RPS 2202 - Personalised risk prediction of breast cancer

Author Block: J. Peters¹, D. Van Der Waal¹, M. Smid-Geirnaerdt¹, C. Van Gils², M. Broeders¹; ¹Nijmegen/NL, ²Utrecht/NL
Purpose: Breast density could impact screening strategies. Women with dense breasts face higher breast cancer risk. Furthermore, lesion masking may shorten the pre-clinical detectable phase (PCDP), increasing interval cancer rates. This study examines the associations of automated breast density measures with breast cancer risk and PCDP duration.
Methods or Background: Digital mammograms were used from 60,739 participants in a prospective Dutch screening cohort (PRISMA study, 2014-2019). Dense volume (DV,cm3), volumetric breast density (VBD,%) and Volpara Density Grade (VDG1-4) were assessed using Volpara version 1.5.0. Breast cancer diagnoses were ascertained through linkage with the Netherlands Cancer Registry. Participants with prior breast cancer (N=73) or screen-detected breast cancer at study entry (N=401), were excluded. Information on time to screen-detected and interval cancers was used to fit a three-state (1:cancer-free, 2:pre-clinical detectable cancer, 3:clinical cancer) Markov regression model. Hazard ratios (HRs) were calculated for the effects of breast density on state transition intensities 1->2 (=breast cancer risk) and 2->3 (=1/PCDP duration).
Results or Findings: After a median 4.2 years (IQR 3.9–4.6) we observed 430 screen-detected and 316 interval cancers. Log-transformed VBD and DV were positively associated with increased breast cancer risk (HR 1.12 [95%CI 1.05-1.21] and HR 1.32 [95%CI 1.05-1.35] per one-standard-deviation increase, respectively). Both measures were associated with shorter PCDP (HR 1.48 [95% CI 1.30-1.69] and HR 1.19 [95%CI 1.05-1.35]). Mean PCDP duration was 1.63 years [95%CI 1.11–2.41] for women with highest density (VDG4), compared to 3.41 years [95%CI 3.31-3.52] for VDG1-3.
Conclusion: Breast density is associated with breast cancer risk and PCDP duration. VBD has the strongest association with PCDP, indicating a reduced sensitivity of biennial mammography, while DV contains more information on breast cancer risk.
Limitations: Other confounders than age, e.g. BMI, were not yet included in the model.
Funding for this study: The PRISMA study is funded by ZonMw and KWF.
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: CMO Arnhem-Nijmegen

Author Block: D. Capra, O. Hoda, C. B. Monti, M. Zanardo, F. Sardanelli; Milan/IT
Purpose: Mammography could offer two sex-specific biomarkers to spotlight cardiovascular risk in women: breast arterial calcifications (BAC) and breast density. We conducted a prospective case-control study evaluating the association between BAC and gynaecological and cardiovascular risk factors.
Methods or Background: Consecutive women showing BAC and age- and breast density-matched controls referred for annual mammography were prospectively enrolled. We recorded anthropometric variables, traditional cardiovascular risk factors and gynaecological risk factors. Breast density was classified as low breast density (BI-RADS categories A and B) or high breast density (BI-RADS C and D).
Results or Findings: 72 BAC patients and 72 controls were enrolled (median age 70.0 years, IQR 62.5 to 78 years). Women with BAC had a younger age at menopause (50 vs 52 years, p=0.008), and showed associations with breastfeeding (p=0.041) and parity (p=0.038). Women with BAC show a borderline significant trend towards the use of anti-ipertensive medications (p=0.092). No other differences between cases and controls were observed (p>0.101). Higher breast density was significantly associated with younger age (p<0.001), lower body weight (p<0.001), lower systolic blood pressure (p=0.003), and higher HDL cholesterol (p=0.017), whereas lower breast density was associated with longer time since menopause (p=0.003) and use of anti-ipertensive medications (p<0.001).
Conclusion: Women with BAC have a younger menopausal age, which represents a precocious shift towards a less favourable cardiometabolic hormonal balance. Similarly, women with low breast density show an unfavourable cardiovascular risk profile, using more often anti-ipertensive medications, having a higher systolic blood pressure and lower levels of HDL cholesterol.
Limitations: Age and breast density matching reduces the statistical power to observe associations between breast density and cardiovascular risk factors among women of similar age.
Absence of follow up to record cardiovascular events.
Funding for this study: General Electric Healthcare supported this study.
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: Ethics committe approval number 90/INT/2020, 08/09/2020. All participants signed informed consent.

Author Block: E. F. Conant¹, C. Parghi², P. Hall³, M. Eriksson³; ¹Philadelphia, PA/US, ²Addison, TX/US, ³Stockholm/SE
Purpose: Image-derived AI-based risk models demonstrate ability to predict breast cancer risk using digital mammography (DM) and digital breast tomosynthesis (DBT) imaging data. However, a direct comparison of performances within the same screening population has yet to be conducted.
Methods or Background: We conducted a nested case-control study including women aged 35-98 from Solis and UPenn screening cohorts, between 2014 and 2021. Participants were followed for two screens, with cancer diagnosed before August 2022. Two image-based ProFound AI Risk models, one for DM and one for DBT, estimated absolute 1-year breast cancer risks at study-entry. We assessed models’ discriminatory performance (AUC) controlling for screening site and classified risks according to U.S. Preventive Services Task Force (USPSTF) thresholds.
Results or Findings: Study included 780 women with incident breast cancer (mean age 63.4±11.2) and 7,481 controls (mean age 57.0±10.6). Cancers were diagnosed on average 1.3±0.5 years (range 4 months to 4 years) after study-entry. At study entry, AUCs of DM and DBT models were 0.71 (95% CI: 0.69-0.73) and 0.75 (95% CI: 0.73-0.77), respectively (p<0.01). Comparing UPenn and Solis data, similar estimates were observed for respective models. Based on USPSTF guidelines, 14% of women were classified as high-risk. Among this group, DM model predicted 40% (95% CI: 36-43%) of future breast cancers, compared to 48% (95% CI: 44-52%) by DBT model (p<0.01). Non-significant differences in proportions of future breast cancers were observed comparing sites.
Conclusion: The image-derived DM and DBT AI-risk models predicted 40-48% of future breast cancers at study-entry in two U.S. screening populations. The DBT model predicted a significantly higher proportion of future cancers compared to the DM model emphasizing the need for some women to obtain supplemental screening.
Limitations: Study is limited to only 2 sites in U.S.
Funding for this study: iCAD, Inc.
Swedish Research Council
Swedish Breast Cancer Association
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: The study has been reviewed and approved.

Author Block: G. Gennaro¹, L. Bucchi², A. Ravaioli², F. Caumo¹; ¹Padova/IT, ²Meldola/IT
Purpose: To evaluate the impact of different breast density metrics on the personalization of breast cancer screening.
Methods or Background: The RIBBS study (ClinicalTrials.gov NCT05675085) is a personalized breast screening study targeting young women. The protocol used digital breast tomosynthesis (DBT) and double reading to stratify participants based on individual breast cancer risk and breast density. In this protocol, breast density determined the need for supplemental ultrasound (US), while breast cancer risk guided the frequency of screening. A quantitative software tool provided volumetric breast density, with a 25% threshold used to identify women who needed supplemental US. This study compares stratification based on this volumetric approach with stratification using categorical breast density metrics, both objective and human.
Results or Findings: A total of 10,269 women, all aged 45 years, were enrolled in the RIBBS study. Of these, 1,904 women (18.5%) had ≥25% breast volumetric density and underwent additional US. Using the categorical breast density provided by the same software, 41.1% of participants would have been categorized as having a BIRADS “d” category density, 2.2 times higher than that identified through the volumetric metric. If BIRADS categorization had been performed by human readers, considering every DBT classified as “d” by at least one reader, the percentage would have dropped to 32.3%, still 1.7 times higher than the current protocol.
Conclusion: This study shows that the choice of breast density metrics can significantly influence the stratification process in personalized breast cancer screening. Quantitative metrics allow more precise stratification than categorical approaches, improving the feasibility of supplemental imaging in clinical practice.
Limitations: Changes in screening performance due to the use of categorical breast density for supplemental US remain unassessed, as the current results are based solely on the applied protocol.
Funding for this study: This specific subanalysis had no funding
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: The RIBBS study was approved by the Ethics Committee with the following code "RIBBS 2019/37"

Author Block: J. Gjesvik, N. Moshina, S. Sagstad, M. Larsen, Å. S. Holen, M. B. Bergan, S. Hofvind; Oslo/NO
Purpose: The evidence on longitudinal changes in mammographic density in breasts developing cancer is limited. We aimed to analyse mammographic density among women developing and not developing breast cancer over three consecutive screening rounds in BreastScreen Norway.
Methods or Background: In this retrospective cohort study, 66,696 women aged 50-69 years with three consecutive screening examinations performed in Rogaland and Hordaland counties, 2007-2020, were included. A total of 909 women were diagnosed with screen-detected and 287 with interval cancer. Mammographic density data was obtained from an automated software (Volpara 1.5.0 and 1.5.4.0) and included absolute (cm3) and percent (%) dense volume per woman and breast. A linear mixed-effects model with a fixed effect for each woman was applied on a breast level to define the changes in absolute and percent dense volume. The model was adjusted for age at entry, breast volume, and history of benign breast disease.
Results or Findings: Mean age for women not developing breast cancer was 62.5 years (standard deviation, SD: 5.1), while it was 62.3 (SD: 4.4) for women with screen-detected cancer and 61.9 (SD: 4.8) for interval cancer. Absolute and percent dense volume decreased over time in all women. In breasts developing cancer the rate of decrease was lower for absolute dense volume, estimate=0.004 (95% CI: 0.002-0.007, p=0.041), compared to breasts not developing cancer. The rate of decrease was also lower for percent dense volume, estimate=0.003 (95%CI 0.000-0.007, p=0.053), in breasts developing versus not developing cancer.
Conclusion: Absolute dense volume decreased to a lower degree in breasts developing versus not developing cancer. Longitudinal changes in absolute dense volume could be used for more precise breast cancer risk prediction and screening personalization.
Limitations: The study is retrospective and the population is fairly homogenous.
Funding for this study: No funding
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: Regional Commitee for Medical and Health Research Ethics

Author Block: J. Gjesvik¹, N. Moshina¹, C. Lee², D. L. Miglioretti³, S. Hofvind¹; ¹Oslo/NO, ²Seattle, WA/US, ³Davis, CA/US
Purpose: Investigate whether an artificial intelligence algorithm (AI) trained for detecting breast cancer scored the breast developing breast cancer and the breast not developing breast cancer differently years before diagnosis.
Methods or Background: In this retrospective cohort study, we included women aged 50-69 who attended three consecutive biennial screening rounds between 2004 and 2018, as part of BreastScreen Norway. A total of 116 495 women were included in the final study population, 1265 with screen-detected breast cancer detected at, and 342 with interval cancer diagnosed within two years after, the third screening round. We used a commercial AI algorithm to score each breast with a risk score between 0 and 100.
Results or Findings: For women developing screening-detected breast cancer the mean AI-score at the first screening round for the breast developing breast cancer was 19.2 (SD: 28.6), and 82.7 (SD: 26.7) after the third screening round. The score was 9.5 (SD: 19.0) in the first and 5.0 (SD: 15.7) in the third screening round for the breast not developing breast cancer. For interval cancer, the mean scores for breasts developing cancer were 17.8 (SD:26.3) and 33.1 (SD: 33.8), respectively, and mean scores for breasts not developing cancer were 10.5 (SD: 19.9) and 8.4 (SD: 18.7), respectively. For women not developing breast cancer, the mean AI score was 7.1 (SD: 15.2) in the first and 6.4 (SD: 14.5) in the third screening rounds, respectively.
Conclusion: AI-scores were higher in breasts developing cancer up to 6 years before it was diagnosed. The findings suggests that commercial AI algorithms for breast cancer detection might be considered for identifying women at higher risk of developing breast cancer.
Limitations: This is a retrospective study, and the population is mostly homogenous.
Funding for this study: Funded by the Norwegian Cancer Society (Pink Ribbon)
Has your study been approved by an ethics committee? Yes
Ethics committee - additional information: Regional Committee for Medical and Health Research Ethics, Norway

Author Block: S. D. Verboom¹, J. G. Mainprize², J. Peters¹, M. Broeders¹, M. Yaffe², I. Sechopoulos¹; ¹Nijmegen/NL, ²Toronto, ON/CA
Purpose: To externally validate a lesion masking prediction model for mammograms, Mammatus, developed on a North American cohort, in a larger retrospective breast cancer screening cohort from one screening center in The Netherlands.
Methods or Background: A total of 935 digital mammograms from the Dutch Breast Cancer Screening Program with a unilateral invasive breast cancer that was either screen detected or diagnosed within 24 months after a negative screening (interval cancer) were included. All mammograms were retrospectively evaluated for the visibility of malignant masses using all available diagnostic imaging and clinical information. Mammatus was applied on the contralateral mammogram to eliminate the influence of the lesion.

The area under the receiver operator characteristics (ROC) curve (AUC) when using Mammatus to distinguish examinations with screen-detected cancers (assumed low masking risk) from interval cancers (assumed high masking risk) was computed. The AUC was compared to that of the original cohort and to that obtained using volumetric breast density (VBD) as a predictor. A second three-category ROC analysis was performed, with interval cancers that were retrospectively visible classified as intermediate lesion masking.
Results or Findings: Mammatus achieved an AUC of 0.70 (95%CI 0.67-0.74) for distinguishing between screen-detected- (n=632) and interval-cancer exams (n=303). This performance did not differ from the original study (AUC=0.75 (0.68-0.82), p=0.20), and outperformed VBD (AUC=0.66 (0.62-0.70, p<0.002). The three-category ROC analysis showed that Mammatus outperformed VBD at identifying low risk of lesion masking (AUC=0.74 (0.70-0.77)), however, not for identifying high risk (AUC=0.69 (0.65-0.74)).
Conclusion: Mammatus performed well in predicting breast cancer-masking risk in a Dutch screening cohort. This suggests that adding information other than density improves prediction of lesion masking.
Limitations: There is no ground truth of lesion masking risk, therefore the best possible approximation is used.
Funding for this study: aiREAD financed by the Dutch Research Council (NWO), Dutch Cancer Society (KWF), and Health Holland (HH)
Has your study been approved by an ethics committee? Not applicable
Ethics committee - additional information: The Radboudumc ethics committee declared that this study falls outside the scope of the Dutch Medical Research involving Human Subjects Act and could be carried out without approval of an Institutional Review Board.