An Investigation into Childhood Leukaemia in Northampton

"We want to find our why our children are getting leukaemia"

5.1 Do five cases mean there is something nasty going on?
5.2 Clusters can occur by chance
5.3 Apparent clusters can be due to chance
5.4 So much for theory but what about practice?
5.5 "Is there a link?"

5.1

Do five cases mean there is something nasty going on?

We have said that childhood leukaemia is an uncommon disease and, from the above figures, it is clear that there are more cases of childhood leukaemia in the Pembroke Road area than one would normally expect.

"I won't accept it as a coincidence" ⁸

A parent

"... the cases in Pembroke Road, although extremely serious and distressing for the families concerned, could have occurred by chance." ⁹

Director of Public Health

It is quite natural that people worry that there might be something in the area that has caused leukaemia in these children. Most people, including medical professionals, have the gut feeling that this "can't possibly be a coincidence". Yet Northamptonshire Health Authority have consistently maintained that this cluster could be a chance occurrence and this view has been confirmed by all of the child cancer experts and epidemiologists we have talked to. So why do we say this?

This section explains why we believe that the Pembroke Road leukaemia cluster is likely to be just be coincidence. It is very important to understand this as it is the key to understanding why it is not yet possible for epidemiologists and doctors to tell us what caused leukaemia in our children.

5.2

Clusters can occur by chance

Firstly, random (i.e. chance) events cluster. Random distribution is not uniform but features clustering. Figure 2 below shows a matrix of 100 squares containing 40 random numbers between 0-99 drawn from random number tables, and marked as crosses on the grid ¹⁰.

The average number of crosses expected in each square is 0.4 but one square contains 4 crosses, that is, 10 times the expected number and one contains three crosses. These clusters have occurred merely by chance.

This phenomenon of clustering of chance events, without any underlying "cause" or risk factor, is seen time and time again. Take the UK National Lottery as an example: since it started in November 1994 to the time of writing, there have been 56 draws. Each week 7 balls (6 plus the bonus ball) are drawn from a possible 49 balls. Thus 392 balls have been drawn to date. On average one would expect any particular number to have come up 8 times. However, in over a year the number 39 has only come up once while the number 5 has come up 15 times. The risk of drawing a 5 was 15 times the risk of drawing a 39 during this period! However, this is does not mean that the lottery is fixed, that there is "something nasty" going on behind the scenes, it is simply due to chance.

5.3

Apparent clusters can be due to artefact

We are able to understand the world we live in because of our natural ability to distinguish patterns. We are constantly categorising what we see in order to simplify and make sense of what is about us. Sometimes the patterns that we see are imposed by us rather than being an objective feature of the world. When this happens the pattern is not real but an artefact. There is considerable evidence that shows that many of the disease "clusters" we see (for many diseases not just leukaemia) are due to the way we look at the world, that is they are artefactual.

The key problem lies in how we define what we are going to call a cluster. We need to have boundaries that limit what we are talking about. Boundaries are not just the lines we draw defining the area or place where we think the cluster is; boundaries are needed to delimit everything that defines the cluster. Criteria that need to be specified to define what counts as a "case" for the purpose of delimiting the cluster include:

What types of illness must a person have to count as a case?
What geographic area or place is the cluster in (i.e. where do we stop and start counting people with the right illness as part of our cluster)?
What age must a person be to count as a case?
Do people with the right illness have to be born, or diagnosed, or resident at some time prior to diagnosis, in the geographic area defining the cluster to count as a case?
What time period does a person have to have become ill in to count as a case?

"An even more serious problem with some surveys of leukaemia.. is a tendency to draw the boundaries after seeing the data. It is essential to draw all boundaries (in space, time, age group and type of disease) before conducting the survey. Otherwise it is easy to create highly improbable incidences by choosing tight boundaries around apparent clusters... Studies by people who are not aware of the problem can lead to apparently frightening results." ¹¹

The New Scientist

Our natural analytical attitude means that we subtly and unconsciously shift boundaries around to maximise the clustering effect. This can be illustrated by considering how the Pembroke Road cluster has been defined in relation to some of the criteria listed above.

5.3.1 - What types of illness counts as a "case"?

Are we interested in any cancer in Pembroke Road area or only leukaemia? What about lymphoma, that is often counted in the same group as leukaemia and shares some of its risk factors? Are we going to count all the different sorts of leukaemia or only the one that most children get, which is known as ALL. Some of the children in Table 1, do not have ALL but CML should they be considered as part of the cluster? Normally we do not actually ask these questions because it just seems natural to select the case definition that makes the cluster seem bigger: we think we are just describing an objective feature of the world. In fact we are subconsciously drawing boundaries to maximise the effect. Thus in the Pembroke Road area we are not very interested in other childhood cancers or adult cancers because the incidence of these is unremarkable, however, if they had been high we would have immediately added them to the pot!

In fact Northamptonshire Health Authority did check with the Oxford Cancer Intelligence Unit about other cancers in both children and adults. We looked at all cancers in the NN5 7** post-code sector for the decade 1983-1992. We did not find more people with cancer than expected ¹².

5.3.2 - What areas or places constitute part of the cluster?

This problem of shifting boundaries is probably best illustrated by the defining of geographical boundaries. We have already seen that chance events cluster. Therefore, when we think we may have a cluster we need to ask ourselves not only whether it is a cluster but also whether it is the sort of cluster we would expect to see occasionally because of chance clustering? It is important to look at the statistical significance of a cluster in the context of what is expected from the overall distribution. We do not naturally do this. We do the opposite. It is our natural inclination to mentally draw as tight a line around a cluster as we can. So, for example, our leukaemia cluster is not in Northamptonshire nor Northampton nor even the Spencer Estate but in "the Pembroke Road area", i.e. the smallest area that can be drawn around the five cases. Note also that it is not just Pembroke Road because this would not have maximised the number of cases making up the cluster (3 of the children with leukaemia are from Pembroke Road on lives in Countess Road nearby). This natural behaviour can make it seem as if there is a cluster when there is not one, or that there is a bigger cluster than there really is. This phenomenon is known as The Texas Sharpshooter's Procedure ¹³: the sharpshooter first fires at the side of the barn and then paints a bulls eye around the bullet hole.

If one reviews the press coverage of the leukaemia cluster in the Pembroke Road area one can see this shifting boundary phenomenon in action. Here are just a few examples:

The case definition is shifted to include adults:: "The toll of cancer cases in Pembroke Road, Northampton has risen to six. In a new development the Chronicle and Echo has been told of two more victims in the street. Now two adults are known to have had the cancer" ¹⁴^¶

The case definition is shifted to include other diagnoses:: "The latest victims to emerge include father-of-three who was diagnosed with lymphatic cancer - linked with leukaemia - in 1989. He died aged 50 in 1992." ¹²

The case definition is shifted to include a transient population:: "Mother-of-five said her brother was diagnosed with leukaemia. He had stayed with her when she was still living at Pembroke road." ¹²

The geographic boundaries are extended from the Pembroke Road area to capture extra cases:: "We can now reveal that two more leukaemia cases have been found in nearby Kings Heath and Duston." ¹⁵

^{¶ NB This paragraph also shifts the place of residence of
one of the children with leukaemia.}

In the light of all the above considerations it is not surprising to learn that there are frequent reports of clusters of childhood leukaemia ¹⁶.

5.4

So much for theory but what about practice?

5.4.1 - Methodological research

If our minds naturally generate artefactual clusters by unconsciously shifting all the boundaries until a cluster seems obvious and real, can we distinguish real from artefactual clusters?

We have seen that problems with disease cluster come from identifying the cluster while unconsciously defining its boundaries. To get round this what investigators have to do is to define the boundaries first (so that they cannot be manipulated sub-consciously) and then proceed to find out if there is a cluster based on a firm case definition and, if there is, how likely it is to have occurred by chance - only if the Texas sharpshooter draws a bulls eye on the barn before he shoots can we really know whether he is a sharpshooter or not!

5.4.2 - Have leukaemia clusters been looked for in this way, starting first with a clear case definition?

Yes.

One of the most comprehensive investigations into childhood leukaemia clusters was undertaken using data from the Childhood Cancer Research Group Register ⁵. A very large set of anonymised data, relating to over 9,000 case of childhood leukaemia and non-Hodgkin lymphoma, was made available to a number of different research groups for independent analysis. The results of the different analyses were all published together. These showed that there was some general evidence of a moderate degree of clustering of childhood leukaemia across the country ¹⁷. However the study did not attempt to determine the causes.

Another study ¹⁸ that attempted to clarify the significance of space-time clusters of leukaemia in children examined the incidence of cancer during childhood (0-14 years) in Los Angeles county over a five year period. Did this reveal statistically significant clustering of childhood leukaemia? No. Space-time clusters of childhood leukaemia were looked for within each of 31 regions of the county. These geographic areas were defined in advance of examining the cancer cases. One cluster of 7 cases of leukaemia was found in one of the regions which, had it been considered in isolation, would have been statistically significant (that is, appear to be unlikely to have been a chance occurrence) but when considered statistically in relation to the overall distribution of cases was simply what one might have expected because chance events tend to cluster. What is really interesting about this study is that the authors then looked at the map and deliberately drew tight boundaries around a few cases next door to each other to "create" leukaemia clusters (like the Texas sharpshooter drawing his bulls eye after shooting). When they did this they managed to create 9 leukaemia clusters where the rates (and the statistical significance) of childhood leukaemia were as high or higher than in two of the most famous post hoc leukaemia clusters.

Thus not only are there good theoretical reasons why what appears to be an obvious and statistically significant cluster probably is not, this evidence from methodological research is a practical demonstration that striking clusters can simply be due to artefact. Evidence like this suggests that we must be very cautious before concluding that clusters are not due to chance: there does not have to be a hidden cause for us to see a clear cluster that "couldn't possibly be a coincidence".

5.5

"Is there a link?"

The fact that apparently statistically significant individual disease clusters can be due to chance or artefact is quite difficult to grasp at first as it flies in the face of our "common sense" and intuition. This situation is not helped when the Health Authority say things like "the cluster is consistent with being a chance occurrence" and will not say "the cluster is just a chance occurrence" or that the cases are "probably just a coincidence" but will not categorically rule out the possibility of their having some common underlying cause.

"I want the Health Authority to investigate if there is any link between the cases"

A parent

"Yes, I can understand why a cluster of this sort is probably just due to chance but surely one can still look at these cases and check to see if they have anything in common that might have caused them" ¹⁹

This often gets a response along the lines that "yes, it may be consistent with being a chance occurrence but is it a chance occurrence or is there a link between the cases?"

This is a very sensible question. For example, if someone tossed a coin five times in a row and it came up heads every time, this is consistent with being a chance occurrence (five identical consecutive tosses will happen with a chance of about 1 in 16), but you might still want to examine the coin to see whether it was a two-headed coin or not!

It is quite natural that people wonder whether the Health Authority ought to do a special investigation into the Pembroke Road cluster to see whether that is due to chance or not (indeed there have been numerous demands from the press and the public that we do just that).

5.5.1 - Could a local investigation tell us if the cases are linked?

No, unfortunately, a special epidemiological study would not answer this question.

Why not?

There are three main reasons why such a study would not tell us whether the cases are linked or what caused them.

Firstly, part of the problem lies in what has already been talked about, the fact that the cluster could be due to chance, as the Cancer Research Campaign point out:

"Clusters of childhood leukaemia have been reported but their investigation is problematic since most of the clusters are statistically likely to be due to chance." ²⁰

Secondly, we do not know what causes leukaemia. If we get more than one case of meningitis in Northamptonshire, even if there are only two cases, the Health Authority immediately investigates the cases to see if they have any common link, in case they might be part of an outbreak, i.e. have a common cause. One of the differences between this and leukaemia is we know what causes meningitis: we know what to look for; there are laboratory tests that can confirm whether two patients have got the same infection or not; there are appropriate preventive measures that can be instituted. Another difference between this sort of individual disease cluster and leukaemia is that they occur much closer together in time making the identification of common links much easier (as happens also in outbreaks of food poisoning). Cancers have a long period of latency, that is to say, it can be a long time between the events that cause a cancer and the appearance of the disease itself, making it very difficult to establish common links.

If we do a special study into the Pembroke Road area we know we will find connections between the families, such as

the children may have been born at the same hospital
the families may use the same shops and buses
the children will play on the same playgrounds and breath the same air

but this does not help us distinguish whether the cases are connected or a coincidence. Only if there was something highly unusual that was common to the children with leukaemia would there be any chance of establishing a link.

Thirdly, we have no clear hypothesis about what might have caused these cases of leukaemia.

A few of the hypotheses that have been mentioned about what might have caused children in the Spencer Estate to get leukaemia:

a spillage of nuclear fuel
a spillage of aviation fuel
weed killers or pesticides used on the railway line
storage of petrol tankers or petro-chemical contaminants from the railways
contamination by chemical aerosol from the nearby cleaning of chemical drums
renewal of pipe linings from the water distribution system
infections
living near a fire hydrant
population mixing
electric or magnetic fields
living near the site of a tannery
radon gas

The initial suggestion, that there had been an accident involving radioactivity on the railway, was found not to be true. Since then there have been a great number of further hypotheses advanced (see box above). With the lack of a clear primary hypothesis, a study would have to be very large to be able to distinguish between different possibilities. As one epidemiologist from the Imperial Cancer Research Fund, Cancer Epidemiology Unit in Oxford has commented:

" unless a specific hypothesis is defined in advance, the relevance of a single geographic cluster of disease can rarely be interpreted. The statistical power of such investigations is usually low and only marked increases in risk can be detected. Investigations of specific hypotheses about defined sources of environmental contamination are more likely to result in conclusive findings than are in-depth studies of individual clusters." ²¹

The Health Authority reached the conclusion in 1993 that a special epidemiological study could not help us find any answers and we consulted national leukaemia and cancer experts and epidemiologists who confirmed this conclusion. We know that local residents sought and received independent confirmation of this from Professor Cartwright, a Professor of Cancer Epidemiology at the Leukaemia Research Fund back in 1993:

"At the present time it is not possible to distinguish random from the non-random and this is one reason why most cluster investigations do not come up with any conclusions. It would be impossible, I suspect, in this instance to take this very much further." ²²

Sadly, despite a great deal of effort and time spent in discussing and explaining this to local journalists, they have continued to demand the impossible "on behalf of the families". We believe that they are doing the families of children with leukaemia and the local community a disservice. It is vital that the right sort of studies are undertaken to address this very serious problem - futile investigations distract from the real work that needs to be done. Despite very many investigations into individual childhood leukaemia clusters, none have been conclusive. At an International Symposium on Leukaemia Clustering held in Canada the Vice President of the Epidemiology and Statistics Department of the American Cancer Society admitted that at that time he did not know of any investigation of an individual leukaemia cluster that had concluded it was due to anything other than chance ²³. At the same meeting it was noted by another senior epidemiologist:

"One take-home message for me is that other epidemiologists from around the world seem to agree that formal investigation of clusters have not been very productive those of us in cancer agencies have to figure out how to communicate these problems so that we can concentrate our efforts on what we feel would be productive work." ²⁴

We hope this report contributes to this effort.

One local paper in Northamptonshire has argued that

"For the anxious parents of the Spencer Estate, any effort to pinpoint the cause, whatever the chances of success, is better than none." ²⁵

We disagree: we feel that it would be irresponsible for the Health Authority to embark on a course of action that had no chance of success and to do so as a "public relations exercise" would just demonstrate contempt for the people we serve.

"I want to know whether the Health Authority is willing to put any money into finding out the cause" ²⁶

-a parent quoted in a local newspaper

Just for the record, the question of cost in time or money played no part whatsoever in the decision of the Health Authority not to institute a special study. The decision was taken solely on methodological and scientific grounds. However, we do think it is our duty as a Health Authority to use the resources we have available in a responsible way. Fortunately appropriate scientific studies are being done (e.g. the current UK Childhood Cancer Study on which an estimated £6,000,000 will be spent ²⁷).

5.5.2 - How does the Health Authority decide when to do a special epidemiological investigation?

How does the Health Authority decide when to do a special epidemiological investigation and whether it would stand a reasonable chance of succeeding? Following an international conference to discuss the investigation of disease clusters, a checklist of characteristics of clusters that suggest a special investigation may have some chance of success was published. This is reproduced in the table below. In the right hand column of the table we discuss whether the Pembroke Road cluster has this characteristic or not. The more characteristics a cluster has the more likely it is to be able to produce some sort of result.

Checklist to identify clusters where special investigations may have some chance of finding the cause or identifying new preventive measures that can be taken ²⁸
Cluster Characteristic	Is this characteristic seen in the Pembroke Road cluster?
There are at least 5 cases and the relative risk is very high (20 or more).	NO - Although we have enough cases, the relative risk is not high enough (even with the most extreme interpretation of the data the relative risk was only 5.6).
The disease is one for which a unique and detectable class of agents has been responsible in the past, or the pathophysiological mechanism is well understood.	NO - we do not know the cause of leukaemia, epidemiologists have only been able to identify a few risk factors and the overwhelming majority of cases are not associated with these known risk factors. NO - we do not yet have a detailed understanding of what is going wrong in the body when it starts to produce too many blood cells.
This agent is persistent in the environment and can be measured there.	N/A - this is not applicable since no such agent has yet been identified.
The agent is persistent or leaves a physiological marker in the bodies of people who have been exposed to it but which is rare in the normal population, so that it can be used as an index of exposure.	N/A - this is not applicable since no such agent has yet been identified.
People who live in the same neighbourhood have different levels of exposure to the possible cause so that effect of exposure can be assessed.	NO - The routes of exposure to the hazards that have been suggested as possible causes of leukaemia in Pembroke Road are such that everyone in the Pembroke Road area is likely to have been exposed. So even if we had a biomarker of exposure or effect for one of the suggested environmental hazards (which we do not), everyone in the community would test positive to some extent.
The plausible route of exposure is such that subjects would be able to accurately assess their own relevant exposure on a questionnaire or it could be reconstructed from records.	NO
It would be feasible to carry out a multicommunity study consisting of several similarly exposed and some unexposed communities.	NO - This is not feasible in Northampton, it needs to be done on a larger scale. Fortunately this is what is being done in the UK Childhood Cancer Study.
The cluster represents an as-yet-uninvestigated, endemic space cluster, rather than a space-time cluster. This suggests a stable, persistent problem and perhaps a persistent agent to be found in the environment.	This was a theoretical possibility, however, on investigation we found that there were NO previous cases of childhood leukaemia in this area prior to the identification of the first case of this "cluster".Moreover by tracing the history of the land in the Spencer Estate we found nothing unusual or alarming.

There is not one feature of the cluster in Pembroke road that would suggest an investigation may be able to help find the causes of the cases!

Back to Section 4

Return to the Index Page

Forward to Section 6

© Northamptonshire Health Authority, reproduced by kind permission of Dr Amanda Burls, Sen Reg in Public Health Medicine.

Created using: Lightning HTML Editor Version 2.10.1997