Double reading in mammographic screening

Abstracts

OBJETIVO: Avaliar a eficácia da dupla leitura de mamografias de rotina para rastreamento do câncer de mama, em ambiente privado (não acadêmico). MATERIAIS E MÉTODOS: Durante um ano, 22.024 mamografias de rastreamento, em mulheres assintomáticas, foram submetidas a dupla leitura independente, por dois radiologistas especializados em imaginologia mamária. Foram determinados o número de carcinomas adicionais encontrados com a dupla interpretação e o número absoluto de reconvocações geradas por tal prática. RESULTADOS: Foram detectados nove carcinomas adicionais com a dupla leitura, sendo 55,5% ductal in situ e o restante, ductal invasivo até 1,5 cm. Nosso índice de reconvocação absoluto gerado pela dupla interpretação foi de apenas 1,8%, com achados falso-positivos em 98%. CONCLUSÃO: O uso da dupla leitura no rastreamento mamográfico aumentou o índice de detecção de câncer em 8,5% e todos os tumores encontrados foram estádio 0 ou 1.

Rastreamento mamográfico; Detecção de câncer mamário; Dupla leitura

ORIGINAL ARTICLE

Double reading in mammographic screening^*

Ana Paula Atihé Benveniste^I; Adriana Helena Padovan Grassmann Ferreira^II; Vera Lúcia Nunes Aguillar^III

^ITitular Member of Colégio Brasileiro de Radiologia e Diagnóstico por Imagem, MD Radiologist at the Sector of Breast Imaginology of Laboratório Fleury

^IIPost Graduate at Universidade Federal de São Paulo, MD Radiologist at the Sector of Breast Imaginology of Laboratório Fleury

^IIIDoctor by Universidade Federal de São Paulo, MD Radiologist responsible for the Sector of Breast Imaginology of Laboratório Fleury

^{Mailing address}

ABSTRACT

OBJECTIVE: To evaluate the efficacy of double reading on routine screening mammography for the detection of breast cancer in a non-academic environment.

MATERIALS AND METHODS: During one year, 22,024 screening mammograms of asymptomatic women were submitted to independent double reading by two radiologists specialized in breast imaginology. Data analyzed were the number of additional carcinomas found after a second reading and the absolute number of recalls on account of the second reading.

RESULTS: Nine additional carcinomas were detected after the second reading: 55.5% were ductal carcinoma in situ and the remainder were ductal invasive carcinomas up to 1.5 cm. Our absolute recall rate generated by the double reading was only 1.8%, with a percentage of false-positive findings of 98%.

CONCLUSION: Double reading increased cancer detection rate by 8.5% and all tumors detected were stage 0 or 1.

Keywords: Screening mammography; Breast cancer detection; Double reading.

INTRODUCTION

The objective of the radiologist in mammographic screening is to detect breast cancers smaller than 1.0 cm to 1.5 cm. It is not an easy task due to the complex structure of the breast, the subtle findings of malignancy, the radiologists' weariness or distraction and the low probability of the disease. The rate of breast cancer detection by mammography screening is estimated to be from six to ten cases per thousand studies, if one considers only prevalent tumors found on the first routine mammogram, and from two to four tumors found on subsequent examinations (incidental cancers).

Even with intense training and experience, it is eventually possible for any radiologist not to see a lesion in a study — generally detected by the same radiologist in an earlier reading —, due to the limitations of human perception. In order to decrease the errors of perception in mammography screening, some methods were developed, including double reading: two professionals independently read the mammographic study to increase the detection rate of the disease.

Several publications document the increase in the detection rate of breast cancer with the adoption of the double reading method. Of the 208 cancers found by Bird, in three years of experience with mammographic screening, 11 were detected only by the second observer, which meant an increase of 5.3% in the sensitivity of the study⁽¹⁾ . Tabar et al .reported 15% additional findings with the second reading of mammograms⁽²⁾. Kopans⁽³⁾, Thurfjell et al.⁽⁴⁾. Anderson et al.⁽⁵⁾ and Hulka et al.⁽⁶⁾ published articles on the subject in 1994, and reported similar experiences as the previously mentioned authors. Thurfjell et al.⁽⁴⁾ encountered 15% more cancers after the second reading of 11,343 mammograms, while Kopans⁽³⁾, Hulka et al.⁽⁶⁾ and Anderson et al.⁽⁵⁾ observed an increases in the rate of detection of 7%, 7.7% and 10% respectively. According to Warren and Duffy from Cambridge University, the double reading procedure resulted in the detection of one additional case for every 1,000 screened women (33 tumors among 33,734 women), in relation to the single reading of the mammographic studies⁽⁷⁾. In 1993, one more study was published⁽⁸⁾, affirming that the second reading of mammograms detected up to 15% more cases than the single reading of the studies. Finally, a recent research coordinated by Harvey et al.⁽⁹⁾ demonstrated an increase of 6.3% in cancer detection with the adoption of the double reading method.

However, the cost/benefit factor of double reading has been challenged on account of the extra expenses caused by the procedure (salaries, supplementary images and eventual biopsies requested after the recalls), not covered by the health insurance companies, besides the possible psychological stress caused to patients due to false positive results which invariably increase with the double reading method^(5,10).

The objective of this work is to assess the efficacy of the second reading in routine mammograms for early cancer detection in a private service (non-academic).

MATERIALS AND METHODS

This study was conducted at Laboratório Fleury, a private clinic in the city of São Paulo, Brazil. For one year, (September of 2001 to August of 2002), 22,024 routine mammograms for early breast cancer detection in asymptomatic women were submitted to independent double reading: Two radiologists screened each study at different times. The second observer was not blind to the findings of the first observer. The role of the second radiologist was detecting eventual lesions not identified on the first screening.

The mammograms made for diagnosis on patients with clinical findings were excluded from the study.

The recalls caused by technical error (positioning and inadequate technique) were also excluded as well as the recalls prompted by the first reading. Therefore, only the recalls generated by the second reading of mammograms are considered in the present study.

When the second radiologist disagreed from the first report, the case was submitted to a group review to pursue a consensus (consensus double reading), with the objective of reducing the recall rate.

The mammographic findings which caused patients to be called for supplementary views were classified according to the terminology of the Breast Imaging Reporting and Data System (BI-RADS™) manual, in: micro calcifications, nodules, architecture distortion areas or asymmetries. The patients with mammographic findings considered as category 3 after the second reading, were followed up for at least one year. All patients with studies in which the final category was altered to 4 or 5 were conducted to percutaneous or surgical biopsies.

The following data were analyzed: 1) number of biopsies indicated and number of additional cancers found with the second reading; 2) total number of recalls generated by the second reading; 3) percentage of women recalled for each mammographic finding; 4) total percentage of false positive recalls and percentage of false negative recalls for each radiological finding.

The service has ten mammographs in eight different locations. Of these locations, four (which generate 80% of the total number of mammographic studies) have physicians trained in mammography, all of them with more than 5 year experience, reading the studies with the patients still in house. The mammograms made at the other four locations are sent to a central location for the first reading with the group specialized in breast imaging. At this same central location, the mammographic studies are submitted to the second reading by this group.

The mammograms were made in conventional mammographs, either GE model DMR or Lorad model M IV, using Kodak Min-R2000 films and intensifying screens, which were developed in Kodak Model M-35 and Min-R processors dedicated to mammography.

In order to calculate the rate of cancer detection, we took the statistical data of our 2003 audit as a basis, as we were not audited in 2001. In 2003, 25,288 routine mammograms were made in the institution, of which 395 were followed by biopsies (BI-RADS 4 or 5), resulting in 121 cancer diagnoses. Therefore, we estimated that we should expect the detection of 105 malignant lesions amongst the 22,024 routine mammograms made in the period.

RESULTS

After the independent double reading by two radiologists, out of 22,024 screening mammograms, 389 women were recalled — an absolute recall rate of 1.8% —, of which 79.5% were classified as BI-RADS 1 or 2 only with the necessary supplementary views; 12% were classified as BI-RADS 3 and 8.5% as BI-RADS 4 which were submitted to biopsies. The second reading generated almost 10% more biopsies (34 out of 389) (Table 1).

Of the cases classified as category 3 after the recall, five (all calcification cases) were elevated to category 4 during follow-up due to increase in number of particles or development of pleomorphism. One of these was confirmed as ductal carcinoma in situ after histopathological correlation as shown in Table 2.

Amongst the patients with indicated biopsies, one died due to chronic nephropathy before being submitted to the procedure. Of the remaining 33 cases, 23 benign lesions, one atypical ductal hyperplasia and nine carcinomas were diagnosed in the histopathological studies, with a predictive malignancy value of 27.3% for the biopsies (Table 3).

Of the nine additional cancer cases found with the second reading, five (55.5%) were ductal carcinomas in situ and four (44.5%) were invasive, of which two (50%) were smaller than 1 cm (0.6 cm and 0.7 cm) and the other two were 1.5 cm and 1.3 cm. The mammographic findings in the nine cancers were two asymmetries — one with distortion and the other with clustered micro calcifications —, five micro calcifications cases and two nodules (Table 4).

The total rate of false positive recalls was 97.7% (380 in 389 recalls), considering the atypical ductal hyperplasia case as a false positive recall (Table 5).

DISCUSSION

The double reading method in mammographic screening consists of two radiologists independently interpreting the same study, with the objective of detecting a greater number of initial tumors. Various authors have demonstrated in literature the increase of cancer detection rate when the double reading method is used: Bird⁽¹⁾, 5%; Harvey et al.⁽⁹⁾, 6,3%; Kopans⁽³⁾, 7%; Hulka et al.⁽⁶⁾, 7,7%; Anderson et al.⁽⁵⁾, 10%; Thurfjell et al.⁽⁴⁾, 15%; and Tabar et al.⁽²⁾, 15%.

In the present work, one observed an increase of 8.5% (nine additional tumors) in the detection rate of cancers with the practice of double interpretation of studies — comparable to results published in the literature^(1–10) —, and all additional carcinomas found with such procedure were stage 0 or 1 (55% ductal in situ, and the remainder, ductal invasive up to 1.5 cm).

One should emphasize that the most significant drawback of double reading in mammographic screening is the added cost it may cause, mainly due to the recalls generated by the second reading (which result in supplementary views) not supported by the health insurance companies, and eventual false positive biopsies^(8,9). In the study by Ciatto et al., for example, the double interpretation of mammograms increased the rate of recalls in 15%, causing extra non reimbursable expenses, in spite of the increase of four to six percent in cancer detection⁽¹⁰⁾. Our absolute rate of recalls generated by the second interpretation of studies was only 1.8%, similar to the rate encountered by Harvey et al.⁽⁹⁾, which was 1.5%, and inferior to the rate encountered by Warren and Duffy⁽⁷⁾, which was 3%. Such recall rates must not be confused with the total rate of recalls for patients submitted to screening mammograms, estimated to be from 10 to 15% in the United States of America. Our rate of recalls is limited to patients called back after the second interpretation of their studies (the absolute rate of recalls).

According to Kopans⁽³⁾ and Sickles⁽¹¹⁾, it is fundamental that an efficient double reading system be developed for mammographic screening so that the method becomes more acceptable from a cost/benefit point of view. Some measures are recommended to reduce the number of recalls and consequently the additional cost, such as:

1 – Encourage patients, when they are setting their appointments for mammography, to bring previous studies on the day of the test (preferably the last two studies) for comparative analysis. According to Sickles⁽¹¹⁾, one can expect a decrease of 50% on the recall rate when mammograms are interpreted along with previous studies.

2 – Provide the radiologist with special motorized negatoscopes ("mammoviewer") in which the studies are fed into the equipment by auxiliary personnel so that the radiologist has more available time for the actual interpretation of the study.

3 – Insist on a consensus double reading strategy, in which the studies are independently interpreted by two radiologists, and, if one of them suggests a recall caused by doubt or disagreement, the study is reviewed by a third observer in order to reach a consensus, thus reducing the number of recalls and consequently the extra expenses caused by the adoption of the double reading method, as proposed by Mucci et al.⁽¹²⁾, who reported a reduction of up to 51% in the recall rate by adopting the third radiologist interpretation in cases in which the two first observers could not reach a consensus (consensus double reading). Brown et al.^(13). after comparing three different strategies in a screening program — only single reading, double reading without consensus (in which all disagreement cases were recalled and consensus double reading —, concluded that the third strategy (consensus double interpretation) was as efficient as double reading without consensus, although more cost effective due to the reduced number of recalls.

A 55% majority of our recalls on the second interpretation were due to asymmetries (without associated calcifications or distortions), of which 94% represented tissue superposition or fibro glandular tissue area remaining in additional images. The remainder 6% of asymmetries recalled were classified as BI-RADS 3 and 4 (only one case in category 4), after supplementary study and no case was malignant in the imaging follow-up or histopathological correlation. These data closely match those published by Sickles em 1998⁽¹⁴⁾, whose work showed that 82% of non-calcified asymmetries could be interpreted as superposition of normal breast structures in two conventional mammographic projections (53% of the cases) or after the supplementary views (29% of them).

Amongst the 143 cases of micro calcification that we recalled after the second reading, five were carcinomas and one was an atypical ductal hyperplasia at the anatomopathologic study. The malignancy rate for the calcifications was 3.5%. We found a high number of BI-RADS 3 cases on the recalled calcifications (23%), probably due to an exaggeration in classification: One of these cases was diagnosed as a carcinoma in situ in the imaging follow-up.

Learning to ignore radiological findings with a very low malignancy probability, as reported by Wolverton and Sickles⁽¹⁵⁾, is another way to try to reduce the recall rate. On the work of these authors, subtle mammographic findings highly suggestive of benignancy, for which no further studies were requested. These findings — partially circumscribed low density opacities, asymmetric densities, areas of asymmetric fibro glandular tissue and partially clustered calcifications — were signaled on the films and reevaluated on the next routine mammogram: 74% of the cases were unchanged, 21% of the alterations had diminished or disappeared, 4% were a little more apparent, but considered to be due to technical differences on imaging, and only 1% of the findings (in six patients) had actually grown. Among these six patients who were submitted to other diagnostic studies, one low grade ductal carcinoma in situ was diagnosed. which represented a rate of 0.2% per mammographic finding or 0.3% by patient submitted to mammographic screening, almost identical rates to the incidence of cancer in asymptomatic population. In our study, we also observed that 91.2% (355 in 389) of the recalled mammographic findings corresponded to asymmetries without calcifications or distortion, subtle and partially clustered micro calcifications or partially circumscribed nodules, classified as category 2 or 3 after supplementary views. Of all these cases, only one proved to be malignant in the follow-up. These data confirm that recalls of these cases will not necessarily increase the detection of tumors.

A promising option to double reading in mammographic screening is "Computer Aided Detection" (CAD). This method has been developed to improve the performance of the radiologist working on single reading of mammographic screenings, and not to be the first interpretation of the study, as its role is clearly inferior to the human interpretation for several types of activities, such as signaling focal asymmetries or developing asymmetries, comparing images one breast to the other's, and comparing a new study with a previous one^(9,16,17).

Recently published works have shown that CAD presents a similar efficacy to human double reading, increasing the cancer detection from 7.4% to 21.2%^(18,19), with a better performance in the detection of micro calcifications than in the detection of nodules. On the other hand, CAD increases the interpretation time of the studies, due to the time spent by the radiologist to evaluate the false positive "marks" generated by the system. Even though in more recent versions of CAD the number of false positive marks is generally smaller, three fourths of the studies receive at least one false-positive signing. (Average of two to four marks per four view study), which means that the observing physician has to perform a double reading of his own cases to assess the signs generated by the CAD. Fortunately most of these signs are easily recognizable as false positive, and the most significant drawback in these cases, is the post CAD⁽¹⁵⁾ time spent with human review of its findings. The combination of CAD with digital mammography should reduce the time required for the double interpretation of the studies, as the images are digitally created. However, it increases even further the cost of the equipment.

A recent work by Destounis et al.⁽²⁰⁾ evaluated the role of CAD in routine mammograms, considered normal after double reading, with the purpose of reducing the rate of perception errors. The conclusion was that CAD reduced by one third the false-negative results (from 31% to 19%).

CONCLUSION

In this study, the adoption of double reading of screening mammograms in asymptomatic women increased the rate of cancer detection by 8.5%, without a significant increase in the absolute rate of recalls. All carcinomas detected on the second interpretation were stage 0 (55.5%) or stage 1, which present a lower mortality rate and lower treatment cost.

The consensus double reading can reduce the costs associated with the method as it reduces "unnecessary" recalls. Finally, it is likely that CAD will eventually play the role of the second professional in the second interpretation of mammographic studies.

REFERENCES

Received June 24, 2004.

Accepted after revision February 2, 2006.

• 1. Bird RE. Screening mammography: approach to interpretation and value of double reading. RSNA Categorical Course in Breast Imaging 1995;7376.
• 2. Tabar L, Fagerberg G, Duffy SW, Day NE, Gad A, Grontoft O. Update of the Swedish two-county program of mammographic screening for breast cancer. Radiol Clin North Am 1992;30:187210.
• 3. Kopans DB. Double reading. Radiol Clin North Am 2000;38:719724.
• 4. Thurfjell EL, Lernevall KA, Taube AAS. Benefit of independent double reading in a population-based mammography screening program. Radiology 1994;191:241244.
• 5. Anderson EDC, Muir BB, Walsh JS, Kirkpatrick AE. The efficacy of double reading mammograms in breast screening. Clin Radiol 1994;49:248251.
• 6. Hulka CA, Morse H, McCarthy KA, et al Value of double reading in screening mammography. Radiology 1994;193(Suppl):239.
• 7. Warren RML, Duffy SW. Comparison of single reading with double reading of mammograms and change in effectiveness with experience. Br J Radiol 1995;68:958962.
• 8. Anttinen I, Pamilo M, Soiva M, Roiha M. Double reading of mammography screening films one radiologist or two? Clin Radiol 1993;48:414421.
• 9. Harvey SC, Geller B, Oppenheimer RG, Pinet M, Riddell L, Garra B. Increase in cancer detection and recall rates with independent double interpretation of screening mammography. AJR 2003;180:14611467.
• 10. Ciatto S, Del Turco MR, Monroe D, et al Independent double reading of screening mammograms. J Med Screen 1995;2:99101.
• 11. Sickles EA. Successful methods to reduce false-positive mammography interpretations. Radiol Clin North Am 2000;38:693700.
• 12. Mucci B, Athey G, Scarisbrick G. Double reading of screening mammograms: the use of a third reader to arbitrate on disagreements. Breast 1999;8:6365.
• 13. Brown J, Bryan S, Warren R. Mammographic screening: an incremental cost effectiveness analysis of double versus single reading of mammograms. BMJ 1996;312:809812.
• 14. Sickles EA. Findings at mammographic screening on only one standard projection: outcomes analysis. Radiology 1998;208:471474.
• 15. Wolverton DE, Sickles EA. Clinical outcome of doubtful mammographic findings. AJR 1996;167: 10411045.
• 16. Kopans DB. Accuracy of mammographic interpretation. N Engl J Med 1994;331:15211522.
• 17. Sickles EA. The use of computer-aided detection in mammographic interpretation: detection versus distraction. RSNA Categorical Course in Diagnostic Radiology Physics: Advances in Breast Imaging Physics, Technology and Clinical Applications 2004;219221
• 18. Vyborny CJ, Giger ML, Nishikawa RM. Computer-aided detection and diagnosis of breast cancer. Radiol Clin North Am 2000;38:725740.
• 19. Brem RF, Baum J, Lechner M, et al Improvement in sensitivity of screening mammography with computer-aided detection: a multiinstitutional trial. AJR 2003;181:687693.
• 20. Destounis SV, DiNitto P, Logan-Young W, Bonaccio E, Zuley ML, Willison KM. Can computer-aided detection with double reading of screening mammograms help decrease the false-negative rate? Initial experience. Radiology 2004;232:578584.

*

Study developed at the Sector of Breast Imaginology of Laboratório Fleury, São Paulo, SP, Brazil.

Publication Dates

• Publication in this collection
  25 May 2006
• Date of issue
  Apr 2006

History

• Accepted
  09 Feb 2006
• Received
  24 June 2004