Causal relationships between breast cancer risk factors based on mammographic features.

BACKGROUND: Mammogram risk scores based on texture and density defined by different brightness thresholds are associated with breast cancer risk differently and could reveal distinct information about breast cancer risk. We aimed to investigate causal relationships between these intercorrelated mammogram risk scores to determine their relevance to breast cancer aetiology. METHODS: We used digitised mammograms for 371 monozygotic twin pairs, aged 40-70 years without a prior diagnosis of breast cancer at the time of mammography, from the Australian Mammographic Density Twins and Sisters Study. We generated normalised, age-adjusted, and standardised risk scores based on textures using the Cirrus algorithm and on three spatially independent dense areas defined by increasing brightness threshold: light areas, bright areas, and brightest areas. Causal inference was made using the Inference about Causation from Examination of FAmilial CONfounding (ICE FALCON) method. RESULTS: The mammogram risk scores were correlated within twin pairs and with each other (r = 0.22-0.81; all P < 0.005). We estimated that 28-92% of the associations between the risk scores could be attributed to causal relationships between the scores, with the rest attributed to familial confounders shared by the scores. There was consistent evidence for positive causal effects: of Cirrus, light areas, and bright areas on the brightest areas (accounting for 34%, 55%, and 85% of the associations, respectively); and of light areas and bright areas on Cirrus (accounting for 37% and 28%, respectively). CONCLUSIONS: In a mammogram, the lighter (less dense) areas have a causal effect on the brightest (highly dense) areas, including through a causal pathway via textural features. These causal relationships help us gain insight into the relative aetiological importance of different mammographic features in breast cancer. For example our findings are consistent with the brightest areas being more aetiologically important than lighter areas for screen-detected breast cancer; conversely, light areas being more aetiologically important for interval breast cancer. Additionally, specific textural features capture aetiologically independent breast cancer risk information from dense areas. These findings highlight the utility of ICE FALCON and family data in decomposing the associations between intercorrelated disease biomarkers into distinct biological pathways.

Genetic Aspects of Mammographic Density Measures Associated with Breast Cancer Risk.

Cumulus, Altocumulus, and Cirrocumulus are measures of mammographic density defined at increasing pixel brightness thresholds, which, when converted to mammogram risk scores (MRSs), predict breast cancer risk. Twin and family studies suggest substantial variance in the MRSs could be explained by genetic factors. For 2559 women aged 30 to 80 years (mean 54 years), we measured the MRSs from digitized film mammograms and estimated the associations of the MRSs with a 313-SNP breast cancer polygenic risk score (PRS) and 202 individual SNPs associated with breast cancer risk. The PRS was weakly positively correlated (correlation coefficients ranged 0.05−0.08; all p < 0.04) with all the MRSs except the Cumulus-white MRS based on the "white but not bright area" (correlation coefficient = 0.04; p = 0.06). After adjusting for its association with the Altocumulus MRS, the PRS was not associated with the Cumulus MRS. There were MRS associations (Bonferroni-adjusted p < 0.04) with one SNP in the ATXN1 gene and nominally with some ESR1 SNPs. Less than 1% of the variance of the MRSs is explained by the genetic markers currently known to be associated with breast cancer risk. Discovering the genetic determinants of the bright, not white, regions of the mammogram could reveal substantial new genetic causes of breast cancer.

Familial Aspects of Mammographic Density Measures Associated with Breast Cancer Risk.

Cumulus, Cumulus-percent, Altocumulus, Cirrocumulus, and Cumulus-white are mammogram risk scores (MRSs) for breast cancer based on mammographic density defined in effect by different levels of pixel brightness and adjusted for age and body mass index. We measured these MRS from digitized film mammograms for 593 monozygotic (MZ) and 326 dizygotic (DZ) female twin pairs and 1592 of their sisters. We estimated the correlations in relatives (r) and the proportion of variance due to genetic factors (heritability) using the software FISHER and predicted the familial risk ratio (FRR) associated with each MRS. The ρ estimates ranged from: 0.41 to 0.60 (standard error [SE] 0.02) for MZ pairs, 0.16 to 0.26 (SE 0.05) for DZ pairs, and 0.19 to 0.29 (SE 0.02) for sister pairs (including pairs of a twin and her non-twin sister), respectively. Heritability estimates were 39% to 69% under the classic twin model and 36% to 56% when allowing for shared non-genetic factors specific to MZ pairs. The FRRs were 1.08 to 1.17. These MRSs are substantially familial, due mostly to genetic factors that explain one-quarter to one-half as much of the familial aggregation of breast cancer that is explained by the current best polygenic risk score.

Novel mammogram-based measures improve breast cancer risk prediction beyond an established mammographic density measure.

Mammograms contain information that predicts breast cancer risk. We developed two novel mammogram-based breast cancer risk measures based on image brightness (Cirrocumulus) and texture (Cirrus). Their risk prediction when fitted together, and with an established measure of conventional mammographic density (Cumulus), is not known. We used three studies consisting of: 168 interval cases and 498 matched controls; 422 screen-detected cases and 1197 matched controls; and 354 younger-diagnosis cases and 944 controls frequency-matched for age at mammogram. We conducted conditional and unconditional logistic regression analyses of individually- and frequency-matched studies, respectively. We estimated measure-specific risk gradients as the change in odds per standard deviation of controls after adjusting for age and body mass index (OPERA) and calculated the area under the receiver operating characteristic curve (AUC). For interval, screen-detected and younger-diagnosis cancer risks, the best fitting models (OPERAs [95% confidence intervals]) involved: Cumulus (1.81 [1.41-2.31]) and Cirrus (1.72 [1.38-2.14]); Cirrus (1.49 [1.32-1.67]) and Cirrocumulus (1.16 [1.03 to 1.31]); and Cirrus (1.70 [1.48 to 1.94]) and Cirrocumulus (1.46 [1.27-1.68]), respectively. The AUCs were: 0.73 [0.68-0.77], 0.63 [0.60-0.66], and 0.72 [0.69-0.75], respectively. Combined, our new mammogram-based measures have twice the risk gradient for screen-detected and younger-diagnosis breast cancer (P ≤ 10-12 ), have at least the same discriminatory power as the current polygenic risk score, and are more correlated with causal factors than conventional mammographic density. Discovering more information about breast cancer risk from mammograms could help enable risk-based personalised breast screening.

Interval breast cancer risk associations with breast density, family history and breast tissue aging.

Interval breast cancers (those diagnosed between recommended mammography screens) generally have poorer outcomes and are more common among women with dense breasts. We aimed to develop a risk model for interval breast cancer. We conducted a nested case-control study within the Melbourne Collaborative Cohort Study involving 168 interval breast cancer patients and 498 matched control subjects. We measured breast density using the CUMULUS software. We recorded first-degree family history by questionnaire, measured body mass index (BMI) and calculated age-adjusted breast tissue aging, a novel measure of exposure to estrogen and progesterone based on the Pike model. We fitted conditional logistic regression to estimate odds ratio (OR) or odds ratio per adjusted standard deviation (OPERA) and calculated the area under the receiver operating characteristic curve (AUC). The stronger risk associations were for unadjusted percent breast density (OPERA = 1.99; AUC = 0.66), more so after adjusting for age and BMI (OPERA = 2.26; AUC = 0.70), and for family history (OR = 2.70; AUC = 0.56). When the latter two factors and their multiplicative interactions with age-adjusted breast tissue aging (p = 0.01 and 0.02, respectively) were fitted, the AUC was 0.73 (95% CI 0.69-0.77), equivalent to a ninefold interquartile risk ratio. In summary, compared with using dense breasts alone, risk discrimination for interval breast cancers could be doubled by instead using breast density, BMI, family history and hormonal exposure. This would also give women with dense breasts, and their physicians, more information about the major consequence of having dense breasts-an increased risk of developing an interval breast cancer.

Predicting interval and screen-detected breast cancers from mammographic density defined by different brightness thresholds.

BACKGROUND: Case-control studies show that mammographic density is a better risk factor when defined at higher than conventional pixel-brightness thresholds. We asked if this applied to interval and/or screen-detected cancers. METHOD: We conducted a nested case-control study within the prospective Melbourne Collaborative Cohort Study including 168 women with interval and 422 with screen-detected breast cancers, and 498 and 1197 matched controls, respectively. We measured absolute and percent mammographic density using the Cumulus software at the conventional threshold (Cumulus) and two increasingly higher thresholds (Altocumulus and Cirrocumulus, respectively). Measures were transformed and adjusted for age and body mass index (BMI). Using conditional logistic regression and adjusting for BMI by age at mammogram, we estimated risk discrimination by the odds ratio per adjusted standard deviation (OPERA), calculated the area under the receiver operating characteristic curve (AUC) and compared nested models using the likelihood ratio criterion and models with the same number of parameters using the difference in Bayesian information criterion (ΔBIC). RESULTS: For interval cancer, there was very strong evidence that the association was best predicted by Cumulus as a percentage (OPERA = 2.33 (95% confidence interval (CI) 1.85-2.92); all ΔBIC > 14), and the association with BMI was independent of age at mammogram. After adjusting for percent Cumulus, no other measure was associated with risk (all P > 0.1). For screen-detected cancer, however, the associations were strongest for the absolute and percent Cirrocumulus measures (all ΔBIC > 6), and after adjusting for Cirrocumulus, no other measure was associated with risk (all P > 0.07). CONCLUSION: The amount of brighter areas is the best mammogram-based measure of screen-detected breast cancer risk, while the percentage of the breast covered by white or bright areas is the best mammogram-based measure of interval breast cancer risk, irrespective of BMI. Therefore, there are different features of mammographic images that give clinically important information about different outcomes.

Breast Cancer Risk Associations with Digital Mammographic Density by Pixel Brightness Threshold and Mammographic System.

Purpose To compare three mammographic density measures defined by different pixel intensity thresholds as predictors of breast cancer risk for two different digital mammographic systems. Materials and Methods The Korean Breast Cancer Study included 398 women with invasive breast cancer and 737 control participants matched for age at mammography (±1 year), examination date, mammographic system, and menopausal status. Mammographic density was measured by using the automated Laboratory for Individualized Breast Radiodensity Assessment (LIBRA) software and the semiautomated Cumulus software at the conventional threshold (Cumulus) and at increasingly higher thresholds (Altocumulus and Cirrocumulus, respectively). Measures were Box-Cox-transformed and adjusted for age, body mass index, and menopausal status. Conditional logistic regression was used to estimate risk associations. For calculation of measures of predictive value, the change in odds per standard deviation (OPERA) and the area under the receiver operating characteristic curve (AUC) were used. Results For dense area, with use of the direct conversion system the OPERAs were 1.72 (95% confidence interval [CI]: 1.38, 2.15) for LIBRA, 1.58 (95% CI: 1.27, 1.97) for Cumulus, 2.04 (95% CI: 1.60, 2.59) for Altocumulus, and 3.48 (95% CI: 2.45, 4.47) for Cirrocumulus (P < .001). The corresponding AUCs were 0.70, 0.69, 0.76, and 0.89, respectively. With use of the indirect conversion system, the corresponding OPERAs were 1.50 (95% CI: 1.28, 1.76), 1.36 (95% CI: 1.16, 1.59), 1.40 (95% CI: 1.19, 1.64), and 1.47 (95% CI: 1.25, 1.73) (P < .001) and the AUCs were 0.64, 0.60, 0.61, and 0.63, respectively. Conclusion It is possible that mammographic density defined by higher pixel thresholds could capture more risk-predicting information with use of a direct conversion mammographic system; the mammographically bright, rather than white, regions are etiologically important. © RSNA, 2017.

Mammographic density defined by higher than conventional brightness thresholds better predicts breast cancer risk.

Background: Mammographic density defined by the conventional pixel brightness threshold, and adjusted for age and body mass index (BMI), is a well-established risk factor for breast cancer. We asked if higher thresholds better separate women with and without breast cancer. Methods: We studied Australian women, 354 with breast cancer over-sampled for early-onset and family history, and 944 unaffected controls frequency-matched for age at mammogram. We measured mammographic dense area and percent density using the CUMULUS software at the conventional threshold, which we call Cumulus , and at two increasingly higher thresholds, which we call Altocumulus and Cirrocumulus , respectively. All measures were Box-Cox transformed and adjusted for age and BMI. We estimated the odds per adjusted standard deviation (OPERA) using logistic regression and the area under the receiver operating characteristic curve (AUC). Results: Altocumulus and Cirrocumulus were correlated with Cumulus (r ∼ 0.8 and 0.6 , respectively) . For dense area, the OPERA was 1.62, 1.74 and 1.73 for Cumulus, Altocumulus and Cirrocumulus , respectively (all P < 0.001). After adjusting for Altocumulus and Cirrocumulus , Cumulus was not significant ( P > 0.6). The OPERAs for percent density were less but gave similar findings. The mean of the standardized adjusted Altocumulus and Cirrocumulus dense area measures was the best predictor; OPERA = 1.87 [95% confidence interval (CI): 1.64-2.14] and AUC = 0.68 (0.65-0.71). Conclusions: The areas of higher mammographically dense regions are associated with almost 30% stronger breast cancer risk gradient, explain the risk association of the conventional measure and might be more aetiologically important. This has substantial implications for clinical translation and molecular, genetic and epidemiological research.

Longitudinal Study of Mammographic Density Measures That Predict Breast Cancer Risk.

Background: After adjusting for age and body mass index (BMI), mammographic measures-dense area (DA), percent dense area (PDA), and nondense area (NDA)-are associated with breast cancer risk. Our aim was to use longitudinal data to estimate the extent to which these risk-predicting measures track over time.Methods: We collected 4,320 mammograms (age range, 24-83 years) from 970 women in the Melbourne Collaborative Cohort Study and the Australian Breast Cancer Family Registry. Women had on average 4.5 mammograms (range, 1-14). DA, PDA, and NDA were measured using the Cumulus software and normalized using the Box-Cox method. Correlations in the normalized risk-predicting measures over time intervals of different lengths were estimated using nonlinear mixed-effects modeling of Gompertz curves.Results: Mean normalized DA and PDA were constant with age to the early 40s, decreased over the next two decades, and were almost constant from the mid-60s onward. Mean normalized NDA increased nonlinearly with age. After adjusting for age and BMI, the within-woman correlation estimates for normalized DA were 0.94, 0.93, 0.91, 0.91, and 0.91 for mammograms taken 2, 4, 6, 8, and 10 years apart, respectively. Similar correlations were estimated for the age- and BMI-adjusted normalized PDA and NDA.Conclusions: The mammographic measures that predict breast cancer risk are highly correlated over time.Impact: This has implications for etiologic research and clinical management whereby women at increased risk could be identified at a young age (e.g., early 40s or even younger) and recommended appropriate screening and prevention strategies. Cancer Epidemiol Biomarkers Prev; 26(4); 651-60. ©2017 AACR.