Policy #:252 Latest Review Date: October 2016

Transcription

1 Name of Policy: Digital Breast Tomosynthesis Policy #:252 Latest Review Date: October 2016 Category: Radiology Policy Grade: B Background/Definitions: As a general rule, benefits are payable under Blue Cross and Blue Shield of Alabama health plans only in cases of medical necessity and only if services or supplies are not investigational, provided the customer group contracts have such coverage. The following Association Technology Evaluation Criteria must be met for a service/supply to be considered for coverage: 1. The technology must have final approval from the appropriate government regulatory bodies; 2. The scientific evidence must permit conclusions concerning the effect of the technology on health outcomes; 3. The technology must improve the net health outcome; 4. The technology must be as beneficial as any established alternatives; 5. The improvement must be attainable outside the investigational setting. Medical Necessity means that health care services (e.g., procedures, treatments, supplies, devices, equipment, facilities or drugs) that a physician, exercising prudent clinical judgment, would provide to a patient for the purpose of preventing, evaluating, diagnosing or treating an illness, injury or disease or its symptoms, and that are: 1. In accordance with generally accepted standards of medical practice; and 2. Clinically appropriate in terms of type, frequency, extent, site and duration and considered effective for the patient s illness, injury or disease; and 3. Not primarily for the convenience of the patient, physician or other health care provider; and 4. Not more costly than an alternative service or sequence of services at least as likely to produce equivalent therapeutic or diagnostic results as to the diagnosis or treatment of that patient s illness, injury or disease. Page 1 of 26

2 Description of Procedure or Service: Digital breast tomosynthesis uses modified digital mammography equipment to obtain additional radiographic data that are used to reconstruct cross-sectional slices of breast tissue. Tomosynthesis may improve the accuracy of digital mammography by reducing problems caused by overlapping tissue. Tomosynthesis involves some additional imaging time and radiation exposure, although a recently improved modification may change this. Conventional mammography produces two-dimensional (2D) images of the breast. Overlapping tissue on a 2D image can mask suspicious lesions or make benign tissue appear suspicious, particularly in women with dense breast tissue. As a result, women may be recalled for additional mammographic spot views. Inaccurate results may lead to unnecessary biopsies and emotional stress, or to a potential delay in diagnosis. The spot views are often used to evaluate microcalcifications, opacities or architectural distortions or to distinguish masses from overlapping tissue, as well as to view possible findings close to the chest wall or in the retroareolar area behind the nipple. The National Cancer Institute (NCI) reports that approximately 20% of cancers are missed at mammography screening. Average recall rates are approximately 10%, with an average cancer detection rate of 4.7 per 1,000 screening mammography examinations. The Mammography Quality Standards Act audit guidelines anticipate two to ten cancers detected per 1,000 screening mammograms. Interval cancers, which are detected between screenings, tend to have poorer prognoses. Digital breast tomosynthesis was developed to improve the accuracy of mammography by capturing three-dimensional (3D) images of the breast, further clarifying areas of overlapping tissue. Developers proposed that its use would result in increased sensitivity and specificity, as well as fewer recalls due to inconclusive results. Digital breast tomosynthesis produces a 3D image by taking multiple low-dose images per view along an arc over the breast. During breast tomosynthesis, the compressed breast remains stationary while the x-ray tube moves approximately one degree for each image in a degree arc, acquiring images. These images are projected as cross-sectional slices of the breast, with each slice typically one-mm thick. Adding breast tomosynthesis takes about 10 seconds per view. In one study in a research setting, the mean time to interpret the results was 1.22 (standard deviation [SD]=1.15) minutes for digital mammography and 2.39 (SD=1.65) for combined digital mammography and breast tomosynthesis. With conventional 2D mammography, breast compression helps decrease tissue overlap and improve visibility. By reducing problems with overlapping tissue, compression with breast tomosynthesis may be reduced by up to 50%. This change could result in improved patient satisfaction. A machine equipped with breast tomosynthesis can perform 2D digital mammography, 3D digital mammography, or a combination of both 2D and 3D mammography during a single compression. The radiation exposure from tomosynthesis is roughly equivalent to a mammogram. Therefore, adding tomosynthesis to mammography doubles the radiation dose, although it still is below the maximum allowable dose established in the U.S. Mammography Quality Standards Act. Page 2 of 26

3 Studies typically compare one- or more commonly, two-view breast tomosynthesis alone or combined with standard 2D mammography with standard 2D mammography alone. A 2014 TEC assessment (updated in 2015) focuses on two-view tomosynthesis. The FDA Radiological Devices Panel, which reviewed this new modality in 2011, recommended that two-view breast tomosynthesis is preferable to one-view tomosynthesis (both used in combination with full-field digital mammography). In May 2013, the FDA approved new tomosynthesis software that will permit creation of a 2D image (called C view) from the tomosynthesis images. As a result, the 2D mammography may become unnecessary, thereby lowering the radiation dose. In other words, only the tomosynthesis procedure will be needed and both 2D and 3D images will be created from them. It is too early to gauge how traditional mammography plus tomosynthesis compares to the C view plus tomosynthesis. Policy: Digital Breast Tomosynthesis (DBT) does not meet Blue Cross and Blue Shield of Alabama s medical criteria for coverage and is considered investigational in the screening or diagnosis of breast cancer. Blue Cross and Blue Shield of Alabama does not approve or deny procedures, services, testing, or equipment for our members. Our decisions concern coverage only. The decision of whether or not to have a certain test, treatment or procedure is one made between the physician and his/her patient. Blue Cross and Blue Shield of Alabama administers benefits based on the member s contract and corporate medical policies. Physicians should always exercise their best medical judgment in providing the care they feel is most appropriate for their patients. Needed care should not be delayed or refused because of a coverage determination. Key Points: The most recent update reviewed the literature through August 18, Primary outcomes reported in current evidence include the number of cancers detected and the number of unnecessary recalls and biopsies. Improvement in sensitivity and specificity of testing is an intermediate outcome that will impact the ultimate health outcomes, but is not by itself sufficient to establish that outcomes are improved. If the sensitivity of breast cancer detection is improved by tomosynthesis, then the number of cases detected will increase. If the specificity of cancer detection is improved, then the number of recalls and biopsies for patients without cancer will decrease. If tomosynthesis is performed during screening, the number of unnecessary recalls may decline. Other relevant screening outcomes, not available or estimable from current evidence, include: (1) breast-cancer specific mortality, (2) interval cancer rates, and (3) overdiagnosis or detection of cancers that would otherwise not go on to cause symptoms or death. If tomosynthesis is performed as part of the diagnostic workup, after a woman is recalled for questionable findings during screening, then a lower false-positive rate could prevent unnecessary biopsies. Page 3 of 26

4 Screening The 2015 TEC Assessment identified twelve studies that addressed the use of mammography with or without digital breast tomosynthesis for screening. [A]n explorative analysis of the first half of the study population in the ongoing Malmö Breast Tomosynthesis Screening Trial, published subsequent to the TEC Assessment, is also described. These studies, as well as more recent studies, are described next and are summarized in Table 1. The Assessment concluded: Current evidence on use of breast tomosynthesis plus mammography versus mammography alone for screening is insufficient to permit conclusions regarding the effect on health outcomes of adding breast tomosynthesis. Most available studies are retrospective and many did not use adequate reference standards. Also, most subjects did not serve as their own controls. The risk of bias for most of the studies was high. In addition, there is substantial variation across studies concerning (1) the magnitude of the impact of adding tomosynthesis on recall rates, and (2) whether any increases in cancer detection rates are statistically significant. Variations may be due to inadequate sample sizes, and the effect of other factors such as whether the screening is the initial or repeat examination. Adding tomosynthesis to mammography may also increase over detection and entails greater radiation exposure. Therefore, the available evidence does not permit conclusions regarding the effect of the tomosynthesis on health outcomes. Table 1 Studies of Digital Breast Tomosynthesis for Breast Cancer Screening Cancers Recalls/1000 PPV for Study No. Cancers/ Detected/1000 Screens Recalls No. Patients Screens (95% (95% CI) (95% CI) CI) Prospective Studies Patients served as their own controls Skaane (2013) 121/12,621 Mammo Mammo + DBT p value STORM,a Mammo Mammo + DBT p value MBST (2015) (exploratory results) Mammo Mammo + DBT p value Sumkin (2015) Mammo Mammo + DBT Bernaldi (2016, STORM-2) Mammo Mammo + DBT Synthetic mammo 59/ /7500 6/ % 29.1% /9672 False-positive recall (%) 3.42(3.07 to 3.80) 3.97 (3.59 to 4.38) 4.45% (4.05 to 11% 19% 24% 24% (3.8 to 7.3) 8.1 (6.2 to 10.4) 6.3 (4.6 to 8.3) 8.9 (6.9 to 11.3) Small number of cancers see text 6.3 (4.8 to 8.1) 8.5 (6.7 to 10.5) 8.8 (7.0 to 10.8) PPV for Biopsies (95% CI) Page 4 of 26

5 + DBT 4.89) Rafferty (2013) c 51/997 Study 1 Reported Mammo separately for Mammo + DBT cancer and Study 2 noncancer cases Mammo Mammo + DBT Retrospective Studies 43% 56% 47% 50% Multireader study; patients apparently served as their own controls Good (2008) 35/125 Gur (2009) (2011) Reported DM alone separately for DBT alone cancer and DM before DBT noncancer cases DM + DBT Patients given choice to receive tomosynthesis and did not serve as their own controls Destounis (2014) 5/1048 Mammo Mammo + DBT p value % 50.0% Durand (2015) d Mammo alone Mammo + DBT p value Greenberg (2014) e Mammo alone Mammo + DBT p value Haas (2013) d DM 105/17, /59,617 71/13,158 DM + DBT p value Pre-post implementation of tomosynthesis Friedewald (2014) e Mammo alone Mammo + DBT p value Lourenco (2015) DM DM + DBT p value McCarthy (2014) DM DM + DBT p value 123 (117 to 130) 78 (73 to 84) 155 (138 to 175) 134 (119 to 152) 120 (113 to 128) 84 (77 to 91) < /454, (89 to 124) 91 (73 to 108) 113/25, /26, (88 to 99) 64 (60 to 68) 104 (98 to 109) 88 (83 to 92) % (2.6 to 3.4) 4.5% (3.8 to 5.4) < NS 4.9 (4.2 to 5.7) 6.2 (5.2 to 7.5) % 6.4% 5.8% 7.2% NS 4.4% (3.2 to 5.6) 6.2% (4.9 to 7.5) (3.8 to 4.7) 5.4 (4.9 to 6.0) NS 4.6 (3.3 to 5.8) 5.5 (4.3 to 6.6) NS 21.5% (18.9 to 24.5) 22.7% (19.5 to 26.6) NS 24.2% 29.2% 30.2% 23.8% 24.7% (18.6 to 30.9) 25.4% (20.6 to 30.2) NS Page 5 of 26

6 Rose (2013) (recalls at BI- RADS=0) DM DM + DBT p value McDonald (2016) DM DBT (year 1) DBT (year 2) DBT (year 3) p value (vs DM, year 1, 2, and 3) Conant (2016) f DM DM + DBT p value Sharpe (2016) DM DBT p value 107/23,357 /23,958 /198, /85, ,, < % 10.1% 4.4% 6.2% 6.5% 6.7% 0.06, 0.03, % 6.4% g NS NS <0.002 NS for DM vs DBT, all years BI-RADS: Breast Imaging Reporting and Data System; CI: confidence interval; DBT: digital breast tomosynthesis; DM: digital mammography; Mammo: mammography; Med: medium; : not reported; PPV: positive predictive value; UN: uncertain. a Data from Houssami (2014); other outcomes are from Ciatto (2013). b Enriched samples include more cancer cases than expected in a routine screening population. c Twenty-seven women for whom there was no follow-up were not included in these results. d Patient samples overlap in the Haas and Durand studies. e Adjusted estimates reported in this table. f Overlap with data in McDonald (2016). g PPV defined as number of cancers diagnosed per number of positive screens. Prospective Studies The strongest evidence for using mammography and breast tomosynthesis for screening women for breast cancer comes from the interim results of a large 2013 trial in Norway. The sample consisted of 12,621 women with 121 screening-detected cancers who underwent routine screening. The cancer detection rate was 6.1 per 1000 screenings for mammography alone and 8.0 per 1000 screenings for mammography plus digital breast tomosynthesis. Cancers missed by digital breast tomosynthesis were missed due to reading errors, either detection or interpretation. After adjusting for reader differences, the ratio of cancer detection rates for mammography versus mammography plus breast tomosynthesis was 1.27 (98.5% confidence interval [CI]: 1.06 to 1.53; p=0.001). The authors note that they did not ascertain any improvement in detecting ductal carcinoma in situ (DCIS) by adding breast tomosynthesis; the additional cancers detected were largely invasive. The false-positive rate was 61.1 per 1,000 screenings for mammography alone and 53.1 per 1,000 screenings for mammography plus breast tomosynthesis. A reduction in the false-positive rate would decrease the number of women recalled after screening for additional imaging or biopsy. In Norway, as in much of Europe, women are screened every other year, and two readers independently interpret the images, which differs from usual practice in the U.S. After adjusting for differences across readers, the ratio of false-positive rates for mammography alone versus mammography plus breast tomosynthesis was 0.85 (98.5% CI: 0.76 Page 6 of 26

7 to 0.96; p). For this interim analysis, only limited data were available about interval cancers so conventional absolute sensitivity and specificity could not be estimated. The second study (STORM) examined comparative cancer detection for traditional mammography with or without breast tomosynthesis in a general Italian, asymptomatic screening population of 7,292 women. The reference standard was pathology for women undergoing biopsies; women with negative results on both mammography and breast tomosynthesis were not followed up, so neither the sensitivity nor specificity could be calculated. Mammography plus breast tomosynthesis revealed all 59 cancers, while 20 of them were missed by traditional mammography (p<0.0001). The incremental cancer detection of using both modalities was 2.7 cancers per 1,000 screens (95% CI: 1.7 to 4.2). There were 395 false-positive results: 181 were false positive using either mammography or both imaging modalities together; an additional 141 occurred using mammography only and 73 occurred using mammography and breast tomosynthesis combined (p<0.0001). In preplanned analyses, the researcher found that the combined results of mammography and digital breast tomosynthesis yielded more cancers in both age groups (<60 versus >60 years) and breast density categories (1, least dense, and 2 versus 3 and 4, most dense). Both of these studies were considered to have a medium risk of bias using the Quality Assessment of Diagnostic Accuracy Studies (QUADAS)-2 tool. In both studies, follow-up of women with negative imaging findings was insufficient. Subsequent to the TEC Assessment, Lång et al reported exploratory results from the first half of the Malmö Breast Tomosynthesis Screening Trial (MBTST), comparing one view (MLO) DBT (a lower radiation dose than digital mammography [DM]) with two view DM. The MBTST is a one arm, single-institution, prospective study. Randomly selected women in Malmö, Sweden (aged 40 to 74 years) were offered one view DBT and two view DM. A sample size of 15,000 was specified to detect an improvement in cancer detection sensitivity from 63% to 88% (power of 80%); 7500 were included in the exploratory analysis. In Sweden, breast cancer screening is offered to women between ages 40 and 55 every 18 months and every 24 months thereafter to age 74. The primary outcomes were cancer detection rates, recall rates, and positive predictive values (PPVs). Six experienced readers interpreted images (mean, 26 years of experience; range, 8-41). Blinded double reading was carried out for DBT and DM with rule-based arbitration for disagreements. Pathologic findings after abnormal imaging (additional DM, ultrasound, and indicated biopsy) were the basis for outcomes. Women in this exploratory analysis were followed at least one year for development of cancer ascertained through the South Swedish Cancer Registry. Of 10,547 women invited, 71.1% participated with 20% undergoing their first screening test. Cancers were detected in 47 of the DM and 67 of the DBT reading arms (21 by DBT alone; one by DM alone; 46 by both modalities). The detection rate for DBT was 8.9 per 1000 screens (95% CI, 6.9 to 11.3) and for DM was 6.3 per 1000 screens (95% CI, 4.6 to 8.3; p). DCIS detection rates were similar between both modalities. Following arbitration, the recall rate was lower for DM (2.6%; 95% CI, 2.3 to 3.0) than for DBT (3.8%; 95% CI, 3.3 to 4.2; p). PPVs were similar for both modalities at 24%. In these explorative MBTST results, DBT achieved a higher cancer detection rate. In contrast to other studies, however, recall rates were lower for DM. Tomosynthesis-detected cancers were Page 7 of 26

8 generally smaller, of lower grade, more often node negative, and found among women somewhat younger. Regarding this finding, the authors noted, It is not clear whether this represents earlier diagnosis and/or overdiagnosis from DBT screening, since the present study was not designed to address these issues. Additionally, DBT-alone detected cancers tended to be in women with dense and fatty breasts. Although the findings are from an explorative analysis, the authors stated that...one-view DBT may be feasible as a stand-alone technique for breast cancer screening. The partial results include only half of the planned sample accrual with follow-up insufficient to assess interval cancer rates. Experienced readers were used and all studies read independently by two readers. Final results will require consideration of broader applicability. Rafferty et al in 2013 compared the results of mammography alone versus breast tomosynthesis plus mammography among 997 subjects with mixed indications: 780 were women undergoing routine screening, and 217 were women scheduled for biopsy. Two reader studies were conducted. Some of these results were included in the submission to the U.S. Food and Drug Administration (FDA) for premarketing application approval of Hologic, Inc. s Selenia Dimensions tomosynthesis system. Readers were trained in interpreting tomosynthesis images, and the training was augmented between the first and second reader studies to emphasize how to read certain lesions that were often misinterpreted in the first reader study. In both reader studies, the area under the receiver operating characteristic curve for mammography plus breast tomosynthesis was greater than for mammography alone; the difference for the second study was 6.8% (95% CI: 4.1% to 9.5%, p). For noncancer cases, adding breast tomosynthesis to mammography changed the mean recall rate across readers for study two from 48.8% (95% CI: 28.2% to 69.1%; SD=12.3%) to 30.1% (95% CI: 19.8% to 41.3%; SD=7.6%) for the combined modalities. Almost all of the improvement among readers was attributable to non-calcification cases, including masses, asymmetries, and architectural distortions. Risk of bias was considered high due to: substantial patient dropout (16%) after enrollment; use of an investigational tomosynthesis system during the early part of the study that may have yielded results different from the FDA-approved system; and enrichment of the screening sample by inclusion of the biopsy cases. In 2016, Bernardi et al published findings of the prospective population-based STORM-2 study comparing DBT with acquired (standard) mammography images and to synthetic mammography images. The study included asymptomatic women attending a breast cancer screening program in Italy. Synthetic mammography images, using this technique, offer the advantage that women do not need two sets of images, which reduces their exposure to radiation. Double-reading was conducted for mammography images. Findings are shown previously in Table 1. The cancer detection rate did not differ significantly in the DBT plus acquired mammography and DBT plus synthetic mammography groups (p=0.58). The false-positive recall rates were significantly higher in both DBT groups compared with mammography alone. Retrospective Studies In three related articles, Good et al and Gur et al compared diagnostic performance of digital mammography alone versus digital mammography plus tomosynthesis. One article reported that the recall rate among noncancer cases was 42% (95% CI: 0.38 to 0.45) for digital mammography alone and 28% (95% CI: 0.25 to 0.31) for digital mammography plus breast tomosynthesis (p<0.0001). The analogous rates for cancer cases were 88% (95% CI: 0.84 to 0.91) for digital Page 8 of 26

9 mammography alone and 93% (95% CI: 0.90 to 0.96) for digital mammography plus breast tomosynthesis. The sensitivity of digital mammography alone was 60% and increased to 72% when breast tomosynthesis was added (p=0.034, but the authors note the small number of positive findings). These articles did not describe the sample, the time between digital mammography and breast tomosynthesis, or how the reference standard was verified. Therefore, risk of bias is unknown. In 2016, McDonald et al compared the diagnostic performance of DM prior to the introduction of DBT, to years 1, 2, and 3 of DBT. As shown in Table 2, breast cancer detection rates were not significantly higher with DBT than with DM. Recall rates, and the PPV of recalled cases, were significantly higher with DBT than with DM. In a related article from the same center, Zuckerman et al, from the same center, compared the performance of DM plus acquired DBT to DM plus synthetic DBT and found the latter to have similar cancer detection (5.03 per 1000 vs 5.45 per 1000, respectively) and lower recall rates (8.8% vs 7.1%, respectively; p).41 Also in 2016, Conant et al reported on screening data from three centers participating in the PROSPR consortium. Data from this study may overlap with data from the McDonald (2016) study, with data collected for both studies at the same center (ie, the University of Pennsylvania described above). The analysis compared the diagnostic accuracy of DM alone and DM plus DBT in asymptomatic women presenting for breast cancer screenings. Outcomes, shown in Table 2, were reported for cases with at least one year of follow-up. The study, which included a total of 198,881 women, found significantly higher cancer detection rates and lower recall rates with DM plus DBT versus DM alone. In a 2016 study, Sharpe et al conducted a prospective study with a retrospective cohort analysis evaluating diagnostic accuracy of breast cancer screening before and after introduction of tomosynthesis. The authors reported on 85,852 asymptomatic women undergoing breast cancer screening, 80,149 (93.4%) women were screened with mammography and 5703 (6.6%) screened with DBT. The study found significantly higher cancer detection rates and lower recall rates with DBT compared with mammography, although it is important to note that the subgroup of patients who received DBT also included a higher proportion of patients with risk factors for breast cancer and baseline examinations. A large reading study was conducted in the U.K. National Health Service. Because the study included a mix of screening and diagnostic patients, case selection bias limits extrapolation of the results to a screening population. The TOMosynthesis with digital MammographY (TOMMY) trial (ISRCTN ) compared the performance of mammography plus tomosynthesis with mammography alone among women at moderate-to-high risk for breast cancer undergoing screening (n=1040 including two cancers) and women who had been recalled for a diagnostic workup after having a screening mammogram (n=6020 including 1158 cancers). After enrollment, these women had a mammogram and tomosynthesis; a synthesized mammogram (C-View) was also reconstructed from the tomosynthesis results. Comparisons were mammography alone versus mammography plus tomosynthesis versus synthetic mammogram plus tomosynthesis. The reference standard was pathology results for women who had a biopsy; the report did not note follow-up for other subjects. Sensitivity was 87% (95% CI, 85 to 89) for mammography only, 89% (95% CI, 25 to 91) for mammography plus Page 9 of 26

10 tomosynthesis, and 88% (95% CI, 86 to 90) for C-View images plus tomosynthesis. None of these differences was statistically significant. Specificity was 58% (95% CI, to 60) for mammography alone, 69% (95% CI, 67 to 71) for mammography plus tomosynthesis, and 71% (95% CI, 69 to 73) for C-View images plus tomosynthesis. Specificity for each of the combined modalities was statistically greater compared with mammography alone (p). This result also held for all age and breast density groups. When breast density was divided into two groups (more or less than 50% density), sensitivity for women with greater breast density was improved with mammography plus tomosynthesis compared with mammography alone (93% vs 86%; p=0.03). Sumkin et al (2015) enrolled women aged 34 to 56 years of age undergoing baseline screening mammography at a single institution. High-risk women were preferentially included. All women underwent DM and DBT (Selenia Dimension, Hologic). DM and DBT images were interpreted by different radiologists (from a group of 14 having 5-28 years of experience) who were instructed not to discuss the evaluation with each other until completed. A total of 1074 women (mean age, years; SD=3.75) were studied; the number invited to participate was not reported. Just over half (50.8%) had a Breast Imaging-Reporting and Data System (BI- RADS) rating of 3, and 40.1% a BI-RADS rating of 2. The recall rate based on DM alone was 38.4% versus 25.5% with DM and DBT. Changes in recall rates varied among radiologists 5 of 11 who interpreted more than 15 images demonstrated statistically significant improvements with DBT. Two invasive cancers were detected 1 by DM alone and 1 by DBT. Four cases of DCIS were found all by both modalities. Among a younger and generally higher risk target population undergoing an initial exam, the recall rate was high for either modality. Although the addition of DBT reduced the recall rate, there was considerable variability among radiologists. Only two invasive cancers were detected (1 by each modality) preventing conclusions concerning cancer detection rates. Lack of follow-up, no report of exclusions and possible selection bias, and adequacy of blinding in interpretation are potential limitations. To obtain FDA-approval of its Mammomat tomosynthesis device, Siemens submitted a similar reading study that included a mix of screening and diagnostic patients (NCT ). As previously noted, case selection bias limits extrapolation of the results to a screening population. Of 698 enrolled women who presented for screening or for workup of an actionable lesion, 300 women with adequate follow-up (e.g., repeat mammogram at one year for negative cases) were selected for the reader study. Fifty women (17%) had biopsy-proven cancer, 85 (28%) had biopsy-proven benign lesions, and 165 (55%) had normal mammograms with one-year followup. Twenty-two readers with at least one year of experience reading digital mammograms underwent tomosynthesis training; mammograms and one and two view tomosynthesis images for all participants were read in randomized and blinded fashion. The primary efficacy end point was area under the ROC curve for probability of malignancy (POM) for each breast. Breast POM scores were derived from lesion POM confidence scores on a 0 to 100 scale, lesion localization, and breast BI-RADS scores. The area under the ROC curves was for mammography versus mammography plus two view tomosynthesis. Mean difference was (95% CI, to 0.139), a statistically significant result (p). Mean sensitivity increased for all subgroups evaluated (e.g., dense breasts, breasts with microcalcifications). No increase in specificity was observed. Limitations of the study that may impact interpretation of results include exclusion of 12 cases due to abnormal one year follow-up. Sensitivity analyses showed Page 10 of 26

11 reduction in the observed effect size, but the magnitude of this reduction was not reported. Additional issues include use of a novel breast POM score formed by combining lesion scores, and use of ROC analysis as the primary end point. ROC curves span the full range of potential discrimination thresholds, some of which may fall outside the range of clinical usefulness. This study is currently unpublished. C-View Tomosynthesis In a study of C-View tomosynthesis (N=236), Zuley et al (2014) compared diagnostic accuracy of synthesized 2D mammography (C-View) and digital mammography, both alone and in combination with 3D breast tomosynthesis. Area under the ROC curve was and for synthesized and digital mammography, respectively; with the addition of 3D tomosynthesis, values increased to and 0.939, respectively. In the second half of a study by Skaane et al (2014), after improvements to 2D image processing were made, there was no statistical difference in cancer detection rates, positive predictive values (PPVs), or false-positive rates (noncancer recall rates) between synthesized and digital mammography (both in combination with tomosynthesis). Mean glandular radiation dose for a single mammographic view was 45% less in the synthesized mammography group compared with the digital mammography group (mean, 1.58 mgy vs 3.53 mgy, respectively). Section Summary The available studies provided some evidence that adding breast tomosynthesis to mammography may increase the accuracy (and possibly the sensitivity) of screening while reducing the number of women who are recalled unnecessarily. However, the available studies have methodological limitations. Several studies did not have adequate follow-up of women with negative screening results; one larger study provided interim results. There is also a lack of data concerning interval cancers, breast-cancer mortality, and potential overdiagnosis. Other studies were retrospective case reviews; patients had mixed or unclear indications for screening. More recently, prospective and large retrospective studies have reported cancer detection rates with reduced false recall rates. This evidence is from nonrandomized designs with a lack of longterm follow-up to assess false negative results. Therefore, performance of digital breast tomosynthesis in the screening setting cannot be determined with certainty. Several studies of synthesized 2D mammography showed comparable diagnostic performance with digital mammography and lower radiation exposure.. Diagnosis Lei et al in 2014 conducted a systematic review with meta-analysis of seven studies (total number of patients, 2014; total number of lesions, 2666) that compared digital breast tomosynthesis with digital mammography in patients with Breast Imaging-Reporting and Data System (BI-RADS) two or higher breast lesions. All studies were rated high quality using the QUADAS tool. As shown in Table 2, compared with histologic diagnosis, performance of both imaging modalities was approximately similar; PPVs were low (57% for breast tomosynthesis and 50% for digital mammography), and negative predictive values (NPV) were high. Statistical heterogeneity in these analyses was considerable (I2 approximately 90%). Studies used both one view (n=4) and two view (n=3) breast tomosynthesis. Pooled sensitivity and specificity for only one view breast tomosynthesis studies were 81% and 77%, respectively; for two view studies, pooled sensitivity and specificity were 97% and 79% respectively. Page 11 of 26

12 Table 2 Side-by-Side Comparison of Digital Breast Tomosynthesis and Digital Mammography Diagnostic Performance Compared with Histologic Diagnosis: Pooled Results Digital Breast Tomosynthesis, pooled estimate (95% CI) Digital Mammography, pooled estimate (95% CI) Sensitivity 90% (87 to 92) 89% (86 to 91) Specificity 79% (77 to 81) 72% (70 to 74) PPV 57% (53 to 61) 50% (46 to 53) NPV 96% (95 to 97) 95% (94 to 97) DOR (8.70 to 77.95) (5.61 to 47.04) LR (2.31 to 5.30) 2.83 (1.77 to 4.52) LR 0.15 (0.06 to 0.36) 0.18 (0.09 to 0.38) AUC ROC AUC ROC: area under the summary receiver operating characteristic curve; DOR, diagnostic odds ratio (ratio of the odds of positivity in cases to the odds of positivity in controls = [LR+] [LR ]); LR+, positive likelihood ratio (ratio of the probability of positivity in cases to the probability of positivity in controls = sensitivity [1- specificity]); LR, negative likelihood ratio (ratio of the probability of a negative result in cases to the probability of a negative result in controls = [1-sensitivity] specificity); NPV, negative predictive value; PPV, positive predictive value Calculated by author The 2014 TEC Assessment identified six studies address the use of breast tomosynthesis in the diagnostic setting, (i.e., when there are suspicious findings on screening mammography or if the woman is symptomatic). Findings of these studies are displayed in Table 3. Table 3 Studies of Diagnostic Digital Breast Tomosynthesis in 2014 TEC Assessment Study (Year) AUC Sensitivity Specificity PPV NPV Bernardi (2012) Screening mammo plus DBT 100% 74% 38% 100% Brandt (2013) Diagnostic mammo plus ultrasound % 94% DBT plus US Reader % 94% Reader % Reader % 89% Tagliafico (2012) Mammo spot view % 94% DBT % 100% Zuley (2013) BI-RADS 3-5 = positive Screening plus diagnostic mammo % 15% Screening mammo plus DBT % 26% BI-RADS 4-5 = positive Screening plus diagnostic mammo 89% 43% Screening mammo plus DBT 90% 52% BI-RADS 5 = positive Screening plus diagnostic mammo 33% 98% Screening mammo plus DBT 39% 98% Michell (2012) Film plus digital mammo % 51% 42% 98% Film plus digital mammo plus DBT % 74% 59% 100% AUC: area under the curve; BI-RADS: Breast Imaging Reporting and Data System; DBT: digital breast tomosynthesis; mammo: mammography. Page 12 of 26

13 The studies noted in Table 3 varied considerably in types of suspicious mammographic findings (e.g., calcifications vs noncalcifications); patient sample; and comparators to breast tomosynthesis (e.g., 2-view mammography, mammographic spot views, ultrasound). One study had a medium risk of bias49; the remainder, a high risk of bias using the QUADAS-2 tool. In a study of 158 women consecutively recalled after screening mammography, breast tomosynthesis was evaluated as a possible triage tool to reduce the number of false-positive results. The results of the diagnostic assessment (including ultrasound and needle biopsy where performed) were used as the reference standard. Breast tomosynthesis eliminated 102 of the 158 recalls, all of which were unnecessary (i.e., false-positive results on mammography). No cancers were missed on breast tomosynthesis. The performance of breast tomosynthesis did not vary by breast density or age group, but the reduction in recalls was greater for asymmetric densities and distortions, and nodular opacities with regular margins. The authors note that the decline in recall rates following the use of breast tomosynthesis was higher in this study than in blinded comparisons of digital mammography and breast tomosynthesis. Tagliafico et al (2012) compared the performance of mammographic spot views versus tomosynthesis among 52 consecutive recalled women with a BI-RADS rating on initial screening of 0 (which means Need Additional Imaging Evaluation and/or Prior Mammograms for Comparison ). Women with calcifications were excluded. The study was designed as a noninferiority analysis for areas under the receiver operating characteristic (ROC) curve, sensitivity, and specificity, with a non-inferiority margin of delta=0.05, so that if breast tomosynthesis were non-inferior to mammographic spot views, breast tomosynthesis could be performed right after screening mammography to avoid a recall. The sensitivity and specificity were extremely high for both modalities, and there was no statistically significant difference between them. Brandt et al (2013) compared diagnostic mammography to breast tomosynthesis among 146 women with abnormalities on screening mammography with no calcifications in a simulated clinical setting. The breast tomosynthesis rating was based on readers ratings and their confidence that no additional studies were needed, as well as ultrasound results in some cases. The reference standard was either the results of the entire clinical workup, including biopsy if performed, or follow-up for women not undergoing biopsy (86% of entire sample).there was not a statistically significant difference between diagnostic mammography and breast tomosynthesis in sensitivity or specificity. Two of these three studies found no difference in sensitivity and specificity between breast tomosynthesis and a clinical workup that consisted of diagnostic mammographic images or a more comprehensive diagnostic work-up. The third study examined the use of breast tomosynthesis to triage women recalled after screening and substantially reduced the recall rate. Michell et al (2012) evaluated 738 women with 759 lesions recalled after screening with film mammography. In this unblinded study, the incremental value of breast tomosynthesis added to film and digital mammography was assessed. The reference standard consisted of pathology results or follow-up for 18 to 36 months. Adding breast tomosynthesis to film and digital mammography results increased the area under the ROC curve from ( ) to ( ) (p=0.001). The complete sensitivity (counting ratings of three to five as Page 13 of 26

14 positive) increased from 39.7% for digital mammography to 58.3% when breast tomosynthesis was added; no confidence intervals or p values were reported. The specificity increased from 51% to 74.2% when breast tomosynthesis was added to digital mammography. The difference in areas under the ROC curve after the addition of breast tomosynthesis was statistically significant for soft-tissue lesions, but not for microcalcifications. Zuley et al (2013) compared diagnostic mammography images to dual-view breast tomosynthesis in 217 lesions (72 [33%] malignant) among 182 women. In this retrospective study, women who had undergone diagnostic mammography and breast tomosynthesis were included. The sample included women with clinical symptoms such as a palpable lump, or findings on mammograms, ultrasound, or magnetic resonance imaging (MRI). Women with only calcifications were excluded. Area under the ROC curve was 0.83 (95% CI, 0.77 to 0.83; range across readers ) for diagnostic mammography, and 0.87 (95% CI, 0.82 to 0.92; range across readers, ) for DBT (p). Skaane et al (2012), authors of the Norse trial also wrote another article on their initial experience with digital breast tomosynthesis in a clinical setting. The study sample comprised 129 women with a palpable lump (23%) or abnormal findings on screening mammography (42%), and women undergoing surveillance after prior biopsy or breast conservation surgery (35%). Twenty-five cancers were diagnosed, including 18 invasive cancers, five ductal carcinomas in situ (DCIS), and two other cancers. Thirty-eight percent of the women had dense breasts (BI-RADS density categories of three or four). Of the 50 women in the reader study, 23 had cancer, five of which were DCIS. All three readers missed two of the DCIS cases. Two cancers were detected in women with dense breasts, increasing cancer detection by 8%. Several studies assessing diagnostic digital breast tomosynthesis have been published subsequent to the TEC Assessment. These studies are summarized in Table 4. These studies reported that addition of tomosynthesis to digital mammography increased diagnostic accuracy overall, with improvements in true positive rates (sensitivity) exceeding improvements in true negative rates (specificity). However, PPV remained low (approximately 50%). Differences in test performance between studies (i.e., between Rafferty 2014 and Thibault 2013) are likely due to the difference in technologies studied (2-view digital mammography plus 1-view tomosynthesis vs 1-view digital mammography plus 1-view tomosynthesis, respectively), but also to differences in sample size (310 vs 130, respectively), setting (U.S. vs Europe, respectively), number (15 vs 7, respectively), training (150 cases vs 20 cases, respectively) of readers. Table 4 Studies of Diagnostic Digital Breast Tomosynthesis 2014 to present AUC Sens Spec PPV NPV Rafferty 2014, N=310 DM % 86% 47% 92% DM + 1-view DBT % 86% 50% 94% DM + 2-view DBT % 85% 50% 95% Gennaro 2013, N=463 DM 76% 1-view (CC) DM + 1-view DBT 79% Thibault 2013, N=130 DM % 53% 53% 74% Page 14 of 26

15 1-view (CC) DM + 1-view DBT % 64% 58% 73% DM + 1-view DBT + US % 52% 55% 79% AUC, area under the receiver operating characteristic curve; CC, cranio-caudal; DBT, digital breast tomosynthesis; DM, digital mammography (2-view unless noted otherwise); MLO, mediolateral-oblique;, not reported; US, ultrasound Note: 1-view DBT is MLO unless noted otherwise. Statistically significant difference from DM ᵇStatistically significant difference from 1-view DBT Tagliafico et al (2015) compared digital mammography versus digital mammography plus tomosynthesis in BI-RADS classification of microcalcification clusters at three centers in Italy. Six radiologists evaluated images from 107 women with biopsy-proven microcalcification clusters who had both digital mammography and tomosynthesis images available and minimum follow-up of 12 months. All clusters were visible on both mammography and tomosynthesis, and 41 (38%) were malignant. Of 11 discordant results, three were malignant and were incorrectly under classified as benign (R2) by mammography plus tomosynthesis; eight (including one highrisk lesion) were nonmalignant and were incorrectly classified as malignant (R3 or R4) by mammography alone. Overall sensitivity and specificity were 100% (95% CI, 91 to 100) and 95% (95% CI, 87 to 99), respectively, for mammography, and 91% (95% CI, 79 to 98) and 100% (95% CI, 95 to 100) for mammography plus tomosynthesis. Overall intra- (based on repeated analysis of 50 image sets two months later) and interobserver agreement was good (weighted kappa range, ) for both mammography and tomosynthesis. The study suggests that tomosynthesis may miss a small proportion of malignant lesions (here 3/41 [7%]) in patients with microcalcification clusters. Cornford et al compared the accuracy of GE DBT with supplementary two view mammographic views in the diagnosis of screen-detected abnormalities. From 322 patients, 342 abnormalities were studied in a prospective analysis. Microcalcifications were uncommon (3.8%), owing to women not being invited to participate if they were the main mammographic finding. Overall discrimination of the modalities was not statistically different (p=0.12), with AUCs of (95% CI, to 0.968) for DBT versus (95% CI, to 0.948) for standard mammographic supplementary views. Section Summary This mixed set of articles provides evidence of either a similar diagnostic performance between breast tomosynthesis and other approaches or an advantage for breast tomosynthesis. Some concerns have been raised concerning classification of microcalcification clusters with DBT alone. The mixed patient populations, differences in references standard, use of different imaging tests to compare to breast tomosynthesis, and variations in follow-up make it difficult to draw a conclusion from these studies. Summary The evidence for digital breast tomosynthesis (DBT) for individuals who are asymptomatic at average risk of breast cancer includes results from several studies in which women served as their own controls and separate observational studies. Relevant outcomes are overall survival, disease-specific survival, test accuracy and validity, change in disease status, and treatmentrelated morbidity. Norse and Italian screening studies published in 2013 provide the strongest Page 15 of 26

16 evidence available to date on the use of mammography plus DBT versus mammography alone for screening women for breast cancer. Partial results (50% of anticipated enrollment) from the Malmö Breast Tomosynthesis Screening Trial (MBTST) using one view DBT offers similar evidence. This evidence suggests that use of mammography plus breast tomosynthesis may modestly increase the number of cancers detected, with a decrease in the number of women who undergo unnecessary recalls or biopsies. For example, in interim analysis of the Norse screening trial, the ratio of cancer detection rates per 1000 screens for mammography plus breast tomosynthesis versus mammography alone was 1.27 (98.5% confidence interval [CI], 1.06 to 1.53; p=0.001). The ratio of false-positive rates for mammography plus breast tomosynthesis versus mammography alone was 0.85 (98.5% CI, 0.76 to 0.96; p). Results from half the anticipated sample of the MBTST demonstrated improved sensitivity with one view DBT, but not lower recall rates. A decrease in the false-positive rate would reduce unnecessary diagnostic workups and their consequences. However, the potential for overdiagnosis cannot be ascertained because of the study designs and interval cancer rates are not yet available. Other studies were retrospective case reviews; patients had mixed or unclear indications for screening. Prospective and large retrospective studies have reported modestly higher cancer detection rates with reduced false-recall rates. The nonrandomized designs lack long-term follow-up to assess false-negative results. Long-term effects of additional radiation exposure also are unknown. Adding tomosynthesis to mammography may increase the radiation dose depending on the specific equipment and protocols used, although it still is below the maximum allowable dose established in the U.S. Mammography Quality Standards Act. The evidence is insufficient to determine the effects of the technology on health outcomes. The evidence for DBT in individuals who have abnormal findings on breast imaging or clinical exam includes multiple observational studies and one meta-analysis. Relevant outcomes are test accuracy and validity, and treatment-related morbidity. Studies show either a similar diagnostic performance between breast tomosynthesis and other approaches or an advantage for breast tomosynthesis, with two view DBT having better performance characteristics that one view DBT. Some concerns have been raised regarding classification of microcalcification clusters with DBT alone. Mixed patient populations, differences in reference standards, use of different imaging tests being compared with breast tomosynthesis, and variations in follow-up make it difficult to draw conclusions from these studies. The evidence is insufficient to determine the effects of the technology on health outcomes. Practice Guidelines and Position Statements American College of Radiology (ACR) The ACR s Statement on Breast Tomosynthesis dated November 24, 2014, included the following: A new digital technology, breast tomosynthesis has shown to be an advance over digital mammography, with higher cancer detection rates and fewer patient recalls for additional testing. This is extremely important. The medical community has long sought ways to improve breast cancer screening accuracy. Better sensitivity will likely translate into more lives saved. Lower recall rates result in fewer patients who may experience short-term anxiety awaiting test results. This is important evidence that tomosynthesis will have a positive impact on patient care. Page 16 of 26

17 As this technology is used in clinical practice, we anticipate that further studies will clarify its impact on long-term clinical outcomes, including reduced mortality. It will also be important to further elucidate which subgroups of women might benefit most from these exams (by age, breast density, frequency of examination, etc.). To be clear: tomosynthesis is no longer investigational. Tomosynthesis has been shown to improve key screening parameters compared to digital mammography. While the College encourages more studies to clarify the clinical role(s) of tomosynthesis and its long-term outcomes, it is clear that tomosynthesis represents an advance in breast imaging. The ACR s Appropriate Criteria, last reviewed in 2016, gave DBT a rating of 9 ( usually appropriate ) for the use on women that were high risk, intermediate risk, as well as average risk for breast cancer. The ACR definitions of breast cancer risk are: High-risk women: women with a BRCA gene mutation and their untested first- degree relatives, women with a history of chest irradiation between the ages of 10 30, women with 20% or greater lifetime risk of breast cancer. Intermediate-risk women: women with personal history of breast cancer, lobular neoplasia, atypical ductal hyperplasia, or 15% 20% lifetime risk of breast cancer. Average-risk women: women with <15% lifetime risk of breast cancer, breasts not dense. American College of Obstetricians and Gynecologists In its 2011 practice bulletin on breast cancer screening, the American College of Obstetricians and Gynecologists notes that digital breast tomosynthesis is one of several screening techniques that were considered but not recommended for routine screening. A 2015 committee opinion no. 625 on management of women with dense breasts diagnosed by mammography states, The American College of Obstetricians and Gynecologists does not recommend routine use of alternative or adjunctive tests to screening mammography in women with dense breasts who are asymptomatic and have no additional risk factors. Breast tomosynthesis or thermography are not cited in the document as alternative tests. American Academy of Family Physicians In 2016, the American Academy of Family Physicians issued a clinical preventive service recommendation concerning breast cancer. The recommendation states that there is insufficient evidence for an assessment of the benefits and harms of DBT as a primary screening method for breast cancer. The recommendation also states that there is insufficient evidence for an assessment of benefits and harms of DBT as adjunctive screening for breast cancer in women identified as having dense breast tissue on an otherwise negative screening mammogram. National Comprehensive Cancer Network The National Comprehensive Cancer Network (NCCN) guidelines (v ) state: Multiple studies show a combined use of digital mammography and tomosynthesis appears to improve Page 17 of 26