5 The epidemiology of vCJD
In 1990, six years before the probable link between BSE and vCJD was established, the CJD Surveillance Unit was set up in Edinburgh. All suspected TSEs in humans have to be referred to this Unit, which maintains the UK's official data on all forms of these diseases. Figure 10 is a plot of the number of cases referred to the Unit from 1990 to 2004. [R D]
Describe any trends that you can see in these data.
Although there was considerable fluctuation from year to year, the annual number of cases referred to the Unit did increase from 1990 (when there were about 50) to 2001 (when there were about 180). Since then there has been a noticeable decrease in the number of cases. This is especially evident for 2004.
What might account for the growth in the annual number of suspected cases referred to the CJD Surveillance Unit during the 1990s?
There might have been a genuine increase in the incidence of human TSE diseases during this period - possibly attributable largely to the appearance on the scene of vCJD in the mid-1990s. Alternatively, as a result of increased awareness of BSE and vCJD, members of the medical professions may have become more alert to the possibility that some patients, whose deaths might previously have been attributed to other causes (such as Alzheimer's disease), might be suffering from a TSE disease.
The reduction in the number of suspected TSE cases reported to the CJD Surveillance Unit in 2004 might reflect either a recent decline in awareness about TSEs among members of the medical profession or greater proficiency in eliminating the possibility that a patient might have a TSE disease.
Figure 11 is a plot of the number deaths in the UK from definite or probable sporadic, iatrogenic, familial and variant CJD, and also GSS, from 1990 to 2004.
Do any of these data throw light on whether there has been a genuine increase in the number of human TSE cases since 1990 or whether increased reporting accounts for the apparent rise?
Clearly, the increase in the number of vCJD cases must have been genuine, as there were no reported deaths from this entirely new disease before 1995 and then some every year since then. There is no reason to suppose that the number of people dying of sporadic CJD has increased in recent years. Therefore, the overall upward trend in the number of deaths from sporadic CJD (from fewer than 30 in 1990 to more than 70 in each of 2002 and 2003) suggests greater reporting of suspected cases.
It is impossible to detect trends in the data for iatrogenic and familial CJD, and GSS, because of the small number of cases. The fluctuations are probably random. Nevertheless, it is rather worrying that deaths from iatrogenic CJD appear to be continuing despite all the precautions that have been taken.
What might deaths from iatrogenic CJD as recently as 2003 and 2004 reflect?
These relatively recent deaths may well reflect the typically long incubation periods of TSEs in humans. Most of these patients probably contracted the disease through medical treatments (such as the use of growth hormones derived from people with undiagnosed CJD) many years previously, before the dangers were fully appreciated.
Figure 12 shows the number of deaths in the UK from definite or probable vCJD from 1990 to 2004 plotted separately. The relatively sharp decline between 2003 and 2004 may be partly due to delays in confirming that patients who died in 2004 definitely did die from vCJD (which requires post-mortem examination of brain tissue) but it may indicate that the worst has passed.
Once vCJD had been recognised and its probable link with BSE widely accepted, it became a very high priority to assess the likely magnitude of the vCJD epidemic. Epidemiologists at Imperial College London have periodically estimated the probable number of deaths in the UK from vCJD. Their 1997 prediction was that the disease might cause up to 10 million deaths. This figure was revised downwards to 50 000 in 2002 and 7000 in 2003 (with the strong likelihood that it would drop further).
What do you suppose was the main reason for such dramatic revisions of these predictions? [C R D]
Looking at Figure 12, it is clear that any prediction made in 1997 must have been based on very little data. Indeed, it might be argued that it was irresponsible to place such a premature prediction in the public domain. By 2002 and 2003, rather more data were available and therefore the pattern of vCJD cases with time would have been clearer.
In what senses might the later predictions be regarded as 'improvements' on the earlier ones?
Hopefully, more recent predictions are more accurate than earlier ones. A more accurate prediction is one that is closer to the final number of cases (although, of course, only time will tell). The availability of more data would also allow the researchers to increase the precision of their predictions (i.e. reduce the random uncertainty surrounding them).
It is important to distinguish between accuracy and precision. A prediction might be accurate but known imprecisely (i.e. the true value falls somewhere within a fairly wide range of possible values). A prediction might also be precise but inaccurate (i.e. although a fairly narrow range of possible values is quoted, the true value in fact lies outside this range). Of course, normally the aim is to make predictions that are both accurate and precise (that is, with the true value falling within a fairly narrow range of possible values). In this case, the researchers' predictions were given in the form of their best estimate of the number of eventual deaths from vCJD together with upper and lower 95% confidence limits for this estimate (see Box 7).
Box 7: Statistical probability
When researchers have to make statements about a population - for instance, about the number (or proportion) of organisms having a particular genotype or the mean value of a particular measurement (such as height) - they generally have to employ probabilistic terminology. The reason for this is that it is usually impossible and/or undesirable to count or measure all the members of the relevant population. Instead, a sample is drawn from the population; the sample is then counted or measured and the data obtained form the basis for a statement about the entire population. For such a statement to have any validity, the sample must be representative of the population from which it is drawn. That is, the sample must not be either deliberately or accidentally biased in any way. One way to help guard against accidental bias is to make sure the sample size is sufficiently large that random fluctuations have little effect on the data. It can be surprising just how small a 'sufficiently large' sample can be - for instance, about 30 measurements can often adequately represent a very much larger population. A good test of the representativeness of a sample is to check that other similar-sized samples drawn from the same population produce similar data.
The same issues arise when it is necessary to make predictions about future trends on the basis of the limited information that might currently be available (as in the case of trying to predict the number of eventual vCJD deaths from the small number of cases that had occurred up to any particular point in time).
Probabilistic statements often take the form of giving for a population a best estimate (for example, of the proportion of a particular genotype, of the mean height of organisms or of the eventual total number of vCJD cases) together with an indication of how much higher or lower than this the true figure might actually be. The latter information is often provided as upper and lower 95% confidence limits. Effectively, these claim that the probability that the true value is (or will be) either greater than the upper confidence limit or less than the lower confidence limit is 5% (or, equivalently, 0.05 or 1 in 20). Of course, there is a relatively small possibility that the true value is outside these limits. This is the nature of probabilistic statements. The reason why 95% is often used for confidence limits is that, conventionally, a result expected on fewer than 5% of occasions is regarded as statistically significant. The closer the upper and lower confidence limits are to the best estimate - or the smaller the difference between the upper and lower confidence limits - the greater the precision (that is, the smaller the random uncertainty) of the statement or prediction.
In 2003 (based on data up to the end of 2001), the Imperial College team's best estimate of the number of deaths from vCJD in the UK by 2080 was 200 with upper and lower 95% confidence limits of 7000 and 10 respectively.
Compare this lower 95% confidence limit of 10 deaths with the number of deaths that had already occurred by the end of 2001.
According to Figure 12, by the end of 2001 there had already been considerably more than 10 deaths from vCJD.
What the researchers were saying was that, although they thought that the total number of deaths would be close to 200, the number could well be either higher or lower than this. However, they believed that the probability that this number would either be greater than 7000 or less than 10 was less than 5% (0.05 or 1 in 20). The number of deaths by 2001 had already exceeded the lower 95% confidence limit. Similarly, the final number of deaths is expected to fall well short of the upper 95% confidence limit. [C]
In what way(s) might a reader potentially interpret a newspaper headline such as 'Up to 7000 [or 50 000 or 10 million] UK deaths from vCJD'? [C]
Expressing the prediction in this way might have the effect of encouraging readers to assume that it is likely that vCJD will cause the deaths of many more than 200 people.
What challenges can you see in reporting the researchers' predictions more comprehensively than implied by the above hypothetical headline? [C]
Eye-catching headlines are necessary to induce most people to read a newspaper article. While headlines should not be misleading, they must usually convey a simple message.
It would be difficult enough to convey the subtleties of 95% confidence limits in the body of an article. The challenge of doing so in a headline would be even greater. On the other hand, perhaps readers - having followed the BSE/vCJD episode as it developed over the years - would prefer more informative headlines such as 'Scientists predict 10-7000 vCJD deaths in UK with 95% confidence'.
The predictions discussed above are based on:
the number of actual cases of vCJD reported - which is why their precision has increased over the years (i.e. their random uncertainty has decreased);
a mathematical model of how vCJD spreads in the population (see Box 8).
Box 8: Mathematical and computer models
Increasingly, we read that scientists have employed mathematical models of the real world - usually run on computers - in their research. The history of mathematical modelling goes back a long way. Sir Isaac Newton (1642-1727) encapsulated his law of gravitational attraction between two bodies in a single elegant equation. These days the phenomena that scientists are trying to understand (e.g. the Earth's climate) involve so many interactions that they can seldom be represented nearly so simply. Fortunately, powerful computers are now available to run models based on multiple interacting equations - equations that would have taken many person-years to solve not so long ago. Of course, any model incorporates assumptions about the real world that some researchers would accept and others might not. It is therefore incumbent upon researchers to make their assumptions clear and to justify them. One interesting way of validating a computer model is not just to use it to make predictions, but to use it to make so-called retrodictions; that is, to run the model from starting conditions that were believed to apply at some stage in the past and to see if it predicts reasonably accurately the conditions that are known to pertain today.
As noted in Box 8, any mathematical model incorporates a number of assumptions. A very important assumption built into the Imperial College model was that people contracted vCJD only from eating contaminated meat.
What other possible routes of infection does this assumption ignore?
It ignores the possibility that vCJD might be contracted from infected surgical instruments or through blood transfusion.
In fact, although at the time of writing (2005) there have been no cases of vCJD attributable to infected surgical instruments, we have seen that there have been two cases of vCJD that may have been contracted through transfusion of blood from donors who subsequently died of vCJD.
Given what was known about genetic susceptibility to vCJD when the 2003 prediction was made, what other assumption built into the model will probably have to be revised?
Until 2004, it was assumed that only the 40% of the population homozygous for methionine at position 129 of the PrP protein (i.e. those that had the genotype MM) was susceptible to vCJD. With confirmation that a person whose genotype was MV had contracted vCJD (although vCJD wasn't the cause of their death), the model will have to be revised to allow for the possibility that people whose genotype is MV (and probably also VV) might develop vCJD, given sufficient time.
At present, there isn't a routine test that could be given to members of the general public to ascertain whether or not they are harbouring PrPSc protein - and therefore whether they are at risk of developing vCJD. Even if there were such a test, its use to improve estimates of the number of vCJD cases that there might eventually be would be hugely controversial. If it were possible to warn people that they had a much higher than average possibility of developing the disease - and assuming they wanted to know - would it be right not to tell them? On the other hand, imagine the impact that such information would have on anyone's life. [C R D]
In 2004, the results from tests on 13 000 preserved tonsil samples kept from tonsillectomy operations were published. These suggested that the number of vCJD cases might be around 4000 - much higher than the Imperial College team's estimate of around 200. However, this figure was an extrapolation from just three samples that tested positive for PrPSc (two of which were doubtful, a situation not uncommon in clinical diagnoses). [C]
Even if all three of these tests were reliable, why should one be cautious about the figure of 4000 cases of vCJD?
Extrapolation from even three reliable positive results would be problematic because of the effects of random fluctuation. Imagine the impact on the prediction of even one fewer or one additional positive result.
There are plans to test about 100 000 fresh tonsil samples over the next few years. Suggest a reason why this survey may not throw a great deal of light on the eventual extent of the vCJD epidemic. [D]
Most tonsillectomy operations are performed on fairly young children, most of whom were likely to have been born long after there was any significant risk of eating beef contaminated with PrPSc protein.
Suggest some reasons why reasonably reliable estimates of how the vCJD epidemic is likely to develop are needed. [D]
The health service would certainly need to plan ahead if there might eventually be 50 000 victims of vCJD - let alone 10 million - rather than about 200 victims. Hard decisions might also have to be taken about how much resource to devote to trying to develop preventative treatment or a cure for vCJD depending on the likely number of victims. If the epidemic does eventually fade away having claimed about 200 lives - with 148 having already died by 2004 - then no pharmaceutical company is going to embark upon the necessary research and development programme.
Although it is generally accepted that most vCJD victims contracted the disease in the mid- to late 1980s through eating contaminated meat as teenagers or young adults, there are other possibilities. One of these is that they may have been infected as early as 1970 through eating contaminated baby food. Since their gut walls are more permeable, babies may be more susceptible than adults to infection from food.
What additional assumptions about BSE and vCJD have to be made if this hypothesis is to be taken seriously?
One would have to accept that BSE existed about 15 years prior to its 'official' recognition. Further assumptions are that beef was included in baby foods at that time (which should not be too difficult to establish) and that some of this beef might have been contaminated with BSE (virtually impossible to establish after all these years). Another implication is that the incubation period of vCJD is about 25 years rather than about 10 years.
Part of the evidence cited in support of the baby food hypothesis is the decline in the annual number of vCJD cases in recent years (Figure 12) and the continuing relatively low average age of victims. It is argued that this might represent a decline in a 'first wave' of vCJD, rather than the disease's disappearance.
If so, what would comprise a possible 'second wave' of vCJD?
People who ate contaminated meat in the 1980s and are currently incubating vCJD, but who do not yet display any symptoms of the disease.