Clinical Oncology (1990) 2:148--152 © 1990 The Royal College of Radiologists
Clinical Oncology
Original Article How Many Tests Are Required in the Diagnosis of Palpable Breast Abnormalities? J. R. H a r d y ) , T. J. P o w l e s 1, I. J u d s o n 1, C. H e r o n 2, M . W i l l i a m s 2, G . C h e r r y m a n 2, J. H u s b a n d 2, D . C o s g r o v e 3, M . B l a s z c y z y k 3, H . D . S i n n e t t 4, S. E . A s h l e y 5 a n d P. A . T r o t t 6 1Medical Breast Unit, Departments of 2Radiology, 3Nuclear Medicine and Ultrasound, 4Surgery, 5Computing and
6Cytopathology, Royal Marsden Hospital, Downs Road, Sutton, Surrey SM2 5PT, UK
Abstract. Palpable breast nodules in 143 patients attending our primary diagnosis breast clinics were assessed by clinical examination, needle cytology, mammography, ultrasonography and magnetic resonance imaging (MRI). The diagnostic accuracy of all test combinations was compared with the final diagnosis of malignant or benign disease. Two-test combinations increased the sensitivity of diagnosis over that of the individual tests to between 93 % and 100% except for MRI/mammography and MRI/cytology. The combinations of three or more tests increased the sensitivity further, but at the expense of an increased false-positive rate. MRI does not appear to have an important role in the primary diagnosis of breast cancer. Mammography is necessary because of the possibility of occult or multifocal disease. Clinical examination was associated with a high false-positive rate. The combination of cytology and ultrasound was best at correctly diagnosing malignancy, but in this series only 42% of patients underwent ultrasound examination. The role of breast ultrasound together with needle aspiration cytology for the diagnosis of malignancy in palpable breast nodules deserves further evaluation. Keywords: Breast cancer; Diagnosis; Mammography; Ultrasound; Magnetic resonance imaging; Fine needle aspiration cytology
INTRODUCTION The main objective of a primary diagnosis breast clinic is to detect 100% of the breast cancers, whilst Correspondence and offprint requests to: Dr T. J. Powles, Medical Breast Unit, Royal Marsden Hospital, Downs Road, Sutton, Surrey SM2 5PT, UK.
subjecting the minimum number of patients with benign disease to unnecessary surgical biopsy. Several diagnostic techniques are available following initial clinical assessment of a palpable breast nodule. Breast cancers have characteristic clinical features indicating malignancy, such as skin tethering, or deep fixation. Unfortunately these changes are often late, reflecting advanced disease. It is more difficult to distinguish clinically between an early cancer and a benign nodule or lump. Cytological examination of cells aspirated from a breast lump may differentiate frankly malignant or suspicious looking cells from those with benign features (Kline et al., 1979). Needle aspiration is a relatively easy method of obtaining cells for examination and can be used in all palpable lesions (Kline et al., 1979; Duguid et al., 1979; Dixon et al., 1984). A definitive diagnosis has been reported in over 90% of cancers using this method alone (Thomas et al., 1978; Dixon et al., 1984; Hansell et al., 1988) although other studies have failed to achieve this low false-negative rate (Kline et al., 1979; Duguid et al., 1979). Although mammography is well established as a sensitive technique for the diagnosis of breast cancer (Yaffe et al., 1987), 10%-20% of cancers are still missed (CahiU et al., 1981; Moskowitz et al., 1983). Furthermore, the sensitivity of this technique decreases in the examination of dysplastic or younger breasts due to superimposed densities with low contrast (Heywang et al., 1986). Ultrasound examination has the advantage of no radiation exposure and can distinguish solid from cystic masses in dense breasts (Heywang et al., 1986). It is unreliable in distinguishing between benign and malignant solid lesions, however, and cannot reliably detect microcalcification (Heywang et al., 1986). Its major use to date has been as an
Diagnosis of Palpable Breast Abnormalities
adjunct to mammography in a clinically abnormal breast (Hilton et al., 1986). Magnetic resonance imaging (MRI) relies on biochemical rather than physical differences between normal and malignant structures (Chu et al., 1987). MRI optimally displays breast tissue close to the chest wall, eliminates image overlap in dense breasts, provides information about the internal structure of masses but fails to identify microcalcification (Heywang et al., 1986). A limiting factor is that this is an expensive test requiring long scanning times (Dash et al., 1986). Imaging of the breast by MRI has only been possible in the last few years and there is too little evidence to date to support a significant role for this technique in the diagnosis of breast disease at present (Sickles, 1987). Subjecting a patient to multiple diagnostic tests is expensive and time consuming. The aim of this study was to identify those tests that singly or in combination could detect a consistently high proportion of cancers (i.e. high sensitivity) while retaining a low false-positive rate (i.e. high specificity), so that a minimal number of patients with benign disease are referred for surgical biopsy. Women referred to our primary diagnosis breast clinic underwent five diagnostic tests and the sensitivity and specificity of each of these tests alone and in combination with each of the others was analysed against the final diagnosis of malignancy or benign disease as determined from histology following surgery or following clinical resolution of disease.
M A T E R I A L S AND M E T H O D S A total of 143 consecutive patients referred to our primary diagnosis breast clinics between March and December 1987 with palpable breast nodules were included in this study. Their median age was 50 years (range 25-80). All patients were first assessed clinically and the breast lumps graded by a clinician before any other investigations were undertaken as: probably benign (PB), possibly malignant (PsM) or probably malignant (PM). All 143 patients underwent mammography. This was initially done using xeroradiography and subsequently by two-view film mammography using cranio-caudal and lateral oblique veiws with additional views as indicated. Films were independently reviewed by one of four consultant radiologists and reported as probably normal (PN), benign disease (BD), probably benign (PB), probably malignant (PM) or malignant disease (MD). The clinical data were written on the mammography request form. All 143 patients underwent MRI of the breast. Scans were obtained using T1 and T2 weighted sequences with a limited sequence of flash images
149
using a Siemens magneton scanner operating at 1.5 Tesla with the manufacturers breast coil. These were reviewed and reported by a consultant radiologist as probably normal (PN), benign disease (BD), probably benign (PB), probably malignant (PM) or malignant disease (MD). MRI examinations were made without knowledge of other test results. Cells were obtained by fine-needle aspiration of the breast lumps in 116 patients. In the remaining patients no cells suitable for cytological examination were obtained (three patients), or the test was not done for technical reasons, doctor preference or patient refusal (24 patients). All aspiration specimens were assessed by a consultant cytologist and graded on the 1-5 Papanicolaou scale: grades (1) and (2) benign, (3) and (4) suspicious and (5) conclusive evidence of malignancy. The physicians opinion as to the nature of the lesion was stated on the cytology request form. After the start of this study the new Acuson 128 ultrasound scanner became available for breast examination of patients attending the breast clinic and the following 60 patients entered were studied with this modality. Breast ultrasound examinations were done using an Acuson 5MHz linear array (538) using supine direct-contact scanning. Dynamic and static criteria were used to classify all lesions as benign (B), probably benign (PB), equivocal (PsM), probably malignant (PM), or malignant (M). Clinical and mammography data were generally available prior to sonography. For the purpose of this analysis all results were categorized into either benign (including PN, BD, PB) or malignant (which included PsM, PM, MD) groups. Patients were managed as follows: 1. Those in whom the clinical diagnosis of carcinoma was unequivocal underwent metastatic staging prior to surgical excision and definitive primary treatment. 2. Those with any suspicious test required surgical excision biopsy prior to metastatic staging and primary treatment if a diagnosis of malignancy was confirmed. 3. Those with negative tests but a persistent breast lump for more than six weeks also underwent surgical excision. 4. Those patients with negative tests whose breast lumps were shown to be cystic on aspiration or no longer palpable at follow-up visit did not have surgical excision. The final diagnosis of malignancy or benign disease against which the tests were analysed was obtained from histology following surgical excision in all cases except in those patients in whom the
150
J . R . Hardy et al.
lump or nodule resolved without treatment. In these patients the diagnosis was made from clinical follow-up. The median follow-up period of those patients who did not undergo surgery was 16 months (range 1-22 months). The prevalence of disease and the sensitivity and specificity of each test was assessed singly and in combinations of two, three, four and five tests. In the analysis of multiple tests a true-positive result was defined if any of the tests was positive or suspicious and the histology malignant. A falsenegative result was defined when none of the tests was positive or suspicious but the histology was malignant. Similarly a true-negative result was defined as benign histology with no test positive or suspicious and a false-positive result when any test was positive or suspicious with benign histology. The false negative and false positive rates are referred to as (100 - sensitivity) % and (100 specificity) % respectively. The difference in sensitivity and specificity between test combinations was analysed using McNemar's test (Armitage and Berry, 1987). The prevalence of malignancy in any test group was compared using a test for differences between two proportions (Armitage, 1971).
city (71%) while cytology was the most specific test (93%). There was no significant difference in prevalence (the proportion of cancers in the populations of patients receiving each test) for the subsamples undergoing cytology and ultrasound as compared to the population prevalence for the whole sample. When the ten combinations of two tests are considered (Table 2) the sensitivity of correctly predicting malignancy is 100% in three cases, each of which includes ultrasound (ultrasound/cytology, ultrasound/MRI and ultrasound/mammography). The sensitivity is between 93% and 98% for all other combinations except MRI/mammography (89%) and MRI/cytology (88%). The false-positive rate of the two-test combinations increases to 18%42% with clinical/mammography being the least specific test combination (58%) and only ultrasound/cytology and MRI/cytology having a specificity of >80%. Combinations of three tests (Table 3) increases the sensitivity further at the expense of a greater reduction in specificity (falsepositive rates 26%-47%). In five of the ten combinations the sensitivity is 100% with mammography/ MRI/cytology having the lowest sensitivity at 93%. However, only four combinations have a specificity >70%.
RESULTS
Table 2. The accuracy of two-test combinations in predicting malignancy
When each of the tests is analysed independently (Table 1) the sensitivities of the individual tests ranged from 76% to 96% as follows: clinical (92%), cytology (80%), ultrasound (96%), MRI (76%), mammography (86%). Ultrasound had the greatest sensitivity (96%) compared to MRI which was the least sensitive test (76%). Specificity ranged from 71% to 93%: clinical (71%), cytology (93%), ultrasound (84%), MRI (91%), mammography (78%). Clinical examination had the lowest specifiTable 1. The accuracy of individual tests in predicting malignancy
Test
PREV. a
SENS. b
SPEC. c
Clinical Cytology Ultrasound MRI Mammography
66/143 59/116 23/60 66/143 66/143
92 80 96 76 86
71 93d 84 91 78
(%)
(%)
aprevalence (PREV.) = no. of cancers/no, of patients having the test. bSensitivity (SENS.) = true positives/true positives + false negatives. cSpecificity (SPEC.) = true negatives/false positives + true negatives. aThe 7% false positive rate does not include any of those samples reported as Papanicolaou grade 5.
Tests
PREV.
SENS.
Clinical Cytology Clinical Ultrasound Clinical MRI Clinical Mammography Mammography Cytology Ultrasound Cytology Ultrasound MRI Ultrasound Mammography MRI Cytology MRI Mammography
59/116
98
65
23/60
96
65
66/143
95
66
66/143
95
58
59/116
93
75
21/49
100
82
23/60
100
76
23/60
100
76
59/116
88
82
66/143
89
77
(%)
SPEC.
(%)
All combinations of four tests achieved 98%100% sensitivity but with a further increase in the false-positive rate (32%-50%) and when all five tests were combined the sensitivity of 100% is balanced against a false-positive rate of 54% (Table 4).
151
Diagnosis of Palpable Breast Abnormalities Table 3. The accuracy of three-test combinations in predicting malignancy
DISCUSSION
Tests
PREV.
Clinical Mammography Cytology Clinical Ultrasound Cytology Clinical Mammography Ultrasound Clinical MRI Cytology Clinical MRI Ultrasound Clinical Mammography MRI MRI Ultrasound Cytology Mammography MRI Cytology Mammography MRI Ultrasound Cytology Mammography Ultrasound
59/116
98
53
21/49
100
61
23/60
100
57
59/116
98
58
23/60
100
57
66/143
95
57
21/49
100
71
59/116
93
74
23/60
100
73
21/49
100
71
The test information was not collected independently for all examinations in this study. For example, prior knowledge of some other test results was available at the time of ultrasound and mammography. Therefore it is not possible to compare the independent contributions of each test. When excluding a diagnosis of malignancy in the 'real world' however, information from other tests is usually available and the results of multiple tests are considered. It is, therefore, important to consider the contribution of each individual test to any test combination as applied to everyday clinical practice. In this study, MRI was shown to be a relatively specific test having a low false-positive rate. It does not appear to be as sensitive as clinical examination, ultrasound and probably mammography, but this may simply reflect the fact that the examiners had less information available for this test. MRI is an expensive examination in terms of both patients' time and money. It does not appear to add to any combination in that it does not improve the sensitivity or lower the false-positive rate. This study suggests that MRI is unlikely to have a major role in the diagnosis of breast cancer. However, it has shown considerable promise in the diagnosis of metastatic disease (Daffner et al., 1986). Ultrasound was the most sensitive individual test but it may well be relevant that only about half the total number of patients underwent this examination and the results of clinical and mammographic examinations were generally available at the time of scanning. Whenever ultrasound is included in a combination, the sensitivity is 100% (except when combined with the clinical test), because of the high sensitivity of this examination. In this group of patients, clinical assessment did not improve the accuracy of diagnosis and served only to increase the false-positive rate, thus increasing the number of patients having unnecessary surgery. This is presumably because of the 'possibly malignant' category into which a large number of breast lumps were placed at initial assessment. This is inevitable in standard clinical practice, however, where the emphasis must be on never missing a malignancy even if it means subjecting the patient to unnecessary investigations. Although mammography was shown to have a significant false-positive and false-negative rate its use is still indicated in the management of a palpable breast lump because of its ability to detect multifocal disease and occult contralateral primaries. Therefore, although it contributes little to the other tests in the diagnosis of malignancy, it must be included in any combination. Needle aspiration cytology has been shown to be a specific test with a very low false-positive rate.
SENS. (%)
SPEC. (%)
Table 4. The accuracy of four-test combinations and all tests in predicting malignancy
Tests
PREV.
SENS.
SPEC.
Clinical Mammography Ultrasound Cytology Clinical MRI Ultrasound Cytology Clinical Mammography MRI Cytology Clinical Mammography MRI Ultrasound Mammography MRI Ultrasound Cytology Clinical Mammography MRI Ultrasound Cytology
21/49 (43%)
100
50
2t/49 (43%)
100
50
59/116 (51%)
98
51
23/60 (38%)
100
54
21/49 (43%)
100
68
21/49 (43%)
100
46
(%)
(%)
152
Whilst a Papanicolaou grade 5 report always indicated cancer, some of the suspicious cases reported as grade 3 or 4 and included in the malignant category for this analysis were benign, thus accounting for the small number of false-positive reports. This examination can also contribute valuable information as to the nature of the lesion from the consistency of the mass and resistance to the needle (Kline et al., 1979). When considering the combinations, the use of three tests added little to the sensitivity of two-test combinations that is already close to 100% and in many cases led to an increase in the false-positive rate. The non-clinical examinations reduce the false-positive rate without affecting sensitivity (and therefore the diagnosis of malignancy), except for MRI that does not appear to improve either the sensitivity or the false-negative rate. The combination of four tests was no more sensitive in diagnosing malignancy than the threeor two-test combinations and the false-positive rate was higher the greater the number of tests in any combination, especially when the results of clinical assessment were included. The 100% sensitivity at the expense of an increased false-positive rate seen in most cases of >2-test combinations in this study may reflect the fact that all equivocal or suspicious results were considered as positives. An alternative method of analysis utilizing a uniform numerical scoring system for all examinations and relating total score to sensitivity and specificity should be considered in future studies. In summary, MRI does not appear to have an important role in the primary screening of breast cancer. Mammography contributed little when combined with other tests, but must be done because of the possibility of occult, multifocal or bilateral disease. Clinical assessment did not improve the accuracy of diagnosis and served only to increase the false-positive rate. Patients will always be examined in our clinics however. The combination of cytology and ultrasound examination was best at correctly diagnosing malignancy in a palpable breast nodule. This is in accord with several other studies that have shown both tests to have relatively high sensitivity and specificity in the investigation of breast disease both when used alone and in combination with other examinations (Thomas et al., 1978; Dixon et al., 1984; Heywang et al., 1985; van Dam et al., 1988). Only a relatively small number of patients underwent ultrasound examination in this study but the prevalence of cancers in this group was not significantly different from the prevalance of cancer in the whole test population. Future studies are planned to
J.R. Hardy et al.
analyse the sensitivity and specificity of ultrasound performed with no knowledge of other test results in a larger group of patients, to see if this examination can improve the accuracy of diagnosis of malignancy in breast lesions.
References Armitage P (ed.) (1971). Statistical Methods in Medical Research. Blackwell Scientific, Oxford, 129 pp. Armitage P, Berry G (1987). In Statistical Methods in Medical Research, 2nd edn, Blackwell, Oxford, pp. 120-123. Cahill CJ, Boulter PS, Gibbs NM, Price JL (1981). Features of mammographically negative breast tumours. British Journal of Surgery, 68, 882-884. Chu DZ, Yamanashi WS, Frazer J, Hazlewood CF, Gallager HS, Boddie AW et al. (1987) Proton NMR of human breast tumours: correlation with clinical prognostic parameters. Journal of Surgical Oncology, 36, 1-4. Daffner RH, Lupetin, AR, Dash N, Deeb ZL, Sefczek RJ, Schapiro RL (1986). MRI in the detection of malignant infiltration of bone marrow. American Journal of Roentgenology, 146, 353-358. Dash N, Lupetin AR, Daffner RH, Deeb ZL, Sefczek RJ, Schapiro RL (1986). Magnetic resonance imaging in the diagnosis of breast disease. American Journal of Roentgenology, 146, 119-125. Dixon JM, Anderson TJ, Lamb J, Nixon J, Forrest APM (1984). Fine-needle aspiration cytology in relationships to clinical examination and mammography in the diagnosis of a solid breast mass. British Journal of Surgery, 71,593-596. Duguid HL, Wood RAB, Irving AD, Preece PE, Cuschieri A (1979). Needle aspiration of the breast with immediate reporting of material. British Medical Journal, ii, 185-187. Hansell DM, Cooke JC, Parsons CA (1988). The accuracy of mammography alone and in combinationwith clinical examination and cytology in the detection of breast cancer. Clinical Radiology, 39, 150-153. Heywang SH, Dunner PS, Lipsit ER, Glassman LM (1985). Advantages and pitfalls of ultrasound in the diagnosis of breast cancer. Journal of Clinical Ultrasound, 13, 525-532. Heywang SH, Fenzl G, Hahn D,Krischke I, Edmaier M, Eiermann Wet al. (1986). MR imaging of the breast: comparison with mammography and ultrasound. Journal of Computer Assisted Tomography, 10, 615-620. Hilton SVW, Leopold GR, Olson LK, Wilson SA (1986). Realtime breast sonography: application in 300 consecutive patients. American Journal of Roentgenology, 147, 479--486. Kline TS, Joshi LP, Neal HS (1979). Fine-needle aspiration of the breast: diagnosis and pit falls. Cancer, 44, 1458-1464. Moskowitz M, Feig SA, Cole-Beuglet C, Fox SH, Haberman JD, Libshitz HI et al. (1983). Evaluation of new imaging procedures for breast cancer. American Journal of Roentgenology, 140, 591-595. Sickles EA (1987). Computed tomography scanning, transillumination and magnetic resonance imaging of the breast. Recent Results in Cancer Research, 105, 31-35. Thomas JM, Fitzharris BM, Redding WH, Williams JE, Trott PA, Powles TJ et al. (1978). Clinical examination, xeromammography and fine-needle aspiration cytology in diagnosis of breast tumours. British Medical Journal, ii, 1139-1141. Van Dam PA, van Goethem MLA, Kersschot E, Vervliet J, Van den Veyver IBM, De Schepper Aet al. (1988). Palpable solid breast masses: retrospective single- and multimodality evaluation of 201 lesions. Radiology, 166, 435--439. Yaffe MJ, Nishikawa RM, Fenster A (1987). Research in new imaging methods for detection of breast cancer. In Fundamental Problems in Breast Cancer, eds. Paterson AHG, Lees AW, pp. 17-34. Martinus Nijhoff, Boston. Received for publication July 1989 Accepted February 1990
Clinical Oncology
Original Article How Many Tests Are Required in the Diagnosis of Palpable Breast Abnormalities? J. R. H a r d y ) , T. J. P o w l e s 1, I. J u d s o n 1, C. H e r o n 2, M . W i l l i a m s 2, G . C h e r r y m a n 2, J. H u s b a n d 2, D . C o s g r o v e 3, M . B l a s z c y z y k 3, H . D . S i n n e t t 4, S. E . A s h l e y 5 a n d P. A . T r o t t 6 1Medical Breast Unit, Departments of 2Radiology, 3Nuclear Medicine and Ultrasound, 4Surgery, 5Computing and
6Cytopathology, Royal Marsden Hospital, Downs Road, Sutton, Surrey SM2 5PT, UK
Abstract. Palpable breast nodules in 143 patients attending our primary diagnosis breast clinics were assessed by clinical examination, needle cytology, mammography, ultrasonography and magnetic resonance imaging (MRI). The diagnostic accuracy of all test combinations was compared with the final diagnosis of malignant or benign disease. Two-test combinations increased the sensitivity of diagnosis over that of the individual tests to between 93 % and 100% except for MRI/mammography and MRI/cytology. The combinations of three or more tests increased the sensitivity further, but at the expense of an increased false-positive rate. MRI does not appear to have an important role in the primary diagnosis of breast cancer. Mammography is necessary because of the possibility of occult or multifocal disease. Clinical examination was associated with a high false-positive rate. The combination of cytology and ultrasound was best at correctly diagnosing malignancy, but in this series only 42% of patients underwent ultrasound examination. The role of breast ultrasound together with needle aspiration cytology for the diagnosis of malignancy in palpable breast nodules deserves further evaluation. Keywords: Breast cancer; Diagnosis; Mammography; Ultrasound; Magnetic resonance imaging; Fine needle aspiration cytology
INTRODUCTION The main objective of a primary diagnosis breast clinic is to detect 100% of the breast cancers, whilst Correspondence and offprint requests to: Dr T. J. Powles, Medical Breast Unit, Royal Marsden Hospital, Downs Road, Sutton, Surrey SM2 5PT, UK.
subjecting the minimum number of patients with benign disease to unnecessary surgical biopsy. Several diagnostic techniques are available following initial clinical assessment of a palpable breast nodule. Breast cancers have characteristic clinical features indicating malignancy, such as skin tethering, or deep fixation. Unfortunately these changes are often late, reflecting advanced disease. It is more difficult to distinguish clinically between an early cancer and a benign nodule or lump. Cytological examination of cells aspirated from a breast lump may differentiate frankly malignant or suspicious looking cells from those with benign features (Kline et al., 1979). Needle aspiration is a relatively easy method of obtaining cells for examination and can be used in all palpable lesions (Kline et al., 1979; Duguid et al., 1979; Dixon et al., 1984). A definitive diagnosis has been reported in over 90% of cancers using this method alone (Thomas et al., 1978; Dixon et al., 1984; Hansell et al., 1988) although other studies have failed to achieve this low false-negative rate (Kline et al., 1979; Duguid et al., 1979). Although mammography is well established as a sensitive technique for the diagnosis of breast cancer (Yaffe et al., 1987), 10%-20% of cancers are still missed (CahiU et al., 1981; Moskowitz et al., 1983). Furthermore, the sensitivity of this technique decreases in the examination of dysplastic or younger breasts due to superimposed densities with low contrast (Heywang et al., 1986). Ultrasound examination has the advantage of no radiation exposure and can distinguish solid from cystic masses in dense breasts (Heywang et al., 1986). It is unreliable in distinguishing between benign and malignant solid lesions, however, and cannot reliably detect microcalcification (Heywang et al., 1986). Its major use to date has been as an
Diagnosis of Palpable Breast Abnormalities
adjunct to mammography in a clinically abnormal breast (Hilton et al., 1986). Magnetic resonance imaging (MRI) relies on biochemical rather than physical differences between normal and malignant structures (Chu et al., 1987). MRI optimally displays breast tissue close to the chest wall, eliminates image overlap in dense breasts, provides information about the internal structure of masses but fails to identify microcalcification (Heywang et al., 1986). A limiting factor is that this is an expensive test requiring long scanning times (Dash et al., 1986). Imaging of the breast by MRI has only been possible in the last few years and there is too little evidence to date to support a significant role for this technique in the diagnosis of breast disease at present (Sickles, 1987). Subjecting a patient to multiple diagnostic tests is expensive and time consuming. The aim of this study was to identify those tests that singly or in combination could detect a consistently high proportion of cancers (i.e. high sensitivity) while retaining a low false-positive rate (i.e. high specificity), so that a minimal number of patients with benign disease are referred for surgical biopsy. Women referred to our primary diagnosis breast clinic underwent five diagnostic tests and the sensitivity and specificity of each of these tests alone and in combination with each of the others was analysed against the final diagnosis of malignancy or benign disease as determined from histology following surgery or following clinical resolution of disease.
M A T E R I A L S AND M E T H O D S A total of 143 consecutive patients referred to our primary diagnosis breast clinics between March and December 1987 with palpable breast nodules were included in this study. Their median age was 50 years (range 25-80). All patients were first assessed clinically and the breast lumps graded by a clinician before any other investigations were undertaken as: probably benign (PB), possibly malignant (PsM) or probably malignant (PM). All 143 patients underwent mammography. This was initially done using xeroradiography and subsequently by two-view film mammography using cranio-caudal and lateral oblique veiws with additional views as indicated. Films were independently reviewed by one of four consultant radiologists and reported as probably normal (PN), benign disease (BD), probably benign (PB), probably malignant (PM) or malignant disease (MD). The clinical data were written on the mammography request form. All 143 patients underwent MRI of the breast. Scans were obtained using T1 and T2 weighted sequences with a limited sequence of flash images
149
using a Siemens magneton scanner operating at 1.5 Tesla with the manufacturers breast coil. These were reviewed and reported by a consultant radiologist as probably normal (PN), benign disease (BD), probably benign (PB), probably malignant (PM) or malignant disease (MD). MRI examinations were made without knowledge of other test results. Cells were obtained by fine-needle aspiration of the breast lumps in 116 patients. In the remaining patients no cells suitable for cytological examination were obtained (three patients), or the test was not done for technical reasons, doctor preference or patient refusal (24 patients). All aspiration specimens were assessed by a consultant cytologist and graded on the 1-5 Papanicolaou scale: grades (1) and (2) benign, (3) and (4) suspicious and (5) conclusive evidence of malignancy. The physicians opinion as to the nature of the lesion was stated on the cytology request form. After the start of this study the new Acuson 128 ultrasound scanner became available for breast examination of patients attending the breast clinic and the following 60 patients entered were studied with this modality. Breast ultrasound examinations were done using an Acuson 5MHz linear array (538) using supine direct-contact scanning. Dynamic and static criteria were used to classify all lesions as benign (B), probably benign (PB), equivocal (PsM), probably malignant (PM), or malignant (M). Clinical and mammography data were generally available prior to sonography. For the purpose of this analysis all results were categorized into either benign (including PN, BD, PB) or malignant (which included PsM, PM, MD) groups. Patients were managed as follows: 1. Those in whom the clinical diagnosis of carcinoma was unequivocal underwent metastatic staging prior to surgical excision and definitive primary treatment. 2. Those with any suspicious test required surgical excision biopsy prior to metastatic staging and primary treatment if a diagnosis of malignancy was confirmed. 3. Those with negative tests but a persistent breast lump for more than six weeks also underwent surgical excision. 4. Those patients with negative tests whose breast lumps were shown to be cystic on aspiration or no longer palpable at follow-up visit did not have surgical excision. The final diagnosis of malignancy or benign disease against which the tests were analysed was obtained from histology following surgical excision in all cases except in those patients in whom the
150
J . R . Hardy et al.
lump or nodule resolved without treatment. In these patients the diagnosis was made from clinical follow-up. The median follow-up period of those patients who did not undergo surgery was 16 months (range 1-22 months). The prevalence of disease and the sensitivity and specificity of each test was assessed singly and in combinations of two, three, four and five tests. In the analysis of multiple tests a true-positive result was defined if any of the tests was positive or suspicious and the histology malignant. A falsenegative result was defined when none of the tests was positive or suspicious but the histology was malignant. Similarly a true-negative result was defined as benign histology with no test positive or suspicious and a false-positive result when any test was positive or suspicious with benign histology. The false negative and false positive rates are referred to as (100 - sensitivity) % and (100 specificity) % respectively. The difference in sensitivity and specificity between test combinations was analysed using McNemar's test (Armitage and Berry, 1987). The prevalence of malignancy in any test group was compared using a test for differences between two proportions (Armitage, 1971).
city (71%) while cytology was the most specific test (93%). There was no significant difference in prevalence (the proportion of cancers in the populations of patients receiving each test) for the subsamples undergoing cytology and ultrasound as compared to the population prevalence for the whole sample. When the ten combinations of two tests are considered (Table 2) the sensitivity of correctly predicting malignancy is 100% in three cases, each of which includes ultrasound (ultrasound/cytology, ultrasound/MRI and ultrasound/mammography). The sensitivity is between 93% and 98% for all other combinations except MRI/mammography (89%) and MRI/cytology (88%). The false-positive rate of the two-test combinations increases to 18%42% with clinical/mammography being the least specific test combination (58%) and only ultrasound/cytology and MRI/cytology having a specificity of >80%. Combinations of three tests (Table 3) increases the sensitivity further at the expense of a greater reduction in specificity (falsepositive rates 26%-47%). In five of the ten combinations the sensitivity is 100% with mammography/ MRI/cytology having the lowest sensitivity at 93%. However, only four combinations have a specificity >70%.
RESULTS
Table 2. The accuracy of two-test combinations in predicting malignancy
When each of the tests is analysed independently (Table 1) the sensitivities of the individual tests ranged from 76% to 96% as follows: clinical (92%), cytology (80%), ultrasound (96%), MRI (76%), mammography (86%). Ultrasound had the greatest sensitivity (96%) compared to MRI which was the least sensitive test (76%). Specificity ranged from 71% to 93%: clinical (71%), cytology (93%), ultrasound (84%), MRI (91%), mammography (78%). Clinical examination had the lowest specifiTable 1. The accuracy of individual tests in predicting malignancy
Test
PREV. a
SENS. b
SPEC. c
Clinical Cytology Ultrasound MRI Mammography
66/143 59/116 23/60 66/143 66/143
92 80 96 76 86
71 93d 84 91 78
(%)
(%)
aprevalence (PREV.) = no. of cancers/no, of patients having the test. bSensitivity (SENS.) = true positives/true positives + false negatives. cSpecificity (SPEC.) = true negatives/false positives + true negatives. aThe 7% false positive rate does not include any of those samples reported as Papanicolaou grade 5.
Tests
PREV.
SENS.
Clinical Cytology Clinical Ultrasound Clinical MRI Clinical Mammography Mammography Cytology Ultrasound Cytology Ultrasound MRI Ultrasound Mammography MRI Cytology MRI Mammography
59/116
98
65
23/60
96
65
66/143
95
66
66/143
95
58
59/116
93
75
21/49
100
82
23/60
100
76
23/60
100
76
59/116
88
82
66/143
89
77
(%)
SPEC.
(%)
All combinations of four tests achieved 98%100% sensitivity but with a further increase in the false-positive rate (32%-50%) and when all five tests were combined the sensitivity of 100% is balanced against a false-positive rate of 54% (Table 4).
151
Diagnosis of Palpable Breast Abnormalities Table 3. The accuracy of three-test combinations in predicting malignancy
DISCUSSION
Tests
PREV.
Clinical Mammography Cytology Clinical Ultrasound Cytology Clinical Mammography Ultrasound Clinical MRI Cytology Clinical MRI Ultrasound Clinical Mammography MRI MRI Ultrasound Cytology Mammography MRI Cytology Mammography MRI Ultrasound Cytology Mammography Ultrasound
59/116
98
53
21/49
100
61
23/60
100
57
59/116
98
58
23/60
100
57
66/143
95
57
21/49
100
71
59/116
93
74
23/60
100
73
21/49
100
71
The test information was not collected independently for all examinations in this study. For example, prior knowledge of some other test results was available at the time of ultrasound and mammography. Therefore it is not possible to compare the independent contributions of each test. When excluding a diagnosis of malignancy in the 'real world' however, information from other tests is usually available and the results of multiple tests are considered. It is, therefore, important to consider the contribution of each individual test to any test combination as applied to everyday clinical practice. In this study, MRI was shown to be a relatively specific test having a low false-positive rate. It does not appear to be as sensitive as clinical examination, ultrasound and probably mammography, but this may simply reflect the fact that the examiners had less information available for this test. MRI is an expensive examination in terms of both patients' time and money. It does not appear to add to any combination in that it does not improve the sensitivity or lower the false-positive rate. This study suggests that MRI is unlikely to have a major role in the diagnosis of breast cancer. However, it has shown considerable promise in the diagnosis of metastatic disease (Daffner et al., 1986). Ultrasound was the most sensitive individual test but it may well be relevant that only about half the total number of patients underwent this examination and the results of clinical and mammographic examinations were generally available at the time of scanning. Whenever ultrasound is included in a combination, the sensitivity is 100% (except when combined with the clinical test), because of the high sensitivity of this examination. In this group of patients, clinical assessment did not improve the accuracy of diagnosis and served only to increase the false-positive rate, thus increasing the number of patients having unnecessary surgery. This is presumably because of the 'possibly malignant' category into which a large number of breast lumps were placed at initial assessment. This is inevitable in standard clinical practice, however, where the emphasis must be on never missing a malignancy even if it means subjecting the patient to unnecessary investigations. Although mammography was shown to have a significant false-positive and false-negative rate its use is still indicated in the management of a palpable breast lump because of its ability to detect multifocal disease and occult contralateral primaries. Therefore, although it contributes little to the other tests in the diagnosis of malignancy, it must be included in any combination. Needle aspiration cytology has been shown to be a specific test with a very low false-positive rate.
SENS. (%)
SPEC. (%)
Table 4. The accuracy of four-test combinations and all tests in predicting malignancy
Tests
PREV.
SENS.
SPEC.
Clinical Mammography Ultrasound Cytology Clinical MRI Ultrasound Cytology Clinical Mammography MRI Cytology Clinical Mammography MRI Ultrasound Mammography MRI Ultrasound Cytology Clinical Mammography MRI Ultrasound Cytology
21/49 (43%)
100
50
2t/49 (43%)
100
50
59/116 (51%)
98
51
23/60 (38%)
100
54
21/49 (43%)
100
68
21/49 (43%)
100
46
(%)
(%)
152
Whilst a Papanicolaou grade 5 report always indicated cancer, some of the suspicious cases reported as grade 3 or 4 and included in the malignant category for this analysis were benign, thus accounting for the small number of false-positive reports. This examination can also contribute valuable information as to the nature of the lesion from the consistency of the mass and resistance to the needle (Kline et al., 1979). When considering the combinations, the use of three tests added little to the sensitivity of two-test combinations that is already close to 100% and in many cases led to an increase in the false-positive rate. The non-clinical examinations reduce the false-positive rate without affecting sensitivity (and therefore the diagnosis of malignancy), except for MRI that does not appear to improve either the sensitivity or the false-negative rate. The combination of four tests was no more sensitive in diagnosing malignancy than the threeor two-test combinations and the false-positive rate was higher the greater the number of tests in any combination, especially when the results of clinical assessment were included. The 100% sensitivity at the expense of an increased false-positive rate seen in most cases of >2-test combinations in this study may reflect the fact that all equivocal or suspicious results were considered as positives. An alternative method of analysis utilizing a uniform numerical scoring system for all examinations and relating total score to sensitivity and specificity should be considered in future studies. In summary, MRI does not appear to have an important role in the primary screening of breast cancer. Mammography contributed little when combined with other tests, but must be done because of the possibility of occult, multifocal or bilateral disease. Clinical assessment did not improve the accuracy of diagnosis and served only to increase the false-positive rate. Patients will always be examined in our clinics however. The combination of cytology and ultrasound examination was best at correctly diagnosing malignancy in a palpable breast nodule. This is in accord with several other studies that have shown both tests to have relatively high sensitivity and specificity in the investigation of breast disease both when used alone and in combination with other examinations (Thomas et al., 1978; Dixon et al., 1984; Heywang et al., 1985; van Dam et al., 1988). Only a relatively small number of patients underwent ultrasound examination in this study but the prevalence of cancers in this group was not significantly different from the prevalance of cancer in the whole test population. Future studies are planned to
J.R. Hardy et al.
analyse the sensitivity and specificity of ultrasound performed with no knowledge of other test results in a larger group of patients, to see if this examination can improve the accuracy of diagnosis of malignancy in breast lesions.
References Armitage P (ed.) (1971). Statistical Methods in Medical Research. Blackwell Scientific, Oxford, 129 pp. Armitage P, Berry G (1987). In Statistical Methods in Medical Research, 2nd edn, Blackwell, Oxford, pp. 120-123. Cahill CJ, Boulter PS, Gibbs NM, Price JL (1981). Features of mammographically negative breast tumours. British Journal of Surgery, 68, 882-884. Chu DZ, Yamanashi WS, Frazer J, Hazlewood CF, Gallager HS, Boddie AW et al. (1987) Proton NMR of human breast tumours: correlation with clinical prognostic parameters. Journal of Surgical Oncology, 36, 1-4. Daffner RH, Lupetin, AR, Dash N, Deeb ZL, Sefczek RJ, Schapiro RL (1986). MRI in the detection of malignant infiltration of bone marrow. American Journal of Roentgenology, 146, 353-358. Dash N, Lupetin AR, Daffner RH, Deeb ZL, Sefczek RJ, Schapiro RL (1986). Magnetic resonance imaging in the diagnosis of breast disease. American Journal of Roentgenology, 146, 119-125. Dixon JM, Anderson TJ, Lamb J, Nixon J, Forrest APM (1984). Fine-needle aspiration cytology in relationships to clinical examination and mammography in the diagnosis of a solid breast mass. British Journal of Surgery, 71,593-596. Duguid HL, Wood RAB, Irving AD, Preece PE, Cuschieri A (1979). Needle aspiration of the breast with immediate reporting of material. British Medical Journal, ii, 185-187. Hansell DM, Cooke JC, Parsons CA (1988). The accuracy of mammography alone and in combinationwith clinical examination and cytology in the detection of breast cancer. Clinical Radiology, 39, 150-153. Heywang SH, Dunner PS, Lipsit ER, Glassman LM (1985). Advantages and pitfalls of ultrasound in the diagnosis of breast cancer. Journal of Clinical Ultrasound, 13, 525-532. Heywang SH, Fenzl G, Hahn D,Krischke I, Edmaier M, Eiermann Wet al. (1986). MR imaging of the breast: comparison with mammography and ultrasound. Journal of Computer Assisted Tomography, 10, 615-620. Hilton SVW, Leopold GR, Olson LK, Wilson SA (1986). Realtime breast sonography: application in 300 consecutive patients. American Journal of Roentgenology, 147, 479--486. Kline TS, Joshi LP, Neal HS (1979). Fine-needle aspiration of the breast: diagnosis and pit falls. Cancer, 44, 1458-1464. Moskowitz M, Feig SA, Cole-Beuglet C, Fox SH, Haberman JD, Libshitz HI et al. (1983). Evaluation of new imaging procedures for breast cancer. American Journal of Roentgenology, 140, 591-595. Sickles EA (1987). Computed tomography scanning, transillumination and magnetic resonance imaging of the breast. Recent Results in Cancer Research, 105, 31-35. Thomas JM, Fitzharris BM, Redding WH, Williams JE, Trott PA, Powles TJ et al. (1978). Clinical examination, xeromammography and fine-needle aspiration cytology in diagnosis of breast tumours. British Medical Journal, ii, 1139-1141. Van Dam PA, van Goethem MLA, Kersschot E, Vervliet J, Van den Veyver IBM, De Schepper Aet al. (1988). Palpable solid breast masses: retrospective single- and multimodality evaluation of 201 lesions. Radiology, 166, 435--439. Yaffe MJ, Nishikawa RM, Fenster A (1987). Research in new imaging methods for detection of breast cancer. In Fundamental Problems in Breast Cancer, eds. Paterson AHG, Lees AW, pp. 17-34. Martinus Nijhoff, Boston. Received for publication July 1989 Accepted February 1990
