Cancer Clusters: Findings Vs Feelings

David Robinson

In This Article

Why Are Investigations of Potential Environmental Cancer Clusters So Often Inconclusive?

A variety of factors often work together to create the appearance of a cluster where nothing abnormal is occurring. Looking for clusters is analogous to drawing a bull's eye after you have thrown darts at the wall at random. In this situation, there is possibly a place in which a bull's eye can be drawn that will leave multiple darts in close proximity to some common center. According to the American Cancer Society, cancer was diagnosed in an estimated 1,268,000 Americans in 2001.[9] Finding clusters in cancer data is, thus, something like looking for patterns in the location of more than a million darts thrown at a dartboard the size of the United States.

The definition of what geographic area is to be investigated in a cancer cluster study is often problematic. If the hypothesis that cancer rates in a certain area may be elevated provides the initial impetus for the study, the natural temptation is to study only the area that includes the cases that inspired the study. This problem is called "preselection bias" because it involves researchers preselecting the geographic area of a study based on what they already know an investigation of certain areas would reveal.[10] In much the same way as gerrymandering -- including certain voters in an electoral district -- can shape the outcome of elections, preselection bias -- including certain patients in the geographic area of a study -- can shape the outcome of cancer cluster investigations.

The problem of "drawing the bull's eye" applies not only to space, but also to time. A study of 2 clusters in an Ontario town noted that "the tendency is to include all years in which cases were reported [in the date range chosen for analysis], thereby maximizing, and magnifying, any effect which may be present."[11]

A third way in which the bull's-eye problem can skew results is in the selection of which cancer to include as part of a possible cluster. In the case of possible pediatric cancer clustering in Toms River, New Jersey, investigators began by looking at every category of childhood cancer and included in their investigation those categories of cancer whose rates were significantly elevated in Toms River.[12] The threshold of significant elevation that was used meant that for every 20 cancer categories examined, 1 would qualify as significantly elevated. These problems can be even further increased when more categories are considered -- for example, age groups and gender.

These sorts of expansions are problematic because the greater the number of possible cancers, areas, and time periods that are evaluated as potential clusters, the greater the chance that randomly distributed cases will appear as a cluster. In addition, the links that have been proven between exposure to carcinogenic chemicals and elevated incidence of cancer have entailed elevated rates of extremely specific cancers: DES, in high doses, elevates risk of vaginal adenocarcinoma, exposure to VCM elevates risk of hepatic angiosarcoma. One thing these documented instances of elevated prevalence have in common is that the chemical agent consistently elevates risk of a specific cancer, not of all cancers equally.

Often in these debates, however, a burgeoning set of effects is putatively linked to a single cause. This is aptly illustrated through one of the most widely publicized cancer cluster cases in recent years, the Erin Brockovich case. Dramatized in a major film with Julia Roberts portraying Ms. Brockovich, a paralegal who worked with local residents, the case dealt with the release of chromium-6 into the Hinkley, California water supply by Pacific Gas and Electric. The suit blamed the chemical for dozens of symptoms, from nosebleeds to breast cancer, miscarriages, Hodgkin's disease, and spinal deterioration.[13] Workers who inhale large amounts of chromium-6 over long periods have been shown to be at elevated risk of developing lung and sinus cancers. But chromium-6 has never been shown to be related to any other human cancer, or to be carcinogenic to any degree when dissolved in drinking water.[13]

Some clustering is to be expected as the result of chance alone.[13] It is reasonable that people should seek explanations for higher-than-expected cancer rates, but epidemiology does not always offer an identifiable cause.

Sometimes, public pressure can impel public health officials to undertake an investigation they do not believe is warranted. Investigations undertaken after experts have concluded that nothing out of the ordinary is occurring are unlikely to produce noteworthy results.

Community members who raise concerns about possible clusters will frequently explain themselves in terms of a "common sense" feeling that something is wrong. Often, they are not inclined to wait patiently for an in-depth, methodical investigation by public health authorities.

An investigation into childhood cancer in Toms River, New Jersey, provides insight into the pressures that can work against balanced scientific inquiry. Toms River is the location of 2 "Superfund" sites, places the Environmental Protection Agency (EPA) has designated as a high priority for clean-up due to the presence of hazardous waste.

A nurse in a Philadelphia pediatric oncology ward noticed that many of her patients were from the Toms River area, and speculated that an environmental cause might be elevating the pediatric cancer rates in Toms River. When parents brought their concerns to the attention of state authorities, in 1996, the state evaluated the cancer rates and found no cause for alarm. A spokeswoman from the New Jersey Department of Health explained that the state, based on existing data about cancer rates, did not think a comprehensive cluster investigation would be economical or useful, because the numbers of childhood cancers were "not statistically meaningful."[14]

Nonetheless, the state moved to address community concern with a series of investigations into possible sources of cancer risk, including the Superfund sites. The parents brought a sense of urgency to the discussion. "This is a terrible disease, and these kids suffer.... These kids don't have time to wait. I have two other children, and I'm scared to death," said one mother of a childhood cancer victim.[15]

"In my heart and in my mind, I have no question. Now, it's up to the scientists to use logic and common sense to get at the truth," said Linda Gillick, chairwoman of a citizen's committee organized to address the issue and the mother of another cancer victim.[16]

Where parents were certain, scientists were not. The data on cancer rates that were available when community members first raised concerns did not show more cancer than scientists might have expected to be found in a random distribution in Toms River. Residents prevailed on their congressional representatives to ask federal officials for an investigation that state health officials said would be futile.[14] Ultimately, the study was undertaken as a joint effort between state officials and the federal Agency for Toxic Substances and Disease Registry.[14]

As part of her group's effort, Linda Gillick traveled to Washington, DC to defend a special line-item allotment of $1 million for the Toms River study in one of Congress's annual appropriations bills.[17] Ultimately, Congress passed the item.

Concerned citizens thus had a doubly decisive impact on the issue. After convincing Congressional representatives to circumvent state cancer experts and launch a federal investigation, the citizen activists intervened again to increase the funding for the study over the amount allotted it in the normal budget process. At both junctures, public concern and fear overrode the decisions of administrators charged with setting public health priorities based on scientific findings.

The study, which took more than 5 years to complete, concluded that "no single risk factor evaluated appears to be solely responsible for the overall elevation of childhood cancer incidence in Dover Township."[18] The study found that most of the childhood cancer cases in the area have no explanation; the only supportable environmental link was that between prenatal exposure to contaminated drinking water and pediatric leukemia in girls.

Dr. Eddy Bresnitz, a New Jersey state epidemiologist, explained that even the narrow relationship found in the study might be a fluke. "Due to the relatively low number of study subjects and other factors, chance cannot be excluded as a possible explanation for the findings."[19]

"You can't have a child with leukemia living two houses down from a child with a tumor, drinking the same water and breathing the same air, and tell me they didn't get cancer from exposure," Linda Gillick told the New York Times. "That's my common sense speaking."[20]

Scientific studies linking elevated cancer risk to environmental causes have generally involved years-long latency periods between exposure to carcinogenic factors and development of cancer. The DES cases did not become obvious until more than 10 years after its use, and exposure to VCM in vinyl plants takes years to cause cancer. Even smoking and sun exposure, the two most widely documented avoidable cancer risk factors, can take half a lifetime to make their effect apparent. The latency problem surfaces in two ways in community-inspired cancer cluster investigations.

First, some of the people who were exposed to the environmental chemical under investigation may have moved away from the area before the investigation began. If they subsequently develop cancer in their new homes, their absence diminishes the perceptibility of the cluster. If they remain healthy, their absence from the area effectively increases the apparent magnitude of the cluster.

Second, it is possible that some of the cancer cases that occur within the investigated area may not be attributable to the local environment. If some of the people who are diagnosed with cancer moved into the area shortly before being diagnosed, steps must be taken to assure that their cancer cases are not attributed to local causes.

The most significant problem plaguing data about possible cancer cases is that cancer is typically not a reportable disease. The government keeps extensive, complete records of the incidence of many infectious diseases -- such as tuberculosis and venereal disease -- in order to track and counter potential outbreaks. For cancer, however, no such record exists. Recently, several states have begun statewide cancer registries. These are helpful to some degree, but they lack historical data, are plagued by physician compliance problems, and may not be able to keep accurate account of diagnoses made out of state. This last issue is particularly problematic, since many definitive cancer diagnoses are made at major medical centers for patients who come from out of state in search of top expertise.[21]

These problems can lead to under- or overascertainment of the number of actual cancer cases within a given area, and may also not be spatially neutral. If fewer cases are detected near a state border, for example, because parents are having their children diagnosed in the next state, this may lead to an artificial impression of spatial clustering.

In the absence of reporting requirements, the NCI runs a program called SEER (Surveillance, Epidemiology, and End Results), which documents cancer prevalence in a sample of the US population to determine the baseline levels of various cancers. The program uses information from hospitals, pathology laboratories, physicians, and death certificates to determine who has cancer, supplemented by population surveys. The SEER program has been in operation since 1973 and has quality control procedures in place that maximize the accuracy and completeness of its results. In addition, many states provide additional support for the maintenance of cancer registry information beyond that provided by SEER.

These programs are helpful, but long-term historical information about cancer incidence is only available for some parts of the country. The population surveyed by SEER -- a subset of the total US population -- is designed to be a representative sample of the national population. If the local area in which a cancer cluster investigation is conducted differs demographically from the national population, the expected cancer levels established by SEER may not apply to the area being studied.

The data collection problem is significant because the only way to determine whether the cancer rate in a community is abnormally high is to compare it with an expected rate. The expected rate forms the "denominator" in a prevalence figure, the normal level of cancer that is used as a reference to determine whether the rate in a given area is elevated. Without an accurate expected rate, there is no way to decide whether the level of cancer in a given community is cause for concern.

As discussed above, the nature of random distributions is such that some amount of clustering may be expected to occur simply by chance. It is conventional among scientists to regard an elevated cancer rate as "statistically significant" if chance alone would produce as much or more elevation less than 5% of the time. This is commonly written in the scientific literature as "P < .05," where P is the probability of seeing such an elevation if only chance is at work. With this criterion, if one examines the cancer rates in 100 neighborhoods, and cancer cases are occurring randomly, one should expect to find about 5 neighborhoods with statistically significant elevations.

Any unusual amount of cancer will tend to provoke concern, regardless of whether it stems from chance or a more concrete cause. As a result, the finding that there is a substantial elevation in cancer rates suggests that further investigation into possible causes may be warranted, but does not in itself establish that any particular cause is at work.

When a group of people who live in geographic proximity to each other exhibit an elevated rate of cancer, the rate may reflect characteristics other than geography that those in the affected area share. Characteristics like similar diets and exercise patterns may tend to be geographically "clustered" because low-income people who eat disproportionately more fatty foods live near each other, because health-conscious suburbanites live in the same neighborhood, or because rates of smoking tend to differ from one community to the next. In any of these cases, a geographic cluster might be proved to exist even if there were no chemical carcinogen in the environment.

No matter how many possible environmental cancer causes are contemplated, it will always remain possible that some heretofore ignored chemical in the environment is elevating cancer rates. Thus, investigations of possible environmental causes for cancer can be extended almost indefinitely, as more and more possible carcinogens are examined.

The many ways in which a "bull's eye" can be drawn, problems of latency, the lack of reportability of cancer, the similar behaviors and backgrounds of people who live near each other, and the vagaries of chance all reduce the likelihood that investigations into proposed environmental cancer clusters will confirm environmental hazards as a source of human cancer. Such confirmation may be achieved in the future, and these problems are not reasons to dismiss efforts to identify environmental cancer clusters. But it is also not advisable to ignore these conundrums simply because their presence hinders efforts to find a cause of cancer.


Comments on Medscape are moderated and should be professional in tone and on topic. You must declare any conflicts of interest related to your comments and responses. Please see our Commenting Guide for further information. We reserve the right to remove posts at our sole discretion.