Chanock S, Hayes RB, Linet M, Kim S, Yin S, Rothman N, Smith MT. Hematotoxicity in workers exposed to low levels of benzene. Science. 2004 Dec 3;306(5702):1774-6.
Liermann M, Steel A, Rosing M, Guttorp P. Random denominators and the analysis of ratio data. Environmental and Ecological Statistics. March 2004 11(1): 55–71
McHale CM, Zhang L, Lan Q, Vermeulen R, Li G, Hubbard AE, Porter KE, Thomas R, Portier CJ, Shen M, Rappaport SM, Yin S, Smith MT, Rothman N. Global gene expression profiling of a population exposed to a range of benzene levels. Environ Health Perspect. 2011 May;119(5):628-34. doi: 10.1289/ehp.1002546.
Price PS, Rey TD, Fontaine DD, Arnold SM. A reanalysis of the evidence for increased efficiency in benzene metabolism at airborne exposure levels below 3 p.p.m. Carcinogenesis. 2012 Nov;33(11):2094-9. doi: 10.1093/carcin/bgs257.
Price PS, Rey TD, Fontaine DD, Arnold SM. Letter to the editor in response to 'Low-dose metabolism of benzene in humans: science and obfuscation' Rappaport et al. (2013). Carcinogenesis. 2013 Jul;34(7):1692-6. doi: 10.1093/carcin/bgt101. Epub 2013 Mar 25.
Rappaport SM, Yeowell-O'Connell K, Smith MT, Dosemeci M, Hayes RB, Zhang L, Li G, Yin S, Rothman N. Non-linear production of benzene oxide-albumin adducts with human exposure to benzene. J Chromatogr B Analyt Technol Biomed Life Sci. 2002 Oct 5;778(1-2):367-74.
Rappaport SM, Kim S, Lan Q, Li G, Vermeulen R, Waidyanatha S, Zhang L, Yin S, Smith MT, Rothman N. Human benzene metabolism following occupational and environmental exposures. Chem Biol Interact. 2010 Mar 19;184(1-2):189-95. doi: 10.1016/j.cbi.2009.12.017.
Rappaport SM, Kim S, Thomas R, Johnson BA, Bois FY, Kupper LL. Low-dose metabolism of benzene in humans: science and obfuscation. Carcinogenesis. 2013 Jan;34(1):2-9. doi: 10.1093/carcin/bgs382. 
Rappaport SM, Kim S, Lan Q, Vermeulen R, Waidyanatha S, Zhang L, Li G, Yin S, Hayes RB, Rothman N, Smith MT. Evidence that humans metabolize benzene via two pathways. Environ Health Perspect. 2009 Jun;117(6):946-52. doi: 10.1289/ehp.0800510.
Schirrmeister A., Flora B., The Coming Wave of Benzene Litigation. Presentation at National Association of Railroad Trial Counsel Special Litigation Conference XVIII. Februaryy 7-8, 2008. Lake Tahoe, CA.
Steinmaus C, Smith AH, Jones RM, Smith MT. Meta-analysis of benzene exposure and non-Hodgkin lymphoma: biases could mask an important association. Occup Environ Med. 2008 Jun;65(6):371-8. doi: 10.1136/oem.2007.036913.
Steinmaus C, Smith AH, Smith MT. Regarding "meta-analysis and causal inference: a case study of benzene and non-Hodgkin lymphoma": an incomplete analysis. Ann Epidemiol. 2011 Jan;21(1):67-9

Thomas R, Hubbard AE, McHale CM, Zhang L, Rappaport SM, Lan Q, Rothman N, Vermeulen R, Guyton KZ, Jinot J, Sonawane BR, Smith MT. Characterization of changes in gene expression and biochemical pathways at low levels of benzene exposure. PLoS One. 2014 May 1;9(5):e91828. doi: 10.1371/journal.pone.0091828.

Vlaanderen J, Lan Q, Kromhout H, Rothman N, Vermeulen R. Occupational benzene exposure and the risk of chronic myeloid leukemia: a meta-analysis of cohort studies incorporating study quality dimensions. Am J Ind Med. 2012 Sep;55(9):779-85. doi: 10.1002/ajim.22087.
Walter RB, Appelbaum FR, Estey EH, Bernstein ID. Acute myeloid leukemia stem cells and CD33-targeted immunotherapy. Blood. 2012 Jun 28;119(26):6198-208. doi: 10.1182/blood-2011-11-325050.
Weed DL. Meta-Analysis and Causal Inference: A Case Study of Benzene and Non-Hodgkin Lymphoma . Ann Epidemiol. 2010 May;20(5):347-55. doi: 10.1016/j.annepidem.2010.02.001.
Wong O. Risk of acute myeloid leukaemia and multiple myeloma in workers exposed to benzene. Occup Environ Med. 1995 Jun;52(6):380-4.

Chapter 5

How Large are Human Health Risks Caused by Antibiotics Used in Food Animals?
Consistent with findings in risk psychology about what triggers strong emotional responses and concern (see Chapter 12), risk of food poisoning from consumption of food contaminated with disease-causing bacteria and antibiotic-resistant “superbugs” sparks strong political passions, dramatic media headlines, and heated science-policy debates (Chang et al., 2014). A widespread concern is that use of animal antibiotics on farms creates selection pressures that favor the spread of antibiotic-resistant bacteria such as methicillin-resistant Staphylococcus aureus (MRSA) (discussed in Chapter 6), multi-drug resistant (MDR) Salmonella, or E. coli (CBS, 2010). The most common effects of food-borne illness are diarrhea and possibly fever, vomiting and other symptoms of food poisoning. However, more serious harm, or death, may occur in vulnerable patients. This is especially likely if food-borne bacterial infections are resistant to usually recommended antibiotic therapies, as might happen if the infections are caused by bacteria from farms where antibiotics are used for purposes of growth promotion or disease prevention. Patients with immune systems compromised by chemotherapy, AIDS, organ transplants, or other sources can have risks hundreds or thousands of times greater than those of consumers with healthy immune systems. Fear that use of animal antibiotics on farms contributes to a rising tide of antibiotic-resistant bacterial infections has spurred many scientists, physicians, activists, journalists, and members of the public to call for elimination of the use of antibiotics as animal growth promoters.

These concerns and calls for action have usually left unquantified how many excess deaths, treatment failures, or days of illness each year are caused in the United States by antibiotic-resistant bacteria arising specifically from animal antibiotic use, as opposed to other sources. Most human cases of MRSA and other resistant infections are health care associated. They arise, for example, from inadequate hand washing and infection control in hospitals (Kallen et al., 2010). Although media reports sometimes conflate stories on food-borne resistance with statistics reflecting hospital-acquired cases (e.g., CBS, 2010), these are in fact quite distinct etiologies. They can now often be discriminated by identifying specific molecular markers for animal-associated as compared to hospital-associated strains of bacteria, allowing source-tracking based on molecular profiles of the bacteria found in infected patients. How many infections and fatalities per year arise among hospital patients, butchers, slaughterhouse workers, farmers, or the general public from livestock operations, meat handling, and consumption remains a topic of continuing interest, and such source-tracking is providing increasingly powerful molecular biological tools for obtaining answers. Responsible risk management is supported best by understanding how large the human health risks are now, and how much they would be changed by proposed interventions. The size of the risk depends on care taken to reduce microbial loads by participants throughout the food chain, including use of microbial safety controls during farming, transportation, slaughter, production and packaging, storage, retail, and food preparation and cooking.

This first part of this chapter introduces methods of quantitative risk assessment (QRA) for quantifying the number of adverse human health impacts per year caused by animal antibiotic use. Next, we summarize quantitative estimates and bounds on human health harm obtained by applying these methods to available data for several types of resistant bacteria (“drugs and bugs”) of greatest concern for public health in the United States. Finally, we discuss implications of such quantitative estimates for prudent risk management. Throughout the chapter, human health risks are expressed as expected numbers of illnesses, fatalities, illness-days, or quality-adjusted life years (QALYs) lost to illness per year (for population risks) or per capita-year (for individual risks). QRA can help to inform and improve risk management decisions and policies by predicting how changes in the food production process, such as greater or lesser use of antibiotics on the farm, will affect human health risks, including individual and population risks for such subpopulations as well as for the whole population of concern.
This section reviews methods of quantitative microbial risk assessment (QMRA) and antimicrobial risk assessment. It expands upon and updates the brief summary in Cox (2008).
Farm-To-Fork Risk Simulation Models
When enough data and understanding are available, the effects of alternative risk management actions on risks created by bacteria in food – both antibiotic-resistant and antobiotoc-susceptible strains – can be quantified by simulating microbial loads of bacteria along the chain of steps leading from production to consumption for each intervention. If the conditional frequency distribution of microbial loads leaving each step (e.g., slaughter, transportation, processing, storage, etc.) can be estimated, given the microbial load entering that step, and given the controls applied (e.g., use of antibiotic sprays, refrigeration, etc.), then the effects on microbial loads of alternative risk management policies can be quantified and compared. Microbial loads are typically expressed in units such as colony-forming-units (CFUs) of bacteria per unit (e.g., per pound, per carcass, etc.) of food. If, in addition, dose-response relations are available to convert microbial loads in ingested foods to corresponding probabilities of illnesses, together with measures of illness severity (e.g., illness-days, QALYs lost, fatalities, etc.), then the effects of alternative risk management policies on human health can also be estimated and compared. As an example, Figure1 shows how the frequency distribution of illnesses per year caused by Vibrio parahaemolyticus in oysters are predicted to change if refrigeration time requirements that accomplish different degrees of reduction in microbial loads are implemented. The underlying quantitative microbial risk assessment (QMRA) model simulates the changes in microbial loads at successive stages from harvesting to consumption; its main logical structure and data inputs are shown in Figure 2.

Figure 1: Frequency distributions of number of Vibrio parahaemolyticus (Vp) illnesses per year from oysters with and without mitigation from cooling requirements.

The discipline of applied microbiology supplies empirical growth curves and kill curves for log increase or log reduction, respectively, in microbial load from input to output of a step. These curves describe the output:input ratio (e.g., a most likely value and upper and lower statistical confidence limits) for the microbial load passing through a stage as a function of variables such as temperature, pH, and time.
figure v-1. schematic representation of the vibrio parahaemolyticus risk assessment model

Figure 2: Structure of the quantitative microbial risk assessment (QMRA) model that allows quantitative risk estimates such as those in Figure to be made.

A “farm-to-fork” simulation model can be constructed by concatenating many consecutive steps representing stages in the food production process. Each step receives a microbial load from its predecessor. It produces as output a microbial load value sampled from the conditional frequency distribution of the output microbial load, given the input microbial load, as specified by the microbial growth model describing that stage. Measured frequency distributions of microbial loads on animals (or other units of food) leaving the farm provide the initial input to the whole model. The key output from the model is a frequency distribution of the microbial load, x, of pathogenic bacteria in servings of food ingested by consumers.

Risk-reducing factors such as antimicrobial sprays and chilling during processing, freezing or refrigeration during storage, and cooking before serving are often modeled by corresponding reduction factors for microbial loads. (These may be represented as random variables, e.g., with log-normal distributions and geometric means and variances estimated from data.) The complete model is then represented by a product of factors that increase or decrease microbial loads, applied to the empirical frequency distribution of initial microbial loads on which the factors act. Running the complete farm-to-fork model multiple times produces a final distribution of microbial loads on servings eaten by consumers. Some farm-to-fork exposure models also consider effects of cross-contamination in the kitchen, if pathogenic bacteria are expected to be spread to other foods by poor kitchen hygiene practices (e.g., failure to wash a cutting board after use.) As an example of the output from such a model, Figure 3 shows an example of the distribution of microbial loads of Salmonella in servings of chicken well.

Figure 3: Average dose (CFU Salmonella) per serving in meals prepared from contaminated broilers. Source:

In summary, farm-to-fork simulation models can estimate the frequency distributions of microbial loads ingested by consumers in servings of food. As already illustrated in Figure 1, QMRA can also estimate of how these frequency distributions would change if different interventions (represented by changes in one or more of the step-specific factors increasing or decreasing microbial load) were implemented. For example, enforcing a limit on the maximum time that ready-to-eat meats may be stored at delis or points of retail sale before being disposed of limits the opportunity for bacterial growth prior to consumption. Changing processing steps (such as scalding, chilling, antimicrobial sprays, etc.) can also reduce microbial loads. Such interventions shift the cumulative frequency distribution of microbial loads in food leftward, other things being held equal. If some fraction of the microbial load at each stage can be identified as resistant to antibiotics used to treat food-borne illnesses caused by consuming contaminated meat or other (possibly cross-contaminated) food, then the QMRA models can also be used to predict exposures to resistant bacteria in food.
Dose-Response Models for Food-borne Pathogens
Once a serving of food (e.g., chicken, oysters, hamburger, deli meats, etc.) reaches consumer, the probability that an ingested dose will cause infection and illness is described by dose-response models. Several parametric statistical models have been developed to describe the relation between quantity of bacteria ingested in food and resulting probability of illness. One of the simplest is the following exponential dose-response relation:
r(x) = Pr(illness | ingest microbial load = x CFUs) = 1 – e-x.
This model gives the probability that an ingested dose of x colony forming units (CFUs) of a pathogenic bacterium will cause illness. r(x) denotes this probability. The function r(x) is a dose-response curve. is a parameter reflecting the potency of the exposure in causing illness. Sensitive subpopulations have higher values of than the general population.

More complex dose-response models (especially, the widely used Beta-Poisson model) have two or more parameters, e.g., representing the population distribution of individual susceptibility parameter values and the conditional probability of illness given a susceptibility parameter. The standard statistical tasks of estimating model parameters, quantifying confidence intervals or joint confidence regions, and validating fitted models can be accomplished using standard statistical methods such as maximum likelihood estimation (MLE) and resampling methods. The excellent monograph by Haas et al. (1999) provides details and examples. It notes that “It has been possible to evaluate and compile a comprehensive database on microbial dose-response models.” Chapter 9 of this monograph provides a compendium of dose-response data and dose-response curves, along with critical evaluations and results of validation studies, for the following: Campylobacter jejuni (based on human feeding study data), Cryptosporidium parvum, pathogenic E. coli, E. coli O157:H7 (using Shigella species as a surrogate), Giardia lamblia, non-typhoid Salmonella(based on human feeding study data), Salmonellatyphosa, Shigella dystenteriae, S. flexneri,Vibrio cholerae, Adenovirus 4, Coxsackie viruses, Echovirus 12, Hepatitis A virus, Poliovirus I (minor), and rotavirus. Thus, for many food-borne and water-borne pathogens of interest, dose-response models and assessments of fit are readily available.

Despite this base of relatively well-developed and validated dose-response models, however, two important challenges remain in developing dose-response models for specific strains of pathogenic bacteria, including antibiotic-resistant strains. Figure 4 illustrates the first, and Figure 5 the second. In Figure 4, the best-fitting model in a specific class of parametric models (the “naïve” Beta-Poisson model, which provides a widely-used approximate mathematical model for response probabilities for different doses (CFUs) ingested) provides a clearly biased description of the observed feeding trial data, under-estimating all observed response probabilities for Log Dose < 5. Fgure 5 illustrates the problem of low-dose extrapolation, in which the dose-response relation at doses far below the range of observed data depends greatly on which specific model is assumed.

Figure 4. Even the best-fitting dose-response model in a specified parametric family, such as the Beta-Poisson (BP) family, may provide a biased description of data. Here, a best-fiting dose-response model for Salmonella data systematically under-estimates risk at low doses. Source:

Figure 5. Multiple dose-response models that fit the available experimental data equally well may make very different predictions for risks outside the range of observed data. Source:

Because of these challenges, dose-response models may be highly uncertain for specific strains of pathogens, and hence risk projections based on them may also be very uncertain. Characterizing this uncertainty is a key step in QMRA that uses dose-response models.
Quantitative Risk Characterization for QMRA and Risk Management
Sampling values of exposures x from the frequency distribution predicted by a farm-to-fork model (expressed in units of bacteria-per-serving) and then applying the dose-response relation r(x) to each sampled value of x produces a frequency distribution of the risk-per-serving in an exposed population. This information can be displayed in various ways to inform risk management decision-making.

For example, Figure 6 shows how the (base 10 logarithm of) risk-per-serving of chicken for salmonellosis is reduced by a mitigation strategy that encourages consumers to cook chicken properly before eating it, based on the exposure sub-model in Figure 3, the Beta-Poisson dose-response model in Figure 5, and assumptions about how mitigation measures will affect the distribution of cooking practices.

Figure 6. A display showing how salmonellosis risk-per serving of chicken is reduced by better cooking practices. Source:

Other displays showing the expected number of illnesses per year in a population, expected illnesses per capita-year in the overall population and for members of sensitive subpopulations, and frequency distributions or upper and lower confidence limits around these expected values are typical outputs of risk characterization. If particular decisions are being considered, such as a new standard for the maximum times and/or temperatures at which ready-to-eat meats can be stored before being disposed of, then plotting expected illnesses per year against the decision variables (i.e., maximum times or temperatures, in this example) provide the quantitative links between alternative decisions and their probable health consequences needed to guide effective risk management decision-making. Figure 7 illustrates the key concept of informing risk management decisions by showing how a measure of risk (here, expected deaths per year) varies with decisions about maximum allowed storage times and temperatures. Such displays, linking actions to their probable consequences, provide the essential information needed to inform risk management decisions.

figure vi-5: graph showing mortality rate for elderly vs. storage temperature for 5 storage times, mortality increasing with storage temperature and time.

Figure 7. Expected elderly mortalities per year from Listeria monocytogenes if different maximum storage times and temperatures are allowed.

In recent years, many efforts have been made to simplify the standard approach to quantitative microbial risk assessment just summarized, especially for application to antimicrobial-resistant bacteria. Because farm-to-fork exposure modeling and valid dose-response modeling can require data that are expensive and time consuming to collect, or that are simply not available when risk management decisions must be made, simpler approaches with less burdensome data requirements are desirable. It is tempting to use simple multiplicative models, such as the following:
Risk = Exposure × Dose-Response Factor × Consequence per case
where: Risk = expected number of excess illness-days per year, Exposure is measured in potentially infectious meals ingested per year in a population, Dose-Response Factor = expected number of illnesses caused per potentially infectious meal ingested, and Consequence per case is measured in illness-days (or fatalities) caused per illness.

While such models have attractive simplicity, they embody strong assumptions that are not necessarily valid, and thus can produce highly misleading results. Specifically, the assessment of Dose-Response Factor requires attributing some part of the causation of illness-days to Exposure. Similarly, estimating the change in Dose-Response Factor due to an intervention that changes microbial load may require guess-work. There is often no valid, objective way to make such attributions based on available data. The risk assessment model – and, specifically, the attribution of risk to particular food sources – may then become a matter of political and legal controversy.

For example, suppose that the Dose-Response Factoris estimated by dividing the observed value of Risk in a population in one or more years by the contemporaneous values of (Exposure × Consequence). Then, this value will always be non-negative (since its numerator and denominator are both non-negative). The model this implies a non-negative linear relation between Exposure and Risk, even if there is no causal relation at all (or is a negative one) between them. (By analogy, one could divide the number of car accidents in Florida in a year by the number of oranges consumed in Florida that year, but the resulting “car accidents per orange consumed” ratio, although certainly positive, would not in any way imply a causal relation, or that reducing consumption of oranges would reduce car accidents per capita-year. Replacing car accidents with food-borne illnesses such as ciprofloxacin-resistant campylobacteriosis and oranges with chicken servings improves the intuitive plausibility but not the logic or credibility of such calculations.) In addition, it is a frequent observation that some level of exposure to bacteria in food protects against risk of food-borne illnesses, for example, by stimulating acquired immunity. Thus, the use of simple multiplicative model implying a necessarily non-negative linear relation between Exposure and Risk may be incorrect, producing meaningless results (or, more optimistically, extreme upper bounds on estimated risks) if the true relation is negative or non-linear.

Unfortunately, past estimates of risk of antibiotic-resistant illnesses caused by consumption of foods contaminated with resistant bacteria attributed to farm use of antibiotics have often simply assumed that some fraction of total resistant infections is caused by farm use of antibiotics (e.g., Barza and Travers, 2002) or that the ratio of estimated excess resistant cases per year to servings of food per year could be interpreted causally (on a logical par with the car-accidents-per-orange-consumed example above). For example, Chang et al (2014) describe a case in which the United States Food and Drug Administration (FDA) used an assumption that excess cases of fluoroquinolone (FQ)-resistant campylobacteriosis were proportional to consumption of chicken exposed to FQ on the farm to estimate that between 4,960 and 14,370 patients per year could have compromised treatment with ciprofloxacin (Bartholomew et al., 2003, cited in Chang et al., 2014). This model was used to support a risk management decision to withdraw fluoroquinolone use in poultry in the United States. But the subsequent decade of experience showed that the withdrawal had no detectable causal impact in reducing levels of ciprofloxacin resistance in the United States (Chang et al., 2014). As in the car crashes per orange analogy, FDA had interpreted a positive ratio as causal, and discovered only after the fact that reducing the denominator had no real-world effect on reducing the numerator.

Thus, great caution should be taken when using such simplified risk assessment models. In general, they may be useful in making rapid calculations of plausible upper bounds in certain situations (for example, if the true but unknown dose-response relation between exposure and risk is convex, or upward-curving), but should not be expected to produce accurate risk estimates unless they have been carefully validated (Cox, 2006).
An alternative method is available for quantifying upper bounds on the adverse human health consequences per year that could be prevented by reducing or eliminating antibiotic uses in agriculture. This method, which has advantages of simplicity, logical soundness, and reliance only on readily available data, does not attempt to simulate in detail the microbial loads traversing different pathways (e.g., food water, environment, co-selection in other bacteria, etc.) or to quantify the relevant dose-response relations. Instead, it begins with the total number of adverse events per year that might be caused by antibiotic use (e.g., total treatment failures caused by antibiotic resistance), and then uses molecular biological data to estimate upper bounds on the fraction of all such cases that might be caused by (and preventable by eliminating) animal antibiotic uses. We call this the empirical upper-bounding approach. We will illustrate it in some detail for ampicillin-resistant E. faecium (AREF) bacteria, and then summarize results from applying the approach to other drug-bug pairs.
Case study: Ampicillin-resistant E. faecium (AREF) bacteria
This section, adapted from Cox et al. (2009), illustrates the empirical upper-bounding approach for potential risks to human health from use of penicillin drugs in agriculture. It illustrates how to do quantitative risk assessment when neither all pathways from farm to consumer (or patient) nor relevant dose-response relations are known with enough confidence to permit useful simulation of microbial loads and illnesses.

Penicillin-based drugs are approved for use in food animals in the United States to treat, control, and prevent diseases and, to a lesser extent, to improve growth rates (FDA-CVM, 2007, Sechen, 2006, AHI, 2006). Concerns that penicillin use might increase the risk of antibiotic resistance in human enterococcal infections from non-human sources, thus leading to increased morbidity and mortality (WHO, 2005), have made approved feed usages of penicillins in food animals a controversial topic for several decades in the United States (IOM, 1989, FDA, 2000, FDA, 2003). The following sections develop a plausible upper bound on the potential for continued use of penicillin drugs in food animals to harm human health by increasing the number of antibiotic-resistant enterococcal infections in human patients. After summarizing relevant background for the hazard of greatest concern – infection of intensive care unit (ICU) patients with ampicillin-resistant E. faecium (AREF) bacteria – the following sections focus on quantifying the fraction of such resistant infections that might be prevented by discontinuing the use of penicillin drugs in food animals.

Risk to human health arises because some strains of enterococci may become opportunistic pathogens, potentially resistant to multiple drugs, that infect patients who are already seriously ill (typically in ICUs) with immune systems weakened by organ transplants, chemotherapy, AIDS, or other causes. Indeed, enterococcal infection is the second most common hospital-acquired infection in the United States (Varman et al., 2006). These infections can prolong illness and increase patient mortality. Vancomycin-resistant enterococci (VRE) are of particular concern because of their virulence and resistance to even some recently developed antibiotics. Vancomycin-resistant E. faecium (VREF) can cause serious and often fatal disease in vulnerable populations, such as liver transplant patients and patients with hematologic malignancies (Rice, 2001).

Although many enterococcal infections, including VRE, resolve without antimicrobial treatment (Varman et al., 2006; Rice, 2001), in severe cases for which antimicrobial treatment is provided, penicillin and ampicillin are often the leading choices. Most E. faecium infections in ICU patients in the U.S. are now resistant to vancomycin (Edmond et al., 1999; Jones et al., 2004). Patients with vancomycin resistant E. faecium (VREF) have worse outcomes than those with vancomycin susceptible strains – longer hospital stays and higher mortality (Webb et al., 2001). As noted by Rice (2001), virtually all VREF are also ampicillin resistant: “More than 95% of VRE recovered in the United States are E. faecium; virtually all are resistant to high levels of ampicillin.” Hence, our risk assessment treats VREF as being (at least approximately) a subset of ampicillin-resistant E. faecium (AREF). Since most VREF are AREF (although many AREF are not VREF), and assuming that changes in animal penicillin use would not significantly affect vancomycin resistance (consistent with historical data), we focus on human (ICU patient) infections with vancomycin-susceptible strains of ampicillin-resistant E. faecium. Presumably, this is the subpopulation that might experience decreased ampicillin resistance if discontinuing animal penicillin drugs were to replace some AREF cases with ampicillin-susceptible cases. For patients with VREF, we assume that AREF would persist (due to the observed co-occurrence of AREF in VREF strains), so that no benefit from reduced AREF would be achieved for these patients.

Recognizing that a farm-to-fork (model is not practical for AREF, due to data and knowledge gaps in release, exposure, and dose-response relations, we instead start with more readily available human data on ICU case loads and resistance rates, similar to the approach in Cox & Popken, 2004. We then work backward to estimate a plausible upper bound on the annual number of human patient mortalities that might be prevented by discontinuing penicillin use in food animals.

For purposes of conservative (i.e., upper-bound) risk assessment, we define a potentially preventable mortality to occur whenever the following conditions hold: (1) An ICU patient dies, following (2) an E. faecium infection that (3) is resistant to ampicillin (AREF) (and hence might have benefited had ampicillin resistance been prevented). The infection was: (4) Vancomycin-susceptible (and hence might have also been ampicillin-susceptible, had it not been for penicillin use in food animals); (5) not known to have been contracted from the hospital environment (and hence might have been prevented by actions external to the hospital, such as elimination of AREFs from food animals); (6) could have come from food animals (i.e., has a genotype or resistance determinants of the types found in food animals). (7) The patient tolerated penicillin (i.e. was not allergic, and hence might have benefited from ampicillin, had it not been for resistance). We propose that the conjunction of these seven conditions should be interpreted as necessary for a mortality to have been caused (with non-negligible probability) by resistance due to use of penicillin in food animals, even though it is not sufficient (e.g., the infecting strain might have had some other origin than food animals, or the patient might have died anyway, even if the infection had been ampicillin-susceptible). Accordingly, the following sections estimate a plausible upper bound on annual preventable mortalities from AREF infections based on the following product of factors:

Preventable AREF mortalities per year (Total number of ICU infections per year) (fraction caused by E. faecium) (fraction of ICU E. faecium infections that are AREF and exogenous, i.e., not known to be of nosocomial origin) (fraction of these exogenous AREF cases that are vancomycin-susceptible) (fraction of vancomycin-susceptible exogenous AREF cases that might have come from food animals) (fraction of these cases that are penicillin-tolerant) (excess mortality rate for AREF cases compared to ASEF cases).
That is, we first quantify the expected annual number of AREF cases in the U.S. that might benefit from ampicillin treatment if food animal uses of penicillin were halted (i.e., cases that are penicillin-tolerant and vancomycin-susceptible and that might have been caused by resistance determinants from food animals). Then, we multiply this number by the excess mortality rate for resistant as opposed to susceptible cases. Each of the foregoing factors can be estimated from available data, as discussed in detail in Cox et al. (2009) and summarized in Table 1.

Table 1 shows key parameter estimates, calculations, assumptions, and resulting risk estimates. When presenting point estimates, it is customary to also present interval estimates to inform decision-makers about the plausible range of estimated values. In this analysis, however, the key uncertainties have little to do with statistical sampling error, and they are not adequately characterized by confidence limits. Rather, they arise from uncertainty about the validity and conservatism of the assumptions in Table 1. Qualitatively, the main uncertainty is about whether a non-zero risk to human health exists from animal use of penicillin drugs. We have assumed that there is, but there is no clear empirical proof that the risk is non-zero. To bridge this knowledge gap, Table 1 incorporates several conservative qualitative assumptions that jointly imply that the risk is non-zero. Other quantitative parameter values presented, and their implied risk estimate of  0.135 excess mortalities/year, are intended to be realistic, data-driven values (rather than extreme upper-bounds or 95% upper confidence limits) contingent on these conservative qualitative assumptions. The most important conservative elements in Table 1 are the following qualitative assumptions:

  • Transfer of ampicillin resistance from food animal bacteria to bacteria infecting human patient occurs. The assumption that ampicillin-resistant strains and/or determinants are transferred from strains in food animals to human ICU patients is fundamental to the assessment in Table 1. Such transfer has never been shown to occur, but may be possible.

  • Withdrawing animal drug use would immediately and completely prevent the problem. Table 1 assumes that halting penicillin use in food animals would immediately eliminate all ampicillin resistance from the cases in Table 1. This is a deliberately extreme assumption. In reality, halting use might have little or no impact on the already very low levels of ampicillin resistance.

  • Resistance increases patient mortality. The assumption that ampicillin resistance causes an increase in the mortality rates of the patients in Table 1 is made even though, in reality, no statistically significant difference in mortality rates has been found between resistant and non-resistant cases (Fortun et al., 2002).

Table 1 Summary of AREF Risk Calculation Using Empirical Upper Bounding


More Conservative Value

Less Conservative Value


N = ICU infections/year

N = 315,000

N = 104,372.5

FDA-CVM 2004

Pent = fraction of ICU infections caused by Enterococcus



(Wisplinghoff et al., 2004)

FDA-CVM 2004

PEF,ent = fraction of enterococcal infections caused by E. faecium.


FDA-CVM 2004

fraction of enterococcal infections caused by E. faecium that are exogenous (non-nosocomial)

 0.17

Cox and Popken 2004. (May be smaller now due to spread of CC-17)

fraction of exogenous cases that are ampicillin-resistant


Willems et al., 2005

fraction of exogenous ampicillin-resistant cases that are vancomycin-susceptible


Jones et al., 2004

fraction of exogenous ampicillin-resistant vancomycin-susceptible cases possibly from food animals


(0.069 assumed)

Data of Leavis et al., 2006

fraction of exogenous ampicillin-resistant cases with penicillin-tolerant host


Lee et al., 2000

fraction of these cases that would become ampicillin-susceptible if penicillin use in food animals were terminated


(1 is assumed.)

Conservative assumption

Increase in mortality risk per case, due to ampicillin resistance


(0.06 is assumed.)

Fortun et al., 2002, conservative assumption

RISK =  0.135 potential excess mortalities/year

315000*0.10*0.25*0.17*0.187*0.14*0.069*0.844*0.06 = 0.135

104372.5*0.09*0.25*0.17*0.187*0.155*0.069*0.844*0.06 0.04 mortalities/year

Product of preceding factors

Source: Cox, Popken and Mathers, 2009
With these assumptions, the calculations in Table 1 predict that excess mortalities per year in the entire United States population could be as high as 0.135, or about one excess mortality expected once every seven to eight years on average, if current conditions persist. This risk is concentrated among ICU patients already at high risk of such infections. With less conservative assumptions, the estimated risk falls to about 0.04 excess mortalities per year, i.e., about one excess mortality every 25 years in the United States under current conditions. The multiplicative calculation in Table 1 makes sensitivity analysis of these results to changes in the values of specific factors especially straightforward: the final risk estimate is directly proportional to each factor listed.

The more conservative risk estimate of 0.135 excess mortalities per year equates to an average individual risk rate in the most at-risk group (ICU patients) of approximately 0.135/315,000 = 4.3  10-7 excess mortalities per ICU patient. For the United States population as a whole, this corresponds to an average individual risk of approximately 0.135/300E6 = 4.5  10-10 excess fatalities per person-year, or a lifetime risk of about 80  (6 x 10-10) = 3.6  10-8 excess risk of mortality per lifetime (for an assumed 80-year lifetime). If the less conservative risk estimate of 0.04 excess mortalities per year is used, these individual and population risks are reduced by a factor of 0.04/0.135, or more than three-fold. If one or more of the key qualitative assumptions listed above are violated, then the true risk could be as low as zero.

The main conclusion from these calculations is that not more than 0.04 excess mortalities per year (under conservative assumptions) to 0.14 excess mortalities per year (under very conservative assumptions) might be prevented in the whole U.S. population if current use of penicillin drugs in food animals were discontinued, and if this successfully reduced the prevalence of antibiotic-resistant E. faecium infections among intensive care unit (ICU) patients. The true risk could well be zero, if one or more of the conservative assumptions above is false.
Summary of Results from Applying Empirical Upper-Bounding Risk Assessment to Other Antibiotic-Resistant Bacteria
Antimicrobial risk analyses have now been completed for several antimicrobial-resistant bacteria of public health concern using empirical upper-bounding approaches. Among the results now available are the following.

  • For streptogramins, banning virginiamycin has been estimated to prevent from 0 to less than 0.06 statistical mortalities per year in the entire United States population (Cox and Popken 2004; see also FDA-CVM, 2004). More data tend to reduce such upper bounds, which in part reflect uncertainties n the data available at the time of the study.

  • For macrolide-resistant campylobacter, Hurd and Malladi (2008) concluded that “the predicted risk of suboptimal human treatment of infection with C. coli from swine is only 1 in 82 million; with a 95% chance it could be as high as 1 in 49 million. Risks from C. jejuni in poultry or beef are even less.” (This analysis followed the FDA approach of interpreting simple ratios as if they applied that reducing exposures in the denominator would proportionally reduce cases in the numerator. Thus, the results may have no predictive validity if this assumption turns out to be incorrect, similar to the case of fluoroquinolones discussed by Chang et al. (2014).)

  • For tetracyclines, Cox and Popken (2010) concluded that “As a case study, examining specific tetracycline uses and resistance patterns suggests that there is no significant human health hazard from continued use of tetracycline in food animals. Simple hypothetical calculations suggest an unobservably small risk (between 0 and 1.75E-11 excess lifetime risk of a tetracycline-resistant infection), based on the long history of tetracycline use in the United States without resistance-related treatment failures.”

  • For MRSA, Cox and Popken (2014) “construct a conservative (plausible upper bound) probability estimate for the actual human health harm (MRSA infections and fatalities) arising from [livestock-associated] ST398-MRSA from pigs. The model provides plausible upper bounds of approximately one excess human infection per year among all U.S. pig farm workers, and one human infection per 31 years among the remaining total population of the United States. These results assume the possibility of transmission events not yet observed, so additional data collection may reduce these estimates further.”

Such quantitative risk estimates suggest that banning agricultural uses of these antibiotics might create small human health benefits (perhaps reducing compromised treatments due to resistance by a few cases per century), but are unlikely to make any measurable difference in improving public health. This finding disagrees with the passionate convictions of many experts who advocate prompt bans as urgently needed to slow the spread of resistance (Chang et al., 2014).

Since the empirical upper-bounding approach was originally developed in the early 2000s with support from the animal antibiotic industry, results such as those just cited are sometimes viewed with suspicion (ibid.) A virtue of quantitative risk assessment in helping to inform (and perhaps occasionally resolve) politically charged debates over what to do is that the logic, data sources, and calculations used are completely transparent and easy to summarize, as in Table 1, so that anyone interested can check the logic and conclusions and experiment with varying the assumptions. However, even if quantitative risk assessment proves to be too controversial to support trusted conclusions, it is often still possible to manage risks pragmatically using principles discussed next.


Even without QRA, it is often possible to apply process quality improvement ideas to control the microbial quality of food production processes – including both susceptible and resistant bacteria. This approach has been developed and deployed successfully (usually on a voluntary basis) using the Hazard Analysis and Critical Control Points (HACCP) approach summarized in Table 2. The main idea of HACCP is to first identify steps or stages in the food production process where bacteria can be controlled, and then to apply effective controls at those points, regardless of what the ultimate quantitative effects on human health risks may be. Reducing microbial load at points where it can be done effectively has proved very successful in reducing final microbial loads and improving food safety.
This chapter has introduced and illustrated key ideas used to quantify and manage human health risks from food contaminated by bacteria, both antibiotic-resistant and antibiotic-susceptible. Somewhat similar ideas apply to other food-borne hazards, from pesticide residues to mad cow disease, i.e., risk assessment can be carried out by modeling the flow of contaminants through the food production process (together with any increases or decreases at different steps), resulting in levels of exposures in ingested foods or drinks. These are then converted to quantitative risks using dose-response functions.
Table 2: Summary of Seven HACCP Principles

  • Analyze hazards. Potential hazards associated with a food and measures to control those hazards are identified. The hazard could be biological, such as a microbe; chemical, such as a toxin; or physical, such as ground glass or metal fragments.

  • Identify critical control points. These are points in a food's production – from its raw state through processing and shipping to consumption by the consumer – at which the potential hazard can be controlled or eliminated. Examples are cooking, cooling, packaging, and metal detection.

  • Establish preventive measures with critical limits for each control point. For a cooked food, … this might include… minimum cooking temperature and time required to ensure the elimination of any harmful microbes.

  • Establish procedures to monitor the critical control points. Such procedures might include determining how and by whom cooking time and temperature should be monitored.

  • Establish corrective actions to be taken when monitoring shows that a critical limit has not been met for example, reprocessing or disposing of food if the minimum cooking temperature is not met.

  • Establish procedures to verify that the system is working properly – for example, testing time-and-temperature recording devices to verify that a cooking unit is working properly.

  • Establish effective recordkeeping to document the HACCP system. This would include records of hazards and their control methods, the monitoring of safety requirements and action taken to correct potential problems. Each of these principles must be backed by sound scientific knowledge: for example, published microbiological studies on time and temperature factors for controlling food-borne pathogens.

Source: USDA/FDA, 2004;
The practical successes of the HACCP approach provide a valuable reminder that quantitative risk assessment (QRA) is not always a prerequisite for effective risk management. It may not be necessary to quantify a risk in order to reduce it. Reducing exposures at critical control points throughout the food production process can reduce exposure-related risk even if the size of the risk is unknown.

Where QRA can make a crucial contribution is in situations where there is doubt about whether an intervention is worthwhile. For example, QRA can reveal whether expensive risk-reducing measures are likely to produce correspondingly large health benefits. It may be a poor use of resources to implement expensive risk-reducing measures if the quantitative size of risk reduction procured thereby is very small. QRA methods such as farm-to-fork exposure modeling and dose-response modeling (Haas et al., 1996), or empirical upper-bounding approaches based on multiplicative models (Cox, 2006), can then be valuable in guiding effective risk management resource allocations by revealing the approximate sizes of the changes in human health risks caused by alternative interventions. A detailed example is given in the next chapter.

Animal Health Institute (AHI). (2006). Animal Health Companies Meet Increase Market Need for Antibiotics. News Release from the Animal Health Institute, Washington, D.C, Oct. 12, 2006. Available at
American Veterinary Medical Association (AVMA). (2008). American Veterinary Medical Association Judicious Therapeutic Use of Antimicrobials. December, 2005. Available at
Barza M, Travers K. (2002). Excess infections due to antimicrobial resistance: the "Attributable Fraction".Clin Infect Dis., 34 Suppl 3:S126-30.
Biavasco, F., Foglia, G., Paoletti, C., Zandri, G., Magi, G., Guaglianone, E., Sundsfjord, A., Pruzzo, C., Donelli, F., & Facinelli, B. (2007). VanA-type Enterococci from humans, animals and food: species distribution, population structure, Tn1546-typing and location, and virulence determinants. Applied and Environmental Microbiology, 73(10), 3307-19.
CBS. (2010, 6/16/2010). "Animal Antibiotic Overuse Hurting Humans? Katie Couric Investigates Feeding Healthy Farm Animals Antibiotics. Is it Creating New Drug-Resistant Bacteria? ." CBS Special News Report: Katie Couric Investigates, from
Chang Q, WangW, Regev-Yochay G, Lipsitch M, and Hanage WP (2014). Antibiotics in agriculture and the risk to human health: how worried should we be? Evolutionary Applications.
Cox LA Jr. Managing Food-borne Risks. Wiley Encyclopedia of Quantitative Risk Analysis and Assessment, 2008.
Cox, L.A., Jr. Quantitative Health Risk Analysis Methods: Modeling the Human Health Impacts of Antibiotics Used in Food Animals. Springer. New York. 2006.
Cox LA Jr., Popken DA, Mathers J. Human health risk assessment of penicillin / aminopenicillin resistance in enterococci due to penicillin use in food animals. Risk Analysis.2009 Jun;29(6):796-805
Cox, L.A. & Popken, D.A. (2004). Quantifying human health risks from virginiamycin used in chickens. Risk Analysis,24(1), 271-288.
Deshpande, L.M., Fritsche, T.R., Moet, G.J., Biedenbach, D.J. &Jones, R.N.(2007). Antimicrobial resistance and molecular epidemiology of vancomycin-resistant enterococci from North America and Europe: a report from the SENTRY antimicrobial surveillance program. Diagnostic Microbiology and Infectious Disease,58(2), 163-70.
Edmond, M.B., Wallace, S.E., McClish, D.K., Pfaller, M.A., Jones, R.N., & Wenzel, R.P.(1999). Nosocomial bloodstream infections in United States hospitals: a three-year analysis. Clinical Infectious Diseases, 29(2), 239-44.
FAO/WHO/OIE (2008). Joint FAO/WHO/OIE Expert Meeting on Critically Important

Antimicrobials. Report of a meeting held in FAO, Rome, Italy, 26–30 November 2007. FAO, Rome,

Italy, and WHO, Geneva, Switzerland.
US Food and Drug Administration – Center for Veterinary Medicine (FDA-CVM). (2007). FDA Database of Approved Animal Drug Products. FDA Center for Veterinary Medicine, VMRCVM Drug Information Lab. Available at
US Food and Drug Administration – Center for Veterinary Medicine (FDA-CVM). (2004). Risk Assessment of Streptogramin Resistance in EnterococcusFaecium Attributable to the Use of Streptogramins in Animals. Draft for Comment, Nov 23, 2004. Available at
US Food and Drug Administration – Center for Veterinary Medicine (FDA-CVM). (2003). Guidance for Industry # 152: Evaluating the Safety of Antimicrobial New Animal Drugs with Regard to their Microbiological Effects on Bacteria of Human Health Concern. US Dept of Health and Human Services, Food and Drug Administration, Center for Veterinary Medicine. Available at
US Food and Drug Administration – Center for Veterinary Medicine (FDA-CVM). (2000). Review of Agricultural Antibiotics Policies. Available at
Fortun, J., Coque, T.M., Martin-Davila, P., Moreno, L., Canton, R., Loza, E., Baquero, F. & Moreno, S. (2002). Risk factors associated with ampicillin resistance in patients with bacteraemia caused by Enterococcusfaecium. Journal of Antimicrobial Chemotherapy,50(6), 1003-9.
Guardabassi, L., &Dalsgaard, A.(2004). Occurrence, structure, and mobility of Tn1546-like elements in environmental isolates of vancomycin-resistant enterococci. Applied and Environmental Microbiology,70(2),984-90.
Haas C.N., J.B. Rose, and C.P. Gerba. 1999. Quantitative Microbial Risk Assessment, John Wiley & Sons, New York.
Hanrahan, J., Hoyen, C., &Rice, L.B.(2000). Geographic distribution of a large mobile element that transfers ampicillin and vancomycin resistance between Enterococcus faecium strains. Antimicrobial Agents and Chemotherapy,44, 1349–51.
Hurd HS, Malladi S. (2008). A stochastic assessment of the public health risks of the use of macrolide antibiotics in food animals.Risk Analysis, 28(3), 695-710.
Institute of Medicine (IOM). (1989). Human Health Risks with the Subtherapeutic use of Penicillin or Tetracyclines in Animal Feed. Report by the Committee of Human Health Risk Assessment of Using Subtherapeutic Antibiotics in Animal Feeds, Institute of Medicine, IOM-88-89. National Academy Press, Washington, DC.

Jones, M.E., Draghi, D.C., Thornsberry, C., Karlowsky, J.A., Sahm, D.F., &Wenzel, R.P.(2004). Emerging resistance among bacterial pathogens in the intensive care unit – a European and North American surveillance study (2000–2002).Annals of Clinical Microbiology and Antimicrobials,3(14). Online journal article available at

Kuhn, I., Iversen, A., Finn, M., Greko, C., Burman, L.G., Blanch, A.R., Vilanova, X., Manero, A., Taylor, H., Caplin, J., Dominguez, L., Herrero, I.A., Moreno, M.A., & Mollby, R. (2005). Occurrence and relatedness of vancomycin-resistant enterococci in animals, humans, and the environment in different European regions. Applied and Environmental Microbiology,71(9),5383-90.
Leavis, H.L., Willems, R.J., van Wamel, W.J., Schuren, F.H., Caspers, M.P., &Bonten, M.J.(2007). Insertion sequence-driven diversification creates a globally dispersed emerging multiresistant subspecies of E. faecium. PLoS Pathogens3(1) e7.

Leavis, H.L., Bonten, M.J., &Willems, R.J.(2006). Identification of high-risk enterococcal clonal complexes; global dispersion and antibiotic resistance. Current Opinions in Microbiology, 9(5), 454-460.

Leavis H.L., Willems, R.J., Top, J., Spalburg, E., Mascini, E.M., Fluit, A.C., Hoepelman, A., de Neeling, A.J.,&Bonten, M.J.(2003). Epidemic and nonepidemic multidrug-resistant Enterococcusfaecium. Emerging Infectious Diseases,9(9), 1108-15.

Lee, C.E., Zembower, T.R., Fotis, M.A., Postelnick, M.J., Greenberger, P.A., Peterson, L.R., &Noskin, G.A. (2000). The incidence of antimicrobial allergies in hospitalized patients: implications regarding prescribing patterns and emerging bacterial resistance. Archives of Internal Medicine, 160(18), 2819-22.

Macovei, L. & Zurek, L. (2006). Ecology of antibiotic resistance genes: characterization of enterococci from houseflies collected in food settings. Applied and Environmental Microbiology, 72(6), 4028-35.
Murray, B.E. (2000). Vancomycin-resistant enterococcal infections. The New England Journal of Medicine,342(10), 710-721.
National Antimicrobial Resistance Monitory System (NARMS). (2002-2005). NARMS Retail Meat Annual Reports. Available at
Patton, B.S., Cast, W.R., Kocher, M.E., Matthews, J.O., Griffith, R.W., Hurd, H.S. &Dickson, J.S. (2006). The Effect of Swine Production System on Bacterial Prevalence and Antibiotic Resistance. Poster Presentation at IAFP (International Association for Food Protection) 2006, Calgary, Alberta, Canada, Aug 13-16, 2006. Abstract available at
Rice, L.B. (2001). Emergence of vancomycin resistant enterococci. Emerging Infectious Diseases, 7(2), 183-187.
Rodrigues, J., Poeta, P., Martins, A., & Costa, D. (2002). The importance of pets as reservoirs of resistant Enterococcus strains, with special reference to vancomycin. Journal of Veterinary Medicine B, 49(6), 278-80.
Sechen, S. (2006). The review of animal production drugs by FDA. FDA Veterinarian 21(1) 8-11. Available at
Suppola, J., Kolho, E., Salmenlinna, S., Tarkka, E., Vuopio-Varkila, J., &Vaara, M. (1999). VanA and VanB incorporate into an endemic ampicillin-resistant vancomycin-sensitive Enterococcus faecium strain: effect on interpretation of clonality. Journal of Clinical Microbiology,37(12), 3934-9.
Top, J., Willems, R., Blok, H., de Regt, M., Jalink, K., Troelstra, A., Goorhuis, B., & Bonten, M. (2007). Ecological replacement of Enterococcusfaecalis by multiresistant clonal complex-17 Enterococcusfaecium.Clinical Microbiology and Infection,13(3),316-9.
Varman, M., Chatterjee, A., Abuhammour, W., &Johnson, W.C.(2006). Enterococcal infection. Online article available at edited July 26, 2006).
Webb, M.,Riley, L.W., &Roberts, R.B. (2001). Cost of hospitalization for and risk factors associated with vancomycin-resistant Enterococcus faecium infection and colonization.Clinical Infectious Diseases,33(4), 445-52.
World Health Organization (WHO). (2005). Critically Important Antibacterial Agents for Human Medicine for Risk Management Strategies of Non-Human Use. Report of a WHO Working Group Consultation, 15-18 Feb, 2005. Canberra, Australia. Available at
Willems, R.J., Top, J., van Santen, M., Robinson, D.A., Coque, T.M., Baquero, F., Grundmann, H. & Bonten, M.J. (2005). Global spread of vancomycin-resistant Enterococcusfaecium from distinct nosocomial genetic complex. Emerging Infectious Diseases,11(6), 821-8.
Willems, R.J., Horman, W., Top, J., van Santen-Verheuvel, M., Tribe, D., Manzioros, X., Gaillard, C., Vandenbroucke-Grauls, C.M.J.E., Mascini, E.M., van Kregten, E., van Embden, J.D.A., & Bonten, M.J.(2001). Variant esp gene as a marker of a distinct genetic lineage of vancomycin resistant Enterococcus faecium spreading in hospitals.Lancet, 357(9259),853-855.
Willems, R.J., Top, J., van den Braak, N., van Belkum, A., Endtz, H., Mevius, D., Stobberingh, E., van den Bogaard, A., &van Embden, J.D.A. (2000). Host specificity of vancomycin-resistant Enterococcus faecium.Journal of Infectious Diseases,182(3),816-23.
Wisplinghoff, H., Bischoff, T., Tallent, S.M., Seifert.H., Wenzel, R.P., & Edmond, M.B. (2004). Nosocomial bloodstream infections in US hospitals: analysis of 24,179 cases from a prospective nationwide surveillance study. Clinical Infectious Diseases, 39(3), 309-17.

Chapter 6

Quantitative Risk Assessment of Human Risks of Methicillin-Resistant Staphylococcus aureus (MRSA) from Swine Operations

Dostları ilə paylaş:
1   ...   26   27   28   29   30   31   32   33   ...   57

Verilənlər bazası müəlliflik hüququ ilə müdafiə olunur © 2019
rəhbərliyinə müraciət

    Ana səhifə