Kurian, Allison W.

Abstract Polygenic risk scores (PRSs), which quantify inherited susceptibility to complex traits and diseases, have emerged as valuable tools for risk stratification and precision medicine. Despite their promise, PRS developed on European cohorts often demonstrate substantially reduced predictive accuracy in non-European populations, due to differences in genetic architecture. The disproportionate representation of European ancestry cohorts in genome-wide association studies (GWAS) leads to inequitable deployment of PRS technologies across diverse populations. Here, we introduce PRANA (Polygenic Risk Adaptation via Neural-network Architecture), a deep learning framework that adapts an existing PRS developed on one population to other ancestries. Unlike methods that require large-scale GWAS in the target population, PRANA leverages pre-trained PRS models derived from European cohorts and adapts them using modestly sized cohorts from the target population. We evaluated PRANA on seven complex traits in South Asian, East Asian and Ashkenazi Jewish populations, as well as in selected smaller East Asian subpopulations where the scarcity of training data poses a particular challenge. PRANA mostly improved predictive performance of the baseline PRS models by 5%-20% in terms of effect size (β) and Nagelkerke’s R², and, in most cases, outperformed existing cross-ancestry multi-PRS approaches. These results highlight PRANA as a scalable and practical strategy to reduce disparities in genomic risk prediction and advance the equitable application of PRS in diverse populations.

Abstract Background The association of germline pathogenic and likely pathogenic variants (GPVs) in hereditary breast cancer genes with underlying tumor biology and clinical outcomes remains incompletely understood. This study characterized differences in somatic alterations and intrinsic subtypes between sporadic and hereditary breast cancers associated with GPVs in ATM, BRCA1, BRCA2, CHEK2, or PALB2. Methods This retrospective cohort study included women with breast cancer and an ATM, BRCA1, BRCA2, CHEK2, or PALB2 GPV who had undergone tumor sequencing and whole transcriptome RNA expression analysis. Clinicopathological features, intrinsic subtypes, somatic alterations, and survival were compared by GPV status and immunohistochemistry-defined subtype and to sporadic cases. All statistical significance tests were 2-sided. Results The 4988 women with breast cancer in this study included 98 BRCA1, 126 BRCA2, 74 PALB2, 54 ATM, and 83 CHEK2 GPVs. Compared with sporadic cases, hormone receptor–positive/(ERBB2) formerly HER2–negative tumors in BRCA1 GPVs were statistically significantly enriched for basal subtype (45.5% vs 11.4%; P < .001), while CHEK2 carriers had a higher prevalence of luminal A subtype (80.4% vs 60.3%; P = .006). In hormone receptor–positive/(ERBB2)-negative breast cancers, BRCA1 GPVs were enriched for TP53 alterations (84.6% vs 29.8%; q < 0.001), ATM GPVs with FGFR1 alterations (35.4% vs 12.7%; q = 0.04), and BRCA2 GPVs with APC alterations (10.1% vs 1.5%; q = 0.004). Conversely, BRCA2 GPVs were inversely associated with PIK3CA alterations (13.0% vs 34.1%; q = 0.005) and CHEK2 GPVs with TP53 alterations (8.0% vs 29.8%; q = 0.02). Conclusions The GPVs in BRCA1, BRCA2, ATM, CHEK2, and PALB2 are associated with distinct intrinsic breast cancer subtypes and somatic genomic alterations. These findings may enhance precision in risk stratification and guide personalized treatment strategies.

The NCCN Clinical Practice Guidelines in Oncology (NCCN Guidelines) for Genetic/Familial High-Risk Assessment: Breast, Ovarian, Pancreatic, and Prostate are intended to serve as a resource for health care providers to identify individuals who may benefit from cancer risk assessment and genetic counseling and testing; help guide decisions related to genetic testing; and facilitate a multidisciplinary approach in the comprehensive care of individuals at increased risk for hereditary breast, ovarian, pancreatic, and prostate cancer. The current guidelines focus primarily on assessment of pathogenic and likely pathogenic (P/LP) variants associated with increased risk of breast, ovarian, pancreatic, and prostate cancer and recommended approaches to genetic counseling/testing and care strategies in individuals with these P/LP variants associated with increased risk of these cancers. These NCCN Guidelines Insights summarize the panel’s most recent recommendations regarding screening for prostate cancer and pancreas cancer, as well as testing criteria for nonepithelial ovarian cancer.

PURPOSE Automated curation of breast cancer treatment data with minimal human involvement could accelerate the collection of statewide and nationwide evidence for patient management and assessing the effectiveness of treatment pathways. The primary challenges are the complexity and inconsistency of structured clinical data streams and accurate extraction of this information from free-text clinical narratives. MATERIALS AND METHODS We proposed a hybrid two-phase information extraction framework that combined a Unified Medical Language System parser (phase-1) with a fine-tuned large language model (LLM; phase-2) to extract longitudinal treatment timelines from time-stamped clinical notes. Our framework was developed through end-to-end joint learning as a question-answering model, where the model was trained to simultaneously answer five questions, each corresponding to a specific treatment. RESULTS We fine-tuned and internally validated the model on 26,692 patients with breast cancer (diagnosed between 2013 and 2020) receiving treatment at Mayo Clinic and externally validated the model on 162 randomly selected patients from Stanford Healthcare. Zero-shot LLM (out-of-the-box) had high specificity but low sensitivity, indicating that although these frameworks are useful for generic language understanding, they are lacking in terms of targeted clinical tasks. The proposed model achieved 0.942 average AUROC on the internal and 0.924 on the external data, demonstrating only marginal drop in performance when evaluated on external. The proposed model also achieved better trade-off between sensitivity (average: 79.2%) and specificity (average: 76.2%) compared with rule-based (average sensitivity: 70.5%, average specificity: 68.1%) and structured codes (average sensitivity: 64.1%, average specificity: 83.5%). CONCLUSION The proposed framework can extract temporal information about cancer treatments from various time-stamped clinic notes, regardless of the setting of treatment administration (inpatient or outpatient) or time frame. To support the cancer research community for such data curation and longitudinal analysis, we have packaged the code as a docker image, which needs minimal system reconfiguration and shared with an open-source academic license.

10512 Background: Polygenic risk scores (PRSs) combine information from single-nucleotide polymorphisms (SNPs) across the genome to explain substantial genetic breast cancer (BC) susceptibility. Previous studies have demonstrated that a multiple-ancestry PRS (MA-385) based on 56 ancestry-informative and 329 BC-associated SNPs is accurate for diverse populations and ranks among the most important known factors affecting the risk of BC development. However, medical guidelines call for more evidence before endorsing the clinical use of PRS, including studies to evaluate possible interactions of SNPs with environmental and hormonal risk factors. Here, we use longitudinal outcomes from the UK Biobank (UKB) to explore interactions of MA-385 and the five most informative individual BC-associated SNPs with the widely used Tyrer-Cuzick (TC) risk model and individual TC risk factors. Methods: The study cohort included 197,509 female UKB participants with no history of cancer at the time of study enrollment. We used Cox proportional hazards models to test associations of MA-385, individual BC SNPs, TC, and individual TC risk factors with BC outcomes. Effect modification of MA-385 and individual SNPs by clinical risk factors was evaluated by including interaction terms in the models. All models were adjusted for age at UKB enrollment and 10 principal components. Hazard ratios (HRs) and 95% confidence intervals (CIs) are reported per standard deviation, and significance tests were performed using two-sided p-values based on likelihood-ratio test statistics. Results: After a median follow-up of 11.8 years, 7,419 (3.8%) participants were diagnosed with BC.In a model including both MA-385 and TC, MA-385 was a significantly better predictor of BC development (HR = 1.56; 95% CI 1.53-1.60, p = 3.2 x 10 -329 ) over the TC model (HR = 1.21; 95% CI 1.18-1.23; p = 5.4 x 10 -69 ). We found no evidence of interaction between MA-385 and TC ( p = 0.31) , nor between individual SNPs and TC. After Bonferroni adjustment for multiple testing (102 tests), we found no evidence of interaction between MA-385 or individual SNPs with clinical TC factors. The strongest evidence for interaction was found between a BC SNP on chromosome 2 (rs13387042) and age at UKB enrollment (unadjusted p = 0.002; Bonferroni adjusted p = 0.23). Conclusions: In a longitudinal analysis of UKB, MA-385 was a highly significant predictor of BC risk and substantially improved prediction over TC. In contrast, interactions of MA-385, and individual BC SNPs, with TC and individual clinical factors in the TC model were not statistically significant. Clinical TC factors have minimal, if any, impact on the strength of the association between MA-385 and BC. These results may help to alleviate concerns expressed in guidelines that SNPs interact with environmental or hormonal risk factors.

Abstract There are limited prospective data on whether established risk factors modify breast cancer risk in women with pathogenic variants (PV) in BRCA1/2 and virtually no risk modification data for ATM, CHEK2, or PALB2. We conducted a nested case–control study in the Women’s Health Initiative (WHI), randomly selecting women with and without breast cancer for DNA sequencing. We evaluated breast cancer odds associated with obesity, family cancer history, smoking, alcohol, parity, breastfeeding, oophorectomy, tubal ligation, neighborhood socioeconomic status, and menopausal hormone therapy (MHT) with estrogen and progestin or estrogen only in PV carriers and noncarriers. In exploratory analyses, we grouped genes by established predisposition for estrogen receptor (ER)-positive (ATM and CHEK2) or ER-negative (BRCA1 and PALB2) disease. Multivariable models with interaction terms were used to assess differential risk modification by PVs. Among 12,957 WHI participants, 287 carried PVs. Breastfeeding was modestly associated with reduced risk for PALB2 [OR = 0.08; 95% confidence interval (CI), 0.00–0.92; P value = 0.042]. With one-sided 95% CI, power was sufficient to exclude OR ≥2.0 with obesity for ATM and BRCA2; smoking and alcohol for CHEK2; no breastfeeding for ATM; no oophorectomy for BRCA2 and CHEK2; no tubal ligation for CHEK2; and neighborhood socioeconomic status for all genes. Estrogen + progestin MHT was modestly associated with increased risk for ER-positive PVs (OR = 7.31; 95% CI, 1.14–64.20; P = 0.036). PVs did not modify risk (interaction P ≥ 0.05). BRCA1/2, ATM, CHEK2, and PALB2 PV carriers do not have breast cancer OR ≥2.0 with many established risk factors. However, MHT warrants additional study in PV carriers. Significance: There is limited information on whether established risk factors increase breast cancer risk from PVs. In the WHI, PV carriers had no substantial (≥2-fold) increase with most risk factors, except potentially MHT in ATM or CHEK2 carriers. The results may inform counseling and research on MHT.

Abstract Background: Breast cancer has been associated with monogenic, polygenic, and epidemiologic (clinical, reproductive, and lifestyle) risk factors, but studies evaluating the combined effects of these factors have been limited. Methods: We extended previous work in breast cancer risk modeling, incorporating pathogenic variants (PV) in six breast cancer predisposition genes and a 105-SNP polygenic risk score (PRS), to include an epidemiologic risk score (ERS) in a sample of non-Hispanic White women drawn from prospective cohorts and population-based case–control studies, with 23,518 cases and 22,832 controls, from the Cancer Risk Estimates Related to Susceptibility (CARRIERS) Consortium. Results: The model predicts 4.4-fold higher risk of breast cancer for postmenopausal women with no predisposition PV and median PRS, but with the highest versus lowest ERS. Overall, women with CHEK2 PVs had >20% lifetime risk of breast cancer. However, 15.6% of women with CHEK2 PVs and a family history of breast cancer, and 45.1% of women with CHEK2 PVs but without a family history of breast cancer, had low (<20%) predicted lifetime risk and thus were below the threshold for MRI screening. CHEK2 PV carriers at the 10th percentile of the joint distribution of ERS and PRS, without a family history of breast cancer, had a predicted lifetime risk similar to the general population. Conclusions: These results illustrate that an ERS, alone and combined with the PRS, can contribute to clinically relevant risk stratification. Impact: Integrating monogenic, polygenic, and epidemiologic risk factors in breast cancer risk prediction models may inform personalized screening and prevention efforts.

PURPOSE Patient-centered outcomes (PCOs) are pivotal in cancer treatment, as they directly reflect patients' quality of life. Although multiple studies suggest that factors affecting breast cancer–related morbidity and survival are influenced by treatment side effects and adherence to long-term treatment, such data are generally only available on a smaller scale or from a single center. The primary challenge with collecting these data is that the outcomes are captured as free text in clinical narratives written by clinicians. MATERIALS AND METHODS Given the complexity of PCO documentation in these narratives, computerized methods are necessary to unlock the wealth of information buried in unstructured text notes that often document PCOs. Inspired by the success of large language models (LLMs), we examined the adaptability of three LLMs, GPT-2, BioGPT, and PMC-LLaMA, on PCO tasks across three institutions, Mayo Clinic, Emory University Hospital, and Stanford University. We developed an open-source framework for fine-tuning LLM that can directly extract the five different categories of PCO from the clinic notes. RESULTS We found that these LLMs without fine-tuning (zero-shot) struggle with challenging PCO extraction tasks, displaying almost random performance, even with some task-specific examples (few-shot learning). The performance of our fine-tuned, task-specific models is notably superior compared with their non–fine-tuned LLM models. Moreover, the fine-tuned GPT-2 model has demonstrated a significantly better performance than the other two larger LLMs. CONCLUSION Our discovery indicates that although LLMs serve as effective general-purpose models for tasks across various domains, they require fine-tuning when applied to the clinician domain. Our proposed approach has the potential to lead more efficient, adaptable models for PCO information extraction, reducing reliance on extensive computational resources while still delivering superior performance for specific tasks.

PURPOSE To develop recommendations for germline mutation testing for patients with breast cancer. METHODS An ASCO–Society of Surgical Oncology (SSO) panel convened to develop recommendations based on a systematic review and formal consensus process. RESULTS Forty-seven articles met eligibility criteria for the germline mutation testing recommendations; 18 for the genetic counseling recommendations. RECOMMENDATIONS BRCA1/ 2 mutation testing should be offered to all newly diagnosed patients with breast cancer ≤65 years and select patients >65 years based on personal history, family history, ancestry, or eligibility for poly(ADP-ribose) polymerase (PARP) inhibitor therapy. All patients with recurrent breast cancer who are candidates for PARP inhibitor therapy should be offered BRCA1/ 2 testing, regardless of family history. BRCA1/ 2 testing should be offered to women who develop a second primary cancer in the ipsilateral or contralateral breast. For patients with prior history of breast cancer and without active disease, testing should be offered to patients diagnosed ≤65 years and selectively in patients diagnosed after 65 years, if it will inform personal and family risk. Testing for high-penetrance cancer susceptibility genes beyond BRCA1/ 2 should be offered to those with supportive family histories; testing for moderate-penetrance genes may be offered if necessary to inform personal and family cancer risk. Patients should be provided enough pretest information for informed consent; those with pathogenic variants should receive individualized post-test counseling. Variants of uncertain significance should not impact management, and patients with such variants should be followed for reclassification. Referral to providers experienced in clinical cancer genetics may help facilitate patient selection and interpretation of expanded testing, and provide counseling of individuals without pathogenic germline variants but with significant family history. Additional information is available at www.asco.org/breast-cancer-guidelines .

The NCCN Guidelines for Genetic/Familial High-Risk Assessment: Breast, Ovarian, and Pancreatic focus primarily on assessment of pathogenic/likely pathogenic (P/LP) variants associated with increased risk of breast, ovarian, pancreatic, and prostate cancer, including BRCA1 , BRCA2 , CDH1 , PALB2 , PTEN , and TP53 , and recommended approaches to genetic counseling/testing and care strategies in individuals with these P/LP variants. These NCCN Guidelines Insights summarize important updates regarding: (1) a new section for transgender, nonbinary and gender diverse people who have a hereditary predisposition to cancer focused on risk reduction strategies for ovarian cancer, uterine cancer, prostate cancer, and breast cancer; and (2) testing criteria and management associated with TP53 P/LP variants and Li-Fraumeni syndrome.