DR. DONNELLY: Can I have the first slide, please? I guess we can think of this whole area and my whole talk as being a discussion of risk, both relative and absolute. And both of those two ways of looking at risk need to be kept in mind throughout this presentation.
I will be also going from an area of relative predictability, relative certainty to an area of relative uncertainty and unpredictability when I shift in the talk from BSE to variant CJD.
You have the even more difficult task of then adding on an additional level of uncertainty. If we knew the prevalence of variant CJD infection in people who lived in Britain for the whole time, what about people who lived there only a brief time or visited? What would their risk be via blood?
I realize all of these things are difficult to put together, but I will try to keep in mind throughout this presentation showing what we know with relative certainty and where we have to do sensitivity analysis to look at the range of what we do and do not know.
Looking at the BSE epidemic through Great Britain, -- and this shows the BSE epidemic of cases, which peaked in 1992 -- you see over 174,000 cases of confirmed BSE in Great Britain.
Keep in mind that it was only in 1988 when the disease BSE became notifiable in Great Britain. So we know through a number of sources that there was under-reporting prior to that.
It was first diagnosed in 1986, but the BSE inquiry, which is ongoing in Great Britain, has identified certain cases that were seen and diagnosed to be spongiform encephalopathy in 1985. So there were even earlier cases definitely documented.
You can see that the epidemic, which had peaked in 1992, has declined considerably since then and is declining to very low levels.
You can see here to some extent the geographic distribution of BSE. This is shown in number of cases per 1,000 cattle. So it just shows the geographic distribution of BSE throughout Great Britain, both that it was geographically disbursed, but you can also see correlation that those counties that had relatively high incidence in, say, 1993 also were the same as the ones that had relatively high incidence in 1991. It's interesting that the disease showed geographical dispersion almost immediately once it was recognized.
I don't think there is a whole lot that can be gained by speculation on where it started because we have the problem of where it was diagnosed versus where it was started. So there would be a period of time when vets were getting to know and recognize the disease that probably determines its earlier pattern, rather than its actual spread of the infectious agent.
In looking at the demographics of BSE, how many cattle were infected, when they were infected, when they were slaughtered. We use a technique called back calculation. This was first developed to use for HIV and AIDS and is a technique that statisticians use in diseases of long incubation period.
Now, long incubation period is bad in some respects in that the key regulation brought in that turned the BSE epidemic from increasing to decreasing was the ruminant feed ban, which was brought in 1988. That made it illegal for the feeding of ruminant protein to other ruminants.
Now, there is considerable evidence, both through surveys as well as through our own work, that this was not immediately completely effective. But it did turn the tide of the epidemic.
Unfortunately, that long incubation period meant that although the tide of incidence of infections turned in 1988, it wasn't until 1992 that we saw the turn in the tide of BSE cases. And that was a function of this long incubation period.
The long incubation period and varied incubation period means that we can look at cases that we see now and get information about past and even relatively recent incidence of infections. And that helps us do projections of future cases.
So the basic approach is if all animals were infected relatively young, then if we see only, say, 20 percent of animals survive to age 5, for each case that we see at age 5, that we can think of representing 5 infections. So we can work from the cases that we have seen and the time period that we have seen them over and work backward over to the infections that that represents.
Now, we need information to do that, both on the actual demographics in terms of the number of animals born each year and their proclivity of survival. We also need information or a form for the incubation period distribution. That is the time from when an animal is infected to when it experiences the clinical onset of disease and another distribution that ties together exposure and susceptibility with age. So this represents the sort of age-specific susceptibility exposure to infection.
Now, we could get information about the incubation period through experiments in cattle. And there have been experiments where cattle were experimentally dosed through the oral route and then watched over a period of time to get information about, among other things, the incubation period. That takes an extremely long period of time, a large sample size, to get an idea of what such a varied and long incubation period disease would require.
But also there is reason to think that the incubation period depends on dose. So how would you know the dose that the population of cattle in Great Britain were receiving? So we are going for an empirical estimate of the incubation period borne out to fit the BSE epidemic.
Now, this looks complicated, but it's not, actually. We have got the incubation period coming in. First let me tell you what the formula is actually representing.
For animals born at a certain time, we cross-tabulated all the animals that had BSE in Great Britain that we analyzed by the year in which they were born and the age at which they experienced BSE.
So you imagine this cross-tabulation table. And we know from the agricultural annual census the number of cattle that were born in each year. So we know our denominator. We just need to figure out, then, what the processes were that generated the cases that we saw.
So on the furthest right-hand side, we see an animal that was maternally infected. So there is greater complexity in figuring out what the time-dependent rate of maternal transmission is.
We know through various studies that there was a rate of approximately ten percent maternal transmission in the last six months of the maternal incubation period. But obviously, then, through time, that depends on how many cows are in that incubation stage.
Those animals that were maternally infected then experienced an incubation period of duration U at the onset of age U. Those that were feed-infected, which is the term, then, further to the left, which we have here, is a combination of the feed risk. And that's of absolute time.
So if this animal was infected at age A, it experienced the feed risk at that time of K at T naught plus A. It had age-related exposure G of A. And that means if it was infected at age A, it had an incubation period of U minus A. That, of course, only applies to those animals that were not maternally transmitted.
So in the square brackets, we have the term for animals being infected and onsetting at age U. We then have to add in the probability that an animal actually survived to age U given when it was born. And we have a survival curve where the majority of animals are slaughtered by three years.
So when we have a long incubation period disease, that means the majority of animals infected actually were not seen to be clinical cases of disease. So they were slaughtered for human consumption.
We then add an additional term of the probability that a case gets reported because, as I noted, the disease BSE was only made notifiable in 1988. So prior to that time, we have under-reporting, which is important to include.
Through fitting such a model, we were able to get a very good fit to the data. You can see here the data for various cohorts. These are the animals that were born in 1987. And you can see that when we look at the number of cases by age, -- this is age naught to eight years -- you see a very good fit of the model to the data.
We have done considerable sensitivity analysis. I think you have a big pack of publications. What we find in fitting these data is you have to get a very precise fit to the data. That requires very precise estimates of the incubation period. You can't fit the data with an incubation period that differs very much from this in form. It doesn't provide the good fit to the data that we need.
Similarly, the age of infection distribution representing exposure susceptibility has to have this relatively odd peak form. This peaks between sort of 6 to 18 months of age and suggests this is a key point in the animal's life when it's most susceptible to infection.
Now, the investigation we have done into feeding practices suggests that this is not just a function of when cattle are fed ruminant protein or protein-supplemented feed because it seems that animals typically receive protein supplements from the first few days of life and then receive considerably more protein supplement after the first lactation.
So to have the key time be between 6 and 18 months suggests that it is something biological in their susceptibility. Now, this is key, having an early infection is key, in interpreting what number of infections generated the cases that we have actually observed.
This is our estimated feed risk profile, which is the function I called K in the earlier formula. What it shows, this is plotted in the highest resolution that we ever fit. You actually don't need this much resolution to get a good fit to the data.
The key aspects of this are up through 1988, the approximately exponential rise in infection incidence. Now, you see lots of spikes as well. We believe this reflects the seasonality in the use of protein supplements, that you need more protein supplements in the winter than in the summer, and that's reflected here.
You see the key in this feed risk profile -- and this is feed to cattle -- peaked in 1988. That was when regulations were brought in that turned the tide of this infection incidence profile. And the infection incidence then dropped considerably in 1989 as a result of the regulations that were brought in.
Now, it's important to note that we did not tell the model in any way that regulations were brought in 1988 or what the effect might have been. We used the data to tell us what the feed risk was.
So, although you might look at this and think "Oh, well, they didn't work absolutely" and that is certainly true, there was a belief at the time very optimistically that this would just absolutely stop feed-borne infections. And it obviously didn't.
But 1989 would not have just been the same as 1988. On the basis of these trends, we would have expected it to be considerably higher, going up in an exponential manner.
And since we estimate in the 1988 cohort of cattle approximately ten percent of animals born in the 1988 cohort were infected with the agent of BSE, although 174,000 cases is considerably a bad epidemic, it could have been much, much worse.
So, although obviously the earlier they brought in regulations, the better, had it been a year later, it would have been a considerably worse situation.
So you can see here that then we have blips of infection later. It's really difficult under the most recent years to get good estimates. You have the least information when looking at current cases about the most recent estimates. But the key thing to consider there is that in 1996, in light of the announcement in March of 1996 about variant CJD cases, further regulations were brought in that restricted the feeding of any mammal protein to mammals.
So one of the suggestions for the reason for this leakage of the ruminant-to-ruminant feed ban was that there may have been the use of, say, pig feed, which could contain cattle or sheep protein, feeding that to cattle because it was on the farm, the farmer needed it, looked like pretty much the same thing.
Another possible route would have been the contamination of equipment if equipment in a feed mill was used for making pig feed and then it was used for making cattle feed.
I think the key thing is that it was extremely effective. We have analyzed this data in another way as well, which was -- I don't have time to go into the details, but it was looking at what is called the basic reproduction number. That is the average number of new cases per initial case.
So if each case or each infection generated on average one or more infections, then the epidemic will be stable or grow. If on average one infection generates less than one secondary infection, the epidemic will die out.
Under all of the scenarios we considered, you see the epidemic dropping to basic reproduction numbers well under one. So all of the suggestions we have are that the epidemic is dying out.
Here you can see in the context of the cases that were observed, so in purple is the annual case incidence, the epidemic of infections. That is shown here in red. The green represents at each year end how many infected animals were alive.
The key thing is to look at the difference in both time shifting from earlier to later and in magnitude that the magnitude of the epidemic of infections is considerably greater than the epidemic of cases.
So we estimated that some 900,000 cattle were infected through 1996. This manifested in over 174,000 cases, as I pointed out. And the difference between those, then, is largely animals that were slaughtered for human consumption. They were slaughtered over a range of incubation stages, which I will address in a moment, but that gives you an idea of the magnitude of the epidemic and the potential risk.
That was Great Britain, constituting Wales, England, and Scotland. Here is a separate analysis for Northern Ireland, which experienced an order of magnitude lower infection incidence. So in Northern Ireland, we estimated some 11,000 animals infected and of those, over 9,000 slaughtered for consumption.
So, again, you see the characteristic shift between estimate of the infection incidence compared to the case incidence. It was earlier and of greater magnitude.
Now, as you may know, the export ban on Northern Irish beef was lifted before that of Great Britain due to both the lower magnitude of infections and also of greater tracing of animals, which was historically due to greater TB incidence.
So it was interesting to consider the time period over which you are looking for travel. This gives an indication of the estimated total number of infected animals slaughtered per year. So this might be a basis where you start to think of translating the risk from cattle into risk to humans.
Classified here is animals slaughtered over and under 30 months. As you can see, the majority of animals slaughtered for consumption are under 30 months, but as the epidemic progresses and new infections are at a much decreased level, the majority of those animals being slaughtered for consumption are actually over 30 months. This is key because of the regulations brought in 1996, which restricted human consumption to animals slaughtered at under 30 months.
So while they didn't eliminate infected animals, that wasn't their basis, what they did was to distinguish between animals at higher and lower risk. Animals under 30 months were at lower risk because of the infection incidence profile going down considerably. And also for those animals that were under 30 months and were infected, they would be at an earlier incubation stage, which may lead to less or lower infectiousness.
Now, also, the other reason I showed this was to show the magnitude. If you imagine that all animals slaughtered for consumption that were infected were equally infectious, that would lead to peak at about 1989-1990 for potential infections.
If the key, though, is animals slaughtered in the last year of their incubation period, those animals that hadn't yet reached clinical stage but were near it, then we see the peak of infectiousness. Again, if those animals in the last year of incubation were all equally infectious, it would peak at about 1992.
So it is very difficult to think about: one, dividing up the risk into the '80s and the '90s; but also thinking about if you're thinking in terms of person time that necessarily one year at a certain time equals one year at another because there are temporal changes.
I also show this because you can see the detail of the estimated number of infected animals slaughtered for consumption under 30 months at extremely low levels.
So if it is these animals that provide the majority of the risk, we're talking a handful of animals. This is particularly important in Great Britain because, in addition to the restriction of animals over and over 30 months, there were regulations brought in 1989 that specified both bovine offal ban, which restricted those tissues believed to be potentially the most infectious, as well as an additional regulation brought in more recently, highly controversial in some areas, beef on the bone. And beef on the bone was banned to restrict exposure to dorsal root ganglia, which is found to infect mice.
Of course, then, in Britain we only worry about animals under 30 months because those are the only ones being consumed. And it was found that dorsal root ganglia was infectious to mice in the last year of incubation period. That is why people were particularly interested in this being just a handful of animals. That ban is still in place but highly under discussion.
This gives you an indication of the confidence intervals for -- these are animals slaughtered under 30 months of age over the recent time period and next year. Both gives you an indication of those animals in all incubation stages.
So we're talking on the order of between 150 and 50 over this time period as well as just, in green, the handful of animals that are slaughtered within 12 months of onset if you're just considering that last period to be potentially infectious.
Now, that is all looking at what is here, which is a BSE epidemic, where we have considerable data, we have done sensitivity analyses. Everything that fits the data has very similar results to what I have shown you here: a peak in susceptibility; a long, approximately five-year on average, incubation period; and number of infected animals slaughtered each year dropping considerably. But to consider variant CJD cases, you have many steps between the BSE epidemic, over which we know a considerable amount, and how we translate that into variant CJD cases.
For this particular meeting, I should have drawn another arrow from variant CJD cases to blood donors, which would be those people who are preclinical but giving blood.
Now, a number of issues come in here highlighting the specified bovine offal ban, which may have considerably reduced potentially infectious material in meat; heterogeneity in consumption rates, which may play a role in your deferral decisions, looking at those who ate more or less meat; potential for infection; dose response susceptibility heterogeneity. I bring that up because there was a mention of risk factors, potentially genetic. And you probably know that all of the variant CJD cases that have been identified to date have been methionine, methionine homozygotes. They have that in common with approximately 40 percent of the British population.
It may not mean that it is just those 40 percent who are potentially infected. It may mean that we have genetic effects in the incubation period. So it may be that we have shorter incubation periods for some genetic groups and longer for others.
Now, I won't go through these formulas because they are even more complicated. The key assumption we made here -- and I am happy to go through this with people at some point later if they want to -- is that we assume a linear dose response.
Because what I am going to tell you and conclude is that we still have a lot of predictability, it can still be an extremely large or extremely small epidemic. The fact that we made a linear dose response assumption has not led to any undefendable restrictions in what could happen.
Our goal here was to find the widest range of potential epidemic scenarios that were consistent with the data. And you see here coded by color in the number of cases between now and 2040 the smallest epidemics, shown in white here, correspond to short incubation periods with a range of standard deviations. So each one of these points represents an epidemic scenario that was consistent with the data that was observed.
This is the annual incidence of cases, 3 in 1995, 10, 10, and 16 in last year. A better way to distinguish these epidemic scenarios is in terms of a parameter we call R. That is the mean number of humans infected by one maximally infectious bovine.
Now, quite logically, if that number is very small, then we will have much more smaller epidemics. And that corresponds to small incubation periods.
As R increases in magnitude, we get larger epidemics. And this may be useful only in that we may be able to get some idea from the meat industry on what the largest potential R could be. How widely is the meat from one infected animal spread between consumers? Is it through a relatively small number or could it be as high as 100 or 1,000?
What we have been able to show through these analyses was that there remains uncertainty. We took forward this analysis because there were people saying they could tell exactly what was going to happen. It was going to be a large number of cases.
One person was quoting in 1996 two million by 2000. He doesn't say that any more. There are also people who say they have looked at the data and they can show that it is absolutely going to be small, there will be no more than one or two hundred cases.
So I think we have to keep in mind that although it is nice to know the answer, we have to admit when we don't. And so far we can't restrict what potential epidemic scenarios could take place.
Over the next one or two years, if the number of cases stays on the order that they are now, predictability will increase considerably and we can put a useful upper bound on the epidemic. Now I would say we are at the point where we cannot.
So thank you.
CHAIRMAN BROWN: Well, thank you very much for a colorful and I would have to say courageous presentation in view of the mathematical formulas. I think the point is well-taken, and it was missed by a lot of people when various modeling studies began to be published.
The major point in some of these studies was the point that you just heard. There is almost total uncertainty about the extent of what is going to happen in terms of the numbers of new cases of new variant and the numbers of people who may currently be incubating the disease.
The uncertainty is so great that it almost seems pointless to dot i's and cross t's with respect to how are we going to estimate any possibility of risk to the U.S. blood donor and recipient population. This is the huge, major, complete unknown, and it is not going to get more known before the day is out.
We now have a presentation by Dr. Philip Comer, who will give us the Det Norsk Veritas risk assessment. Dr. Comer?