mammography screening and the increase in breast...

Vol. 6, 201-208, March 1997 Cancer EpidemioIo�j�, Biomarkers & Prevention 201

Mammography Screening and the Increase in Breast Cancer Incidence

in Hawaii1

Gertraud Maskarinec,2 Lynne Wilkens, and Lixin Meng

Cancer Research Center of Hawaii, Honolulu, Hawaii 96813

Abstract

This ecological study investigated the association betweenmammography utilization and breast cancer incidence inHawaii with the hypothesis that geographic areas withhigh mammography use have higher breast cancer

incidence than geographic areas with low mammographyuse. Insurance claims for mammograms received during1992 and 1993 were combined with breast cancerincidence data from the Hawaii Tumor Registry and datafrom the 1990 Census ZIP File. The claims data wereobtained from four private and three public health plans

and covered approximately 85% of women 40 years ofage and older. Age-specific breast cancer incidence ratesfor the 79 ZIP code areas were regressed onmammography rates and selected aggregate demographicvariables using multiple linear regression. An estimated

42% of women 40 years of age and older had received atleast 1 mammogram during 1992 and 1993, with thehighest rate (45%) in women ages 50-64 years old.Overall, 23% of the variation in age-specific breastcancer incidence could be predicted by mammographyutilization, 23% by increasing age, and 4% by highereducation. The relationship between mammography useand breast cancer incidence was strongest for women 50-64 years old and for localized disease. The magnitude ofthe association between breast cancer incidence andmammography utilization was comparable to the increase inbreast cancer rates observed in Hawaii during the mid-1980s, supporting the hypothesis that the sharp increase inbreast cancer incidence was attributable to screening andearly detection. However, the long-term 1% increase inbreast cancer incidence requires alternate explanations.

Introduction

As in many Western populations, breast cancer is the mostcommon cancer among women in Hawaii, with 672 invasive

cases plus 1 15 in situ cases diagnosed in 1993. Breast cancer

incidence rates in Hawaii (Fig. 1) have been increasing between

Received 6/I 1/96; revised 12/6/96; accepted 12/10/96.

The costs of publication of this article were defrayed in part by the payment of

page charges. This article must therefore be hereby marked advertisement inaccordance with 18 U.S.C. Section 1734 solely to indicate this fact.

I Supported by a research grant (R03 CA63305) from the National Cancer Institute,

Prevention and Control Branch. The data presented here were part of a dissertation at

the University of Hawaii in partial ftilfillment of the requirements for the degree of

Doctor in Philosophy in Biomedical Sciences (biostatistics/epidemiology).2 To whom requests for reprints should be addressed, at Cancer Research Centerof Hawaii, 1236 Lauhala Street, Honolulu, HI 96813.

1960 (34 per 100,000 women, age-adjusted to the 1970 popu-

lation of the United States) and 1993 (102 per 100,000 women),

gradually at first and then more rapidly between 1983 and 1987.A similar gradual increase of 1% per year until 1982 followedby an annual 4% increase during the 1980s was observed in thenationwide SEER3 program (1). Most of the increase in mci-

dence was due to localized tumors and occurred in women 50years of age and older. In contrast to breast cancer incidence,breast cancer mortality rates in the United States have beencomparatively constant at 26 deaths per 100,000 women of all

ages (1), with a small decline of 5.5% observed in Caucasianwomen between 1989 and 1992 (2). Mortality in Hawaii hasbeen constant at around 22 deaths per 100,000 women since1976. Breast cancer incidence and mortality in Hawaii are

considerably higher in Caucasian and Hawaiian women than inwomen of Asian descent (3). According to an unpublished

report by the Hawaii Tumor Registry, during 1990-1993 theinvasive breast cancer incidence rate was I 16 cases per 100,000

Caucasian women, 1 1 1 cases per 100,000 Hawaiian women, 91cases per 100,000 Japanese women, 82 cases per 100,000

Chinese women, and 68 cases per 100,000 Filipino women(age-adjusted to the 1970 population of the United States).

Well-established breast cancer risk factors, such as late age atfirst live birth, nulliparity, first-degree family history, and higher

socioeconomic status, can explain an estimated 41% of the pop-ulation-attributable risk according to a recent study (4). Including

age at menarche, age at menopause, and exposure to radiationincreases the proportion of explainable risk to 45-55%. Becausevery few ofthese factors for breast cancer can be modified through

interventions, early diagnosis through mammography and regularbreast exams remain the major strategies to decrease morbidity and

mortality from breast cancer (5). Currently, the National CancerInstitute (6, 7) recommends annual mammography screening forwomen 50 years of age and older, whereas the American Cancer

Society (8) advocates regular mammograms every 2 years startingat age 40 and annually after age 50.

The number of mammography facilities in Hawaii rose

rapidly during the 1980s, especially between 1983 and 1987(Fig. 1 ). Of the 44 facilities in the state, 21 facilities are locatedinside the city of Honolulu, 13 are located on Oahu outsideHonolulu, 3 are located on the island of Hawaii, 2 are located

on Kauai, 4 are located on Maui, and 1 is located on Molokai.Because of the large increase in mammography screening dur-

ing the l980s, it was suggested that changes in screeningpractices were responsible for the increased number of breast

cancer cases detected. Several statistical studies (9-12) thatmodeled breast cancer rates based on mammography ratesassessed through telephone surveys or medical records reviewcould explain the increase only partly by mammography. 0th-

3 The abbreviations used are: SEER, surveillance, epidemiology. and end results;CPT, current procedural terminology; HMO. health maintenance organization;BRFSS, Behavioral Risk Factor Survey System.

on May 28, 2018. © 1997 American Association for Cancer Research. cebp.aacrjournals.org Downloaded from

http://cebp.aacrjournals.org/

120

1960 1965 1970 1975 1980 1985 1990 1995

Yur

202 Mammography Use and Breast Cancer Incidence

100

80

I60

4�I

20

Fig. 1. The number of mammography facilities and breast cancer incidence

(adjusted to the 1970 population of the United States), Hawaii, 1960-1965.

ers (13-15) concluded that the 4% annual increase between1982 and 1987 was entirely attributable to early detection

through screening, whereas the annual 1 % long-term increasein breast cancer incidence was not predictable by screening butmay have been due to changing risk factor patterns among

consecutive birth cohorts. Trends in breast cancer incidencesince 1982 were also found to be compatible with the supply ofmammography units in the United States (16). Shapiro (17) hasemphasized the importance of developing more appropriatepopulation-based data systems with information on how tumorswere detected to serve as tools to evaluate the effects of screen-

ing on breast cancer incidence in the future.

Several features make Hawaii a unique location to explorethe relationship between mammography utilization and breastcancer incidence. Because of the state’s island nature, mosthealth care is received within the state. Insurance coverage ishigh: less than 5% of the population are estimated to be unin-sured (18). Insurance coverage for mammography has beenlegislatively mandated since 1991. Therefore, over 90% ofwomen in Hawaii ages 40 years and older were able to receiveregular mammograms at no or at low cost. Local insurance

carriers are willing to provide insurance claims for researchpurposes. Claims data have the advantage of being economicalbecause they are already collected and put into an electronicformat. They also cover large segments of the population, allow

follow-up, use standardized codes [International Classificationof Diseases 9 (19) for diagnosis and CPT (20) for procedures],do not need informed consent to be studied, and do not rely onsubject recall (21). Most published research related to insuranceclaims comes from Medicare (22-24), which covers most of theelderly population in the United States, or from Canada (25-27), where the national health care system covers 99% of the

residents. In a Medline search covering the years 1983-1995,no publications using insurance claims data for the study ofmammography utilization or breast cancer were found.

This study examined geographic variations in mammography

rates and breast cancer incidence rates to estimate what part of the

recent increase in breast cancer incidence may be attributable toincreased mammography utilization. The specific objectives wereto determine mammography utilization rates for 1992 and 1993 by

age group and geographic area in Hawaii to associate mammog-

raphy utilization in 1992 and 1993 with breast cancer incidence in1992 and 1993 on the smallest geographic levels for which data are

available and to determine the association between mammographyutilization and tumor stage at diagnosis for breast cancer casesdiagnosed in 1992 and 1993.

Materials and Methods

Study Design and Population. An ecological study designwith ZIP code area as the group unit of observation was used

for this project because ZIP code was the smallest geographicunit common to all databases. The study population consisted

of all women resident in the state of Hawaii who were 40 yearsof age or older in 1993, estimated at 234,400 by the United

States Bureau of the Census (28). Hawaii’s ethnic distributionin the 1990 census was 33% Caucasian, 22% Japanese, 13%Hawaiian/part Hawaiian, 15% Filipino, and 6% Chinese, with

some smaller groups constituting the rest of the population (28).

Data Sources. Mammography use was assessed through in-

surance claims data covering an estimated 85% of women

40-64 years old and close to 100% of women 65 years of ageand older. Claims with all procedure codes identifying mam-mograms were extracted from the databases of the participating

health plans. Procedure codes 76092, 40300, Z8003, Z5030,

Z5026, and Z5027 indicated screening mammograms, whereascodes 76091, 76090, 8737, and 40100 specified diagnostic mam-

mograms. Over 54% of the data were provided by the Blue

Cross/Blue Shield insurer; 24% were provided by Kaiser Perma-

nente, a HMO; 16% were provided by Medicare; 3% were pro-vided by Medicaid; 2% were provided by other private health

plans; and less than 1% was provided by the Civilian Health andMedical Program of the Uniformed Services, the health plan for

military dependents. The two largest health plans also covered

30% of the Medicare population through prepaid enrollment con-tracts with the Health Care Financing Administration. Missingwere mammograms paid out-of-pocket, paid by mainland insurers,

or provided to active military personnel.

Data elements of insurance claims used for this projectincluded patient information (identification number, sex, birthdate, and zip code), date of service, and up to five diagnostic(International Classification of Diseases; Ref. 19) and proce-dure (CPT; Ref. 20) codes. All duplicate claims were deleted,

i.e. claims from the same health plan that referred to the samemammography appointment. However, a mammogram paid for

by more than one health plan was counted twice becauseduplicates could not be eliminated by name. To minimize

double-counting, data for women 65 years of age and olderfrom Medicaid and from private insurers other than the Blue

CrossfBlue Shield insurer were not included because thesewomen have primary coverage through Medicare.

In addition to the 79 geographic ZIP codes in the ZIP file,

the postal service uses another 60 ZIP codes (29) for post officeboxes and major institutions. Whenever possible, the closest

geographic ZIP code was assigned to the 3.5% of insuranceclaims with one of these ZIP codes.

Strict confidentiality about the identity of women was

maintained at all times by managing and analyzing the datawithout identifiers. Encrypted account numbers were used to

track women over the 2-year period. Research proposals weresubmitted to all agencies providing data, to the Internal Review



Cancer Epidemiologj�, Biomarkers & Prevention 203

Board at the HMO, and to the University of Hawaii’s Corn-rnittee on Human Studies.

Data on breast cancer incidence included all cases identi-

fled by the Hawaii Tumor Registry, which is part of the SEERprogram. Quality control reviews have shown that case ascer-

tainment through the Hawaii Tumor Registry has been virtually

complete (30). All cases resident in Hawaii with codes 74.x

according to the International Classification of Disease for

Oncology (31) diagnosed during 1992 and 1993 were extracted.The nine SEER stage codes were summarized into in situ,

localized, and advanced.

The 1990 census ZIP file on CD-ROM (32), referred to as

the “ZIP code data file,” served as the population denominatorand as the source for demographic and socioeconomic infor-

mation. Because intercensal estimates were not available forindividual ZIP code areas, the population size in each ZIP code

area was increased by the state-wide population growth. Therates from 1990-1992 were 1 1.7% for the age group 40-49

years, 3.5% for the age group 50-64 years, and 7.5% for the

age group 65 years and older. The respective growth rates for

these groups from 1992-1993 were 3.8, 2.3, and 3.4%.The ZIP census file includes data by age and gender for only

broad ethnic groups: Caucasians, African-Americans, nativeAmericans, and Asian-Pacific islanders, a category that includes

most of the major ethnic groups in Hawaii. The breakdown into

more specific groups (Japanese, Chinese, Hawaiian, Korean, Sa-

moan, and so forth) by ZIP code is not detailed by age and gender

in the ZIP code data file and is not available from any other source.Therefore, four summary ethnic variables measuring the percent-age of persons with Caucasian, Japanese, Filipino, or Hawaiian

ancestry in each ZIP code area were used in this study.Data sets for the BRFSS since 1987 were obtained from

the Hawaii Department of Health. Annual mammography rates

were calculated by dividing the number of women 40 years ofage and older who reported a mammogram during the last 2

years by the number of women 40 years of age and older who

participated in the survey.

Statistical Analysis. After computing the number of womenwho received at least one mammogram during the 2-year pe-

riod, mammography rates for three age groups (40-49 years,

50-64 years, and 65 years and older) were calculated by

dividing the number of women with at least one mammogramduring 1992 and 1993 by the estimated number of women

living in the geographic area. Crude breast cancer incidence

rates were calculated for all stages combined and by three

categories (in situ, localized, and advanced). The analyses inthis study were first performed with all mammograms, regard-less of coding, and then repeated with the subset of mammo-

grams coded as screening mammograms. Because of the sim-

ilarity in results, only findings from the analysis using allmammograms are presented in this report. To determine areas

of high and low mammography utilization, a z-test to test fordifferences between proportions (33) was applied to analyze the

differences in mammography rates between individual ZIP

code areas and the entire state.Linear regression was used to predict aggregate breast

cancer incidence within zip codes; the models were weighted

by the number of women 40 years of age and older in 1992 and

1993 using an iterative procedure (34). Incidence was modeled asa function of mammography utilization, using both univariate andmultivariate analyses (35, 36). The stepwise method with a 0.15

significance level for entry into the model was applied to develop

final regression models. The only variable that was kept in everymodel was an indicator variable for ZIP code areas on Maui for

which claims for mammograms provided by the HMO in the first

6 months of 1992 were missing. Breast cancer incidence andmammography rates were treated as continuous variables withage-specific rates for the three age groups. Several potential con-founders were treated as continuous variables: the percentage of

the population 25 years and older with more than a high schooleducation in a ZIP code, the percentage of households with anannual income of less than $30,000, and the percentage of thepopulation that is Japanese, Caucasian, and Hawaiian. The van-

ables had an approximately normal distribution after weighting forpopulation size. Dummy variables were created for several poten-tial confounders: high proportion (more than 15%) of the popula-tion with Filipino ancestry, high proportion (more than 15%) of

military and veteran population, the presence of a mammographyfacility in the ZIP code area, and rural versus urban residence. Agewas included in the form of two categorical variables, one mdi-

cating 50-64 years versus 40-49 years and the other indicatingage 65 years and older versus 40-49 years. The assumptions of

multiple linear regression, i.e. linearity of association, constantvariance of error terms, independence of error terms, and normal-

ity of error terms were verified.Incidence rate data is often modeled with a log-linear

model (37). Our results were unchanged after log-transformingthe rates. Therefore, only the results based on the untrans-

formed rates are shown. All analyses were repeated using astepwise logistic regression modeling aggregate cancer rates

and including the same predictor variables. Finally, predicted

breast cancer incidence rates for different levels of mammog-raphy utilization were calculated according to the final linear

and logistic models, using mean population values for thedemographic variables (38). All data management and analysiswere performed on a personal computer using PC-SAS, release6.10 (SAS Institute, Cary, NC).

Results

Mammography Use. The final data set included 131,490mammograms received by women 40 years of age and older

during 1992 and 1993. Of these, 89,705 (68.2%) were coded asscreening mammograms, and 41 ,785 (3 1 .8%) were coded asdiagnostic mammograms. Because 26% of women receivedmore than I mammogram during the 2-year period, biennialrates are based on 97,610 women who received at least 1mammogram during 1992 and 1993 (Table 1). An estimated

42% of all women ages 40 years and older received at least 1mammogram in either 1992 or 1993, and 33% of women

received at least 1 screening mammogram (Table 1 ). Mammog-raphy rates were highest among women 50-64 years old (45%)and were lowest for women 40-49 years old (38%).

The counties of Honolulu and Hawaii had slightly higher

mammography rates than the counties of Kauai and Maui, butoverall, the differences by county were rather small. Mammog-raphy rates by ZIP code areas ranged from 9% to over 100%.The two ZIP code areas with rates above 100% belong to small

communities whose population was probably underestimatedbecause of rapid growth during recent years. Thirty-seven ZIPcode areas had a mammography rate higher than the state-wide

rate, and 23 areas had a rate lower than the state-wide rate.Visual inspection of the mapped data helped to identify geo-graphic patterns of screening participation (Fig. 2).

Breast Cancer Incidence. A total of I ,39 1 cases of breast

cancer were diagnosed among women ages 40 years and olderduring 1992 and 1993. The age-specific incidence rates were177, 321, and 419 per 100,000 women peryear for the three agegroups, respectively. The state-wide age-adjusted breast cancer



Oahu

Molokai

t::��:7 Maui

LanaiS�3

Kahoolawe

Honolulu

Big Island of Hawaii

LIUU

Lower Utilization than State

Utilization Similar to State

Higher Utilization than State


Table 1 Mammog raphy utilization during 2 years (rates per 100 women) by county, age grou p. year. and m ammography type, Hawaii, 1992-1993

Year Age group (yrs)County

State no. of mammogramsHawaii Honolulu Kauai Maui Rate

1992/1993 (overall)

1992/1993 screening

40-49

50-64

�65

All ages

40-49

50-64

�65

All ages

41.1

43.5

46.0

43.5

30.7

33.6

32.0

32.1

38.3

45.7

44.3

42.8

30.5

37.7

32.7

33.7

39.5

48.2

31.4

40.0

31.4

37.8

23.9

31.3

35.8

43.3

33.9

37.9

27.2

35.1

23.1

28.7

38.4

45.4

42.9

42.2

30.3

37.0

31.3

32.9

31.445

35.287

30,878

97,610

24,762

28.793

22,531

76.086

Niihau

Fig. 2. Mammography utilization by ZIP code area, Hawaii, 1992-1993.

incidence rate was 302 per 100,000 women ages 40 years andolder (adjusted to the 1990 Hawaii resident population ofwomen ages 40 years and older). Eighteen ZIP code areas hadno case, and another 17 had less than 5 cases of breast cancer.Thirty-seven (47%) ZIP code areas had no in situ case, and only25 (32%) had more than 3 in situ cases.

Relationship between Mammography Use and Breast Can-cer Incidence. A univariate analysis revealed (Table 2) thateducation, income, urban residence, mammography use, age 65

years and older, and the percentage of Japanese were positivelyrelated to breast cancer incidence, whereas the percentage ofFilipinos was inversely related. However, in a multivariate

analysis with all variables, education, age, and mammographyuse remained as the significant predictors. The stepwise pro-cedure resulted in a model with 50% of the variance in breast

cancer incidence (Table 3) predicted by a combination of mam-mography utilization (23%) and demographic variables. Age

Miles

0 20 40

and educational achievement were significant, contributing

more than 20% of the variance. The breast cancer incidence ratein a ZIP code area increased with age, mammography use, and

educational attainment. Excluding ZIP code areas with less than100 women 40 years of age and older did not change the results

of the analysis.The variation of in situ disease was not very predictable

(R2 0.16); in addition to mammography utilization and age,urban residence and the presence of a mammographic facility in

the area entered the model (Table 3). The explained variancewas highest for localized disease (R2 = 0.45); mammography

use contributed 17%, age contributed 24%, and post-secondaryeducation contributed 4% of variation. The rate for advanced

stage disease showed a very weak relationship with mammog-raphy (partial R2 0.03), and the model was not highly

predictive (R2 = 0.19).Mammography use (Table 3) predicted the largest amount of



Cancer Epidemiology, Blomarkers & Prevention 205

Table 2 Regressi on#{176}of breas t cancer incidence o n selected factors, Hawaii, 1992-1993

Variable Mean SD Mm MaxLinear Regression Logistic Regression

Univariate Multivariate” Univariate Multivariate”

Percent of population with post-secondary education 0.5 1 0. 1 1 0. 15 0.72 0.0032 0.0056 1 .080 1.324

0.601 0.604 0XKXl1 0.03

Households <$30,000 annual income 0.36 0.12 0.1 1 0.90 -0.0023 0.0008

0.609 0.49

-0.784

0.0004

-0.119

0.73

Percent of population Caucasian 0.32 0. 14 0.03 0.93 0.00005 -0.0016 0.015 1 0.424

0.95 0.32 0.94 0.44

Percent of population Hawaiian 0.12 0.09 0.02 0.67 -0.0016 0.0016 -0.589 1.252

0.20 0.25 0.08 0.04

Percent of population Japanese 0.25 0.1 1 0.01 0.49 0.0031 -0.0010 1.059 0.622

0.601 0.52 0.0001 0.3

Urban residence (yes/no) 0.70 0.46 0.00 1 .00 0.0005 0.0005 0. 1 77 0.177

0.03 0.13 0.004 0.09

Mammography facility in area 0.57 0.50 0.00 1.00 -0.0003 -0.00004 -0.086 0.030

0.25 0.84 0.02 0.64

Missing Kaiser data� 0.08 0.28 0.00 1.00 -0.0003 0.0002

0.51 0.66

-0.092

0.37

0.039

0.78

High (> 15%) proportion military and veteran 0.63 0.49 0.00 1.00 -0.0003 -0.0003 -0.095 -0.143

0.21 0.15 0.08 0.04

High (> 15%) proportion Filipino 0.40 0.49 0.00 1 .00 -0.0006 0.0002 -0. 19 0.126

0.01 0.56 0.0607 0.26

Mammography utilization 0.42 0.07 0.09 1 .54 0.0074 0.0054 1 .72 1 1.197

0.0001 0.6001 0.0601 0.0601

Age 50-64 yrs 0.34 20.9 0 1 0.0003 0.0010 0.101 0.523

0.18 0.tTXX3J 0.07 00001

Age 65 yrs and older 0.31 20.5 0 1 0.0017 0.0021 0.53 0.813

0.0001 0.0001 0.0601 OJXIOI

a Unit is ZIP code area; Mm, minimum; Max, maximum. P values are italicized.

b Model includes all variables in the table.

‘ Areas on Maui with missing Kaiser data for the first 6 months of 1992.

variation in breast cancer among women 50-64 years old (partial

R2 0.36). Again, educational achievement was positively related

to breast cancer incidence, and the presence of a mammographyfacility in the area was inversely related to breast cancer incidence.

For women 40-49 years old, only 15% of the variance wasattributable to any of these factors. Among women 65 years andolder, mammography utilization contributed 24% of the variance.

The proportion of military and veteran residents showed a weakinverse association with breast cancer incidence.

Plotting breast cancer incidence rates predicted from thefinal linear and logistic models (Table 3) illustrates the

central research question of whether mammography use canexplain the increase in breast cancer incidence in women 40years and older in the 1980s (Fig. 3). The 1982 and 1992/1993 incidence rates are represented by the two dotted lines(age-adjusted to the 1990 Hawaii female population 40 years

and older). According to the linear regression model, the1982 incidence rate corresponds to a mammography rate of

approximately 25%. According to the logistic regressionmodel, the 1982 incidence rate corresponds to a mammog-raphy rate of approximately 13%. Therefore, if the mam-mography rate was somewhere between 13 and 25% in 1982,the results of this study are consistent with the hypothesisthat the entire increase in breast cancer incidence was due togrowing mammography use. Extrapolating from the BRFSSrates of 38% in 1987, 60% in 1990, and 71% in 1993,mammography utilization was probably below 20% in 1982,

compatible with the 13-25% estimated in the geographicmodel. The small number of mammography facilities in thel980s also supports the presumption of low use of mam-

mography screening (Fig. I) during that time.

Discussion

Mammography utilization was directly associated with breast

cancer incidence on a geographic level after adjusting for

potential confounders using linear or logistic regression. Mam-

mography use, age, and educational achievement combined

predicted half of the variation in breast cancer incidence. The

finding that higher education is an important predictor of breast

cancer incidence is consistent with the literature (39). The

association between mammography utilization and breast can-

cer incidence was strongest for localized disease and for women

50-64 years old. The weak relationship between mammogra-

phy and the incidence of in situ cancer contradicts the 5-fold

increase of in situ incidence between 1983 and 1992 correlating

with the widespread adoption of mammography screening (40).

This discrepancy may be a result of the relatively small number

of in situ cases, a consequence of regional differences in the

quality of mammographic images, or, more likely, due togeographic differences in follow-up and biopsy rates. A

recent analysis of SEER data (41) suggested that variation in

criteria for performing breast biopsy after mammographymay be responsible for the wide range of in situ breast cancer

incidence rates (between 12 and 22 per 100,000 women)

among the 9 SEER regions. Biopsy rates related to abnor-malities in mammograms have not been published for

Hawaii. However, an informal survey among several of the

44 licensed mammography facilities in the state suggested a

large variation in follow-up practices and a wide range in

biopsy rates.

This study did not seek to explain the secular trend in breast

cancer incidence observed since the 1960s. This increase has been




Table 3 Stepwise regression” of breast cancer incidence, all mam mograms, Hawaii, 1992-1993

Stage/age Significant variables” Partial R2 Model R2 Parameter estimate SE

All stages

All ages

In situ

All ages

Localized

All ages

Advanced

All ages

All stages

40-49 yrs

All stages

50-64 yrs

All stages

�65 yrs

Missing Kaiser data’

Mammography utilization

Age 65 and older

Age 50-64 yrs

Percent of population with post-secondary education

Intercept

Missing Kaiser data�

Urban residence


Mammography facility in area

Age 65 yrs and older

Age 50-64 yrs

Intercept





Age 50-64 yrs

Intercept



Age 50-64 yrs


Intercept





Intercept

Missing Kaiser data�




Intercept

Missing Kaiser datac


High (>15%) proportion military

Intercept

<0.01

0.23

0. 18

0.05

0.04

NAd

0.02

0.07

0.02

0.02

0.01

0.02

NA

<0.01

0.21

0. 17

0.04

0.03

NA

<0.01

0.09

0.07

0.03

NA

0.02

0.06

0.04

0.03

NA

<0.01

0.36

0.07

0.02

NA

<0.01

0.24

0.04

NA

<0.01

0.23

0.41

0.46

0.50

NA

0.02

0.09

0. 1 1

0.13

0.14

0. 16

NA

<0.01

0.21

0.38

0.42

0.45

NA

<0.01

0.09

0. 16

0. 19

NA

0.02

0.08

0.12

0. 15

NA

<0.01

0.36

0.43

0.45

NA

<0.01

0.24

0.28

NA

-0.00003

0.0060

0.002 1

0.0009

0.0029

-0.0020

0.0004

0.0003

0.0005

-0.0001

0.0002

0.0002

0.00003

-0.00015

0.0014

0.004 1

0.0019

0.0005

-0.0015

0.00007

0.0005

0.0003

0.001 1

-0.0000

-0.0002

0.0028

0.0016

0.0003

-0.0003

-0.0001

0.0073

0.0041

-0.0005

-0.0019

0.0007

0.0086

-0.0007

0.0008

0.00028

0.0009

0.0002

0.0002

0.0007

0.0005

0.0001

0.00007

0.0003

0.00006

0.00008

0.00008

0.0001

0.0002

0.0001

0.0007

0.0005

0.0001

0.0003

0.00013

0.00009

0.00008

0.0004

0.0002

0.0003

0.0012

0.0008

0.0002

0.0007

0.0006

0.0015

0.0014

0.0003

0.0006

0.0007

0.0017

0.0004

0.0007

a Unit is ZIP code area.

b The following variables did not meet the 0.15 significance level for entry into the model: percent of households with less than $30,000 annual income, high Filipino,

percent Hawaiian, percent Japanese, or percent Caucasian population.

� Areas on Maui with missing Kaiser data for the first 6 months of 1992.d NA, not applicable.

partially attributed to reproductive factors (4) and dietazy factorssuch as alcohol consumption and carbohydrate intake (42). Prob-

ably other factors as yet undetermined are related to this gradualincrease. However, the findings of the study suggest that the rapid

rise in breast cancer incidence during the l980s can be attributedto mammography utilization alone.

This analysis estimated that at least 42% of women 40years of age and older in the state of Hawaii had at least 1mammogram during 1992 and 1993, much lower than the 71%of women who reported a mammogram during the last 2 years

in the BRFSS. Even inflating the estimate from the insuranceclaims data to account for missing information for 15% of thetarget population and assuming they had similar mammographyutilization only gives a state-wide rate of close to 50% (42%divided by 0.85). Selection bias in the BRFSS participants dueto a low participation rate (34% of phone calls resulted in a

completed interview) and a possible tendency of participants togive answers that conform to social desirability might explainthe large discrepancy.

A potentially serious challenge to the validity of anyecological study is the argument that the observed associa-

tion was due to the influence of the response variable (breast

cancer) on screening behavior (43), also referred to as cause-and-effect bias. If women at high risk for breast cancer weremore likely to utilize screening services than women at lowrisk for breast cancer, the hypothesis of this study wouldhave to be rejected. However, because the ability to predictindividual risk for breast cancer remains very limited as

compared to the estimation of breast cancer incidence inpopulations (4, 44-46), it seems unlikely that such a self-

selection for screening has actually occurred, except for the

small proportion of women with a family history or a priordiagnosis of breast cancer.

The unavailability of data for 15% of the population andthe lack of a unique identification number may have affectedthe results of this study through selection bias and double-counting. Accurate denominators for intercensal years on theZIP code level would have been desirable. As it was, the

number of women ages 40 years and older had to be estimatedunder the false assumption of equal population growth in allZIP code areas. The ecological nature of the ethnicity mayexplain the fact that ethnic variables were not significant in the



.-*- Linear Model

-.-- Logistic Model

Incidence 1992/93- - . - Incidence 1982

Mammography Rat.

Cancer Epidemiology, Biomarkers & Prevention 207

a§a

a

0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5

Fig. 3. Model of predicted breast cancer incidence rates, Hawaii, 1982-1993.

multivariate models, although it is known that ethnicity is arecognized determinant of breast cancer incidence. Also, be-cause of the strong association between ethnicity and socioeco-nomic status, ethnic variables lost significance in the statisticalmodel after adding education and income variables.

Miscoding of screening mammography as diagnostic

may have occurred because not all providers had changed

their coding procedures since 199 1 , when coverage forscreening mammography became mandatory. It is also pos-sible that some mammograms were purposely miscoded toobtain the higher reimbursement for diagnostic mammo-grams. At Kaiser Permanente, where coding is not related toreimbursement, only 17% of mammograms were coded di-

agnostic. However, the analyses based on all mammogramswould not be affected by this misclassification. The lack of

information on known and suspected risk factors for breast

cancer such as age of menarche and menopause, age at firstlive birth, diet, body weight, and estrogen levels probablyaccounts for some of the large amount of unexplained van-ance in breast cancer incidence.

This study has shown that it was possible to aggregate data

from different public and private payors and to use them forresearch purposes. Three features of this research project [thepopulation-based approach, a closed geographic area in whichmost health care is received within the state, and the availability of

insurance claims for a large proportion of the population (85% ofwomen 40 years of age and older)] provided a unique opportunityto study the association between mammography use and breastcancer incidence. In addition, the ecological approach meant thatindividuals did not have to be contacted, and refusal to participatedid not occur; thus selection bias was not a great problem except

for the data from unavailable data sources. Using insurance claimsdata may have great potential for geographic areas with healthinsurance systems that cover the majority of the population such as

Canada, Great Britain, and the Scandinavian countries. This typeof surveillance is useful in identifying population segments with

low mammography use so that women in these subgroups can betargeted for interventions.

Considering possible biases and alternative explanations,how are the results of this study to be interpreted? Given the

unlikelihood of some of the potential biases, the hypothesis thatpopulation-based mammography screening is related to in-creased breast cancer incidence seems plausible. However, with

a perfect data set, the strength of the true association is morelikely to be weaker than the one found here because grouping

by geography tends to exaggerate relationships (43). Mammog-naphy use in this context was a surrogate for other factors, such

as health care-seeking behavior, access to health care, andbeliefs about disease, probably resulting in a higher associationthan the “true” association between mammography screeningand breast cancer (47). Caucasian women in Hawaii haveamong the highest breast cancer incidence rates among SEER

areas, possibly due to the availability of and insurance coveragefor mammography. Participation in mammography screeningwas less than 50% in the United States (48, 49) as compared to

71% in Hawaii (50).A significant question is whether all screening-detected

cancers would have ever become clinically apparent orwhether some small tumors would have remained asymp-tomatic until the end of a woman’s life. There is evidence

that screening-detected breast cancers have a slower growthrate and a lower malignant potential then symptomatic breast

cancers (5 1 , 52). Four autopsy studies (53-56) using rela-tively unselected series of autopsies in populations with alow use of mammography screening found invasive breast

cancer in 1.0, 1.3, 0.9, and 1.5% of women 15-98 years ofage previously undiagnosed with breast cancer, respectively.The prevalence of in situ breast cancer in autopsies dependedon the intensity of breast tissue sampling. Two studies fromDenmark (54, 55) found a prevalence of 18%, whereas a

study from Australia (56) detected in situ cancers in 13% ofautopsied women. The high rate of undetected breast cancers

in women who died from other causes supports the idea that

cancerous lesions develop in a large proportion of femalebreasts and never become clinically apparent. Research intothe pathology of tumor growth suggests that cancerous le-sions do not always progress to become invasive and thatsome of them may even regress (57). Therefore, if the

Danish autopsy studies were correct, improved imagingtechniques have the potential to detect early stage breastcancer cases in as many as 19% of women. Because thecumulative lifetime breast cancer incidence for women in the

United States (39) is currently estimated at 1 1%, breastcancer incidence rates may continue to increase until every prey-

alent breast cancer case has been detected. However, a largeproportion of screening-detected breast cancers will experience no

benefit from treatment because they may have never progressed ormay have even regressed ifleft undetected. Treatment itself causesmorbidity and some mortality, in addition to the psychologicaltrauma of being diagnosed with breast cancer. One challenge for

future research is to identify characteristics of mammographiclesions that distinguish cancers that require treatment from rela-tively benign cancers that will not influence a woman’s life ex-pectancy.

Acknowledgments

We thank the staff at the Hawaii Department of Health, Office of Health Status

Monitoring; the personnel at the Hawaii Tumor Registry; G. M’s dissertationcommittee; and the health plans who prepared the data sets, in particular, Dr.

Andy White of the Hawaii Medical Service Association and Darrell Kikuchi of

Kaiser Permanente.

References

I. Ries, L. A. G., Miller, B. A., Hankey, B. F., Kosary, C. L., Harras, A., and

Edwards, B. K. SEER Cancer Statistics Review, 1973-1991: Tables and Graphs.NIH PubI. No. 94-2789. Bethesda, MD: National Cancer Institute, 1994.




2. Smigel, K. Breast cancer death rates decline for white women. J. NatI. Cancer

Inst., 87: 173, 1995.

3. Oyama, N., and Johnson, D. B. Hawaii Health Surveillance Program Survey

Methods and Procedures. Research & Statistics Report No. 54. Honolulu, Hawaii:

Hawaii State Department of Health, 1986.

4. Madigan, M. P., Ziegler, R. G., Bemchou, J., Byrne, C., and Hoover, R. N.

Proportion of breast cancer cases in the United States explained by well-estab-

lished risk factors. J. Natl. Cancer Inst., 87: 1681-1685, 1995.

5. Hurley, S. F., and Kaldor, J. M. The benefits and risks of mammographic

screening for breast cancer. Epidemiol. Rev., 14: 101-130, 1992.

6. Chalmers, T. C. Screening for breast cancer: what should national healthpolicy be? J. Natl. Cancer Inst., 85: 1619-1621, 1993.

7. Fletcher, S. W., Black, W., Harris, R., Rimer, B. K., and Shapiro, S. Report of

the International Workshop on Screening for Breast Cancer. J. NatI. Cancer Inst.,

85: 1644-1656, 1993.

8. American Cancer Society. Cancer Risk Report: Prevention and Control, p. 17.

Atlanta, GA: American Cancer Society, 1994.

9. White, E., Lee, C. Y., and Kristal, A. R. Evaluation of the increase in breast

cancer incidence in relation to mammography use. J. Natl. Cancer Inst., 82:

1546-1552, 1990.

10. Glass, A. G., and Hoover, R. N. Rising incidence of breast cancer: relation-

ship to stage and receptor status. J. Natl. Cancer Inst., 82: 693-696, 1990.

1 1. Lantz, P. M., Remington, P. L., and Newcomb, P. A. Mammography screen-

ing and increased incidence of breast cancer in Wisconsin. J. Nail. Cancer Inst.,

83: 1540-1546, 1991.

12. Luff, J. M., Sung, J. F., Chow, W-H., Greenberg, R. S., and Flanders, W. D.

Does increased detection account for the rising incidence of breast cancer? Am. J.

Public Health, 81: 462-465, 1991.

13. Newcomb, P. A., and Lantz, P. M. Recent trends in breast cancer incidence,

mortality, and mammography. Breast Cancer Res. Treat., 28: 97-106, 1993.

14. Feuer, E. J., and Wun, L. M. How much of the recent breast cancer incidencecan be explained by increases in mammography utilization? A dynamic popula-

tion model approach. Am. J. Epidemiol., 136: 1423-1436, 1992.

15. Miller, B. A., Feuer, E. J., and Hankey, B. F. The significance of the rising

incidence of breast cancer in the United States. V. T. DeVita, S. Hellman, and S.

A. Rosenberg (eds.), Important Advances in Oncology 1994, pp. 193-207.

Philadelphia: J. B. Lippincott Co., 1994.

16. Kessler, L. G., Feuer, E. J., and Brown, M. L. Projections ofthe breast cancer

burden to U.S. women: 1990-2000. Prey. Med., 20: 170-182, 1991.

17. Shapiro, S. More on screening and breast cancer incidence. J. Natl. Cancer

Inst., 83: 1522-1523, 1991.

18. Hawaii Medical Service Association Foundation. Health Trends in Hawaii,

pp. 84-85. Honolulu, HI: Hawaii Medical Service Association, 1995.

19. International Classification of Disease, 9th Revision, Clinical Modification.

Salt Lake City, UT: Medicode, 1995.

20. American Medical Association. Physician’s Current Procedural Terminology

1995. Chicago, IL: American Medical Association, 1994.

21. Roos, L. L., Nicol, J. P., and Cageorge, S. M. Using administrative data for

longitudinal research: comparisons with primary data collection. J. Chronic Dis.,

40: 41-49, 1987.

22. Potosky, A. L., Riley, G. F., Lubitz, J. D., Mentnech, R. M., and Kessler, L.

G. Potential for cancer-related health services research using a linked Medicare

tumor registry database. Med. Care (Phila.), 31: 732-748, 1993.

23. Potosky, A. L., Miller, B. A., Albertsen, P. C., and Kramer, B. S. The role of

increasing detection in the rising incidence of prostate cancer. J. Am. Med.

Assoc., 273: 548-552, 1995.

24. Fisber, E. S., Whaley, F. S., Krushat, W. M., Malenka, D. J., Fleming, C., Baron,

J. A., and Hsia, D. C. The accuracy ofMedicare’s hospital claims data: progress hasbeen made, but problems remain. Am. J. Public Health, 82: 243-248, 1992.

25. Young, T. K., Roos, N. P., and Hammerstrand, K. M. Estimated burden of

diabetes mellitus in Manitoba according to health insurance claims: a pilot study.

Can. Med. Assoc. J., 144: 318-324, 1991.

26. Roos, L. L., Mustard, C. A., Nicol, J. P., Comm, B., McLerran, D. F.,

Malenka, D. J., Young, T. K., and Cohen, M. M. Registries and administrative

data: organization and accuracy. Med. Care (Phila.), 31: 201-212, 1993.

27. Mossey, J. M., Havens, B., Roos, N. P., and Shapiro, E. The Manitoba

longitudinal study on aging: description and methods. Gerontologist, 21: 551-

558, 1981.

28. Hawaii Department of Business, Economic Development & Tourism. State of

Hawaii Data Book 1993-94: A Statistical Abstract, p. 35. Honolulu, HI: State of

Hawaii, 1994.

29. United States Postal Service. 1993 National Five Digit ZIP Code & Post

Office Directory. Washington, DC: The National Information Data Center,

1993.

30. Goodman, M. T., Yoshizawa, C. N., and Kolonel, L. N. Descriptive epide-

miology of thyroid cancer in Hawaii. Cancer (Phila.), 61: 1272-1281, 1988.

31. Percy, C., Van Holten, V., and Muir, C. International Classification of

Disease for Oncology, p. 13. Geneva: WHO, 1990.

32. U.S. Department of Commerce, Bureau of the Census. 1990 Census of Popula-

tion and Housing, summaiy tape file 3B, ZIP codes 7. 8, 9. CD9O-3B-03, 1993.

33. Armitage, P. Statistical Methods in Medical Research, pp. 1 1 1-1 15. Oxford,United Kingdom: Blackwell Scientific Publications, Ltd., 1971.

34. Steel, R. G. D., and Tome, J. H. Principles and Procedures of Statistics: ABiomedical Approach, pp. 269-270. New York: McGraw-Hill, Inc., 1980.

35. Hair, J. F., Anderson, R. E., Tatham, R. L., and Black, W. C. Multivariate

Data Analysis, pp. 78-166. Englewood Cliffs, NJ: Prentice Hall, 1995.

36. Littell, R. C., Freund, R. J., and Spector, P. C. SAS System for Linear

Models. Cary, NC: SAS Institute, Inc., 1991.

37. Breslow, N. E., and Day, N. E. Statistical Methods in Cancer Research, Vol. I:

The Design and Analysis ofCase Control Studies, p. 57. Lyon. France: [ARC, 1980.

38. Rothman, K. J. Modem Epidemiology. Boston: Little, Brown & Co., 1986.

39. Kelsey, J. L., and P. L. Hom-Ross. Breast cancer: magnitude of the problem

and descriptive epidemiology. Epidemiol. Rev., 15: 7-16, 1993.

40. Emster, V. L., Barclay, J., Kerlikowske, K., Grady, D., and Henderson, I. C.

Incidence of and treatment for ductal carcinoma in situ of the breast. J. Am. Med.

Assoc., 275: 913-918, 1996.

41. Choi, W. S., Parker, B. A., Pierce, J. P., and Greenberg, E. R. Regional

differences in the incidence and treatment of carcinoma in situ of the breast.

Cancer Epidemiol., Biomarkers & Prey., 5: 317-320, 1996.

42. Ewertz, M., and Duffy, S. W. Incidence of female breast cancer in relation to

prevalence or risk factors in Denmark. Int. J. Cancer, 56: 783-787, 1994.

43. Morgenstem, H. Uses of ecologic analysis in epidemiologic research. Am. J.

Public Health, 72: 1336-1344, 1982.

44. Gail, M. H., Brinton, L. A., Byar, D. P., Corle, D. K., Green, S. B., Schairer,

C., and Mulvihill, J. J. Projecting individualized probabilities ofdeveloping breast

cancer for white females who are being examined annually. J. Natl. Cancer Inst.,

81: 1879-1886, 1989.

45. Spiegelman, D., Colditz, G. A., Hunter, D., and Hertzmark, E. Validation of

the Gail et al. model for predicting individual breast cancer risk. J. Nail. CancerInst., 86: 600-607, 1994.

46. Bondy, L. B., Lustbader, E. D., Halabi, S., Ross, E., and Vogel, V. G.

Validation of a breast cancer risk assessment model in women with a positive

family history. J. Nail. Cancer Inst., 86: 620-625, 1994.

47. Schwartz, S. The fallacy of the ecological fallacy: the potential misuse of a

concept and the consequences. Am. J. Public Health, 84: 819-824, 1994.

48. Centers for Disease Control and Prevention. Mammography and clinical

breast examinations among women aged 50 years and older: behavioral risk factor

surveillance system, 1992. Morb. Mortal. Wkly. Rep., 42: 737-741, 1993.

49. Anderson, L. M., and May, D. S. Has the use of cervical, breast, and

colorectal cancer screening increased in the United States? Am. J. Public Health,85: 840-842, 1995.

50. Hawaii Department of Health, Health Promotion and Education Branch.Hawaii’s Health Risk Behaviors, 1993, pp. 105-109. Honolulu, HI: Hawaii

Department of Health, 1995.

51. Klemi, P. J., Jocnsuu, H., Toikkanen, S., Tuominen, J., Raesaenen, 0.,

Tyrkkoe, J., and Parvinen, I. Aggressiveness of breast cancers found with and

without screening. Br. Med. J., 304: 467-471, 1992.

52. Henderson, C. Biologic variations of tumors. Cancer (Phila.), 69: 1888-

1895, 1992.

53. Bartow, S. A., Pathak, D. R., Black, W. C., Key, C. R., and Teaf, S. R.

Prevalence of benign, atypical, and malignant breast lesions in populations at

different risk for breast cancer. Cancer (Phila.), 660: 2751-2760, 1987.

54. Nielsen, M., Jensen, J., and Andersen, J. Precancerous and cancerous breastlesions during lifetime and at autopsy. Cancer (Phila.), 54: 612-615, 1984.

55. Nielsen, M., Thomsen, J. L., Primdahl, S., Dyreborg, U., and Andersen, J. A.

Breast cancer and atypia among young and middle-aged women: a study of 110

medicolegal autopsies. Br. J. Cancer, 56: 814-819, 1987.

56. Bhatal, P. S., Brown, R. W., Lesueur, G. C., and Russell, I. S. Frequency of

benign and malignant breast lesions in 207 consecutive autopsies in Australianwomen. Br. J. Cancer, 51: 271-278, 1985.

57. Clark, W. H. The role oftumor progression in prevention ofcancer and reduction

of cancer mortality. P. Greenwald, B. S. Kramer, and D. L Weed (cdx.), Cancer

Prevention and Control, pp. 135-159. New York: Marcel Dekker, Inc.. 1995.



1997;6:201-208. Cancer Epidemiol Biomarkers Prev G Maskarinec, L Wilkens and L Meng incidence in Hawaii.Mammography screening and the increase in breast cancer

Updated version

http://cebp.aacrjournals.org/content/6/3/201

Access the most recent version of this article at:

E-mail alerts related to this article or journal.Sign up to receive free email-alerts

Subscriptions

Reprints and

[email protected] at

To order reprints of this article or to subscribe to the journal, contact the AACR Publications

Permissions

Rightslink site. Click on "Request Permissions" which will take you to the Copyright Clearance Center's (CCC)

.http://cebp.aacrjournals.org/content/6/3/201To request permission to re-use all or part of this article, use this link



http://cebp.aacrjournals.org/cgi/alerts

mailto:[email protected]



mammography screening and the increase in breast...

Documents