Available via license: CC BY-NC 4.0
Content may be subject to copyright.
Numeracy
Advancing Education in Quantitative Literacy
*:7@80
F
>>@0 =?4.70
Cancer Clusters in Delaware? How One
Newspaper Turned O!cial Statistics into News
Victor W. Perez
University of DelawareA4.?:=;@/070/@
Joel Best
University of Delaware5:07-0>?@/070/@
Rachel J. Bacon
Pennsylvania State University=5-,.:9@/070/@
@?3:=>=0?,49.:;D=423?:1?304=8,?0=4,7@9/0=, =0,?4A0:88:9>#:9:880=.4,7M=4-@?4:9!4.09>0
&0.:8809/0/4?,?4:9
%0=0E*4.?:=+0>?:07,9/,.:9&,.307,9.0=7@>?0=>4907,B,=0:B$90#0B>;,;0=(@=90/$H.4,7
'?,?4>?4.>49?:#0B> Numeracy*:7>>=?4.70
$3M;/C/:4:=2
A,47,-70,?3M;>.3:7,=.:88:9>@>10/@9@80=,.DA:74>>,=?
Cancer Clusters in Delaware? How One Newspaper Turned O!cial
Statistics into News
Abstract
I0K,2>34;90B>;,;0=1:=?30>?,?0:107,B,=0?30 News Journal3,>-00949>?=@809?,749/4>>0849,?492
491:=8,?4:91=:8>?,?02090=,?0/=0;:=?>:1 cancer clusters ?:4?>=0,/0=>:A0=?30;,>?D0,=>I0>?:=40>
;=:A4/0.:7:=1@78,;>:1.09>@>?=,.?>/0>429,?0/,>.7@>?0=>:L09:9?301=:9?;,20,9//0?,47?30?D;0>:1
070A,?0/.,9.0=>1:@9/49?30>0?=,.?>,9/?30;@=;:=?0/=07,?4:9>34;:1070A,?0/.,9.0==,?0>?:7:.,749/@>?=D
;:77@?4:9I:@23?30 News Journal ,7>:;=:A4/0/4?>=0,/0=>B4?3,/A4.0,-:@?49?0=;=0?492?30>0/,?,B4?3
.,@?4:94?@9.=4?4.,77D;=0>09?0/?30>0/,?,)>492?30>?,?0G>@9@>@,7/0J94?4:9,9/80,>@=0809?:1070A,?0/
.,9.0=49.4/09.0,> cancer clusters4??=,9>1:=80/<@0>?4:9,-70>?,?4>?4.>49?:,9,7,=8492;@-74.4>>@0I4>
,=?4.70.=4?4.,77D0C,8490>?30>090B>=0;:=?>,9/?30>?,?02090=,?0/=0;:=?>?30D@?474E0/
Keywords
;@-74.30,7?3/4>0,>0.7@>?0=>>?,?4>?4.>
Creative Commons License
I4>B:=64>74.09>0/@9/0=, =0,?4A0:88:9>M=4-@?4:9#:9.:880=.4,7!4.09>0
Cover Page Footnote
*4.?:=+%0=0E4>,9>>4>?,9?%=:10>>:=49?300;,=?809?:1':.4:7:2D,9/=4849,7@>?4.0,??30
)94A0=>4?D:107,B,=04>=0>0,=.349?0=0>?>49.7@/0?30>:.4,7.:9>?=@.?4:9:130,7?3,9/09A4=:9809?,7
69:B70/200?0,.30>.:@=>0>4949?=:/@.?:=D>:.4:7:2D=0>0,=.380?3:/>,9/>?,?4>?4.>,9/09A4=:9809?,7
,9/30,7?3>:.4,78:A0809?>
:070>?4>%=:10>>:=:1':.4:7:2D=4849,7@>?4.0,??30)94A0=>4?D:107,B,=04>8:>?=0.09?-::6>
,=0 e Student Loan Mess B4?3=4.0>?,9/ Kids Gone Wild B4?3 ,?37009:270
&,.307,.:94>,2=,/@,?0>?@/09?,?%099'?,?0)94A0=>4?D'304>B:=6492?:B,=/>30="49':.4:7:2D
,9/08:2=,;3DB4?3.:9.09?=,?4:9>4930,7?3,9/>;,?4,780?3:/>
I4>,=?4.704>,A,47,-7049#@80=,.D 3M;>.3:7,=.:88:9>@>10/@9@80=,.DA:74>>,=?
Introduction
The principal headline on the front page of the Sunday, September 12, 2010, issue
of Wilmington, Delaware’s News Journal must have alarmed many readers:
“Mapping Out Clues to a Cancer Mystery”; the subhead was “Study Focuses on
Clusters to Find Cause of Del.’s High Rates” (Montgomery 2010: A1). Next to
the story was a small map of the state and readers were invited to view an
“enlarged, detailed map” on page A6. In fact, the larger map covered about half
the page and identified 45 census tracts that had unusually high incidence rates
(i.e., new cases) of cancer (see Figure 1).
Delaware is a small state—less than 100 miles long, and only 9-35 miles
wide— and the News Journal is the state’s principal daily newspaper. Circulation
is highest for the paper’s Sunday edition, thus the story was meant to attract
maximum attention. Moreover, many readers probably recognized this story as
one in a series, originating with the 2007 front-page headline that reported a
“confirmed cancer cluster” near the Indian River power plant in Millsboro had
been identified (Nathans 2007: A1). Several similar pieces in 2008 examined the
cancer cluster issue, and after the 2010 front-page story, another would appear in
2014—each accompanied by its own maps (Barrish 2008; Barrish and
Montgomery 2008; Miller 2008; Montgomery 2008; Montgomery and Miller
2014; Shortridge 2008).
These stories made frequent reference to a familiar fact—familiar at least to
Delaware residents—that the state has had higher cancer rates than the national
average. Many residents assume that this is related to the long-standing presence
of the chemical industry in Delaware, that pollution from chemical plants and
other types of industry found around the state undoubtedly explains elevated
cancer risks. For instance, the News Journal quoted one woman (Montgomery
and Miller 2014: A8):
My husband is a cancer patient. I lost my mother to cancer. Everybody I know, truly
everybody, has a close connection to cancer in some way. I think my whole family has
always assumed it was the chemical companies, and we are in some way paying the price
for pollution.
Our purpose is to critically examine these newspaper stories, particularly the
ways their maps communicate information about public health, and to further
examine the reports from state agencies that provided the basis for these news
reports. In a sense, these stories presented data and let the facts speak for
themselves, except that the impression conveyed was both alarming and
misleading.
1
Perez et al.: Cancer Clusters in the News Media
Published by Scholar Commons, 2015
Figure 1. From The News
Journal, September 12, 2010
© Gannett. All rights
reserved. Used by
permission and protected by
the Copyright Laws of the
United States. The printing,
copying, redistribution, or
retransmission of this Content
without express written
permission is
prohibited.
http://www.delaw
areonline.com/
2
Numeracy, Vol. 8 [2015], Iss. 1, Art. 7
http://scholarcommons.usf.edu/numeracy/vol8/iss1/art7
DOI: http://dx.doi.org/10.5038/1936-4660.8.1.7
Thinking about the 2010 Cancer Map
Consider the map in Figure 1. It identifies 45 census tracts with high rates of
different types of cancers, and features notations explaining which types have
elevated rates in 14 of the tracts. The elevated cancer incidence rates of ten of the
census tracts are included in a “Top Ten” list, while 4 census tracts are referred to
by way of text pop-outs on the map, describing how much more cancer they have
(in percent) relative to the state as a whole. Further, note the types of cancer
identified in the census tract areas included in the “Top Ten” list: 1 tract had an
elevated rate for one type of cancer, 5 for two types, and the remaining 4 for three
to five types (Montgomery 2010). Overall, the map mentions elevated rates for at
least 10 different types of cancer (i.e., prostate, ovarian, etc.).
Right away the map presents a statistical issue. The story notes that
Delaware has 196 census tracts, and that the state report classified rates for 23
types of cancer. Delaware census tracts average approximately 4,000 residents,
but the number of residents in each tract varies greatly across the state.
Additionally, for rarer types of cancer a tract might have only “one or two new
cases a year to a few dozen” (Montgomery 2010: A6). If the number of residents
in a census tract is small, small fluctuations in the number of cancer diagnoses can
result in significant fluctuations in the incidence rate. Additionally, if the number
of cancer diagnoses in a tract is small, even in tracts with higher numbers of
residents, confidence intervals can be wide and denote a significant amount of
uncertainty in interpreting the incidence rate. Thus, it is easy for rates to vary in
both thinly populated tracts and in tracts with fewer diagnoses, both resulting in
apparent “clusters” of cases (Gelman and Nolan 2002). For example, the census
tract located west of Dover, Delaware’s capital, was number 4 on the top-ten list
of census tracts with elevated incidence rates, having a rate of 817.5 per 100,000
for the time period 2002-06 (DHHS 2010a). The average number of cancer
diagnoses in this census tract was 19.2 cases per year for the years 2002-06
(DHHS 2010b), and it had a population of approximately 4,000 during this time
period (the Census estimate for the time period 2005-2009 is approximately
4,400) (U.S. Census Bureau 2014). Ultimately, this tract had a cancer incidence
confidence interval ranging from a lower limit of 647.2 to an upper limit of
1018.9 cases per 100,000, which is a wide range of uncertainty around the rate of
817.5 per 100,000. Upon close inspection, we know that this range of uncertainty
is due to the small number of cases diagnosed, however, the lack of confidence
intervals around the incidence rates on the “Top Ten” list could easily lead
readers to misinterpret the presented rate.
The map also presents us with a geographic puzzle. It doesn’t suggest much
3
Perez et al.: Cancer Clusters in the News Media
Published by Scholar Commons, 2015
in the way of clear spatial patterns; the 45 tracts with one or more high cancer
rates are sprinkled throughout the state. If we look at the “Top Ten” tracts and the
specific types of cancer elevated within them, there is relatively little evidence of
contiguous tracts sharing a particular sort of cancer: the tracts designated 2 and 3
share a border, and both have elevated rates of lung cancer, but the one labeled 2
has a second type of cancer, and the tract labeled 3 has four other types for which
the other tract does not have elevated rates; similarly the tracts labeled 4, 10 and 5
are contiguous, and all three had high rates of prostate cancer, but each had high
rates for one or two other types of cancer that were not shared with its immediate
neighbor. If cancer rates were indeed patterned, we might expect more instances
of high rates for particular types of cancer being found in adjacent census tracts
(i.e., a spatial pattern).
In other words, while Delaware residents who looked at the News Journal’s
2010 map might have been alarmed to discover that cancer rates were above
average in or near the census tracts where they lived, the information conveyed in
the map is insufficient to make a convincing case that there are actual cancer
clusters in Delaware.
Comparing Maps
The confusion increases when we consider the two maps accompanying a 2014
News Journal story. In line with this story’s upbeat headline (“Conquering
Cancer”), the story featured a smaller map showing “more than 30 census tracts
[with] . . . significantly above average cancer rates” for 2002-2006 (Montgomery
and Miller 2014: A1). The 2010 story discussed above also had been based on
2002-2006 data, but there is no explanation why the earlier story identified 45
problematic tracts, vs. only 30 in this newer map (the tracts identified in the newer
map all seem to have been among those specified in the 2010 story).
Even more interesting is the 2014 story’s larger map identifying 14 “Hot
Spots”—census tracts that “recorded consistently high overall cancer rates for
periods from 2001-2009" (Montgomery and Miller 2014: A1). Comparing this
map with the 2010 map, and with the smaller 2014 map is confusing. Of the 14
“Hot Spots,” only 2 were among the “Top Ten” tracts identified in the 2010 map,
and only 5 are identified on the smaller 2014 map showing 30 tracts with high
cancer rates during 2002-2006. Presumably, cancer clusters should display some
stability–we might expect that a cluster would reappear on map after map (i.e., a
temporal pattern). The discovery that, year after year, a particular type of cancer
is unusually common in a particular census tract would suggest that the high
cancer incidence rate is not merely a product of random variation and could
possibly be connected to environmental sources.
The News Journal defines a “Hot Spot” as a census tract with an elevated rate
4
Numeracy, Vol. 8 [2015], Iss. 1, Art. 7
http://scholarcommons.usf.edu/numeracy/vol8/iss1/art7
DOI: http://dx.doi.org/10.5038/1936-4660.8.1.7
of all-site cancer across any two or more of the time periods analyzed in a recent
state report examining cancer incidence for the years 2001-05, 02-06, 03-07, 04-
08, and 05-09 (DHHS 2013a). Thus, a “Hot Spot” is a census tract that has an
elevated cancer incidence rate, regardless of type of cancer, in two or more of the
time periods analyzed by the state in one of its 2013 reports. However, this
should be interpreted with some degree of caution because it does not mean that
any one specific type of cancer is elevated consistently for any specific time
period in a tract. It is true that some tracts have elevated cancer diagnoses for
some types of cancer over periods of time, but a more precise examination of the
types of cancer, and for how long they are elevated (and if consecutively across
time periods) is warranted. The fact that the Delaware data do not display clear
temporal or spatial patterns–at least as they are portrayed in the News Journal’s
maps–suggests that these findings should be interpreted with caution.
The News Journal’s 2014 coverage seemed more reserved than its earlier
articles. It conceded that the reports by state agencies (Montgomery and Miller
2014: A8):
have yet to reveal a verified, environmentally caused “cluster” of the disease. The tract
studies grew out of public concern about a possible pollution-related cancer hot spot in
the Millsboro area in 2006. Health officials eventually did find higher than average rates
for bladder and male lung cancers in one Millsboro area census tract, between 2001 and
2006, but follow-up studies never confirmed [a] pollution link and rates diminished in
subsequent reports.
In sum, the News Journal maps sought to summarize a great deal of official
data, but it did so in an uncritical fashion. Neither the number of census tracts
having higher cancer rates, nor their geographic proximity, nor the temporal
patterns in the findings gave a clear sense that the paper had indeed identified
health risks. These stories may have commanded readers’ attention, but it’s not
clear that they served the public interest. But, of course, the News Journal did not
originate the data it mapped; those data came from official agencies. Next, in an
attempt to provide some clarity to what the statistics and the maps actually
demonstrate, we will consider another puzzling issue: what is a cancer cluster, and
how did the state of Delaware identify them?
What is a Cancer Cluster?
To its credit, the News Journal did a reasonable job of providing its readers with
the general limitations of using “cancer cluster” maps as evidence of industrial or
corporate harm on public health. Further, without the luxury of being able to go
into detail about the recent state reports re-analyzing cancer incidence data in its
news stories, confusion among News Journal readers was likely. To boot, there
are numerous issues in identifying clusters and drawing conclusions about the
5
Perez et al.: Cancer Clusters in the News Media
Published by Scholar Commons, 2015
environmental impact on public health from these data, such as too few cases to
sustain statistical validity (i.e., the “small numbers problem”), resident mobility
and migration, resident health behavior, utilization of cancer screening, timing of
disease, cancer registries, and statistical chance (Goodman et al. 2014; Margai
2010). Even though the News Journal has presented these issues to readers, the
chief problem is that the very use of the term “cancer cluster” by the News
Journal was erroneous.
A cancer cluster falls under the rubric of a non-communicable disease cluster,
which the National Cancer Institute (NCI), a division of the National Institutes of
Health (NIH), defines as “the occurrence of a greater than expected number of
cases of a particular disease within a group of people, a geographic area, or a
period of time” (National Cancer Institute 2014a). The News Journal used data
from reports generated by Delaware’s Department of Health and Human Services
(DHHS), Division of Public Health (DPH). These reports are unusual, as they did
not identify cancer clusters using a standard method known as relative risk, where
researchers calculate the expected number of cancer diagnoses and then compare
that number to what was observed (i.e., the number of diagnoses that actually
happened); what they did is different (and will be explained later). To further
complicate the use of the term “cancer cluster” in the News Journal articles, the
NIH reserves the use of cluster to a high incidence of one type of rare cancer (not
many types lumped together) that affects an age group not usually impacted by
that type of cancer or disease (National Cancer Institute 2014b). Delaware
officials, on the other hand, began by identifying census tracts in which the
combined rate of all cancers exceeded the statewide rate, and labeled those tracts
as elevated.
In general, if the public is concerned about cancer in their local area, they can
catalyze an investigation by contacting the state’s health department, which then
seeks to identify the population at risk (e.g., children of a certain age), estimates
the expected number of cancer cases in that group in a given time period, and then
compares that value to the number of cancer cases that actually occurred in that
group. Again, the maps used by the News Journal do not reflect this approach;
instead, they start with the geographic boundaries of census tracts (not a specific
group at risk) and compare cancer incidence rates of census tracts to the overall
state rate to determine if a tract is a “cluster” (i.e., elevated relative to the state).
Using census tracts and incidence rates to identify clusters invites uncertainty in
their interpretation because: 1) the incidence rates are made up of numerous
cancers lumped together, including the most common types of cancer that are so
much more common than other types that they skew the overall rate; and 2) tracts
will include people who have the disease but are unrelated to the real at-risk
group (i.e., the people we suspect have cancer due to exposure to environmental
burdens) (National Cancer Institute 2014b).
6
Numeracy, Vol. 8 [2015], Iss. 1, Art. 7
http://scholarcommons.usf.edu/numeracy/vol8/iss1/art7
DOI: http://dx.doi.org/10.5038/1936-4660.8.1.7
Thus, what the state reports contain and what the News Journal stories
reported are age-adjusted cancer incidence rates for 5-year time periods for
individual census tracts, which are then compared to the state as a whole. There
were two ways that cancer incidence in census tracts were presented using this
method. First, the “all-site” incidence rate, which lumped together diagnoses for
all 23 types of cancer, including the “big four” of lung, prostate, colorectal, and
breast cancer. These four cancer types make up the majority of diagnoses. There
are lots of types of each of these cancers (i.e., there are dozens of types of breast
cancer), lots of ways people get them, and they have the most opportunity for
chance variation. In short, they’re so common and varied that including them all
in cluster investigations invalidates the method of finding true “clusters.”
Second, the state investigators went on to examine incidence rates for 23
specific types of cancer in those tracts that had elevated all-site cancer incidence.
Adding to the complexity of including so many types of cancer is the need to
calculate an incidence rate per 100,000 people for all of Delaware’s census tracts,
even though many of the tracts were so small that they averaged only 25 to 35
new cancer cases of all types during the time period being analyzed (DHSS
2010b). Thus, we have a complicated mix of using incidence rates that involve
combining five years worth of data at once, extrapolated from relatively small
numbers of new cases, for many types of cancer, which are then compared to the
incidence rate for the state as a whole. Got it? So, how did these statistics
become cancer clusters?
Deconstruction of State Reports
The method of identifying elevated tracts in state reports and the availability of
mapping software to present census tracts in color-coded maps, combined with
using the term “cluster” in news stories, resulted in the “cancer cluster issue” the
readers of the News Journal imagine. Arguably the most influential report used
by the News Journal in recent years was published by the DPH in May 2010.
Using year 2000 Census population estimates and census tracts, the report
mapped all-site cancer incidence rates for each census tract, for the years 2002-
2006 as a whole (i.e., summed together) (DHSS 2010a). We will use that report
to describe how clusters came to be identified in the News Journal. (It is worth
noting again that there have been several DPH reports that used different
methodologies for calculating and mapping cancer incidence, resulting in various
numbers of elevated tracts across reports – this will be discussed in more detail
later.)
The 2010 DPH report contained startling information: Delaware had 45
census tracts that had statistically significantly higher cancer incidence rates for at
least one type of cancer than the overall rate for the entire state. These areas
7
Perez et al.: Cancer Clusters in the News Media
Published by Scholar Commons, 2015
quickly became “clusters” in the News Journal’s September, 2010 front-page
story (Montgomery 2010). Indeed, the article also printed links to interactive
online maps that the News Journal had created that allowed people to find their
census tract and get more detailed information on it.
1
However, what the Delaware DPH identified in its reports were actually
census tracts with higher age-adjusted all-site cancer incidence rates, relative to
the state, over the 5-year time period from 2002-06. These reports do not identify
clusters in the same way that the NIH does, but instead give spatial and temporal
data on cancer incidence (Margai 2010). As we noted above, this is not the same
as a cancer cluster. So, how did the state identify the census tracts with higher
levels of cancer incidence rates, and how did these tracts become “clusters” in the
News Journal? Furthermore, what are the issues in creating and understanding
this type of data? Though the following paragraphs are expository, spelling out
precisely how the state generated these data is of vital importance to
understanding the issue.
This is where things get a little complicated, so we’ll lay out the basics first,
drawing on the 2010 report from the DPH. First, the address of a person
diagnosed with cancer is assigned to a census tract and this address-to-tract data is
validated. For the years 2002-2006, cumulatively (i.e., all diagnoses within that
time period added together), all malignant cancer cases that were diagnosed were
included (benign tumors and basal and squamous cell cancers were excluded).
Because of low census tract validity (i.e., cases where the address of a person
could not be assigned to a census tract with confidence), 599 malignant diagnoses
were excluded from the analysis at the census tract level (471 of these exclusions
came from Sussex—the state’s most rural county), but were retained for use in
calculating the state’s overall cancer incidence rate. In all, 22,161 diagnoses from
2002-2006 were included in the analysis (DHSS 2010a).
Before age-adjusted cancer incidence rates could be calculated, it was
necessary to calculate population estimates for census tracts and the state, and to
divide them into 5-year age groups so that age could be taken into account (cancer
diagnoses occur more frequently among older groups). Let’s start with the census
tracts, using year 2000 Census guidelines. Of the 197 census tracts that the state
of DE had in the year 2000 Census, 1 was not populated, leaving 196 census
tracts. Using these data, the DPH calculated 5-year age group proportions by
gender for each census tract, providing estimates of annual population for every
tract in Delaware. Census tract population estimates for the time period 2002-
2006 ranged from 3,132 to 65,136 people across tracts (DHSS 2010a).
1
These maps have been replaced by a single, all-encompassing interactive map at
http://www.delawareonline.com/story/news/health/2014/01/24/map-incidence-cancer-
delaware/4834779/ (last accessed Dec 22, 2014)
8
Numeracy, Vol. 8 [2015], Iss. 1, Art. 7
http://scholarcommons.usf.edu/numeracy/vol8/iss1/art7
DOI: http://dx.doi.org/10.5038/1936-4660.8.1.7
Given these estimates for the census tracts’ populations, how did the report
generate age-adjusted cancer incidence rates for them? First, the DPH ran a
cross-tabulation of age group by census tracts, providing a way to show all the
people, according to age group, with a cancer diagnosis in every tract. This gave
the number of people in each tract, by age grouping, with a cancer diagnosis.
With this information, both crude and age-adjusted incidence rates could be
calculated for each 5-year age group and entire census tracts, starting with
specific age groups. First, the crude incidence rate was derived by taking the
number of cancer diagnoses in a particular age group (e.g., 40-44 year olds) in a
census tract, then dividing that number by the population estimate for that age
group in the tract; next, multiplying that value by 100,000 to get a cancer
incidence rate of that particular age group in that census tract per 100,000 people.
Now, what about a cancer incidence rate for an entire tract and everyone in it
together (i.e., not grouped by age)? That’s an easy one: just take the number of
cancer diagnoses for 2002-2006 in the tract, divide it by the 2002-2006 population
estimate for that tract, and then multiply it by 100,000. Want to get, finally, your
age-adjusted cancer incidence rates for a census tract? Take the product of the
crude incidence rate for each 5-year age group and its corresponding year 2000
Census population weight, then sum all age group incidence rates, and you have
an age-adjusted cancer incidence rate per 100,000 for a census tract.
Ultimately, in order for elevated tracts to appear, census tracts were
compared to the state as a whole to see how they differed—this is essentially a
two-step process. First, we have to calculate the age-adjusted cancer incidence
rate for the state of Delaware. Using the 22,161 cases from 2002-2006, the DPH
followed similar calculation procedures used to get age-adjusted incidence rates
for census tracts in order to calculate the incidence rate for the state as a whole,
which turned out to be 507 per 100,000. This rate is an estimate based on the
known number of cancer diagnoses and the population estimates used by the
DPH. Therefore, the rate is not necessarily exact, but we can situate it within a
confidence interval, or an interval around this estimate within which we are pretty
sure the true cancer incidence rate for the state would fall (i.e., 95% sure).
Remember that a confidence interval is calculated using a relatively simple
formula involving the age-adjusted rate of a tract, the square root of the number of
cancer cases in that tract, and the value 1.96, which reflects a standard deviation
value. The 95% confidence interval around the state’s all-site cancer incidence
rate for the years 2002-2006 ranged from a lower limit of 500.4 to a high of 513.6
per 100,000 (DHSS 2010a). Notice that this confidence interval is relatively
narrow around the estimate, indicating an incidence rate that has a good degree of
precision because it involves all of the cases (over 20,000 diagnoses for the 5-year
time period) and the entire state’s population.
In order for census tracts to be identified as elevated, the state compared
9
Perez et al.: Cancer Clusters in the News Media
Published by Scholar Commons, 2015
incidence rate confidence intervals of individual census tracts to the state’s.
Specifically, a tract’s incidence rate for all-site cancer must fall above the state’s
and the confidence interval for the tract’s incidence rate must not overlap with the
state’s, either. For example, census tract 122, an area located immediately west of
the City of Wilmington, had an all-site cancer incidence rate of 670.4 per
100,000, with a 95% confidence interval ranging from 556.9 to 783.9 (DHSS
2010a). Notice how much wider the confidence interval for the census tract is
than the state’s, highlighting the spread in values necessary in order to be 95%
confident that the true number of cancer diagnoses in this tract falls within that
range. (This is because the 2002-2006 cancer incidence for that census tract was
only an average of 26.8 cases per year for the time period from 2002-2006 (a
“small number problem”) (DHSS 2010b)) Remember that the confidence interval
for the state of Delaware had an upper limit of 513.6 and a lower limit of 500.4
per 100,000. Now, compare the two confidence intervals: the lower limit of the
confidence interval for census tract 122 (556.9 per 100,000) does not overlap with
the upper limit of the state’s confidence interval (513.6 per 100,000), making it
“elevated.” This is how a statistically significant difference in cancer incidence
rate between a census tract and the state came to be identified in official reports,
and, subsequently, how “cancer clusters” appeared in the News Journal.
In other words, in order to label census tracts as statistically significantly
higher, lower, or non-significantly different than the state in terms of cancer
incidence over time, the DPH used confidence intervals for census tract incidence
rates and compared them to the state’s confidence interval. If a census tract’s
cancer incidence rate was higher than the state’s, and the respective confidence
intervals did not overlap, the DPH deemed that census tract to have an elevated
cancer incidence rate that was not due to chance. Thus, “cancer clusters” were
census tracts with confidence intervals that do not overlap with the state’s
incidence rate for the years 2002-2006 (cumulatively). This approach is not
unprecedented, but certainly not the typical definition of a cancer cluster,
although it is a very effective way for the News Journal to report cancer data
according to census tract and incidence.
How the News Journal Made Cancer Clusters “Real”
The census tract analyses in the DPH reports are the result of legislation that
required such analyses be done, and news reporting on these investigations was
inevitable in a state where there is a history of concern about cancer-environment
links. It is easy to see how a census tract map, color-coded on the front page of
the state’s leading newspaper, could be understood to be a “cluster” to readers.
Although true cancer clusters are rare, the term is useful for citizens who feel that
industry pollution has affected their health in a specific area, and for news media
10
Numeracy, Vol. 8 [2015], Iss. 1, Art. 7
http://scholarcommons.usf.edu/numeracy/vol8/iss1/art7
DOI: http://dx.doi.org/10.5038/1936-4660.8.1.7
outlets making claims about connections among toxic substances, pollution, and
human health. In communities that have an anecdotal history of high cancer rates,
as well as communities with long-standing contentious relationships with local
industry, the idea of the cancer cluster is a powerful rhetorical tool for making
claims about environmental justice. With the availability of mapping software
and this type of public health data, organized by incidence rate and already geo-
coded by census tract, the News Journal was able to provide to the public visually
attractive and convincing data on the relationship between cancer and industrial
pollution throughout DE, appealing to a variety of communities.
The News Journal has been instrumental in championing the cause of citizens
who perceive that their area’s high cancer rates are the result of industrial
pollution since they began publishing a series of stories on the Indian River power
plant emissions in the Millsboro area of Delaware, and the disproportionately high
number of resident-reported cancers (Nathans 2007). Concern about elevated
cancer incidence begins with residents, when one or more concerned citizens ask
the state to investigate cancer in their area. In 2007, the News Journal portrayed a
familiar trajectory for these investigations (Nathans 2007: A1):
For years, residents in the small towns around the Indian River power plant have noticed
friends and relatives falling sick in greater numbers than they thought normal. Years
after citizen activists first asked the state for data to establish a pattern, the Division of
Public Health has finally confirmed what they suspected: There’s a cluster of cancer
cases near the coal burning plant – the state’s worst polluter.
The 2007 story included only a small map that displayed the Millsboro area and
its close surroundings, but soon the News Journal was producing statewide maps
of cancer incidence by census tract in a series of articles, many of which were
based on the release of new and updated state reports. In 2008, for example, the
front-page map of the entire state of Delaware showcased 8 areas designated as
clusters, and also included a list of high incidence areas and the locations of the
top 20 largest polluters in the state (Barrish 2008: A1). The map was made
possible by the state’s DPH 2008 report that reported cancer incidence rates
according to the 27 Census County Division (CCD) areas in Delaware, using
methods similar to those described earlier for census tracts, for determining
whether a CCD had a statistically significantly higher incidence rate than the
state’s (DHHS 2008).
These maps are integral to the claimsmaking activities of the News Journal
and reify the existence of cancer clusters in Delaware, as does the very use of the
term. The state’s DPH reports, because of the data they included and the way that
incidence rates were geo-coded, were easily transferable onto visual displays that
made for attractive, easily understood wellsprings of information for the public.
Connections to environmental burdens were easy to proclaim, supported by such
visual evidence, as one journalist noted in 2008 (Montgomery 2008: A1):
11
Perez et al.: Cancer Clusters in the News Media
Published by Scholar Commons, 2015
While state officials cautioned about speculation about potential causes, each of the
clusters identified in the state report to be released today was found in an area with clear
environmental burdens, from industrial and farm pollution to jammed and smog-laced
interstate highways to heavy reliance on shallow, private groundwater supplies that are
vulnerable to contamination.
To be sure, the News Journal did not misrepresent the data that was provided by
the state–it reported them with its own maps and discussed the many limitations
of these data. However, it did use the opportunity to call county divisions and
census tracts with statistically significantly higher incidence rates cancer clusters,
and use this information to make claims about environmental burdens and their
connection to cancer incidence in Delaware. Only in the early DPH reports did
the state actually use the term “cancer cluster,” but that term quickly fell out of
use in later reports that analyzed census tracts. The idea of a cancer cluster is an
effective rhetorical tool for making claims stick to environmental justice issues,
suggesting a pattern of excessive cancer in a specific location that needs
explanation. The News Journal helped to reify cancer clusters in the state of
Delaware through the use of state reports, but those early reports were about to
get a facelift and fundamentally change the number of “cancer clusters” in
Delaware.
From 8 to 59 to 45 to 11: The Problem of Cancer
Incidence Statistics and the Frontier of “Hot Spots”
Since 2008, there have been no fewer than 5 comprehensive reports on cancer in
Delaware that included incidence-rate-by-census-tract analyses. Recent analyses
of cancer incidence by census tract using updated population data, as well as
secondary analyses of older reports also using updated population data, have
identified significantly fewer “clusters” or elevated incidence areas in the state.
In this section, we briefly return to the ways that the state of Delaware has
analyzed cancer data and reported on it, how that has changed over time, and how
the News Journal’s reporting on cancer incidence, while remaining as true as
possible to these reports, ultimately created a befuddling picture of the issue.
In 2008, the state of Delaware’s Division of Public Health used the 27
Census County Divisions (CCD) in Delaware, along with 5-year all-site cancer
incidence rates of each CCD to determine if CCDs had elevated cancer incidence
rates, using methods similar to those described earlier for census tracts (with
confidence intervals for CCDs and comparing them to the state as a whole)
(DHSS 2008). The result of that investigation revealed that 8 of the CCDs were
elevated. The DPH has always used 5-year interval data to include as many cases
as possible to gain some statistical power, but also to “average out” any year-to-
year fluctuation of cancer diagnoses. A year later, the DPH started analyzing
12
Numeracy, Vol. 8 [2015], Iss. 1, Art. 7
http://scholarcommons.usf.edu/numeracy/vol8/iss1/art7
DOI: http://dx.doi.org/10.5038/1936-4660.8.1.7
cancer incidence rates in Delaware using census tracts, using both year 2000
Census and Delaware Population Consortium data, including the 196 tracts that
had people living in them, and subsequently revealing 29 census tracts with
elevated cancer incidence rates for the years 2001-2005 (DHHS 2009). Then, in
2010, the Delaware DPH released yet another analysis for the years 2002-2006
(described in detail earlier), using year 2000 Census data and 196 census tracts,
revealing now 45 tracts with statistically significantly elevated cancer incidence
rates (DHSS 2010a). Another report followed, using year 2000 Census data and
196 census tracts for the years 2003-2007 to reveal a startling 59 tracts with
elevated rates (DHSS 2012).
However, with the introduction of newer, updated population and census tract
data from the Census, things changed dramatically. Armed with year 2010
Census data, which partitions DE into 214 census tracts, the new population
estimates used in calculations of incidence rates, and the resulting confidence
intervals around them, revealed only 11 elevated tracts in the years 2004-2008
and 9 for the years 2005-2009 (DHSS 2013b). Still another, very recent report
analyzing the years 2006-2010 revealed 11 census tracts with elevated cancer
incidence rates, relative to the state (DHSS 2014). Further, the availability of
2010 Census data allowed for the re-calculation of population estimates and the
re-calculation of cancer incidence rates for the years 2001-2005, 2002-2006, and
2003-2007, revealing 15, 10, and 10 elevated census tracts, respectively (DHSS
2013a). In sum, the number of areas officially identified as having high cancer
rates has varied widely, from 8 to 59.
The News Journal has kept up with drastically changing official statistics on
cancer incidence throughout Delaware in reporting on this issue and is now
focusing on those tracts with some consistency in elevated incidence rates,
designating 14 of them as “hot spots” that serve as areas for increased efforts for
screening and prevention. Indeed, one can view these consistently elevated areas
using their online, interactive map.
2
These hot spots may be useful in identifying
areas that at least have some measurable level of elevated cancer incidence over
some time period during the years 2001-2009, but unless these tracts have the
same type of cancer diagnoses that are elevated over time, and are not merely a
tract with any cancer diagnosis elevated over time, temporal patterns in elevated
tracts provide little value.
Anyone following this series of stories since 2007, however, is confronted
with a confusing array of stories, maps, and statistics, and the term “cancer
cluster,” used often in early News Journal stories, has now faded from the
headlines.
2
http://www.delawareonline.com/story/news/health/2014/01/24/map-incidence-cancer-
delaware/4834779/.
13
Perez et al.: Cancer Clusters in the News Media
Published by Scholar Commons, 2015
Conclusion
Cancer is a hot button issue for Delaware residents, many of whom live in areas
with long histories of controversy concerning industry, pollution, and disease. In
the News Journal stories, not only can residents read about and connect with
others’ experiences with cancer, but they are also exposed to public health data
that they may view as objective, scientific, and authoritative. The ease with
which the News Journal could obtain, interpret, and disseminate the state’s cancer
incidence data is something we need to recognize because it allowed—in
relatively short order—the troubling social issue of “cancer clusters” to take on a
life of its own through the widespread dissemination of questionable maps and
statistics.
Discussions of popular epidemiology often take the form of a morality tale, in
which activists expose corporate wrongdoing. Thus, Nathans (2007) cites the
Millsboro area cancer cluster investigation as the beginning of the DPH’s efforts
to look into the issue more broadly for the entire state. A resident of the area—a
local doctor—is credited, along with a handful of others, with catalyzing that
investigation. As noted earlier, the impetus for cancer cluster investigations is
most often one or more community members who believe that cancer in their area
is unusually high. To be sure, citizen activists and citizen science alliances can
have positive results on community and individual health by shedding light on an
issue in a community, providing direction to scientists for asking new questions,
and integrating the community’s experiences into the scientific paradigm that
guides public health investigations (Brown 1992). In order to do this, and in order
to raise awareness, the public must have access to good public health data and the
media outlets that present it must do so reliably. The News Journal did not
misrepresent the data it used from Delaware state reports, but the “cancer cluster”
issue is one that is already so complex, so statistically tenuous, and so emotionally
charged, that the series of stories since the 2006 Millsboro investigation have
made for a confusing and misleading saga.
The Delaware case illustrates the problems with statistical “facts.” The News
Journal did not invent numbers; it relayed data first presented in a series of
official DPH reports. And those DPH reports were efforts to collate and make
sense of reports of medical diagnoses. The officials who wrote the reports
explained their methods, although they can be criticized for applying the term
cancer cluster (at least early on in their reports) with its alarming implications to
phenomena that do not fit the definition of the term. In turn, the News Journal’s
maps of those “clusters” conveyed a troubling impression to its readers.
What is striking is the failure to ask obvious questions: why were there not
clearer geographic patterns; and why did the findings vary so much from report to
report, so that the number and locations of high-cancer areas shifted from one
14
Numeracy, Vol. 8 [2015], Iss. 1, Art. 7
http://scholarcommons.usf.edu/numeracy/vol8/iss1/art7
DOI: http://dx.doi.org/10.5038/1936-4660.8.1.7
report to the next? In a world where numbers are equated with facts, and where
maps are increasingly used to display numeric data, there is a risk that colorful
images will displace critical thinking (Monmonier 1996).
In sum, what the DPH reports and what the News Journal provides its readers
are temporal and spatial arrangements of average cancer incidence over time, by
census tract, but these are not cancer clusters and they should be handled
cautiously as evidence of the effect of local industry pollution. In the most recent
News Journal stories, the tracts with elevated rates are being referred to as “hot
spots” of disease, areas to focus cancer prevention efforts. Stopping the use of the
term cancer cluster is a step in the right direction, as this undoubtedly will have an
effect on how the issue is interpreted by the public. However, the identification of
census tracts with elevated cancer incidence - the “hot spots” – should still be
interpreted with a healthy degree of caution when trying to determine what the
data really mean and how they serve as causal evidence of the effect of local
industry and pollution. This is not to say that the census tract analysis doesn't
have any value, as areas with high levels of cancer incidence (at least relative to
the state using confidence intervals) may reveal underlying social conditions that
also influence health.
We do not mean to suggest that either the DPH or the News Journal were
guilty of especially bad practices. They were trying to make data available for
public consumption. No doubt other sorts of agencies in other states find
themselves trying to produce data to address a wide range of public concerns, and
other news media find themselves reporting on those data. But Delaware’s small
size makes it easier to understand how these data were produced, disseminated,
and interpreted. This case reveals many issues in thinking critically about a
relatively small data set, which suggests that there are special challenges to
promoting numeracy about big data.
Acknowledgments
The authors would like to thank the reviewers for their helpful suggestions and
recommendations to improve the paper.
References
Barrish, Cris. 2008. “Eight Cancer Clusters Discovered in Delaware.”
Wilmington News Journal, April 24: A1, A6.
———, and Jeff Montgomery. 2008. “No Money for Cancer Studies, Del. Says.”
Wilmington News Journal, April 27: A1, A8‒9.
Brown, Phil. 1992. Popular epidemiology and toxic waste contamination: Lay and
professional ways of knowing. Journal of Health and Social Behavior 33(3):
267‒281. http://dx.doi.org/10.2307/2137356
15
Perez et al.: Cancer Clusters in the News Media
Published by Scholar Commons, 2015
Delaware Health and Social Services, Division of Public Health. 2008. Average
Annual Age-Adjusted Cancer Incidence Rates, 2000-2004, at the Delaware
Sub-County Level. Delaware Division of Public Health. (downloaded
October 6, 2010)
———. 2009. Delaware Cancer Incidence by Census Tract: Methods. Delaware
Division of Public Health. (downloaded February 2, 2014)
———. 2010a. Cancer Incidence and Mortality in Delaware: May 2010.
Delaware Division of Public Health. (downloaded August 19, 2011)
———. 2010b. Analysis of Census Tracts with 2002-06 Elevated All-Site Cancer
Rates: August 25, 2010. Delaware Division of Public Health. (downloaded
August 2, 2011)
———. 2012. Delaware’s Cancer Incidence and Mortality Report: 2003-2007.
Delaware Division of Public Health. (downloaded February 2, 2014)
———. 2013a. Secondary Analysis of Census Tracts with Consistently-Elevated
All-Site Cancer Rates in Delaware, 2001-2005, 2002-2006, 2003-2007, 2004-
2008 and 2005-2009. Delaware Division of Public Health. (downloaded
February 2, 2014)
———. 2013b. Cancer Incidence and Mortality in Delaware: April 2013.
Delaware Division of Public Health. (downloaded February 2, 2014)
———. 2014. Cancer Incidence and Mortality in Delaware, 2006-2010. Delaware
Division of Public Health. (downloaded May 2, 2014)
DHSS. See Delaware Health and Social Services, Division of Public Health.
Gelman, Andrew, and Deborah Nolan. 2002. Teaching Statistics: A Bag of
Tricks. New York: Oxford University Press.
Goodman, Michael, Judy S. LaKind, Jerald A. Fagliano, Timothy L. Lash, Joseph
L. Wiemels, Deborah M. Winn, Chirag Patel, Juliet Van Eenwyk, Betsy A.
Kohler, Enrique F. Schisterman, Paul Albert, and Donald R. Mattison. 2014.
Cancer cluster investigations: Review of the past and proposals for the future.
International Journal of Environmental Research and Public Health
11:1479‒1499. http://dx.doi.org/10.3390/ijerph110201479
Margai, Florence. 2010. Environmental Health Hazards and Social Justice:
Geographical Perspectives on Race and Class Disparities. Washington,
D.C.: Earthscan.
Miller, J.L. 2008. “Del. Cancer Maps Raise Questions about Incidence Rates.”
Wilmington News Journal, December 19: A1‒A2.
Monmonier, Mark. 1996. How to Lie with Maps, 2
nd
ed. Chicago: University of
Chicago Press.
Montgomery, Jeff. 2008. “High Incidence Seen in Most Polluted Areas.”
Wilmington News Journal, April 24: A1, A6‒A7.
———. 2010. “Mapping Out Clues to Cancer Mystery.” Wilmington News
Journal, September 12: A1, A6‒A7.
16
Numeracy, Vol. 8 [2015], Iss. 1, Art. 7
http://scholarcommons.usf.edu/numeracy/vol8/iss1/art7
DOI: http://dx.doi.org/10.5038/1936-4660.8.1.7
———, and Beth Miller. 2014. “Conquering Cancer.” Wilmington News Journal,
January 26: A1, A8‒9.
Nathans, Aaron. 2007. “Delaware Confirms Cancer Cluster.” Wilmington News
Journal, August 5: A1, A11‒A12.
National Cancer Institute. 2014a. National Cancer Institute Fact Sheet: Cancer
Clusters. “Defining Disease Clusters.”
(www.cancer.gov/cancertopics/factsheet/Risk/clusters)
———. National Cancer Institute Fact Sheet: Cancer Clusters. “Facts About
Cancer Clusters.” (www.cancer.gov/cancertopics/factsheet/Risk/clusters)
Shortridge, Dan. 2008. “State Zeroes in on Millsboro’s Cluster.” Wilmington
News Journal, May 18: A1, A6.
U.S. Census Bureau. 2014. American Community Survey, American Fact Finder.
Report generated on October 26, 2014 at
http://factfinder2.census.gov/faces/tableservices/jsf/pages/productview.xhtml
?fpt=table
17
Perez et al.: Cancer Clusters in the News Media
Published by Scholar Commons, 2015
