- Email: [email protected]
The Visibility of Cancer on Previous Mammograms in Retrospective Review
Clinical Radiology (2001) 56, 40±43 doi:10.1053/crad.2000.0567, available online at http://www.idealibrary.com on
The Visibility of Cancer on Previous Mammograms in Retrospective Review È R IN EN ², J. IS O L A ³ , I. SAARENMAA, T. SALMINEN*, U. GEIGER, P. HEIKKINEN², S. HYVA È RKKA È INEN³³, V . K A T A J A §, M- L . K O K K O * * , R . K O K K O ³ ³ , E . K U M P U L A I N E N § , A . K A J . P A K K A N E N , P . P E L T O N E N § § , A . P I I R O N E N * * , A . S A L O * * , M- L . T A L V I A L A , M. HAKAMA*k Pirkanmaa Cancer Society, HaÈmeenkatu 5A, *University of Tampere School of Public Health, Tampere, ²Radiological Clinic, North Karelia Central Hospital, Joensuu, ³Department of Biomedical Sciences, University of Tampere, Tampere, §Oncological Clinic, Kuopio University Hospital, Kuopio, **Department of Diagnostic Radiology, Tampere City Hospital, ²²Oncological Clinic, TampereUniversity Hospital, Tampere, ³³RoÈntgentutka, Tampere, §§Cancer Society of Finland, Mammography Screening Unit, Joensuu, kFinnish Cancer Registry, Helsinki, Finland Received: 17 March 2000
Accepted: 19 June 2000
AIM: To study how many tumours were visible in restrospect on mammograms originally reported as normal or benign in patients coming to surgery with proven breast cancer. The effect of making the pre-operative mammogram available was also assessed. MATERIALS AND METHODS: Three hundred and twenty initial mammograms of consecutive new breast cancer cases were analysed by a group of radiologists in the knowledge that all patients were later diagnosed with breast cancer. The ®lms were read twice, ®rst without and then with the later (pre-operative) mammograms available. The parenchymal density in the location of the tumour was classi®ed as fatty, mixed or dense, and the tumours were classi®ed as visible or not visible. The reasons for the invisibility of the tumour in the earlier examination were analysed. RESULTS: Fourteen per cent (45) of cancers were retrospectively visible in earlier mammograms without the pre-operative mammograms having been shown, and 29% (95) when pre-operative mammograms were shown. Breast parenchymal density decreased with age and the visibility of tumours increased with age. When considered simultaneously, the effect of age (over 55 vs under 55) was greater (OR 2.9) than the effect of density (fatty vs others) (OR 1.5). The most common reasons for non-detection were that the lesion was overlooked (55%), diagnosed as benign (33%) or was visible only in one projection (26%). Growing density was the most common (37%) feature of those lesions originally overlooked or regarded as benign. CONCLUSIONS: Tumours are commonly visible in retrospect, but few of them exhibit speci®c signs of cancer, and are recognized only if they grow or otherwise change. It is not possible to differentiate most of them from normal parenchymal densities. Saaremaa, I. (2001). Clinical Radiology 56, 40±43. q 2001 The Royal College of Radiologists Key words: mammography, breast neoplasms, retrospective visibility.
INTRODUCTION
Several studies have shown that some breast cancers ± even some palpable ones [1] ± are not visible in a mammogram [1±5], which reduces the effectiveness of screening and clinical mammography. On the other hand, many breast cancers which were originally not detected in an earlier mammogram are visible on retrospective review [6±9]. Finland is a country with a relatively high incidence of and moderately high mortality from breast cancer. In Finland there has been an organized screening programme for breast cancer since 1987 covering women in the age group 50±59 [10]. The health services are Author for correspondence: Irma Saarenmaa, Pirkanmaa Cancer Society, HaÈmeenkatu 5A, 33100 Tampere, Finland. Fax +358 3 24 99 222. 0009-9260/01/0560040+04 $35.00/0
largely publicly funded, and a nationwide cancer registry has operated since 1953. This background enabled us to run a population-based study of breast cancer cases in two Finnish areas representing a high risk and a low risk area. In this study we examined how many cancers were visible on initial mammograms by a retrospective review and what kind of changes originally remained unnoticed. MATERIALS AND METHODS
Oncologists collected the clinical data of consecutive histologically or cytologically veri®ed new breast cancers in the period 1996±1997 in two different areas of Finland. The study areas covered 440 000 and 180 000 inhabitants, respectively. q 2001 The Royal College of Radiologists
41
THE VISIBILITY OF CANCER ON PREVIOUS MAMMOGRAMS IN RETROSPECTIVE REVIEW
Fig. 1 ± Breast parenchymal density in the location of the tumour classi®ed as fatty (left), mixed or dense (right).
Averages of 300 and 100 new breast cancers annually were respectively diagnosed in the study areas. Each area has only one oncological clinic where almost all new breast cancer patients were referred for treatment or follow-up. Our material consisted of 90% of all new breast cancers in the study areas during the study period. In this series we focus on those 320 new breast cancer cases for which both initial and pre-operative mammograms and original patient ®les were available for a review. In all the cases the initial mammograms were originally interpreted as normal or benign. A group of radiologists (minimum 3, maximum 8 who, to ensure neutrality, were not all working in the same unit) interpreted the initial mammograms in two ways: the ®rst reading was made by radiologists who were aware that all the women had later been diagnosed with breast cancer, but who did not know the exact location and
type of the tumour (Reading A). The second reading was made by showing the pre-operative mammograms taken at the time breast cancer was diagnosed, and if mammographically visible, the exact location of the tumour was known (Reading B). The radiologists classi®ed the tumours as visible or not visible in the earlier mammogram. The tumours were classi®ed as not visible if the radiologists' consensus opinion on the earlier mammogram was that it was normal or clearly benign, and visible if it was at least possibly malignant requiring histological veri®cation. The radiological features of missed cancers were analysed and the radiologists made suggestions as to why the tumour was not originally detected. The breast parenchymal density at the location of the tumour was classi®ed into three groups (fatty, mixed or dense) by comparison with model mammograms (Fig. 1). Clinical and histopathological data
Table 1 ± Number and prevalence of retrospectively visible cancers in earlier mammograms originally interpreted as normal or benign without (Reading A) and with (Reading B) preoperative mammograms by age of the patient and density of the breast at the location of the tumour Pattern
Age 36±55 Total
56±76
Visible Reading A n (%)
Total Reading B n (%)
Total
Visible
Total
Reading A n (%)
Reading B n (%)
Visible Reading A n (%)
Reading B n (%)
Fatty Mixed Dense
27 50 37
2 (1) 7 (14) 2 (5)
6 (22) 10 (20) 5 (14)
88 88 30
14 (16) 13 (15) 7 (23)
35 (40) 28 (32) 11 (37)
115 138 67
16 (14) 20 (14) 9 (13)
41 (36) 38 (28) 16 (24)
Total
114
11 (10)
21 (18)
206
34 (12)
74 (36)
320
45 (14)
95 (30)
42
CLINICAL RADIOLOGY
Table 2 ± Crude and adjusted odds ratios with 95% con®dence intervals (CIs) of retrospective visibility of breast cancer in earlier mammograms by age, density and time lag Variable Age 35±55 56±77 Density Mixed/dense Fatty Time lag <2 years $2 years
Crude OR (95% CI)
Adjusted OR* (95% CI)
1 2.48 (1.43±4.31)
1 2.88 (1.60±5.18)
1 1.55 (0.95±2.53)
1 1.48 (0.88±2.50)
1 0.50 (0.31±0.81)
1 0.38 (0.23±0.64)
* Adjusted for all above mentioned variables.
were recorded from the patient ®les of the oncological clinic. Data was analysed in two age groups: under 56, 56 and over. Cross-tabulations were made by SPSS for Windows 95 software and the statistical analysis was based on two-tailed Fisher's exact test using the approximation of Woolf for the differences and chi-squared test for trends and logistic regression. RESULTS
Breast cancer incidence was 72 per 100 000 women years in 1995 in Finland. In the study areas the ®gures were 77.0 and 67.6. In our material there were 320 new breast cancers among women aged 36±76 for whom earlier mammograms were available and breast parenchymal density in location of the tumour could be de®ned. The mean age of the women studied was 59. The density of breast parenchyma at the location of the tumour was strongly dependent on the age of the patient (P < 0.0001). Breast parenchyma density at the location of the tumour decreased with age, being dense in 32% of younger (36±55) women and dense in 15% of older (56±76) women. The retrospective visibility of cancers by age and breast density at the location of the tumour in readings A and B is shown in Table 1. Without pre-operative mammograms (Reading A) the group of radiologists found a suspect lesion in 45 (14%) cases. There were no signi®cant differences in visibility by age or density of the breast. Additionally, there were 10 cases in which a suspect lesion was found at a different location to the subsequent proven cancer, i.e., a false-positive reviewing result. When the exact location of the tumour was known (Reading Table 3 ± Reasons why cancer was not detected originally but was retrospectively visible in reading B Not noticed (overlooked) Diagnosed as benign Visible in only one projection Further examinations misleading Dense breast parenchymal pattern Diagnostic imaging work-up incomplete Low mammography quality Total reasons Total mammograms Reasons per mammogram
55 33 26 13 12 4 3 146 95 1.5
Table 4 ± Mammographic ®ndings in earlier mammograms after reading B in cases in which mammogram was originally interpreted benign or normal
Growing density Architectural distortion Asymmetric density Microcalci®cations Other Typical or probably malignant Total
n
%
35 19 16 13 8 4 95
37 20 17 14 8 4 100
B) 29% (95/320) of the cases were retrospectively visible. The visibility due to the radiologists having seen the pre-operative mammograms improved the visibility from 10 to 18% among younger women and from 12 to 36% among older women. Visibility was highest among women with fatty breasts. The age and density of the breast are correlated. Moreover, the visibility was affected by the time period between the two mammogram examinations, and this should be considered as a confounding factor. We therefore considered the effect of age, density and lag simultaneously. The effect of age (over 55 vs under 55) was OR 2.9 (95% CI 1.6±5.2) and it was stronger than that of density (fatty vs other) (OR 0.7, 95% CI 0.4±1.2) (Table 2). We also assessed why the missed cancers were not originally found (Table 3). For most of the lesions there was more than one reason for the undectectability. Most commonly the lesion had not been noticed (overlooked, 55 cases) and often it had probably seen but falsely interpreted as benign (33 cases). In 26 cases the lesion was visible only in one projection, in 13 cases further examinations, including magni®cation mammography, were later evaluated as misleading, and in 12 cases dense breast parenchymal pattern obscured the lesion. Table 4 shows the mammographic features of the retrospectively visible tumours are shown. Most (77%) of the lesions were either growing densities, architectural distortions or asymmetric densities. Microcalci®cations were also common (14%). Only four patients had changes on the earlier mammogram which were classi®ed as malignant on review. DISCUSSION
The visibility of cancer in a mammogram depends on the quality of the mammogram, the density of the breast, the size and the radiological characteristics of the tumour and the experience and skills of the radiologist [2]. The density of the breast depends on the age of the woman [11]. On retrospective review, with knowledge of the exact location of tumour, 29% of the cases were visible, which is consistent with earlier publications (6±9). The percentage of retrospectively visible cancers increased with age, being 18% among women under 55 and 36% among women over 55. This means that 18% of the younger women (under 55) with previous examinations could theoretically have been diagnosed earlier by mammography. This percentage is much lower than the 35% estimated as false negatives in mammograms by Joensuu et al. [12] for women under 50. Our result may truly be better because most radiologists were aware of the poor results of
THE VISIBILITY OF CANCER ON PREVIOUS MAMMOGRAMS IN RETROSPECTIVE REVIEW
diagnosing young women in the earlier study. We also made extensive use of complementary examinations and may have been more careful when diagnosing young women. Of the retrospectively visible cancers, 41 out of 95 were in fatty breastareas,35beingintheolderagegroup.Birdetal.[3>]foundmost of the missed lesions in women with dense breasts and few in women with fatty breasts, which is contrary to our result. On the other hand, in ourstudywetookintoaccountonlythebreastdensityinthelocationof thetumourandnotinthewholebreast,andwealsoclassi®edasvisible even minimal densities. Due to mixed breast architecture there are in most cases translucent areas in otherwise dense breasts where even slight densities are easily visible. These are analysed as normal parenchyma or as benign lesions until the next mammogram shows a tendency to grow, or some other suspicious features. We estimated that dense breast parenchyma obscured the lesion in only 13%. This is less than expected, since the unfavourable effect of dense breast parenchyma on mammography screening results is well known [13±16]. Kolb et al. [15] have shown that ultrasound (US) can be helpful in detecting cancers in dense breasts, however, US is only performed if there is a reasonable suspicion of cancer, and not as a screening method for all women with dense breasts. It was estimated that in 4% of cases an extra projection or some other procedure could have been helpful. Mammogram quality was estimated to be low in only 3% of cases, which is reasonable considering that our study included mammograms from all mammography units in the study areas. Age and density correlated and so we estimated their separate effect on visibility. Visibility is also affected by time period between mammograms. We therefore accounted for the confounding effect of lag. To the best of our knowledge this has not been done in previous studies. After adjustment for age (over 55 vs under 55) there was a three-fold difference in visibility, whereas the effect of density (fatty vs others) showed only a 1.5 fold difference. Age therefore was more important than density. As the density was estimated by the subjective judgements of the radiologists, it is more liable to misclassi®cation than age. As misclassi®cation results are biased towards the null value, the smaller effect of density compared to age may be partly due to a misclassi®cation error. An analysis of retrospectively visible lesions revealed that overlooking was more common than misinterpretation of a perceived abnormality (55/33). This is contrary to the results of Bird et al. [3]. In our study the most important reason for missing the lesion was its visibility only in one projection (27%) which is more than the 10% reported by Bird et al. [3]. The next important reason was the presence of misleading further examinations, for example, false interpretation of further mammography, such as coned compression views; ultrasonography, and ®ne needle or core biopsies (14%). Among retrospectively visible cancers (Reading B) 77% were growing densities, architectural distortions and asymmetries. All such lesions could also be called asymmetric densities, which Harvey et al. [6] also found to be the most often missed lesions, and which can often be detected only by reviewing or follow-up. Maes et al. [17] also found such non-speci®c minimal signs of cancer common, but in the mammography screening project they were still found to have favourable staging in the next screening round. This is similar to our ®nding based on screening material [9]. Fourteen per cent of the retrospectively visible lesions were microcalci®cations, which is slightly less
43
than the 18% reported by Bird et al. [3]. It is, however, possible that our de®nition of mammographic lesions is not consistent with the de®nition used by earlier study groups. Reading A resulted in 45 retrospectively visible cancers and a further 10 were false-positive cases. Reading B resulted in 95 retrospectively visible cancers. We conclude that retrospectively visible cancers are common, however, most of them can only be detected when the exact location of the tumour is known. Thus, in routine mammographic practice, which involves prospective interpretation, it is impossible to detect most of these lesions, almost all of which demonstrate subtle or non-speci®c features of malignancy.
REFERENCES 1 Coveney EC, Geraghty JG, O'Laoide R, Hourihane JB, O'Higgins NJ. Reasons underlying negative mammography in patients with palpable breast cancer. Clin Radiol 1994;49:123±125. 2 Ma L, Fishell E, Wright B, Hanna W, Allan S, Boyd NF. Case-control study of factors associated with failure to detect breast cancer by mammography. J Natl Cancer Inst 1992;84:781±785. 3 Bird RE, Wallace TW, Yankaskas BC. Analysis of cancers missed at screening mammography. Radiology 1992;184:613±617. 4 Hollingsworth AB, Taylor LDH, Rhodes DC. Establishing a histologic basis for false-negative mammograms. AJR 1993;166:643±647. 5 Coergen SK, Evans J, Cohen GPB, MacMillan JH. Characteristics of breast carcinomas missed by screening radiologists. Radiology 1997; 204:131±135. 6 Harvey JA, Fajardo LL, Innis CA. Previous mammograms in patients with impalpable breast carcinoma: retrospective vs blinded interpretation. AJR 1993;161:1167±1172. 7 van Dijck JAAM, Verbeek ALM, Hendriks JHCL, Holland R. The current detectability of breast cancer in a mammographic screening program. A review of the previous mammograms of interval and screen-detected cancers. Cancer 1993;72:1932±1938. 8 Jones RD, Mclean L, Young JR, Simpson W, Neilson F. Proportion of cancers detected at the ®rst incident screen which were false negative at the prevalent screen. The Breast 1996;5:339±343. 9 Saarenmaa I, Salminen T, Geiger U, et al. The visibility of cancer on earlier mammograms in a population-based screening programme. Eur J Cancer 1999;35:1118±1122. 10 Hakama M, Elovainio L, Kajantie R, Louhivuori K. Breast cancer screening as public health policy in Finland. Br J Cancer 1991;64:962± 964. 11 Stomper PC, D'Souza DJ, DiNitto P, Arrendo MA. Analysis of parenchymal density on mammograms in 1353 women 25±79 years old. AJR 1996;167:1261±1265. 12 Joensuu H, Asola R, Holli K, Kumpulainen E, Nikkanen V, Parvinen L-M. Delayed diagnosis and large size of breast cancer after a false negative mammogram. Eur J Cancer 1994;30A:1299±1302. 13 Sala E, Warren R, McCann J, Duffy S, Day N, Luben R. Mammographic parenchymal patterns and mode of detection: implications for the breast screening programme. J Med Screen 1998;5:207±212. 14 van Gils CH, Otten JDM, Verbeek ALM, Hendriks JHCL, Holland R. Effect of mammographic breast density on breast cancer screening performance: a study in Nijmegen, the Netherlands. J Epidemiol Community Health 1998;52:267±271. 15 Kolb TM, Lichy J, Newhouse JH. Occult cancer in women with dense breasts: detection with screening US ± diagnostic yield and tumour characteristics. Radiology 1998;207:191±199. 16 Simpson W, Neilson F, Young JR and the Northern Region Breast Screening Radiology Audit Group. The identi®cation of false negatives in a population of interval cancers: a method for audit of screening mammography. Breast 1995;4:183±188. 17 Maes RM, Dronkers DJ, Hendrics JHCL, Thijssen MAO, Nab HW. Do non-speci®c minimal signs in biannual mammographic breast cancer screening programme need further diagnostic assessment? BJR 1997; 70:34±38.
The Visibility of Cancer on Previous Mammograms in Retrospective Review È R IN EN ², J. IS O L A ³ , I. SAARENMAA, T. SALMINEN*, U. GEIGER, P. HEIKKINEN², S. HYVA È RKKA È INEN³³, V . K A T A J A §, M- L . K O K K O * * , R . K O K K O ³ ³ , E . K U M P U L A I N E N § , A . K A J . P A K K A N E N , P . P E L T O N E N § § , A . P I I R O N E N * * , A . S A L O * * , M- L . T A L V I A L A , M. HAKAMA*k Pirkanmaa Cancer Society, HaÈmeenkatu 5A, *University of Tampere School of Public Health, Tampere, ²Radiological Clinic, North Karelia Central Hospital, Joensuu, ³Department of Biomedical Sciences, University of Tampere, Tampere, §Oncological Clinic, Kuopio University Hospital, Kuopio, **Department of Diagnostic Radiology, Tampere City Hospital, ²²Oncological Clinic, TampereUniversity Hospital, Tampere, ³³RoÈntgentutka, Tampere, §§Cancer Society of Finland, Mammography Screening Unit, Joensuu, kFinnish Cancer Registry, Helsinki, Finland Received: 17 March 2000
Accepted: 19 June 2000
AIM: To study how many tumours were visible in restrospect on mammograms originally reported as normal or benign in patients coming to surgery with proven breast cancer. The effect of making the pre-operative mammogram available was also assessed. MATERIALS AND METHODS: Three hundred and twenty initial mammograms of consecutive new breast cancer cases were analysed by a group of radiologists in the knowledge that all patients were later diagnosed with breast cancer. The ®lms were read twice, ®rst without and then with the later (pre-operative) mammograms available. The parenchymal density in the location of the tumour was classi®ed as fatty, mixed or dense, and the tumours were classi®ed as visible or not visible. The reasons for the invisibility of the tumour in the earlier examination were analysed. RESULTS: Fourteen per cent (45) of cancers were retrospectively visible in earlier mammograms without the pre-operative mammograms having been shown, and 29% (95) when pre-operative mammograms were shown. Breast parenchymal density decreased with age and the visibility of tumours increased with age. When considered simultaneously, the effect of age (over 55 vs under 55) was greater (OR 2.9) than the effect of density (fatty vs others) (OR 1.5). The most common reasons for non-detection were that the lesion was overlooked (55%), diagnosed as benign (33%) or was visible only in one projection (26%). Growing density was the most common (37%) feature of those lesions originally overlooked or regarded as benign. CONCLUSIONS: Tumours are commonly visible in retrospect, but few of them exhibit speci®c signs of cancer, and are recognized only if they grow or otherwise change. It is not possible to differentiate most of them from normal parenchymal densities. Saaremaa, I. (2001). Clinical Radiology 56, 40±43. q 2001 The Royal College of Radiologists Key words: mammography, breast neoplasms, retrospective visibility.
INTRODUCTION
Several studies have shown that some breast cancers ± even some palpable ones [1] ± are not visible in a mammogram [1±5], which reduces the effectiveness of screening and clinical mammography. On the other hand, many breast cancers which were originally not detected in an earlier mammogram are visible on retrospective review [6±9]. Finland is a country with a relatively high incidence of and moderately high mortality from breast cancer. In Finland there has been an organized screening programme for breast cancer since 1987 covering women in the age group 50±59 [10]. The health services are Author for correspondence: Irma Saarenmaa, Pirkanmaa Cancer Society, HaÈmeenkatu 5A, 33100 Tampere, Finland. Fax +358 3 24 99 222. 0009-9260/01/0560040+04 $35.00/0
largely publicly funded, and a nationwide cancer registry has operated since 1953. This background enabled us to run a population-based study of breast cancer cases in two Finnish areas representing a high risk and a low risk area. In this study we examined how many cancers were visible on initial mammograms by a retrospective review and what kind of changes originally remained unnoticed. MATERIALS AND METHODS
Oncologists collected the clinical data of consecutive histologically or cytologically veri®ed new breast cancers in the period 1996±1997 in two different areas of Finland. The study areas covered 440 000 and 180 000 inhabitants, respectively. q 2001 The Royal College of Radiologists
41
THE VISIBILITY OF CANCER ON PREVIOUS MAMMOGRAMS IN RETROSPECTIVE REVIEW
Fig. 1 ± Breast parenchymal density in the location of the tumour classi®ed as fatty (left), mixed or dense (right).
Averages of 300 and 100 new breast cancers annually were respectively diagnosed in the study areas. Each area has only one oncological clinic where almost all new breast cancer patients were referred for treatment or follow-up. Our material consisted of 90% of all new breast cancers in the study areas during the study period. In this series we focus on those 320 new breast cancer cases for which both initial and pre-operative mammograms and original patient ®les were available for a review. In all the cases the initial mammograms were originally interpreted as normal or benign. A group of radiologists (minimum 3, maximum 8 who, to ensure neutrality, were not all working in the same unit) interpreted the initial mammograms in two ways: the ®rst reading was made by radiologists who were aware that all the women had later been diagnosed with breast cancer, but who did not know the exact location and
type of the tumour (Reading A). The second reading was made by showing the pre-operative mammograms taken at the time breast cancer was diagnosed, and if mammographically visible, the exact location of the tumour was known (Reading B). The radiologists classi®ed the tumours as visible or not visible in the earlier mammogram. The tumours were classi®ed as not visible if the radiologists' consensus opinion on the earlier mammogram was that it was normal or clearly benign, and visible if it was at least possibly malignant requiring histological veri®cation. The radiological features of missed cancers were analysed and the radiologists made suggestions as to why the tumour was not originally detected. The breast parenchymal density at the location of the tumour was classi®ed into three groups (fatty, mixed or dense) by comparison with model mammograms (Fig. 1). Clinical and histopathological data
Table 1 ± Number and prevalence of retrospectively visible cancers in earlier mammograms originally interpreted as normal or benign without (Reading A) and with (Reading B) preoperative mammograms by age of the patient and density of the breast at the location of the tumour Pattern
Age 36±55 Total
56±76
Visible Reading A n (%)
Total Reading B n (%)
Total
Visible
Total
Reading A n (%)
Reading B n (%)
Visible Reading A n (%)
Reading B n (%)
Fatty Mixed Dense
27 50 37
2 (1) 7 (14) 2 (5)
6 (22) 10 (20) 5 (14)
88 88 30
14 (16) 13 (15) 7 (23)
35 (40) 28 (32) 11 (37)
115 138 67
16 (14) 20 (14) 9 (13)
41 (36) 38 (28) 16 (24)
Total
114
11 (10)
21 (18)
206
34 (12)
74 (36)
320
45 (14)
95 (30)
42
CLINICAL RADIOLOGY
Table 2 ± Crude and adjusted odds ratios with 95% con®dence intervals (CIs) of retrospective visibility of breast cancer in earlier mammograms by age, density and time lag Variable Age 35±55 56±77 Density Mixed/dense Fatty Time lag <2 years $2 years
Crude OR (95% CI)
Adjusted OR* (95% CI)
1 2.48 (1.43±4.31)
1 2.88 (1.60±5.18)
1 1.55 (0.95±2.53)
1 1.48 (0.88±2.50)
1 0.50 (0.31±0.81)
1 0.38 (0.23±0.64)
* Adjusted for all above mentioned variables.
were recorded from the patient ®les of the oncological clinic. Data was analysed in two age groups: under 56, 56 and over. Cross-tabulations were made by SPSS for Windows 95 software and the statistical analysis was based on two-tailed Fisher's exact test using the approximation of Woolf for the differences and chi-squared test for trends and logistic regression. RESULTS
Breast cancer incidence was 72 per 100 000 women years in 1995 in Finland. In the study areas the ®gures were 77.0 and 67.6. In our material there were 320 new breast cancers among women aged 36±76 for whom earlier mammograms were available and breast parenchymal density in location of the tumour could be de®ned. The mean age of the women studied was 59. The density of breast parenchyma at the location of the tumour was strongly dependent on the age of the patient (P < 0.0001). Breast parenchyma density at the location of the tumour decreased with age, being dense in 32% of younger (36±55) women and dense in 15% of older (56±76) women. The retrospective visibility of cancers by age and breast density at the location of the tumour in readings A and B is shown in Table 1. Without pre-operative mammograms (Reading A) the group of radiologists found a suspect lesion in 45 (14%) cases. There were no signi®cant differences in visibility by age or density of the breast. Additionally, there were 10 cases in which a suspect lesion was found at a different location to the subsequent proven cancer, i.e., a false-positive reviewing result. When the exact location of the tumour was known (Reading Table 3 ± Reasons why cancer was not detected originally but was retrospectively visible in reading B Not noticed (overlooked) Diagnosed as benign Visible in only one projection Further examinations misleading Dense breast parenchymal pattern Diagnostic imaging work-up incomplete Low mammography quality Total reasons Total mammograms Reasons per mammogram
55 33 26 13 12 4 3 146 95 1.5
Table 4 ± Mammographic ®ndings in earlier mammograms after reading B in cases in which mammogram was originally interpreted benign or normal
Growing density Architectural distortion Asymmetric density Microcalci®cations Other Typical or probably malignant Total
n
%
35 19 16 13 8 4 95
37 20 17 14 8 4 100
B) 29% (95/320) of the cases were retrospectively visible. The visibility due to the radiologists having seen the pre-operative mammograms improved the visibility from 10 to 18% among younger women and from 12 to 36% among older women. Visibility was highest among women with fatty breasts. The age and density of the breast are correlated. Moreover, the visibility was affected by the time period between the two mammogram examinations, and this should be considered as a confounding factor. We therefore considered the effect of age, density and lag simultaneously. The effect of age (over 55 vs under 55) was OR 2.9 (95% CI 1.6±5.2) and it was stronger than that of density (fatty vs other) (OR 0.7, 95% CI 0.4±1.2) (Table 2). We also assessed why the missed cancers were not originally found (Table 3). For most of the lesions there was more than one reason for the undectectability. Most commonly the lesion had not been noticed (overlooked, 55 cases) and often it had probably seen but falsely interpreted as benign (33 cases). In 26 cases the lesion was visible only in one projection, in 13 cases further examinations, including magni®cation mammography, were later evaluated as misleading, and in 12 cases dense breast parenchymal pattern obscured the lesion. Table 4 shows the mammographic features of the retrospectively visible tumours are shown. Most (77%) of the lesions were either growing densities, architectural distortions or asymmetric densities. Microcalci®cations were also common (14%). Only four patients had changes on the earlier mammogram which were classi®ed as malignant on review. DISCUSSION
The visibility of cancer in a mammogram depends on the quality of the mammogram, the density of the breast, the size and the radiological characteristics of the tumour and the experience and skills of the radiologist [2]. The density of the breast depends on the age of the woman [11]. On retrospective review, with knowledge of the exact location of tumour, 29% of the cases were visible, which is consistent with earlier publications (6±9). The percentage of retrospectively visible cancers increased with age, being 18% among women under 55 and 36% among women over 55. This means that 18% of the younger women (under 55) with previous examinations could theoretically have been diagnosed earlier by mammography. This percentage is much lower than the 35% estimated as false negatives in mammograms by Joensuu et al. [12] for women under 50. Our result may truly be better because most radiologists were aware of the poor results of
THE VISIBILITY OF CANCER ON PREVIOUS MAMMOGRAMS IN RETROSPECTIVE REVIEW
diagnosing young women in the earlier study. We also made extensive use of complementary examinations and may have been more careful when diagnosing young women. Of the retrospectively visible cancers, 41 out of 95 were in fatty breastareas,35beingintheolderagegroup.Birdetal.[3>]foundmost of the missed lesions in women with dense breasts and few in women with fatty breasts, which is contrary to our result. On the other hand, in ourstudywetookintoaccountonlythebreastdensityinthelocationof thetumourandnotinthewholebreast,andwealsoclassi®edasvisible even minimal densities. Due to mixed breast architecture there are in most cases translucent areas in otherwise dense breasts where even slight densities are easily visible. These are analysed as normal parenchyma or as benign lesions until the next mammogram shows a tendency to grow, or some other suspicious features. We estimated that dense breast parenchyma obscured the lesion in only 13%. This is less than expected, since the unfavourable effect of dense breast parenchyma on mammography screening results is well known [13±16]. Kolb et al. [15] have shown that ultrasound (US) can be helpful in detecting cancers in dense breasts, however, US is only performed if there is a reasonable suspicion of cancer, and not as a screening method for all women with dense breasts. It was estimated that in 4% of cases an extra projection or some other procedure could have been helpful. Mammogram quality was estimated to be low in only 3% of cases, which is reasonable considering that our study included mammograms from all mammography units in the study areas. Age and density correlated and so we estimated their separate effect on visibility. Visibility is also affected by time period between mammograms. We therefore accounted for the confounding effect of lag. To the best of our knowledge this has not been done in previous studies. After adjustment for age (over 55 vs under 55) there was a three-fold difference in visibility, whereas the effect of density (fatty vs others) showed only a 1.5 fold difference. Age therefore was more important than density. As the density was estimated by the subjective judgements of the radiologists, it is more liable to misclassi®cation than age. As misclassi®cation results are biased towards the null value, the smaller effect of density compared to age may be partly due to a misclassi®cation error. An analysis of retrospectively visible lesions revealed that overlooking was more common than misinterpretation of a perceived abnormality (55/33). This is contrary to the results of Bird et al. [3]. In our study the most important reason for missing the lesion was its visibility only in one projection (27%) which is more than the 10% reported by Bird et al. [3]. The next important reason was the presence of misleading further examinations, for example, false interpretation of further mammography, such as coned compression views; ultrasonography, and ®ne needle or core biopsies (14%). Among retrospectively visible cancers (Reading B) 77% were growing densities, architectural distortions and asymmetries. All such lesions could also be called asymmetric densities, which Harvey et al. [6] also found to be the most often missed lesions, and which can often be detected only by reviewing or follow-up. Maes et al. [17] also found such non-speci®c minimal signs of cancer common, but in the mammography screening project they were still found to have favourable staging in the next screening round. This is similar to our ®nding based on screening material [9]. Fourteen per cent of the retrospectively visible lesions were microcalci®cations, which is slightly less
43
than the 18% reported by Bird et al. [3]. It is, however, possible that our de®nition of mammographic lesions is not consistent with the de®nition used by earlier study groups. Reading A resulted in 45 retrospectively visible cancers and a further 10 were false-positive cases. Reading B resulted in 95 retrospectively visible cancers. We conclude that retrospectively visible cancers are common, however, most of them can only be detected when the exact location of the tumour is known. Thus, in routine mammographic practice, which involves prospective interpretation, it is impossible to detect most of these lesions, almost all of which demonstrate subtle or non-speci®c features of malignancy.
REFERENCES 1 Coveney EC, Geraghty JG, O'Laoide R, Hourihane JB, O'Higgins NJ. Reasons underlying negative mammography in patients with palpable breast cancer. Clin Radiol 1994;49:123±125. 2 Ma L, Fishell E, Wright B, Hanna W, Allan S, Boyd NF. Case-control study of factors associated with failure to detect breast cancer by mammography. J Natl Cancer Inst 1992;84:781±785. 3 Bird RE, Wallace TW, Yankaskas BC. Analysis of cancers missed at screening mammography. Radiology 1992;184:613±617. 4 Hollingsworth AB, Taylor LDH, Rhodes DC. Establishing a histologic basis for false-negative mammograms. AJR 1993;166:643±647. 5 Coergen SK, Evans J, Cohen GPB, MacMillan JH. Characteristics of breast carcinomas missed by screening radiologists. Radiology 1997; 204:131±135. 6 Harvey JA, Fajardo LL, Innis CA. Previous mammograms in patients with impalpable breast carcinoma: retrospective vs blinded interpretation. AJR 1993;161:1167±1172. 7 van Dijck JAAM, Verbeek ALM, Hendriks JHCL, Holland R. The current detectability of breast cancer in a mammographic screening program. A review of the previous mammograms of interval and screen-detected cancers. Cancer 1993;72:1932±1938. 8 Jones RD, Mclean L, Young JR, Simpson W, Neilson F. Proportion of cancers detected at the ®rst incident screen which were false negative at the prevalent screen. The Breast 1996;5:339±343. 9 Saarenmaa I, Salminen T, Geiger U, et al. The visibility of cancer on earlier mammograms in a population-based screening programme. Eur J Cancer 1999;35:1118±1122. 10 Hakama M, Elovainio L, Kajantie R, Louhivuori K. Breast cancer screening as public health policy in Finland. Br J Cancer 1991;64:962± 964. 11 Stomper PC, D'Souza DJ, DiNitto P, Arrendo MA. Analysis of parenchymal density on mammograms in 1353 women 25±79 years old. AJR 1996;167:1261±1265. 12 Joensuu H, Asola R, Holli K, Kumpulainen E, Nikkanen V, Parvinen L-M. Delayed diagnosis and large size of breast cancer after a false negative mammogram. Eur J Cancer 1994;30A:1299±1302. 13 Sala E, Warren R, McCann J, Duffy S, Day N, Luben R. Mammographic parenchymal patterns and mode of detection: implications for the breast screening programme. J Med Screen 1998;5:207±212. 14 van Gils CH, Otten JDM, Verbeek ALM, Hendriks JHCL, Holland R. Effect of mammographic breast density on breast cancer screening performance: a study in Nijmegen, the Netherlands. J Epidemiol Community Health 1998;52:267±271. 15 Kolb TM, Lichy J, Newhouse JH. Occult cancer in women with dense breasts: detection with screening US ± diagnostic yield and tumour characteristics. Radiology 1998;207:191±199. 16 Simpson W, Neilson F, Young JR and the Northern Region Breast Screening Radiology Audit Group. The identi®cation of false negatives in a population of interval cancers: a method for audit of screening mammography. Breast 1995;4:183±188. 17 Maes RM, Dronkers DJ, Hendrics JHCL, Thijssen MAO, Nab HW. Do non-speci®c minimal signs in biannual mammographic breast cancer screening programme need further diagnostic assessment? BJR 1997; 70:34±38.