Recall and Cancer Detection Rates for Screening Mammography: Finding the Sweet Spot

Transcription

1 Women s Imaging Original Research Grabler et al. Optimal Recall and Cancer Detection Rates for Screening Mammography Women s Imaging Original Research Paula Grabler 1 Dominique Sighoko 2 Lilian Wang 3 Kristi Allgood 4 David Ansell 5 Grabler P, Sighoko D, Wang L, Allgood K, Ansell D Keywords: American College of Radiology guidelines, cancer detection rates, recall rate, sweet spot DOI: /AJR Received December 14, 2015; accepted after revision July 21, Department of Diagnostic Radiology and Nuclear Medicine, Rush University Medical Center, Rush Oak Park Hospital, 520 S Maple Ave, Oak Park, IL Address correspondence to P. Grabler (Paula_Grabler@rush.edu). 2 Metropolitan Chicago Breast Cancer Task Force, Rush University Medical Center, Chicago, IL. 3 Department of Diagnostic Radiology, Breast Imaging Section, Northwestern Memorial Hospital, Chicago IL. 4 Sinai Urban Health Institute, Sinai Health System, Chicago, IL. 5 Department of Internal Medicine, Rush University Medical Center, Chicago, IL. This article is available for credit. AJR 2017; 208: X/17/ American Roentgen Ray Society Recall and Cancer Detection Rates for Screening Mammography: Finding the Sweet Spot OBJECTIVE. The purpose of this study is to identify the optimal screening mammography recall rate range on the basis of cancer detection rates among breast imaging specialists at an academic institution. MATERIALS AND METHODS. Medical outcome audit data collected in accordance with the Mammography Quality Standards Act from September 1, 2007, through August 31, 2012, were reviewed. Cancer detection rates were calculated from 984 screen-detected cancers identified in 188,959 total digital screening mammograms. The percentages of minimally invasive and early-stage cancers were also calculated. The 75 annual recall rates were analyzed two ways. First, they were separated into recall groups to assess cancer detection rate variation by the recall categories using rate ratios: less than 10%, 10% to less than 12%, 12% to less than 14%, and 14% or higher. Next, a linear regression with bootstrap bias correction was performed to assess changes in cancer detection rate with each unit increase in the recall rate up to 20%, with the recall category of less than 7% taken as reference. Annual cancer detection rates for a physician were grouped according to annual percentage recall rate. RESULTS. Statistically significantly higher cancer detection rates were seen for recall rates 12% or higher, with rate ratios of 1.75 (95% CI, ) and 2.06 (95% CI, ) for the recall groups 12% to less than 14% and 14% and higher, respectively, compared with the less than 10% group. When taking the category 12% to less than 14% as the reference, there were no statistically significant differences between recall groups 12% to less than 14% and 14% or higher in cancer detection rate. A statistically significant increase in the cancer detection rate with each unit increase in the recall rate was seen only for recall rates 12% or higher. CONCLUSION. These observations suggest that the sweet spot for optimal cancer detection is in the recall rate range 12% to less than 14% with the incremental benefit above this to be relatively small. A recall rate less than 10% may be too low. T he Mammography Quality Standards Act [1] was enacted in 1992 to improve the quality of screening mammography and outcomes for patients with breast cancer. One requirement of the legislation is that each facility conducts an annual medical outcome audit to evaluate its performance in detecting breast cancer. The American College of Radiology [2] recommends that all mammography facilities compare their audit data with established benchmarks [3 5]. Performance parameters include recommended recall rates (percentage of screening mammograms for which additional evaluation is required), cancer detection rates, percentage of minimally invasive cancers, percentage of stage 0 and 1 cancers, and percentage of node-negative cancers [2 5]. These performance parameters reflect the goals of screening mammography, which involve finding a high percentage of cancers, detecting early cancers, and practicing within an acceptable range for which additional imaging and biopsies are recommended. In particular, various investigators, using data from the Breast Cancer Screening Consortium, retrospectively identified target recall rates between 5% and 12% using the sensitivity of mammography screening and data from centralized breast cancer tumor registries [4 6]. These recall ranges were recommended on the basis of the trade-off between the number of additional women who needed to be recalled to detect one additional cancer. However, the use of sensitivity measures to set recall guidelines may be problematic because they do not take into account the size 208 AJR:208, January 2017

2 Optimal Recall and Cancer Detection Rates for Screening Mammography and stage of the breast cancer detected [4 6], which influence both prognosis and treatment. Minimally invasive cancers ( 1 cm with no lymph node involvement) not only have better prognoses, but they also may not require adjuvant chemotherapy as part of the treatment regimen [7]. Even though ductal carcinoma in situ (DCIS) typically has an excellent prognosis and does not require adjuvant chemotherapy, critics question whether all DCIS should be considered significant cancers, given the uncertainty of which ones will progress to invasive cancer [8]. Thus, an ideal recall rate would detect more minimally invasive cancers and fewer low-risk DCIS. It is critical to note that the data from which the Breast Cancer Screening Consortium recall recommendations were based are now years old and predate the near universality of digital mammography. Furthermore, the performance recommendations have focused on minimally acceptable performance criteria rather than guidelines for optimal performance [4]. In this study, we aim to identify the optimal recall rate range, the sweet spot, defined as a recall range for optimal breast cancer detection using screening digital mammography. Materials and Methods Study Data The study group for this institutional review board approved HIPAA-compliant research included 188,959 digital screening mammograms performed at a single Midwest academic institution and its satellite imaging sites. Informed consent was waived by the institutional review board of Northwestern Memorial Hospital. The mammograms were interpreted by 19 dedicated breast imaging specialists. Screening examinations were performed on women who were asymptomatic or had nonfocal breast pain, had no history of breast cancer, and did not have breast implants. Most of these women began screening at or after the age of 40 years, with some obtaining a baseline mammogram at approximately the age of 35 years. High-risk women were offered screening before age 40 years but not earlier than age 25 years. White women comprised 63% of women screened, African American women comprised 16%, Asian women comprised 3%, and 17% of women were of unknown race (race was self-identfied by the patients). The mammography equipment was all digital, with most mammograms obtained using a Selenia (Hologic) system and a small percentage obtained using a Senographe Care (GE Healthcare) system. Tomosynthesis was not used. The women screened represented a group undergoing periodic screening as part of Recall Rate (%) All Subtotal (I 2 = 79.3%, p = 0.008) DCIS Subtotal (I 2 = 74.8%, p = 0.019) Invasive Subtotal (I 2 = 46.3%, p = 0.155) Minimally Invasive Subtotal (I 2 = 0.0%, p = 0.505) their periodic health maintenance. Quality assurance data compiled for the Mammography Quality Standards Act annual medical outcome audits for the study period for all 19 designated breast imaging radiologists were entered into an Excel (version 2013, Microsoft) spreadsheet. Standard measures were generated. Recalls from screening were based on findings on a mammogram requiring additional imaging or comparison with earlier mammograms. Breast density alone, in the absence of a mammographic finding, was not an indication for recall. The annual recall rates were calculated for each of the radiologists as well as for the group Cancer Detection Rate Ratio (95% CI) 1.34 ( ) 1.75 ( ) 2.06 ( ) 1.64 ( ) 1.04 ( ) 1.73 ( ) 1.85 ( ) 1.39 ( ) 1.56 ( ) 1.77 ( ) 2.22 ( ) 1.80 ( ) 1.96 ( ) 2.44 ( ) 2.77 ( ) 2.32 ( ) Fig. 1 Rate ratios of different cancer detection rates by recall rate categories. Solid diamonds and horizontal lines denote rate ratios and their respective 95% CIs. Vertical line denotes reference rate ratio (1.00). Open diamonds denote subtotal rate ratio and 95% CIs. DCIS = ductal carcinoma in situ. Fig. 2 Plot of regression coefficient of cancer detection rates by unit change in recall rates. Diamonds denote means, and vertical lines and whiskers denote standard error. Linear Regression Coefficient for Cancer Detection Rate Recall Rate (%) There were 75 annual recall rates calculated for the 5-year study period for all the radiologists. The radiologists ranged in experience from 1 to 30 years after training. Of the 75 reader years, 56 were for radiologists who had a minimum of 5 years experience at the beginning of the study. Those with less than 5 years experience had all completed a fellowship in breast imaging. Each audit year went from September 1 to the following August 31. Our data include five complete audit years from September 1, 2007, through August 31, In addition, the annual cancer detection rates (number of breast cancers per 1000 mammo- AJR:208, January

3 Grabler et al. grams) for each of the radiologists and the group were obtained. Annual cancer detection rates were calculated for the overall number of cancers as well as for various subcategories of breast cancer, which included DCIS, all invasive cancers, minimally invasive cancers ( 1 cm without lymph node involvement) and early-stage invasive cancers ( 2 cm without lymph node involvement). The percentages of DCIS, minimally invasive cancers (DCIS plus invasive cancers 1 cm with no lymph node involvement), and early-stage cancers (DCIS plus invasive cancers 2 cm with no lymph node involvement) relative to the overall number of cancers were also generated [2]. Statistical Analysis The annual recall rates, the mean number of screening mammograms read annually, the cancer detection rate, the positive predictive value of an abnormal screening mammogram (PPV1), and the percentage of minimally invasive and early stage cancers were calculated for the entire group for the 5-year study period. The recall rates were evaluated in two ways. First, they were separated into recall groups to assess cancer detection rate variation by the recall categories. Second, changes in cancer detection rate with each unit increase in the recall rate were analyzed. In the first analysis, the 75 annual recall rates were divided into four groups: less than 10%, 10% to less than 12%, 12% to less than 14%, and 14% or higher. The group less than 10% was chosen because it is below the mean recall performance reported by the Breast Cancer Screening Consortium investigators [3, 5]. The second group (10% to < 12%) reflects recall rates above the mean Breast Cancer Screening Consortium rate but within performance range recommended by an expert panel based on the Breast Cancer Screening Consortium data [4]. The third group (12% to < 14%) includes recall rates that are above current recommended performance ranges [2 5] but lower than the highest historically published recall guideline [9]. Finally, the fourth group (%) comprises recall rates that are above all recommended performance criteria. The mean number of screening mammograms read annually was calculated for each group. The annual cancer detection rate for each physician was then sorted by recall rates according to the above groups. The cancer detection rate were then calculated separately for overall cancers, DCIS, invasive cancers, and minimally invasive cancers. The percentages of minimally invasive and early-stage cancers were also calculated according to the recall groups. There was relatively little variation in the year-to-year recall rates by reader. TABLE 1: Screening Mammography Characteristics Across Recall Rate Groups Characteristic Total To assess the statistical significance of the observations among the four recall rate groups, a Pearson chi-square test was used. The rate ratios (RRs) with 95% CIs were calculated for the different cancer detection rate according to the recall categories, with the lowest recall group (%) taken as the reference. In the second analysis, to estimate the coefficient of a unit change in the cancer detection rate according to the recall rates, a regression model controlled for heteroscedasticity with bootstrap bias-corrected and 95% CIs was used. In this regression model, different recall rate categories were generated. A cutoff at each 1.0 unit of the recall rate was generated, with the category 0 representing recall rates of less than 7%, the category 7 for recall rates of 7% to less than 8%, the category 8 for recall rates of 8% to less than 9%, and the category 9 for recall rates of 9% to less than 10%. Following the same rational, categories 10 (10% to < 11%), 11 (11% to < 12%), 12 (12% to < 13%), 13 (13% to < 14%), 14 (14% to < 15%), 15 (15% to < 16%), 16 (16% to < 17%), 17 (17% to < 18%), 18 (18% to < 19%), 19 (19% to < 20%), and 20 ( 20%) were generated. There were no data in the range 17% to less than 18%; therefore, no category was generated for this range. All statistical analyses were conducted using STATA software (release 14, StataCorp). Recall Rate Group % 10% to < 12% 12% to < 14% % Total no. of screening mammograms 188,959 49,534 43,598 25,858 69,969 No. of recalled patients 25, ,209 No. of cancers detected DCIS Invasive cancers No. of minimally invasive cancers (< 1 cm) No. of minimally invasive cancers (DCIS and invasive cancers < 1 cm) No. of early-stage cancers (DCIS and invasive cancers < 2 cm) Rates Overall detection rate, no. of cancers/1000 mammograms PPV DCIS cancer detection rate, no. of cancers/1000 mammograms Invasive cancer detection rate, no. of cancers/1000 mammograms Minimally invasive cancer detection rate, no. of cancers/1000 mammograms Percentages of cancer types Minimally invasive cancer (DCIS and invasive cancers < 1 cm) Early-stage cancers (DCIS and invasive cancers < 2 cm) DCIS Recall rate (%) Note DCIS = ductal carcinoma in situ, PPV1 = positive predictive value of an abnormal screening mammogram. 210 AJR:208, January 2017

4 Optimal Recall and Cancer Detection Rates for Screening Mammography Results During the study period, the total number of screening mammograms performed and interpreted by 19 breast imaging specialists was 188,959. The mean annual number of screening mammograms read per radiologist was The mean recall rate was 13.3% (range, %). There were 984 breast cancers detected during this period. The mean cancer detection rate was 5.2 cancers per 1000 screening mammograms (range, 3 10 cancers/1000 mammograms for the individual radiologists). The PPV1 of an abnormal screening mammogram was 3.9%. Among cancers, 65.8% (647/984) were minimally invasive, 81.9% (806/984) were early stage, and 39.2% (386/984) were DCIS. The highest cancer detection rates were seen in the high recall rate groups (6 cancers/1000 screening mammograms for the 12% to < 14% group and 6.8 cancers/1000 screening mammograms for the % group) as compared with the low recall groups (3.3 cancers/1000 screening mammograms for the < 10% group and 4.4 cancers/1000 screening mammograms for the 10% to < 12% group). Similarly, the high recall groups had the highest rate of DCIS, minimally invasive cancers, and invasive cancers as compared with the low recall groups. The data on cancer detection rates and cancers are presented in Table 1. Figure 1 displays the RR for the cancer detection rates with the lowest recall rate group (%) as the reference. The recall groups 12% to less than 14% and 14% and higher had the highest relative cancer detection rates. Specifically, the recall rate group 12% to less than 14% detected 75% more cancers than the lowest category (%) (RR, 1.75; 95% CI, ) and the highest recall group (%) detected more than twice the cancers than the lowest recall group (%) (RR, 2.06; 95% CI, ). Similarly, for DCIS detection rate, the RRs were 1.73 (95% CI, ) for the 12% to less than 14% group and 1.85 (95% CI, ) for the 14% and higher group. With regard to minimally invasive cancer detection rates, for those with recall rates in the 12% to less than 14% groups, the RRs were 2.4 times more likely to detect minimally invasive cancers than those with recall rates less than 10% (RR, 2.44; 95% CI, ). Those with recall rates 14% or higher were 2.77 times more likely to detect minimally invasive cancer (RR, 2.77; 95% CI, ) than those in the lowest recall group. When taking the category 12% to less than 14% as the reference, there were no statistically significant differences between recall groups 12% to less than 14% and 14% and higher in cancer detection rate. To estimate the effect of each unit increase in recall rate on the cancer detection rate, a linear regression model was used. Taking the lowest recall rate in our data as the reference (< 7%), we found that recall rates above this level increasingly detect cancer (Fig. 2). However, only the recall rates falling within the high recall category ( 12%) were statistically significantly more likely to detect cancer than the less than 7% recall category. The highest coefficient for an increase in overall cancer detection was seen in category 14 (14% to < 15%), with a coefficient of 4.61 (95% CI, ) (Table 2). A statistically significant increase in the cancer detection rate for invasive breast cancer was also observed in recall rate category 14, as well (coefficient, 3.13; 95% CI, ) (Table 3). With regard to the cancer detection rate for DCIS, the increase was smaller and less stable, with the highest coefficient for cancer detection rate seen in recall rate category 18 (2.52; 95% CI, ) (Table 3). These data are presented in Table 2. Discussion In our academic practice, the cancer detection rate (5.2 cancers/1000 mammograms), the percentage of minimally invasive cancers (65.8%), and the proportion of early-stage cancers (81.9%) were, on average, higher than the numbers published by Rosenberg et al. [3] on the basis of Breast Cancer Screening Consortium data (cancer detection rate, 4.7 cancers/1000 mammograms; 51.7% minimally invasive cancers; 75% early-stage cancers) [2]. Radiologists in our practice with a recall rate below 10% found significantly fewer breast cancers of all types when compared with those in our practice with recall rates above 12%. This suggests that the mean recall recommendation by the Breast Cancer Screening Consortium (10%) may be too low. In contrast, those breast imagers in our practice operating at recall rates 12% or higher found significantly more cancers of all types compared with those operating within the minimally acceptable 5 12% recall performance range recommended by other experts based on Breast Cancer Screening Consortium data [4]. Looking at those breast imagers with recall rates above the Breast Cancer Screening Consortium experts recommended upper performance recall limit of 12% [4], the linear regression coefficients in cancer detection rate between breast imagers with recall rates of 12% to less than 14% and those with recall rates 14% or higher were similar. This suggests that the incremental benefit of a recall rate higher than 12% to less than 14% was limited when it comes to detecting cancer. Critics have speculated that higher recall and cancer detection rates may be due to the detection of insignificant cancers (e.g., some TABLE 2: Regression Coefficient of Each Unit Increase in Recall Rate According to Cancer Detection Rate Recall Rate Category (%) a Coefficient (95% CI) Standard Error p 7 to < ( 0.83 to 0.49) to < ( 0.56 to 2.17) to 0.59 ( 0.36 to 1.53) to < ( ) to < ( 0.63 to 2.26) to < ( ) to < ( ) to < ( ) to < ( ) to < ( ) to < to < ( ) to < ( ) ( ) Note Dashes ( ) indicate that there are no data. a The reference is < 7%. AJR:208, January

5 Grabler et al. TABLE 3: Regression Coefficient of Each Unit Increase in the Recall Rates According to the Cancer Detection Rates for Ductal Carcinoma In Situ (DCIS) and Invasive Cancers Recall Rate Category (%) a Cancer Detection Rate for DCIS Cancer Detection Rate for Invasive Cancers Coefficient (95% CI) p Coefficient (95% CI) p 7 to < ( 1.03 to 0.14) ( ) to < ( 0.85 to 0.80) ( ) to 0.26 ( 1.30 to 0.78) ( ) to < ( 0.64 to 0.30) ( ) to < ( 1.21 to 1.14) ( ) to < ( ) ( ) to < ( ) ( ) to < ( ) ( ) to < ( 1.77 to 1.38) ( ) to < ( ) ( ) to < to < ( ) ( ) to < ( 1.60 to 4.15) ( ) ( ) ( ) 0.00 Note Dashes ( ) indicate that there are no data. a The reference is < 7%. DCIS that may never progress to invasive cancer) [7]. This study found no differences in the percentage of DCIS relative to the total number of cancers detected for each recall subgroup. In addition, compared with the lowest recall category, the higher recall groups found significantly more invasive cancers and, specifically, significantly more minimally invasive cancers, which offer women the best prognosis for long-term disease-free survival. The fact that there is no significant increase in the DCIS cancer detection rate at any point of recall in the low recall rate category (< 12%) and just a marginal increase in the DCIS cancer detection rate up to a recall rate of 14% reinforces the clinical importance of this finding of higher invasive cancer detection rate at higher recall rates. Gur et al. [10], who correlated recall and cancer detection rates in an academic breast imaging practice in 2004, emphasized the importance of optimizing practice parameters to maximize early detection. Their mean recall rate of 11% is in contrast to the 13% mean recall rate in our study. This may reflect the sole use of digital mammography in our practice [10]. Although critics may raise concerns regarding additional workups and false-positive results from higher recall rates, studies of women s preferences show that many women prefer the anxiety of a higher recall rate if it results in a higher breast cancer detection rate [11]. How do we explain the difference in our findings and the current external performance parameters? Contrary to this study, the current performance parameters for mammography recall and cancer detection rate are based on data from the Breast Cancer Screening Consortium. Although it represents a large number of community radiologists and large number of follow-up examinations, the Breast Cancer Screening Consortium has the following four limitations: first, the data come from heterogeneous practices that predate the nearly universal digital mammography era; second, the published performance standards represent average-to-minimal performance standards rather than optimal performance standards; third, the Breast Cancer Screening Consortium performance standards represent the trade-off between finding breast cancers and eliminating unnecessary imaging workups based on sensitivity and specificity data, which are not readily available to the average breast imager; and fourth, using sensitivity and specificity data to determine performance standards does not take into account the size of the breast cancers detected. What is the sweet spot for optimal mammography recall rates? On the basis of our findings, a recall rate range of 12% to less than 14% appears optimal. The incremental benefit of higher recall rates is relatively small. A recall rate of less than 10% seems to be too low. Consistent with national data, we experienced wide variability in recall and cancer detection rates [12, 13]. Of interest are the factors that determine the recall rates of any breast imager. A small pilot study at our institution that looked at differences in recall rates for calcifications versus other mammographic findings did not show significant differences between the high and low recall groups. A larger and more extensive study of the characteristics of the recall findings may shed more light on recall variability. One limitation of this study is that it represents one academic practice, which may affect the generalizability of the findings. Although there were 75 radiologist years and 188,959 screening mammograms determining the results, these findings could be strengthened by comparisons across a number of similar expert practices. In addition, the number of baseline mammograms and number of mammograms with no comparisons, factors that could increase recall rate, were unknown but were thought to be a small proportion of overall screening mammograms. Finally, although this study provides a statistical assessment of optimal recall rates as they relate to cancer detection rate in the age of digital mammography, this is unlikely to be the last word on the subject. Early data from tomosynthesis research suggest that this newer modality can maximize cancer detection rate and significantly lower mammography recall rates [14]. Although promising, its influence on current national recommendations will not be realized until this modality receives widespread acceptance and its findings are validated. Until then, we recommend raising the acceptable recall rate to 14% to optimize breast cancer detection and better reflect expert practice. Understanding that increasing recall rates does not guarantee improved cancer detection rates, breast imagers must compare their own practice numbers with these reference standards to assess the effect. Our observations from this expert breast imaging practice suggest that the sweet spot for optimal cancer detection is in the recall rate range 12% to less than 14%. The incremental benefit above this range is relatively small. A recall rate less than 10% may be too low. 212 AJR:208, January 2017

6 Optimal Recall and Cancer Detection Rates for Screening Mammography Acknowledgments We thank all the radiologists whose work contributed to this data, particularly Steven Whitman (deceased), for his guidance throughout this research project. References 1. Monsees BS. The Mammography Quality Standards Act: an overview of the regulations and guidance. Radiol Clin North Am 2000; 38: D Orsi CJ, Sickles EA, Mendelson EB, et al. ACR BI-RADS Atlas, Breast Imaging Reporting and Data System, 5th ed. Reston, VA: American College of Radiology, Rosenberg RD, Yankaskas BC, Abraham LA, et al. Performance benchmarks for screening mammography. Radiology 2006; 241: Carney PA, Sickles EA, Monsees BS, et al. Identifying minimally acceptable interpretive performance criteria for screening mammography. Radiology 2010; 255: Breast Cancer Surveillance Consortium (BCSC). Abnormal interpretations for 2,016,691 screening mammography examinations from : based on BCSC data through 2009 ( HHSN C). BCSC website. breastscreening.cancer.gov/statistics/ benchmarks/screening/2009/table3.html. Updated September 26, Accessed March 20, Schell MJ, Yankaskas BC, Ballard-Barbash R, et al. Evidence-based target recall rates for screening mammography. Radiology 2007; 243: National Comprehensive Cancer Network (NCCN). NCCN clinical practice guidelines in oncology (NCCN guidelines): breast cancer, version NCCN website. physician_gls/pdf/breast.pdf. Published April 1, Accessed February 6, Allegra CJ, Aberle DR, Ganschow P, et al. NIH state-of-the-science conference statement: diagnosis and management of ductal carcinoma in situ (DCIS). NIH Consens State Sci Statements 2009; 26: Agency for Healthcare Research and Quality (AHRQ). Imaging efficiency: percentage of patients with mammography screening studies that are followed by a diagnostic mammography, ultrasound or MRI of the breast in an outpatient or office setting within 45 days. AHRQ website. www. qualitymeasures.ahrq.gov/content.aspx?id= Published August 29, Updated May 7, Accessed February 6, Gur D, Sumkin JH, Hardesty LA, et al. Recall and detection rates in screening mammography. Cancer 2004; 100: Ganott MA, Sumkin JH, King JI, et al. Screening mammography: do women prefer a higher recall rate given the possibility of earlier detection of cancer? Radiology 2006; 238: Elmore JG, Jackson SL, Abraham L, et al. Variability in interpretive performance at screening mammography and radiologists characteristics associated with accuracy. Radiology 2009; 253: Miglioretti DL, Gard CC, Carney PA, et al. When radiologists perform best: the learning curve in screening mammogram interpretation. Radiology 2009; 253: Friedewald SM, Rafferty EA, Rose SL, et al. Breast cancer screening using tomosynthesis in combination with digital mammography. JAMA 2014; 311: FOR YOUR INFORMATION This article is available for CME and Self-Assessment (SA-CME) credit that satisfies Part II requirements for maintenance of certification (MOC). To access the examination for this article, follow the prompts associated with the online version of the article. AJR:208, January