Animal & Veterinary
Panel Discussions on CVM RA Model
Dr. Wesley Long
DR. LONG: Okay. Great. We are going to move on now. I would like for the panel members to come up. And I am going to talk about the rules for the panel discussion as they come to the stage.
We have had some really interesting information today. We have had the risk assessment presented to us. We have had a lot of really good questions that I am sure that CVM -- well, I think they have probably thought about most of these things and debated some of these things. It just shows that we have an intelligent audience that is able to draw out these issues that clearly may need further consideration.
The format for this panel discussion actually allows each panel member, I am going to give them about eight minutes to go through the seven questions that are posed that are on everyone's agenda under "Panel Discussion on CVM Risk Assessment Model."
After each person gets a turn to address those seven questions -- and let me just tell the panel members that you can just go through and tick them off. And that is what I am going to do and I am going to be very brief and I get to go first.
You can choose one of those points that is most important to you that you think really needs to be addressed. If you want to go outside of the questions, as well, you have that option. But I will be running the time clock and it will be right up here. And if you could keep your eye on it when it is your turn, we need to ensure that we stick within the time frame.
Following this then, there will be an opportunity for the public to address questions to the panel. Following that will be a public comment period for comments for the public. So with that, I am going to sit down and take my turn.
MS. : Wes?
DR. LONG: Yes?
MS. : Could you have them turn the lights on in our part?
DR. LONG: You bet.
MS. : Here in the back.
DR. LONG: I forgot to say that there are a few people here who have get to be introduced. So as it becomes their turn, I will go ahead and pick them off. Most of these people have been introduced as being prior speakers today.
Okay. My preference is just to tick off through these questions in a fairly quick fashion and give fairly simplistic answers to each one. The first question, what are the positive aspects of the model. And not to be facetious, but I think one of the great positive aspects is that it is done and it is out in the public and it is here to simulate -- stimulate -- not simulate, we are done simulating for right now -- to stimulate discussion amongst all of you, amongst risk assessment peers and scientists, and to try to get -- you know, it adds to the limited pool of these types of assessments that we have on these types of products.
I personally don't have trouble with the assumptions that were made. And I will get back to that as I work my way down the list. I forgot to start the clock on myself.
Okay. Limitations of the model, I guess, you know, this model is not going to ever be everything to everyone. And certainly the model does its best to address the question that was put before the risk assessors and certain it is what I think was necessary for CVM to move forward.
So as we saw in the examples, we saw some pathway analysis models, we saw a qualitative risk assessment model. Dr. McEwen showed us some pathway-related type modeling which I think is very useful and plays a role. But it wasn't the subject of this exercise.
In terms of significant data gaps, I will take the easy way out and say that Kathy Hollinger seemed to have covered them pretty well. And, yes, there are data gaps. But, no, I don't think those data gaps are of -- certainly, we can fill in the FoodNet data over time. We can collect better information and data. But I don't think it should stop us from using this assessment.
What aspects would I consider changing, I think I am going to defer on that question. Can this model be used to help CVM or the industry reduce the level of risk, that is sort of a -- you can answer that question in a lot of different ways. I guess directly, it is not going to help industry reduce the risk.
There is no mention of interventions that you can use to control the levels. There is -- it is not intended to be a recipe or a HACCP-type thing for you to insert controls in the appropriate point to achieve an appropriate level of reduction.
But what it does do is it gives CVM now the tool to work on this risk management decision which I think is really where the next step is in this process, that we now have a an estimate of the risk to human health. And, you know, as I listened to the comments today, this -- and as David said, somewhere within his range he has got the right number.
And I think I agree with him. But I think that fine-tuning that is always going to be a goal that we will have and we will continue to re-evaluate. But it shouldn't stop us from moving forward.
Should CVM evaluate other antimicrobial-pathogen combinations, I think absolutely they should. You know, the reason they told us they chose this one is because they had the best data. And, of course, that is critical to the assessment. But in terms of a comparative risk assessment so that they can prioritize their resources the best, I -- because I am not in the veterinary field, I don't know if Campy and fluoroquinolones are the number one issue or if perhaps it is another organism-drug combination.
And as far as alternative approaches, I think that this farm-to-fork approach which we have heard talked about a number of times today is a valid follow-up to this. And I think that it is going to require significant industry involvement to take on a farm-to-fork approach. And perhaps industry should take the lead in that type of approach. And that is my answer to the seven questions.
Paula is up next. And I don't think Paula got introduced properly. Did you before?
DR. FEDORKA-CRAY: Yes.
DR. LONG: Okay. Take it away.
DR. FEDORKA-CRAY: Do I need an introduction?
DR. LONG: You need no introduction. Here you go.
Dr. Paula Fedorka-Cray
DR. PAULA FEDORKA-CRAY: I will take Wes' extra four minutes. I think there are positive aspects to the model. And I will go with 1 and 2 and what do I see as the limitations of the model. And I think the answer to both is probably the research gaps. The positive aspect is is it identifies the research gaps that we can all focus on now to make better assessments as time goes on.
And I think that because we have an idea of some of the thought processes that have been identified here, that we will think about future risk assessments in a different way and begin to gather information. And I will address that as we go further on.
I also think that there was an assumption made that fluoroquinolone use in poultry is going to result in an adverse human health impact. And that is probably one of the largest and the most contentious issue here that we can have from both sides.
And I think that if you look at it from different perspectives and if we all step back and perhaps look at it in a more objective way, that we can see that really those questions -- that question could probably be answered in a number of different ways. And I think that we are all calling upon ourselves though to provide as much information as we possibly can to make the correct assessment as time is going on.
In my opinion, not only for job security -- I always say that, but it always comes more and more evident. But there are some significant data gaps that have been addressed. One of the things that I am particularly struck with is that obviously I missed it in the literature that humans -- someone said that humans can't become reinfected with Campylobacter. And I think my body missed that lesson at some point in time.
And that brings to mind one of the most intriguing questions, at least in my limited capacity up here, is to think that we are in fact exposing ourselves to multiple isolates that results in increased disease so that we may be immune to one different type of isolate that keeps changing over time.
This may lead then to the tremendous genetic diversity that a lot of people will talk about in the Campylobacter species and really confounds what we may be able to do in the long term then if we can't control that in some other way. And so from a research aspect, that may require us to look at this question very differently than we have in the past.
And it also suggests that there may be increased stresses in our immune system so that while we are busy containing one particular isolate that may be more virulent, other isolates have the potential to take over and cause the disease that we may then see. And this may be way we have this disparity between coli and jejuni.
The other thing that I think that I brought out was the difference in culture methodologies and the selection of isolates over time and how that may change, how that is different. I mean, one of the things that we are addressing is how it is different on the farm versus the slaughter plant versus retail. And I think those will become very critical issues at some point in time.
And I think that we really can't discount an environmental impact. Because of the global nature and the ubiquity of this organism itself, that really if we have already tipped the balance and if we already have an environment that is saturated with some type of bacteria, that there may only be X amount that we can do to, in fact, lower the graph, if you will. And then that begs again for interruption of the system in different ways.
I think that we are gaining some evidence that there is going to be increased -- that with increased resistance, there is an increased likelihood of colonization with prolonged shedding. And that is a virulence that speaks to pathogenesis. And I think that we have to go and we have to look along those lines.
I am wondering if we can't just do something very simple by suggesting that if we know that there was fluoroquinolone use on any farm at any one particular time, if they can't be slaughtered last in the queue in a slaughter facility and see how that may or may not affect what goes on in a processing situation.
And maybe that is not totally feasible and there are all kinds of implications for that. But that speaks to somewhat the Danish system with the Salmonella and slaughtering animals last after they have a known serologic change.
One of the things that has nothing to do with research but is a way that all of you can probably influence the process is we tend to only publish positive results. If we see something, we publish it. Well, what about the negative results? You know, what about the times that we know there is no impact whatsoever and it really never gets out there because a journal only wants to publish positive results?
And so we have this gap then and people saying, you know, you have to go to a meeting and actually ask a lot of questions and then have them say, "Oh, no, well, we've done that. Don't bother going there", or something. And so it would require tremendous change on a lot of different levels to actually have information published that would suggest that, in fact, something that wasn't observed is just as important as something that could be observed.
And then one of the things I think that we should do is look at -- in fact, look more closely at the role of Campy coli and some of the other Campylobacters and see if we haven't missed something. And bacteria have a unique way of out-foxing us no matter what we think.
And I think that if we look at the number of papers and information that has been published over time, that because we haven't really solved any one problem in its entirety as far as bacterial species go, I think that it speaks that we always need to look at things differently if we keep coming up with the same answer which is, "I don't know why this is happening."
And then one of the things that I think it might -- we might beg to ask is how many people have had repeat infections and what probability that is over time. And if we could look at a small number of people and look at the isolates that they may be shedding, that may give us some idea of what has happened to this population dynamics.
What aspects of the model would I consider changing? I am always intrigued by having one risk creating a second risk and if there is a probability of that. So that if you take away the use of, say, an antimicrobic at some level, do we create a second risk by allowing for an increase in some disease, whether it be bacterial or viral, and then where that will ultimately lead with exposure to an effect on public health.
And I think that that is something that we should at least consider because we are trending new ground here. And we really don't have a good answer for that. And I think we would be remiss if we didn't at least think about it. Maybe you are thinking that's plenty; we've thought about it enough. But I think it is something we should think about.
Can this model be used to help CVM or the industry reduce the level of risk? I think any time you have information, that can be useful. And if we are all walking away, there is always something positive. My mother has always told me there is a positive aspect to everything in life. So I think we should -- I will tell my mom that it works this way, too.
How should CVM evaluate other antimicrobial-pathogen combinations? I think it would be -- I think that I don't really know the answer to that. I think it would be good to have other risk assessments done. One thing that would be nice would be to have enough of a lead time and enough of an idea perhaps of what may or may not be going on so that the submission of data can come from a large number of sources that may be able to provide additional information that can be used in the risk assessment model.
We all have access to information and data that other people may not have whether that be published or unpublished works. And I think that having the opportunity to provide that, whether it ends up being used or not, is not a final call. But I think having the opportunity to provide it provides for more interactive processes and may also provide for more useful information.
And I also think that in some ways if it wasn't cost prohibitive that it would be nice to see a model done where we already know a lot of the data and we have a very, very good idea of what the expected outcome is going to be. I just -- you know, numbers and calculators and punching things in are all very nice.
But I think there would be some merit to seeing something like that and that it would give a much higher level of confidence in the thought process in seeing how everything is going on with the bacterial-drug combination. I know that there are other risk assessments that have been done. But we are looking at something more specific here.
And an alternative to risk assessment approaches that CVM should consider, I don't necessarily know now that we have gotten into it that anyone is ever going to get out of this. And I think it becomes the new rage and the thing to do. It's like Pokemon. And, you know, we have Pokemon now and we have risk assessments now. And we will be going on to some other things, too.
But I really do think that we should not lose sight of the fact that what we are really talking about here is reducing pathogens. If we reduce pathogens, then it follows -- at least in my assessment, it would follow that we reduce the percentage of resistant pathogens, too, or that would be a good starting point.
And I think that we have to keep sight of that fact and we have to keep working toward the goal of, in fact, reducing those types of pathogens regardless of whether they are resistant or not. And prudent use becomes absolutely critical in this for all of our constituents.
And then I think that we really have to look at implementation of alternatives. Since the issue isn't going to go away and if some of these other assumptions are true, then this begs -- we can develop a vaccine. If immunity happens once, I would be over -- all -- you know, that low dose, avirulent, get it over with one day. You know, it is like a flu shot. And then, of course, probiotics and other issues, too.
And I think that we really have to look at the implementation, actually put them into practice now and see if we can use something else while we are trying to fix this issue, too. Thanks.
DR. LONG: Scott?
Dr. Scott McEwen
DR. McEWEN: Yes, you already heard from me, so I won't sort of reiterate too much stuff. I think -- what are the positive aspects of the model, I think first and foremost has been said. It is done. It looks to me like an excellent job, so compliments to the group on the whole sort of process. Again, I underscore that we have the model. In fact, there is a public meeting on it and there is wide-ranging discussion and there is people from all different fields that can sort of take punches on it and add to it. I think that is terrific.
I think it is very important that regulatory agencies with the kind of stature of FDA does this kind of thing because I think that sends out a great signal to a lot of other places. You know, sitting at a university, I think this is going to have reverberations in our graduate training program and of people that are going to have to develop the skills to sort of get involved in this kind of thing which I think is excellent.
Lester mentioned that the teaching value of this sort of exercise. So it has I think a lot more impact than the specific topic and issue, Campylobacter and fluoroquinolone resistance, though I think a lot of people in the room would probably -- we all have our own interests and I think that is a major one for me.
I think the -- another positive aspect that hasn't really come out is my understanding is that in a lot of public applications of risk assessment, if there is uncertainty and default assumptions are made, those are usually to favor public health. And there is good reasons for doing that.
If we don't really know how it is working, then sometimes we make worse case assumption. And then the onus is on other people to go out and get more data to show that that is not the case and we should redefine that.
I think FDA seems to have done a good job of balancing that, not sort of gone overboard on that particular aspect I think is a positive thing. And as people have said, the explicit assumption description, sensitivity analysis, I think all that -- transparency although I don't like the word, I think that inspires a lot of confidence in the process and helps with growth and people understanding it and that kind of thing. So I think -- and there are other positive aspects, as well.
What are the limitations? Again, we talked about the what I would say ecologic nature. There is probably a better word for it than that. And I think that -- although I understand why it was set up that way and I think there are good reasons for it, I think I would sort of like to see the effects of maybe refining some of the parameter. I am not an expert in sort of parameter estimation.
But just as one example, there was the -- we saw in Dave's literature how the effect that the -- the prevalence of fluoroquinolone resistance in slaughterhouse isolates had, how important that was. And yet I think as I kind of read it, it looked like what the -- the way the data were collected suggested that the standard errors might be under-estimated based on the sort of possibility of clustering at sort of slaughterhouse levels.
And, again, I don't really know. I suspect, as Dr. Cox said, that some of these things don't have much impact on the sort of outcome, but it would be good to sort of evaluate that.
I think the assumption that an infected -- person infected with a resistant organism in his treatment -- corresponds to the treatment failures, well, a reasonable one I guess I wonder about that. And I would like to hear about clinical experience in that area.
Do you feel there were significant data gaps, I think everybody would like to see more direct evidence if you want to call it that or evidence of drug use in these animal populations and resistant selection and so on. I think we need more of that. Whether it takes the form of this kind of assessment, I don't know. It could be -- there are lots of other approaches to addressing that as I said.
What aspects would you consider changing, one thing I don't know a lot about but I am interested in is separating out the variability and uncertainty components. I think we sort of talk about those things as equivalent. And in some cases, there are pragmatic reasons for doing that. But I think it would be good for people to have an idea of how much of the influence on the outcome is due to how uncertain the parameters are versus their variability biologically and other ways.
Can this model be used to help CVM and other industries to reduce risk, yes. If it is -- and as I see it, it hasn't yet been used to look at the effects of interventions and test hypotheses. But I think that would be a great thing to do.
I guess we -- to be strictly speaking, as Steve said, you have to define what is an acceptable level of risk. And that hasn't been defined yet. So you could make the argument that the level of disease out there is acceptable. I personally don't believe that, but that could be stated. In that case, then there is -- under that scenario, there might not be a need to reduce risk. But, again, I think that is not true.
As always, I would like to see some economic assessments used in conjunction with evaluating different risk management strategies to see what kind of collateral damage might be done, if you will, in other industries and sort of weigh that in the equation somehow.
How should CVM evaluate other antimicrobial-pathogen combination? Again, I ran out of time, had too many slides and had some there to sort of reinforce what Louise was talking about on the qualitative assessments. I think that given the large number of pathogen-drug combinations, I think it is unlikely we will be doing full blown quantitative assessments on all of them.
And I think -- quantitative that is. And so I think qualitative assessments are going to be important. And we need to have better ways of doing that, more structured ways of doing it. And I think that will move along.
I think, again, that there is a merit in having what we called in a previous talk a tiered approach to this. We have a sort of screening level of qualitative assessment. It looks like there is no problem. We don't need to go any further. And as the ante is up for a variety of public health or cost reasons, then we could start to engage in more quantitative assessments.
And I believe that has worked in other fields. I think it could work here, as well. I think we have to move into ways of assessing the quality of information, scientific information that the GENACAR Report from Australia gets into this quite a bit. The weight of evidence approach I guess, the evidence-based medicine approach of somehow weighing how well a study was done, how representative it is. And I get a sense of how believable it is into the equation. So with that, I will pass on to Louise.
Dr. Louise Kelly
DR. KELLY: I think I need a cushion this time. I can't sit on there. It is too uncomfortable. Anyway, well, to start off, what are the positive aspects? Well, just for David and my best Glasgowegian accent, I think it is pure dead brilliant. I really do. I think it is a great attempt at considering this problem that we have been looking at ourselves in the Veterinary Laboratories Agency.
And it is not an easy task. It is a difficult -- risk assessment, development of these models is not simple. A lot of people think that it is. It is a difficult task. And I think you have done a really great job.
And all the way through reading this report, I think it has been completely transparent. Everything has been laid out. All the assumptions are laid out. And from this, we could then if you want to take it back and try and reproduce the results yourself.
And I think the transparency is always put down as one of the most important, crucial elements of a good risk assessment. And I think this falls through throughout the whole report.
In addition to this, I feel that there has been a real team effort involved in the development of this model. It has been multi-disciplined really. It is not just a mathematician who is sitting in an office developing the model. There has been input from every possible background. And, again, I think that is very important. So I really do think this model has a lot of positive aspects.
The limitations of the model I think really depends on what perspective you are coming from and what particular question you are trying to address. And in this aspect, I think the model has addressed the question it was asked to address. And it has done that very well. So from that point of view, I really don't see that there are many limitations.
Obviously, if we were considering estimating this risk from another perspective, for example, considering control on the farm level, then that would be a different risk question that we would be trying to address and a different type of model would be required for that type of problem.
So really it comes down to defining your question in the first instance. And that has been done here and followed through. And I think, therefore, for that particular question, the limitations are really limited in themselves.
Significant data gaps, well, it has been acknowledged that there are data gaps within this risk assessment. But they have been laid out in the report and they have been accounted for by adequate uncertainty assumptions. And I think I am right in saying that the separation of variability and uncertainty has been undertaking in the model to a large extent.
I think you are nodding, David? Yes. Because essentially what the final outcome for Stage 3 was nominal expected value. And that itself is described by a variability distribution Poisson process. So that has been accounted for.
Aspects of the model that I would consider changing, none really, again, for this particular question. But, again, depending on a different perspective and a different type of question, maybe a different approach would have to be accounted for.
And, again, as Scott mentioned, I think the idea of integrating risk assessment models with economic analysis is a very good idea because these drugs do have benefits both to the animal and to the human. And we have to consider that. To a benefit-cost analysis, benefit-risk analysis if you like would be another good way to go.
Can this model be used to help manage the risk? Well, I think that we have to remember that risk assessment models, and this one included, are dynamic tools and they are tools. The aim is not to concentrate on the final numbers that come out of these models. They have to be appreciated as being tools which need to be updated as new information becomes available.
And the estimates that come out of this model are really based on 1998 data which can be updated. And, therefore, it is dynamic and can respond to monitoring information which needs to be undertaken at the same time. So I think it can be used as a regulatory tool.
Other antimicrobial-pathogen combinations, yes, I think that that should be undertaken. But I think we have to pay cross-consideration to the type of modeling approach that we might need to use for these different combinations. It would be a danger to consider that this model developed from Campylobacter and fluoroquinolones could be used for any other species and drug combinations.
Each problem, each combination in this way has different aspects, different processes that need to be considered. And, therefore, the thought process has to begin again for the different combinations. So we need to remember that we can't just simply present new figures for new bugs and for new drug combinations. We need to think again about the whole process.
And the alternative risk approaches, well, we presented today -- well, I presented today our farm-to-fork type approach. That is another method that can be used. But, again, we have to consider what exact risk question we are trying to address and really the available data that we can use to fit within a model. And it all depends on the problem that we are trying to consider for our risk assessment. And with that, I will now step down again.
DR. LONG: Steve?
Dr. Steve Anderson
DR. ANDERSON: All right. Well, I think -- first of all, I think the CVM team needs to be congratulated, as well, because I think it is a very good product that they have generated. They have used the sort of quantitative methodology and they have supplied the actual spreadsheet which I think is great because, again, I will echo everybody else's sentiments, is that we now have the spreadsheet.
And you can take that. And it provides that transparency. You can take that and work with that on your own and see how you agree with the model, as well. So I think there is the transparency component that is an excellent part of it.
The model makes full and complete use of the available data. The surveillance data and the monitoring data and the CDC data, it kind of brings all of those things together and ties those together very well.
The study also recognizes the uncertainty. And I think that is a reasonable thing, especially when the risk managers take this and start working with the actual risk assessment product. It will be a good -- good to recognize that there is uncertainty in the values generated.
The limitations that I would say that I see are I would like to see probably the pathogen load or the concentration on the carcass considered. And I think that is really important. In our model, we can take pathogen load, hold it constant and raise the prevalence. And you will get an increase in the incidence of disease or illness.
The same thing, you can hold prevalence constant, raise the concentration or the pathogen load by the same amount, and you will get similarly increases in the number of illnesses. So I think those two things actually work together. I don't think you should actually exclude one or the other. And I also think those to things, the pathogen load and the prevalence, they kind of work together in that ultimate dose.
The other problem or limitation that I see is in the market data was used to give the final consumption amount of poultry which was 50 some-odd pounds. And I think that you could use the consumer survey for food intake data set or the NHANES data set that actually tracks consumption patterns and get a little bit better handle on the actual consumption data because market data is going to over-estimate what people consume because that includes wastage as well as what is actually consumed. So actual consumption data is what is really needed.
The other limitation I see is it doesn't really provide many options since it is such a simple model for interventions. And the ultimate intervention it seems that I can think of would be controlling or banning the use of the product. So having a more complex model may be more difficult. But you also have the increasing opportunities to -- for mitigations and suggesting mitigations.
The data gaps I think have been covered adequate. The things that I would consider changing, again, I would really strongly urge that sort of the concentration or the pathogen load be added. And that can be derived from the USDA baseline data where the prevalence of Campylobacter was originally derived on the carcass. Again, I would use the consumer data, as well.
And then the next question was can this model be used to help CVM or industry reduce the level of risk. And I would say it is a good start. I would suggest maybe another year of data. But we are already at the end of '99. So I presume that the '99 data can be entered into it, as well. And then I think it is a useful tool. I think it is a simple model. But that may be the nice thing about it. It contributes to the understanding of how those figures were derived.
Now, how should CVM evaluate other pathogen and antimicrobial combinations, and I think that is a case-by-case basis. I think in future risk assessments, you need to consider other sort of pathogen-specific things like growth and how the drug is administered. In poultry, it may be administered quite differently, in water. In cattle, it may be injected. Those are significant. How the resistance is acquired and spread also is important.
I think of this as more of a horizontal risk assessment in many ways because it is a very simple model. And perhaps doing a more farm-to-fork process model -- process-based model might be useful.
Alternative risk assessment approaches, again, would be a farm-to-fork model. The other possibility is the Canadians are also finishing a Campylobacter risk assessment study. And I think you could put the resistance data into -- and the resistance trends into that risk assessment and also sort of better derive what the relationship is between the animal prevalence of resistance and then how that relates to the incidence of fluoroquinolone-resistant illness. And I will stop there.
DR. LONG: Thanks, Steve. Dr. Lipsitch is next. He is Assistant Professor of Epidemiology at the Harvard School of Public Health -- oh?
DR. LIPSITCH: Randy is in between us. I don't --
DR. SINGER: It doesn't matter.
DR. LIPSITCH: I have some slides. So I can put them on while you --
DR. LONG: Oh, okay. Randy is up next, okay. Well, Randy, even though the program -- or even though his badge says he is in California is -- maybe he wishes he was in California at this time of year. But he is Assistant Professor of Epidemiology at University of Illinois at Champagne-Urbana.
Dr. Randy Singer
DR. SINGER: Okay. Well, I would like to thank CVM for inviting me to participate. Rather than walking through this list of points, I see a lot of them as interrelated. So I would like to discuss the relevant points, you know, together.
Well, first there is the question of what are the positive aspects. And I think that does play directly into what are the negative aspects. The positive aspects, like has been said, is that the process is started. And, you know, this is a great first start at -- it has outlined some important areas for further research.
But it plays directly into what is really in my mind an important negative that we need to be careful about. And that is to reiterate something that Doug Powell said this morning about risk communication. I really truly believe that the public does not understand risk. They don't understand probabilities. But when they know that FDA, CVM and a group of experts are talking about a risk assessment model, they are going to hear the words, "chicken", "antimicrobial resistance", "resistant bacteria", and "risk."
They are not going to ask, "What is my risk or what is the probability?" They just hear the buzzwords. And to them, a product gets singled out as a risky product, especially with the media play today that we see with antimicrobial products.
So the risk communication aspect doesn't take place just between us. It doesn't take place when the model is finished. I think there is going to need to be some careful consideration of how just our convening here is related to the public and so that an unfair negative impact isn't seen in a singled-out industry.
The next I guess questions that interrelate are how to use this model and what are some of the data gaps. I see some of these coming together. Well, one of the purpose of designing a risk model -- one of the tools is a hypothesis-generating tool. Another might be that it outlines key areas where we need more data. It might help us establish some thresholds.
But one of the key ideas in my read of this model is we want to outline -- well, that is quick. That's all I get because I am an assistant professor.
DR. LONG: Go ahead. Finish.
DR. SINGER: In the -- in a risk assessment model, you often want to identify foci for risk reduction strategies. And I understand that this is at the point of being called a simple model.
In reading through it, at the very end of Section 5 if you all had a chance to make it that far, there is this variable thrown out called Pmax which is defined as the maximum acceptable prevalence of fluoroquinolone-resistant Campylobacter on chickens. And it is suggested that this might be the threshold that we set.
Perhaps in a processing plant, if Pmax is overshot, then something has to happen. Maybe fluoroquinolones are pulled or something. So this is a possible risk reduction site. The concern I have is that many broiler producers currently don't use fluoroquinolones. So if you are in their processing plant you find that they have overshot this Pmax, well, then what do we do?
And that brings up in my mind kind of a disconnect of the model. Where we want to invoke a risk reduction strategy is at the farm level. We are not interested in telling people when they get to the people -- well, maybe we are, but what antibiotic they should receive or -- we are more concerned about how do we manage it on the farm. And yet the farm component is completely absent from the model.
So while I do understand that it was meant at this point to look at the risk through consumption, to me if it is really going to have the risk of risk reduction, we need to include the farm component and get a better understanding of the relationship between fluoroquinolone usage, the development of fluoroquinolone resistance and then that transfer mechanism as it might occur to humans.
So that addresses this data gap. And then how might we manage it? At this point, I don't see the model as being so useful except in identifying key areas where we need to collect more data.
Another issue I would like to bring up -- and this is maybe being an epidemiologist, thinking of causal inference all the time -- is how we need to really be careful how we assume the causal nature. Some have said today that we aren't assuming any causality. And some have said, well, we are definitely assuming causality. We are assuming that the fluoroquinolone resistance we see in chickens, that was Campylobacter, are the fluoroquinolone-resistant Campylobacter that we see in humans.
So we are not -- we don't seem to all agree on whether or not this is a causal connection. The problem I have is with the methods that we might even use to assess causality. I have been looking recently at some of these molecular epidemiologic techniques from their actual methodologic aspect; I mean, how we actually use them.
And what you find is this difficult situation where if two isolates are different, then they are probably different. But if the isolates have the same fingerprint, what can you say? It would be nice to say that they are identical and so that is the source. But one of the other explanations is that you just don't have enough resolution.
To pull out in my read of the New England Journal of Medicine study from Minnesota, while, yes, they found identical DNA fingerprints in the Campylobacter -- in the resistant Campylobacter in humans and the resistant Campylobacter on domestic chickens, they also found some of those same fingerprints in the resistant fingerprints from internationally acquired infections.
So if we can't even -- we don't have the resolution to do a trace-back within our own country for domestically acquired cases, I think it is difficult to assign this causal link. And I am just trying to -- again, it is -- it needs to be done cautiously so that we don't incriminate any single producer or single industry, etcetera. How much time do I have?
DR. LONG: I'll give you two more minutes.
DR. SINGER: Okay. Well, one of the other concerns I have -- and maybe this is just my own personal thing. Maybe most of the other statisticians, mathematicians here wouldn't agree, is that in my background of a Bayesian analysis, the purpose of that prior probability is to take into account the expert opinion, to take into account our uncertainty coming into the problem.
But the way this model has been written is that every prior distribution was modeled as a uniform 0-1 which converts to a beta 1-1 which for those of you who don't know anything about probability distributions means that it has very little weight. So if there is a lot of data that were collected, those get weighted very heavily and the prior means nothing.
So what that means is that the entire uncertainty in the model in my mind is coming from a statistical uncertainty generated by that beta distribution. It does not allow us to account for biological uncertainty, nor does it allow us to account for differences between the various studies that were interconnected in this model.
In a meta-analysis which is typically where you would take many different studies and try to reach some end product, you account for the different study designs by weighing them differently and by adding uncertainty factors.
And so my concern is that we haven't done enough
-- it might not ultimately make a difference at all in these probabilities. But without yet having had a chance to really go through the model in detail, I am concerned that there isn't enough uncertainty in the model inputs.
And so the comment was made that they were impressed that there is very little uncertainty in the model outputs. And that to me says that, well, that's obvious. There might not have been enough uncertainty in the model inputs. So I would -- as we continue to develop this process would just like to explore more the use of expert opinion and uncertainty into the model.
DR. LONG: Thank you. Okay. Where did he go? There he is. Okay. Now, this is Mark Lipstich -- is that right?
DR. LIPSITCH: Lipsitch.
DR. LONG: Lipsitch. He is Assistant Professor of Epidemiology at Harvard School of Public Health. And his research uses mathematical models to study the transmission dynamics of infectious diseases.
Dr. Mark Lipsitch
DR. LIPSITCH: Thank you. Thanks for the invitation to come here. I have a few comments that are on sort of various topics. I think they are responsive to the questions, but I haven't really tried to key them to the questions.
So I will briefly talk about the strengths of the model as I see them. And then talk a little about the limitations and then the question of setting thresholds and responding.
I would like to start by commending FDA on the -- I am not going to -- and Dr. Vose on really a very impressive model. And I think if anything, the concern is that we may be spoiled by having such a nice model for a system where there is so much data. I mean there are certainly significant gaps.
But I think my strongest point today is going to be that we can't know everything and that this may, in fact, be really as good as we are going to do for any pathogen that might be of interest. Having said that, I have four brief comments about limitations of the model.
The first of those is that the report makes very specific predictions about the number of excess fluoroquinolone-resistant Campylobacter cases, the result from use of fluoroquinolones in chickens. What is not totally clear is the toll of these additional infections on human health and welfare, although there was some discussion of that today in terms of additional days of disease.
It would be possible to make such estimates using those sorts of data on differences in the duration of disease. And it would also be important to consider separately the impact of resistance and potential treatment failure on the rare or a bit more severe cases of invasive disease.
And finally, also to consider the effect of resistance on the duration of symptoms in untreated patients, as there has been some suggestion that even in the absence of treatment, resistant isolates may cause worse disease. And the last point here is we might want to know how these effects are different in different sub-groups of the U.S. population that might be at elevated risk such as immunocompromised people.
The second limitation is that the model is really a static model. And the flip side of that is it is an easily updated model. So -- but I think it will be important to consider how these estimates are changing over time as the number of resistant isolates possibly increases. And we heard discussion earlier from the Minnesota data that the prevalence of resistance in Campylobacter appears to be increasing already this year over last.
The third point is on the pathogen burden which has been mentioned a little bit today. But I think it is important to emphasize that the model assumes that the human health impact of fluoroquinolone use in chickens is the increased likelihood of exposure to resistant Campylobacter. But it doesn't consider another issue which is mainly the effect of fluoroquinolones on pathogen load.
So if, for example, fluoroquinolone use substantially increased the load of Campylobacter or other pathogens in chickens, then that would increase the risk of Campylobacteriosis or other disease for an individual consuming that chicken. And that would be an additional impact that just isn't factored into the model, but which could be I think fairly easily obtained if one had data on changes in pathogen load following treatment.
And the last point of the limitations that I want to make is it is very important to remember that although the harms of Campylobacter are probably the most readily quantified, they are not the only ones. And they may not be even the most important human health consequences of fluoroquinolone use. And this is not so much a limitation of the risk assessment as a concern that it should be viewed in the proper context.
And this goes back to what I was saying about not being spoiled by the high quality of the data on this topic. Non-typhoidal Salmonella infections, for example, account for almost ten percent of food-borne illnesses, less than Campylobacter. But one-quarter of all hospitalizations for food-borne pathogens and almost a third of all deaths, about 553 per year in this country according to the CDC, and that is more than five times the number caused by Campylobacter.
And high level fluoroquinolone resistance remains rare in food-borne Salmonella in this country, but lower susceptibility reflected in increased MICs is being observed in Salmonella in the U.S. and other countries that use fluoroquinolones in poultry. And this trend appears to be worsening.
These Salmonella with reduced susceptibility are frequently only one mutation away from full resistance to quinolones and that makes them an ideal substrate for the development of higher level resistance, either upon further veterinary exposure or in humans who are treated.
Now, scenarios like that are undoubtedly harder to quantify precisely than the immediate problem of resistance in Campylobacter. And it was very sensible to start with what is most easily quantified. However, we know that each of those steps in the scenarios is possible and that once fully resistant Salmonella appear in our flocks, it may be at a considerable selective advantage, although that is an area where we certainly need more data.
The fact that we don't yet have a noticeable clinical problem shouldn't make us conclude that we can wait until the clinical problem because obvious. And I put up this data from vancomycin-resistant enterococcus just to make the point that I think there is a relevant parallel.
If you look at enterococcal use -- sorry, vancomycin use in human medicine starting from the '70s, you had almost -- well, well over a decade of use of the agent before resistance became a problem. And that may be sort of like the stage of which we are now in, something like Salmonella.
But what is important to see is that once it appeared, it increased very, very rapidly. And it has been hard to get rid of. And so my point here is simply that this is not the same pathogen. There are a lot of potential differences. But that focusing on where there is a big problem and a quantifiable problem shouldn't distract us from what could be later on a greater problem.
So, finally, I want to comment on the question of how risks can be reduced and in particular, on what might be done if the level of risk were judged to have reached a level that is unacceptable, that Pmax I believe. I must say that having read the section of the draft report on establishing regulatory thresholds four or five times, I still don't understand the solution that is being proposed. But I think that what is being proposed is that when a level of human health impact is judged unacceptable, the Agency would take some mitigating action which I suppose would mean restricting some use of fluoroquinolones. And the problem with that approach is that resistance is not something that can be simply switched off by curtailing the use of a drug. And this is the point that Dr. McEwen made a little earlier.
Once we reach a level of human health impact that is judged unacceptable in either -- in any pathogen, even if we recognize it right away and take very strong action, we might continue to have the resistance problem for some years following that intervention.
As far as I know, there are no data that addresses what happens in Campylobacter following a reduction in use of fluoroquinolones. But we have some reasonable parallels in -- potential parallels in human infections. And I just wanted to show one example from what is really universally the success story that everyone cites for why we should reduce antibiotic use in human infections.
And the orange line shows the reduction in erythromycin use in Finland from a level of three, about a six-fold reduction. And while this is cited as a great success story, what you see is that following that reduction, we have several years of continuing increase in resistance and then a decrease. And the decrease was about two-fold in about five years.
And so the -- to summarize, the reduction -- when you take mitigating action, the reduction can be delayed and it can be slow. And so when thinking about thresholds in a risk mitigation context, it is very important to realize that the threshold has to be set below what is unacceptable because you can't simply switch things off. And I will stop there.
DR. LONG: Thank you, Mark. Okay. The final word comes from Dr. David Bell. He is an Assistant to the Director for Antimicrobial Resistant at the National Center for Infectious Diseases at CDC. He is a specialist in pediatric infectious diseases in public health.
David Bell, M.D.
DR. BELL: Thank you. I am going to be able to shorten my remarks because I agree with virtually everything that Dr. Lipsitch has just said. CDC commends the FDA and Dr. Vose on developing this model. CDC believes that it reflects the available data well and we agree with the overall approach and the overall conclusions.
This analysis provides additional insight into the harm that fluoroquinolone use in poultry is currently causing to humans in the United States. The model is also a useful step for assessing what impact could result from more serious fluoroquinolone-resistant infections such as Salmonella.
We have some suggestions for minor adjustments that we will provide as follow-up. But one of them, for example, is to consider the harm caused to the 135,000 people who are estimated to acquire infections with Campylobacter and not be treated with antibiotics. And this refers to the increased length of illness, that we have emerging data to assess.
We would like to see if the model could be used more predictively to get some idea of the consequences for the future if current trends continue. In terms of fluoroquinolone resistance and Campylobacter in humans, it is increasing approximately two percent per year.
Fluoroquinolone use in humans is also increasing. And the impact of these two trends may also need to be considered. For example, the rate of fluoroquinolone treatment of 55 percent of the cases is probably going to rise. I want to connect that with a little clinical insight, if you will.
Fluoroquinolones are by far the best drug for the empiric treatment of bacterial gastroenteritis and its complications. This drug is oral. It is safe. It covers the spectrum of likely pathogens. And it really is the drug that all of us would want presenting with an infection that was thought to be a bacterial gastroenteritis or its complications.
The drug has not to date been used in children. That is primarily because in baby rabbits, it causes cartilage damage. However, the evolving collective thought in the pediatric infectious disease community based on studies and increased experience in certain unusual situations in which it has been given to children is that this is an effect in rabbits only, not in children particularly in short courses.
And it is quite possible that fluoroquinolone use will continue in adults and will begin in children. And so I think that I would just offer this to the modelers as something to consider in assessing the -- using this model, the trends that can be expected if no action is taken to mitigate the current hazard. Thank you.
DR. LONG: Thank you, Dr. Bell. Okay. I am going to stand up again. What I want to do is get an assessment of how many of you might want to have comments during the public comment period so that we can gage how much time we have for questions of the panel. Can I see a show of hands, who is planning on commenting during the public comment period?
(Show of hands.)
I see one. I see David is going to comment then, three, four, five. Okay. Okay. So I think what we can probably do if we were to limit those comments to about three minutes, then allowing a few more people might stand up, we can spend at least 15 minutes here if there are questions to address to the panel. You can step up to the microphone at this time.
(Away from microphone.)
DR. VOSE: --- have a point ---
DR. LONG: That would be great if David would clear up some points. Yes.