Difference between revisions of "Dose response models for Pathogenic Escherichia coli"

From QMRAwiki
Jump to: navigation, search
 
Line 1,110: Line 1,110:
  
 
<br style="clear:both" />
 
<br style="clear:both" />
 +
 +
 +
 +
[[Category:Dose Response Model]]

Latest revision as of 21:50, 11 November 2011

Pathogenic Escherichia coli

Author: Kyle S. Enger
If you want to download this chapter in pdf format, please click here
If you want to download the excel spreadsheet of tables, please click the captions of tables. If you want to download a specific figure, just click on the figure


Overview

Escherichia coli usually exists as a commensal bacterium in the mammalian large intestine, benefiting itself as well as the host. However, there are several well-established pathotypes of disease-causing E. coli (Kaper 2004, Nataro 1998):

  • Enteropathogenic (EPEC)
    • Attaches to small intestinal wall and produces ‘attaching and effacing lesions’, in which microvilli are destroyed and the bacteria become perched on pedestals on the surface of the epithelial cell. This ability is encoded on the locus of enterocyte effacement (LEE) pathogenicity island.
    • Causes an inflammatory response and diarrhea, but seldom in persons older than 2y; it can also be isolated from healthy older persons.
    • Primarily found in developing countries.
  • Enterohemorrhagic (EHEC)
    • This is discussed in more detail in its own chapter.
  • Enterotoxigenic (ETEC)
    • Attaches to small intestinal wall.
    • Produces a heat-labile (LT) and/or heat-stable (ST) toxins, both of which cause secretion from the small intestinal wall, leading to mild to severe watery diarrhea. LT is an immunogenic multisubunit protein similar to cholera enterotoxin. ST is a nonimmunogenic polypeptide containing 18-19 amino acids.
    • Primarily found in developing countries and is a major cause of diarrhea in weaned infants, as well as traveler’s diarrhea.
    • Can be shed even by immune asymptomatic individuals.
  • Enteroaggregative (EAEC)
    • Loosely classified group, some of which may be nonpathogenic.
    • Produces a thick biofilm (‘stacked brick’ configuration) in the small or large intestines.
    • Thought to cause persistent diarrhea (lasting >14d).
    • Can produce many different secretory toxins and cytotoxins, but not ST or LT.
  • Enteroinvasive (EIEC)
    • Actually invades the epithelial cells of the large intestine, where it multiplies.
    • Usually produces watery diarrhea similar to that of EPEC and ETEC, sometimes inflammatory colitis or dysentery.
    • Particularly closely related to Shigella sp. (which are now thought to be subgroups of E. coli); much pathogenesis from EIEC and Shigella sp. is mediated by the pWR100 virulence plasmid.
  • Diffusely adherent (DAEC)
    • Attaches to the small intestinal wall and induces formation of projections which wrap around the bacterium.

Enterotoxigenic Escherichia coli (ETEC) is the most common type of diarrheagenic E. coli (Qadri 2005). It may also be the most common cause of childhood diarrhea in the developing world, responsible for approximately 1/7 of diarrheal episodes in children aged less than 1y and almost ¼ of diarrheal episodes in 1-4 year olds (Wenneras 2004). It can also cause severe dehydrating cholera-like disease in adults (Qadri 2005). Diagnosis is complicated since many other Gram-negative bacteria produce similar toxins, so toxins as well as the E. coli bacterium must be tested for in order to yield accurate results (Wenneras 2004).

ETEC can often be detected in apparently healthy people. In developing countries among healthy 0-11 month olds, and 1-4 year olds, 11.7% and 7.1%, respectively, are estimated to be colonized with ETEC (Wenneras 2004).

Feeding studies of ETEC or EPEC in healthy volunteers typically give 2-3g of NaHCO3, which neutralizes stomach acid and reduces the infectious dose (Levine et al. 1977). However, it has been suggested that food as a vehicle would have a similar acid-neutralizing effect, so feeding studies given with NaHCO3 may better represent natural foodborne infection.(Levine et al. 1977) ETEC and EPEC generally have high ID50, and partly as a consequence of this, they do not appear to be transmitted person-to-person; a study of ETEC-infected volunteers co-housed with uninfected volunteers did not result in any transmission of infection.(Levine 1980) Food was all served individually to the volunteers over the course of the experiment, so there was no opportunity for ETEC to spread via that route.(Levine 1980)




Summary of data and models

There are many human feeding studies of various E. coli types and strains, which can be pooled in various ways to yield different dose response models. Many of these have small sample sizes and cannot be used on their own to reliably fit a dose response model. An important factor is whether the dose was given with bicarbonate, which would neutralize some stomach acid and possibly increase infectivity.

Haas, Rose, and Gerba (1999) fitted beta-Poisson models to EPEC strains O111 (Ferguson et al. 1952) and O55 (June et al. 1953), as well as EIEC strains 4608 and 1624 (DuPont et al. 1971). Diarrhea was the response used.

Haas, Rose, and Gerba (1999) fitted a beta-Poisson model to several pooled datasets describing the disease response from ETEC, EPEC, and EIEC. One strain (EPEC O111) was found to differ from the rest, and was excluded.

Powell et al. (2000) pooled 2 human trial datasets (Levine et al. 1978, Beiber et al. 1998) for EPEC to produce a beta-Poisson model and a Weibull-gamma model.

Table X.X: Summary of dose response data and models
Experiment number Reference(data) Reference(model) Host type Pathogen type Route Dose units Response Best Fit Model Optimized parameters ID50
38 DuPont et al. 1971 human ETEC B7A oral with milk cells diarrhea None (2 doses)
39 DuPont et al. 1971 human EIEC 4608 oral with milk cells diarrhea Exponential k = 9.70E-09 7.14E+07
40 DuPont et al. 1971 human EIEC 1624 oral with milk cells diarrhea Exponential k = 1.22E-08 5.70E+07
99 DuPont et al. 1971 human ETEC B2C oral with milk cells diarrhea None (2 doses)
42 June et al. 1953 Haas, Rose, and Gerba 1999 human ETEC O55 oral cells diarrhea Beta-Poisson α = 8.69E-02

N50 = 1.94E+05

1.94E+05
43 Ferguson et al. 1952 Haas, Rose, and Gerba 1999 human EPEC O111 oral cells diarrhea Beta-Poisson α = 2.61E-01

N50 = 3.39E+06

3.39E+06
142 Coster et al. 2007 human ETEC B7A oral with NaHCO3 CFU diarrhea None (2 doses)
143 .. human ETEC H10407 oral with NaHCO3 CFU diarrhea None (2 doses)
144 Graham et al. 1983 human ETEC H10407 oral with NaHCO3 CFU diarrhea None (1 dose)
145 Levine et al. 1979 human ETEC B7A oral with NaHCO3 cells diarrhea None (2 doses)
147 .. human ETEC B7A oral with NaHCO3 cells diarrhea None (1 dose)
151 .. human ETEC E2528-C1 oral with NaHCO3 cells diarrhea None (1 dose)
153 Tacket et al. 2000 human EPEC 2348/69 oral with NaHCO3 CFU diarrhea None (1 dose)
155 Donnenberg et al. 1998 human EPEC 2362-75 oral with NaHCO3 CFU diarrhea None (2 doses)
157 .. human EPEC 2348/69 oral with NaHCO3 CFU diarrhea None (1 dose)
159 .. human EPEC 2348/69 oral with NaHCO3 CFU diarrhea None (1 dose)
161 Clements et al. 1981 human ETEC 214-4 oral with NaHCO3 CFU diarrhea None (1 dose)
162 .. human ETEC TD225-C4 oral with NaHCO3 CFU diarrhea None (1 dose)
163 .. human ETEC H10407 oral with NaHCO3 CFU diarrhea None (1 dose)
164 .. human ETEC B7A oral with NaHCO3 CFU diarrhea None (2 doses)
165 Levine et al. 1977 human ETEC 214-4 oral with milk cells diarrhea or vomiting Beta-Poisson α = 2.50E-01

N50 = 9.10E+07

9.10E+07
168 Levine et al. 1982 human ETEC H10407 oral with NaHCO3 cells diarrhea None (2 doses)
169 .. human ETEC B7A oral with NaHCO3 cells diarrhea None (1 dose)
170 Levine et al. 1980 human ETEC H10407 oral with NaHCO3 cells diarrhea None (1 dose)
172 .. human ETEC 214-4 oral with NaHCO3 cells diarrhea None (1 dose)
214 Bieber et al. 1998 human EPEC B171-8 oral with NaHCO3 CFU diarrhea Exponential k = 1.97E-09 3.51E+08
216 Levine et al. 1978 human EPEC 2348/69 oral with NaHCO3 CFU diarrhea None (2 doses)
217 .. human EPEC E851/171 oral with NaHCO3 CFU diarrhea None (poor fit)
218 .. human EPEC E74/68 oral with NaHCO3 CFU diarrhea None (all responses negative)
Pooled models
38, 39, 40, 42, 99, 144 See above Haas, Rose, and Gerba 1999 human ETEC, EPEC, EIEC oral CFU diarrhea Beta-Poisson α = 1.78E-01

N50 = 8.60E+07

8.60E+07
214, 216, 217 See above Powell 2000 human EPEC oral CFU diarrhea Beta-Poisson α = 2.21E-01

N50 = 6.85E+07

6.85E+07




Optimized models: uncertainty and fitting analyses.

Human hosts exposed orally to EIEC 4608; response is mild to severe diarrhea

Table X.X. Dose response data
Dose
(cells)
Diarrhea Total
Yes No
1.00E+04 0 5 5
1.00E+06 0 5 5
1.00E+08 5 3 8
DuPont et al. 1971.


Table X.X. Goodness of fit and model selection
Model Deviance Goodness of fit Model comparison
Value DF χ2 p χ21, DF p
exponential 0.0 2 6.0 0.952 3.8 1.000
beta-Poisson 0.1 1 3.8 0.751
Conclusion: Exponential fits better than beta-Poisson; cannot reject good fit for exponential.
Table X.X: Parameters (k) and values (ID50) for the best fit model; percentiles from 104 bootstrap iterations
Parameter MLE estimate Percentiles
0.5% 2.5% 5% 95% 97.5% 99.5%
k 9.70E-09 1.33E-09 2.86E-09 4.66E-09 2.04E-08 5.07E-08 5.07E-08
ID50 7.14E+07 1.37E+07 1.37E+07 3.40E+07 1.49E+08 2.43E+08 5.23E+08


Figure X.X. Histogram showing uncertainty of k parameter estimates (bootstrap realizations).
Figure X.X. Exponential model plot, with confidence bounds around optimized model.





Human hosts exposed orally to EIEC 1624; response is mild to severe diarrhea

Table X.X. Dose response data
Dose
(cells)
Infection Total
Yes No
1.00E+04 0 5 5
1.00E+06 1 8 9
1.00E+08 3 2 5
DuPont et al. 1971.


Table X.X. Goodness of fit and model selection
Model Deviance Goodness of fit Model comparison
Value DF χ2 p χ21, DF p
exponential 0.9 2 6.0 0.224 3.8 0.085
beta-Poisson 0.0 1 3.8 0.901
Conclusion: Exponential fits better than beta-Poisson; cannot reject good fit for exponential.
Table X.X: Parameters (k) and values (ID50) for the best fit model; percentiles from 104 bootstrap iterations
Parameter MLE estimate Percentiles
0.5% 2.5% 5% 95% 97.5% 99.5%
k 1.22E-08 1.97E-09 2.19E-09 4.40E-09 4.03E-08 1.17E-07 4.03E-07
ID50 5.70E+07 1.72E+06 5.92E+06 1.72E+07 1.58E+08 3.17E+08 3.52E+08


Figure X.X. Histogram showing uncertainty of k parameter estimates (bootstrap realizations).
Figure X.X. Exponential model plot, with confidence bounds around optimized model.




Human hosts exposed orally to ETEC 0111; response is slight to severe illness

Table X.X. Dose response data
Dose
(organisms)
Illness Total
Yes No
7.00E+06 7 4 11
5.31E+08 8 4 12
6.50E+09 11 0 11
9.00E+09 12 0 12
Ferguson et al. 1952.


Table X.X. Goodness of fit and model selection
Model Deviance Goodness of fit Model comparison
Value DF χ2 p χ21, DF p
exponential 51.4 3 7.8 0.000 3.8 0.000
beta-Poisson 6.4 2 6.0 0.041
Conclusion: Neither the exponential nor beta-Poisson fits well; beta-Poisson is less bad.
Table X.X: Parameters (α & N50) for the beta-Poisson model, percentiles from 104 bootstrap iterations
Parameter MLE estimate Percentiles
0.5% 2.5% 5% 95% 97.5% 99.5%
α 2.61E-01 -- -- -- -- -- --
N50 3.39E+06 -- -- -- -- -- --
A very similar model fit has been published by Haas, Rose, and Gerba (1999).


Figure X.X. Scatterplot showing uncertainty of α & N50 parameter estimates (bootstrap realizations).
Figure X.X. Beta-Poisson model plot, with confidence bounds around optimized model.




Human hosts exposed to ETEC 214-4 (ST) orally in milk; response is diarrhea or vomiting

Table X.X. Dose response data
Dose
(organisms)
Illness Total
Yes No
1.00E+06 0 4 4
1.00E+08 3 2 5
1.00E+10 4 1 5
Levine et al. 1977.


Table X.X. Goodness of fit and model selection
Model Deviance Goodness of fit Model comparison
Value DF χ2 p χ21, DF p
exponential 0.5 2 6.0 0.000 3.8 0.000
beta-Poisson 1.7 1 3.8 0.466
Conclusion: Beta-Poisson fits better than exponential; cannot reject good fit for beta-Poisson.
Table X.X: Parameters (α & N50) for the beta-Poisson model, percentiles from 104 bootstrap iterations
Parameter MLE estimate Percentiles
0.5% 2.5% 5% 95% 97.5% 99.5%
α 2.50E-01 -- -- -- -- -- --
N50 9.10E+07 -- -- -- -- -- --
FILL IN CITATION OF MODEL FIT HERE.


Figure X.X. Scatterplot showing uncertainty of α & N50 parameter estimates (bootstrap realizations).
Figure X.X. Beta-Poisson model plot, with confidence bounds around optimized model.





Human hosts exposed to EPEC B171-8 (serotype O11:NM) orally with buffer; response is diarrhea

Table X.X. Dose response data
Dose
(CFU)
diarrhea Total
Yes No
5.00E+08 3 2 5
2.50E+09 6 0 6
2.00E+10 2 0 2
Bieber et al. 1998.


Table X.X. Goodness of fit and model selection
Model Deviance Goodness of fit Model comparison
Value DF χ2 p χ21, DF p
exponential 0.6 2 6.0 0.950 3.8 1.000
beta-Poisson NA 1 NA NA
Conclusion: Exponential fits better than beta-Poisson; cannot reject good fit for exponential.
Table X.X: Parameters (k) and values (ID50) for the best fit model; percentiles from 104 bootstrap iterations
Parameter MLE estimate Percentiles
0.5% 2.5% 5% 95% 97.5% 99.5%
k 1.97E-09 7.78E-10 1.02E-09 1.03E-09 3.23E-09 3.23E-09 5.67E-08
ID50 3.51E+08 1.22E+07 2.14E+08 2.14E+08 6.76E+08 6.76E+08 8.90E+08
FILL IN CITATION OF MODEL FIT HERE.


Figure X.X. Histogram showing uncertainty of k parameter estimates (bootstrap realizations).
Figure X.X. Exponential model plot, with confidence bounds around optimized model.





Human hosts exposed to various ETEC, EPEC and EIEC strains orally; response is diarrhea

Table X.X. Dose response data
Dose
(CFU)
Diarrhea Total
Yes No
1.00E+05 0 5 5
1.00E+05 0 5 5
1.00E+06 0 5 5
1.00E+06 1 8 9
1.00E+08 1 4 5
1.00E+08 5 3 8
1.00E+08 3 2 5
1.00E+08 2 3 5
1.43E+08 6 2 8
2.70E+08 9 7 16
1.73E+09 5 2 7
5.33E+09 6 2 8
1.00E+10 4 1 5
1.00E+10 3 2 5
1.60E+10 7 1 8
June et al. 1953, DuPont et al. 1971, Graham et al. 1983.


Table X.X. Goodness of fit and model selection
Model Deviance Goodness of fit Model comparison
Value DF χ2 p χ21, DF p
exponential 114.3 14 23.7 0.000 3.8 0.000
beta-Poisson 14.9 13 22.4 0.924
Conclusion: Beta-Poisson fits better than exponential; cannot reject good fit for beta-Poisson.
Table X.X: Parameters (α & N50) for the beta-Poisson model, percentiles from 104 bootstrap iterations
Parameter MLE estimate Percentiles
0.5% 2.5% 5% 95% 97.5% 99.5%
α 1.78E-01 -- -- -- -- -- --
N50 8.60E+07 -- -- -- -- -- --
A very similar model fit has been published by Haas, Rose, and Gerba (1999).


Figure X.X. Scatterplot showing uncertainty of α & N50 parameter estimates (bootstrap realizations).
Figure X.X. Beta-Poisson model plot, with confidence bounds around optimized model.





Human hosts exposed to EPEC orally; response is illness

Table X.X. Dose response data
Dose
(NA)
NA Total
Yes No
1.00E+06 0 4 4
1.00E+06 1 4 5
1.00E+08 1 4 5
5.00E+08 3 2 5
2.50E+09 6 0 6
1.00E+10 3 2 5
1.00E+10 5 0 5
2.00E+10 2 0 2
Levine et al. 1978, Bieber et al. 1998.


Table X.X. Goodness of fit and model selection
Model Deviance Goodness of fit Model comparison
Value DF χ2 p χ21, DF p
exponential 31.0 7 14.1 0.000 3.8 0.000
beta-Poisson 16.5 6 12.6 0.081
Conclusion: Beta-Poisson fits better than exponential; cannot reject good fit for beta-Poisson.
Table X.X: Parameters (α & N50) for the beta-Poisson model, percentiles from 104 bootstrap iterations
Parameter MLE estimate Percentiles
0.5% 2.5% 5% 95% 97.5% 99.5%
α 2.21E-01 -- v -- -- -- --
N50 6.85E+07 -- -- -- -- -- --
A very similar model fit has been published by Powell (2000).


Figure X.X. Scatterplot showing uncertainty of α & N50 parameter estimates (bootstrap realizations).
Figure X.X. Beta-Poisson model plot, with confidence bounds around optimized model.