In addition, if we wish to look at the trends in cancer near the plant, or further afield, we need the numbers of cases of cancer and the populations for defined areas, which are more distant from the plant. There is one other refinement. Some cancers are naturally more prevalent (incidence and deaths) in poor people or disadvantaged people. The effect is large for lung cancer, the most common cancer in men, and is believed to be due to cigarette smoking behaviour. Some people think it is because poverty and disadvantagement may have effects through immune system suppression. For some other cancers, notably breast cancer and leukaemia, there is a weaker effect in the opposite direction. Thus there is more breast cancer in less disadvantaged or richer women. This could be because of differences in nutrition (more dairy) or behaviour (having fewer children later in age), or a combination. But these effects should be controlled for, and this can be accomplished in various ways, using indices of disadvantagement. The easiest way to control is to adjust the expected numbers by multiplying by a weighting factor calculated from the relationship between Social Class and cancer rates in the national population. Social Class makeup of an area is obtained directly from the census. Many other indices, Carstairs or Townsend, which include unemployment, multiple occupancy of houses, car ownership etc. can be calculated from data in the census.

There are basically two sources for these kinds of data, national databases and questionnaires. The problem with national databases, as far as our examination of point sources of risk are concerned, is that generally speaking the establishment will not give out the data, so no one can find out anything. This seems to be true for the USA and Europe as well as the UK and it is getting worse. I will discuss the problem further in my new book but there is one tremendously useful exception. Although the rules for releasing such data are held to refer to both incidence (developing the disease) and mortality (dying of it) [M Quinn, 1992], since about 1999, ward level cancer mortality (deaths) for all malignancy, lung cancer, breast, stomach and prostate cancer annually in England and Wales from 1995 has been sold by the UK Office for National Statistics. This oversight, which I am sure is regretted because of the use we have put to the data, is probably because when ONS was split off from government by Mrs Thatcher, they were told they had to be self sufficient, and so they began to sell of the data to make ends meet. Anyone could buy this data. So for about £500 we do now have seven years of cancer mortality data for the main cancer types by ward, 1995 to 2001. The oversight was noted belatedly following some studies I carried out in 2000 and 2001 near nuclear sites in the UK and these received a great deal of Press coverage (since we found cancer increases near nuclear sites) and by 2002, the gate had been shut. No more data was to be released. I have an application into the Freedom of Information Act at the moment but don’t hold your breath.

In the UK, wards are the smallest sizes of area for which the population makeup can be obtained. The UK census in 1981, 1991 and 2001 gives numbers of men and women in each 5-year age group for all the wards in the UK, and these data are also available from the ONS for a fee. Before I turn to an explanation of how to proceed from here, I will briefly examine two other sources of data, questionnaires and local doctors' records.

If we wish to examine cancer risk in a smaller area than a ward, or if locals can take the matter into their own hands, they can bypass the cancer registries and knock on doors. Green Audit developed this local area cancer questionnaire technique originally because it seemed that we were going to be refused small area data for Ireland. In the event we did get some numbers from the new National Cancer Registry in Cork, but by then we had gone ahead in the small area around Carlingford and Greenore in County Louth. Later, we used the method in Burnham on Sea, near the nuclear power station at Hinkley Point in Somerset. I will give an account of the questionnaire method. Briefly, the method consists of defining an area and visiting all the houses in the area with a questionnaire. This asks the head of the household, or some responsible person to record the ages and sex of each person living in the house, together with any case of cancer diagnosed in the last ten years, the type of cancer and the age at diagnosis. This information allows the calculation of risk on the basis of the population living in all the houses.

The other possibility is to approach local GP surgeries and ask if anonymised totals of cancer incidence can be made available for research. For a meaningful result, we also need the total age breakdown of the surgery population. It may seem unlikely that these data will be forthcoming, but we were able to use this approach in Ireland and the results were very valuable. I will now turn to the basic method used to convert the data into meaningful patterns of risk.

1. Ward level data

The method is essentially the same whatever the source of data. We have to calculate the expected numbers of cancer deaths or diagnosed cases in the area of interest for the period we are studying. We then must compare these with the observed or recorded numbers. It is best to look at as long a period as possible in order to get the largest number of cases into the study. This is because the larger the number of cases, the less likelihood there is of any effect we find being a consequence of chance. The procedure is simple but quite tedious if many wards are involved. We start with the population of each ward making up the area close to our proposed source of risk. The source of risk may be a nuclear site, or it may be a coastal strip where radioactive materials are known, by measurement, to accumulate. To give an example, I will show how the sums are done for an area we have been interested in, the ward of Burnham on Sea North in Somerset. The 1991 census population of females in this ward is shown in Table 2.1 below together with the England and Wales death rates from breast cancer and calculated annual expected numbers of deaths. These populations can be obtained from ONS as computer text files; and this is the best way, since they can be imported directly into a spreadsheet program and the calculations done from within it. There are, therefore, no transcription errors.

The rates for the cancer in question are calculated from national figures published by ONS. Table 2 of Series DH2, Mortality by Cause gives the numbers of deaths in each 5-year age group, and also the population. So that it is easy to follow this, I calculate the rate for one year in Table 2.2

These calculations, and the others that follow, are very tedious, but are made straightforward and can be done in blocks of as many wards as you like, using the computer. The most widely used software that will deal with these calculations quickly and efficiently is Microsoft EXCEL, but other spreadsheet programs will do just as well e.g LOTUS 123. Whole columns containing hundreds of populations of wards can be multiplied by rates to give new columns of expected cases and these can be added for all the age groups and the result put into a new column with a few clicks of a mouse. For those who are able to do these procedures by batch programming, the whole process can be automated, and indeed we are now developing a system here that will enable us to obtain cancer mortality risks in any ward in Britain.

In Table 2.1, the 1991 female population of the ward of Burnham on Sea North is given by 5-year age group. In this table we are calculating the expected annual numbers of breast cancer deaths, based on the rates in England and Wales. Each 5-year age group population is thus multiplied by the appropriate rate, given in column (C) to give the expected numbers in column (D). All ages are then added up to give the total expected number of deaths, 2.178. What this tells us is that if Burnham on Sea North has exactly the same risk as the population of England and Wales, and the population had not changed from 1991, there would be 2.178 deaths from breast cancer every year. What we found, in the first of these studies we did in 2000 was that there were 17 recorded deaths from breast cancer in the four years 1995 to 1998. The expected number would be 2.178 x 4 years = 8.7 deaths. Thus we can calculate the age standardised mortality risk or Standardised Mortality Ratio as:

SMR = 17/8.7 = 1.95

This is a valuable discovery. We have found that there is almost twice the probability of dying of breast cancer here than the mean for England and Wales. And yet Burnham on Sea is not in the middle of some industrial slum area surrounded by chemical factories: It is a pretty little seaside town, where people go on holiday and children play in the sand. But it is also directly opposite and down wind of the Hinkley Point nuclear power station complex just across the bay.

The same calculation can be done for any cancer type and for men and women combined. All that we need is the appropriate rates, the ward population by sex and 5-year groups and the observed numbers.

Table 2.1 1991 female census population of Burnham North ward with expected numbers of deaths from breast cancer (D) calculated by multiplying each age group population (B) by the average England and Wales mortality rate (D).

* calculated from annual figures of deaths tabulated by cause published by ONS (series DH2) for England and Wales

Table 2.2 Calculating the England and Wales breast cancer mortality rates for 1995

2. Questionnaire and GP studies.

The approach in both of these types of study is exactly the same. The questionnaire responses give the numbers of cancers diagnosed in each household, the type of cancer, the sex, age and year at diagnosis and the sex, age and number of everyone living in the house. When all these data are added together we have a sample population, drawn randomly from the total population of the area being canvassed. We know the sex and age breakdown of this population, at the time of the questionnaire and can calculate the annual expected numbers of cases of all cancers or any specific cancer from the national database. We then compare this expected number with the reported numbers over any period we choose to see if there is an excess. There are some particular problems with this approach for questionnaires. The main one is that as we go back in time from the date of the questionnaire, there will be leakage of people with cancer from the sample due to deaths. Thus we always find that the number of annual cases reported falls off for earlier years. This is shown in Table 2.3. below, for the 2001 Burnham on Sea North questionnaire undertaken by 'Parents Concerned about Hinkley' (PCAH). This survey was a response to denials by the Somerset Health Authority that there was any increase in breast cancer in the ward following the first mortality study we carried out in 2000 [Busby, Dorfman, Rowe 2000]. The survey obtained answers from addresses that added up to about one third of the census population of the ward.

Table 2.3 All cancer cases by year of diagnosis reported in 2002 in Burnham on Sea North according to PCAH questionnaire. Numbers fall off in earlier years because many have died and their families have moved away.

Now, on the basis of the population defined by the response, the expected number of all types of cancer was 11 per year, so if we choose to look at 2000 and 2001 together, this will define a risk of 29/22 or 1.32. For year 2000 alone, the risk is 17/11 = 1.55. The risk we calculate for all cancers falls off rapidly with time. This is because the numbers will be dominated by lung cancer, which is usually fatal, and for all cancers, lung and certain other cancers which have poor survival, the questionnaire method is of little use. But for looking at types of cancer that are treatable, or are less immediately fatal, like breast cancer and leukaemia, the method is valuable. It also has some great advantages. The first is that you can believe the results. If there is a significant excess of cancer shown by a survey like this, then you have immediate connection with the people who are reporting this. You know where they live and can relate this to the source of pollution by dividing your area up into bands of distance from the source, as we were able to do in Carlingford in Ireland. Best of all, you know that no one in the establishment has altered the data or re-evaluated the numbers and retrospectively removed cases that might indicate a problem. I show in the new book Wolves of Water that all these things happen.

The PCAH questionnaire showed that the Somerset Health Authority were covering up a real effect. It showed a doubling of the breast cancer incidence rate in the ward, confirming what we found in the earlier ward level mortality study. In reality the true risk will always be higher than the calculated risk because of this leakage of population, so if the questionnaire shows a problem, there is one. In addition to locating sufferers on a map so that their distance from the source of pollution can be assessed, we can also ask questions about lifestyle e.g. smoking, eating habits etc. which may be valuable indicators of exposure routes or other stresses. Some 18 months after our publication of the results of these studies, the local Cancer Registry were forced by the local Health Authority and concerns driven by media to conduct an official examination of the data to see if our results were correct. They found that the increases in breast cancer and leukemia we found were real. However, they argued that these were chance findings and not related to the nuclear site. Nevertheless, I make this point that the small study carried out by concerned citizens gave approximately the right answer and was validated by the cancer registry, who had refused earlier to release data to us or the citizens of Burnham. The media had a field day.

GP studies involve exactly the same calculations as we did for the mortality studies. The only difference is that the age breakdown of the study population is that of the surgery patient list. You have to be able to persuade the doctor to get the data out; often a very difficult exercise.

Part of my purpose here is to show that anyone can do such calculations and that people who are concerned about illness near any source of pollution, nuclear plants, landfill sites, incinerators, can check out what is going on for themselves. But we can't get too carried away at this stage; there is one more question we must address before we are secure in our belief that there is a problem at Burnham on Sea North, or anywhere where we find an apparent increased risk. We have to establish, as Bradford Hill's Canon asks: could this doubling of the risk of death occurred by chance? Here we have to deal with some statistical methods that enable us to answer this question.

I think am going to upset the epidemiologists here. As I mentioned, in the last fifteen years, the discipline has collapsed under the intellectual weight of statistical methods developed to answer the simple question about whether an event might be a real indicator of an underlying causal relation (e.g. radiation and leukaemia) or merely an example of the random play of chance. Part of the problem we have to deal with is that the levels of risk we will find for adult cancers are usually modest. In Burnham on Sea, downwind from the Hinkley Point nuclear power station we found in 2001 that there was a 2-fold excess of breast cancer based on 17 deaths over the four years 1995-98. In public health epidemiology the diseases that are being tracked are usually more immediate and graphic than the development of cancer some ten to twenty years after the initial causal exposure. Public Health departments deal with more mundane question. Some people eat at a restaurant and suffer food poisoning: why? Twenty out of thirty who attended a picnic are in hospital vomiting with a high fever: what was the cause? It is unlikely to be chance, in these cases. But with the cancer risk situation, it may be chance, and we have to establish whether it is. I am going to avoid the complexity of modern statistical methods and offer a few simple tests that will let us know whether our discoveries are statistically significant or not. When we discover some increased risk level and report it, the health authorities will usually respond immediately with the cry: this is a random cluster of cases! In order to see if this is so, for small numbers of cases up to 40, the most important weapon in our armoury is the Cumulative Poisson Probability Table. This is a table which is published in most compilations of statistical tables and which enables us to see at a glance what the probability of observing some number of events is, given that the expectation is some other number of events. The tables are constructed from the Poisson Equation, which is an approximation to the Normal Distribution (or Bell-Shaped Curve) for rare events in the wings of the distribution. If this means nothing, don't worry, just use the tables and assume they tell you what you wish to know. I refer you to Statistical Tables by J Murdoch and JA Barnes (Macmillan1998), but all these tables are the same. In the case of Burnham on Sea and breast cancer deaths from 1995-1998 we expect 8.7 and find 17. In the Cumulative Poisson Tables we look along the columns of expected numbers to 8.7 and then look down the rows to observed numbers 17. The Table gives the value 0.008. This value means that for an expectation of 8.7 deaths, you could expect to find 17 or any number fewer deaths with a probability of 0.008. That is 1 chance in 1/0.008 or 125. It tells us that if we looked at about 125 wards of the same size as Burnham on Sea, we would expect to find one ward with this apparent increase in breast cancer by chance alone.

The level of statistical significance conventionally accepted by statisticians as showing that a finding is 'significant' is taken to be below 0.05 (1 chance in 20) so in this case we certainly achieve this level. The numbers of cases involved in the calculation is very critical to this question of significance. Between 1995 and 1998 there were 17 breast cancer deaths. By 2001, another three years, the number of deaths was 14 more taking the total to 31. In the seven years the expectation is (assuming the same population) increased to 2.178 x 7 = 15.3. We now have a new risk 31/15.3 = 2.03. In the Poisson Tables this gives us a probability of 0.0002 or one chance in 5000 of these events being due to chance. There is not much difference between SMR's of 1.95 and 2.03, but the significance has hugely increased because the numbers are greater. This shows the importance of getting as many people into a study as possible.

Thus for the previous case, we would have:

= 246.49/15.3

= 16.1

Our calculation gives 16.1 and since this is greater than 10.82, the statistical significance is lower than 0.001, which we found in the Poisson Tables also. Some researchers demand that we also give the 95% confidence limits, and there is some reason for this. These are rather harder to work out than the p-values we have just calculated but can be obtained easily enough using a computer program. There are a great deal of excellent statistical computer programs which can be used to deal with these issues, and I will say a bit more about these below. Before I do this I need to briefly address the small numbers problem, since this is usually referred to by health officials trying to dismiss the significance of a finding.

We saw that the p-value from breast cancer mortality in Burnham North over the 5-year period was 0.008 and the RR was 1.95. This means that if we are looking for some effect near a nuclear site or other putative source of risk where we are studying a number of wards we should be aware that for every 125 wards there would be an excess of 1.95 by chance alone. In our first study we looked at about 109 wards, so finding the effect in one of them was not unexpected. On the other hand we already had a hypothesis suggesting the pattern of risk, close to the contaminated mud where Burnham on Sea was. This suggested the problem was a real one. So we have to look at more than just the value in one particular ward: we have to look for a pattern on several wards which share some attribute that can act as a surrogate for the exposure. In our case, it was proximity to the intertidal sediment. As it happened, increasing the period to 2001 increased the risk slightly but hugely increased the statistical significance to 0.0002. For this to have been a chance finding we would have to study about 5000 wards.

But there is yet another pitfall for the amateur epidemiologist: referred to (by the epidemiological community) as the 'Texas Sharpshooter'. In this, the argument goes, the student examines all the different cancers in all the different wards near a point source of pollution. This gives a large number of results (for different cancers) and, by chance, one of these will show a high risk (merely because there are lot of numbers). At this point, the student decides that this particular cancer is being caused by the releases. This apparently is not allowed. It is like the 'Texas Sharpshooter' firing his pistol at the wall of a barn and then after he has hit the wall, drawing a bullseye around his bullethole. Whilst I have sympathy with this argument, I get extremely irritated at the way it is mindlessly trotted out by establishment or government to deny real problems. I discuss a specific example of this in Wolves, but at this stage I want to point out that such an argument would make it impossible to discover anything on first notice. What if we lived near a factory that produced a seriously carcinogenic substance that caused throat cancer? We do not know this, but we may have some (plausible) general belief that pollution is a bad thing and causes disease. So we do a cancer study of the area using 100 wards (to get a control) and look at all cancer types and find that the ward near the plant has a 4-fold excess of throat cancer. We are not allowed to conclude anything from this because of the Texas Sharpshooter.

4 Adjusting for Disadvantagement.

It is quite commonly stated, when attention is drawn to high levels of cancer incidence in an area where there is a source of some environmental pollution, that the high rates are a consequence of socioeconomic factors. The argument generally follows the reasoning that people who are poor or unemployed smoke or drink too much alcohol and eat too many chips, not enough fresh fruit and vegetables. These behaviour patterns, it is argued, are associated with elevated cancer risk. It is also said that the genetic constitution of people in the lowest socio-economic groups is, in some un-elaborated way, 'poor' and this also adds to the cancer risk. Furthermore, these people get ill more often owing to their damp houses and overcrowded conditions with higher levels of infective illness or virus infection. Possibly also, the toffs might say, their lower class behaviour, too much boozing and sex.

The study of deprivation and health is associated with a large body of literature. Deprivation indices using data easily extracted from the Census (such as that first suggested by Townsend) include such indicators as multiple dwelling occupancy, car ownership, home ownership, unemployment rates and so forth. The report by Carstairs and Morris (1991) discussed the development and use of the now widely utilised 'Carstairs Index' of deprivation. However, even if it is subsequently shown that the area in question has a higher index of socioeconomic deprivation, great care must be exercised in attributing the cancer to the deprivation. A common method to examine trend in cancer incidence near a putative source of pollution is to use least-squares multiple regression to examine trend with distance. This is a method that has become much in vogue since the development of computers. It can be used for looking at data where there are multiple possible causal factors for some measurement of interest and in principle enables us to see what are the most important contributors. I will discuss it further below. In its approach to effects near a point source, distance may be entered as one of the independent variables as a surrogate for pollution level and we can also include some measure of socio-economic deprivation. Implication of causation by the pollutant is flagged up by high statistical significance for the slope coefficient of the regression line or curve. Implication of socioeconomic deprivation may also be confirmed in this way. This, however, may be entirely spurious since it is only generally poor areas where sources of pollution exist. Either they are built in poor areas because there is insufficient political power in such areas to oppose the project or else, following the identification of such a source, people rich enough to move away, do so. Furthermore, it is an extraordinary fact that in the UK there is in force a statutory planning instrument encouraging the siting of environmentally polluting sites in areas of high unemployment.

Examination of the relationship between cancer incidence and the Carstairs index of deprivation in Scotland reveals that, except for lung cancer in males and cervical cancer in females (many sexual partners and papillovirus transfer) the differences between the most deprived and least deprived are not very great. (Carstairs and Morris, 1991). Similar data for England has been obtained from the OPCS longitudinal study and uses Social Class as a measure of disadvantagement (Leon, 1988). The direction of the trend in cancer with increasing levels of deprivation may be positive as well as negative. In our work we follow Leon (1988) in defining a 'positive' effect as being an increase in cancer risk in going up the socioeconomic scale and a 'negative' effect as an increase in cancer in going down the socioeconomic scale. A large number of studies have investigated such effects both in the UK and elsewhere in the world and in general, the results indicate that lung cancer in males and cervical cancer in females suffer negative effects. Positive effects exist for colon cancer, female breast cancer, and leukemia. A Table of Carstairs deprivation index and cancer incidence in Scotland is shown in Table 4.1 below.

We used both Carstairs, and a Welsh Office index of deprivation for regression analysis in the Wales small area study but in our cancer mortality studies we carry out an adjustment of the expected numbers of cases in a ward using the relation between mortality and Social Class in England and Wales discovered by Leon in 1988. Since this is a straightforward calculation, I show how it is done. The relationship is given in Table 4.2 for the main cancers of interest. The trends with Social Class are similar to the trends shown by Carstairs indices. In practice, the adjustment is made to the expected numbers of cases. To use our example of breast cancer mortality in Burnham on Sea, first we have to obtain from the Census the numbers of each household in the ward in each of the Social Class categories as a proportion of all the households.

Table 4.1 Standardised Incidence Ratio of cancer sites by Carstairs deprivation score in Scotland 1979-82. (Source: Carstairs and Morris, 1991) The ranges of status in the rural areas we have examined are given in bold.

Table 4.2 Standardised Mortality Ratio (0-64yrs) by Social Class for some common cancers (1971-72)