JNCI J Natl Cancer Inst 2002 Ernster 1546 54
description
Transcript of JNCI J Natl Cancer Inst 2002 Ernster 1546 54
Detection of Ductal Carcinoma In Situ in WomenUndergoing Screening MammographyVirginia L. Ernster, Rachel Ballard-Barbash, William E. Barlow, Yingye Zheng,Donald L. Weaver, Gary Cutter, Bonnie C. Yankaskas, Robert Rosenberg, PatriciaA. Carney, Karla Kerlikowske, Stephen H. Taplin, Nicole Urban, Berta M. GellerBackground: With the large number of women having mam-mographyanestimated28.4millionU.S. womenaged40years and older in 1998the percentage of cancers detectedas ductal carcinoma in situ (DCIS), which has an uncertainprognosis, has increased. Wepooleddatafromsevenre-gional mammography registries to determine the percentageofmammographicallydetectedcancersthatareDCISandthe rate of DCIS per 1000 mammograms. Methods: We ana-lyzeddataon653 833mammogramsfrom540738womenbetween40and84yearsofagewhounderwentscreeningmammographyat facilities participatinginthe NationalCancerInstitutesBreastCancerSurveillanceConsortium(BCSC)throughout1996and1997. Mammographyresultswere linked to population-based cancer and pathology reg-istries. We calculatedthe percentage of screen-detectedbreastcancersthatwereDCIS,therateofscreen-detectedDCIS per 1000 mammograms by age and by previous mam-mography status, and the sensitivity of screening mammog-raphy. Statistical testsweretwo-sided. Results: Atotal of3266casesofbreastcancerwereidentified, 591DCISand2675 invasive breast cancer. The percentage of screen-detected breast cancers that were DCIS decreased with age(from28.2%[95%confidence interval (CI) =23.9%to32.5%]forwomenaged4049yearsto16.0%[95%CI=13.3% to 18.7%] for women aged 7084 years). However, therate of screen-detected DCIS cases per 1000 mammogramsincreasedwithage(from0.56[95%CI=0.41to0.70]forwomen aged 4049 years to 1.07 [95% CI = 0.87 to 1.27] forwomen aged 7084 years). Sensitivity of screening mammog-raphy in all age groups combined was higher for detectingDCIS (86.0% [95% CI = 83.2% to 88.8%]) than it was fordetecting invasive breast cancer (75.1% [95% CI = 73.5% to76.8%]). Conclusions: Overall, approximately1inevery1300 screening mammography examinations leads to a diag-nosis of DCIS. Given uncertainty about the natural historyof DCIS, theclinical significanceof screen-detectedDCISneeds further investigation. [JNatl Cancer Inst 2002;94:154654]Incidencerates of ductal carcinomainsitu(DCIS) of thebreast, a noninvasive form of breast cancer, have risen dramati-callyintheUnitedStatesandelsewheresincetheearly1980s(15).TheincreaseinDCISincidenceparallelsanincreaseinthe use of screening mammography, which makes the detectionof DCIS much more likely than in the past. DCIS is a diverseprocess of the breast epithelium that is, by definition, confinedwithin the breast duct above the basement membrane. The cel-lular appearance of DCIS varies from low-grade lesions similarto atypical hyperplasia to high-grade or anaplastic lesions. Be-causeDCISisexcisedwhendetected, itisnotpossibletoac-curately estimate the fraction of untreated cases that would prog-ress to invasive malignancy. In addition to undergoing surgicaltreatment (lumpectomy or mastectomy), women with DCIS arefrequentlytreatedwithradiationor hormonetherapy(1,6,7).Moreover, inadditiontoothertreatment, axillarylymphnodedissectionmayalsobeperformedfor womenwithDCIS, al-though this procedure is not standard of care (6,7).It is unlikelythat there will ever be definitive studies todemonstratewhetherdetectionofDCISbymammographyhascontributed to the reduction in breast cancer mortality associatedwith screening mammography. However, it seems reasonable toconclude that some women benefit from having DCIS detectedthrough screening mammography, whereas other women do not,dependingonwhetherthediseasewaslikelytoprogress. Be-causeit is not possibletoscreenwomenfor invasivebreastcancer alone, thedetectionof DCISispart of anyscreeningmammographyprogram. Arecent surveyof 479womeninthe United States found that few women had heard about DCISor were aware that some forms of breast cancer might not pro-gress (8). When these women were informed, however, 60% ofthem wanted to take the possibility of DCIS detection into ac-count when deciding about participating in screening mammog-raphy. Noinformationiscurrentlyavailableabout therateofscreen-detectedDCIS. Therefore, wepooleddatafrommam-mographyregistriesacrosstheUnitedStatestodeterminethepercentageof DCISamongall screen-detectedbreast cancersandtherateofscreen-detectedDCISper1000mammogramsperformed.Affiliations of authors: V. L. Ernster, Department of EpidemiologyandBiostatistics, School of Medicine, Universityof California, SanFrancisco;R. Ballard-Barbash, Applied Research Program, Division of Cancer Control andPopulationSciences, NationalCancerInstitute, Bethesda, MD;W. E. Barlow,Center for Health Studies, Group Health Cooperative, Seattle, WA, and Depart-ment of Biostatistics, University of Washington, Seattle; Y. Zheng, DepartmentofBiostatistics, UniversityofWashington;D. L. Weaver, (DepartmentofPa-thology), B. M. Geller (HealthPromotionResearch), Universityof Vermont,College of Medicine, Burlington, VT; G. Cutter, Center for Research Design andStatistical Methods, University of Nevada, Reno; B. C. Yankaskas, Departmentof Radiology, University of North Carolina, Chapel Hill; R. Rosenberg, Depart-ment of Radiology, Universityof NewMexico, Albuquerque; P. A. Carney,Norris Cotton Cancer Center/Dartmouth-Hitchcock Medical Center/Departmentof Community and Family Medicine, Dartmouth Medical School, Lebanon, NH;K. Kerlikowske, Department of Epidemiology and Biostatistics, School of Medi-cine, andGeneralInternalMedicineSection, DepartmentofVeteransAffairs,University of California, San Francisco; S. H. Taplin, Center for Health Studies,Group Health Cooperative, Seattle; N. Urban, Fred Hutchinson Cancer ResearchCenter, Division of Public Health, Seattle.Correspondence to: Rachel Ballard-Barbash, M.D., M.P.H., Applied ResearchProgram, National Cancer Institute, EPN4005, 6130ExecutiveBlvd., MSC7344, Bethesda, MD 20892-7344 (e-mail:[email protected]).See Notes following References. Oxford University Press1546 ARTICLES Journal of the National Cancer Institute, Vol. 94, No. 20, October 16, 2002 by guest on July 3, 2015http://jnci.oxfordjournals.org/Downloaded from METHODSStudy PopulationWeobtainedpatient historyandclinical andradiologicas-sessment data from screening mammography examinations fromthe mammography registries of the Breast Cancer SurveillanceConsortium (BCSC) performed from January 1996 through De-cember 1997. The mammography registries were located in Col-orado, New Hampshire, New Mexico, North Carolina, San Fran-cisco (CA), Vermont, and western Washington State. TheBCSC, fundedbytheNationalCancerInstitute, anditsconfi-dentiality procedures are described elsewhere (9,10). Each mam-mographyregistryhadInstitutionalReviewBoardapprovaltocollect data from mammography facilities for research purposes.All patient identifiers wereremovedfromthecollecteddatabeforetheyweretransferredtotheBCSCsStatisticalCoordi-nating Center for analysis.Thepresent analysis is limitedtotheresults of screeningmammographyexaminations for womenbetween40and84years old who had no self-report of previous breast cancer. To beeligible for inclusion, the womans mammography examinationhadtobedesignatedasascreeningexaminationratherthanadiagnostic examination by the radiologist. We excluded unilat-eral screening examinations and examinations that did not havean assessment code indicating whether they had been considerednegativeor positivefor anabnormalityindicativeof cancer.Assessment codes were based on the American College of Radi-ologysBreast ImagingReportingandDataSystem(BI-RADS)(11). We alsoexcludedmammographic examinations if thewomenhadhadanybreast imaging(includingultrasoundoranothermammogram)inthepreceding9months,becauseim-agingwithinthisperiodmayindicatethatthescreeningmam-mographicexaminationwasnot atruescreeningexaminationbut ratherafollow-upexamination. Weused9monthsratherthanalongertimeperiodasthecutoffbecausewomensome-times schedule their annual mammographic examination amonth or two early. All screening examinations from the samewomanoverthe2-yearstudyperiodwereincluded,subjecttothe 9-month exclusion rule.Mammogram Assessment, Categorization, and CancerDiagnosesMammograms were linked to either cancer or pathology reg-istries in the same geographic region as the mammography reg-istry. Not all mammography facilities in a particular geographicregionare includedinthe BCSCfor that region, sonot allmammograms for that region were ascertained; however, ascer-tainment of cancers is believed to be almost complete. StatisticsbystateforeachcancerregistryaremaintainedbytheNorthAmericanAssociationof Central Cancer Registriesandshowadjustedcompletionratesbasedonobservedversusexpectedrates for all cancers in excess of 94.3% in all states included inthis studyexcept for Vermont, whichdidnot havestatisticsavailable(12).CancerrecordsfromtheNationalCancerInsti-tutes Surveillance, Epidemiology, and End Results (SEER) Pro-gram1wereusedforthethreeBCSCsitesthat arelocatedinSEER registry areasthat is, New Mexico, San Francisco, andwestern Washington Stateand a fourth site, Colorado, used anon-SEERcancerregistrythat usesSEERguidelinesfordatacollection. For the other three sites, where there can be a longdelaybetweencancerdiagnosisandappearanceofthecancerrecord in the local cancer registry, pathology registries served astheprimarysourceofcasereporting, supplementedbycancerregistry data. In this article, we present combined data from allmammography registries participating in the BCSC, recognizingthat there may be some variation in cancer ascertainment acrossregistries and, therefore, potential variation in cancer incidencerates.Cases werecategorizedas either DCISor invasivebreastcancer based on either cancer registry or pathology registry data.Cases ofinsitu lesions that were specifically coded as lobularcarcinoma insitu(LCIS) were excluded, but other cases ofinsitulesionswereincluded(forexample, insitulesionsthatwerecodedasnot otherwisespecified). Thus, someinsitulesions other than DCIS were categorized as DCIS, but the num-ber of these was small. If a cancer case was coded within 60 daysof diagnosis as having both invasive and in situ components, itwas categorized as invasive breast cancer rather than as DCIS.For comparison with the DCIS findings, we also provide somefindings for invasive breast cancer; however, the data on inva-sive breast disease were not examined in detail.Throughoutthisarticlethetermscreen-detectedreferstoscreening mammography and not to other forms of breast cancerdetection. Weusedtheinitial mammographicassessment cat-egory assigned to each screening mammography examination todetermine whether cases of DCIS and invasive breast cancer hadbeen screen-detected. We considered cases of DCIS or invasivebreast cancer to be screen-detected (i.e., positive) if the patientsmost recent eligible screening mammography examination in the365days prior todiagnosis resultedinanyof thefollowingBI-RADSassessment codes: 0(needs additional evaluation);3 (probably benign finding), with a recommendation for imme-diate work-up; 4 (suspicious abnormality); or 5 (highly sugges-tiveofmalignancy). CasesofDCISorinvasivebreast cancerwereconsiderednon-screen-detected(i.e., negative)ifthepa-tientsmostrecenteligiblescreeningmammographyexamina-tion in the 365 days prior to diagnosis had resulted in a BI-RADSassessment code of 1 (negative); 2 (benign finding); or 3 (prob-ablybenignfinding), withnorecommendationfor immediatework-up or with an unknown recommendation. These case defi-nitions were used to generate the numerators for calculating ratesper 1000mammograms for screen-detectedandnon-screen-detectedDCISandinvasivebreast cancer. Thedenominatorsusedincalculatingrateswerethenumberofscreeningmam-mography examinations, not the number of women. The follow-up period after a screening examination was 365 days or the nextscreeningexamination, whichever camefirst. This follow-upperiodensuredthat adiagnosis of cancer was allocatedtoasingle screening mammography examination.Data AnalysisWe first determined the sensitivity of screening mammogra-phyfordetectingDCISandinvasivebreastcancer.Sensitivitywasdefinedasthenumberofcaseswithapositivemammo-graphic assessment at screening divided by the total number ofwomen. We then calculated the percentage of all screen-detectedand non-screen-detected breast cancers that were DCIS, and therates of screen-detected and non-screen-detected DCIS per 1000mammograms by age group (4049 years, 5059 years, 6069years, and 7084 years) and for all ages combined. The denomi-nator used to calculate the DCIS rate in each category was thetotal number of screeningexaminationsinthat category. WeJournal of the National Cancer Institute, Vol. 94, No. 20, October 16, 2002 ARTICLES 1547 by guest on July 3, 2015http://jnci.oxfordjournals.org/Downloaded from further cross-classified the percentage and rate variables by pre-vious mammographystatus as determinedbywomens self-reportedmammographyhistory(screeningordiagnosticmam-mogram). At the time of screening mammography, women wereroutinelyaskediftheyhadhadapriormammographicexami-nationandthedateoforapproximatetimeinyearssincethatexamination. Women who did not provide complete informationabout historyof previousmammographywerecategorizedashavingnopreviousmammography;461 129(71%)womenre-porteda historyof a prior mammographic examinationand192 704(29%)womendidnot report havingapriormammo-graphic examination.Data from individual BCSC sites on percentages of cancersthat are DCIS and rates of DCIS per 1000 mammograms are notreported, but ranges of these variables over the seven BCSC sitesarepresented(seeTables3and4). Ninety-fivepercentconfi-dence intervals (CIs) were calculated for sensitivity, for percent-age of cancers that are DCIS, and for the rate of DCIS per 1000mammogramsbasedonthepooledBCSCdataandwerenotcorrectedformultiplecomparisons.Weassumedanormalap-proximationtothebinomialdistributionforproportionsandaPoissondistributionfor eachcancer rate. Standardchi-squaretestswereusedasstatisticaltestsforproportions,andPoissonregressionwas usedas the statistical test for comparisonofcancer rates. Statistical tests were two-sided. We also plotted thedistributionsof time-to-diagnosisfor casesdefinedasscreen-detected and those defined as non-screen-detected.RESULTSDemographic and Clinical VariablesAcrossthesevenBCSCsites, datafrom653833screeningmammographyexaminations(rangingfrom43 624to150387examinations among 540 738 women were available for evalu-ation (Table 1). Thus, approximately 20% of women contributedmore thanone screeningmammographyexaminationtotheanalysis. Basedonquestionnairedatacollectedat thetimeofscreening, 29%ofallmammographicexaminationswereclas-sifiedasfirst ever: 33%for ages4049years, 27%for ages5059 years, 28% for ages 6069 years, and 29% for ages 7084years. Race was self-reported for 86% of the women who werescreened; of these women, 79% were white, 5% were African-American, 2%were Asian/Pacific Islander, 2%were NativeAmerican, and12%wereother/mixed. Ofthe81%ofwomenwhorespondedtoaquestionaboutwhethertheywereofHis-panic origin, 4% reported being Hispanic. Educational level waseither not asked for or not reported for 39% of women. Of thosewomenwhodidreporteducationallevel, 34%completedcol-lege, 57% completed high school but not college, and 9% did notcompletehighschool. Becausethis demographicinformationwas not reported for many women, we did not analyze the resultsof screening mammography data by race or educational level.Linkageof mammographyscreeningdataandcancer casereports resultedintheidentificationof 3266cases of breastcancer (2675 invasive and 591 DCIS) occurring within 365 daysfollowingscreeningmammography. The distributions of thenumber of screening mammography examinations, DCIS cases,andinvasivebreastcancercasesbyagearegiveninTable1,where we further distribute by positive (screen-detected) cases.Sensitivity of Screening Mammography for DetectingDCISThe sensitivityof screeningmammographyfor detectingDCISwashigher thanthat for invasivebreast cancer. Of allDCIS cases diagnosed during the study period, 86.0% (508/591,95%CI 83.2%to88.8%) wereassociatedwithapositivescreeningmammographyassessment (comparedwith75.1%(2010/2675, 95% CI 73.5% to 76.8%) of all invasive breastcancer cases (Table 1). Sensitivity of screening mammographyfordetectingDCISdidnot differstatisticallysignificantlybyage (P .47). By comparison, sensitivity for detecting invasivebreast cancer increased statistically significantly with age(Ptrend