Do women with dense breasts have higher radiation dose during screening mammography?

Transcription

1 Received: 8 June 6 Revised: 6 September 6 Accepted: 9 September 6 DOI:./tbj.8 ORIGINAL ARTICLE Do women with dense breasts have higher radiation dose during screening mammography? Jonathan V. Nguyen MD Mark B. Williams PhD James T. Patrie MS Jennifer A. Harvey MD Department of Radiology and Medical Imaging, University of Virginia Health System, Charlottesville, VA, USA Department of Public Health Sciences, University of Virginia Health System, Charlottesville, VA, USA Correspondence Jonathan V. Nguyen, MD, Department of Radiology and Medical Imaging, University of Virginia Health System, Charlottesville, VA, USA. jvn6c@virginia.edu Abstract Radiation dose during screening mammography is a concern among women. The purpose of this study was to evaluate the relative contribution of breast density to the radiation dose from screening mammography. This IRB approved retrospective study involved collecting patient age, weight, height, compressed breast thickness, and average glandular dose for each exposure for 44 sequential patients undergoing screening mammography at our institution. Automated volumetric density software was used to quantify breast density. The relationship of predictors was evaluated by univariate and multivariate analysis. Median patient age was 58 years and median body mass index (BMI) was 6.8. Median volumetric breast density was 5.8% (range.7-.5%). Median compressed breast thickness was 6.4 mm (range mm). Univariate analysis showed positive associations between radiation dose and both breast thickness and BMI, an inverse association with age, and no association with density. Multivariate regression analysis demonstrated a significant association between dose and age (P=.), laterality (P<.), BMI (P=.8), density (P<.), and breast thickness (P<.). Decomposition of the multivariate regression model coefficient of determination showed that breast thickness was the primary determinant of dose, accounting for 76% of the 58% of the dose variability, followed by density (8%), laterality (4%), age (<%), and BMI (<%). Compressed breast thickness had the greatest effect on average glandular dose. Breast density has a minor impact, while BMI and age have minimal impact on dose. KEYWORDS breast density, mammography, radiation dose INTRODUCTION Screening mammography has reduced the mortality from breast cancer in the United States., The risk of radiation-induced breast cancer is small relative to the benefit achieved with the early detection of breast cancer. However, fear of radiation is a common concern among women of screening age, and a commonly cited reason for not obtaining screening mammograms. 4,5 Women are becoming more self-aware about mammographic breast density. There are currently 4 states with breast density notification laws. 6 The nature of these laws varies between states, but all require that women be informed if they have dense breast tissue. As women become more self-aware of breast density, women may become less willing to undergo a screening test that may potentially increase the risk of radiation-induced breast cancer if the test is of less benefit to them personally due to dense breast tissue. Breast J. 8;4:5 4. wileyonlinelibrary.com/journal/tbj 7 Wiley Periodicals, Inc. 5

2 6 NGUYEN ET AL. The absorbed glandular dose (AGD) to patients during digital screening mammography can vary widely. There have been several studies in the literature examining the factors that affect the AGD from a mammogram. Larger compressed breast thickness (CBT) has been shown to be positively associated with AGD. 7 9 Body mass index (BMI) has also demonstrated a positive correlation with dose. 8 However, the specific relationship between breast density and radiation dose has not been well studied. The purpose of this study was to evaluate the relative contribution of breast density to the radiation dose from digital screening mammography. MATERIALS AND METHODS. Study population Our Institutional Review Board approved this retrospective study and granted a waiver of consent. The study was compliant with the Health Insurance Portability and Accountability Act. Sequential women undergoing digital screening mammography at our institution between June, 4, and June 7, 4 were included. Inclusion criteria were female, age 8-89 years, undergoing screening mammography with Breast Imaging Reporting and Data System (BI-RADS) assessment of or (negative or benign findings respectively). Patients were excluded if they had missing data necessary for analysis or history of breast augmentation or personal history of breast cancer, as treatment changes may affect breast density.. Data collection Patient demographic information, including race, weight, and height was collected from self-reported entries in a questionnaire filled out at the time of the screening mammogram. BMI was calculated using self-reported height and weight. BI-RADS breast density categories were obtained from the mammography report in order to assess the distribution of breast densities in the study population, but were not otherwise used in the analysis. Mammograms were obtained using Senograph D (GE Healthcare, Fairfield, CT) or Selenia (Hologic, Danbury, CT) mammography machines. Radiation dose was determined by the mammographic unit, which calculates an AGD in mgy for every exposure. The parameters for each exposure were extracted from the mammographic unit readout using the DICOM headers. These included CBT and AGD and were obtained for each view. Volumetric breast density (VBD) was calculated quantitatively using automated density software (Volpara, version.4.5; Volpara Solutions, Wellington, New Zealand). All data were gathered and entered into a data-base program (Excel; Microsoft, Bothwell, Washington). Variables collected from the DICOM headers (CBT and AGD) for left and right breast for the same patient were kept separate, due to the possibility of size or breast density differences between breasts. The CBT and AGD recorded for a given breast were obtained by averaging the CC and MLO views.. Statistical analysis.. Data summarization Categorical data were summarized as frequencies and percentages, while continuous scaled data (eg, age) were summarized by the median, the interquartile range (IQR), and the range of the measurement distribution... Mammogram radiation exposure analytic overview Relationships between radiation dose on digital mammography and VBD (%), CBT (mm), breast laterality (left and right breast), BMI (kg/m ), and age (years) were analyzed by way of univariate and multivariate regression. For the univariate and multivariate regression analyses, the radiation dose data were rescaled to the log (mgy) scale, so that the homogeneity of variance assumption and the normality assumption of the univariate and multivariate regression models were satisfied... Univariate regression analyses A generalized estimating equation (GEE) version of univariate linear regression was utilized to examine pairwise associations between radiation dose (log [mgy]), VBD, CBT, BMI, and patient age. In order to have a comparative measure of association, which is not scale-dependent like the regression slope coefficient, the pairwise associations were estimated on a common correlation coefficient scale with range to. With respect to hypothesis testing and confidence interval construction, to account in variance estimation for the cluster nature of data collection (bilateral breast sampling), the Huber and White sandwich estimator was utilized to derive the GEE regression model variance-covariance matrix [, ]. A P.5 decision rule was selected a priori as the null hypothesis rejection criterion establishing statistical significance associations...4 Multivariate regression analyses A GEE version of multivariate linear regression was utilized to examine partial-associations (ie, unique associations) between radiation dose (log [mgy]) and VBD, CBT, breast laterality (left and right breast), BMI, and patient age. With regard to hypothesis testing, due to the cluster nature of data collection (bilateral breast sampling), the Huber and White sandwich estimator was utilized to derive the multivariate GEE regression model variance-covariance matrix. Type III Wald tests were utilized to test the null hypothesis that there is no partial-association (ie, unique-association) between the measurements of the predictor variable (eg, breast density) and radiation dose (log [mgy]) after accounting for co-associations between radiation dose (log [mgy]) and the remaining regression predictor variables. For continuous scaled covariates, partial-associations were quantitatively expressed as a percentage change in geometric mean

3 NGUYEN ET AL. 7 radiation dose (log [mgy]) between a subpopulation of patients at the rd quartile of the covariate distribution and a subpopulation of patients at the st quartile of the covariate distribution, while for breast laterality, the partial-association was quantitatively expressed as percentage change in the geometric mean radiation dose of the right breast relative to left breast. A P.5 decision rule was selected a priori as the null hypothesis rejection criterion establishing statistical significance partial-associations...5 Software The statistical software package Spotfire Splus version 8. (TIBCO, Palo Alto, CA) was used to conduct the statistical analyses. RESULTS Five hundred and sixty-six digital mammographic exams were performed at our institution during the study period. A total of 44 (76.7%) examinations met study criteria. The median age of the study patients was 58 years (IQR: 5-66 years, range 5-89 years). The study population was predominantly non-hispanic Caucasian (8.8%), followed by Hispanic Caucasian (.8%), and Black (.9%). Race information was undocumented for 6 patients (4.5%). Median BMI was 6.8 kg/m (IQR:.-.7 kg/m, range kg/m ). Distribution of reported BI-RADS breast density categories was: almost entirely fatty 5 (8.8%), scattered fibroglandular densities 85 (4.6%), heterogeneously dense (.5%), and extremely dense (5.%). Median VBD was 5.8% (IQR:.9-8.7%, range.7-.5%). Median CBT was 6.4 mm (IQR: mm, range mm). Median glandular doses for the left and right breasts were.68 mgy (IQR: mgy, range mgy) and.75 mgy (IQR:.6-.8 mgy, range mgy), respectively. Univariate relationships between digital mammogram radiation dose (log [mgy]) and VBD, CBT, breast laterality, patient BMI, and patient age demonstrate that VBD has no correlation with radiation dose (r=.6; 95% CI: [.6,.4], P=.) (Figure ). Radiation dose did increase with increasing CBT and BMI, but decreased with advancing age. Interestingly, the radiation dose was slightly higher Radiation Dose (log[mgy]) Radiation Dose (log[mgy]) Radiation Dose (log[mgy]) Left Right (A) Volumetric Breast Density (%) (B) Compressed Breast Thickness (mm) (C) Breast Laterality Radiation Dose (log[mgy]) Radiation Dose (log[mgy]) (D) Age (years) (E) BMI (kg/m ) FIGURE Univariate relationships between digital mammogram radiation dose (log [mgc]) and volumetric breast density (VBD, %) (correlation: r=.6; 95% CI: [.6,.4], P=.) (A), compressed breast thickness (CBT; mm) (correlation: r=.68; 95% CI: [.6,.75], P<.) (B), breast laterality (right vs left, P<.) (C), patient body mass index (BMI; kg/m ) (correlation: r=.9; 95% CI: [.,.7], P<.) (D), and patient age (years) (correlation: r=.9; 95% CI: [.8,.], P<.) (E). For the continuous scale covariate, the red lines identify the regression line of best fit [Color figure can be viewed at wileyonlinelibrary.com]

4 8 NGUYEN ET AL. for the right than the left breast. Univariate correlations between VBD, CBT, BMI, and patient age demonstrate significant interactions between all continuous scaled studied factors (Table ). Breast density was inversely correlated with CBT, BMI, and age. Likewise, CBT correlated positively with BMI and inversely with age. Increasing age correlated with decreasing BMI. Multivariate regression revealed significant unique associations between digital mammography radiation dose (log [mgy]) and VBD (P<.), CBT (P<.), breast laterality (P<.), BMI (P=.4), and patient age (P=.8; Table ). After accounting for co-explained radiation dose variability, VBD and CBT were positively associated with radiation dose, while BMI and patient age were negatively associated with radiation dose. In comparing left and right breasts, the right breast was predicted by the multivariate model to receive on average a larger dose of radiation than left breast (Table S). Overall, the multivariate model predicted 58% of the variability in radiation dose. Decomposition of the multivariate regression model coefficient of determination (ie, R ) showed that CBT was the primary determinant of radiation dose accounting for approximately 76% of the 58% of the variability in dose explained by the multivariate regression model, followed by breast density (8%), breast laterality (4%), age (<%), and BMI (<%) (Figure ). Comparison between the rd TABLE Univariate correlations between the continuous scaled regression model predictor variables Correlate variable X Volumetric breast density (%) Correlate variable Y Compressed breast thickness (mm) Correlation coefficient, r [95% CI] P-value.5 [.6,.4] <. Age (years). [.,.] <. BMI (%).5 [.6,.4] <. Compressed Age (years).7 [.6,.8] <. breast BMI (%).59 [.5,.67] <. thickness (mm) Age (years) BMI (%).5 [.4,.6]. TABLE summary Source Volumetric breast density Compressed breast thickness Multivariate regression model analysis of variance Degree of freedom Type III Wald chi-squared P-value 45.6 <. 4. <. Breast laterality 9.97 <. BMI 4..4 Age Regression model <. R =.58 R adjusted=.57 Compressed Breast Thickness Volumetric Breast Density Breast Laterality Age BMI quartile relative to the st quartile of CBT was associated with a 5.% increase in radiation dose. In contrast, the dose only increased by.7% between the rd and st quartiles of breast density (Figure ). Although a significant predictor of radiation dose in the univariate analysis, once co-explained radiation dose variability was accounted for by other factors, BMI was no longer a predictor of radiation dose. 4 DISCUSSION (.8%) (.5%) (.9%) (.8%) (75.7%) Overall Model R = Attributable Percentage of Overall Model R FIGURE Ranking of the multivariate predictors of radiation dose (log [mgy]). Rank order based on the attributable percentage of the regression model R associated with the unique covariation between the measurements of the predictor variable and radiation dose (log [mgy]) [Color figure can be viewed at wileyonlinelibrary.com] The results of our study demonstrate that breast density is a minor determinant of radiation dose from digital screening mammography, accounting for just % of the dose as a multivariate predictor. Women with dense breast tissue should not defer mammographic screening due to concerns about greater radiation dose. Compressed breast thickness was the dominant factor in determining radiation dose in our study, accounting for 8% of the radiation dose. In clinical practice, not only does increasing compression improve image quality due to separation of structures, our study reinforces that compressing the breast is important in minimizing radiation to the patient. Patient age and BMI both affect radiation, but have minimal effects after adjustment for other factors in our multivariate analysis. Increasing CBT increases radiation dose during a mammographic exposure. Larger breast thickness results in larger overall attenuation, for which the system automatic exposure control compensates by the selection of higher tube voltage (kvp) and/or current-time product (mas) settings. Higher breast density also increases attenuation because of the higher linear attenuation coefficient of glandular tissue compared to adipose tissue, although our study found that this

5 NGUYEN ET AL FIGURE Covariate adjusted estimates for the percent increase in radiation dose, for compressed breast thickness (CBT) =7.5 mm vs CBT=5.5, volumetric breast density (VBD)=8.7 mm relative to VBD=.9 mm, breast laterality=right relative to left breast, Age=66. years relative to Age=5. years, and body mass index (BMI)=.7 relative to BMI=.. Note that for the continuous scale variables, the comparison is the rd quartile value relative to the st quartile value of the empirical distribution. Closed circles identify the point estimate for the % increase in radiation dose, and vertical lines identify the 95% confidence interval for the % increase in radiation dose Increase in Radiation Dose (%) Compressed Breast Thickness Volumetric Breast Density Breast Laterality Predictor Variable Age BMI effect on dose is minor compared with the effect of CBT. We also found that radiation dose decreased with increasing patient age and BMI, likely due to their associations with decreasing breast density. These factors became nonsignificant after multivariate analysis, which compensated for CBT as well as other factors. In, Hendrick et al. 7 likewise found that AGD increased with increasing compressed thickness among four different types of mammographic machines. Our average compressed thickness (6. mm) was similar to their data set. In, Schubauer-Berigan et al. 8 found a positive association between BMI and radiation dose, while our analysis found the opposite. However, the previous study used a univariate analysis, while our study used a multivariate analysis controlling for CBT. As shown in the univariate correlation between variables in our study, BMI is highly correlated to CBT, breast density, and age. Patients with a large BMI would be more likely to have larger CBT and lower breast density, which our data demonstrate have their own individual effects on radiation dose. It is interesting that the right breast overall had a minor but higher radiation dose compared to the left breast although the cause is not clear. The left breast is known to be slightly larger than the right in most women and breast cancer is likewise slightly more common in the left than right breast. A strength of our study was the use of automated software to quantitatively calculate breast density. This allows for optimal use of density as a variable, since categorization into groups, such as BI- RADS density categories, decreases the ability to detect associations with other variables due to grouping. In addition, automated volumetric density measures have high reliability as shown in a study demonstrating low variability with repeated breast density measurements. In contrast, the use of BI-RADS mammographic density categories has well-known moderate intraobserver and interobserver variability., Our study has several limitations. Our study data set comes from a single institution that was largely non-hispanic Caucasian with low racial diversity. In addition, only two machine types were used, and machine-specific programming of the exposure control (AEC) can result in differences in the choice of tube voltage, external filtration, tube current, and exposure time, each of which affects dose. The results of our study may be different in a population with different demographic factors and when using different machines. In our study, approximately 7% of our patient population were reported to have fatty or scattered breast density. Including more women with dense breasts may possibly change the results of our study. Lastly, we used the radiation dose as calculated by the machine, which may differ from one unit model to another. Most machines do not account for breast glandular density in their dose calculations, since it is unknown at the time of exposure. In the Monte Carlobased simulations of Boone 4 and Dance et al., 5 breast glandularity is inversely related to dose for a given breast thickness, x-ray spectrum, and entrance air kerma. Inclusion of VBD in those calculations in the future could alter conclusions regarding the actual clinical dose-density relationship. In summary, our study demonstrates the majority of the radiation dose during digital screening mammography is driven by CBT and is only minimally associated with breast density. Women with dense breast tissue should not have significant concern over increased radiation dose compared with women with nondense breast tissue. REFERENCES. Howlader N, Noone A, Krapcho M, et al. SEER Cancer Statistics Review, Accessed February 6, 6.. Berry DA, Cronin KA, Plevritis SK, et al. Effect of screening and adjuvant therapy on mortality from breast cancer. N Engl J Med. 5;5: Yaffe MJ, Mainprize JG. Risk of radiation-induced breast cancer from mammographic screening. Radiology. ;58:98-5.

6 4 NGUYEN ET AL. 4. Aro AR, de Koning HJ, Absetz P, Schreck M. Two distinct groups of non-attenders in an organized mammography screening program. Breast Cancer Res Treat. ;7: Rimer BK, Keintz MK, Kessler HB, Engstrom PF, Rosan JR. Why women resist screening mammography: patient-related barriers. Radiology. 989;7: Are You Dense? 6. nse/. Accessed March, Hendrick RE, Pisano ED, Averbukh A, et al. Comparison of acquisition parameters and breast dose in digital mammography and screen-film mammography in the American College of Radiology Imaging Network digital mammographic imaging screening trial. AJR Am J Roentgenol. ;94: Schubauer-Berigan MK, Frey GD, Baron L, Hoel DG. Mammography dose in relation to body mass index, race, and menopausal status. Radiat Prot Dosimetry. ;98: Helvie MA, Chan HP, Adler DD, Boyd PG. Breast thickness in routine mammograms: effect on image quality and radiation dose. AJR Am J Roentgenol. 994;6: Senie RT, Rosen PP, Lesser ML, Snyder RE, Schottenfeld D, Duthie K. Epidemiology of breast carcinoma II: factors related to the predominance of left-sided disease. Cancer. 98;46: Alonzo-Proulx O, Mawdsley GE, Patrie JT, Yaffe MJ, Harvey JA. Reliability of automated breast density measurements. Radiology. 5;75: Berg WA, Campassi C, Langenberg P, Sexton MJ. Breast Imaging Reporting and Data System: inter- and intraobserver variability in feature analysis and final assessment. AJR Am J Roentgenol. ;74: Kerlikowske K, Grady D, Barclay J, et al. Variability and accuracy in mammographic interpretation using the American College of Radiology Breast Imaging Reporting and Data System. J Natl Cancer Inst. 998;9: Boone JM. Glandular breast dose for monoenergetic and highenergy X-ray beams: Monte Carlo assessment. Radiology. 999;: Dance DR, Skinner CL, Young KC, Beckett JR, Kotre CJ. Additional factors for the estimation of mean glandular breast dose using the UK mammography dosimetry protocol. Phys Med Biol. ;45: 5-4. SUPPORTING INFORMATION Additional Supporting Information may be found online in the supporting information tab for this article. How to cite this article: Nguyen JV, Williams MB, Patrie JT, Harvey JA. Do women with dense breasts have higher radiation dose during screening mammography?. Breast J. 8;4:5 4.