Citations

Abstract Background Artificial intelligence (AI)–based, mammography breast cancer risk prediction models show improved discriminatory accuracy relative to clinical risk models. However, data on their calibration are limited. This study compared model performance of 3 clinical breast cancer risk models: Gail, Tyrer–Cuzick v8, and Breast Cancer Surveillance Consortium v3 with the MIRAI AI-risk model. Methods Digital mammograms were ascertained from a screening mammography cohort of 12 308 women within the Mayo Clinic Biobank with 250 incident breast cancers (176 invasive) within 5 years. We predicted 5-year breast cancer risk, estimated discriminatory accuracy (C-index), and calibration (observed-to-expected ratio) of both overall and invasive breast cancer and compared estimates using bootstrapping approaches. Results MIRAI demonstrated similar or improved discriminatory accuracy of overall breast cancer (C-index = 0.71, 95% confidence interval [CI] = 0.68 to 0.74) and invasive breast cancer (C-index = 0.71, 95% CI = 0.67 to 0.75) compared with clinical models (overall breast cancer: C-index = 0.59-0.68; invasive breast cancer: C-index = 0.60-0.68). MIRAI’s calibration for risk of overall breast cancer (observed-to-expected ratio = 0.96, 95% CI = 0.85 to 1.08) was improved compared with Gail (observed-to-expected ratio = 1.22, 95% CI = 1.07 to 1.38) and Breast Cancer Surveillance Consortium (observed-to-expected ratio = 1.38, 95% CI = 1.22 to 1.56) but similar to TC with volumetric percent density and polygenic risk score (observed-to-expected ratio = 0.99, 95% CI = 0.87 to 1.13). However, for low-risk women (approximately 50%), MIRAI overestimated risk of overall breast cancer. MIRAI also overestimated risk of invasive breast cancer across the risk spectrum (observed-to-expected ratio = 0.68, 95% CI = 0.58 to 0.78), whereas clinical models had good calibration (observed-to-expected ratio = 0.86–0.99). Conclusion MIRAI demonstrated stronger discriminatory accuracy than clinical models for 5-year overall and invasive breast cancer risk prediction but overestimated risk for both breast cancer endpoints. AI-based risk models should consider discriminatory accuracy and calibration for invasive cancer before implementation.

Abstract Deep learning models enable the prediction of clinical endpoints from whole-slide images (WSIs), but many such models function as “black boxes”, lacking transparency about whether and which histomorphological patterns drive their predictions, hindering interpretability and clinical adoption. Here we propose a human-in-the-loop explanation framework, MorphoXAI, which provides both local and global interpretability for deep learning models by incorporating human-expert interpretations. At the global level, it reveals the histomorphological patterns on which the model consistently relies to distinguish between classes of WSIs, as well as the patterns associated with confusion between classes. At the local level, it indicates which of these patterns are used in the prediction of an individual WSI and which regions within the slide correspond to such patterns. We validated our method on a deep learning model trained for ovarian tumor histologic subtype prediction. The results show that our framework generates explanations that accurately reflect the histomorphology underlying the model’s predictions at both global and local levels. For interpretability and clinical utility in diagnostic contexts, human evaluation results showed that our explanations were easy to interpret, rich in diagnostic features, and directly helpful for diagnostic decision-making, thereby enhancing pathologist-AI collaboration. Our work highlights that unifying global and local explanations and grounding them in expert-interpreted morphology enhances the interpretability and verifiability of deep learning models, thereby facilitating the transparent deployment of such models in clinical practice.

Abstract Background The incidence of early onset breast cancers (BCs) has increased, paralleling rising trends in delayed childbearing. We hypothesize that a distinct postpartum BC (PPBC) subtype, identifiable by time since last birth (TSLB) and biomarker expression, contributes to this trend. Methods We applied GeoMx Digital Spatial Profiling (DSP) to measure associations between TSLB and 71 proteins in 640 BCs from women aged ≤40 years included in the Young Women’s Breast Cancer Study. We analyzed data using univariable linear regression and multivariable sliced inverse regression to account for higher order interactions among biomarkers. Results In keratin-rich segments, progesterone receptor (PR) (P = 1.00 × 10−4) and PTEN (P = 1.00 × 10−3) were associated with longer TSLB; multivariable analyses revealed positive associations for GZMB (P = 4.00 × 10−4), SMA (P = 1.00 × 10−4), and NF-1 (P = 1.00 × 10−3). In keratin-poor segments, univariable significant positive associations were found for PR (P = 2.00 × 10−4) and PTEN (P = 1.00 × 10−3), whereas CD20 (P = 3.00 × 10−4) and CTLA4 (P = 4.00 × 10−5) were negatively associated; multivariable significant associations were found for fibronectin (P = 3.00 × 10−5) and pan-Akt (P = 1.00 × 10−3). Associations persisted after adjustment for multiple comparisons and BC molecular subtypes. Associations including for PR, PTEN, and CD20 were strongest among women with the shortest TSLB. OPAL multiplex immunofluorescence assays for PR, PTEN, CD20, SMA, and CTLA4 replicated several DSP findings, particularly when stratified by subtype and with compartment matching. In TCGA, RNA species linked to proteins associated with TSLB correlated strongly with a T-cell exhaustion signature previously linked to poor prognosis among premenopausal women. Conclusion These data support PPBC as a biologically coherent phenotype defined by TSLB and biomarker profile, with potential implications for prevention and therapy.

Background: Discovering risk biomarkers in small benign breast disease (BBD) biopsies is constrained by scarce tissue and microanatomic heterogeneity of terminal duct lobular units (TDLUs). We tested whether tissue-sparing tissue microarray (TMA)–based Digital Spatial Profiling (DSP) can deliver reproducible, biologically coherent protein measurements across benign lobules and breast cancers (BCs), and how well DSP aligns with standard immunoassays. Methods: We performed a pilot using tissues from the Mayo Clinic BBD cohort using TMAs representing four contexts: terminal duct lobular units (TDLUs) from BBD biopsies preceding BC and matched BBD-controls, subsequent BCs, and BC-associated TDLUs. We profiled 79 proteins by DSP (37 retained after QC) and benchmarked against chromogenic IHC and OPAL immunofluorescence. Reproducibility was evaluated using intraclass correlation coefficients (ICCs), cross-platform agreement (weighted kappa), marker correlations, and mixed-effects models with false-discovery-rate (FDR) control. Results: We analyzed 368 BBD-TDLU cores (88 cases; 88 controls), 204 BC cores and 110 BC-associated TDLU cores. ICCs were highest in BC tissues, and lower in BC-associated TDLUs and BBD-TDLUs. Agreement was slight–to-fair in TDLUs but moderate (ER/PR) to substantial (BCL2) in BC. DSP recapitulated expected immunologic correlations (CD45 with T-cell, B-cell, and macrophage markers) and tissue-type gradients (BC > BC-associated TDLUs > BBD-TDLUs). Exploratory case–control differences in BBD-TDLUs did not persist after FDR control. Conclusions: TMA-based DSP is feasible in archival breast tissues and yields biologically coherent, cross-platform-benchmarked profiles that are particularly robust in BC, while conserving scarce TDLUS and clarifying current limits of single-marker risk stratification from benign lobules. These data provide a foundation for refined sampling and expanded panels in future TDLU-focused studies.

Abstract BRCA-associated homologous recombination deficiency (HRD) is present in ~ 50% of high-grade serous carcinomas (HGSC) and predicts sensitivity to platinum-based therapy. However, there is little understanding of why some patients with BRCA-deficient tumors experience unexpectedly poor outcomes. We profiled 154 tumors, enriched for patients with BRCA-deficient tumors that experienced short overall survival (≤ 3 years, n = 42), using whole-genome, transcriptome, and methylation analyses. All but one BRCA-deficient tumor exceeded an accepted HRD genomic scarring threshold. However, patients with BRCA1-deficient HGSC with a more elevated HRD score survived significantly longer. Patients with BRCA2-deficient HGSC and loss of NF1 survived twice as long as those without NF1 loss, whereas PIK3CA or RAD21 amplification defined BRCA2-deficient HGSC with exceptionally short survival. BRCA1-deficient tumors in short survivors had evidence of immunosuppressive c-kit signaling and EMT. In a large HGSC cohort (n = 1,389) including 282 individuals with pathogenic germline BRCA variants (gBRCApv), the location of the mutation within functional domains stratified clinical outcomes. Notably, residual disease after primary surgery had limited prognostic effect in gBRCApv-carriers compared to non-carriers. Our findings indicate that tumor HR proficiency in the context of therapy response and survival is not a binary property, and highlight genomic and immune modifiers of outcomes in BRCA-deficient HGSC.

Objective: To evaluate the documentation of menopause-related International Classification of Diseases-10 (ICD-10) codes in the electronic health record (EHR) among midlife women with moderate or greater menopause symptoms receiving primary care. Methods: This cross-sectional study from the Hormones and Experiences of Aging (HERA) cohort included women aged 45-60 years receiving primary care at one of 4 Mayo Clinic sites who completed a one-time survey between March 1, 2021 and June 30, 2021. The survey captured demographic data, menopause symptoms using the Menopause Rating Scale (MRS), health care utilization, and treatment. Women with an MRS score ≥12 were included. The primary outcome was documentation of a menopause-related ICD-10 code in the EHR in the 12 months before survey completion. Results: Of 5,254 women with completed surveys, 2,414 (49%) had an MRS score ≥12 and were included. Among these, 1,519 (63%) reported seeking care for their menopause symptoms in the past 12 months, but only 345 (23%) had a menopause-related ICD-10 code. Women with an ICD-10 code had higher MRS scores (18 [IQR: 14-22] vs 17 [IQR: 14-20]; P = 0.002) and were more likely to use systemic hormone therapy (HT; 26% vs 9%; P < 0.001), and vaginal HT (20% vs 6%; P < 0.001). Conclusions: Menopause-related ICD-10 diagnosis codes were under-documented in women with significant menopause symptom burden. Those with a code were more likely to report menopause treatment. These findings highlight a critical gap between symptom burden and diagnosis coding in the EHR, underscoring the need to improve identification and management of menopause symptoms.

Objective: To compare the prevalence of female sexual dysfunction (FSD) and distress between sexually active midlife women (50-64 y) and older women (65+ y) presenting for care at women’s health clinics at a tertiary care center. Methods: This cross-sectional study included women aged 50 and above who received care at Mayo Clinic women’s health clinics in Rochester, MN; Scottsdale, AZ; and Jacksonville, FL, between May 1, 2015, and August 31, 2022. Sexual function and distress were compared between midlife and older women using the Female Sexual Function Index (FSFI), Female Sexual Distress Scale-Revised (FSDS-R), and self-reported sexual health concerns. Results: Among 3,465 sexually active women, older women were less likely to report loss of sexual desire (33.4% vs. 47.6%; P <0.001) and reduced genital sensation (13.0% vs. 16.9%; P =0.024) compared with midlife women. There were no significant differences for vaginal dryness, painful intercourse, or arousal/orgasm difficulties. FSFI scores were higher in older women for desire (3.0 vs. 2.4; P <0.001) and lubrication (3.9 vs. 3.6; P <0.001). While total FSFI scores were similar between older and midlife women (21.2 vs. 22.2; P =0.11), sexually related distress was lower in older women compared with midlife women (13.0 vs. 15.0; P =0.015). The prevalence of FSD (defined as FSFI score ≤26.55 and FSDS-R score ≥11) was similar between groups (51.8% vs. 56.2%; P =0.056). Conclusions: Older women experienced FSD at similar rates as midlife women but reported less sexual distress, potentially reflecting lower expectations regarding sexual function. Addressing sexual health concerns in older women may enhance quality of life.

Objective: To explore potential differences in physical activity and sedentary behavior volumes and patterns among postmenopausal women with and without premenopausal bilateral oophorectomy (PBO). Methods: Women with a history of PBO (n = 50) and age-matched postmenopausal referent women (n = 50) were recruited. Participants wore accelerometers on both ankles for 7 days. Volume metrics of sedentary behavior and physical activity, such as step counts, active time, and sedentary time, as well as the sedentary behavior and habitual physical activity distribution, and accumulation patterns, were quantified from the accelerometer data and compared between groups. Results: Metrics indicating volume of sedentary behavior and physical activity were not statistically different between groups. PBO was significantly associated with higher variability in stepping bout time ( P = 0.022), indicating a potentially more complex walking pattern. In addition, PBO was significantly associated with lower variability in sedentary break time ( P = 0.012), and lower activity time Gini index ( Z = −2.428, P = 0.015). This suggests that women with PBO may have broken up their sedentary time with shorter and less variable activity bouts, and because they had relatively shorter average daily active time, they might be at a higher risk of subsequent adverse health outcomes such as low bone mineral density. Conclusions: Although there were no differences in overall activity volume, some differences in activity patterns emerged between women with PBO and referent women. The study highlights the need for longitudinal research to understand how physical activity and sedentary behavior patterns evolve in postmenopausal women with a history of PBO.

Three-dimensional radiomic analysis of digital breast tomosynthesis outperformed two-dimensional measures in breast cancer risk assessment.

AbstractBackground and AimsSex‐related hormones and proteins may underlie sex differences in alcohol use disorder characteristics and consequences. Previous reports suggest steroid sex hormones may influence alcohol consumption behaviors while proteins that regulate their circulation levels are rarely investigated. Following up on our earlier study of individual sex‐related hormones' associations with alcohol dependence (AD), this study measured the associations between the combinations of sex‐related hormones and proteins and AD in a larger sample.DesignSex‐stratified case‐control comparison and clustering of plasma sex‐related hormone and protein levels, plus a genome‐wide association study on the AD‐associated hormone and protein combinations.SettingAddiction treatment programs in the United States.ParticipantsFour hundred treatment‐seeking recently abstained AD patients (median = 24.5 days) and 388 age‐and‐sex‐matched controls from a community biobank (male:female ≈ 2:1).MeasurementsPlasma levels of total testosterone, estrone (E1), estradiol (E2), follicle‐stimulating hormone (FSH), luteinizing hormone (LH), sex hormone binding globulin (SHBG) and albumin.FindingsIn males, we found higher E2 [β = 0.13, Punadjusted < 0.001, false discovery rate (FDR) = 0.005], progesterone (β = 0.01, Punadjusted = 0.025, FDR = 0.054), LH (β = 0.12, Punadjusted = 0.017, FDR = 0.044) and albumin (β = 0.09, Punadjusted < 0.001, FDR = 0.005) in AD patients than controls; in females, we observed lower E1 (β = −0.24, Punadjusted = 0.024, FDR = 0.113) and progesterone (β = −1.86, Punadjusted = 0.026, FDR = 0.113) and higher albumin (β = 0.18, Punadjusted < 0.001, FDR = 0.001) in AD patients than controls. Three principal components (PCs), reflecting hormone signatures, were statistically significantly associated with AD in each sex: PC1 (TT, E2 and SHBG; β = −0.15, P = 0.016), PC3 (E1, progesterone and albumin; β = 0.28, P = 0.007) and PC4 (SHBG and albumin; β = 0.34, P = 0.001) in males; PC3 (progesterone and albumin; β = −0.38, P = 0.009), PC4 (TT and albumin; β = −0.49, P = 0.001) and PC5 (TT and progesterone; β = 0.54, P = 0.004) in females. Statistically significant SNP × group associations were found in females between PC4 and polymorphisms in RP51004I9.1 (top SNP: rs6082693; P = 5.41E‐12) and in males between PC3 and a genomic locus that contained SLC35E4, DUSP18 and OSBP2 (top SNP: rs13053277; P = 1.77E‐07).ConclusionsThree sex‐related hormone and protein signatures appear to be associated with alcohol dependence in each sex. Peripheral levels of sex‐related hormones and proteins, as well as their combinations, could be altered in people with alcohol dependence who have abstained for a few weeks compared with controls, and such differences may be sex‐specific.