Cancer Epidemiology 38 (2014) 619–622

Increased risk of breast cancer in women with false-positive test:The role of misclassification

My von Euler-Chelpin a,*, Megumi Kuchiki b, Ilse Vejborg b

a Department of Public Health, University of Copenhagen, Øster Farimagsgade 5, DK-1014 Copenhagen, Denmarkb Department of Radiology, University Hospital of Copenhagen, Rigshospitalet, Blegdamsvej 9, DK-2100 Copenhagen, Denmark

A R T I C L E I N F O

Article history:

Received 3 March 2014

Received in revised form 16 June 2014

Accepted 22 June 2014

Available online 14 July 2014

Keywords:

Breast cancer

Screening

Mammography

False-positive

Misclassification

A B S T R A C T

Introduction: Studies have shown that women with a false-positive result from mammography screening

have an excess risk for breast cancer compared with women who only have negative results. We aimed

to assess the excess risk of cancer after a false-positive result excluding cases of misclassification, i.e.

women who were actually false-negatives instead of false-positives.

Method: We used data from the Copenhagen Mammography Screening Programme, Denmark. The study

population was the 295 women, out of 4743 recalled women from a total of 58,003 participants, with a

false-positive test during the screening period 1991–2005 and who later developed breast cancer.

Cancers that developed in the same location as the finding that initially caused the recall was studied in-

depth in order to establish whether there had been misclassification.

Results: Seventy-two cases were found to be misclassified. When the women with misclassified tests

had been excluded, there was an excess risk of breast cancer of 27% (RR = 1.27, 95% confidence interval

(CI), 1.11–1.46) among the women with a false-positive test compared to women with only negative

tests. Women with a false-positive test determined at assessment had an excess risk of 27%, while false-

positives determined at surgery had an excess risk of 30%.

Conclusions: The results indicate that the increased risk is not explained only by misclassification. The

excess risk remains for false-positives determined at assessment as well as at surgery, which favours

some biological susceptibility. Further research into the true excess risk of false positives is warranted.

� 2014 Elsevier Ltd. All rights reserved.

Contents lists available at ScienceDirect

Cancer EpidemiologyThe International Journal of Cancer Epidemiology, Detection, and Prevention

jou r nal h o mep age: w ww.c an cer ep idem io log y.n et

1. Introduction

Breast cancer is the most common cancer in women [1].Screening for breast cancer with mammography has proven tolower the mortality from breast cancer in the population [2–5].However, with screening also follows some disadvantages such asfalse-positive tests. A false-positive test refers to women who arerecalled for further assessment after a screening mammogram, andthen found to be free of breast cancer. The risk of getting a false-positive test varies greatly between screening programmes, sowhile the estimated risk over 10 screens in the US varies between58% and 77%, with an estimate of 63% as the most reasonableassumption [6] the corresponding percentage for the here studiedprogramme in Copenhagen, Denmark is close to 16% [7].

* Corresponding author. Tel.: +45 25858078.

E-mail addresses: myeu@sund.ku.dk (M. von Euler-Chelpin),

rqjps480@yahoo.co.jp (M. Kuchiki), Ilse.Vejborg@regionh.dk (I. Vejborg).

http://dx.doi.org/10.1016/j.canep.2014.06.006

1877-7821/� 2014 Elsevier Ltd. All rights reserved.

In a recently published study [8], women with false-positivetest were found to have an increased risk of breast cancer later in

life, compared to women with only negative tests. The increased

risk remained up to 12 years or more after the first false-positive

test, but was somewhat reduced for screening in later time periods,

possibly due to enhanced test quality e.g. improved imaging

techniques, more experienced radiologists, etc. This excess risk of

breast cancer may be due to misclassification of disease status,

meaning that women with an abnormal mammogram were falsely

declared negative, when they should actually having been declared

as having cancer, similar to a study by Peeters et al. [9]. Another

possible explanation could be that these women have some

biological susceptibility for increased risk of breast cancer, such as

benign breast disease [10–17]. What favours the hypothesis of

misclassification is the more than doubled risk at the first screen

following the false-positive test and that the excess risk was higher

in the early technology phase (screened from January 1, 1994 to

December 31, 1998 and followed up to December 31, 2000) than in

the late technology phase (screened from January 1, 2001 to

December 31, 2005 and followed up to December 31, 2007). On the

Table 1Relative risk of breast cancer for women with and without non-misclassified false-

positive screening tests versus women with negative screening tests (invasive and

DCIS).

Cohort Person years

at risk

Breast cancer

Total number

Relative risk,

age adjusted

(95% CI)

Negative test 580,450 1969 1.0 Ref

False positive,

including misclassifications

50,589 295 1.67 (1.45–1.88)

Negative test 580,450 1969 1.0 Ref

False positive,

excluding misclassifications

50,304 223 1.27 (1.11–1.46)

Type 1 FP excl MC 45,282 200 1.27 (1.09–1.46)

Type 2 FP excl MC 5022 23 1.30 (0.86–1.96)

Note: Type 1 FP, established as false-positive after assessment; Type 2 FP,

established as false-positive after surgery; MC, misclassifications.

M. von Euler-Chelpin et al. / Cancer Epidemiology 38 (2014) 619–622620

other hand, excess breast cancer risk for up to 12 years and moreafter the first false-positive test favours the hypothesis ofbiological susceptibility.

To assess the excess risk of breast cancer in women who havehad a false-positive test at screening excluding misclassificationwe re-analysed the data, calculating risk of breast cancer excludingthe women with cancer in the same location as the finding causingthe original recall, as well as studying the role of breast density andcancer morphology in misclassification. To our knowledge this hasnot been done before.

2. Method and materials

We used data from a population-based screening mammogra-phy programme in Copenhagen, Denmark. Screening took placebetween 1991 and 2005. A total of 58,003 women were included inthe analysis out of which 4743 were recalled and subsequentlydeclared negative, i.e. women with a false-positive result. Westudied the 295 women with a false-positive test during thescreening period and who later developed breast cancer (274 withinvasive breast cancer and 21 with ductal carcinoma in situ (DCIS))within the follow-up until April 17, 2008.

Population-based screening mammography started in Copen-hagen, Denmark, in April 1991 and is organised in approximatelybiennial invitation rounds. During our study period all womenaged 50–69 years were, in each invitation round, personallyinvited to screening. Screen-film mammography was usedthroughout the study period, and mammograms were evaluatedindependently by two radiologists. Women with suspiciousfindings were recalled for assessment using the triple testconsisting of clinical examination, mammography, and needlebiopsy. From 1992 onwards, for palpable and/or mammogra-phically uncertain, suspicious, or malignant lesions, mammog-raphy was supplemented with whole breast ultrasoundexamination, as well as ultrasound guided fine needle aspirationcytology and/or histological biopsies. High frequency ultrasounddevices were introduced in 2001, and since 2002 stereotacticbiopsy equipment were used for suspicious micro-calcificationsand impalpable mammographic findings that could not be foundby ultrasound. In the event of inconsistent findings in the tripletest, further investigations were undertaken. If consensus stillcould not be reached, the women were referred to surgicalbiopsy. False-positive tests were defined as Type 1, when the testwas determined negative at assessment and as Type 2 when thetest was determined negative at surgery.

To determine whether there had been a misclassification aradiologist (MK) compared the initial test results that gave thereason for recall with the later cancer location. Cancers thatdeveloped in the same quadrant as the finding that initiallycaused the recall was studied in-depth in order to establishwhether there had been misclassification or not. For each patientthe initial screen was compared with the diagnostic screens andcancer was verified by use of the Danish Pathology Register. Forany uncertain cases a second radiologist (IV) was called in. Ifcertainty could not be reached, the case was not defined asmisclassified.

2.1. Data analysis

Data from the mammography register was linked to data fromthe Cancer Registry and the Danish Pathology Register by means ofthe unique Danish Civil Registration System (CRS) Number. Thestudy included breast cancer (C50), and carcinoma in situ (D05)according to the International Classification of Disease no. 10 (ICD-10). The incidence rate of breast cancer was analysed as a log-linearfunction of attained age (a) and exposure status (s) and expressed

as ln(las) = a + baa + bss, where a is the intercept and b is the slopeof the regression line. Age was divided into 5-year age-groups (50–54, 55–59, 60–64, 65–69, 70–74, 75–79, 80–84, and 85–89 years)and exposure status was divided into false-positive or never false-positive (hereafter called ‘‘negative’’). Person years at risk werecalculated from date of first screen until censoring or end of follow-up. Women contributed person years at risk to the negative groupas long as the screening tests were negative only. Womencontributed person years at risk to the false-positive group fromthe date of the first false-positive test. Women were censored atdeath, breast cancer diagnosis, emigration, or end of follow-up onApril 17, 2008, whichever came first. A full description of themethodology has been reported elsewhere [8].

In the further analysis of the excess risk the cases determined asmisclassified were excluded from the calculations of incidencerate.

Mammographic density was evaluated for the 295 false-positives with cancer by a radiologist (MK) according to standardmethods and in accordance with the Breast Imaging Reporting andData System (BIRADS). In the analysis BIRADS-1 (indicating apredominantly fatty breast) and BIRADS-2 (indicating scatteredfibroglandular densities) were grouped under ‘low mammographicdensity’, while BIRADS-3 (indicating a breast that is heteroge-neously dense) and BIRADS-4, (indicating an extremely densebreast) were grouped under ‘high mammographic density’. For 13cases (4.4%) BIRADS could not be established. The statisticalcalculations were done using SAS version 9.1 (SAS Institute Inc.,Cary, NC).

The Chi-square test was used to compare differences in tumoursize, receptor, and nodal status between the breast cancersdiagnosed in the groups of misclassified and non-misclassified.Data were supplied by the Danish Breast Cancer Cooperative Group(DBCG).

3. Results

A total of 58,003 women were included in the analysis out ofwhich 4743 were recalled. Out of the 295 that later got breastcancer, 72 cases were found to be misclassified, which represents afalse-negative rate of 1.5% (72/4743 recalled women) in womenrecalled for assessment. The excess risk was reduced afterexcluding the misclassified, but there was still a significant excessrisk of breast cancer of 27% (RR = 1.27, 95% CI, 1.11–1.46) amongthe women with a false-positive test compared to women withonly negative tests. Women with a false-positive test determinedat assessment (Type 1) had an excess risk of 27% (RR = 1.27, 95% CI,1.09–1.46), while false-positives determined at surgery (Type 2)had an excess risk of 30% (RR = 1.30, 95% CI, 0.86–1.96), Table 1.

Table 2Relative risk (RR) of misclassification by mammographic density.

Mammographic density Total no. (%) Misclassified (%) Risk of being misclassified RR 95% CI p-value

Low mammographic density 133 (100) 43 (32.3) 1.00 Ref. 0.015

High mammographic density 149 (100) 29 (19.5) 0.60 0.40–0.91

Density data NA 13 (100) 0 – – –

Table 3Invasive breast cancers by screening status, tumour size, receptor, and nodal status; Copenhagen, Denmark, 1991–2008.

Misclassified tests Non- misclassified tests Total p-value

No. % No. % No. %

Total 64 23.4 210 76.6 274 100

Tumour size

�10 mm 20 31.2 50 23.8 70 25.5 0.26

10– 20 mm 25 39.1 90 42.9 115 42.0

>20 15 23.4 55 26.2 70 25.5

Neoadjuvant 0 0.0 1 0.5 1 0.4

Not applicable/missing 4 6.2 14 6.7 18 6.6

ER status

Negative 11 17.2 24 11.4 35 12.8 0.34

Positive 48 75.0 171 81.4 219 79.9

Not applicable/missing 5 7.8 15 7.2 20 7.3

PgR status

Negative 17 26.6 58 27.6 75 27.4 0.75

Positive 23 35.9 85 40.5 108 39.4

Not applicable/missing 24 37.5 67 31.9 91 33.2

Nodal status

Negative 41 64.1 122 58.1 163 59.5 0.52

Positive 20 31.2 70 33.3 90 32.8

Not applicable/missing 3 4.7 18 8.6 21 7.7

HER-2

Negative 10 15.6 53 25.2 63 23.0 0.40

Positive 4 6.3 11 5.2 15 5.5

Not applicable/missing 50 78.1 146 69.5 196 71.6

ER, oestrogen receptor; PgR, progesterone receptor; HER-2, human epidermal growth factor receptor 2.

M. von Euler-Chelpin et al. / Cancer Epidemiology 38 (2014) 619–622 621

The risk of being misclassified followed the same pattern as theexcess risk in being higher in the early technology phase (data noshown).

In the group of women with breasts with high density tissue(BI-RADS density classification 3 or 4) there were a significantlylower proportion of misclassified than for breasts with fatty tissue(BI-RADS density classification 1 or 2), Table 2.

The average time from first false-positive test to time ofdiagnosis was significantly shorter among women with misclassi-fied tests, 4 years versus 7.8 years for the non-misclassified(p < .0001), but the range was nearly equal, 0.5–15.1 years and0.5–16.1 years, respectively. The average age at first recall was 60years for both misclassified and non-misclassified (data notshown). There were no significant differences in tumour size,receptor or nodal status, Table 3.

4. Discussion

Misclassification did not explain all of the excess risk of breastcancer in women with a false-positive result from mammogra-phy screening, though it significantly reduced it. When thewomen with false-positive results defined as misclassified truepositives had been excluded, there still remained an excess riskfor both types of false-positive, those determined at time ofassessment and those at time of surgery. These results suggestthat there might be some underlying biological susceptibilitythat causes some of the excess cancer risk in women with false-positive test.

This study is based on the entire screened population inCopenhagen from 1991 to the end of 2005, but the number of

cases in each group, once stratified on misclassification, is fairlysmall. There is no difference in trend between Type 1 and Type 2false-positives but due to the fact that the women withmisclassified tests had a significantly shorter time between recalland diagnosis than the non-misclassified, the analysis, when themisclassified are excluded, might underestimate the excess cancerrisk. Another limitation was that there was no information on BI-RADS for the entire study population. The decreased risk ofmisclassification for women with high-density breasts might bedue to the use of ultrasound in the assessment. As to the actualcause of misclassification, which undoubtedly is an importantissue, it was beyond the scope of this study.

The cumulative risk over 10 screens for a false-positive resultwhen attending a population-based screening programme inEurope has been estimated to vary approximately between 8 and20% depending on the programme [18]. Earlier studies have shownthat women with false-positive results have an excess risk ofbreast cancer [8,9,19,20]. Few studies have looked at whether theexcess risk can be attributed to misclassifications during work-up.Peeters et al. [9] found a short-term excess risk that was basicallyattributed to misclassification while the long-term risk wasunknown. That is coherent with the results from our study interms of that the excess risk for the non-misclassified is more long-term. Some studies have looked at false-negative results atassessment. Burrell et al. [21] studied the cause of misclassificationin a small study of 28 women, representing around 0.56% of therecalled. False-negative diagnostic assessment was studied byCiatto et al. [22] in a longitudinal study of the Florence Districtscreening programme (1992–2001). They used a 2-year follow-up,and found 57 cases of misclassification, which in this study

M. von Euler-Chelpin et al. / Cancer Epidemiology 38 (2014) 619–622622

represented 0.50% of the recalled. The misclassified at assessmentin our study represents approximately 1.5% of the recalled with anincidence of 0.14% over time at risk, which is coherent with earlierresults considering the low recall rate in our study.

In a newly published study with a follow up period of 17 yearsCastells et al. [23] demonstrate a strong association between falsepositive tests and breast cancer detection, especially amongstwomen where the false-positive test results in cytology or biopsyand in line with our results the cancer detection risk remainedsignificantly higher 4 years or more after the false-positive test. In arecent study, using data from the Breast Cancer SurveillanceConsortium (BCSC) collected from 1996 to 2002 and including2,392,998 screening mammograms, Armstrong et al. [24] found thatincluding pre-test risk prediction using clinical and genomicinformation could improve management of abnormal mammogramsand have a substantial impact on early detection. Improvedmanagement would presumably lead to a decreased risk ofmisclassification. These results must, however, be seen in the contextof the high cumulative risk of false-positive results in the USA.

5. Conclusion

The results indicate that the increased risk is not explained onlyby misclassification, i.e. a false-negative result at assessment. Theexcess risk remains for both Type 1 and Type 2 false-positives,which favours some biological susceptibility. More individualisedscreening is much discussed these days and our results suggestthat it might be worthwhile to further study the increased risk ofwomen with false positive results in this context.

Conflict of interest

None of the authors declare having a conflict of interest.

Authorship contributions

M.V.E. conceived of the study and design, performed thestatistical analysis, participated in the coordination and drafted themanuscript: M.K. and I.V. did the radiological analyses and datacollection, participated in the draft of the manuscript. All authorsread and approved the final manuscript.

Ethical approval

In accordance with Danish legislation, the study was notified tothe Danish Data Inspection Agency (J No 2008-41-2191). Nofurther ethical approval is needed for register based studies inDenmark.

Funding

No external funding was received for this study.

References

[1] GLOBOCAN 2008 database; 2008, http://www-dep.iarc.fr/globocan/databa-se.htm [accessed 26.11.12].

[2] Njor S, Nystrom L, Moss S, Paci E, Broeders M, Segnan N, et al. Breast cancermortality in mammographic screening in Europe: a review of incidence-basedmortality studies. J Med Screen 2012;19(Suppl. 1):33–41.

[3] Broeders M, Moss S, Nystrom L, Njor S, Jonsson H, Paap E, et al. The impact ofmammographic screening on breast cancer mortality in Europe: a review ofobservational studies. J Med Screen 2012;19(Suppl. 1):14–25.

[4] Moss SM, Nystrom L, Jonsson H, Paci E, Lynge E, Njor S, et al. The impact ofmammographic screening on breast cancer mortality in Europe: a review oftrend studies. J Med Screen 2012;19(Suppl. 1):26–32.

[5] Olsen AH, Njor SH, Vejborg I, Schwartz W, Dalgaard P, Jensen MB, et al. Breastcancer mortality in Copenhagen after introduction of mammography screen-ing: cohort study. BMJ 2005. http://dx.doi.org/10.1136/bmj.38313.639236.82.

[6] Hubbard RA, Miglioretti DL, Smith RA. Modelling the cumulative risk of a false-positive screening test. Stat Methods Med Res 2010;5:429–49.

[7] Njor SH, Olsen AH, Schwartz W, Vejborg I, Lynge E. Predicting the risk of a false-positive test for women following a mammography screening programme. JMed Screen 2007;2:94–7.

[8] von Euler-Chelpin M, Risor LM, Thorsted BL, Vejborg I. Risk of breast cancerafter false-positive test results in screening mammography. J Natl Cancer Inst2012;9:682–9.

[9] Peeters PH, Mravunac M, Hendriks JH, Verbeek AL, Holland R, Vooijs PG. Breastcancer risk for women with a false-positive screening test. Br J Cancer1988;2:211–2.

[10] Cote ML, Ruterbusch JJ, Alosh B, Bandyopadhyay S, Kim E, Albashiti B, et al.Benign breast disease and the risk of subsequent breast cancer in AfricanAmerican women. Cancer Prev Res (Phila) 2012;12:1375–80.

[11] Dorjgochoo T, Deming SL, Gao YT, Lu W, Zheng Y, Ruan Z, et al. History ofbenign breast disease and risk of breast cancer among women in China: a case-control study. Cancer Causes Control 2008;8:819–28.

[12] Hartmann LC, Sellers TA, Frost MH, Lingle WL, Degnim AC, Ghosh K, et al. Benignbreast disease and the risk of breast cancer. N Engl J Med 2005;3:229–37.

[13] Hutchinson WB, Thomas DB, Hamlin WB, Roth GJ, Peterson AV, Williams B.Risk of breast cancer in women with benign breast disease. J Natl Cancer Inst1980;1:13–20.

[14] Kabat GC, Jones JG, Olson N, Negassa A, Duggan C, Ginsberg M, et al. A multi-center prospective cohort study of benign breast disease and risk of subse-quent breast cancer. Cancer Causes Control 2010;6:821–8.

[15] Potter JF, Slimbaugh WP, Woodward SC. Can breast carcinoma be anticipated?A follow-up of benign breast biopsies. Ann Surg 1968;6:829–38.

[16] Roberts MM, Jones V, Elton RA, Fortt RW, Williams S, Gravelle IH. Risk of breastcancer in women with history of benign disease of the breast. Br Med J (ClinRes Ed) 1984;6413:275–8.

[17] Samoli E, Trichopoulos D, Lagiou A, Zourna P, Georgila C, Minaki P, et al. Thehormonal profile of benign breast disease. Br J Cancer 2012;1:199–204.

[18] Hofvind S, Ponti A, Patnick J, Ascunce N, Njor S, Broeders M, et al. False-positiveresults in mammographic screening for breast cancer in Europe: a literaturereview and survey of service screening programmes. J Med Screen2012;19(Suppl. 1):57–66.

[19] Castells X, Molins E, Macia F. Cumulative false-positive recall rate and associ-ation with participant related factors in a population based breast cancerscreening programme. J Epidemiol Community Health 2006;4:316–21.

[20] McCann J, Stockton D, Godward S. Impact of false-positive mammography onsubsequent screening attendance and risk of cancer. Breast Cancer Res 2002;5:R11.

[21] Burrell HC, Evans AJ, Wilson AR, Pinder SE. False-negative breast screeningassessment: what lessons can we learn? Clin Radiol 2001;May (5):385–8.

[22] Ciatto S, Houssami N, Ambrogetti D, Bonardi R, Collini G, Del Turco MR.Minority report – false-negative breast assessment in women recalled forsuspicious screening mammography: imaging and pathological features, andassociated delay in diagnosis. Breast Cancer Res Treat 2007;1:37–43.

[23] Castells X, Roman M, Romero A, Blanch J, Zubizarreta R, Ascunce N, et al. Breastcancer detection risk in screening mammography after a false-positive result.Cancer Epidemiol 2013;1:85–90.

[24] Armstrong K, Handorf EA, Chen J, Bristol Demeter MN. Breast cancer riskprediction and mammography biopsy decisions: a model-based study. Am JPrev Med 2013;1:15–22.