Comparison of Direct Digital Mammography, Computed Radiography, and Film-Screen in the French National Breast Cancer Screening Program

Transcription

1 Women s Imaging Original Research Séradour et al. Comparison of Methods Women s Imaging Original Research Brigitte Séradour 1 Patrice Heid 1 Jacques Estève 2 Séradour B, Heid P, Estève J Keywords: breast cancer, breast cancer screening, breast imaging, digital mammography, film-screen mammography DOI: /AJR Received December 6, 2012; accepted after revision June 6, ARCADES Association pour la Recherche et le Dépistage des Cancers du Sein, du Col de l Utérus et des Cancers Colorectaux. Parc Mure, 16 Blvd des Aciéries, Marseille Cedex 10, France. Address correspondence to B. Séradour (secretariat@arcades-depistages.com). 2 Biostatistique, Hospices civils de Lyon, Lyon, France. AJR 2014; 202: X/14/ American Roentgen Ray Society Comparison of Direct Digital, Computed Radiography, and Film-Screen in the French National Breast Cancer Screening Program OBJECTIVE. The purpose of this article was to compare the performance of digital mammography using hardcopy image reading against film-screen mammography in a French national routine population-based screening program with a decentralized organization. The French context offered the opportunity to examine separately computed radiography and direct digital mammography performances in a large cohort. MATERIALS AND METHODS. The study includes 23,423 direct digital mammography, 73,320 computed radiography, and 65,514 film-screen mammography examinations performed by 123 facilities in Bouches du Rhône, France, for women years old between 2008 and We compared abnormal mammography findings rate, cancer detection rate, and tumor characteristics among the technologies. RESULTS. Abnormal finding rates were higher for direct digital mammography (7.78% vs 6.11% for film-screen mammography and 5.34% for computed radiography), particularly in younger women and in denser breasts. Cancer detection rates were also higher for direct digital mammography (0.71% vs 0.66% for film-screen mammography and 0.55% for computed radiography). The contrast between detection rates was stronger for ductal carcinoma in situ. Breast density was the main factor explaining the differences in detection rates. For direct digital mammography only, the detection rate was clearly higher in dense breasts whatever the age (odds ratio, 2.20). Except for grade, no differences were recorded concerning tumor characteristics in which the proportion of high-grade tumors was larger for direct digital mammography for invasive and in situ tumors. CONCLUSION. Direct digital mammography has a higher detection rate than filmscreen mammography in dense breasts and for tumors of high grade. This latter association warrants further study to measure the impact of technology on efficacy of screening. The data indicate that computed radiography detects fewer tumors than film-screen mammography in most instances. R andomized controlled trials have shown that film-screen mammography based screening reduces breast cancer mortality. For other screening modalities, such as digital mammography, only surrogate endpoints such as cancer detection rates, recall rates, and tumor characteristics (e.g., size and stage) have been used. Direct digital mammography was approved by the U.S. Food and Drug Administration in In France, as in the majority of European countries, the shift to digital mammography has been slow, but the dissemination progressed rapidly after its official incorporation into the national breast cancer screening program in Nationwide, 74% of mammography units were digital in 2010, compared with 20% at the end of How- ever there is a large choice in digital mammography systems and technologies with a variable combination of computed radiography and direct digital mammography. The major obstacles to direct digital mammography include the high cost of the units and archiving facilities. Most of the comparative studies have been carried out with direct digital mammography systems, and few studies have examined computed radiography performance in a large screening mammography cohort [1 3]. Recent results of units using direct digital mammography and computed radiography in organized screening programs in Belgium and in Germany have been published but results are not given separately according to the different digital systems [4, 5]. Moreover, only one study AJR:202, January

2 Séradour et al. in the United Kingdom examined cancer detection rates for direct digital mammography using hardcopy image reading in a routine screening program [6]. Thus, our purpose was to compare the accuracy of digital systems using hardcopy image reading with that of film-screen mammography according to age, breast density, and tumor characteristics at diagnosis, within the context of a French mammography screening program including 70% computed radiography and 30% direct digital mammography for women years old (twoview mammograms obtained once every 2 years) in a decentralized organization. Materials and Methods Between January 1, 2008, and December 31, 2010, in the département of Bouches du Rhône (BdR), France, 128,514 women years old attended the organized screening program and were subjected altogether to 162,257 screening mammography examinations; 96,743 were digital mammography and 65,614 were film-screen mammography. The target population is estimated at 260,000 women (attendance rate = 42%). Women received written invitation from the local screening monitoring center, Association pour la Recherche et le Dépistage des Cancers du Sein, du Col de l Utérus et des Cancers Colorectaux (ARCADES). The invitation contained a list of 123 static mammography facilities that participate in the program. The women could consult the radiologist of their choice. The screening location and technology were mainly influenced by the women s residential area. The BdR program is a part of the national French breast cancer screening program [7]. Individual informed consent to use data of the national breast cancer screening program for research was not required by law. Women were, however, informed that their anonymous data would be used for evaluation of the program unless they explicitly refused. Image Acquisition Digital mammography images were acquired in 2010 using 21 units equipped with direct digital mammography (11 ESSENTIAL, GE Healthcare; 9 SELENIA, Hologic; and one INSPIRA- TION, Siemens Healthcare) and 68 units equipped with computed radiography systems (15 systems with CR MM 3.0 Mammo screens, Agfa; 10 with EHR-M2 screens, Carestream; 36 with HR-BD screens, Fuji; three with RP-7M screens, Konica; and four with HR-BD screens, Philips Healthcare). Thirty-four units used film-screen mammography with different types of x-ray systems using high-contrast and low-speed film-screen combinations. In 2008, only 62 units used digital mammography (13, direct digital mammography; 49, computed radiography). Films were printed directly and immediate interpretation was made on a viewbox when women were present in the unit. Both film-screen mammography and digital mammography systems were subjected to a quality control program according to the European Guidelines for Quality Assurance in Screening [8]. Quality control of hardcopy printer units is also mandatory in the French protocol (according to European Reference Organisation for Quality Assured Breast Screen and Diagnostic Service guidelines and based on the American Association of Physicists in Medicine Task Group 18 recommendations). Image Analysis Volumes of mammography and levels of experience were variable between units, but participating radiologists needed to be qualified (3 days of national training and a minimum of 500 mammography readings per year). The first readers were the local radiologists who performed two-view mammography with clinical examination and read the images in their normal clinical setting. Extra views and immediate assessment (magnification, ultrasound, needle core biopsy) are allowed in the program for positive mammographic findings. Positive or abnormal mammographic findings were defined according to the standard BI-RADS definitions ( need additional imaging evaluation and suspicious ) [9]. Negative or benign mammographic findings after screening examination or after immediate assessment were then submitted with previous mammograms, if available, to the screening monitoring center responsible for program coordination, including a centralized second reading. Second reading was performed by placing the images (film-screen mammography and digital mammography) on an alternator viewbox. Previous mammograms were available to the first and second readers for 74% of the screens (film-screen mammograms were not secondary digitized). Second readers were more specialized in breast imaging, and they read a minimum of 1500 examinations per year. In BdR, 15 radiologists took part in the second reading of film-screen mammography and digital mammography images; 10 had more than 10 years of experience in clinical mammography and also in the screening program using film-screen mammography. All second readers switched to digital mammography in their clinical setting. They are responsible for recalls due to technically inadequate imaging. Women with an abnormal mammogram at second reading were recalled to undergo additional assessment, usually in the first reader s facility. Abnormal mammograms were therefore defined as those identified as abnormal by one or the other readers (first or second). BdR does not have a dedicated cancer registry. To follow-up breast cancer incidence, a pathology register maintained since 1990 collects data of breast biopsies from 10 laboratories of the district and questionnaires are sent by the coordination center to the referring physician regarding the final outcome of assessment and pathology results. Data Collected and Statistical Analysis Data routinely collected at the time of each screening examination included age, screening round, breast density (BI-RADS four category classification) [9], hormonal replacement therapy (HRT) use, screening test outcome categorized as positive or negative using standard BI-RADS definitions (in France the BI-RADS category III needs further evaluation rather than early recall), and final outcome of surgical biopsies of abnormal mammography cases. For all malignancies, data included type (invasive carcinoma or ductal carcinoma in situ [DCIS]), size, histologic grade, and nodal status. Performance indicators including abnormal mammography rate, cancer detection rate, biopsy rate, positive predictive value (PPV), histologic types of cancers, nodal status, and grading were calculated separately for the three groups (film-screen mammography, computed radiography, and direct digital mammography). The abnormal mammography rate was defined as the proportion of women for whom additional imaging assessment was recommended. PPV of abnormal mammography was defined as the proportion of women with an abnormal mammography who were later diagnosed with breast cancer. PPV of biopsy was defined as the proportion of surgical biopsies diagnosed as breast cancer. Among 128,514 women between 50 and 74 years old, 162,257 screening mammography examinations were performed (65,514 screen film and 96,743 digital) in the years These screening tests were cross-tabulated by age, rank of examination, breast density, and HRT use. Associations were evaluated with the Pearson chi-square statistic. To eliminate confounding, we evaluated the association between mammography technology and the variables of interest with logistic or multinomial regression. For this analysis, age was split in two categories (< 60 and 60 years). Likewise density was split in two groups (BI-RADS I and II and BI-RADS III and IV). Multinomial regression was especially useful for evaluating the association between the mammography technology and the grade of invasive tumors. Because the middle grade is often used in the situation of uncertainty, it is relevant to study the factors leading to grade 3 rather than 2 and to grade 1 rather than 2. This is more informative than studying the factors leading to score 3 or 1 using logistic regression analysis. 230 AJR:202, January 2014

3 Comparison of Methods TABLE 1: Characteristics of Women in Study Characteristic Film-Screen Direct Digital Computed Radiography Total Total examinations 65,514 23,423 73, ,257 Age (y) ,410 (44.9) 11,429 (48.8) 33,314 (45.4) 74,153 (45.7) ,984 (39.7) 9080 (38.8) 29,549 (40.3) 64,613 (39.5) ,120 (15.5) 2914 (12.4) 10,457 (14.3) 23,491 (14.5) Rank of examination First 16,019 (24.5) 6758 (28.8) 17,257 (23.5) 40,034 (24.7) Subsequent 49,495 (75.5) 16,665 (71.2) 56,063 (76.5) 122,223 (75.3) BI-RADS breast density 1 11,950 (18.3) 3503 (15.0) 14,712 (20.1) 30,165 (18.6) 2 39,679 (60.9) 15,613 (67.0) 46,541 (63.6) 101,833 (63.0) 3 12,169 (18.7) 3,950 (16.0) 10,774 (14.7) 26,893 (16.6) (2.1) 234 (1.0) 1113 (1.5) 2749 (1.7) Not done HRT Positive 4581 (7.4) 1786 (7.9) 4617 (6.6) 10,984 (7.1) Negative 57,731(92.6) 20,712 (92.1) 65,484 (93.4) 143,927 (92.9) Not done Note All data are number with percentage in parentheses. HRT = hormone replacement therapy. Results Tables 1 6 provide a detailed description of the variables of interest and of possible confounders cross-tabulated with mammography technology Table 1 shows the characteristics of the population. Direct digital mammography subjects were younger, more often in the first rank (in the program), and more often on HRT than computed radiography subjects. Film-screen mammography subjects had denser breasts (21% were BI-RADS III or IV compared with 18% for direct digital mammography and 16% for computed radiography). These observed differences were all highly statistically significant. Performance indicators for digital versus screen film mammography are shown in Table 2. Abnormal Findings Rate The rates of abnormal mammography findings were significantly different among the three groups: 5.34% (3920/73,320) for women screened with computed radiography compared with 7.78% (1823/23,423) with direct digital mammography and 6.11% (4007/65,514) with film-screen mammography (p < 0.001). The recall rate because of poor technical quality was less frequent with digital mammography (0.46% for direct digital mammography vs 0.60% for computed radiography and 0.78% for film-screen mammography). Cancer Detection Rate The combined detection rate, including invasive carcinoma and DCIS, was found to be significantly higher in the direct digital mammography group, with a rate of 0.71% (166/23,423) compared with the computed radiography group, with a rate of 0.55% (404/73,320) and the film-screen mammography group, with a rate of 0.66% (432/65,614) (p = 0.006). A logistic regression analysis was performed to adjust for age, mammography rank, HRT use, and breast density. This multivariate model showed TABLE 2: Performance Indicators Indicator Film-Screen Direct Digital Computed Radiography Total Cancer detection Total 432 (0.66) 166 (0.71) 404 (0.55) 1002 (0.62) Invasive cancer DCIS Cancer unspecified Surgical biopsy 493 (0.75) 208 (0.89) 462 (0.63) 1163 (0.72) Abnormal mammography findings 4007 (6.11) 1823 (7.78) 3920 (5.34) 9750 (6.00) Total examinations 65,514 23,423 73, ,257 PPV screening (%) PPV surgical biopsy (%) Note Data in parentheses are percentages. DCIS = ductal carcinoma in situ, PPV = positive predictive value. AJR:202, January

4 Séradour et al. TABLE 3: Age and Density at Cancer Detection Characteristic Film-Screen Direct Digital Computed Radiography Total Cancer detection by age (y) /29,410 (0.47) 66/11,429 (0.58) 134/33,314 (0.40) 339/74,153 (0.46) /25,984 (0.80) 75/9080 (0.83) 188/29,549 (0.64) 470/64,613 (0.73) /10,120 (0.85) 25/2914 (0.86) 82/10,457 (0.78) 193/23,491 (0.82) Cancer detection by BI-RADS breast density Nondense BI-RADS I and II 337/51,629 (0.65) 113/19,116 (0.59) 322/61,253 (0.53) 772/131,998 (0.58) Dense BI-RADS III and IV 92/13,571 (0.68) 50/4184 (1.20) 77/11,887 (0.65) 219/29,642 (0.74) Not done Note Data are number/total with percentage in parentheses. that after this adjustment only computed radiography had a significantly lower detection rate (odds ratio [OR], 0.77; 95% CI, ), whereas film-screen mammography and direct digital mammography had similar performance. After separating invasive and in situ cancers, the direct digital mammography detection rate of invasive cancers was no longer significantly different from that of computed radiography (0.84; ). Contrarily, the computed radiography detection rate of in situ cancers was significantly lower than that of direct digital mammography (0.49; ) and the film-screen mammography detection rate was lower than that of direct digital mammography but not significantly so (0.71; ). Therefore, the difference in the overall detection rate was mainly due to in situ cancers. Biopsy Rate The percentage of surgical biopsies (fineneedle aspiration and core biopsy are not included) performed was larger in the direct digital mammography group (0.89%), compared with the film-screen mammography group (0.75%) and the computed radiography group (0.63%) (p < 0.001). Positive Predictive Value of Abnormal No statistically significant differences were evidenced among the film-screen mammography (10.8%), direct digital mammography (9.1%), and computed radiography (10.3%) groups. Positive Predictive Value of Biopsy The percentage of surgical biopsies resulting in a diagnosis of breast cancer was lower in the direct digital mammography group (79.8%) than in the other groups (computed radiography, 87.5%; film-screen mammography, 87.6%); and this difference was statistically significant (p = 0.014). The age and density results are shown in Tables 3 and 4. The direct digital mammography detection rate was higher in younger women (0.58% vs 0.40% in computed radiography and 0.47% in film-screen mammography) and in denser breasts (1.20% vs 0.65% in computed radiography and 0.68% in filmscreen mammography). On the other hand, the film-screen mammography detection rate was significantly higher in nondense breasts (0.65% vs 0.59% for direct digital mammography and 0.53% for computed radiography). It may be that the age effect is an indirect effect of density. A logistic regression analysis was then performed to measure the effect of density separately in younger women and in older women; the detection rate was always higher in older women after adjusting for density (OR, 1.60, 1.78, and 1.68 for direct digital mammography, computed radiography, and film-screen mammography, respectively; all were highly significant). When screening with direct digital mammography, the frequency of cancer detection in dense breasts was twice that of nondense breasts. The ORs were 2.15 and 2.25 for younger and older women, respectively, both highly significant. These same ORs were 1.32 and 1.39 for computed radiography and 0.99 and 1.24 for film-screen mammography and were not significant except for computed radiography in older women, which had significance of Therefore, when density is taken into account, the detection rate is, as expected, always lower in younger women, and with direct digital mammography the detection rate is higher in denser breasts, as expected. The other methods of screening are unable, or not particularly able in the case of computed radiography, to detect the higher risk in denser breasts. TABLE 4: Age and Density at Abnormal Findings Film-Screen Direct Digital Computed Characteristic Radiography Total Abnormal mammography findings by age (y) /29,410 (6.91) 986/11,429 (8.63) 1932/33,314 (5.80) 4951/74,153 (6.68) /25,984 (5.60) 642/9080 (7.07) 1493/29,549 (5.05) 3591/64,613 (5.56) /10,120 (5.12) 195/2914 (6.69) 495/10,457 (4.73) 1208/23,491 (5.14) Abnormal mammography findings by BI-RADS breast density Nondense BI-RADS I and II 2852/51,629 (5.52) 1249/19,116 (6.53) 3006/61,253 (4.91) 7107/131,998 (5.38) Dense BI-RADS II and IV 1020/13,571 (7.52) 467/4184 (11.16) 774/11,887 (6.51) 2261/29,642 (7.63) Not done Note Data are number/total with percentage in parentheses. 232 AJR:202, January 2014

5 Comparison of Methods TABLE 5: Tumor Characteristics of Invasive Cancers Characteristic Film-Screen Direct Digital Computed Radiography Total Tumor size Known mm 112 (37.2) 48 (45.3) 94 (33.6) 254 (37.0) mm 120 (39.9) 39 (36.8) 126 (45.0) 285 (41.5) > 20 mm 69 (22.9) 19 (17.9) 60 (21.4) 148 (21.5) Missing Nodal Status Known Positive 72 (23.3) 30 (27.0) 65 (22.2) 167 (23.4) Negative 237 (76.7) 81 (73.0) 228 (77.8) 546 (76.6) Missing Tumor grade Known (40.9) 37 (33.0) 105 (35.1) 267 (37.2) (45.4) 48 (42.9) 131 (43.8) 318 (44.4) 3 42 (13.7) 27 (24.1) 63 (21.1) 132 (18.4) Missing Note Data in parentheses are percentages. The results were similar if we looked separately at invasive and in situ cancers. However the significance was lost for the detection of in situ cancers because of small numbers. The advantage of direct digital mammography in denser breasts was not explained by a better detection of in situ cancers. By contrast, the frequency of abnormal mammography results was lower in older women whatever the technology (OR, 0.81), and this frequency was always higher in denser breasts whatever the age of the patient or the technology. However, this significant effect of density was higher for direct digital mammography than for the other technologies (OR, 1.75 for direct digital mammography vs 1.34 for other technologies; both significant). Note that the frequencies of abnormal mammography results in younger women with nondense breasts were 7.08%, 5.28%, and 6.17% when screening with direct digital mammography, computed radiography, and film-screen mammography, respectively. Tumor characteristics are shown in Tables 5 and 6. The proportion of small invasive tumors was quite similar in the three groups (direct digital mammography, computed radiography, and film-screen mammography) without significant differences. Neither did we find any significant difference in the proportions of negative lymph nodes in the direct digital mammography group (73.0%) compared with the computed radiography group (77.8%) and the film-screen mammography group (76.7%). Only the proportion of invasive tumors classified grade 3 was higher in the direct digital mammography group (24.1%) compared with 21.1% in the computed radiography group and 13.7% in the filmscreen mammography group. To further study this latter difference of borderline significance (p = 0.064), we carried out a multinomial regression analysis to determine which factors influenced the choice of grade 1 rather than 2 and the choice of grade 3 rather than 2. Among the available variables, only small size influenced the choice of grade 1 rather than 2, even if film-screen mammography screening had a slight nonsignificant effect on this choice. For the choice of grade 3 rather than 2, the significant factors were node involvement (1.64; ) for digital mammography, and the relative risks were 1.75; and 2.09; for computed radiography and direct digital mammography, respectively. Therefore, the choice of grade 3 rather than 2 is significantly more frequent in tumors detected by digital mammography than in tumors detected by film-screen mammography. Among the 139 DCIS available, the tumors of high grade or intermediate grade (grade 2 or 3) were also more frequent in direct digital mammography, and this difference was significant (p = 0.025). Discussion With the advent of digital screening, there is a need for evaluation of the new technologies to make sure that their performances are at least as efficient as the film-based methods for which randomized studies have shown the ability to reduce mortality. Our study showed that, judged on detection rate, direct digital mammography had better performance than either film-screen mammography or computed radiography and will probably bring at least the same improvement in mortality as film-screen mammography. In particular, it is the only technology that detects correctly more tumors in dense breasts than in nondense breasts for all ages. Despite a high recall rate in younger women with dense breasts, film-screen mammography did not detect more tumors in this group than in women with nondense breasts of the same age. In addition, because direct digital mammography detects more grade 3 tumors for both in situ and invasive tumors, it is unlikely that it produces more overdiagnosis than film-screen mammography. Since 2008, many screening facilities participating in the French decentralized breast cancer screening program have switched to digital mammography, mostly to computed radiography and a few to direct digital mammography. We have been able to set up a comparative study in BdR. This study included TABLE 6: Tumor Characteristics of Ductal Carcinoma In Situ Characteristic Film-Screen Direct Digital Computed Radiography Total Tumor grade Known (33.9) 3 (8.8) 11 (23.9) 34 (24.5) 2 or 3 39 (66.1) 31 (91.2) 35 (76.1) 105 (75.5) Missing Note Data in parentheses are percentages. AJR:202, January

6 Séradour et al. three large concurrent cohorts of participants who were screened with one of the available methods, either film (film-screen mammography) or digital mammography (direct digital mammography or computed radiography), within a single screening program that contains centralized data management and evaluation bases in the town of Marseille. The BdR program has 123 mammographic facilities and uses hardcopy image reading. In our study, screening performances are reported for a decentralized program in the context of daily practice. To date, most studies have been performed in a centralized screening organization [10 14]. Computed radiography systems are used in some countries [15], but few results have been published in large cohorts [1, 2]. Our results differ to some extent from the few studies that explore digital mammography performance in the context of routine screening practice. The largest study carried out in the United States [16] showed somewhat discrepant results associated with a different age range (40 79) and screening strategy. Nevertheless, all studies recognized the role of age, breast density, and hormones [17, 18], but few have been able to disentangle their specific contributions through precise statistical adjustment. The high number of facilities in the BdR explains the economic choice to implement first digital screening with hardcopy reading. To our knowledge, only one other study from 2009 in the United Kingdom compared direct digital mammography with film-screen mammography within the context of a routine screening program by using hardcopy image reading [6]. Those authors found similar cancer detection rates for film-screen mammography and direct digital mammography (0.72% vs 0.68%) and similar recall rates. Pisano et al. [3] also used hardcopy images in the Digital Mammographic Imaging Screening Trial (DMIST), but hardcopy performance results were not analyzed separately. Our study showed that the difference between the performance of direct digital mammography and that of other technologies is clearly seen in the context of hardcopy reading when strict control of the quality of the laser printing is implemented. Two studies in Finland and Sweden [1, 2] evaluated computed radiography performance in population-based screening programs, using computed radiography softcopy reading, and film-screen mammography. The Finnish study of Lipasti et al. [1] was carried out using a unique computed radiography system and film-screen mammography in two cohorts of women years old. Those authors found that computed radiography detected more cancers than film-screen mammography (0.62% vs 0.40%) and more women were recalled for calcifications with the digital technology. The recall rates were very low (1.71 for computed radiography vs 1.59 for film-screen mammography). This differs with our results, obtained with a higher rate of abnormal mammographic findings and hardcopy reading, which show similar detection rates for computed radiography and film-screen mammography in women years old (0.40 for computed radiography vs 0.47 for film-screen mammography) but better detection rates for film-screen mammography than for computed radiography in nondense breasts (0.47 vs 0.36). The Swedish study of Heddson et al. [2] evaluated photon-counting direct digital mammography performance against computed radiography and film-screen mammography performance in three noncontemporaneous cohorts of women years old. The recall rate was around 1% and cancer detection rates were 0.31 for film-screen mammography, 0.38 for computed radiography, and 0.49 for direct digital mammography, but the subjects screened with film were a historical control group recruited a long time before the subjects were screened with digital technology, therefore it is difficult to compare these results with ours and with the Finnish results. In the Belgian study [4] and in the DMIST trial [3], computed radiography systems were used concurrently with direct digital mammography. In the former, a unique computed radiography system and two direct digital mammography systems were compared with film-screen mammography in a population-based screening program, in the latter several digital systems, including one computed radiography and several direct digital mammography systems, were compared with film-screen mammography in a trial comparing the technologies for the same women in several centers. The digital screening performances were compared with those of film, but computed radiography and direct digital mammography were not analyzed separately. Pisano et al. [19] compared the performance of computed radiography and direct digital mammography with that of film-screen mammography, but no direct comparison was made between the two types of digital mammography. Overall there are few data comparing computed radiography and direct digital mammography. Thus, it is difficult to compare our results on computed radiography systems with those of other studies. When the results for invasive tumors and DCIS were taken separately, direct digital mammography showed a significantly higher detection rate only for DCIS but not for invasive cancer. Our findings are partially comparable with three other studies [12 14], but the percentage of DCIS in our study was lower (24% vs 27 33%). After a review of our data, we found that direct digital mammography more often detected cancers that presented with microcalcifications: 20% for filmscreen mammography versus 37.5 for direct digital mammography and 22.8% for computed radiography. Other researchers have reported similar results [12, 20, 21]. Those findings also confirm the results of phantom studies and experimental studies comparing microcalcification detection with computed radiography and direct digital mammography [22]. The significant reduction in DCIS detection with computed radiography (0.07% vs 0.14% for direct digital mammography) in our study is comparable to the reduction of detection of subtle clusters of microcalcifications in experimental images [23]. Abnormal mammography rates and cancer detection rates were significantly higher with direct digital mammography and lower with computed radiography compared with filmscreen mammography. Younger women and women with denser breasts have higher rates of abnormal mammography findings, whatever the technology. The percentage of women needing additional assessment increased 1.6% in the direct digital mammography group compared with film-screen mammography without impact on PPV (Table 2). Those results contradict the recent conclusions of Sala et al. in Spain [24], in which the recall rate was lower with direct digital mammography. Other studies in different European countries have shown divergent results concerning the recall rate and PPV of abnormal screening with direct digital mammography [25]. The impact of digital screening on the abnormal mammography rate is variable according to the baseline recall rate. In our study, the abnormal mammography rate with film-screen mammography (6.6%) was higher than the recommendations laid down in the European guidelines [9], but the French screening program recalls women only if second reading is abnormal [26] and this true recall rate is low in this context (abnormal rate in the second reading, 1.9%). A smaller effect of direct digital mammography on the abnormal mammography and cancer detection rates is expected after the first digital screening, which may detect prevalent tumors not seen previously by the other technologies [27]. Our results confirm the findings reported by the DMIST direct digital mammography de- 234 AJR:202, January 2014

7 Comparison of Methods tects more cancers among younger women and women with dense breasts when looking separately at these factors [28]. We have shown that the age effect is actually an indirect effect of density. The effect of density on detection by direct digital mammography is similar if invasive cancers and DCIS are analyzed separately. The increased rate of cancers detected through clustered microcalcifications with direct digital mammography and with increased contrast resolution is beneficial for invasive cancers, not just for DCIS [14]. The apparent contradiction with the unadjusted detection rate is explained by a slightly higher detection rate of film-screen mammography in nondense breasts and by the relatively low proportion of dense breasts (18% in our study). It is difficult to compare our results with those of other studies in which the cutoff point defining dense breasts may be different despite the use of the BI-RADS scale [29]. Our results showed that computed radiography had a lower detection rate than direct digital mammography in dense breasts and a lower detection rate than film-screen mammography in fatty breasts. We have noted in our detailed data that the detection of microcalcifications by computed radiography was lower than by direct digital mammography and that the detection of masses with computed radiography was lower than with filmscreen mammography. Among detected cancers with a specified diagnosis, direct digital mammography detected 24% of DCIS (37/155) against 15% for computed radiography (55/367) and 19% for film-screen mammography (75/400) (Table 2), and they are of higher grade than for the two other technologies. Those results are concordant with the DMIST and Vestfold studies [3, 11]. Direct digital mammography also detects more grade 3 invasive cancers, although there is a trend for a high detection of grade 1 with film-screen mammography, particularly in fatty breasts. A recent article [30] analyzing the routine results of a population-based screening program found results similar to ours on differential detection rates of digital mammography and film-screen mammography and the more frequent detection of highgrade cancers by digital mammography. The benefits and possible adverse effects of screening are highly dependent on the characteristics of the detected cancers. In this context, we observed that direct digital mammography may detect more life threatening cancers than film-screen mammography and computed radiography at the expense of an increased rate of false-positive findings, especially in surgical biopsies [21]. Filmscreen mammography detects more indolent tumors, especially in nondense breasts, and computed radiography is less efficient than the other two technologies in most instances. As the DMIST study has already shown, the superiority of direct digital mammography is especially obvious in dense breasts in which the performance of film-screen mammography is quite poor. Our results validate this observation in the context of a population-based screening program using hardcopy reading with strict quality control. This observation is enhanced in that direct digital mammography technology is associated with a smaller breast radiation dose than either film-screen mammography or computed radiography [31]. The good performance of direct digital mammography was observed in the context of many radiologic facilities in the private sector. Several facilities installed digital equipment in 2007 and radiologists learned to read digital mammograms and to use them for diagnostic or opportunistic screening before the implementation in the screening program in Any nonrandomized study is subject to bias. In the current study, this bias may be caused by the differences in the population attending the three arms of the study or by the heterogeneity of the radiologists reading the results of the screening examinations. We will address these two possible causes of bias separately. This comparative study was based on three large concurrent series of screening participants from the same population attending a single screening program who had a free choice of radiologist. The population attending the screening program was from Marseilles, a large city (37%), and from the semirural part of the district (63%). There is a large number of screening facilities in the district. The direct digital mammography population was more represented in Marseilles. Women chose the facility in their residential area without knowledge of the technology on the invitation. Women of lower socioeconomic classes are dispersed throughout the district. The population undergoing direct digital mammography compared with film-screen mammography or computed radiography was not exactly similar, with many confounding variables (Table 1). Direct digital mammography subjects were slightly younger, had less dense breasts, were more frequently attending their first examination in the program, and were more often undergoing HRT. The first examination in the program was not necessarily the first mammography for the women, who could have undergone mammography at the recommendation of a gynecologist. In France, gynecologists recommend mammography before the age of 50, more often on direct digital mammography units for opportunistic screening. The difference in breast density between digital mammography and film-screen mammography subjects is mainly explained by contrast differences in the two technologies. After adjusting for the available characteristics of the women in a logistic regression analysis, we found that cancer detection rates and abnormal mammography rates were significantly higher for direct digital mammography than for the two other technologies, but the largest difference was seen between computed radiography and direct digital mammography. Not surprisingly, the small differences observed on the confounding variables had only a small effect on the differences in detection rates. No adjustment was made for clinical examination because only two invasive radiologically occult cancers were found in this way (one by a direct digital mammography unit and one by a film-screen mammography unit). The better performance of direct digital mammography is concordant with the studies by other investigators [3, 11 14]. The bias possibly caused by the heterogeneity of the radiologists is more subtle. The quality of direct digital mammography readers may have played a role in the better screening performance observed in this group. Facilities equipped with direct digital mammography systems more often have a high yearly throughput of mammography, and radiologists working with this technology are often involved in breast imaging. However, the dual reading system, including the assessment of an expert second reader, has certainly corrected for this possible bias. The final comparison was not made between the assessments of the first readers working with the various technologies but rather between the final assessments, including those of the second readers. This argument is strengthened in that the second readers were the same during the 3 years of the study. The quality of the computed radiography systems may change in the future. A new type of computed radiography called NIP for needle image plate was introduced by three manufacturers to replace the current powder phosphor plates. This should improve image quality and lower the radiation dose of computed radiography systems. This new technology appeared on the market in 2011 and was in use only briefly at the time of our study. The same type of argument is valid for our analysis of direct digital mammography units. It referred to the systems available between 2008 AJR:202, January

8 Séradour et al. and In particular, it did not include the new machines. Our study strongly suggests that direct digital mammography may improve the efficacy of breast screening with mammography by earlier detection of life-threatening tumors that were not seen on film-screen mammography, especially in dense breasts. However, before any conclusions can be drawn, we need a longer follow-up of our screening program, which will provide a precise incidence of the interval cancer rates and a complete evaluation of the potential advantage of this new technology. Acknowledgment We thank Françoise-Marie Morin for her help in preparing the manuscript. References 1. Lipasti S, Anttila A, Pamilo M. Mammographic findings of women recalled for diagnostic workup in digital versus screen-film mammography in a population-based screening program. Acta Radiol 2010; 51: Heddson B, Ronnow K, Olsson M, Miller D. Digital versus screen-film mammography: a retrospective comparison in a population-based screening program. Eur J Radiol 2007; 64: Pisano ED, Gatsonis C, Hendrick E, et al.; Digital Mammographic Imaging Screening Trial (DMIST) Investigators Group. Diagnostic performance of digital versus film mammography for breast-cancer screening. N Engl J Med 2005; 353: Van Ongeval C, Van Steen A, Vande Putte G, et al. Does digital mammography in a decentralized breast cancer screening program lead to screening performance parameters comparable with film-screen mammography? Eur Radiol 2010; 20: Weigel S, Decker T, Korsching E, Hungermann D, Böcker W, Heindel W. Calcifications in digital mammographic screening: improvement of early detection of invasive breast cancers? Radiology 2010; 255: Vinnicombe S, Pinto Pereira SM, McCormack VA, Shiel S, Perry N, Dos Santos Silva IM. Fullfield digital versus screen-film mammography: comparison within the UK breast screening program and systematic review of published data. Radiology 2009; 251: Lastier D, Salines E, Rogel A. Évaluation du programme de dépistage organisé du cancer du sein en France: résultats 2009, évolutions depuis Saint-Maurice, France: Institut de veille sanitaire, Perry N, Broeders M, dewolf C, Törnberg S, Holland R, von Karsa L, eds. European guidelines for quality assurance in breast cancer screening and diagnosis. Luxembourg: European Commission, American College of Radiology. The American College of Radiology Breast Imaging Reporting and Data System (BI-RADS), 4th ed. Reston, VA: American College of Radiology, Skaane P, Hofvind S, Skjennald A. Randomized trial of screen-film versus full-field digital mammography with soft-copy reading in populationbased screening program: follow-up and final results of Oslo II study. Radiology 2007; 244: Vigeland E, Klaasen H, Klingen TA, Hofvind S, Skaane P. Full-field digital mammography compared to screen film mammography in the prevalent round of a population-based screening programme: the Vestfold County Study. Eur Radiol 2008; 18: Del Turco MR, Mantellini P, Ciatto S, et al. Fullfield digital versus screen-film mammography: comparative accuracy in concurrent screening cohorts. AJR 2007; 189: Hambly NM, McNicholas MM, Phelan N, Hargaden GC, O Doherty A, Flanagan FL. Comparison of digital mammography and screening-film mammography in breast cancer screening: a review in the Irish Breast Screening Program-. AJR 2009; 193: Karssemeijer N, Bluekens AM, Beijerinck D, et al. Breast cancer screening results 5 years after introduction of digital mammography in a population-based screening program. Radiology 2009; 253: Skaane P. Digital mammography in European population-based screening programs. In: Bick U, Diekmann F, eds. Digital mammography. Berlin, Germany: Springer-Verlag, 2009: Kerlikowske K, Hubbard RA, Miglioretti DL, et al. Comparative effectiveness of digital versus film-screen mammography in community practice in the United States. Ann Intern Med 2011; 155: Carney PA, Miglioretti DL, Yankankas BC, et al. Individual and combined effects of age, breast density and hormone replacement therapy use on the accuracy of screening mammography. Ann Intern Med 2003; 138: Séradour B, Estève J, Heid P, et al. Hormone replacement therapy and screening mammography: analysis of the results in the Bouches du Rhône programme. J Med Screen 1999; 6: Pisano ED, Zuley M, Baum JK, Marques HS. Issues to consider in converting to digital mammography. Radiol Clin North Am 2007; 45: Perry NM, Patani N, Milner SE, et al. The impact of digital mammography on screening a young cohort of women for breast cancer in an urban specialist breast unit. Eur Radiol 2011; 21: Feeley L, Kiernan D, Mooney T, et al. Digital mammography in a screening programme and its implications for pathology: a comparative study. J Clin Pathol 2011; 64: Van Ongeval C, Bosmans H, Van Steen A, et al. Evaluation of the diagnostic value of a computed radiography system by comparison of digital hard copy images with screen-film mammography: results of a prospective clinical trial. Eur Radiol 2006; 16: Warren LM, Mackenzie A, Cooke J, et al. Effect of image quality on calcification detection in digital mammography. Med Phys 2012; 39: Sala M, Salas D, Belvis F, et al. Reduction in false-positive results after introduction of digital mammography: analysis from four populationbased breast cancer screening programs in Spain. Radiology 2011; 258: Skaane P. Studies comparing screen-film mammography and full-field digital mammography in breast cancer screening: updated review. Acta Radiol 2009; 50: Séradour B, Ancelle-Park R. Dépistage organisé des Cancers du Sein: peut-on comparer les résultats du programme français aux résultats internationaux? J Radiol 2006; 87: Bluekens AM, Karssemeijer N, Beijerinck D, et al. Consequences of digital mammography in population-based breast cancer screening: initial changes and long-term impact on referral rates. Eur Radiol 2010; 20: Pisano ED, Hendrick EE, Yaffe MJ, et al. Diagnostic accuracy of digital versus film mammography: exploratory analysis of selected population subgroups in DMIST. Radiology 2008; 246: Ciatto S, Houssami N, Apruzzese A, et al. Categorizing breast mammographic density: intraand interobserver reproducibility of BI-RADS density categories. Breast 2005; 14: Bluekens AM, Holland R, Karssemeijer N, Broeders MJ, den Heeten GJ. Comparison of digital screening mammography and screen-film mammography in the early detection of clinically relevant cancers: a multicenter study. Radiology. 2012; 265: Public Health England website. NHS Breast Screening Programme: mammography equipment equipment evaluation and guidance. www. cancerscreening.nhs.uk/breastscreen//publications/ mammography-equipment.html. Accessed September 10, AJR:202, January 2014