Breast Cancer Risk Prediction Research Articles

Functional partial least squares with censored outcomes: Prediction of breast cancer risk with mammogram images

Functional partial least squares with censored outcomes: Prediction of breast cancer risk with mammogram images

Mammographic texture versus conventional Cumulus measure of density in breast cancer risk prediction: a literature review.

Mammographic textures show promise as breast cancer risk predictors, distinct from mammographic density. Yet, it lacks comprehensive evidence to determine the stronger risk predictor between textures and density, and the reliability of texture-based measures. We searched PubMed database for research publications, published up to November 2023, which assessed breast cancer risk associations(odds ratios[OR]) with texture-based measures and percent mammographic density(PMD), and their discrimination(area under the receiver operating characteristics curve[AUC]), using same datasets. Of 11 publications, for textures, six found stronger associations(P<0.05) with 11%-508% increases on log scale by study and four found weaker associations(P<0.05) with 14%-100% decreases, compared with PMD. Risk associations remained significant when fitting textures and PMD together. Eleven of 17 publications show greater AUCs for textures than PMD(P<0.05); increases were 0.04-0.25 by study. Discrimination of PMD and these textures jointly was significantly higher than PMD alone (P<0.05). Therefore, different textures could capture distinct breast cancer risk information, partially independent of mammographic density, suggesting their joint role in breast cancer risk prediction. Certain textures could outperform mammographic density for predicting breast cancer risk. However, obtaining reliable texture-based measures necessitates addressing various issues. Collaboration of researchers from diverse fields could be beneficial for advancing this complex field.

Deep Learning for Breast Cancer Risk Prediction: Application to a Large Representative UK Screening Cohort.

"Just Accepted" papers have undergone full peer review and have been accepted for publication in Radiology: Artificial Intelligence. This article will undergo copyediting, layout, and proof review before it is published in its final version. Please note that during production of the final copyedited article, errors may be discovered which could affect the content. Purpose To develop an artificial intelligence (AI) deep learning tool capable of predicting future breast cancer risk from a current negative screening mammographic examination and to evaluate the model on data from the UK National Health Service Breast Screening Program. Materials and Methods The OPTIMAM Mammography Imaging Database contains screening data, including mammograms and information on interval cancers, for > 300,000 women who attended screening at three different sites in the UK from 2012 onward. Cancer-free screening examinations from women aged 50-70 years were obtained and classified as risk-positive or risk-negative based on the occurrence of cancer within 3 years of the original examination. Examinations with confirmed cancer and images containing implants were excluded. From the resulting 5264 risk-positive and 191488 risk-negative examinations, training (n = 89285) validation (n = 2106) and test (n = 39351) datasets were produced for model development and evaluation. The AI model was trained to predict future cancer occurrence based on screening mammograms and patient age. Performance was evaluated on the test dataset using the area under the receiver operating characteristic curve (AUC) and compared across subpopulations to assess potential biases. Interpretability of the model was explored, including with saliency maps. Results On the hold-out test set, the AI model achieved an overall AUC of 0.70 (95% CI: 0.69, 0.72). There was no evidence of a difference in performance across the three sites, between patient ethnicities or across age-groups Visualization of saliency maps and sample images provided insights into the mammographic features associated with AI-predicted cancer risk. Conclusion The developed AI tool showed good performance on a multisite, UK-specific dataset. ©RSNA, 2024.

Breast density prediction from low and standard dose mammograms using deep learning: effect of image resolution and model training approach on prediction quality

Purpose. To improve breast cancer risk prediction for young women, we have developed deep learning methods to estimate mammographic density from low dose mammograms taken at approximately 1/10th of the usual dose. We investigate the quality and reliability of the density scores produced on low dose mammograms focussing on how image resolution and levels of training affect the low dose predictions. Methods. Deep learning models are developed and tested, with two feature extraction methods and an end-to-end trained method, on five different resolutions of 15,290 standard dose and simulated low dose mammograms with known labels. The models are further tested on a dataset with 296 matching standard and real low dose images allowing performance on the low dose images to be ascertained. Results. Prediction quality on standard and simulated low dose images compared to labels is similar for all equivalent model training and image resolution versions. Increasing resolution results in improved performance of both feature extraction methods for standard and simulated low dose images, while the trained models show high performance across the resolutions. For the trained models the Spearman rank correlation coefficient between predictions of standard and low dose images at low resolution is 0.951 (0.937 to 0.960) and at the highest resolution 0.956 (0.942 to 0.965). If pairs of model predictions are averaged, similarity increases. Conclusions. Deep learning mammographic density predictions on low dose mammograms are highly correlated with standard dose equivalents for feature extraction and end-to-end approaches across multiple image resolutions. Deep learning models can reliably make high quality mammographic density predictions on low dose mammograms.

Genome-wide association analyses of breast cancer in women of African ancestry identify new susceptibility loci and improve risk prediction.

We performed genome-wide association studies of breast cancer including 18,034 cases and 22,104 controls of African ancestry. Genetic variants at 12 loci were associated with breast cancer risk (P < 5 × 10-8), including associations of a low-frequency missense variant rs61751053 in ARHGEF38 with overall breast cancer (odds ratio (OR) = 1.48) and a common variant rs76664032 at chromosome 2q14.2 with triple-negative breast cancer (TNBC) (OR = 1.30). Approximately 15.4% of cases with TNBC carried six risk alleles in three genome-wide association study-identified TNBC risk variants, with an OR of 4.21 (95% confidence interval = 2.66-7.03) compared with those carrying fewer than two risk alleles. A polygenic risk score (PRS) showed an area under the receiver operating characteristic curve of 0.60 for the prediction of breast cancer risk, which outperformed PRS derived using data from females of European ancestry. Our study markedly increases the population diversity in genetic studies for breast cancer and demonstrates the utility of PRS for risk prediction in females of African ancestry.

Population-specific validation and comparison of the performance of 77- and 313-variant polygenic risk scores for breast cancer risk prediction.

The polygenic risk score (PRS) allows the quantification of the polygenic effect of many low-penetrance alleles on the risk of breast cancer (BC). This study aimed to evaluate the performance of two sets comprising 77 or 313 low-penetrance loci (PRS77 and PRS313) in patients with BC in the Czech population. In a retrospective case-control study, variants were genotyped from both the PRS77 and PRS313 sets in 1329 patients with BC and 1324 noncancer controls, all women without germline pathogenic variants in BC predisposition genes. Odds ratios (ORs) were calculated according to the categorical PRS in individual deciles. Weighted Cox regression analysis was used to estimate the hazard ratio (HR) per standard deviation (SD) increase in PRS. The distributions of standardized PRSs in patients and controls were significantly different (p < 2.2×10-16) with both sets. PRS313 outperformed PRS77 in categorical and continuous PRS analyses. For patients in the highest 2.5% of PRS313, the risk reached an ORof3.05 (95% CI,1.66-5.89; p=1.76×10-4). The continuous risk was estimated as an HRper SDof1.64 (95% CI,1.49-1.81; p<2.0×10-16), which resulted in an absolute risk of 21.03% at age 80 years for individuals in the 95th percentile of PRS313. Discordant categorization into PRS deciles was observed in 248 individuals (9.3%). Both PRS77 and PRS313 are able to stratify individuals according to their BC risk in the Czech population. PRS313 shows better discriminatory ability. The results support the potential clinical utility of using PRS313 in individualized BC risk prediction.

Abstract PS10-07: A second-generation polygenic risk score (PRS) based on genetic ancestry improves breast cancer (BC) risk prediction for all ancestries

Abstract Background: Common genetic variants, mainly single-nucleotide polymorphisms (SNPs) explain substantial genetic susceptibility to BC. PRS have been developed to quantify the combined effects of BC-associated SNPs, providing important information about BC risk. Historically, genome-wide association studies have been conducted in predominantly European populations, resulting in miscalibrated and inaccurate PRS for non-Europeans. We previously described a multiple-ancestry PRS (MA-PRS-149) based on 56 ancestry-informative and 93 BC-associated SNPs. The MA-PRS-149 achieved accuracy for all women by characterizing the genetic ancestry of each BC-associated SNP in terms of three reference ancestries (African, East Asian, and European), applying ancestry-specific SNP risks and frequencies, and combining the results as a weighted sum of three ancestry-specific PRS. Here, we aimed to improve the predictive accuracy of MA-PRS-149, particularly for non-Europeans, through the inclusion of additional BC-associated SNPs. Methods: Women referred for hereditary cancer testing and negative for pathogenic variants in BC-associated genes were divided into consecutive study cohorts to (1) quantify ancestry-specific SNP risks, (2) combine the three ancestry-specific PRS, and (3) pre-specified clinical validation. To select an optimal set of BC-associated SNPs, we developed a novel synthetic stepwise regression methodology that accounts for linkage disequilibrium. Ancestry-specific SNP risks were determined from meta-analyses of literature with clinical cohorts of 57,827 Black/African and 26,992 East Asian women. Ancestry-specific PRS were combined based on a diverse cohort of 157,740 women. Clinical validation was conducted in an independent cohort of 77,774 women. We used multivariable logistic regression adjusted for age, ancestry, and cancer history to test for improved BC risk prediction over clinical factors. We tested for improvement over the MA-PRS-149, and a European PRS, by including additional PRS as covariates. Calibration was assessed through goodness-of-fit tests. All analyses were conducted within the full cohort and ancestral subpopulations. Odds ratios (ORs) and 95% confidence intervals (CIs) are reported per standard deviation within the corresponding population. Results: An optimal set of 383 SNPs (56 ancestry-informative and 327 BC-associated) was included in the final PRS (MA-PRS-383). MA-PRS-383 added significant predictive information to clinical factors in the full cohort and within each ancestry (Table 1). MA-PRS-383 had greater predictive accuracy than MA-PRS-149 or a 383-SNP PRS with European weights. Goodness-of-fit tests showed that MA-PRS-383 was well-calibrated and predicted risk accurately in the tails of the distribution for both European and non-European women. Conclusion: MA-PRS-383 was well-calibrated and substantially improved upon existing PRS in all tested ancestral populations. Incorporation of MA-PRS-383 into BC risk assessment may lead to more accurate identification of women who are most likely to benefit from screening and preventive medications. Citation Format: Timothy Simmons, Elisha Hughes, Dmitry Pruss, Matthew Kucera, Benjamin Roa, Thaddeus Judkins, Thomas Slavin, Victor Abkevich, Ryan Hoff, Srikanth Jammulapati, Susanne Wagner, Dale Muzzey, Jerry Lanchbury, Alexander Gutin. A second-generation polygenic risk score (PRS) based on genetic ancestry improves breast cancer (BC) risk prediction for all ancestries [abstract]. In: Proceedings of the 2023 San Antonio Breast Cancer Symposium; 2023 Dec 5-9; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2024;84(9 Suppl):Abstract nr PS10-07.

Abstract PO5-07-06: Impact of patient-specific factors on quantitative breast parenchymal texture features

Abstract Breast parenchymal texture characterizes the spatial and structural patterns of breast tissue and is an independent predictor of breast cancer risk. Although it has conceptual similarities with another breast composition measure, breast density, parenchymal texture is more complex and is represented by multiple quantitative measures. The extent to which the epidemiology and biological basis of parenchymal texture and breast density differ has not been well described. In this cross-sectional study, we evaluated quantitative measures of parenchymal texture among 308 postmenopausal women aged ≥ 45 years who participated in a routine breast cancer screening at an academic medical center between 2020 and 2022. After providing informed consent, participants completed a short interview-based questionnaire. Additional demographic and breast cancer risk factor information was obtained from the medical record. Parenchymal texture features were measured from craniocaudal views of full-field digital mammograms. Measured texture values were standardized and averaged across left and right breasts. For this analysis, we evaluated associations between features previously validated as being associated with breast cancer risk (fractal dimension, grey-level mean, co-occurrence entropy and inverse difference moment) and patient personal and clinical characteristics reported at the time of screening. Associations between texture features and participant characteristics were evaluated using Pearson correlations (continuous variables) or the Kruskal-Wallis test (categorical variables). Generalized linear models were used to adjust for technical factors associated with image acquisition, including mammography machine, compression force, and software version. P-values less than 0.05 were considered statistically significant. Among the 308 participants, 70% were White, 24% were Black, 1% were Asian, 1% were American Indian, and 4% were another race that was not specified. 4% of participants reported Hispanic ethnicity. The mean age was 64.7 years (range 45 – 84 years). Fewer than half of the women were reported by a radiologist to have dense breasts (heterogeneously dense – 24%, extremely dense – 5%, vs. almost entirely fatty – 21%, scattered fibroglandular densities – 50%). The mean body mass index (BMI) was 29.4 kg/m2 (range 19.4 to 53.1 kg/m2) and 77% of the women were parous. BMI was strongly correlated with fractal dimension (r=0.62, P&amp;lt; 0.01) and grey-level mean (r=-0.48, P&amp;lt; 0.01), and had a weaker correlation with co-occurrence entropy (r=-0.19, P&amp;lt; 0.01). Age was weakly correlated with co-occurrence inverse difference moment (r=-0.12, P=0.03). BI-RADS breast density was inversely associated with fractal dimension, grey-level mean, and co-occurrence entropy (all P&amp;lt; 0.01). Adjustment for technical factors attenuated some associations, but BMI associations with fractal dimension (P&amp;lt; 0.01) and gray-level mean (P&amp;lt; 0.01), and breast density associations with fractal dimension, grey-level mean, and co-occurrence entropy (all P&amp;lt; 0.01) all remained statistically significant. Parity was not associated with any of the features evaluated in unadjusted or adjusted analyses (all P &amp;gt;0.05). These data suggest that, in postmenopausal women, risk-associated parenchymal texture features are associated with measures affected by body size and adiposity, but not with measures related to age or reproductive status. Longitudinal studies are needed to understand the temporality of these associations. Additionally, future studies will address the joint effects of texture features and the impact of endogenous estrogen levels on the features, which will provide insight into the biological mechanisms that influence texture feature associations with personal characteristics and breast cancer risk. Citation Format: Sarah Nyante, Yukie Kajita, Walter Mankowski, Ley Killeya, Despina Kontos, Xianming Tan, Eric Cohen, Cherie Kuzmiak. Impact of patient-specific factors on quantitative breast parenchymal texture features [abstract]. In: Proceedings of the 2023 San Antonio Breast Cancer Symposium; 2023 Dec 5-9; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2024;84(9 Suppl):Abstract nr PO5-07-06.

Abstract PO4-07-03: Machine learning breast cancer risk prediction using sequential past mammograms - a pilot study

Abstract Background and Aim Some studies have reported the use of mammogram AI deep learning algorithms to accurately predict the risk of future breast cancer development in women. In these studies, random and unrelated mammograms were independently used for training of the AI model. We propose a novel deep learning system using sequential past mammograms instead of random studies and aimed to determine if it can further improve risk prediction in a pilot study. Methods Pre-training of the backbone of a Swin-transformer machine learning model was performed by applying region-of-interest lesional masks of benign and malignant on a training set of 1644 mammograms derived from the open-access Curated Breast Imaging Subset of Digital Database for Screening Mammography dataset. To further develop the model, deep learning training and validation were performed on the Cohort of Screen-age Women (CSAW) open-access screening mammogram set which was derived from the long term data of 873 women who developed breast cancer for the first time and 7850 non-cancer women. We then compared the diagnostic accuracy of a model using sequential past mammograms for training against the standard, non-sequential model. For the sequential model, imaging segmentation of the potential cancer site on mammography was performed on all past, pre-cancer mammograms obtained within 60 to 2555 days of cancer diagnosis in the pre-processing stage for breast cancer cases. AI training was then carried out by combining the pre-cancer mammograms belonging to the same woman in a temporal sequence using a transformer encoder. The learning rate started from 1e-6 and decayed 10 times every 10 epochs. For the non-sequential model, we removed the temporal encoder and used random, unrelated mammogram studies for training. Validation was performed on a dataset consisting of 75 pre-cancer mammograms from breast cancer cases and 69 mammograms from non-cancer cases. Results Sensitivity, specificity and overall accuracy on the validation set were 77.3% (58/75), 68.1% (47/69) and (105/144) 72.9% respectively for the sequential model compared to 69.3% (52/75), 69.6% (48/69) and 69.4% (100/144) respectively for the non-sequential model. Conclusion The study showed that deep learning breast cancer risk prediction can be further improved by using sequential past mammograms instead of random mammograms for AI model training. This can also potentially enhance other AI risk prediction models that employ combined mammogram and traditional breast cancer clinical risk factors. A larger validation study will be useful. Contact information: email Lester Leong at lester.leong.c.h@singhealth.com.sg ROC curve for breast cancer risk prediction using AI deep learning training with sequential past mammograms. Citation Format: Lester Leong, Mingjie Xu, Weimin Huang, Erli Zhang, Engracia Loh, Sze Yiun Teo, Geok Hoon Lim, Veronique Kiak Mien Tan, Yirong Sim, Fredrik Strand, Ryan Tan. Machine learning breast cancer risk prediction using sequential past mammograms - a pilot study [abstract]. In: Proceedings of the 2023 San Antonio Breast Cancer Symposium; 2023 Dec 5-9; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2024;84(9 Suppl):Abstract nr PO4-07-03.

Abstract PO3-08-01: Decision curve analysis to compare breast cancer risk predictions for a polygenic integrated clinical risk model with those of a gold standard

Abstract Risk-based guidelines have been developed to provide advice on options that are available to women who are at increased risk of breast cancer. We and others are focused on improving risk prediction models for general population use. For breast cancer, polygenic risk has been identified as important for risk prediction, especially when combined with clinical risk factors such as family history, anthropometric measures, breast imaging measures, and hormonal and reproductive risk factors. When comparing the performance of risk prediction models, the AUC is most often used but it focuses on performance in terms of distinguishing between affected and unaffected individuals and does not take into account differences in the specificity (true negative rate) and sensitivity (true positive rate) of the models being compared. In contrast, decision curve analysis takes sensitivity and specificity into account and enables the comparison of the net benefit of risk prediction tools at the clinical risk thresholds that are used to guide clinical management decisions. Herein, we use decision curve analysis to compare breast cancer risk prediction guidelines for BRISK and the Breast Cancer Risk Assessment Tool (BCRAT) in the UK Biobank, and for BRISK and IBIS version 7 in the Nurses’ Health Study. We evaluated the net benefit at the 5-year risk thresholds (1.67% and 3%) that are used to guide clinical management decisions around risk-reducing medication and at the remaining lifetime risk of 20% that is used to guide supplemental MRI surveillance. In the UK Biobank, BRISK showed a net benefit over BCRAT at the 5-year risk thresholds of 1.67% and 3.0%. The AUCs were 0.649 (95% CI = 0.640, 0.695) for BRISK and 0.567 (95% CI = 0.556, 0.577) for BCRAT and showed that, overall, BRISK discriminated between affected and unaffected women better than BCRAT (P &amp;lt; 0.001). In the Nurses’ Health Study, BRISK showed a net benefit over IBIS at the remaining lifetime risk threshold of 20%. The AUCs were 0.647; 95% CI = 0.627, 0.668 for BRISK and 0.571; 95% CI = 0.546, 0.595 for IBIS, a statistically significant improvement for BRISK over IBIS (P &amp;lt; 0.0001). We also looked at sensitivity, specificity, positive predictive value and negative predictive value even though these risk models are one-step removed from breast cancer diagnosis by screening/risk-reduction options. The positive predictive values are higher for the BRISK (27.1%) model compared to BCRAT (5.5%) in the UK Biobank dataset and higher for BRISK (15.4%) compared to IBISv7 (14.0%) in the Nurses’ dataset. We have shown the application of a statistical tool that can be used to directly compare risk models at clinically relevant thresholds. We have shown that BRISK, which comprises polygenic risk and clinical risk factors, shows higher net benefit at both 1.67% and 3% thresholds compared to BCRAT and at 20% remaining lifetime risk for IBIS version 7. In breast cancer risk prediction, a tool with high specificity (and therefore a necessary trade-off for low sensitivity) is preferred because we do not want to incorrectly classify women as not being at high risk and deny them the opportunity for intervention in the form of more frequent screening, alternative modes of screening or risk-reducing medication. This improvement in general population risk stratification that BRISK can provide has the potential to have a clinically significant effect on women’s health. Citation Format: Erika Spaeth, Bernard A Rosner, Gill Dite. Decision curve analysis to compare breast cancer risk predictions for a polygenic integrated clinical risk model with those of a gold standard [abstract]. In: Proceedings of the 2023 San Antonio Breast Cancer Symposium; 2023 Dec 5-9; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2024;84(9 Suppl):Abstract nr PO3-08-01.

Abstract PS10-09: Development of an absolute risk prediction model for premenopausal breast cancer in an international consortium

Abstract Risk prediction models that have been developed for overall breast cancer risk are based on a limited number of premenopausal cases in individual cohorts. As some risk factors differ in their associations with pre- versus postmenopausal breast cancer, a distinct risk prediction model is needed for premenopausal breast cancer. We developed a risk prediction model for premenopausal breast cancer using 779,601 participants and 9,665 incident cases from 18 prospective studies within the Premenopausal Breast Cancer Collaborative Group (PBCCG), across North America (N=9), Europe (N=6), Australia (N=1), and Asia (N=2). Data were split, within each cohort, into training (2/3) and testing (1/3) datasets. Individual risk was assessed in five-year intervals, using variables reported at the start of the time interval. Cox proportional hazards regression was used to model risk factors in a backwards-selection method, stratified by cohort: age at menarche, age at first birth, parity, breastfeeding (months), height (cm), BMI (kg/m2), BMI at age 18, recent weight change (kg), alcohol consumption (drinks/week), first-degree family history of breast cancer, and personal history of benign breast disease. To enable the use of information from all cohorts despite differences in missing variables by design, cohorts were grouped by available variables and risk models were fit for each group. In cohorts with incomplete data, model coefficients were adjusted based on the correlation of covariates with the missing variables in the complete case dataset. Coefficients were meta-analyzed with inverse variance weighting to obtain final coefficients. Discrimination was evaluated in the testing dataset by calculation of the c-index. Work is ongoing to calibrate the model based on five-year absolute risk using GLOBOCAN continent- and age-specific incidence rates to represent baseline risk. The final model included age at menarche, parity, height, BMI, BMI at age 18, first-degree family history of breast cancer, and history of benign breast disease. Young adulthood BMI and BMI (at start of the 5-year risk interval) were associated with a decreased risk (Hazard Ratio (HR) (95% confidence interval (CI)) per 5 kg/m2 = 0.87 (0.81-0.93) and 0.90 (0.86-0.95), respectively) as was parity (HR (95% CI) = 0.92 (0.90-0.94)). Height was associated with increased risk (HR (95% CI) per 10 cm = 1.14 (1.07-1.21)), while history of benign breast disease and family history were associated with larger increases in risk (HR (95% CI) = 1.64 (1.30-2.06) and 1.76 (1.63-1.90), respectively). Model discrimination was higher than that reported for women under 50 years in existing breast cancer risk prediction models considering clinical factors (AUC (95% CI) = 0.61 (0.59-0.62)). Calibration of absolute 5-year risk is ongoing. Several factors driving risk prediction of postmenopausal breast cancer have similar influence on risk of premenopausal breast cancer, while family history has a stronger influence on premenopausal breast cancer risk. Our model demonstrates acceptable discrimination and will enable individual 5-year absolute risk prediction for premenopausal breast cancer. Citation Format: Kristen Brantley, Michael Jones, Minouk Schoemaker, Hazel Nichols, Anthony Swerdlow, Robert MacInnis, Roger Milne, Tess Clenenden, Yu Chen, Xiao-Ou Shu, Wei Zheng, Woon-Puay Koh, Jian-Min Yuan, Cari Kitahara, Martha Linet, Dale Sandler, Bernard A Rosner, Peter Kraft, A. Heather Eliassen. Development of an absolute risk prediction model for premenopausal breast cancer in an international consortium [abstract]. In: Proceedings of the 2023 San Antonio Breast Cancer Symposium; 2023 Dec 5-9; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2024;84(9 Suppl):Abstract nr PS10-09.

Abstract PS10-06: Contribution of a breast cancer polygenic risk score to mammography artificial intelligence models and breast density for long term breast cancer risk prediction

Abstract Artificial intelligence (AI) models based on deep learning of mammogram images have been developed for lesion detection and diagnosis as well as breast cancer (BC) risk. We and others have shown that these AI models, coupled with breast density, predict long-term breast cancer risk. Common germline genetic variation, in particular a polygenic risk score (PRS) that captures the cumulative impact of hundreds of common risk variants, has been shown associated with 3-fold increased BC risk. The BC-PRS may complement imaging risk factors to improve identification of women at high BC risk and better target prevention strategies. We evaluated the contribution of a BC-PRS to two imaging-based AI models (Transpara AI cancer detection system and Mirai 5-year risk) and breast density measures to long-term BC risk. We hypothesized that the BC-PRS will improve long-term risk prediction above AI and density measures. We examined this hypothesis within a cohort of 10,271 women enrolled to the Mayo Clinic Biobank between 2009 and 2015. Eligible women had no prior cancer, were ages 35 to 85, had a full-field digital screening mammogram within two years of enrollment and genotyping by Regeneron. Incident invasive and DCIS BC was identified through the Mayo tumor registry and follow-up questionnaires, and person years were calculated from date of mammogram through date of BC, last mammogram at Mayo or last questionnaire. We estimated Transpara AI malignancy score (1-10), 5-year Mirai risk and Volpara volumetric density measures (volumetric percent density, dense volume) on screening mammograms at enrollment and obtained clinical BI-RADS density. We calculated the BC-PRS using 268 SNPs (of the 313-SNP score) and the weighted sum of the SNPs using odds ratios from the largest BC GWAS of European ancestry to date. We performed cox proportional hazards regression adjusting for age and BMI to estimate hazard ratios (HRs), 95% confidence intervals (CI), and C-statistics (AUC) to describe the contribution of BC-PRS (per standard deviation (SD)) to the association of AI scores and breast density with BC risk. Likelihood ratio tests (LRT) and bootstrapping were used to compare model performance with vs. without inclusion of BC-PRS, at 10-years follow-up. Comparisons to the Tyrer-Cuzick model are ongoing. Over a median of 9.1 years (Interquartile range, 6.8 to 10.6), 446 incident BC were identified of which 285 invasive and 103 DCIS were confirmed by pathology report (60 were self-reported); 399 (256 invasive) occurred within 10 years and were used in primary analyses. Transpara score [HR per 1 unit=1.18 (1.14, 1.23), AUC=0.655 (0.628, 0.682)] and Mirai risk [HR per SD log risk=1.54 (1.43, 1.65), AUC=0.681 (0.655, 0.706)] were both associated with BC. All three breast density measures were also significantly associated with BC, including in models with AI scores (Table). Addition of the BC-PRS to models with AI scores and breast density improved both model fit (All PLRT &amp;lt; 0.001; Table) and discrimination, with increases in AUC ranging from 0.033 to 0.034 for models with Transpara and density and 0.025 to 0.026 for models with Mirai and density (all P-values &amp;lt; 0.001; Table). Analyses restricted to invasive BC showed identical results. But, the BC-PRS contribution at 5-year follow-up was attenuated. In this cohort study, we found BC-PRS contributed to long-term risk prediction of BC above imaging AI models and breast density. Imaging and genetic risk factors are complementary and may outperform existing risk models and provide more accurate risk assessment to aid in precision prevention strategies. Citation Format: Celine Vachon, Christopher Scott, Imon Banerjee, Ramon Correa Moderno, Dan Hursh, Matthew Jensen, Yvonne Wang, Kathleen Brandt, Karla Kerlikowske, Sandhya Pruthi, Fergus Couch, Stacey Winham. Contribution of a breast cancer polygenic risk score to mammography artificial intelligence models and breast density for long term breast cancer risk prediction [abstract]. In: Proceedings of the 2023 San Antonio Breast Cancer Symposium; 2023 Dec 5-9; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2024;84(9 Suppl):Abstract nr PS10-06.

Abstract PO2-28-09: Long-term Breast Cancer Risk Prediction in Black Women: External Validation of a Mammography-driven AI Model

Abstract Despite the demonstrated potential of artificial intelligence (AI) in breast cancer risk assessment for personalizing screening recommendations, further validation is required regarding AI model bias and generalizability. We performed external validation in Black women of a mammography-driven AI breast cancer risk model (Mirai) originally developed on screening cohorts primarily consisting of White women. In this institutional review board-approved, Health Insurance Portability and Protection Act (HIPAA)-compliant study under a waiver of consent, we retrospectively analyzed a case–cohort sample nested within the core academic breast cancer screening practice of BJC Healthcare, the hospital partner of Washington University in St. Louis. For the purposes of this validation study, relying on 2D digital mammography (DM) images, we focused on Black women presenting for annual DM screening (Selenia or Selenia Dimensions; Hologic) between 2008 and 2018. Eligible breast cancer cases were derived from all women with a breast cancer diagnosis (with associated biopsy-confirmed tumor pathology via institutional cancer registry) after negative (BI-RADS 1 or 2) DM screening 1 to 5 years prior to cancer diagnosis. We also identified a random sample of controls, defined as women who had negative (BI-RADS 1 or 2) DM screening, with 1 to 5 years of screening follow-up without a cancer diagnosis. Risk scores for all DM exams were calculated via the Mirai model. Performance was evaluated using concordance-index (C) analyses and associated 95% confidence intervals (CIs) for the entire cohort, as well as for study subgroups of invasive versus in-situ cancer and cancer molecular subtypes. We analyzed 1368 DM screening exams, including 672 DM exams from 391 women diagnosed with breast cancer (mean age, 58 years; standard deviation, 10 years) and 696 DM exams from 406 controls (mean age, 55 years; standard deviation, 10 years). The overall C-index was 0.62 [95% CIs 0.60–0.64] for all Black women, which was lower compared to previously reported validation results for Mirai in studies of similar design on predominantly White and racially diverse screening cohorts (C-index = 0.67-0.78). There was no evidence of a significant difference between invasive and in-situ cancer (C-index = 0.64 [95% CIs 0.61–0.66] vs. 0.64 [95% CIs 0.61–0.67]). Compared to other cancer molecular subtypes, performance was significantly higher among triple-negative (C-index = 0.67 [95% CIs 0.62–0.71]) and estrogen receptor (ER) negative cancer (C-index = 0.66 [95% CIs 0.61–0.70]). A previously developed mammography-driven AI model showed overall good performance in long-term breast cancer risk assessment in a dataset of Black women only, particularly for triple-negative and ER- cancer types. However, performance was lower compared to previously reported validation results from similar studies on predominantly White and racially diverse screening cohorts. Our results suggest that further refinements are needed towards more accurate breast cancer risk assessment in Black women. Long-term breast cancer risk prediction performance (C-indices) of Mirai in the full cohort of Black women and in study subgroups. Citation Format: Juanita Hernandez Lopez, Zhixin Sun, Shirin Shoushtari, Ulugbek Kamilov, Debbie Bennett, Aimilia Gastounioti. Long-term Breast Cancer Risk Prediction in Black Women: External Validation of a Mammography-driven AI Model [abstract]. In: Proceedings of the 2023 San Antonio Breast Cancer Symposium; 2023 Dec 5-9; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2024;84(9 Suppl):Abstract nr PO2-28-09.

Abstract PO5-08-10: Breast cancer risk prediction performance of polygenic risk score in Taiwanese female with dense breast: A nested case-control study

Abstract Introduction: Breast cancer is the most common cancer among women in Taiwan and the incidence rate of breast cancer has been increasing steadily over the past 30 years. To address this, a nationwide screening program with biennial mammography for women aged 40-69 was implemented in 2004, which has contributed to a significant reduction in breast cancer mortality. Asian women are more likely to have dense breast tissue, and studies have shown a correlation between breast density and the risk of developing breast cancer, with higher breast density values associated with increased risk. Polygenic risk scores (PRS) are calculated based on an individual's genetic information and can estimate their risk of developing a particular disease. Although PRS can identify low-penetrance risk variants, its use in breast cancer risk estimation is typically limited to gene-only models, and integration with breast cancer screening databases is rare. Hence, the aim of this specific study was to evaluate the predictive capacity of PRS in women with dense breast tissue by exploring potential genetic markers and constructing a PRS to investigate the genetic factors associated with clinical characteristics and the risk of developing breast cancer. Methods: The PRS was developed using genetic information from the Taiwan Precision Medicine Initiative genotyping data. A total of 6335 patients were enrolled, and 101 single nucleotide polymorphisms (SNPs) were selected as candidate markers based on the PGS Catalog (PGS000001). Clinical characteristics of the study cohort were summarised using median and range, or frequency and percentage. The distribution of characteristics between breast cancer and controls was estimated using an independent two-sample t-test, chi-square, and Fisher’s exact test. The association of the PRS and related characteristics with breast cancer risk in the study cohort was estimated using univariate and multivariate binomial logistic regression. Harrel’s C-index was reported to demonstrate the predictive performance of PRS in both univariate and multivariate models. All p-values were two-sided, and a p less than .05 is considered statistically significant. All analyses were performed using R 4.1.2 (R core team, 2023) Results: The results showed that breast cancer patients constituted a higher proportion of individuals in PRS Q4 (37.8% vs 24.8% in controls). The model's predictive performance for breast cancer risk increased from 0.565 (95% CI = 0.520-0.611) to 0.699 (95% CI = 0.644-0.755) by adding related clinical characteristics compared to the PRS-only model. Patients characterized with benign breast disease (OR = 0.50, 95% CI = 0.31-0.79, P = 0.004) showed a decreased breast cancer risk, and patients with mastalgia or palpable breast lesion (OR = 6.87, 95% CI = 4.29-10.8, P &amp;lt; 0.001) predicted significant greater breast cancer risk. Subgroup analysis of patients without breast symptoms was conducted as they may be overlooked or underestimated for breast cancer screening. For dense breast patients without symptoms, the high PRS group (Q4) consistently showed a significantly elevated breast cancer risk compared to the low PRS group (Q1-Q3) in both univariate (OR = 2.25, 95% CI = 1.43-3.50, P &amp;lt; 0.001) and multivariate analyses (OR = 2.22, 95% CI = 1.41-3.46, P &amp;lt; 0.001). Conclusion: Breast cancer screening has undergone a shift from a general approach to a personalized, risk-based approach. Besides breast cancer screening using family history as the only risk factor, our proposed model identifies both genetic and clinical risk factors using big data analyses including the EHR, cancer screening, and registry from a single institute. The study suggests that integrating PRS into personalized screening strategies could improve risk prediction for Taiwanese females with dense breasts without prominent symptoms. Table 1. Clinical characteristics of study cohort (n=6335). Table 2. Association of the PRS quartiles in breast cancer risk. Table 3. Predictive performance of PRS for breast cancer risk. Citation Format: Chih Yean Lum, Chih Chiang Hung, Chi-Cheng Huang, Tzu-Hung Hsiao, Sin-Hua Moi. Breast cancer risk prediction performance of polygenic risk score in Taiwanese female with dense breast: A nested case-control study [abstract]. In: Proceedings of the 2023 San Antonio Breast Cancer Symposium; 2023 Dec 5-9; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2024;84(9 Suppl):Abstract nr PO5-08-10.

Promoting Artificial Intelligence for Global Breast Cancer Risk Prediction and Screening in Adult Women: A Scoping Review.

Background: Artificial intelligence (AI) algorithms can be applied in breast cancer risk prediction and prevention by using patient history, scans, imaging information, and analysis of specific genes for cancer classification to reduce overdiagnosis and overtreatment. This scoping review aimed to identify the barriers encountered in applying innovative AI techniques and models in developing breast cancer risk prediction scores and promoting screening behaviors among adult females. Findings may inform and guide future global recommendations for AI application in breast cancer prevention and care for female populations. Methods: The PRISMA-SCR (Preferred Reporting Items for Systematic reviews and Meta-Analyses extension for Scoping Reviews) was used as a reference checklist throughout this study. The Arksey and O'Malley methodology was used as a framework to guide this review. The framework methodology consisted of five steps: (1) Identify research questions; (2) Search for relevant studies; (3) Selection of studies relevant to the research questions; (4) Chart the data; (5) Collate, summarize, and report the results. Results: In the field of breast cancer risk detection and prevention, the following AI techniques and models have been applied: Machine and Deep Learning Model (ML-DL model) (n = 1), Academic Algorithms (n = 2), Breast Cancer Surveillance Consortium (BCSC), Clinical 5-Year Risk Prediction Model (n = 2), deep-learning computer vision AI algorithms (n = 2), AI-based thermal imaging solution (Thermalytix) (n = 1), RealRisks (n = 2), Breast Cancer Risk NAVIgation (n = 1), MammoRisk (ML-Based Tool) (n = 1), Various MLModels (n = 1), and various machine/deep learning, decision aids, and commercial algorithms (n = 7). In the 11 included studies, a total of 39 barriers to AI applications in breast cancer risk prediction and screening efforts were identified. The most common barriers in the application of innovative AI tools for breast cancer prediction and improved screening rates included lack of external validity and limited generalizability (n = 6), as AI was used in studies with either a small sample size or datasets with missing data. Many studies (n = 5) also encountered selection bias due to exclusion of certain populations based on characteristics such as race/ethnicity, family history, or past medical history. Several recommendations for future research should be considered. AI models need to include a broader spectrum and more complete predictive variables for risk assessment. Investigating long-term outcomes with improved follow-up periods is critical to assess the impacts of AI on clinical decisions beyond just the immediate outcomes. Utilizing AI to improve communication strategies at both a local and organizational level can assist in informed decision-making and compliance, especially in populations with limited literacy levels. Conclusions: The use of AI in patient education and as an adjunctive tool for providers is still early in its incorporation, and future research should explore the implementation of AI-driven resources to enhance understanding and decision-making regarding breast cancer screening, especially in vulnerable populations with limited literacy.

Disentangling the relationships of body mass index and circulating sex hormone concentrations in mammographic density using Mendelian randomization.

Mammographic density phenotypes, adjusted for age and body mass index (BMI), are strong predictors of breast cancer risk. BMI is associated with mammographic density measures, but the role of circulating sex hormone concentrations is less clear. We investigated the relationship between BMI, circulating sex hormone concentrations, and mammographic density phenotypes using Mendelian randomization (MR). We applied two-sample MR approaches to assess the association between genetically predicted circulating concentrations of sex hormones [estradiol, testosterone, sex hormone-binding globulin (SHBG)], BMI, and mammographic density phenotypes (dense and non-dense area). We created instrumental variables from large European ancestry-based genome-wide association studies and applied estimates to mammographic density phenotypes in up to 14,000 women of European ancestry. We performed analyses overall and by menopausal status. Genetically predicted BMI was positively associated with non-dense area (IVW: β = 1.79; 95% CI = 1.58, 2.00; p = 9.57 × 10-63) and inversely associated with dense area (IVW: β = - 0.37; 95% CI = - 0.51,- 0.23; p = 4.7 × 10-7). We observed weak evidence for an association of circulating sex hormone concentrations with mammographic density phenotypes, specifically inverse associations between genetically predicted testosterone concentration and dense area (β = - 0.22; 95% CI = - 0.38, - 0.053; p = 0.009) and between genetically predicted estradiol concentration and non-dense area (β = - 3.32; 95% CI = - 5.83, - 0.82; p = 0.009), although results were not consistent across a range of MR approaches. Our findings support a positive causal association between BMI and mammographic non-dense area and an inverse association between BMI and dense area. Evidence was weaker and inconsistent for a causal effect of circulating sex hormone concentrations on mammographic density phenotypes. Based on our findings, associations between circulating sex hormone concentrations and mammographic density phenotypes are weak at best.

Abstract LB328: Measurement of ER and Ki67 coexpression in cytokeratin positive epithelial cells in benign breast disease (BBD) percutaneous biopsies

Abstract Over 1 million women in the U.S. undergo radiologically guided percutaneous biopsies annually. Of these, 75% reveal benign breast disease (BBD), typically surrounded by normal breast lobules, which are proposed to reflect the biology of the breast parenchyma diffusely. With aging, normal lobules undergo involution, and delays in this process are linked to increased breast cancer (BC) risk. Because BBD patients are at increased risk of developing BC, and pathologic classification only stratifies risks among groups of patients, novel methods for individual BC risk prediction are needed to guide precision management. Prior literature has shown that most proliferating epithelial cells in normal lobules do not express estrogen receptor (i.e., ER-/Ki67+), whereas co-expression is frequent in BC precursors and in BCs (i.e., ER+/Ki67+). Thus we developed and applied a novel quantitative dual ER/Ki67 stain to BBD biopsies from the well-characterized contemporary Mayo Clinic BBD cohort (2002-2013) to assess differences by levels of lobular involution and tissue type (BBD vs. normal), as well as associations with BC risk. We identified 100 BBD biopsies preceding BC (cases) and 100 biopsies from cancer-free women (controls), frequency-matched on age and length of follow-up. A multiplex immunofluorescence assay was developed to detect ER/Ki67 coexpression in cytokeratin positive epithelial cells in normal lobules or BBD lesions in FFPE tissue sections. An inverse normal transformation was applied to cell counts and proportions and associations were assessed for levels of lobular involution, tissue type (normal lobule vs. BBD), and case status using t-tests and linear regression models. The final data set included 1152 normal lobules or BBD lesions in biopsies of 177 subjects (88 cases and 89 controls). In univariate analyses, normal lobules showing delayed age related involution included higher counts of ER-/Ki67+, ER+/Ki67- and ER+/Ki67+ cells, and the lowest proportion of ER+/Ki67- cells (p&amp;lt;0.03; all). In separate analyses comparing presence vs. absence, detection of any ER+/Ki67+ cells was suggestively more common in non-involuted vs. involuted lobules. Compared with normal lobules, BBD showed higher counts and proportions of ER+/Ki67- and ER+/Ki67+ cells and lower values for ER-/Ki67+ (all p&amp;lt;0.001). In these preliminary results, marker expression in normal lobules and in BBD did not differ by case-control status. Preliminarily, we find that ER+/Ki67+ cells are increased in normal lobules with delayed age-related involution and in BBD lesions, consistent with their associations with increased BC risk. These findings are consistent with the risk reducing effects of endocrine chemoprevention and may serve as useful biomarkers in prevention therapy. In this analysis, limited to only a few lobules and BBD lesions per biopsy, ER+/Ki67+ cell counts and proportions did not vary significantly by case-control status. Citation Format: Melody L. Stallings-Mann, Robert A. Vierkant, Stacey J. Winham, Bryan McCauley, Matt R. Jensen, Sophie Noel, Laura Pacheco-Spann, Celine M. Vachon, Amy C. Degnim, Derek C. Radisky, Mark E. Sherman. Measurement of ER and Ki67 coexpression in cytokeratin positive epithelial cells in benign breast disease (BBD) percutaneous biopsies [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 2 (Late-Breaking, Clinical Trial, and Invited Abstracts); 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84(7_Suppl):Abstract nr LB328.

249 Defining the Role of Hedgehog Signaling in Breast Cancer Risk of non-Hispanic Black Women

OBJECTIVES/GOALS: The molecular basis of increased risk of triple negative breast cancer in non-Hispanic Black women represents a critical knowledge gap that this research is designed to address; successful completion of this work could lead to better prevention, earlier stage diagnoses, and possible discovery of novel therapeutic strategies for this population. METHODS/STUDY POPULATION: We have recently generated a living tissue cohort of 11 non-Hispanic Black and 25 non-Hispanic White women who underwent breast surgery at Mayo Clinic. Gene expression profiling of normal breast tissue from this cohort has identified a pattern of gene expression differences that have been associated with the development of basal breast cancer and are also reflective of Hedgehog (Hh) signaling. We will identify protein-based biomarkers for Hedgehog signaling within normal breast tissue using immunohistochemistry methods. We will culture primary human mammary epithelial cells and further separate luminal and myoepithelial cells using flow cytometry to then decipher Hedgehog signaling. RESULTS/ANTICIPATED RESULTS: We anticipate identifying and localizing protein-based biomarkers for Hedgehog signaling within myoepithelial cells of non-Hispanic Black women. Using our findings, we aim to create a biomarker risk model for triple negative breast cancer and validate this model within a separate and larger cohort of women to predict breast cancer risk. DISCUSSION/SIGNIFICANCE: In addition to immediate benefits from improved risk prediction, the proposed work has the potential to provide new insight into the driving forces underlying basal breast carcinogenesis and the distinct biological differences that distinguish non-Hispanic Black women from non-Hispanic White women.

Abstract 3429: Using metabolomics to identify biomarkers mediating the associations of adiposity in childhood and early adulthood with mammographic breast density in premenopausal women

Abstract Introduction: Mammographic breast density (MBD), a strong predictor of breast cancer risk, is highly influenced by body mass index (BMI) in childhood and early adulthood, but the mechanisms elucidating this relationship are undetermined. To our knowledge, no study has explored the mediating role of biomarkers on the relationship between early-life adiposity and MBD. The goal of this study is to discover metabolites that mediate the relationship between BMI at ages 10 and 18 with MBD in premenopausal women. Methods: This study includes premenopausal women who had their screening mammogram at Washington University in St. Louis, MO, and provided a fasting blood sample. Metabolon performed untargeted metabolomic profiling, detecting 1,074 metabolites. Metabolites missing &amp;gt;300 observations were excluded, leaving 828 metabolites; the other missing values were imputed using the nearest neighbor method. To mitigate batch effect, we normalized the metabolite peak area data using ComBat. Volumetric percent density (VPD) was calculated in 700 women using Volpara software. BMI at age 10 was estimated using the Stunkard pictogram, and BMI at age 18 was calculated from self-reported weight at 18 and height at study initiation. To assess the mediating role of the 828 metabolites, we performed high dimensional mediation analysis using the hima R package adjusting for potential confounders. Missing values in the covariates and BMI measures included in the mediation analysis were imputed using multivariate imputation by chain equations. Associations were considered significant if FDR p-value &amp;lt;0.1. Results: Four metabolites (glutamate, beta-cryptoxanthin, phytanate, and cortolone glucuronide (1)) mediated the relationship between BMI at age 10 and VPD; and 2 metabolites (glutamate, beta-cryptoxanthin) mediated the relationship between BMI at age 18 and VPD. Glutamate and beta-cryptoxanthin significantly mediated the relationship between both BMI measures and VPD, but glutamate was the strongest mediator across both time points. Glutamate mediated 6.7% (FDR p-value=0.06) and 9.3% (FDR p-value=0.008) of the total effect between BMI at age 10 and 18, respectively, on VPD. Beta-cryptoxanthin mediated 4.1% (FDR p-value=0.06) and 6.3% (FDR p-value=0.04), of the total effect between BMI at age 10 and 18, respectively, on VPD. Additionally, phytanate mediated 2.6% (FDR p-value=0.06) and cortolone glucuronide (1) mediated 2.0% (FDR p-value=0.07) of the total effect between BMI at age 10 and VPD. Conclusion: Amino acid, lipid, cofactor/vitamin, and xenobiotic metabolites mediated the associations of BMI in early-life and VPD in premenopausal women. This innovative study offers insight into the biological mechanisms underlying the link between early-life adiposity and MBD, which can support future research into breast cancer prevention. Citation Format: Kayla R. Getz, Myung Sik Jeon, Lili Liu, Lei Liu, Chongliang Luo, Jingqin Luo, Adetunji T. Toriola. Using metabolomics to identify biomarkers mediating the associations of adiposity in childhood and early adulthood with mammographic breast density in premenopausal women [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 1 (Regular Abstracts); 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84(6_Suppl):Abstract nr 3429.

Abstract 6619: Dynamic prediction with repeated mammogram images improves 5-year breast cancer risk performance

Abstract Background: Increasing use of digital screening mammography opens potential to add individualized dynamic measures of breast image history to predict future breast cancer risk. Current prediction models use a single mammogram measure, usually breast density, at one point in time, to add to prediction models based on established risk factors. Our previous work has shown that change in each breast over time is related to risk of diagnosis with breast cancer. Methods: We apply a dynamic prediction model that includes full field digital mammogram in each breast to predict future 5-year risk. Data from Joanne Knight Breast Health Cohort (JKBHC) serve as training population with 290 pathology-confirmed breast cancer cases and 658 controls sampled with a total number of 8,712 craniocaudal (CC) view mammograms for analysis. Nurses’ Health Study 2 (NHS2) serves as the external validation with 188 cases and 462 controls sampled with a total number of 3,712 CC-views. The dynamic risk prediction model aims to estimate the probability of future breast cancer risk conditional on the past history of the longitudinal mammograms. We train the model on the JKBH cohort, retaining 20% for testing and 15% for validation. We report a time dependent AUC when evaluating model performance. Results: The mean age of women was 56.6 years at entry to cohort; 81.3% were white. Adding trajectory of mammogram images in each breast to a model improved performance. The mean number of mammograms conditional on 5 years of observation in JKBHC was 3.36 (1.49) and in NHS2 2.36 (1.53). We obtained a 5-year dynamic AUC of 0.81 (0.05) in the JKBHC internal validation and 0.77 (0.06) in the NHS2 external validation. Model performance was comparable in pre- and postmenopausal women. Because breasts with high mammographic density hinder detection of tumors we stratify analyses by dense and non-dense at baseline. Among women with dense breasts at baseline (BIRADS C and D), we observe a dynamic 5-year risk AUC = 0.77. In the subgroup with less dense breasts (BIRADS A and B) we observe AUC = 0.74. Conclusions: With this approach we can accommodate the irregularly observed time-varying images and project only partial information prior to doing the prediction. Adding trajectory of mammogram history in each breast improves 5-year breast cancer risk prediction beyond models used in clinic today. Citation Format: Graham A. Colditz, Rulla M. Tamimi, Bernard Rosner, Shu Jiang. Dynamic prediction with repeated mammogram images improves 5-year breast cancer risk performance [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 1 (Regular Abstracts); 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84(6_Suppl):Abstract nr 6619.