Animal & Veterinary
CVM Risk Assessment: Assumptions and Uncertainties
Dr. Kathy Hollinger and Mary Bartholomew
MS. BARTHOLOMEW: Good afternoon.
I hope you have had some time since lunch with the break and everything to get your serotonin levels back up to an acceptable point. And so on that assumption, I am going to get started talking about the assumptions and the statistical uncertainties in our risk assessment.
There are two different sorts of assumptions that are used in this model. The first sort is the type that is used when there is a lack of data. For example, there is information about the rate of seeking care among people with all sorts of diarrheal disease. There is not that same information about the rate of seeking care in patients with Campylobacteriosis.
So we had to make an assumption at that point. So that is one type of assumption. You don't have the data in the specific population of interest. So you make an assumption you can apply the same rate that you have got in a given population to another.
And then the other sort of assumption is a statistical assumption. We have data in the appropriate population, but it -- the parameter of interest is not known with perfect knowledge. So we make the assumption that given the data that we have got, we apply a particular statistical model. And that is our assumption, the given statistical model, to generate the uncertainty distributions about the parameter of interest.
So I am going to talk about a couple of the global assumptions of the first type. And then I will turn it over to Kathy. She will talk about some more of those. And since there is a lack of data involved, she will also consider some of the data gaps.
Our first global assumption is that susceptible and resistant Campylobacter have the same virulence characteristics. At the time that we started the risk assessment model, that is certainly what we thought. And as was mentioned earlier this morning by Dr. Angulo, there may be some indication of a difference -- that this is not true in the future.
So we will be looking for more information when it becomes available. And if that happens, we will have to modify something in the risk assessment.
We also assume that susceptible and resistant Campylobacter have the same survival characteristics from slaughter to the point of human exposure. Again, we have no indication that that is not the case.
We also made the assumption that human susceptibility to infection remains constant in the population.
And that consumption patterns remain constant. And in the short time frame that this risk assessment covers, that is probably true. But you have also heard some information from Dr. Cray earlier this morning that, in fact, if you look over a wide enough period of time, that is not true either.
Well, as we said, our risk assessment model is fairly simple and it is flexible. If we find out different information that leads us to believe that these assumptions are not true, we will update it. And I will turn this over now to Kathy. Dr. Hollinger will cover some of the other fine points of the data assumptions.
DR. HOLLINGER: So I get to give the top ten list so to speak, but in reverse of David Letterman's grouping of the -- his top ten list. Our risk assessment model modeled the measurable human health impact of fluoroquinolone-resistant Campylobacter jejuni and coli that were acquired from poultry sources using the most currently available data to model that risk.
The assumptions I have listed here in order of priority in the model, the impact in the model from my perspective, not necessarily from a mathematical perspective. The first assumption, the one with the most impact on the model, is that fluoroquinolone resistance in people calculated after we removed those people who traveled, those people who used fluoroquinolones prior to culture, and those for whom the time of fluoroquinolone use was unknown was attributed to chickens.
And we removed travelers because it is known that travelers carry very frequently higher levels of resistance than the general population. They don't acquire their disease in the United States. And, therefore, their disease was not tied to domestic drug use in food-borne sources.
Fluoroquinolone use is associated with development of resistance so that those people who had cultures taken after they had used a fluoroquinolone could possibly have had a resistant infection due to that use. And those people for whom the time of the fluoroquinolone use was not known could have then had their fluoroquinolones prior to taking cultures.
So that remaining resistance then was attributed to chickens. And we use the Campy case control study to be able to determine what proportion were travelers and who had used fluoroquinolones and those who did not know when their fluoroquinolone was used.
This assumption represents a data gap. The remaining level of resistance could have been distributed either uniformly across all sources of human infections that were remaining or that resistance could have been attributed to a single source or to certain specific sources.
So this assumption was based upon evidence of fluoroquinolone resistance developing in chickens, humans and when fluoroquinolones are used because there was no food animal fluoroquinolone use other than the use in poultry until late 1998. And there was no fluoroquinolone resistance observed prior to '92 in human cases in the U.S. even though fluoroquinolones had been approved for human use since 1987.
We felt it was unlikely that the increase in domestically acquired fluoroquinolone resistance that was observed in people since 1996 as reported in the Minnesota paper that was published in May of '99 could be attributed to levels of resistant Campylobacter that were uniformly distributed amongst all sources of human infection.
The distribution of resistance in food-borne sources of infection was more likely to be associated with specific exposures linked to drug use and was assumed to be limited predominantly to poultry.
Assumption number two states that the level of risk ascertained in the early 1980s represent the current level of risk in the U.S. population. And this is the risk of acquiring a poultry associated infection. And we modeled this estimate. And we used the literature. And the proportion of cases that could be attributed to exposure to chicken was 48 to 70 percent in the literature.
This wide range was modeled with a uniform distribution to account for the large amount of uncertainty in this parameter. The CDC is currently analyzing a case control study evaluating risk factors for Campylobacteriosis which we expect will provide an update of this estimate and maybe a more precise estimate.
Both assumptions one and two represent data gaps in, you know, precision of estimates and the proportion of human disease attributable to the specific source of infection and then how to determine the level of resistance in specific food-borne sources of infection.
Assumption number three, we had some data for the level of resistance that was observed in broiler chickens. And that data was a sample of only 159 isolates that were collected in a pilot survey. And the collection period was limited from October to December.
The level of resistance in chickens was modeled using the level of resistance in Campylobacter jejuni species alone as there were no data available that were specific to Campylobacter coli.
This may have slightly under-estimated the level of resistance. But because Campylobacter coli represents such a small proportion of human disease, only 2.7 percent in the NARMS isolates in '98, it was unlikely to have much, if any, impact on the overall estimate of the risk.
A prevalence survey is currently being conducted by FSIS that will provide a more robust sample for isolate susceptibility testing in 1999.
The next assumptions have been grouped together because Campylobacter-specific data were not available for the proportion of enteric cases that sought care for either bloody or non-bloody diarrhea for those cases that were requested to submit a stool and did submit a stool for culture for both bloody and non-bloody diarrhea and for the proportion of people that received treatment but never submitted a stool sample.
Rates for these parameters were obtained from population surveys conducted by CDC at FoodNet sites for diarrheal illness or from a survey of physicians that saw patients for diarrheal disease. And I will give one example of the seeking cure assumption.
This assumption states that the rate of seeking cure among people with diarrheal illness is similar to the rate of seeking cure among people with Campylobacteriosis. And then this assumption was divided into two components, one for bloody diarrhea and one for non-bloody diarrhea because the rates of seeking care were expected to be different.
Bloody stools were significant risk factors associated with seeking care in a multi-variate analysis of the population survey data. The rates of seeking cure were obtained from the population survey for persons with diarrheal disease. And diarrheal illness was defined as three or more lose stools within a 24-hour period or diarrhea lasting for more than one day or which resulted in an inability to perform normal activities.
And as a validity check or a cross-check to see if population data could really apply to these parameters for Campylobacteriosis, a comparison of symptoms significant in seeking care for diarrheal illness in Campylobacteriosis was made to determine if this rate was applicable to these Campy rates.
Comparing the groups, a greater proportion of people with culture-confirmed Campylobacter cases were affected by fever and blood in their stool than the people seeking care for diarrheal illness. Therefore, the actual rate of seeking care for Campylobacteriosis may be somewhat under-estimated.
However, because a greater proportion of people with fever and bloody stools would be cultured and possibly enrolled in the case control study, it makes such comparisons somewhat difficult.
Our next assumption was that the incidence rates for culture-confirmed Campylobacter infections in FoodNet catchment are representative of incidence rates for culture-confirmed Campylobacter infections in the United States. We compared demographic characteristics of the FoodNet catchment area population to the U.S. population. And we looked at characteristics available from the U.S. Census Bureau.
And some of those are rural to urban population distribution, age, sex and race. And these characteristics were similar except for fewer Hispanics were represented in the FoodNet catchment area than were in the U.S. population. And we felt that because this comparison of demographic characteristics was so similar between the FoodNet and U.S. populations, that this indicated risk factors for the disease may also be distributed similarly.
And, therefore, rates of disease obtained from FoodNet would be likely to be representative of disease rates in the United States. And the table comparing these demographic characteristics is available in Section 1 of the risk assessment.
Again, we group these assumptions 11 through 13 here because data were not available to describe invasive disease parameters. The invasive disease parameters that we were looking for were the proportion of cases seeking care, the proportion of cases that were requested for and submitted specimens, the sensitivity of culture methods and treatment rates.
Invasive disease is predominantly bloodstream infections and bloodstream culture methods are a fairly good method for isolating Campylobacter. And we felt that probably most of these invasive cases would be detected -- first of all, that they would seek care; that they would be detected through these culture methods; and that they would be treated with an antimicrobial.
Because invasive disease cases represent less than one percent of overall number of cases, we felt that even if we were slightly under-estimating or over-estimating the total number of cases, that it would have very little impact on the overall risk.
Assumption number 14 was the proportion of people not submitting a specimen that received antimicrobials for treatment of diarrheal disease was similar to the proportion of people with Campylobacteriosis that didn't submit a specimen and were treated. We didn't have a data parameter for that from the Campy case control study because all of the cases that were included there were culture-confirmed cases.
So we went to the population survey and looked, again, at diarrheal illness and found that those people who don't submit a stool sample were treated at a rate of around 40 percent compared to the culture-confirmed cases who were treated at a rate of 84 percent.
Assumption number 15 looked at the proportion of people treated with fluoroquinolones. And we used that to -- and it was the same for people with enteric disease and people with invasive Campylobacteriosis and enteric Campylobacteriosis that did not submit a stool for culture.
Again, the treatment rates using fluoroquinolones were obtained from the Campy case controls data. And, again, those were culture-confirmed cases and invasive Campy. And for those people who did not submit a culture, we needed drug use information. And then we assumed that they would be treated at the same rate as other individuals with enteric disease. Okay. Mary.
MS. BARTHOLOMEW: As I alluded to earlier, the second sort of assumption was applied in considering uncertainty distributions. When we didn't know -- when we didn't have perfect knowledge of a population parameter, we would want to estimate the uncertainty that we had about it. It would be in the specific population of interest. But we still needed to show that. We had only a sample out of the total population.
So we had to model the -- we had to model -- make an assumption about the statistical model that would be appropriate for doing so. So, for instance, if we had a proportion like the proportion of people seeking care that we wanted to model, we assumed that that binomial proportion was, in fact, a beta.
If we did not, in fact, have the real proportion, for instance, if the p that we were given -- the estimate for p being the proportion that we were seeking -- was a weighted estimate such as from a population survey, sometimes a population survey is done in such a way that the different areas that are sampled are disproportionate. So then the surveyors will adjust by giving you a weighted proportion.
Given the weighted proportion, we didn't have a numerator. So what we did was we took p* which was the given weighted estimate based on a sample of size n, and then we back calculated and said the success rate for that given p* would be n times p*, or s*, s* being the number of successes, the numerator that we didn't have.
And then we modeled p using that s* as a beta, s* plus one which is the number of successes plus one and the number of failures plus one. That is a kind of standard assumption for modeling binomial proportions.
I will go down a laundry list more or less for the different output variables. When several variables are strung together to create an output, then the output has the product uncertainty. So this was the output that Dr. Vose showed for the total number of Campylobacteriosis cases in the United States for 1998.
And as he mentioned, we should think of this not as the distribution of the number of cases, but as the expected mean. So that the mean could be anywhere from about 0.9 to about, what is it, nine something -- I can't read those numbers from here. But anyway, you can see what they are. And so that is not an estimate --
You can't? Oh, dear. I will tell you. It looks to me like 4.8 is kind of up there in the tail. Anyway, I will tell you the laundry list of the variables that were included. We had to incorporate uncertainty about the proportion of cases with enteric and bloody stools, and the enteric with non-bloody stools or those that had invasive disease in the first place.
Then we had to determine the proportion of each of those types who sought care and the uncertainty distributions about those. We had to develop the proportions of each type who requested -- who were requested and did submit culture specimens and the uncertainty distributions about them and the proportion of cases that were tested who were actually ascertained to be positive by the culturing procedure.
So that this estimate here involves the uncertainties from those four different -- five different sets of proportions.
This is the output variable for Section 3 which ways the fluoroquinolone-resistant Campylobacter infections from chicken that received fluoroquinolone as treatment. And the laundry list of variables of uncertainties -- whose uncertainty distributions had to be included were proportion of Campylobacter due to chicken consumption, proportion of persons seeking care, proportion of those seeking care who receive antibiotic therapy, proportion of those receiving antibiotic therapy who have received fluoroquinolone and the proportion of infection from chicken that are resistant.
And then finally, we have the output variable for Section 4 is the nominal mean total number of people with fluoroquinolone-resistant Campylobacter infection from chicken that we see fluoroquinolone as treatment. And the uncertainty in that variable comes from the prevalence of Campylobacter on chicken carcasses, the prevalence of resistance among Campylobacter isolates in the slaughter plant, the prevalence of fluoroquinolone-resistant Campylobacter on carcasses and amounts of chicken consumed.
And so naturally the probability distributions that you were shown as the final analysis in Section 5 depend on all of the above. And that pretty much describes how we dealt with model uncertainty. Thank you.
DR. LONG: Okay. These two are ready to take questions. I think they have tried to lay out the assumptions and the uncertainties. And we are interested in what you have to say. Please, if you can line up behind the microphone, that would be great.
MR. : I would like to ask just about a couple of further uncertainties and assumptions that you may be making. The first is that any chicken Campylobacter is the same as the Campylobacter that will cause infection in man. Although there clearly are cases where this thing can be made, I think there are many cases where it can't.
And as I understand, the distribution of Campylobacter in chickens doesn't by any means relate very closely to the distribution in a population from humans. That was the first one. Would you like to make a comment on that?
DR. HOLLINGER: Yes. I think that the Campylobacter on chickens very closely parallels the Campylobacter that we find causing infections in people. We have seen strain typing -- first of all, the list that Dr. Cray offered earlier today, if you look at the species level, the predominate isolate from chickens is Campylobacter jejuni. And you find that isolate in human clinical cases.
And if you want to look at the -- you know, looking at strains within Campylobacter jejuni, you can type strains by many different methods. And when you do strain type by either the biotyping, serotyping, or even using a PCR or some of the other RFLP techniques, you find that very -- there are -- there is a lot of overlap between human, poultry and cattle strains. And that is a recent paper out from Denmark.
And then you find a little different association from the Campylobacter coli. You find similar strains amongst swine, people and chickens. So I would say that the evidence is there. It is available in the literature. And it shows that the strains very closely do overlap between humans and poultry.
MR. : Is the question of the burden of contamination of the chicken carcass something that you are taking an assumption about, that any organism whether present in tens, hundreds or millions is equivalent regardless?
DR. HOLLINGER: We have a table in the risk assessment that shows the burden of contamination. The most probable number in the case of this survey, the FSIS baseline surveys, because they used enrichment procedures. And the burden of Campylobacter on chickens is considerably higher than any other food animal species that was sampled.
MR. : But you nevertheless have to assume that any contamination is equivalent to any other in your -- in the way you take these into account, do you not?
DR. HOLLINGER: David, do you have a --
(Away from microphone.)
DR. VOSE: --- but mathematically, it is looking at a quantity of --- and post-slaughter, just the chiller. And if it is infected with --- Campylobacter, then it doesn't really matter from the point of view of the mathematics what the number of bacteria there are in that sample. Clearly, it does matter when it comes to feeding that to a person because a large amount of bacteria, the more likely they are going to be ill of course.
But if the distribution of the number of bacteria that will be in contaminated carcasses remains constant, then the mathematics of this problem remains constant, too. If the distribution changes so if we were to institute some risk management technique that reduced the load, then we would have to make a change in our model which I think turns out to be a reasonably simple thing to do under a certain --- unlimiting assumption. But it does point at -- it is not necessarily considered a given item
--- how many Campylobacter ---
MR. : My final point is this one to do with the seasonality. As I understand it, the reason for this seasonality is far from clear. But it is very dramatic as we saw this morning. I wonder whether this really does raise a question about our knowledge of the epidemiology of this organism which suggests that maybe we are assuming a simplicity of connection here which maybe turned out to be misplaced in time. Thank you.
MS. BARTHOLOMEW: I think the important thing about that is that we are looking at an annualized rate. And so that if there are peaks and valleys, it is really sort of the annualized rate that we are modeling. And if there are significant shifts in that annualized rate, they will maybe have some peaks and valleys.
But if the peaks and valleys increase the following year, then the annualized rate the following year will also increase.
MR. : There is something odd going on though.
MR. CONDON: Yes, Robert Condon. It may affect your estimates to just that seasonality depending on these case control studies. The question I wanted to address was your estimates of the portion of cases coming from poultry chicken primarily and the case control studies that you used. And the values there are based on the factors they looked at. In the Colorado study, they only looked at it as reading a summary, only two things.
So the fact that you got 70 percent out of that study to me doesn't mean anything. If you only look at pets and poultry, you are going to have a higher proportion due to poultry. The more things you look at, the more possible sources, the less you are going to have from poultry. And that is a limitation on the study and that is a bias that you get in your estimates.
DR. HOLLINGER: Right. And I --
MR. CONDON: And that is something you haven't really -- I haven't seen mentioned here, the bias of these estimates yet.
DR. HOLLINGER: Okay. Well, there are some description of that in the risk assessment text itself. But in the university study, it is not that they only looked at two risks. They certainly looked at more risks.
But because they looked at a subset of the population that did not have certain exposures such as raw milk exposure or had not traveled, I believe that, you know, when you say there are biases, certainly the high level of risk in that population was due to their limited exposures. And I think that the reason that that study was included was because we have a lot of uncertainty in what that actual estimate ought to be, a more precise estimate. And we saw that between -- somewhere between 48 and 70 percent we thought would be an estimate for -- or it would be a broad enough range to include maybe the actual estimate for the general population because the general population is certainly an average set of exposures.
MR. CONDON: Well, but there are a couple of issues. One is the -- you've got a study that in your report here -- and I have only had a chance to look at this briefly -- it says it is not representative. Okay. I think at that point when you are trying to make an inference back to the population, you take that study, you put at the top of it, "Sample not representative", you put it away. You don't worry about it. You don't get confused by trying to use it.
I mean, the best example I can think of as far as a nonrepresentative study is in 1936, a political poll was done for who was going to be President. One hundred thousand people were asked. Roosevelt was going to be overwhelmingly defeated was the results of the poll, not even close.
There was a question about the representativeness of the poll. It was a telephone survey. And if you think back in 1936, a lot of people did not have telephones. A lot of the people who did not have a very high income did not have telephones. So there was a bias in that. And that set polling back 30 years.
DR. HOLLINGER: I would say --
MR. CONDON: And so that is where -- once you say the data is not representative, put it away. Don't even try to use it.
DR. HOLLINGER: Wait. No, no, no. You know --
MR. CONDON: Because it is just going to confuse you.
DR. HOLLINGER: Okay. And what I said was it was representative of certain sub-groups in the population, Bob. And the reason it was included was because we knew that people were eating more chicken than they had in the past. We knew that exposures have probably changed since 1981. And we wanted to show that we had little confidence in one single point estimate. That's why. David, did you have something you wanted to say?
(Away from microphone.)
DR. VOSE: Well, I just wanted to reiterate exactly what you said because --
DR. LONG: Come up to the microphone, please.
DR. VOSE: I wanted to reiterate exactly what Kathy has just said. I think -- we are trying very hard to recognize where we have uncertainty. And I think if we had gone to this one study, if you like, that had one figure, I think that that would have been more of a failure than to have taken some -- two studies that were dissimilar and say, well, hell, it is going to be somewhere in there, probably somewhere between the two.
It is much better from our point of view to recognize that we don't know that very well so that we instigate discussions like this because if we picked one single estimate, it is almost certainly going to be wrong. Actually, where we are right now, we are almost certainly going to be right that it is somewhere in where we are. And maybe we can argue, but later.
But -- and you are quite right. There is going to be a bias in there because we have got that higher prevalence. But this is a work in progress. As Dr. Sundlof said, it is a living document. And if this -- okay. Well, if I put in a single estimate, it wouldn't ever have appeared in that spider plot that you all saw. It wouldn't have appeared there as something that is flagged say, hey, we don't know a lot about that.
Because it is there, we are going to have a discussion. And maybe -- I hope it is because it is a very dominant parameter of the model. I hope that we are -- it is going to instigate some more research that will try to get a better estimate of what those values are. So I still -- from a modeling perspective, I prefer a strategy of modeling if you like. I prefer to put it in.
And you will notice it had a uniform distribution. Now, I don't know if any of you will ever read my book. There must be somebody. No? Oh, Louise, hurray.
Were you to read my book, you would see that I loath the uniform distribution. I hate it a lot. But -- and the only time I ever really use it is to make it stand up and to make people shout about it and say, hang on, that's not fair. You know, we've got to know something a little bit better than that. Hence this discussion.
So don't too much pick up the numbers. But certainly if you have some better data, then -- any of you, then, my goodness, we would be very willing to see it.
MS. BARTHOLOMEW: I can add that we have had this sort of discussion sort of internally about this, that we are not all that pleased with the 70 percent as being representative. But we didn't have other things in black and white. And there are ways, in fact, to incorporate expert opinion. We just didn't know whose expert opinion we wanted to incorporate there I guess.
MR. : Maybe I should sit down then. We agree that this is an important estimate. And it would be nice to know precisely what the proportion of Campylobacter cases in this country are attributed to each food commodity. And it would be nice to know how much is due to poultry and other foods.
I think it is in your range of estimates, the 48 percent to the 70 percent, is entirely defendable based upon the current published data. It has been replicated in the United States in smaller studies. And it has been demonstrated in very recent large case control studies in New Zealand, in Denmark and in the United Kingdom.
And whether you decide to put the -- use the uniform distribution, if you decide to put it at 48 percent or 70 percent, it doesn't -- the outcome is just influenced -- you still have this demonstrable outcome. And os if you want -- prefer to use a more conservative estimate, 48 percent or some people actually want to go lower than that, then you still can decrease the outcome by just as much.
But you still have this demonstrable outcome. So I really don't think it is -- it certainly -- to quibble about where to exactly put that estimate would be to speak against the current literature which has already gone through peer review.
MS. LASKEY: I am Tammy Laskey. I am an Epidemiologist at the Food Safety Inspection Service. And this may be a bit of a digression. But the contradiction of having a large proportion of cases associated with raw milk consumption and then such a low prevalence or exposure to raw milk in the population that one can't study it suggests that the probability of becoming infected given that the bacteria are in the raw milk is different than the probability of becoming infected if the bacteria are in the chicken for whatever reason, either a dose or a virulence or a strain, a reason that we don't know.
But it is a piece of very important information. And I would suggest it needs further exploration. It is very intriguing, as well.
DR. HOLLINGER: Well, I think the level of exposure from raw milk to chicken, the comparison, I mean, the difference is going to be huge. Very few people are drinking raw milk. And since I believe 1987, there was a raw milk interstate ban of sale. So raw milk has generally been associated with outbreaks. And that represents less than maybe one percent of all Campy cases.
So raw milk as far as being significant in this risk assessment is probably not. It is probably very, very small compared to poultry.
MS. LASKEY: Right. But I was saying in terms of understanding Campylobacter infections in general and the contribution by raw milk, it is suggesting something different is happening in the raw milk situation than -- because it is a way disproportionate number of cases. Even though it is small, one percent of the population does not drink raw milk. So finding one percent of the cases there is very strange. And I am just bringing this contradiction up as a point for further study.
DR. HOLLINGER: Thank you.
(Away from microphone.)
DR. VOSE: Kathy, does that have to do with the detection of milk before the infection ---
DR. HOLLINGER: I don't think that really, that if one percent of the population is having -- is an outbreak-associated case, fewer cases are raw milk-associated cases, much smaller number. As far as I think what she was getting at was somewhat about the pathogenesis.
And I think the interesting information that was brought to us from Canada about cross-contamination within the kitchen from poultry sources also is very interesting. So it really may be vehicle dependent. And, you know, the infection or susceptibility to infection may be vehicle dependent.
Certainly, Salmonella has shown that -- in fatty foods, that it is protected in the stomach from acid and then is more likely perhaps to cause an infection. So, yes, that is an area that could have more research done to understand. But, again, because it is such a low number of cases, that is a question apart from this risk assessment. Yes?
MS. MORNER: My name is Ann Morner. I work for Bayer in Europe. And I just wanted to draw your attention to Danish results within the Dane Map Surveillance System in which there is a considerably higher resistance level in Campylobacter isolated from retail products compared to isolates from the carcasses at the slaughterhouse indicating that something is happening.
Then I had a question regarding the -- if you have taken into consideration the number of people at risk, whatever 4,000 to 6,000 people being at risk, how many of these cannot be treated with fluoroquinolones because of their age or because of other factors so that they will not be given the fluoroquinolones as a first time choice.
DR. HOLLINGER: Yes. In response to your first question, the Dane Map and Danish situation, that difference has been demonstrated because of imported products. A lot of the imported foods -- and this was also demonstrated in the U.K. That the imported products has higher levels of resistance than did the domestically produced products. So that is one issue.
And as far as the children who were not treated with fluoroquinolones, we only looked at that actual proportion of people who were treated with fluoroquinolones. So those people who had other treatments or who were not treated were not considered in this risk assessment.
MS. MORNER: Thank you.
DR. KRISHINSKY: My name is Beth Krishinsky. I am with Wompler Foods. I just had a question on the volume of boneless, domestically-reared broiler that is consumed -- broiler products that is consumed in the United States. With the changing trends and consumption patterns from cutting up a whole bird at home to eating pre-prepared breaded, fried fast food products in the general population, fast food restaurants, how do you reconcile the exposure to raw chicken as being a source of Campylobacter infection or cross-contamination when an increasing percentage of chicken that is consumed is already precooked and packaged either in a restaurant or in fast food?
DR. HOLLINGER: What can happen in that circumstance is that they can be preparing the food. But after the food is prepared, they can handle or contaminate it. So food handler education would be very important. And I think that it is -- there is a considerable amount of cross-contamination either in restaurants or at homes. And that handling food isn't the only source of people's infections.
DR. KRISHINSKY: Do you think that your assumption of the volume of poultry that is consumed in the United States should be adjusted for products that are already pre-breaded and sold frozen and only deep fat fried at the restaurant site?
DR. HOLLINGER: David has an answer for that one. Excuse me.
DR. VOSE: You've got a good point. And one could do that. The value of K, this mystical K value, implicitly takes into account what you are saying. There is only a certain number that will go out into the consumer's pathway that still contains Campylobacter. And we don't know what that is. We have never tried to address it.
So there is a proportion, if you like, where you could separate that proportion that is already pre-cooked and never goes near a consumer before all the Campylobacter are dead and then that which are uncooked and received by the consumer.
And if we did that, we would say, well, the volume of meat now is much smaller. But the value of K will be correspondingly higher. It would quite amount to the same thing because we were saying that we now have a fewer number of pounds of potentially contaminateable meat. And yet they are producing this level of infection in humans. So this sort of --
(Away from microphone.)
DR. KRISHINSKY: ---.
DR. VOSE: Only if you see the chickens produce infections, well, absolutely right. I mean, of course. But if that is wrong, then, you know, the whole thing is blown out of the water. Yes.
But, absolutely. And I do hope that we make that assumption explicit. If we didn't, then I am making it now. If that is a shock to any of you, I hope not -- okay.
(Away from microphone.)
DR. KRISHINSKY: --- not agree with it.
DR. VOSE: Okay, well, if you don't agree with it, then yes.
DR. LONG: We need to move ahead. We will have one more question and any other comments can be deferred to the public comment section at the end.
MR. BRIAR: Yes, Mike Briar from Alfarma. I am having a little hard time figuring out just how this is going to fit in. But your number one assumption was that all of the resistance came from fluoroquinolone use in poultry. Am I correct about that?
DR. HOLLINGER: That is correct.
MR. BRIAR: And I think it is on page 313, you had a little footnote. And I assume that is based on this 1992 study that showed that there were no human isolates that were fluoroquinolone resistant.
DR. HOLLINGER: Right.
MR. BRIAR: I came across a paper that went from August 22nd, '92 to August 25, '95 which if memory serves me right was just before approval of serafloxicin in poultry from the Medical College of Wisconsin. And they had 12 percent resistance in their isolates as of the point just prior to the approval. I don't know how that figures in with your assumption that --
(Away from microphone.)
MR. : That was --- or that was ---?
MR. BRIAR: It doesn't say, but it is certainly not limited to --
DR. HOLLINGER: Right.
MS. : There is always that --- infections in travelers.
DR. HOLLINGER: We looked at domestically -- yes.
DR. BRIAR: It does not say anything about it.
DR. HOLLINGER: We looked at domestically acquired resistance. And our assumption was that everything was chicken-associated. And we removed the travelers and we removed prior fluoroquinolone use. And for those people who did not know or those cases for whom it wasn't known when they got fluoroquinolone.
As far as any prior fluoroquinolone resistance that was domestically acquired from food-borne sources in the United States, I am not aware of it from the data searches that we have done. But we would be very happy to have that paper and have a look at it and see if it changed --
MR. BRIAR: I don't know how this would figure in. I am just saying that it was rather striking that they did some rather extensive typing and they came up with 40 C. jejuni. And they had 12 percent of them already resistant prior to any use in food animals.
DR. HOLLINGER: We see this --
MR. BRIAR: It looks to me like in your model -- now, I may be wrong. Please correct me if I am. But it looks to me like your model that you would assume that any -- in other words, you are just taking the poultry percentage and applying that to the cases in the --
DR. HOLLINGER: What we did was we removed all the potential sources of resistance that would not be acquired from domestic sources. I believe in Canada also there is a hospital study where they show resistance in people and maybe someone here from Canada can speak up. But they do not use fluoroquinolones in food animals either. But a lot of these infections can be acquired from travelers returning from trips to places where they use fluoroquinolones in food animals.
MR. BRIAR: You know, it doesn't say in the paper, you know, whether that was travel-associated or not. It simply said that they had, you know, the 40 -- actually, there were quite a few more isolates, but 40 C. jejuni. And that was a little bit higher even than what we see from the NARMS data in poultry. That is what struck my --
DR. HOLLINGER: So, you know, this is a call for information. So please, you know, submit it. I would be very happy to look at it. Thank you.