Critique
Eric Mintz, Ph. D.

February 2004

I. Background

Port Hope is a small community with a current population of about 12,000. It follows that the study of rare diseases is limited by the small numbers of cases that would be expected to occur in a limited time period. This is a constant and crucial consideration. Many of the diseases that might be of concern in Port Hope are normally rare ones like brain cancer and leukemia. Any study of rare but important outcomes in Port Hope would want to maximize the number of cases available for study.

A. Mortality Data vs. Incidence Data

Even serious diseases like cancer may have a considerable long-term survival rate. Therefore a mortality study will result in many fewer cancers to study in a time period than would an incidence study. The degree of the difference is related to the case-fatality rate of the cancer. Lung cancer mortality and incident event numbers would not differ markedly because lung cancer is usually fatal. Breast cancer incidence rates are much higher than fatality rates, however, since the case-fatality rate is about 50 percent. Overall, there were a total of 589 incident cancers diagnosed (using postal code for residence) from 1986-1996. During the period 1986-1997 there were 339 cancer deaths. There were over 53 incident cancers and only about 28 cancer deaths that occurred per year during these most closely corresponding periods. Clearly when we use mortality data we lose a large proportion of the cases. This is crucial given the small numbers and the resulting low statistical power available to analyze most cancers, even when using incidence data. It is important that with mortality data many more of the computations will be confidential, since the overall number of deaths will more likely be less than 5.

The bottom line is that the lower number of mortality cases results in a serious limitation for studying cancer in Port Hope, since the mortality study will generally have fewer cases for each disease in each time period. This seriously limits the statistical power associated with the SMR (Standardized Mortality Ratio) analyses and subsequent conclusions that can be made.

Obviously the time from cancer initiation to death is longer than the time to diagnosis. This occurs because in addition to the time between exposure to disease diagnosis; there is an additional period from diagnosis to death. The longer the period from exposure to an outcome, the more difficult it is to make associations and particularly causal connections. Also, people may be more likely to move between diagnosis and death. The residence is assigned at the time of death. Again in this regard, incidence data is superior. It should be noted that because the time intervals are rather long; that many deaths would occur in the same time interval as the diagnosis.

Incidence is a measure directly rated to risk (if surveillance biases are not large) whereas mortality is related to two measures; the risk of getting the disease and the risk of dying from the disease once you get it. Clearly visible better medical care may lower mortality but will not usually affect disease risk. It is the risk of getting disease that is most important in Port Hope since risk is most closely related to causation.

The June 2002 Mortality Study (the mortality study) covers a period of 42 years compared to 26 years by the August 2000 Incidence Study (the incidence study). Taken by its itself, of course, the greater follow-up period would appear to at least partially balance the lower mortality numbers per time period.
But the study in specific time periods (e.g. 10 year periods) will have less power in the mortality study due to smaller numbers. This is very important because the entire study period includes times of higher and lower exposures which when combined will dilute any negative health effects. In other words, trends in time may be masked when we only have sufficient data to look at the entire time period.
This is particularly evident for childhood cancers where mortality data is only presented for the entire period. Presumably this is due to the small numbers resulting in time specific data being of limited use. Since children generally have greater exposures and shorter induction times, the childhood data is of particular interest. Looking at Table 1, it is quite evident that the incidence data for children is much more revealing than the mortality data (e.g. brain cancer).

B. Ecological environmental studies

This mortality study utilizes environmental epidemiology. Environmental epidemiology investigates whether environmental exposures are associated with subsequent excess rates of disease. Ecological studies in particular, such as this one, are blunt tools. There are many limitations and biases that make it difficult to obtain an accurate picture of the relationship between exposure and disease, some of which are described below.

1. Biases
a) misclassification by disease

(1) misdiagnosis

There is considerable misdiagnosis of cancer. The diagnosed cancer may not represent the primary site (where the cancer originated). The degree of misdiagnoses depends on the tumour type as well as other factors. This may be a smaller problem for mortality data when compared to incidence data.

(2) coding errors

There will inevitably be some death certificate coding errors of the tumour type by The Cancer Registry.

b) misclassification by exposure

Misclassification by exposure is most likely far more important than misclassification by disease.

(1) migration

Any ecological study such as this one, makes an implicit assumption that those who lived in Port Hope when they died had lived there long enough to have been exposed there. This is often not the case. Lee’s incidence study found that nearly 1/3 did not live long enough in the community to receive a meaningful radon dose. On the other hand, some people move away to live in a different community at the time of diagnosis. Their tumour that may have been acquired due to exposures in Port Hope, will not be attributed to Port Hope. People may be even more likely to move by the time of their death. These biases may be important. This bias will generally be nondifferential that means it results in an underestimating of risk (towards the null).

(2) residency coding

Residence at death may not be completely or accurately coded. This is discussed at some length in the report.

c) standard rates

The standard rates used to determine the expected numbers of cancers are based on provincial rates. These rates include other areas close to nuclear facilities as well as many urban areas that may have multiple exposures to various industrial and environmental contaminants. Therefore many of the communities included in the standard may have elevated rates. Therefore we could be comparing "dirty" areas to "dirty" areas and may be falsely underestimating risk in Port Hope.

d) Census Estimates

The accuracy of the census data for Port Hope is a possible limitation. Census data inherently will have some under or overcounts. If local census errors are different in direction or magnitude than errors in provincial counts, there would be bias.

2. Limitations

1. This being a descriptive SMR study, there is no information on individual exposures. The principal exposure is living in Port Hope. There is no way to control for confounding for individuals; that is we do not know what other factors besides radiological and heavy metal exposures in Port Hope may have caused cancers or other diseases. Certainly many other common causes are acting besides the one(s) of interest in this study. For these reasons, firm conclusions regarding cause and effect for such studies are generally not possible. The main purpose of this type of preliminary study is to provide leads for further research.

2. Statistical results can only be used as a guidepost. Since so many SMRs were calculated, many (5 or 10% depending on the significance level) would be expected to be statistically significant even if there were no unusual exposures in Port Hope. On the other hand, small numbers result in very low statistical power (ability to detect SMRs that are truly elevated). This is even more of a problem in the mortality study. Therefore elevated but non statistically significant results cannot provide reason for comfort. The pattern of elevated SMRs must be judged as to its suggestibility of problems. On interpretation of this pattern, reasonable people may disagree.

3. We are dealing mainly with chronic diseases such as cancer and circulatory disease, most of which have very long periods from diagnosis to death. It is difficult to assign causation for a chronic disease to an exposure that may have occurred 25 years ago. These problems are less important for some of the cancers such as leukemia that have a shorter induction period.

C. Burden of Proof

The statistical burden of proof is very high. Traditionally we have to be 95% or at least 90% certain (confidence level) that differences are not due to chance before we state they are "likely not chance findings" or statistically significant. That leaves the area of high but less than 90% confidence as a gray zone where results are likely real, but not statistically significant.

The arguments for using a 95% confidence level (5% significance level) centre on it being the historical standard for the burden of proof. There is, however, no logical basis for choosing the 95% level of confidence. In fact, in was originally arrived at based on the cost-benefit associated with agricultural decision making. It is arbitrary. Most people use it simply because others have, not because of its appropriateness. It was not designed with the limitations of environmental epidemiology in mind.

The 95% level would appear unrealistically stringent here for a number of reasons including the following:

1. The confidence level assumes that the data is of high quality. High levels of misclassification by exposure (people are placed in the incorrect exposure categories) and some misclassification by disease (people are incorrectly classified as either being free from or having a given disease) will usually decrease the chance of finding significant differences between groups. This sets the burden of proof much higher than the stated level. We are dealing with highly inexact data here. Most of the biases are non directional errors that would increase the burden of proof by adding noise to the data. The misclassification may be quite large in this study due to the serious sources of bias discussed above.

2. The ramifications of incorrectly finding a significant relationship (false positive) between an exposure and a cancer is not great since conclusions in a study of this type will be based mainly on patterns rather than individual associations.

3. The ability to detect true excesses (statistical power) is very low for many cancers in the mortality study. Using a high confidence level decreases the power.

For these reasons, I advocate a less stringent burden of proof of 90% if one is to be chosen at all.

In any case, the choosing of a significance level is largely a red herring. Significance levels are chosen when we are investigating whether an exposure (or a few) is related to a disease (or a few diseases). They are used to make decisions.

This study (and the incidence study) is not to be used for decision making. It is a preliminary study searching for likely leads. We are looking at hundreds of SMRs. It is not useful to rely on a stringent, strict statistical cutoff as a standard of proof. In so doing, we would be imagining that the quality of evidence is far superior than the reality and not taking into account serious issues affecting the true statistical significance of each finding.

The 95% confidence level, or any other, is not magical. For example, a study comparing leukemia rates in children living around the Pickering plant to lower risk children found approximately a doubling of the risk for children living near the plant. The probability of this finding being due to chance (p-value) was about 5.5%. Therefore the study was interpreted as being “negative” because the results were not statistical significant at the 5% level.. If the p-value had been 4.9% the study would have been seen as “positive”. Clearly these divisions are arbitrary and somewhat foolish.

The only statistical measure that really matters is the p-value, which is based on the results obtained. Interpretation of the results should be based on the p-value, an estimate of the biases affecting the p-value and clinical information, other information (such as the patterns over time and demographic groups) and common sense. Interpreting study results as negative based solely on arbitrary significance points is nonsensical in my view, for all of the above reasons.

The choosing of important findings is an art as well as a science. The p-value along with the consistency of the finding and its plausible relationship to exposures in Port Hope should be taken into account.

For example, in the incidence study, brain cancer was found to be elevated in all time periods for women in Port Hope. This excess was statistically significant at the 5% level for the period 1986-1996 using MOH data only (p-value < .05). It is also highly significant for the total study period (p<.01). This cancer is elevated in men for the 1986-1996 period, but the p-values are not low (could well be a chance finding). Taken alone the male data would not be strongly suggestive of excess brain cancer.

Brain cancer was found to be highly elevated in Port Hope children during the period 1971-1985 (p < .05). These findings taken together show a pattern that is quite suggestive of there being a problem with brain cancer in Port Hope, even though only some of the findings were statistically significant at the 5% level. This is because excesses were found in all three groups and in all time periods.

Children generally have greater exposures and shorter latency periods. That the brain cancer excesses were greatest in children and appeared earlier is supportive of a real excess that is environmental related.

The above example shows that statistical significance (at any level) should be used only as a guidepost or screen for diseases which require further scrutiny. There will be significant results that are due to chance because of the large number of SMR calculations. On the other hand, because of low statistical power, many of the nonsignificant elevations may be important particularly if they are part of a pattern of excess rates.

D. Time Window
.
As previously stated there are long periods between exposure and death for most cancers and other chronic diseases. This period between initial exposure and death will vary by disease. It becomes obvious that one will not uncover excesses if one is looking in the wrong time window. For example, most leukemias have short latency periods of less than 10 years. If it is true that exposure was greatest before the mid 1970s then we would expect the greatest leukemia rates to occur in the 1971-1985 time period.

The expected pattern of excess rates is one of the tools we can employ to separate perhaps spuriously elevated rates from ones that require further scrutiny.

E. Background Summary

Most of the biases discussed above serve as white noise masking any real effect between exposure and disease. Along with the limitations inherent in such studies, the result is that most environmental epidemiological studies are biased towards producing “negative” results. The low quality of the evidence (data) being evaluated often guarantees such findings. This should be taken into account when interpreting the findings.

II. Text Review and Discussion

1 b. Port Hope Background

The Lees study (2) did show an association between radon levels and lung cancer risk in Port Hope after smoking was controlled for. This is minimized in the report, here, and in the discussion.

The report tells us that models show that “an observed excess of cancer would not be expected because the cumulative doses were low." But such models are of dubious worth since they are based on very inexact cumulative doses of exposure and “existing knowledge of dose-response relationships of radiation risk."

Similarly, while it is noted that heavy metals that are of concern in Port Hope are human carcinogens, the Ontario Ministry of the Environment reported that levels observed in Port Hope (in 1991) “were not of sufficient magnitude to expect increases in cancer”. Here we have the additional limitation that exposures were measured at only one point in time.

The increases of circulatory disease found with radiation doses of one Gy or higher is a finding that is potentially very important. Circulatory disease is extremely common, being the leading cause of death in Western Society. Even a weak association between radiation exposure and circulatory disease would have a greater effect than a strong association between radiation and a rare disease. For example, the greater than 20 fold (2000 percent) increase in lung cancer deaths associated with heavy smoking has a lesser public health impact than the 1.4 fold (40 percent) increase in heart disease caused by heavy smoking.

2. Methods

A test for trend in over the time period of the study was done for most causes of death. But why would one expect a trend in time over the study period, if the exposures in Port Hope were (are) significant health predictors?

It is true that there were remediations done in the late 70s that would hopefully reduce the burden of some contaminants. But the long latency of many cancers would result in the effects (especially for the additional time inherent in mortality data) not appearing, for the most part, by the terminal time period of 1986-1997. It is also not clear if exposures were greatest in the 50s and 60s. Therefore, I would not expect most cancers to follow a trend line showing a decreasing rate in time from the 1950s to the 1990s.

The report, throughout, appears to ignore this long latency period from exposure to diagnosis and then to death that we know exists for many cancers. For example, the report states when referring to providing SMRs starting in 1956 “the importance of going back as early as possible is based on providing results for the time period before remediation actions were initiated to determine if greater effects were observed than in later periods." But for cancer, even mortality cases from the 70s would almost universally reflect cases that were initiated years and even decades earlier, before remediation took place.

3. Results

3a. Port Hope residence assignment

There were pockets of very low all cause mortality in Northumberland County; namely Hope Township and Hamilton Township. Both had just over half the expected number of deaths in the period 1986-1997. This contrasts with an SMR of 1.13 or 13% more deaths for all Port Hope residents over that period. However the entire county had an all cause SMR for all residents of 1.01 for the entire period. Assuming that Northumberland included Port Hope (not clear from text), the Northumberland SMR (excluding Port Hope) would have been noticeably lower than Port Hope’s.

3b. Mortality for Cancers Frequently Associated with Radiation and Childhood Cancer

Tables 3a-c are supposed to present SMRs for all cancers combined and radiosensitive cancers. Only lung, breast and leukemia were considered radiosensitive. Notable by omission were brain cancer (where radiation is certainly a suspected cause) colon cancer, stomach and several radiation related rare cancers.

Since there are a smaller number of cases in each time period, censoring due to confidentiality concerns is an even greater limitation than in the incidence study. When the number of cases is less than 5, the actual number of cases and the associated confidence intervals are not released. The result is still flagged as “statistically significant” at the 5% or lesser level, but no other information is given. Much information is lost when the results are dichotomized as either “significant” or “not significant”.

The mortality data is certainly much less useful in assessing childhood cancer in Port Hope. There were only 19 incident cases that occurred in the 26 year period from 1971-1996. Numbers were very small for most cancers resulting in a low ability to detect any high cancer rates that did exist (low statistical power). But the mortality data was much more limiting. Only 11 cancer related deaths occurred during the 42 year period covered by the mortality study. Statistical power, already a problem, is much lower here. For example, for all childhood cancers there was a 48% increase over expected rates and for leukemia a 63% elevation over what might be expected. We cannot say that these two results are not chance happenings, however, due to the low power. Given the lack of information (low statistical power) available to us in this data, statements about statistical significance are of little value.

While the mortality report emphasizes that all SMRs in these tables of “sentinel” sites for radiosensitivity were near one, actually lung cancer was elevated for men and women in different periods. Females showed an SMR for lung cancer of 1.15 over the entire mortality study period. The incidence study fortifies these findings since the female rates were significantly elevated for the 1986-1996 period and over the entire study period (p<.05).

What is notable are the radiosensitive data not highlighted in this section. Brain cancers in females appear at more than 2 times the expected rate in the 1986-1997 period (p<.05), similar to the incidence results. Children showed a near 50% excess of brain cancer over the mortality study period (p>.05) which is not inconsistent with the high rates noted in the earliest time period of the incidence study (p<.05). Colorectal cancer rates were high in women during the 1986-1997 period (SMR = 1.38 p=.1) which closely matches the corresponding incidence results using postal code (1.42 p<.05). I cannot see a valid reason for colorectal and brain cancer results to be omitted from this table since the results are provocative and both cancers have been associated with radiation.

The small percentage excess rates for circulatory disease noted are highly significant because of the great importance of this cause of death. Even small rises in rates have large impacts. High circulatory disease rates are the main reason that some of the all cause mortality rates are significantly high. Male rates all high over the entire study period while female rates were not markedly high until 1986. Female circulatory disease rates rose dramatically from the 1976-1985 period to the 1986-1997 period. There were over 100 more female deaths than expected in the 1986-1997 period. This surprising finding requires further scrutiny because one would not expect traditional risk factors to rise so quickly, particularly for only one sex.

3e. Comparison of Morbidity and Mortality

While mentioning the high brain cancer rate for women found in the 1986-1997 period, the mortality report states that “brain cancer was not significantly elevated in any other time period, or among men”. True enough, but this passes by the fact that the childhood brain cancer incidence rate was highly elevated (SMR = 4.17 , p<.05) during the 1971-1985 period. Female incidence rates were also statistically significantly elevated (p<.05) over the entire incidence study time period.

4. Discussion

4a. Summary of Main Results

Again the authors limit "sentinel” (for radiation) cancers to lung, breast, leukemia and thyroid. This conflicts with the literature (3) and even their own earlier discussion (section 1c.) which shows many more cancers associated with radiation exposure.

Nasal cancers are associated with heavy metal carcinogenic exposures in Port Hope, but are not discussed as a cancer “important in examining possible adverse health effects of industrial activity in Port Hope”. Nasal cancer incidence rates were significantly high for males, most notably in the 1971-1985 period with rates 5 times higher than expected. The mortality study does not indicate excess rates. Most probably this indicates that most of the small number of cases survived. This points out the weakness of mortality data for studying rare cancers that are not usually fatal.

Regarding leukemia, both the incidence and mortality results show a markedly (but not statistically significant) increase over expected rates in the order of 50% over the entire study period. In addition, rates were lower than expected since remediation that indicates that in the period 1971-1985 they must have been even higher. The statistical power, or the ability to detect true excesses is very limited for childhood cancers, which is acknowledged in the report. Yet later in the discussion we are told that “the absence of excess leukemia cancer rates is particularly reassuring”. This latter statement is irreconcilable with the findings of an excess noted with low power The fact that we cannot prove that the excesses noted were not due to chance fluctuations, is hardly grounds for rejoicing.

Then there is the major consideration of brain cancer in children. There was a markedly high rate before remediation found in the incidence study. The mortality study does not contradict the incidence study. It shows an excess in brain cancer over the entire period that is not statistically significant. However, the data is very limited by the very small numbers and the fact that the results are not shown for each time period.

When one considers all of these factors, the discussion of the childhood cancers does not accurately reflect the findings and the inherent limitations of the studies.

The small percentage increases in circulatory deaths, noted in this section, is very significant in terms of number of lives this represents. A small increase in a very common disease may be more important than a very large increase in a rare disease. For, example, during the 42 year study period there was a modest 15% excess over expected in circulatory deaths. But this represents over 300 excess deaths (more than 7 a year). Clearly, more study of the circulatory disease risk factor profile in Port Hope is warranted in an attempt to determine if lifestyle or environmental contamination is most likely responsible for the findings, particularly in light of the unusual female patterns.

4b. Strengths and Limitations of the Report

On page 10 there is a thorough discussion of the effects of misclassification on epidemiological studies. Misclassification occurs when individuals are placed in the wrong exposure or disease categories. Generally, misclassification biases results towards the null. This means that real excesses in rates are more likely to go undetected. The results of misclassification will normally lead the observed SMR to be lower than the true one. As the report states, the possibility for misclassification is even greater for mortality data since there is additional time (and possibly reasons) for individuals to migrate between the time of diagnosis and death.

The problem is that the authors do not appear to take their own beliefs to heart. Most of the analyses, presentations and interpretations are based primarily on strict statistical criteria (statistical significance at the 95% level). Results are important and noted if they meet this requirement and are dismissed if they do not. This is inappropriate for at least three reasons. One is that the results of misclassification are probably significant and most likely reduce the observed SMRs from their true value. Secondly, as the report states, even if there are about 10 cases of cancer, the probability would only be 80% for picking up a true doubling of the cancer rate. In many of the cancers looked at here, the numbers are considerably smaller in many time and gender categories (i.e. leukemia, brain cancer). Also very important is the fact that the standard rates are almost certainly biased upward. This is because Ontario contains many other areas of contamination as well as big cities that generally have higher cancer rates than small areas like Port Hope. This will again lower the SMR. Certainly one has to look at the patterns of SMRs, not just the patterns of statistically significant ones, to most properly analyze data such as this.

The report points out that “over the four periods 1956-1965 through to 1986-1997 respectively there were 0, 5, 6 and 12 disease groupings significantly elevated and 0, 2, 4 and 3 significantly decreased”. This means that even using these strict statistical criteria, in all of the three later time periods, there were more “high” rates than “low” rates. There is only 1/8 probability of this occurring by chance.

The authors state that “the increased mortality observed for esophageal cancer, colorectal cancer, circulatory disease and cirrhosis suggests the possibility of some common lifestyle risk factors”. This may be true, but they do not point out that at least circulatory disease and colorectal cancer are also associated with radiation exposure.

4c. Confounding Risk Factors

Regarding brain cancer mortality we are told that “it was not increased in the previous period (1976-1985) or among men”. However, as previously stated, the incidence results show an increased SIR (Standardized Incidence Ratio) for women for the previous period and both sexes had increased SMRs for the period previous to that one (not statistically significant but still “increased”). Children are not discussed even though their incidence rates are very high and there is no mortality data by period. The mortality data is of little use since there was considerable survival and therefore very small numbers of deaths.

Near the end of page 11, the discussion turns to the ability of ionizing radiation to induce brain/nervous system tumours. The report states that most radiation associated tumours are “benign”. This is of small comfort, since many so called benign brain tumours have considerably mortality due to their position and the risks and difficulties associated with removal. The report does concede that ionizing radiation may induce some types of malignant tumours and some benign tumours. The risk is greater when the exposure occurs during childhood. Despite this discussion, brain cancer is not considered a “sentinel cancer” for radiation in this report. Brain cancers can be induced by ionizing radiation. The only argument is over the minimum dose at which this occurs. The history of medicine has repeatedly shown us that low dose effects are often underestimated.

Since radiation exposure is a factor of interest, it is not a confounder. Therefore it shouldn’t be discussed here as a confounder for brain cancer. Similarly, radon is discussed here with respect to lung cancer when it is a factor of interest rather than a confounder.

The public health importance of the high circulatory disease rates in Port Hope and the lack of information regarding the cause of the high rates make it a priority for investigation. Given these conditions; it is inappropriate that this is dismissed out of hand as not due to exposure. The reasons given are the low exposures and the high rates in the rest of Northumberland County. But the effects of prolonged low radiation exposures are not known, particularly in conjunction with heavy metal exposures.

4d. Concluding Remarks

The first paragraph in this section misrepresents the results. Surprisingly we are told that “the absence of an increase in leukemia rates and other radiosensitive sites, does not support the hypothesis that radiation exposures in the community impact on residents’ health. The absence of excess leukemia rates amongst children is particularly reassuring”. It depends on how you define “excess” and “radiosensitive sites”. As discussed earlier, the leukemia rates in children are well above those expected. The fact that they are not statistically significant is far from reassuring given the low statistical ability to detect real excesses due to small number (low power). Radiosensitive sites are defined in a limited fashion, not including brain cancer and colon cancer which both show some disturbing patterns. Certainly the raised leukemia rates, which were even higher before remediation are not reassuring. Along with the brain cancer and colon cancer results and some of the rare cancers, the available evidence points to their being problems in Port Hope. The fact that this evidence, due to its limited nature is not definitive should not be taken as reassuring. Again strict criterion of statistical significance is used for decision making when information available in this report show such a method to be inadequate and inappropriate.

Are the results of the mortality and incidence childhood cancer studies reassuring? Actually the available data, having the limitations alluded to, points strongly in the opposite direction. Firstly, in the mortality study, all 3 individual cancers looked at showed SMRs greater than one. There is only a 1/8 probability that this would occur by chance.

Referring to brain cancer, the increases noted in women are discounted since “the lack of any excess among men in any period does not provide support for an environmental hypothesis”. This assumes that men and women have been subjected to the same environment. But through personal communication, I have learned that the women were more likely to play with children in some of the high risk areas. Men were more likely to be working. The childhood results are ignored where in an earlier period the incidence was several times the expected rate. The pattern of childhood and female brain cancer increases is consistent with an environmental effect.

The second paragraph makes the point that cancer mortality rates were close to Ontario rates. As noted, the patterns of several cancer rates show cause for concern in that the patterns are consistent with environmental contamination. It is most probable that the Ontario standard rates are higher than a community like Port Hope should have in the absence of major unusual environmental contaminants. It follows that the SMRs that have been calculated generally underestimate the risk in Port Hope. Under normal conditions, because of a “healthy community effect” we would expect most SMRs to be lower than one. Along with the effect of misclassification bias, which is probably larger in mortality studies, the calculated SMRs may be considerably smaller than the true values would be. Given these arguments, there is nothing about the results here that could be deemed reassuring unless your criteria of reassurance is that there be no obvious gross excesses of cancer in Port Hope.

III. Analyses of Results

Several cancers were identified to be of special interest in the incidence study review (1). These were cancers for which there was evidence of increased incidence and there was at least some evidence that the disease may be related to known exposures in Port Hope.

The mortality and incidence study results are compared in Table 1 for these cancers (as well as some additional ones). The fact that none of the time periods are exactly the same makes comparison more difficult. The roughly corresponding time periods for the two studies (e.g. - 1986-1996, 1986-1997) are paired to facilitate comparisons.

General Comments

The incidence and mortality results usually showed similar risks. Despite this, the incidence results are more likely to be statistically significant due to the power issues discussed above. Some causes deserve the more detailed discussion that follows.

All causes (death only):

The SMRs show elevated death rates (p<.05) for males for all time periods and for the total period from 1956-1997. For women, the death rate from all causes is elevated for the entire 42 year period, but this is principally due to a markedly statistically significant (p<.05) elevated rate in the 1986-1997 period. There were 16% more deaths than expected in Port Hope women in this period. Men and women combined (overall heading) show statistically elevated rates for the 1986-1997 time period and for the entire time period. Children show an elevated SMR of 1.18 for the 1986-1997 time period only (p >.05).

All Cancer

Both incidence and mortality results show males to be near expected rates. Women show small excess in rates in the 1986-1996,7 periods when looking at mortality or morbidity data (p>.05). Over the entire study period, there was a statistically elevated female cancer incidence rate, but the mortality rate was near the expected rate (SMR = .99). This difference appears to be due to lower than expected female cancer death rates in the early years covered by the mortality study which were not covered in the incidence study,

The morbidity and mortality rates are very similar for children. Children show almost 50% more cancer cases and deaths than would be expected during the total study period. The SIR for the 1971-1985 period was statistically significant a the 10% level.

When the data for men and women were combined, the rates are not markedly elevated.

Brain Cancer

Brain cancer was discussed at length in the Incidence Study Review (1) because the pattern of elevated rates suggest strongly that there may be a problem. The mortality data are more sparse, particularly for children, where the incidence study yielded a SIR of 4.17 (p<.05) for the 1971-1985 period. Mortality data are only presented for the entire time period where the higher than expected death rates in children were not statistically significant.

Where there is corresponding data, there are not great disparities. For example, women show statistically significant and similarly elevated brain cancer rates in the 1986-1996,7 periods whether considering deaths or incident cases. The only possible disparity is that males show high incident rates in 1986-1996 but lower than expected mortality rates. Men and women combined show similarly increased (p>.05) incidence and mortality rates over the entire period.

Lung Cancer

Lung cancer death SMRs are generally lower than the corresponding SIRS for an unexplained reason. The incidence and mortality patterns are similar for women, however, as the rates are high in the 1986-1996,7 period (p<.05 - incidence rates).

Ovary

Ovarian Cancer rates are low, particularly in the 1986-1996,7 periods. The incident rates show a statistically significant decrease (p<.05).

Colorectal

The incidence study showed high rates for women in the 1986-1996 period. The mortality study showed a similar 38% higher number of deaths than expected. The only difference, is that as has often occurred, the incidence results are statistically significant while the mortality ones are not, due to the lower power. Men do not show a marked increase in this cancer in either study, although the mortality rates are higher in the two later time periods.

Breast Cancer

There is a hint of elevated female breast cancer incidence rates (not statistically significant) but the mortality rates are close to expected values.

Rarer Cancers
(Lip/nose and sinuses/bone)

All of these rarer cancers showed statistically significantly elevated incidence rates for men in at least one time period. Nasal cancer is associated with the heavy metal exposures of Port Hope. This is not reflected in the corresponding mortality results where there were no significant elevations and all SMRs were displayed as <1 or >1 due to confidentiality restrictions. Clearly the mortality data is not very useful for looking at very rare cancers because of universally small numbers.

Leukemia

This cancer that has a relatively large impact on children and known to be radiation sensitive; is of interest for both reasons. Adults show no evidence of problems. As noted in the incidence review, children demonstrate a 41% elevated rate for the entire incidence study time period. This result while suggestive, is not statistically significant. The mortality results are only available summed over the time periods and show a similar but slightly higher SMR.

NHL

Children showed statistically significantly elevated incidence rates for the 1986-1996 period and for the entire period. The mortality study again gives little information, showing a confidential computation for the entire study period only. It shows a rate over one that is not statistically significant. Less mortality data provides less information that limits interpretations and conclusions.

Esophageal

There is evidence that male esophageal cancer rates have been elevated in Port Hope. For the earliest incidence period and the nearest corresponding mortality study period; the SIR and SMR are both well over 2. The incidence result is statistically significant while the mortality result is of borderline statistical significance (p=.05). The mortality SMR shows a 50% increase over expected rates during the entire study period (p >.05). The incidence study showed women to have higher than expected rates in all periods and a 50% excess rate for the entire period (p>.05). The mortality rates were near expected ones for women.

For men and women combined, both incidence and mortality rates were statistically significantly high for the 1971-1985/1976-1985 periods. The SMR for the 66-75 period was 1.86 (p>.05). For the entire time period, both rates were greater than one, but not statistically significant (p>.05).

IV. Conclusions

1. Regarding cancer, when we use mortality data (June 2002 Mortality Report) instead of incidence data, we lose a large proportion of the cases. This is crucial given the small numbers and the resulting low statistical power available to analyze most cancers, even when using incidence data. The ability to detect real increases (statistical power) in many cancers is inadequate and much lower than in the incidence study.

2. Since there are a smaller number of cases in each time period, censoring due to confidentiality concerns was an even greater limitation in the June 2002 Mortality Report than in the August 2000 Incidence Report. When the number of cases was less than 5, the actual number of cases and the associated confidence intervals were not released. The result is still flagged as “statistically significant” at the 5% or lesser level, but no other information is given. Much information is lost when the results are dichotomized as either “significant” or “not significant”.

3. It follows from 1. and 2. that for the study of cancer in Port Hope, incidence data is more informative than mortality data.

4. Most of the analyses, presentations and interpretations in The June 2002 Mortality Report are based primarily on strict statistical criteria (statistical significance at the 95% level). Results are important and noted if they meet this requirement and are dismissed if they do not. This is inappropriate for several reasons including the following:

• there are significant biases in the mortality study that would most probably bias results towards the null value (no excesses)

• the ability to detect true increases (statistical power) is very low for many of the less common cancers

• the large number of comparisons will result in many chance significant findings that must be separated from true excess rates by using qualitative analyses (i.e., patterns and plausibility) along with statistical criteria.

1. Tables 3a-c are supposed to present SMRs for all cancers combined and radiosensitive cancers. Only lung, breast and leukemia were considered radiosensitive. The definition of cancers “sentinel” for radiosensitivity is arbitrary as several radiosensitive cancers are not highlighted. A corresponding chart for cancers associated with heavy metal contamination is not presented.

2. The incidence (August 2000 Incidence Report) and mortality (June 2002 Mortality Report) cancer results generally show close agreement in direction and trend. The incidence results were more often statistically significant, however, due mainly to the more limited information contained in the mortality data.

3. The results of the June 2002 Mortality Study do not in any way contradict the main conclusions of the Incidence Study Review (1). These major conclusions are listed below:

• for several common cancers the evidence from this study suggests that females have high rates (lung cancer, colorectal cancer, all cancers). The SIRs were generally higher than the corresponding ones for males. These findings suggest differential exposures by gender.

• If confounding explains the pattern of lung cancer and colorectal rates it would have to be greater for females. Such a scenario is not a likely one.

• The findings suggest that children have experienced high cancer rates particularly before 1986. The pattern of cancer rates in children is consistent with effects from the higher exposures before remediation.

• The findings taken together show a pattern that is quite suggestive of there being an excess of brain cancer in Port Hope.

• The Canadian Cancer Registry confidentiality restrictions when there were less than five cancers further constrained a study which relied on limited data.

• Further research should look into the feasibility of a case-control study for suggestive sites.

1. The small percentage excess rates for circulatory disease noted in The June 2002 Mortality Report are noteworthy because of the great importance of this cause of death. Even small rises in rates have large impacts. This is the principal reason that some of the all cause mortality rates are high.

2. The reasons for the high circulatory disease death rates are not clear. Female circulatory disease rates rose dramatically from the 1976-1985 period to the 1986-1997 period. There were over 100 more female deaths than expected in the 1986-1997 period due to this cause. This surprising finding requires further scrutiny because one would not expect traditional risk factors to rise so quickly, particularly for only one sex. It is important to determine if the radiation and heavy metal exposures in Port Hope are contributing factors for these findings.

3. When one considers all of the inherent biases and limitations of the data, the discussion of the childhood cancers does not accurately reflect the findings. The statistical power, or the ability to detect true excesses is very low for childhood cancers, which is acknowledged in the report. Yet in the discussion we are told that “the absence of excess leukemia cancer rates is particularly reassuring”. This latter statement is irreconcilable with the findings that show a considerable excess along with very low statistical power Given these conditions, the fact that we cannot prove that the excesses noted were not due to chance fluctuations, is hardly grounds to feel reassured.

4. Similarly the abstract of the June 2002 Mortality report tells us “that the lack of evidence of elevated lung cancer is reassuring." The evidence tells us that for women, both the incidence and mortality were elevated, particularly during the period of 1986-1996. The incidence rates were statistically significantly elevated (p<.05). It is incorrect to call the lung cancer situation “reassuring” in light of these facts.

5. The report, throughout, appears to ignore the long latency period from exposure to diagnosis and then to death that we know exists for many cancers and other chronic diseases. For example, for cancer, even mortality cases from the 70s would almost universally reflect cases that were initiated years and even decades earlier, before remediation took place. This contrasts with assertions made in the report.

6. Nasal cancer incidence rates were significantly high for males, most notably in the 1971-1985 period with rates 5 times higher than expected. The mortality study does not indicate excess rates. Most probably this indicates that most of the small number of cases survived. This points out the weakness of mortality data for studying rare cancers that are not usually fatal.
V. References

1. Mintz, E. Morbidity Study Review. August 2000.

2. Lees, RE, Steele R., Roberts JH. A case-control study of lung cancer relative to domestic radon exposure. Int J Epidemiology 1987;16(1):7-12.

3. Durham Region Health Department. Radiation and Health In Durham Region. November 1996.

4. Ulm K. A Simple Method to Calculate The Confidence Interval Of a Standardized Mortality Ratio (SMR). American Journal of Epidemiology. 1990; 131(2):373-375

Back