• Decrease font size
  • Return font size to normal
  • Increase font size
U.S. Department of Health and Human Services

Animal & Veterinary

  • Print
  • Share
  • E-mail

EMEA Risk Assessment

Dr. Louise Kelly

DR. KELLY: I just achieved being more than four-foot, ten. I have reached five. So I hope you can all see me now. And thank you very much for inviting me here today to take part in this workshop. And what I am going to present to you today is the work that has been done by ourselves at the Veterinary Laboratories Agency in the U.K. for the European Medicines Evaluation Agency.

And this work was done by my boss at the Veterinary Laboratories Agency, Dr. Marian Wildridge. And I am going to present her work to you today.

(Slide.)

So the main focus of this study was to look at one particular problem in relation to antibiotic resistance. And that was to look at problems associated with Salmonella typhimurium and with fluoroquinolone-quinolone class of antimicrobials. So in this particular study, we had to look at one particular organism and one particular class of drugs.

(Slide.)

The study was based on a two-week period. So it was a very short study that was undertaken. And the impetus for this work arose as a result of a vast amount of data that had been collected by the EMEA. So the first aim of our study was to evaluate that data that had been collected. So the study was based on that data and that data only. No other information was collected from any other source.

(Slide.)

Following this evaluation, the second aim of the study was then to extract from this data any data inputs that would be relevant for a risk assessment and in particular, a risk assessment for Salmonella typhimurium and fluoroquinolones.

(Slide.)

Then the third and the main aim then was to present these major inputs, extract it from the way that the data was presented to us in the form that would be relevant for a qualitative risk assessment. So we are talking in terms of qualitative assessment there rather than the quantitative approach that we have been discussing up until now.

Given this extraction then and the search for data inputs, the next main aim was to look at problems associated with the data that had been supplied, so particular problems with the EMEA data irrelevant of any other information that may have been collected from elsewhere.

In relation to this, we would be looking for data efficiencies, inappropriate data collection with specific regard to risk assessment modeling and any areas of missing data, again, for this particular data set.

Then if possible in this very short two-week time period, the aim was to qualitatively assess the risk for one particular risk question, make recommendations on further work that should be undertaken both by the EMEA and any other groups that may have been involved here. And this would be with regard to future data collection and, in addition, risk assessment methodology, both from a qualitative and a quantitative perspective.

(Slide.)

So what we want to look at now is the qualitative risk assessment that was undertaken. So if you remember, I had mentioned that this was done -- the agreement was to do this if it was turned out to be possible in the short time frame. And it was found that some information was available to undertake a short qualitative assessment.

(Slide.)

So today we have been talking about quantitative modeling. How does this differ from qualitative risk assessments and when should we focus on taking this approach rather than the quantitative steps in the first instance? So here we will have just a basic reminder or an introduction to this for those who are not familiar.

(Slide.)

The first and most important step that we found in this process was to define the exact risk question that we are trying to address. And this has been mentioned today in relation to the CVM model. We have to both agree with the assessor and the manager what is the exact question that we are trying to address here. So our first step was to agree with the EMEA what exact question we would be looking at.

(Slide.)

We then move on to elucidate pathways from the particular hazard that we are interested in to a particular unwanted outcome.

(Slide.)

For each step on the pathway gather the data to assess an overall probability first for each step and then for the overall risk that we are interested in.

(Slide.)

And this for a qualitative assessment would be assessed in terms of words such as low, medium, negligible or high. And these are words that are commonly used in risk qualitative assessments.

(Slide.)

So the case study that we were looking at, Salmonella and fluoroquinolone group of antimicrobials.

(Slide.)

The question that we were posed to address by the EMEA, a rather large question here, but essentially it is looking at the risk of adverse human health effects in the European Union only consequent upon the development of antibiotic resistance to fluoroquinolones due specifically to the use of these drugs in farm livestock.

And you will notice here we are not identifying any particular species such as poultry, cattle, pigs. In this first case study, they wanted us to consider all livestock species.

(Slide.)

So our first step based on this risk question was to consider an appropriate or a possible risk pathway to describe in which antimicrobial resistance could be transferred from the farm to the human and have an adverse human health effect. We decided in the first instance to go down the traditional route of the farm-to-fork type approach and map out the different stages that would be necessary in this element.

(Slide.)

So looking at the key stages of the transfer from the farm to the fork and to the human health effects.

(Slide.)

So we began by looking at resistant organisms present in farm livestock. And here we are defining these to be Salmonella typhimurium-resistant organisms to the fluoroquinolone group, then moving on to see how this would transfer from the farm to result in a human exposure to these resistant organisms.

(Slide.)

Following exposure, humans could then either be infected or colonized with resistant organisms. And then this would lead on to an adverse human health effect.

(Slide.)

So our aim was to look at each of these stages in turn, evaluate the EMEA data, and determine how much could we actually use to estimate in a qualitative manner the probabilities that would be necessary to describe these various stages.

(Slide.)

So Stage 1, resistant organisms in farm livestock. Essentially, here our real aim would be to assess the probability of the presence in farm livestock of resistant organisms due to the use of these drugs. And from the EMEA data, we were able to first of all assess in the first aspect presence of Salmonella typhimurium infection.

(Slide.)

And we find that throughout the EU, the data that had been supplied suggested that such prevalence was both variable between different countries and between different livestock species.

(Slide.)

There was a large amount of missing data in this one data set that would allow us to properly interpret or properly estimate a level of prevalence.

(Slide.)

Overall, we concluded that the prevalence of Salmonella typhimurium would be low, but there was a high degree of uncertainty associated with this data.

(Slide.)

We then, following prevalence, attempted to look at, well, if an animal is infected or colonized with Salmonella, then what would be the probability of those organisms being resistant to fluoroquinolones.

(Slide.)

It was complex and contradictory data for this aspect model. Again, variation between different EU countries and species of livestock. There was missing data again and there was a large range reported, again, for the different species and different countries ranging from zero to 86 percent in some cases. Again, overall we concluded it would be low; but, again, a high degree of uncertainty.

(Slide.)

So our overall conclusion for this first stage, how likely is it that resistant organisms would be present on the farm, overall we concluded that it would be low, but with a high degree of variability and uncertainty with regard both to countries and different species.

(Slide.)

So for Stage 2, human exposure to resistant organisms, we have assumed that these resistant organisms have originated on the farm and by some mechanisms, exposure through ingestion is going to result. So for this stage, we would have to look at all stages of the production process and then preparation, cooking in the home of the consumer.

(Slide.)

So our end point here would be to assess the probability of human exposure to these resistant organisms resulting from farm livestock. And the main point that we found here from the data provided, that we could not assume that all resistant organisms present on the food source came from the source animal. The information the EMEA provided did not allow us to assume that.

(Slide.)

So ideally, for this stage, we would have likened to consider each step on the production process an attempt to estimate the probabilities. And we found that this was very difficult to do. So we had to approach it from a different way.

(Slide.)

And instead of trying to estimate the probability of transition of organisms, we looked for data at each end of the food chain position to see what the probability of isolation would be. So we found again it was very little in this case between different stages of production for different livestock species and, therefore, different types of food product, and again within European countries.

(Slide.)

Missing data, again, particularly in different serotypes of Salmonella, overall probability of isolation at any one stage seemed to be low. But, again, very much a high degree of uncertainty.

So given, again, that we considered to some respect the probability of isolation at the different stages of production, how then would preparation in the world of the consumer have an effect on the probability of final exposure?

(Slide.)

This aspect of the exposure process had not been properly addressed within the data collected by EMEA. There was a very limited amount of information to assess in any respect this probability.

(Slide.)

Overall, it appeared that the probability of cooking reducing the level of exposure would result in this significant type of reduction. But this was all we were able to conclude from the information provided.

(Slide.)

So, overall, the probability of ingestion and, therefore, exposure to these organisms, again, low we concluded, but a high degree of uncertainty.

(Slide.)

Variable again between country and species. And, therefore, again, we are looking at a problem with much variability and much uncertainty.

(Slide.)

So Stage 3, given exposure to CARS, then what is the probability of either colonization or infection? For this stage, we are interested in the probability that the Salmonella typhimurium organisms from animal products going through the food chain would actually result in some kind of infection or even colonization.

(Slide.)

Again, the information provided to allow us to do this, for example, in a dose response type approach was very, very limited. So, again, we had to approach it and look for information in a different manner.

(Slide.)

So in this case, we looked for information that would suggest that there was any relationship between human and animal isolates reported in the literature.

(Slide.)

The reported conclusions were equivocal and, therefore, we could not automatically assume that these isolates had originated from farm livestock. Again, this was from this available data.

(Slide.)

Following on from this, we then looked to see, well, what is the actual probability of any randomly selected individual in the country being reported as a case of Salmonella typhimurium infection.

(Slide.)

And from this data, this suggested to be low and in some countries it was very low. But, again, there was variation by country and, again, uncertainty. And much of this uncertainty arose due to the differences in the reporting systems and, therefore, in the differences in the availability.

(Slide.)

But what we found from the information was that even where reporting was mandatory, the probability still appeared to be low.

(Slide.)

So our final stage of this farm-to-fork type model and adverse human health effect, what is the probability of such effects given ingestion and subsequent infection or colonization with resistant Salmonella typhimurium?

(Slide.)

Again, a limited amount of data allowed us to estimate this in a way that we would normally do in the farm-to-fork type approach. So, therefore, we looked at human isolates and the data that would allow us to estimate the probability of those isolates actually being fluoroquinolone resistant.

(Slide.)

David suggested that this was generally low, although there would be appear over the years to be a suggested increase within the U.K.

(Slide.)

Again, it was suggested that more data really here would be required to reduce large amounts of uncertainty. Given then that human -- a random individual may be regarded as a human isolate or fluoroquinolone-resistant, then what would be the probability of this resulting end treatment requirement? This was suggested to be low, but it may be higher for resistant strains than for non-resistant strains.

And the suggested range in the data provided was around ten to 36 percent, so a large amount of uncertainty again.

(Slide.)

So the overall risk from these different stages, trying to combine these, what is the probability of an adverse effect, we concluded from each stage that each stage has a low probability of occurrence. And for most stages, there were some data available to quantify perhaps at a later date.

And for some stages, data was particularly sparse, in particular from the probability of the food point at the point of ingestion actually being contaminated with resistant organisms and the probability of strains isolated from humans being the same and, therefore, coming from strains from livestock.

(Slide.)

So the overall quantification, again, variability due to species and country, large amounts of uncertainty in missing data in particular, with regard to serotypes, denominators and reported methods of isolation. But overall, our initial qualitative assessment suggested that the probability would be low, a large amount of uncertainty and variability.

(Slide.)

So this very short study, what did we conclude from this? A number of recommendations were made. First of all, a risk assessor should be appointed to work closely with experts in the EMEA if any further data has to be collected. So a large amount of data was collected, but none of this was done with the view to undertaking a risk assessment.

(Slide.)

The data sources provided should be revisited in a much longer period of time rather than two weeks. And this should be done with an understanding of risk assessment. And that would allow some estimation of uncertainty.

(Slide.)

It was suspected that there are much data in existence for this probable. But it is not actually available in a format that could be required to allow us to input into either a qualitative or a quantitative assessment.

(Slide.)

So a revised qualitative assessment should be undertaken at some stage to indicate genuine data gaps and, indeed, to consider one more specific question that was undertaken in this study in particular for one livestock species and perhaps for one European country. And then at a later date if appropriate, a quantitative assessment should be undertaken --

(Slide.)

-- concurrently with data collection, perhaps a Stacastic model and would allow an updatable tool to use in a regulatory fashion as the one presented today. Thank you very much for your time.

(Applause.)

DR. LONG: Thank you, Dr. Kelly. I think this is a great example of the use of qualitative risk assessment to help us focus in on what our problems and data gaps are without going through the large quantitative exercise as an immediate first step. Are there any questions? Okay.

We are scheduled to go into a break. And as you may have noticed, we are a little bit behind. And Dr. Sundlof promised we would be out of here by 6:00. I am going to do the best I can during the panel discussion to keep us on track there and maybe trim a minute off of each person so we can make up about eight minutes.

If we take a ten-minute break instead of a 15-minute break, then we can gather back another ten. And 6:00 is still a distinct possibility. So it is 3:00 now by my watch. Back at 3:10.

(Whereupon, a brief recess was taken.)

DR. LONG: The next item is looking at the risk assessment, assumptions and uncertainties by Kathy Hollinger who is a veterinary epidemiologist and by Mary Bartholomew who is a mathematician/statistician both for the Center for Veterinary Medicine. I think that their talk will be important as we move then on into the panel discussion and address the questions that are listed on your agenda. So this is going to be a tag team here. So Mary is going to start us off.