PMCCPMCCPMCC

Search tips
Search criteria 

Advanced

 
Logo of nihpaAbout Author manuscriptsSubmit a manuscriptHHS Public Access; Author Manuscript; Accepted for publication in peer reviewed journal;
 
Ann Intern Med. Author manuscript; available in PMC 2013 July 30.
Published in final edited form as:
PMCID: PMC3726800
NIHMSID: NIHMS492755

Comparative Effectiveness of Digital Versus Film-Screen Mammography in Community Practice in the United States

A Cohort Study
Karla Kerlikowske, MD, Rebecca A. Hubbard, PhD, Diana L. Miglioretti, PhD, Berta M. Geller, EdD, Bonnie C. Yankaskas, PhD, Constance D. Lehman, MD, PhD, Stephen H. Taplin, MD, MPH, and Edward A. Sickles, MD, for the Breast Cancer Surveillance Consortium

Abstract

Background

Few studies have examined the comparative effectiveness of digital versus film-screen mammography in U.S. community practice.

Objective

To determine whether the interpretive performance of digital and film-screen mammography differs.

Design

Prospective cohort study.

Setting

Mammography facilities in the Breast Cancer Surveillance Consortium.

Participants

329 261 women aged 40 to 79 years underwent 869 286 mammograms (231 034 digital; 638 252 film-screen).

Measurements

Invasive cancer or ductal carcinoma in situ diagnosed within 12 months of a digital or film-screen examination and calculation of mammography sensitivity, specificity, cancer detection rates, and tumor outcomes.

Results

Overall, cancer detection rates and tumor characteristics were similar for digital and film-screen mammography, but the sensitivity and specificity of each modality varied by age, tumor characteristics, breast density, and menopausal status. Compared with film-screen mammography, the sensitivity of digital mammography was significantly higher for women aged 60 to 69 years (89.9% vs. 83.0%; P = 0.014) and those with estrogen receptor-negative cancer (78.5% vs. 65.8%; P = 0.016); borderline significantly higher for women aged 40 to 49 years (82.4% vs. 75.6%; P = 0.071), those with extremely dense breasts (83.6% vs. 68.1%; P= 0.051), and pre- or perimenopausal women (87.1% vs. 81.7%; P = 0.057); and borderline significantly lower for women aged 50 to 59 years (80.5% vs. 85.1%; P = 0.097). The specificity of digital and film-screen mammography was similar by decade of age, except for women aged 40 to 49 years (88.0% vs. 89.7%; P< 0.001).

Limitation

Statistical power for subgroup analyses was limited.

Conclusion

Overall, cancer detection with digital or film-screen mammography is similar in U.S. women aged 50 to 79 years undergoing screening mammography. Women aged 40 to 49 years are more likely to have extremely dense breasts and estrogen receptor-negative tumors; if they are offered mammography screening, they may choose to undergo digital mammography to optimize cancer detection.

Primary Funding Source

National Cancer Institute.

Of the 12 445 accredited mammography machines in the United States as of 1 October 2010, 8748 (70.3%) are full-field digital (1). Despite the rapid dispersion of full-field digital mammography, few studies on the accuracy of digital mammography in the United States have been published (24), and no studies have compared this technology with film-screen mammography in U.S. community practice.

Studies comparing digital with film-screen mammography in Europe and the United States have produced conflicting findings (5). DMIST (Digital Mammography Imaging Screening Trial) performed film-screen and digital mammography in asymptomatic U.S. women at the same screening encounter. It found that overall accuracy of film-screen and digital mammography for breast cancer detection was similar (2) but that digital mammography was more accurate in pre- or perimenopausal women younger than 50 years with mammographically dense breasts and less accurate in women aged 65 years or older with non-dense breasts (3). The Oslo II study randomly assigned women aged 45 to 69 years to undergo digital or film-screen mammography and reported higher cancer detection rates and lower specificity with digital than with film-screen mammography (6). In a population-based screening program in Spain, recall rate was higher among women undergoing digital mammography than film-screen mammography, and cancer detection rates were similar (7). The United Kingdom’s breast cancer screening program for women aged 50 years or older found no difference in cancer detection or recall rates (8). In a population-based screening program in the Netherlands, recall rate and detection rates for ductal carcinoma in situ (DCIS) were higher among women undergoing digital mammography than those having film-screen mammography, but detection rates for invasive cancer were similar (9).

The inconsistent results across studies may be due to small numbers of digital examinations and, thus, few cases of breast cancer associated with these examinations (n = 25 to 254). In addition, the studies did not account for correlation among mammography examinations performed at the same facility by the same radiologist or for secular trends in mammography performance. Study design also varies considerably, ranging from randomized, controlled trials to paired examinations, retrospective cohorts, and population-based cohorts. Other factors that may contribute to the divergent results include single- versus double-reading, readers’ experience, and practice environment (5).

We sought to compare the accuracy of digital mammography with that of film-screen mammography and tumor outcomes according to age, breast density, menopause status, and tumor subtype at diagnosis among women aged 40 to 79 years undergoing screening mammography in the Breast Cancer Surveillance Consortium (BCSC). The BCSC (http://breastscreening.cancer.gov) is a large population-based cohort of community-based imaging facilities in the United States.

Methods

Data Source

Data were pooled from 4 mammography registries that participate in the BCSC (10) and collect data from at least 1 facility that performs digital mammography: San Francisco Mammography Registry, Vermont Breast Cancer Surveillance System, New Hampshire Mammography Network, and Carolina Mammography Registry. These registries collect information on breast imaging examinations performed in their defined catchment areas. Each breast-imaging registry annually links women in its registry to a state tumor registry or regional Surveillance, Epidemiology and End Results program that collects population-based cancer data. Three of the 4 registries also link with pathology databases. Each registry obtains annual approval from its institutional review board for consenting processes or a waiver of consent, enrollment of participants, and ongoing data linkages for research purposes. The BCSC Statistical Coordinating Center and each registry adhere to strict confidentiality procedures; comply with the Health Insurance Portability and Accountability Act; and have a Federal Certificate of Confidentiality and other protections of research subjects, radiologists, and mammography facilities.

Participants

The study sample included bilateral digital and film-screen mammography examinations performed between 1 January 2000 and 31 December 2006 among women aged 40 to 79 years who did not have a history of breast cancer or breast implants. Mammography examinations that occurred after 31 December 2006 were not included to ensure at least 12 months for reporting cases of cancer to tumor registries. Cancer ascertainment from cancer registries is estimated to be more than 94.3% complete (11).

Measurements and Definitions

Demographic characteristics and breast health history were obtained by using a self-administered questionnaire (available at http://breastscreening.cancer.gov) that was completed at each screening examination. Women were considered to have a family history of breast cancer if they reported having at least 1 first-degree relative (mother, sister, or daughter) with breast cancer. Postmenopausal women were defined as those who had had both ovaries removed, those who reported that their periods had stopped naturally, those currently using hormone therapy, and those aged 55 years or older. Women were considered to be premenopausal if their menstrual periods had not stopped and perimenopausal if they were not sure whether their periods had stopped. Women were considered to have missing menopausal status if they had had a hysterectomy without bilateral oophorectomy and were not using hormone therapy or if their menopause status could not be determined from available information.

We used self-reported race and ethnicity to categorize women as non-Hispanic white, non-Hispanic black, Hispanic, Asian/Native Hawaiian/Pacific Islander, Native American/Native Alaskan, or other/mixed race.

Time between mammography examinations was determined by using the dates of prior mammography examinations recorded in each mammography registry (87%); if these were not available, we used self-reported information (13%) collected at the screening examination. A mammography examination was determined to be the first if a woman reported no previous examination and no prior mammogram was found in a registry.

Mammographic breast density was assigned in clinical practice by a radiologist at the time of mammography interpretation. The Breast Imaging Reporting and Data System (BI-RADS) density categories used were almost entirely fat, scattered fibroglandular densities, heterogeneously dense, and extremely dense (12).

Our primary measure was the initial screening mammography assessment, which we categorized as positive or negative by using standard BI-RADS and BCSC definitions (12, 13) that indicate whether a woman was recalled to undergo additional evaluation on the basis of screening views only (12, 13), and the association of initial assessment with cancer outcomes. Standard BCSC definitions of true-positive, false-positive, true-negative, and false-negative results were used to calculate breast cancer and recall rates and the sensitivity, specificity, and positive predictive value of mammography (13).

A mammography examination was associated with breast cancer if invasive carcinoma or DCIS was diagnosed within 12 months of and before the next screening examination. Women with lobular carcinoma in situ only were not considered to have cancer. Stage at diagnosis was classified according to the tumor, lymph node, metastasis system based on the criteria of the American Joint Committee on Cancer, 6th edition, as stage 0, I, IIA, IIB, III, or IV (14). Invasive cancers were classified according to their estrogen receptor status, lymph node status, tumor size, and grade.

Statistical Analysis

All analyses were performed by using the screening examination as the unit of analysis; women may have had more than 1 mammography examination during the study period. Frequency distributions of risk factors for breast cancer and BI-RADS density scores and assessments were determined for digital and film-screen examinations.

We modeled mammography performance measures by using binomial generalized linear mixed models with a logit link, including normally distributed facility random effects to account for correlation among mammography examinations performed at the same facility. To account for differences between facilities performing digital mammography and those performing only film-screen mammography, we included a binary indicator of whether a facility performed any digital examinations during the study period. All models were adjusted for factors related to performance and timing of digital examinations: age, examination year, time between screenings (within 1, 2, or ≥3 years, or first screening examination), and BCSC registry. Adjusted performance measures were estimated from these models by using indirect standardization to ensure identical distributions of covariates among digital and film-screen examinations (15, 16), as described elsewhere (17, 18). Performance estimates were calculated for a facility at the median of the distribution of facility random effects. Standard errors for adjusted performance measures were calculated by using the delta method.

Separate performance measures were calculated and reported for digital and film-screen mammography from facilities that either switched to digital (66% of facilities) during the study period or performed both film and digital mammography. Film-screen examinations (n = 994 000) from facilities that did not perform digital mammography during the study period were also included in all analyses to adjust for possible secular trends in mammography performance.

Two-sided statistical tests resulting in P values less than 0.050 were considered statistically significant, and values between 0.050 and 0.100 were considered borderline significant.

Sensitivity analyses were done to evaluate the effect of correlation among mammography examinations interpreted by the same radiologist by including a random effect for radiologist in addition to facility (Appendix, available at www.annals.org). Allowing for within-radiologist correlation among mammography examinations had no qualitative effect on regression model results; this factor was therefore not included in the main model, as it reduced our sample size (because a radiologist identifier was missing for some examinations).

We also conducted a sensitivity analysis of the effect of adjustment for the proportion of digital examinations performed at a facility (Appendix). In this analysis, we decomposed the effect of type of mammography into between-and within-facility components by using the methods of Neuhaus and Kalbfleisch (19); this had no qualitative effect on results and was therefore not included in the main model. The Appendix shows performance estimates for facilities at the 25th and 75th percentiles of the random effects distribution to quantify variability across facilities.

To assess whether examinations performed in the first year (n= 60 383 [26.1%]) after a facility implemented digital mammography affected performance estimates, we excluded these examinations in additional sensitivity analyses.

Role of the Funding Source

This study was funded by the National Cancer Institute. A senior scientist from the National Cancer Institute participated in the study design and preparation of the manuscript. The views expressed in this article do not represent those of the National Cancer Institute, and this organization had no role in the final decision to submit the manuscript for publication.

Results

Among 329 260 women aged 40 to 79 years, 869 286 screening mammography examinations were performed (231 034 digital and 638 252 film-screen) and breast cancer was diagnosed in 4046 women (1054 digital and 2992 film-screen examinations), primarily at nonacademic facilities (83%). Women undergoing digital compared with film-screen mammography were similar in age, race or ethnicity, and time since last mammography examination, and they were equally likely to have a first-degree relative with breast cancer and dense breasts (Table 1). Women undergoing digital mammography were slightly more likely to have an initial BI-RADS assessment that indicated a need for additional imaging evaluation. The median time that facilities had performed digital mammography was 2.3 years.

Table 1
Characteristics of Women Undergoing Digital Versus Film-Screen Mammography

Performance Measures for Digital Versus Film-Screen Mammography

Overall

For the most part, digital and film-screen mammography were similar in rates of breast cancer per 1000 examinations (overall, invasive cancer, and DCIS), cancer detection, false-negative results, breast biopsies, and sensitivity. Recall rate and specificity differed significantly, but the differences were small (Table 2). Excluding digital examinations from the first year during which a facility performed digital mammography did not change the significance of any of our comparisons of the performance of digital and film-screen mammography (data not shown).

Table 2
Performance Measures of Screening Mammography Among Women Undergoing 231 034 Digital Versus 638 252 Film-Screen Examinations*

By Age, BI-RADS Breast Density, and Menopausal Status

Rates of cancer detection per 1000 examinations by decade of age were similar for digital and film-screen mammography (Table 3). Sensitivity to detect invasive cancer or DCIS was borderline significantly higher for digital than for film-screen mammography among women aged 40 to 49 years (82.4% vs. 75.6%; P = 0.071), significantly higher among women aged 60 to 69 years (89.9% vs. 83.0%; P = 0.014), and borderline significantly lower for women aged 50 to 59 years (80.5% vs. 85.1%; P = 0.097). Specificity was similar for digital and film-screen mammography for all decades of age, except women aged 40 to 49 years (88.0% vs. 89.7%; P< 0.001).

Table 3
Performance Measures of Screening Mammography Among Women Undergoing Digital Versus Film-Screen Examinations, by Age, BI-RADS Breast Density, and Menopausal Status*

Rates of cancer detection per 1000 examinations across breast density categories were similar for digital and film-screen mammography (Table 3). Sensitivity was similar for digital and film-screen mammography for all density categories except extremely dense breasts, for which sensitivity was borderline significantly higher for digital than for film-screen mammography (83.6% vs. 68.1%; P = 0.051). Specificity was significantly lower for digital than for film-screen mammography (range, 0.6% to 1%) for all breast density categories (P ≤ 0.010). Sensitivity was borderline significantly higher for digital than for film-screen mammography in pre- or perimenopausal women (87.1% vs. 81.7%; P = 0.057) and specificity was lower (88.7% vs. 90.2%, P< 0.001) (Table 3).

We examined the subgroups in which DMIST (3) reported that the sensitivity of digital mammography differed from that of film-screen mammography. In pre- or perimenopausal women aged 40 to 49 years with extremely dense breasts, among whom there were 66 cases of breast cancer, a nonsignificant trend was seen toward higher sensitivity for digital than for film-screen mammography (86.8% vs. 62.3%; odds ratio, 4.1 [95% CI, 0.7 to 23.3]; P = 0.111). For women aged 65 to 79 years with fatty breasts, among whom there were 48 cases of breast cancer, the sensitivity of digital versus film-screening mammography was similar (83.7% vs. 89.1%; odds ratio, 0.6 [CI, 0.1 to 7.3]; P = 0.69).

By Estrogen Receptor Status

Sensitivity was significantly higher for digital than for film-screen mammography among women aged 40 to 79 years who had estrogen receptor–negative cancer (78.5% vs. 65.8%; P = 0.016), was higher for all decades of age, and was significantly higher for women aged 40 to 49 years (95.2% vs. 54.9%; P = 0.007) (Table 4). Sensitivity was similar for women aged 40 to 79 years who had estrogen receptor–positive cancer (83.5% vs. 82.7%; P = 0.66) and for all decades of age except 60 to 69 years, for which sensitivity was higher for digital than for film-screen mammography (90.6% vs. 82.3%; P = 0.017) (Table 4).

Table 4
Sensitivity of Screening Mammography Among Women Undergoing Digital Versus Film-Screen Examinations, by Age and Estrogen Receptor Status

Tumor Characteristics, by Type of Mammography

The distribution of types of breast cancer by tumor characteristics did not vary for digital and film-screen mammography (Table 5). The proportion of early stage (I and IIA) versus advanced (IIB, III, IV) disease also did not vary by type of mammography (P = 0.168).

Table 5
Characteristics of Invasive Breast Cancer Among Women Undergoing Digital Versus Film-Screen Mammography*

Discussion

Digital mammography has diffused rapidly in countries in which screening mammography is available, despite limited data on the comparative effectiveness relative to film-screen mammography. We compared digital and film-screen mammography in a large number of U.S. community mammography facilities during a period when cancer detection using film-screen mammography has improved (20) and digital mammography technology has advanced. Overall, we found digital and film-screen mammography to be similar in cancer detection rates and the proportion of cancer cases diagnosed at an early stage. Specificity was similar for all decades of age except women 40 to 49 years, for whom specificity was lower for digital than film-screen mammography. In addition, we found that women with extremely dense breasts benefit from the higher sensitivity of digital than film-screen mammography, and we provide new evidence that digital mammography is better at detecting estrogen receptor–negative breast cancer, particularly in women aged 40 to 49 years.

Breast cancer may not be detected on mammography if a radiologist does not identify a visible lesion or a tumor is obscured by normal breast tissue. In addition, an imperceptible tumor may grow quickly and be discovered clinically before the next screening examination. High mammographic breast density is associated with decreased cancer detection on mammography (18, 21), in part because cancerous and normal fibroglandular tissue have similar radiographic attenuation. Breast tumors that are not detected by film-screen mammography tend to be estrogen receptor–negative, high-grade, and large and have high mitotic activity in women with dense tissue patterns (22, 23).

Digital mammography was developed in part to improve the detection of breast cancer in dense breasts by improving the ability to distinguish normal dense breast tissue from isodense invasive cancer. We would therefore expect digital mammography to improve cancer detection in women who have dense breast tissue and those who present with fast-growing invasive cancer in which the tumor is difficult to discern from normal dense tissue. Our results support this supposition: We found that digital mammography had higher sensitivity than film-screen mammography in women with dense breasts and women with estrogen receptor–negative tumors. Thus, women aged 40 to 49 years may benefit most from digital mammography because the proportion of women in this group with extremely dense breasts (about 12% to 15%) (24) and estrogen receptor–negative tumors (25) is higher than that of women aged 50 years or older; however, they may experience additional harms. If 10 000 women aged 40 to 49 years are screened with digital mammography, 2 additional cases of cancer will be identified for every 170 additional false-positive examinations.

DMIST identified subgroups of women in whom digital mammography may perform better than film-screen mammography (for example, those younger than 50 years, those with radiographically dense breasts, and pre- or peri-menopausal women) (2). The main conclusions of DMIST are based on the area under the receiver-operating characteristic curve, constructed from a 7-point mammography assessment scale. These curves estimated the sensitivity and specificity associated with classifying mammography examinations as normal or abnormal at each level of the assessment scale. This comparison of the performance of digital and film-screen mammography, which is based on an interpretive approach not used in clinical practice, magnifies the difference between modalities and tends to have greater statistical significance that stems from comparing sensitivity and specificity across all possible cut points rather than for an overall mammography result of normal or abnormal. When the DMIST investigators compared the overall sensitivity and specificity of digital versus film-screen mammography by using the 7-point scale, they did not find a difference in sensitivity for women of all ages combined, pre- or perimenopausal women, or those with dense breasts and borderline higher sensitivity and slightly lower specificity for women younger than 50 years (2), similar to the results we report for women aged 40 to 49 years. Although our estimates of the sensitivity of digital compared with film-screen mammography tend to be higher for pre- or perimenopausal women aged 40 to 49 years with extremely dense breasts and lower for women aged 65 to 79 years with almost entirely fat breast density, similar to data reported in DMIST (2), the differences were not statistically significant, possibly because of the small number of cases of breast cancer in these subgroups and the somewhat different age and breast density groups than those in DMIST.

We classified mammography examinations as normal or abnormal on the basis of initial BI-RADS assessments as collected in clinical practice because the study goal was to evaluate the influence of screening modality. The BI-RADS assessment does not provide a reliable basis for estimating receiver-operating characteristic curves in screening mammography because the BI-RADS scale is not ordinal, but rather a dichotomy of normal or abnormal assessment (26). Our results on the performance of digital and film-screen mammography in community practice are clinically relevant because we report overall sensitivity and specificity based on the main decision by radiologists during interpretation of screening mammography of whether to recall a woman for further diagnostic evaluation. Diagnostic evaluations are typically performed on a different day, which creates anxiety during the waiting period for some women. Diagnostic evaluations also contribute to health care costs and additional radiation exposure and discomfort. False-positive recalls based on the initial BI-RADS assessment are the most common “harm” of mammography and are thus an important and clinically relevant outcome to measure.

Our study included a large, diverse population-based sample and large number of outcomes. We took into account secular trends in mammography performance and adjusted for facility-level differences by including a facility-specific random effect. Although more than 1000 cases of breast cancer were identified among women undergoing digital mammography, we did not have the statistical power to examine subgroups of women with multiple risk factors that may influence breast cancer detection. Misclassification of BI-RADS density because of modest interrater agreement between radiologists (27, 28) could result in under- or overestimation of performance measures by density category. We evaluated numerous comparisons; some may be significant by chance alone.

We found small differences in the proportion of women with abnormal results by screening modality and did not find differences in tumor characteristics or rate of detection of invasive cancer or DCIS. Of note, we found that the sensitivity of digital and film-screen mammography to detect breast cancer is similar and relatively high among women aged 50 to 79 years. Women who have access to only digital or only film-screen mammography should be encouraged by our results, because both modalities seem to be equally effective in detecting breast cancer, in particular among women aged 50 to 79 years.

An important factor that we did not address in our study is the quality of mammography interpretation. Whether mammography examinations are interpreted at large-volume facilities (29) or by radiologists who are experienced in mammography interpretation (30) may influence the accuracy of mammography at least as much as whether the screening modality is digital or film-screen.

In summary, overall, cancer detection with digital and film-screen mammography is similar in U.S. women aged 50 to 79 years undergoing screening mammography. Women aged 40 to 49 years who are being offered screening mammography may choose to undergo digital mammography to optimize cancer detection, because digital mammography is better at detecting tumors in extremely dense breasts and estrogen receptor–negative tumors, both of which are more likely in this age group.

Appendix Table 2
Performance Measures of Screening Mammography Among Women Undergoing Digital Versus Film-Screen Examinations, by Age, BI-RADS Breast Density, and Menopausal Status, Allowing for Facility and Radiologist Random Effects*
Appendix Table 5
Performance Measures of Screening Mammography Among Women Undergoing Digital Versus Film-Screen Examinations, by Age, BI-RADS Breast Density, and Menopausal Status, Allowing for Between- and Within-Facility Effects*
Appendix Table 8
Performance Measures of Screening Mammography Among Women Undergoing Digital Versus Film-Screen Examinations, by Age, BI-RADS Breast Density, and Menopausal Status, for Facilities at the 25th and 75th Percentile of Performance

Context

Digital mammography is widely adopted despite limited evidence comparing its accuracy with that of film-screen methods.

Contribution

In a large sample of women screened in community settings, digital and film-screen mammography yielded similar cancers detection rates and proportions of early-stage cancer diagnosed. Digital screening had higher sensitivity in women with dense breasts and was better at detecting estrogen receptor–negative cancer, but specificity was lower for women aged 40 to 49 years than for other decades.

Caution

Some subgroups were small, and the study did not examine breast cancer mortality rates.

Implication

Screening methods are similarly effective, but sensitivity and specificity tradeoffs occur in some subgroups.

The Editors

Acknowledgment

The authors thank the participating women, mammography facilities, and radiologists for the data they have provided for this study. A list of BCSC investigators and procedures for requesting BCSC data for research purposes can be found at http://breastscreening.cancer.gov/.

Grant Support: By the National Cancer Institute–funded BCSC cooperative agreement (grants U01CA63740, U01CA86076, U01CA86082, U01CA63736, U01CA70013, U01CA69976, U01CA63731, and U01CA70040) and National Cancer Institute grants R03CA150007 and RC2CA148577. The collection of cancer data used in this study was supported in part by several state public health departments and cancer registries throughout the United States; for a full description of these sources, see http://breastscreening.cancer.gov/work/acknowledgement.html.

Appendix

Results of Regression Models Including Random Effects for Facility and Radiologist

In our primary analyses, we adjusted for clustering at the facility level. However, the same radiologist can interpret multiple mammography examinations leading to clustering among radiologists. Because radiologists are largely nested within facilities, adjustment for facility accounts for much of the correlation between observations that arises from radiologist-level clustering. To explore the sensitivity of our results to within-radiologist clustering, we repeated the analyses reported in Tables 2 to to4,4, but allowed for both facility and radiologist random effects. Results of these analyses are shown in Appendix Tables 1 to to33

Appendix Table 1
Performance Measures of Screening Mammography Among Women Undergoing 223 900 Digital Versus 611 098 Film-Screen Examinations, Allowing for Facility and Radiologist Random Effects*
Appendix Table 3
Sensitivity of Screening Mammography Among Women Undergoing Digital Versus Film-Screen Examinations, by Age and Estrogen Receptor Status, Allowing for Facility and Radiologist Random Effects

Results Decomposing Digital Effect Into Within- and Between-Components Rather Than Using Binary Indicators of Film-Only Versus Digital Mammography

In our primary analysis, we assumed that between-facility differences can be entirely captured by a binary indicator representing whether a facility performed any digital mammography. We performed a sensitivity analysis to model the between-facility effect by adjusting for the proportion of digital mammography performed, rather than using a dichotomous classification of any digital mammography versus none. This analysis follows the approach of Neuhaus and Kalbfleisch (19). We repeated the analyses reported in Tables 2 to to44 by using separate between- and within-facility components for digital mammography effect.

Results of these analyses are shown in Appendix Tables 4 to to6.6. In these results we compare the performance of digital mammography with that of film-screen mammography (within-facility effect) while holding the proportion of digital mammograms performed by the facility (between-facility effect) constant. Odds ratios and P values for digital versus film-screen mammography represent the within-facility effect of digital mammography.

Appendix Table 4
Performance Measures of Screening Mammography Among Women Undergoing 231 034 Digital Versus 638 252 Film-Screen Examinations, Allowing for Between- and Within-Facility Effects*
Appendix Table 6
Sensitivity of Screening Mammography Among Women Undergoing Digital Versus Film-Screen Examinations, by Age and Estrogen Receptor Status, Allowing for Between- and Within-Facility Effects

Performance Measures of Screening Mammography for Facilities at the 25th and 75th Percentiles of the Distributions of Facility Performance

The performance estimates reported in Tables 2 to to44 are for a facility with median performance. To demonstrate the extent of between-facility variability in performance, we report performance estimates for facilities at the 25th and 75th percentiles of the distribution of facility performance for each measure (Appendix Tables 7 to to9).9). We also report the SDs for the facility random effects in our logistic-normal random intercept model.

Appendix Table 7
Performance Measures of Screening Mammography Among Women Undergoing 223 900 Digital Versus 611 098 Film-Screen Examinations for Facilities at the 25th and 75th Percentiles of Performance
Appendix Table 9
Sensitivity of Screening Mammography Among Women Undergoing Digital Versus Film-Screen Examinations, by Age and Estrogen Receptor Status, for Facilities at the 25th and 75th Percentiles of Performance

Footnotes

Potential Conflicts of Interest: Disclosures can be viewed at www.acponline.org/authors/icmje/ConflictOfInterestForms.do?msNum=M10-2769.

Reproducible Research Statement:Study protocol: Available from Dr. Kerlikowske (Karla.Kerlikowske/at/ucsf.edu). Statistical code: Available from Dr. Hubbard (hubbard.r/at/ghc.org). Data set: Available after approval by the BCSC Steering Committee at http://breastscreening.cancer.gov/.

Author Contributions: Conception and design: K. Kerlikowske, R.A. Hubbard, D.L. Miglioretti, B.M. Geller, B.C. Yankaskas, S. Taplin.

Analysis and interpretation of the data: K. Kerlikowske, R.A. Hubbard, D.L. Miglioretti, B.M. Geller, B.C. Yankaskas, E.A. Sickles.

Drafting of the article: K. Kerlikowske, R.A. Hubbard, B.M. Geller, B.C. Yankaskas, C.D. Lehman, S. Taplin.

Critical revision of the article for important intellectual content: K. Kerlikowske, R.A. Hubbard, B.M. Geller, B.C. Yankaskas, C.D. Lehman, S. Taplin, E.A. Sickles.

Final approval of the article: K. Kerlikowske, R.A. Hubbard, D.L. Miglioretti, B.M. Geller, B.C. Yankaskas, C.D. Lehman, S. Taplin, E.A. Sickles.

Statistical expertise: R.A. Hubbard, D.L. Miglioretti.

Obtaining of funding: K. Kerlikowske, D.L. Miglioretti, B.C. Yankaskas, S. Taplin.

Collection and assembly of data: K. Kerlikowske, R.A. Hubbard, D.L. Miglioretti, B.M. Geller, B.C. Yankaskas.

References

1. U.S. Food and Drug Administration. Mammography Quality Standards Act and Program. Accessed at www.fda.gov/Radiation-EmittingProducts/MammographyQualityStandardsActandProgram/default.htm on 17 October 2010.
2. Pisano ED, Gatsonis C, Hendrick E, Yaffe M, Baum JK, Acharyya S, 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;353:1773–1783. [PMID: 16169887] [PubMed]
3. Pisano ED, Hendrick RE, Yaffe MJ, Baum JK, Acharyya S, Cormack JB, et al. DMIST Investigators Group Diagnostic accuracy of digital versus film mammography: exploratory analysis of selected population subgroups in DMIST. Radiology. 2008;246:376–383. [PMID: 18227537] [PubMed]
4. Lewin JM, D’Orsi CJ, Hendrick RE, Moss LJ, Isaacs PK, Karellas A, et al. Clinical comparison of full-field digital mammography and screen-film mammography for detection of breast cancer. AJR Am J Roentgenol. 2002;179:671–677. [PMID: 12185042] [PubMed]
5. Skaane P. Studies comparing screen-film mammography and full-field digital mammography in breast cancer screening: updated review. Acta Radiol. 2009;50:3–14. [PMID: 19037825] [PubMed]
6. Skaane P, Hofvind S, Skjennald A. Randomized trial of screen-film versus full-field digital mammography with soft-copy reading in population-based screening program: follow-up and final results of Oslo II study. Radiology. 2007;244:708–717. [PMID: 17709826] [PubMed]
7. Sala M, Comas M, Macià F, Martinez J, Casamitjana M, Castells X. Implementation of digital mammography in a population-based breast cancer screening program: effect of screening round on recall rate and cancer detection. Radiology. 2009;252:31–39. [PMID: 19420316] [PubMed]
8. Vinnicombe S, Pinto Pereira SM, McCormack VA, Shiel S, Perry N, Dos Santos Silva IM. Full-field digital versus screen-film mammography: comparison within the UK breast screening program and systematic review of published data. Radiology. 2009;251:347–358. [PMID: 19401569] [PubMed]
9. Karssemeijer N, Bluekens AM, Beijerinck D, Deurenberg JJ, Beekman M, Visser R, et al. Breast cancer screening results 5 years after introduction of digital mammography in a population-based screening program. Radiology. 2009;253:353–358. [PMID: 19703851] [PubMed]
10. Ballard-Barbash R, Taplin SH, Yankaskas BC, Ernster VL, Rosenberg RD, Carney PA, et al. Breast Cancer Surveillance Consortium: a national mammography screening and outcomes database. AJR Am J Roentgenol. 1997;169:1001–1008. [PMID: 9308451] [PubMed]
11. Ernster VL, Ballard-Barbash R, Barlow WE, Zheng Y, Weaver DL, Cutter G, et al. Detection of ductal carcinoma in situ in women undergoing screening mammography. J Natl Cancer Inst. 2002;94:1546–1554. [PMID: 12381707] [PubMed]
12. American College of Radiology. The American College of Radiology Breast Imaging Reporting and Data System (BI-RADS) 4th ed. Reston, VA: American Coll Radiology; 2003.
13. Yankaskas BC, Taplin SH, Ichikawa L, Geller BM, Rosenberg RD, Carney PA, et al. Association between mammography timing and measures of screening performance in the United States. Radiology. 2005;234:363–373. [PMID: 15670994] [PubMed]
14. American Joint Committee on Cancer. Manual for Staging of Cancer. 6th ed. Philadelphia: JB Lippincott; 2002.
15. Graubard BI, Korn EL. Predictive margins with survey data. Biometrics. 1999;55:652–659. [PMID: 11318229] [PubMed]
16. Lane PW, Nelder JA. Analysis of covariance and standardization as instances of prediction. Biometrics. 1982;38:613–621. [PMID: 7171691] [PubMed]
17. Kerlikowske K, Miglioretti DL, Ballard-Barbash R, Weaver DL, Buist DS, Barlow WE, et al. Prognostic characteristics of breast cancer among postmenopausal hormone users in a screened population. J Clin Oncol. 2003;21:4314–4321. [PMID: 14645420] [PubMed]
18. Carney PA, Miglioretti DL, Yankaskas BC, Kerlikowske K, Rosenberg R, Rutter CM, 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;138:168–175. [PMID: 12558355] [PubMed]
19. Neuhaus JM, Kalbfleisch JD. Between- and within-cluster covariate effects in the analysis of clustered data. Biometrics. 1998;54:638–645. [PMID: 9629647] [PubMed]
20. Ichikawa LE, Barlow WE, Anderson ML, Taplin SH, Geller BM, Brenner RJ. National Cancer Institute-sponsored Breast Cancer Surveillance Consortium. Time trends in radiologists’ interpretive performance at screening mammography from the community-based Breast Cancer Surveillance Consortium, 1996–2004. Radiology. 2010;256:74–82. [PMID: 20505059] [PubMed]
21. Kerlikowske K. The mammogram that cried Wolfe [Editorial] N Engl J Med. 2007;356:297–300. [PMID: 17229958] [PubMed]
22. Roubidoux MA, Bailey JE, Wray LA, Helvie MA. Invasive cancers detected after breast cancer screening yielded a negative result: relationship of mammographic density to tumor prognostic factors. Radiology. 2004;230:42–48. [PMID: 14695385] [PubMed]
23. Buist DS, Porter PL, Lehman C, Taplin SH, White E. Factors contributing to mammography failure in women aged 40–49 years. J Natl Cancer Inst. 2004;96:1432–1440. [PMID: 15467032] [PubMed]
24. Kerlikowske K, Ichikawa L, Miglioretti DL, Buist DS, Vacek PM, Smith-Bindman R, et al. National Institutes of Health Breast Cancer Surveillance Consortium. Longitudinal measurement of clinical mammographic breast density to improve estimation of breast cancer risk. J Natl Cancer Inst. 2007;99:386–395. [PMID: 17341730] [PubMed]
25. Farhat GN, Walker R, Buist DS, Onega T, Kerlikowske K. Changes in invasive breast cancer and ductal carcinoma in situ rates in relation to the decline in hormone therapy use. J Clin Oncol. 2010;28:5140–5146. [PMID: 21060026] [PMC free article] [PubMed]
26. Jiang Y, Metz CE. BI-RADS data should not be used to estimate ROC curves [Editorial] Radiology. 2010;256:29–31. [PMID: 20574083] [PubMed]
27. Kerlikowske K, Grady D, Barclay J, Frankel SD, Ominsky SH, Sickles EA, et al. Variability and accuracy in mammographic interpretation using the American College of Radiology Breast Imaging Reporting and Data System. J Natl Cancer Inst. 1998;90:1801–1809. [PMID: 9839520] [PubMed]
28. Ciatto S, Houssami N, Apruzzese A, Bassetti E, Brancato B, Carozzi F, et al. Categorizing breast mammographic density: intra- and interobserver reproducibility of BI-RADS density categories. Breast. 2005;14:269–275. [PMID: 16085233] [PubMed]
29. Buist DS, Anderson ML, Haneuse SJ, Sickles EA, Smith RA, Carney PA, et al. Influence of annual interpretive volume on screening mammography performance in the United States. Radiology. 2011;259:72–84. [PMID: 21343539] [PubMed]
30. Smith-Bindman R, Chu P, Miglioretti DL, Quale C, Rosenberg RD, Cutter G, et al. Physician predictors of mammographic accuracy. J Natl Cancer Inst. 2005;97:358–367. [PMID: 15741572] [PubMed]