NR0B2 re-educates myeloid immune cells to reduce regulatory T cell expansion and progression of breast and other solid tumors.

Although survival from breast cancer has dramatically increased, many will develop recurrent, metastatic disease. Unfortunately, survival for this stage of disease remains very low. Activating the immune system has incredible promise since it has the potential to be curative. However, immune checkpoint blockade (ICB) which works through T cells has been largely disappointing for metastatic breast cancer. One reason for this is a suppressive myeloid immune compartment that is unaffected by ICB. Cholesterol metabolism and proteins involved in cholesterol homeostasis play important regulatory roles in myeloid cells. Here, we demonstrate that NR0B2, a nuclear receptor involved in negative feedback of cholesterol metabolism, works in several myeloid cell types to impair subsequent expansion of regulatory T cells (Tregs); Tregs being a subset known to be highly immune suppressive and associated with poor therapeutic response. Within myeloid cells, NR0B2 serves to decrease many aspects of the inflammasome, ultimately resulting in decreased IL1ß; IL1ß driving Treg expansion. Importantly, mice lacking NR0B2 exhibit accelerated tumor growth. Thus, NR0B2 represents an important node in myeloid cells dictating ensuing Treg expansion and tumor growth, thereby representing a novel therapeutic target to re-educate these cells, having impact across different solid tumor types. Indeed, a paper co-published in this issue demonstrates the therapeutic utility of targeting NR0B2.

Development of NR0B2 as a therapeutic target for the re-education of tumor associated myeloid cells.

Immune checkpoint blockade (ICB) has had limited utility in several solid tumors such as breast cancer, a major cause of cancer-related mortality in women. Therefore, there is considerable interest in alternate strategies to promote an anti-cancer immune response. A paper co-published in this issue describes how NR0B2, a protein involved in cholesterol homeostasis, functions within myeloid immune cells to modulate the inflammasome and reduce the expansion of immune-suppressive regulatory T cells (Treg). Here, we develop NR0B2 as a potential therapeutic target. NR0B2 in tumors is associated with improved survival for several cancer types including breast. Importantly, NR0B2 expression is also prognostic of ICB success. Within breast tumors, NR0B2 expression is inversely associated with FOXP3, a marker of Tregs. While a described agonist (DSHN) had some efficacy, it required high doses and long treatment times. Therefore, we designed and screened several derivatives. A methyl ester derivative (DSHN-OMe) emerged as superior in terms of (1) cellular uptake, (2) ability to regulate expected expression of genes, (3) suppression of Treg expansion using in vitro co-culture systems, and (4) efficacy against the growth of primary and metastatic tumors. This work identifies NR0B2 as a target to re-educate myeloid immune cells and a novel ligand with significant anti-tumor efficacy in preclinical models.

Specific Polo-Like Kinase 1 Expression in Nodular Lymphocyte-Predominant Hodgkin Lymphoma Suggests an Intact Immune Surveillance Program.

Nodular lymphocyte-predominant Hodgkin lymphoma (NLPHL) is a rare and relatively indolent B-cell lymphoma. Characteristically, the [lymphocyte-predominant (LP)] tumor cells are embedded in a microenvironment enriched in lymphocytes. More aggressive variants of mature B-cell and peripheral T-cell lymphomas exhibit nuclear expression of the polo-like kinase 1 (PLK1) protein, stabilizing MYC (alias c-myc) and associated with worse clinical outcomes. This study demonstrated expression of PLK1 in the LP cells in 100% of NLPHL cases (n = 76). In contrast, <5% of classic Hodgkin lymphoma cases (n = 70) showed PLK1 expression within the tumor cells. Loss-of-function approaches demonstrated that the expression of PLK1 promoted cell proliferation and increased MYC stability in NLPHL cell lines. Correlation with clinical parameters revealed that the increased expression of PLK1 was associated with advanced-stage disease in patients with NLPHL. A multiplex immunofluorescence panel coupled with artificial intelligence algorithms was used to correlate the composition of the tumor microenvironment with the proliferative stage of LP cells. The results showed that LP cells with PLK1 (high) expression were associated with increased numbers of cytotoxic and T-regulatory T cells. Overall, the findings demonstrate that PLK1 signaling increases NLPHL proliferation and constitutes a potential vulnerability that can be targeted with PLK1 inhibitors. An active immune surveillance program in NLPHL may be a critical mechanism limiting PLK1-dependent tumor growth.

Impact of genomic testing on urologists' treatment preference in favorable risk prostate cancer: A randomized trial.

INTRODUCTION: The Oncotype Dx Genomic Prostate Score (GPS) is a 17-gene relative expression assay that predicts adverse pathology at prostatectomy. We conducted a novel randomized controlled trial to assess the impact of GPS on urologist's treatment preference for favorable risk prostate cancer (PCa): active surveillance versus active treatment (i.e., prostatectomy/radiation). This is a secondary endpoint from the ENACT trial which recruited from three Chicago hospitals from 2016 to 2019. METHODS: Ten urologists along with men with very low to favorable-intermediate risk PCa were included in the study. Participants were randomly assigned to standardized counseling with or without GPS assay. The main outcome was urologists' preference for active treatment at Visit 2 by study arm (GPS versus Control). Multivariable best-fit binary logistic regressions were constructed to identify factors independently associated with urologists' treatment preference. RESULTS: Two hundred men (70% Black) were randomly assigned to either the Control (96) or GPS arm (104). At Visit 2, urologists' preference for prostatectomy/radiation almost doubled in the GPS arm to 29.3% (29) compared to 14.1% (13) in the Control arm (p = 0.01). Randomization to the GPS arm, intermediate NCCN risk level, and lower patient health literacy were predictors for urologists' preference for active treatment. DISCUSSION: Limitations included sample size and number of urologists. In this study, we found that GPS testing reduced urologists' likelihood to prefer active surveillance. CONCLUSIONS: These findings demonstrate how obtaining prognostic biomarkers that predict negative outcomes before treatment decision-making might influence urologists' preference for recommending aggressive therapy in men eligible for active surveillance.

Machine Learning for Digital Scoring of PRMT6 in Immunohistochemical Labeled Lung Cancer.

Lung cancer is the leading cause of cancer death in the U.S. Therefore, it is imperative to identify novel biomarkers for the early detection and progression of lung cancer. PRMT6 is associated with poor lung cancer prognosis. However, analyzing PRMT6 expression manually in large samples is time-consuming posing a significant limitation for processing this biomarker. To overcome this issue, we trained and validated an automated method for scoring PRMT6 in lung cancer tissues, which can then be used as the standard method in future larger cohorts to explore population-level associations between PRMT6 expression and sociodemographic/clinicopathologic characteristics. We evaluated the ability of a trained artificial intelligence (AI) algorithm to reproduce the PRMT6 immunoreactive scores obtained by pathologists. Our findings showed that tissue segmentation to cancer vs. non-cancer tissues was the most critical parameter, which required training and adjustment of the algorithm to prevent scoring non-cancer tissues or ignoring relevant cancer cells. The trained algorithm showed a high concordance with pathologists with a correlation coefficient of 0.88. The inter-rater agreement was significant, with an intraclass correlation of 0.95 and a scale reliability coefficient of 0.96. In conclusion, we successfully optimized a machine learning algorithm for scoring PRMT6 expression in lung cancer that matches the degree of accuracy of scoring by pathologists.

Re-education of myeloid immune cells to reduce regulatory T cell expansion and impede breast cancer progression.

Immune checkpoint blockade (ICB) has revolutionized cancer therapy but has had limited utility in several solid tumors such as breast cancer, a major cause of cancer-related mortality in women. Therefore, there is considerable interest in alternate strategies to promote an anti-cancer immune response. We demonstrate that NR0B2, a protein involved in cholesterol homeostasis, functions within myeloid immune cells to modulate the NLRP3 inflammasome and reduce the expansion of immune-suppressive regulatory T cells (Treg). Loss of NR0B2 increased mammary tumor growth and metastasis. Small molecule agonists, including one developed here, reduced Treg expansion, reduced metastatic growth and improved the efficacy of ICB. This work identifies NR0B2 as a target to re-educate myeloid immune cells providing proof-of-principle that this cholesterol-homeostasis axis may have utility in enhancing ICB.

Quantification of subtype purity in Luminal A breast cancer predicts clinical characteristics and survival.

PURPOSE: PAM50 profiling assigns each breast cancer to a single intrinsic subtype based on a bulk tissue sample. However, individual cancers may show evidence of admixture with an alternate subtype that could affect prognosis and treatment response. We developed a method to model subtype admixture using whole transcriptome data and associated it with tumor, molecular, and survival characteristics for Luminal A (LumA) samples. METHODS: We combined TCGA and METABRIC cohorts and obtained transcriptome, molecular, and clinical data, which yielded 11,379 gene transcripts in common and 1,178 cases assigned to LumA. We used semi-supervised non-negative matrix factorization (ssNMF) to compute the subtype admixture proportions of the four major subtypes-pLumA, pLumB, pHER2, and pBasal-for each case and measured associations with tumor characteristics, molecular features, and survival. RESULTS: Luminal A cases in the lowest versus highest quartile for pLumA transcriptomic proportion had a 27% higher prevalence of stage > 1, nearly a threefold higher prevalence of TP53 mutation, and a hazard ratio of 2.08 for overall mortality. We found positive associations between pHER2 and HER2 positivity by IHC or FISH; between pLumB and PR negativity; and between pBasal and younger age, node positivity, TP53 mutation, and EGFR expression. Predominant basal admixture, in contrast to predominant LumB or HER2 admixture, was not associated with shorter survival. CONCLUSION: Bulk sampling for genomic analyses provides an opportunity to expose intratumor heterogeneity, as reflected by subtype admixture. Our results elucidate the striking extent of diversity among LumA cancers and suggest that determining the extent and type of admixture holds promise for refining individualized therapy. LumA cancers with a high degree of basal admixture appear to have distinct biological characteristics that warrant further study.

Prostate-derived circulating microRNAs add prognostic value to prostate cancer risk calculators.

Prostate cancer is the second leading cause of malignancy-related deaths among American men. Active surveillance is a safe option for many men with less aggressive disease, yet definitively determining low-risk cancer is challenging with biopsy alone. Herein, we sought to identify prostate-derived microRNAs in patient sera and serum extracellular vesicles, and determine if those microRNAs improve upon the current clinical risk calculators for prostate cancer prognosis before and after biopsy. Prostate-derived intracellular and extracellular vesicle-contained microRNAs were identified by small RNA sequencing of prostate cancer patient explants and primary cells. Abundant microRNAs were included in a custom microRNA PCR panel that was queried in whole serum and serum extracellular vesicles from a diverse cohort of men diagnosed with prostate cancer. The levels of these circulating microRNAs significantly differed between indolent and aggressive disease and improved the area under the curve for pretreatment nomograms of prostate cancer disease risk. The microRNAs within the extracellular vesicles had improved prognostic value compared to the microRNAs in the whole serum. In summary, quantifying microRNAs circulating in extracellular vesicles is a clinically feasible assay that may provide additional information for assessing prostate cancer risk stratification.

Regulation of Prostate Androgens by Megalin and 25-hydroxyvitamin D Status: Mechanism for High Prostate Androgens in African American Men.

Vitamin D deficiency is associated with an increased risk of prostate cancer mortality and is hypothesized to contribute to prostate cancer aggressiveness and disparities in African American populations. The prostate epithelium was recently shown to express megalin, an endocytic receptor that internalizes circulating globulin-bound hormones, which suggests regulation of intracellular prostate hormone levels. This contrasts with passive diffusion of hormones that is posited by the free hormone hypothesis. Here, we demonstrate that megalin imports testosterone bound to sex hormone-binding globulin into prostate cells. Prostatic loss of Lrp2 (megalin) in a mouse model resulted in reduced prostate testosterone and dihydrotestosterone levels. Megalin expression was regulated and suppressed by 25-hydroxyvitamin D (25D) in cell lines, patient-derived prostate epithelial cells, and prostate tissue explants. In patients, the relationships between hormones support this regulatory mechanism, as prostatic DHT levels are higher in African American men and are inversely correlated with serum 25D status. Megalin levels are reduced in localized prostate cancer by Gleason grade. Our findings suggest that the free hormone hypothesis should be revisited for testosterone and highlight the impact of vitamin D deficiency on prostate androgen levels, which is a known driver of prostate cancer. Thus, we revealed a mechanistic link between vitamin D and prostate cancer disparities observed in African Americans. Significance: These findings link vitamin D deficiency and the megalin protein to increased levels of prostate androgens, which may underpin the disparity in lethal prostate cancer in African America men.

Prostate-derived circulating microRNAs add prognostic value to prostate cancer risk calculators.

Prostate cancer is the second leading cause of malignancy-related deaths among American men. Active surveillance is a safe option for many men with less aggressive disease, yet definitively determining low-risk cancer is challenging with biopsy alone. Herein, we sought to identify prostate-derived microRNAs in patient sera and serum extracellular vesicles, and determine if those microRNAs improve upon the current clinical risk calculators for prostate cancer prognosis before and after biopsy. Prostate-derived intracellular and extracellular vesicle-contained microRNAs were identified by small RNA sequencing of prostate cancer patient explants and primary cells. Abundant microRNAs were included in a custom microRNA PCR panel that was queried in whole serum and serum extracellular vesicles from a diverse cohort of men diagnosed with prostate cancer. The levels of these circulating microRNAs significantly differed between indolent and aggressive disease and improved the area under the curve for pretreatment nomograms of prostate cancer disease risk. The microRNAs within the extracellular vesicles were the most informative and improved the AUC to 0.739 compared to the existing nomogram alone, which has an AUC of 0.561. The microRNAs in the whole serum improved it to AUC 0.675. In summary, quantifying microRNAs circulating in extracellular vesicles is a clinically feasible assay that may provide additional information for assessing prostate cancer risk stratification.

Cautious Artificial Intelligence Improves Outcomes and Trust by Flagging Outlier Cases.

PURPOSE: Artificial intelligence (AI) models for medical image diagnosis are often trained and validated on curated data. However, in a clinical setting, images that are outliers with respect to the training data, such as those representing rare disease conditions or acquired using a slightly different setup, can lead to wrong decisions. It is not practical to expect clinicians to be trained to discount results for such outlier images. Toward clinical deployment, we have designed a method to train cautious AI that can automatically flag outlier cases. MATERIALS AND METHODS: Our method-ClassClust-forms tight clusters of training images using supervised contrastive learning, which helps it identify outliers during testing. We compared ClassClust's ability to detect outliers with three competing methods on four publicly available data sets covering pathology, dermatoscopy, and radiology. We held out certain diseases, artifacts, and types of images from training data and examined the ability of various models to detect these as outliers during testing. We compared the decision accuracy of the models on held-out nonoutlier images also. We visualized the regions of the images that the models used for their decisions. RESULTS: Area under receiver operating characteristic curve for outlier detection was consistently higher using ClassClust compared with the previous methods. Average accuracy on held-out nonoutlier images was also higher, and the visualizations of image regions were more informative using ClassClust. CONCLUSION: The ability to flag outlier test cases need not be at odds with the ability to accurately classify nonoutliers in AI models. Although the latter capability has received research and regulatory attention, AI models for clinical deployment should possess the former as well.

Association of genetic polymorphisms with local steroid metabolism in human benign breasts.

PURPOSE: Although alterations of concentrations in circulating steroids have been linked to single nucleotide polymorphisms (SNPs) of steroidogenic enzymes, we hypothesized that SNPs of such enzymes located within the breast affect local steroid concentrations more than products of such SNPs absorbed from the circulation. METHODS: Steroids (estradiol, estrone, testosterone, androstenedione, DHEA, DHEA sulfate, progesterone) in nipple aspirate fluid (NAF) were purified by HPLC and they along with serum steroids were quantified by immunoassays. Polymorphisms of the transporter SLCO2B1 and enzymes HSD3B1, CYP19A1, HSD17B12, AKR1C3, CYP1B1, and SRD5A1 were measured in white blood cell DNA. RESULTS: Steroid concentrations in NAF of subjects with homozygous minor genotypes differed from those with heterozygotes, i.e., SLCO2B1 (rs2851069) decreased DHEAS (p = 0.04), HSD17B12 (rs11555762) increased estradiol (p < 0.004), and CYP1B1 (rs1056836) decreased estradiol (p = 0.017) and increased progesterone (p = 0.05). Also, in serum, CYP19A1 (rs10046 and rs700518) both decreased testosterone (p = 0.02) and SRD5A1 increased androstenedione (p = 0.006). Steroids in subjects with major homozygotes did not differ from those with heterozygotes indicating recessive characteristics. CONCLUSIONS: In the breast, SNPs were associated with decreased uptake of DHEAS (SLCO2B1), increased estradiol concentrations through increased oxidoreductase activity (HSD17B12), or decreased estradiol concentrations by presumed formation of 4-hydroxyestradiol (CYP1B1). CYP19A1 was associated with decreased testosterone concentrations in serum but had no significant effect on estrogen or androgen concentrations within the breast. The hormone differences observed in NAF were not usually evident in serum, indicating the importance of assessing the effect of these SNPs within the breast.

Weakly supervised learning on unannotated H&E-stained slides predicts BRAF mutation in thyroid cancer with high accuracy.

Deep neural networks (DNNs) that predict mutational status from H&E slides of cancers can enable inexpensive and timely precision oncology. Although expert knowledge is reliable for annotating regions informative of malignancy and other known histologic patterns (strong supervision), it is unreliable for identifying regions informative of mutational status. This poses a serious impediment to obtaining higher prognostic accuracy and discovering new knowledge of pathobiology. We used a weakly supervised learning technique to train a DNN to predict BRAF V600E mutational status, determined using DNA testing, in H&E-stained images of thyroid cancer tissue without regional annotations. Our discovery cohort was a tissue microarray of only 85 patients from a single hospital. On a large independent external cohort of 444 patients from other hospitals, the trained model gave an area under the receiver operating characteristic curve of 0.98 (95% CI 0.97-1.00), which is much higher than the previously reported results for detecting any mutation using H&E by DNNs trained using strong supervision. We also developed a visualization technique that can automatically highlight regions the DNN found most informative for predicting mutational status. Our visualization is spatially granular and highly specific in highlighting strong negative and positive regions and moves us toward explainable artificial intelligence. Using t-tests, we confirmed that the proportions of follicular or papillary histology and oncocytic cytology, as noted for each patient by a pathologist who was blinded to the mutational status, were significantly different between mutated and wildtype patients. However, based solely on these features noted by the pathologist, a logistic regression classifier gave an average area under the receiver operating characteristic curve of 0.78 in five-fold cross-validation, which is much lower than that obtained using the DNN. These results highlight the potential of weakly supervised learning for training DNN models for problems where the informative visual patterns and their locations are not known a priori. © 2021 The Pathological Society of Great Britain and Ireland. Published by John Wiley & Sons, Ltd.

Genetic Variation and Immunohistochemical Localization of the Glucocorticoid Receptor in Breast Cancer Cases from the Breast Cancer Care in Chicago Cohort.

BACKGROUND: Glucocorticoid, one of the primary mediators of stress, acts via its receptor, the glucocorticoid receptor (GCR/NR3C1), to regulate a myriad of physiological processes. We measured the genetic variation and protein expression of GCR, and the genes that regulate GCR function or response and examined whether these alterations were associated with breast cancer clinicopathological characteristics. METHOD: We used samples from a multiracial cohort of breast cancer patients to assess the association between breast cancer characteristics and the genetic variants of single nucleotide polymorphisms (SNPs) in GCR/NR3C1, FKBP5, Sgk1, IL-6, ADIPOQ, LEPR, SOD2, CAT, and BCL2. RESULTS: Several SNPs were associated with breast cancer characteristics, but statistical significance was lost after adjustment for multiple comparisons. GCR was detected in all normal breast tissues and was predominantly located in the nuclei of the myoepithelial cell layer, whereas the luminal layer was negative for GCR. GCR expression was significantly decreased in all breast cancer tissue types, compared to nontumor tissue, but was not associated with breast cancer characteristics. We found that high nuclear GCR expression was associated with basal cell marker cytokeratin 5/6 positivity. CONCLUSION: GCR expression is reduced in breast cancer tissue and correlates with the basal cell marker CK5/6.

Impact of a Genomic Test on Treatment Decision in a Predominantly African American Population With Favorable-Risk Prostate Cancer: A Randomized Trial.

PURPOSE: The Genomic Prostate Score (GPS), performed on biopsy tissue, predicts adverse outcome in prostate cancer (PCa) and has shown promise for improving patient selection for active surveillance (AS). However, its impact on treatment choice in high-risk populations of African Americans is largely unknown and, in general, the effect of the GPS on this difficult decision has not been evaluated in randomized trials. METHODS: Two hundred men with National Comprehensive Cancer Network very low to low-intermediate PCa from three Chicago hospitals (70% Black, 16% college graduates) were randomly assigned at diagnosis to standard counseling with or without a 12-gene GPS assay. The primary end point was treatment choice at a second postdiagnosis visit. The proportion of patients choosing AS was compared, and multivariable modeling was used to estimate the effects of various factors on AS acceptance. RESULTS: AS acceptance was high overall, although marginally lower in the intervention group (77% v 88%; P = .067), and lower still when men with inadequate specimens were excluded (P = .029). Men with lower health literacy who received a GPS were seven-fold less likely to choose AS compared with controls, whereas no difference was seen in men with higher health literacy (Pinteraction = .022). Among men with low-intermediate risk, 69% had GPS values consistent with unfavorable intermediate or high-risk cancer. AS choice was also independently associated with a family history of PCa and having health insurance. CONCLUSION: In contrast to other studies, the net effect of the GPS was to move patients away from AS, primarily among men with low health literacy. These findings have implications for our understanding of how prognostic molecular assays that generate probabilities of poor outcome can affect treatment decisions in diverse clinical populations.

Performance of prostate health index and PSA density in a diverse biopsy-naïve cohort with mpMRI for detecting significant prostate cancer.

Objective: To compare Prostate Health Index (PHI) and prostate-specific antigen (PSA) density as secondary tests after multiparametric magnetic resonance imaging (mpMRI) in improving the detection accuracy of Gleason grade group (GG) 2-5 prostate cancer (PCa) and in decreasing unnecessary biopsies in a multiethnic biopsy-naïve population. Methods: From February 2017 to February 2020, we recruited consecutive biopsy-naïve men in participating urology clinics for elevated PSA levels. They all had a PHI score, mpMRI, and prostate biopsy. Experienced genitourinary radiologists read all mpMRI studies based on PIRADS version 2.0. Logistic regression models were used to generate receiver operating characteristic curves. Models were tested for effect modification between Race (Black vs White) and both PHI and PSA density, and Race and PIRADS to determine if race impacted their prediction accuracy. Sensitivity, specificity, and predictive values of PHI and PSA density thresholds were calculated by PIRADS scores. The primary outcome was GG2-5 PCa, that is, Gleason score ≥3 + 4. Results: The study included 143 men, of which 65 (45.5%) were self-reported Black. Median age was 62.0 years and 55 men (38.4%) had GG2-5 PCa. Overall, 18.1% had PIRADS 1-2, 32.9% had PIRADS 3, and 49.0% had PIRADS 4-5. For the binary logistic regressions, the interactions between PIRADS and Race (P = .08), Log (PHI) and Race (P = .17), and Log (PSA density) and Race (P = .42) were not statistically significant. Within PIRADS 3 lesions, a PHI ≥49 prevented unnecessary biopsies in 55% of men and missed no GG2-5 PCa, yielding a negative predictive value of 100%. There was no reliable PHI or PSA density threshold to avoid PCa biopsies in PIRADS 1-2 or 4-5. Conclusions: PHI and PSA density can be used after mpMRI to improve the detection of GG2-5 PCa in a biopsy-naïve cohort. PHI may be superior to PSA density in PIRADS 3 lesions by avoiding 55% of unnecessary biopsies. Using both PHI and PSA density in series may further increase specificity and lead to fewer unnecessary biopsies, but further larger studies are warranted to determine the optimal threshold of each biomarker.

Ethnic variation in prostate cancer detection: a feasibility study for use of the Stockholm3 test in a multiethnic U.S. cohort.

BACKGROUND: The Stockholm3 test improves Gleason Grade Group ≥2 (GG ≥ 2) prostate cancer (PC) detection, however it has not been evaluated in an American cohort where clinical practice patterns and ethnicity differ. We aimed to identify subgroups within a Stockholm population with PC risk profiles matching American ethnicity-specific subgroups and compare the detection of PC and describe Stockholm3 performance within these subgroups. METHODS: All men age 49-70 years presenting for prostate biopsies were evaluated at UIC from 2016 to 2019, as well as men in Stockholm from 2012 to 2014 in the STHLM3 study. Propensity scores (PS) were estimated for each person using logistic regression for age, PSA, prostate volume, family history of PC, 5-alpha reductase inhibitor use, and prior biopsy. 3:1 PS matching was performed for Stockholm to Chicago ethnicity-specific cohorts and odds ratios (OR) were computed to compare detection of GG ≥ 2 PC between groups. RESULTS: 504 Chicago men and 6980 Stockholm men were included. In African American (AA) men, 51% had GG ≥ 2 PC detected, while in risk-matched Stockholm men, 34% had GG ≥ 2 PC detected (OR: 2.1, p < 0.001). There was no statistical difference in GG ≥ 2 PC detected when matching Stockholm men to non-Hispanic Caucasian men (31% vs. 24%, OR: 0.7, p = 0.30) or Hispanic Caucasian men (31% vs. 27%, OR: 1.2, p = 0.42). The AUC for the Stockholm3 test of the matched Stockholm cohorts for AA, non-Hispanic Caucasian, and Hispanic Caucasian men was 0.85, 0.89, and 0.90, respectively. CONCLUSIONS: Using statistical techniques to simulate a multi-ethnic Chicago cohort within the STHLM3 population, we found an excess risk of GG ≥ 2 PC among AA men. Our hypothesis that the Stockholm3 may have good predictive value in a multiethnic cohort is strengthened, and that recalibration to at least AA men seems likely to be needed to obtain well-calibrated predictions.

Performance of Prostate Health Index in Biopsy Naïve Black Men.

PURPOSE: The Prostate Health Index is validated for prostate cancer detection but has not been well validated for Gleason grade group 2-5 prostate cancer detection in Black men. We hypothesize that the Prostate Health Index has greater accuracy than prostate specific antigen for detection of Gleason grade group 2-5 prostate cancer. We estimated probability of overall and Gleason grade group 2-5 prostate cancer across previously established Prostate Health Index ranges and identified Prostate Health Index cutoffs that maximize specificity for Gleason grade group 2-5 prostate cancer with sensitivity >90%. MATERIALS AND METHODS: We recruited a "cancer-free" Black control cohort (135 patients) and a cohort of biopsy naïve Black men (158) biopsied for elevated prostate specific antigen. Descriptive statistics compared the prostate cancer cases and controls and the frequency of Gleason grade group 2-5 prostate cancer across Prostate Health Index scores. Receiver operating characteristics compared the discrimination of prostate specific antigen, Prostate Health Index and other prostate specific antigen related biomarkers. Sensitivity and specificity for Gleason grade group 2-5 prostate cancer detection were assessed at prostate specific antigen and Prostate Health Index thresholds alone and in series. RESULTS: Of biopsied subjects 32.9% had Gleason grade group 2-5 prostate cancer. In Blacks with prostate specific antigen from 4.0-10.0 ng/ml, Prostate Health Index and prostate specific antigen had similar discrimination for Gleason grade group 2-5 prostate cancer (0.63 vs 0.57, p=0.27). In Blacks with prostate specific antigen ≤10.0, a threshold of prostate specific antigen ≥4.0 had 90.4% sensitivity for Gleason grade group 2-5 prostate cancer; a threshold of prostate specific antigen ≥4.0 with Prostate Health Index ≥35.0 in series avoided unnecessary biopsy in 33.0% of men but missed 17.3% of Gleason grade group 2-5 prostate cancer. Prostate specific antigen ≥4.0 with Prostate Health Index ≥28.0 in series spared biopsy in 17.9%, while maintaining 90.4% sensitivity of Gleason grade group 2-5 prostate cancer. CONCLUSIONS: The Prostate Health Index has moderate accuracy in detecting Gleason grade group 2-5 prostate cancer in Blacks, but Prostate Health Index ≥28.0 can be safely used to avoid some unnecessary biopsies in Blacks.

Deep Learning to Estimate Human Epidermal Growth Factor Receptor 2 Status from Hematoxylin and Eosin-Stained Breast Tissue Images.

CONTEXT: Several therapeutically important mutations in cancers are economically detected using immunohistochemistry (IHC), which highlights the overexpression of specific antigens associated with the mutation. However, IHC panels can be imprecise and relatively expensive in low-income settings. On the other hand, although hematoxylin and eosin (H&E) staining used to visualize the general tissue morphology is a routine and low cost, it does not highlight any specific antigen or mutation. AIMS: Using the human epidermal growth factor receptor 2 (HER2) mutation in breast cancer as an example, we strengthen the case for cost-effective detection and screening of overexpression of HER2 protein in H&E-stained tissue. SETTINGS AND DESIGN: We use computational methods that reliably detect subtle morphological changes associated with the over-expression of mutation-specific proteins directly from H&E images. SUBJECTS AND METHODS: We trained a classification pipeline to determine HER2 overexpression status of H&E stained whole slide images. Our training dataset was derived from a single hospital containing 26 (11 HER2+ and 15 HER2-) cases. We tested the classification pipeline on 26 (8 HER2+ and 18 HER2-) held-out cases from the same hospital and 45 independent cases (23 HER2+ and 22 HER2-) from the TCGA-BRCA cohort. The pipeline was composed of a stain separation module and three deep neural network modules in tandem for robustness and interpretability. STATISTICAL ANALYSIS USED: We evaluate our trained model through area under the curve (AUC)-receiver operating characteristic. RESULTS: Our pipeline achieved an AUC of 0.82 (confidence interval [CI]: 0.65-0.98) on held-out cases and an AUC of 0.76 (CI: 0.61-0.89) on the independent dataset from TCGA. We also demonstrate the region-level correspondence of HER2 overexpression between a patient's IHC and H&E serial sections. CONCLUSIONS: Our work strengthens the case for automatically quantifying the overexpression of mutation-specific proteins in H&E-stained digital pathology, and it highlights the importance of multi-stage machine learning pipelines for added robustness and interpretability.