To avoid over-treatment of prostate cancer patients following screening for elevated prostate-specific antigen (PSA) levels, keeping patients on active surveillance has been suggested as an alternative to radical treatment. This means reoccurring visits for patients with low-grade cancer to monitor progression. The Prostate Cancer Research International Active Surveillance (PRIAS) study was initiated to research the benefits of active surveillance. Usually, the presence and grade of cancer are determined with needle-core biopsies of the prostate. However, the Gleason grading scale has demonstrated intra- and inter-observer variability. This is a limiting factor for novel automated Gleason grading algorithms. To address this, and simultaneously utilizing a cohort collected within the PRIAS program, we have developed a deep learning-based framework for outcome prediction of patients on active surveillance. Our framework does not use explicit Gleason grades and consists of a pre-trained feature extractor, a feature selector, and an attention-based outcome predictor. We evaluate three feature extractors: the foundation model UNI, an ImageNet pre-trained model, and our Gleason grading network. Using UNI as the feature extractor outperformed the other models, with an average area under the receiver operator characteristic curve (AUC) of 0.996 (95% CI: 0.996 - 0.996). To our knowledge, this is the first end-to-end deep learning-based model for patient-level outcome predictions of prostate cancer patients on active surveillance. We believe that our algorithm could assist the pathologists and facilitate the implementation of prostate cancer screening programs, however, more work is needed in terms of validation and generalization. Our code for training and evaluating our models is publicly available*.

Purpose:To avoid over-treatment of prostate cancer patients following screening for elevated prostate-specific antigen levels, keeping patients on active surveillance has been suggested as an alternative to radical treatment. This means recurring visits for patients with low-grade cancer to monitor progression. Our aim was to develop an artificial intelligence-based model that can identify high-risk patients in a cohort of prostate cancer patients on active surveillance. Approach:We have developed a multiple instance learning-based framework for predicting the longitudinal outcomes for prostate cancer patients on active surveillance. Our models were trained only on whole-slide images with patient-level labels without using explicit Gleason grades. We employed the UNI-2 foundation model and the well-established attention-based multiple instance learning approach. We further evaluated our models by fitting Cox proportional hazards models and testing them on an external dataset. Results:With this approach, we achieved an average area under the receiver operator characteristic curve of 0.958 (95% CI, 0.957 to 0.959). Fitting Cox models to the predicted probabilities achieved a C -index of 0.824 and a hazard ratio of 2.32. However, all models showed a large drop in performance when evaluated on an external dataset. Conclusion:We show that avoiding Gleason grades is beneficial for longitudinal outcome prediction of prostate cancer. Our results suggest that benign prostate tissue contains prognostic information. However, before our models could be used clinically, much more work remains to improve the generalization.

OBJECTIVES:To evaluate a cancer detecting artificial intelligence (AI) algorithm on serial biopsies in patients with prostate cancer on active surveillance (AS). PATIENTS AND METHODS:A total of 180 patients in the Prostate Cancer Research International Active Surveillance (PRIAS) cohort were prospectively monitored using pre-defined criteria. Diagnostic and re-biopsy slides from 2011 to 2020 (n = 4744) were scanned and analysed by an in-house AI-based cancer detection algorithm. The algorithm was analysed for sensitivity, specificity, and for accuracy to predict need for active treatment. Prognostic properties of cancer size, prostate-specific antigen (PSA) level and PSA density at diagnosis were evaluated. RESULTS:The sensitivity and specificity of the AI algorithm was 0.96 and 0.73, respectively, for correct detection of cancer areas. Original pathology report diagnosis was used as the reference method. The area of cancer estimated by the pathologists correlated highly with the AI detected cancer size (r = 0.83). By using the AI algorithm, 63% of the slides would not need to be read by a pathologist as they were classed as benign, at the risk of missing 0.55% slides containing cancer. Biopsy cancer content and PSA density at diagnosis were found to be prognostic of whether the patient stayed on AS or was discontinued for active treatment. CONCLUSION:The AI-based biopsy cancer detection algorithm could be used to reduce the pathologists' workload in an AS cohort. The detected cancer amount correlated well with the cancer length measured by the pathologist and the algorithm performed well in finding even small areas of cancer. To our knowledge, this is the first report on an AI-based algorithm in digital pathology used to detect cancer in a cohort of patients on AS.

Supplemental Materials and Methods and figure text for supplemental figures

Background 18 F-Fluorodeoxyglucose positron emission combined with computed tomography (FDG-PET/CT) has been proposed to improve preoperative staging in patients with bladder cancer subjected to radical cystectomy (RC). Objective Our aim was to assess the accuracy of FDG-PET/CT for lymph node staging ascertained at the multidisciplinary tumour board compared to lymph node status in the surgical lymphadenectomy specimen obtained at RC, and to explore potential factors associated with false-positive FDG-PET/CT results. Design, setting and participants Consecutive patients with bladder cancer undergoing RC with extended lymph node dissection between 2011 and 2019 without preoperative chemotherapy in a tertial referral cystectomy unit were included in the study. Outcome measurements and statistical analyses Sensitivity, specificity, positive and negative predictive values and likelihood ratios were calculated. Potential factors investigated for association with false-positive FDG-PET/CT were; bacteriuria within four weeks prior to FDG-PET/CT, Bacillus Calmette–Guerin (BCG) treatment within 12 months prior to FDG-PET/CT and transurethral resection of bladder tumour (TURB) within four weeks prior to FDG-PET/CT. Results Among 157 patients included for analysis, 44 (28%) were clinically node positive according to FDG-PET/CT. The sensitivity and specificity for detection of lymph node metastasis were 50% and 84%, respectively, and the corresponding positive predictive and negative predictive values were 61% and 76%. Positive and negative likelihood ratios were 3.0 and 0.6, respectively. No association was found between bacteriuria, previous BCG treatment or TURB within 28 days and false-positive FDG-PET/CT results. Conclusions Preoperative FDG-PET/CT prior to RC had a clinically meaningful high specificity (84%) but lower sensitivity (50%) for detection of lymph node metastases compared to lymph node status in an extended pelvic lymphadenectomy template. We could not identify any factors associated with false-positive FDG-PET/CT outcomes.

The current diagnostic pathway for patients with muscle-invasive bladder cancer (MIBC), which involves with computed tomography urography, cystoscopy, and transurethral resection of the bladder (TURB) to histologically confirm MIBC, delays definitive treatment. The Vesical Imaging-Reporting and Data System (VI-RADS) has been suggested for MIBC identification using magnetic resonance imaging (MRI), but a recent randomized trial reported misclassification in one-third of patients. We investigated a new endoscopic biopsy device (Urodrill) for histological confirmation of MIBC and assessment of molecular subtype by gene expression in patients with VI-RADS 4 and 5 lesions on MRI. In ten patients, Urodrill biopsies were guided by MR images to the muscle-invasive portion of the tumor via a flex-ible cystoscope under general anesthesia. During the same session, conventional TURB was subsequently performed. A Urodrill sample was successfully obtained in nine of ten patients. MIBC was verified in six of nine patients, and seven of nine samples contained detrusor muscle. In seven of eight patients for whom a Urodrill biopsy sample was subjected to RNA sequencing, single-sample molecular classifi-cation according to the Lund taxonomy was feasible. No complications related to the biopsy device occurred. A randomized trial comparing this new diagnostic pathway for patients with VI-RADS 4 and 5 lesions and the current standard (TURB) is warranted.Patient summary: We report on a novel biopsy device for patients with muscle -invasive bladder cancer that facilitates histology analysis and molecular character-ization of tumor samples.& COPY; 2023 The Author(s). Published by Elsevier B.V. on behalf of European Association of Urology. This is an open access article under the CC BY-NC-ND license (http://creative-commons.org/licenses/by-nc-nd/4.0/).

Abstract Background 18 F-Flourodeoxyglucose positron emission combined with computed tomography (FDG-PET/CT) has been proposed to improve preoperative staging in patients with bladder cancer subjected to radical cystectomy (RC). Objective Our aim was to assess the accuracy of FDG-PET/CT for lymph node staging compared to lymph node status in the surgical lymphadenectomy specimen obtained at RC, and to explore potential factors associated with false positive FDG-PET/CT results. Design, setting and participants Consecutive patients with bladder cancer undergoing RC with extended lymph node dissection between 2011 and 2019 without preoperative chemotherapy in a tertial referral cystectomy unit were included in the study. Outcome measurements and statistical analyses Sensitivity, specificity, positive and negative predictive values, and likelihood ratios were calculated. Potential factors investigated for association with false positive FDG-PET/CT were; bacteriuria within four weeks prior to FDG-PET/CT, Bacillus Calmette-Guerin (BCG) treatment within 12 months prior to FDG-PET/CT and transurethral resection of bladder tumor (TURB) within four weeks prior to FDG-PET/CT. Results Among 159 patients included for analysis, 46 (29%) were clinically node positive according to FDG-PET/CT. The sensitivity and specificity for detection of lymph node metastasis were 50% and 82%, respectively, and the corresponding positive predictive and negative predictive values were 59% and 76%. Positive and negative likelihood ratios were 2.8 and 0.6, respectively. No association were found between bacteriuria, previous BCG treatment or TURB within 28 days and false positive FDG-PET/CT results. Conclusions Preoperative FDG-PET/CT prior to RC had a clinically meaningful high specificity (82%) but lower sensitivity (50%) for detection of lymph node metastases compared to lymph node status in an extended pelvic lymphadenectomy template. We could not identify any factors associated with false positive FDG-PET/CT outcomes.

Abstract Background There is an urgent need to accurately predict the risk of distant metastasis and metastatic castration-resistant prostate cancer (mCRPC) for treatment decision-making and reducing mortality. An artificial intelliegnce machine learning screening method in combination with liquid biopsy urine test was used to develop a novel gene expression-based algorithm for predicting prostate cancer distant metastasis and mCRPC in newly diagnosed patients. Methods Random forest machine learning algorithm screening was conducted to develop and validate a gene expression-based algorithm to predict the risk of metastasis and mCRPC using liquid biopsy urine samples from the patients with distant metastasis and mCRPC collected from multi-center retrospective (n = 505) and prospective (n = 243) studies with a median follow-up period of 8 and 6 years respectively. The prognostic performance of the algorithm test was assessed using univariate/multivariate Cox regression analyses, Kaplan-Meier disease-free survival plot, and univariate/multivariate discriminant analyses. Results A novel 18-Gene Algorithm urine test was developed and validated. The algorithm showed high accuracy to predict distant metastasis with an area under the curve (AUC) of 0.96 (95% CI 0.87–1.05) and 0.98 (95% CI 0.96–1.02) in the retrospective and prospective cohort respectively. In the prospective cohort, a hazard ratio (HR) to predict metastasis-free survival was 93.8 (95% CI 29.3-300.6) (p < 0.0001). In a prospective mCRPC cohort (n = 205), the algorithm predicted mCRPC-free survival with a HR of 154.4 (95% CI 36.8-647.5) (p < 0.0001) and predicted mCRPC with AUC of 0.98 (95% CI 0.95–1.01). In contrast, currently using clinicopathological parameters, such as Gleason grade and pre-operative PSA, had much lower prognostic power. Conclusions The novel 18-Gene Algorithm is the first highly accurate and non-invasive liquid biopsy urine test to predict distant metastasis and mCRPC in newly diagnosed prostate cancer patients with the potential to improve prostate cancer treatment decision-making and reduce mortality.

PC3M cells overexpressing cyclin A1 initiated macro-metastases in distant organs in xenograft mice.

In recent years, the combination of the digitization of the field of pathology and increased computational power has led to a big increase in research of computer-aided diagnostics using systems based on artificial intelligence (AI). This includes detection and classification of prostate cancer, where several studies have shown great promise in automated prostate cancer grading using deep learning based AI systems. However, there is still work to be done to ensure that these algorithms are invariant to possible variations of the digitized microscopy images they are applied to. A standard method in deep learning to increase the variation of the training data is dataset augmentation. All of these studies apply some augmentation of their data, however, there is a lack of evaluation of different methods and their impact on this crucial part of the AI systems. In this study, we look into different color augmentation methods for the task of segmentation of prostate biopsies. Furthermore, we introduce a novel color augmentation method based on stereographic projection. Our results affirm the importance of studying different augmentation methods and indicate a gain in performance using our method.

Evaluate bone marrow metastasis initiated by PC3M cells via intracardiac injection of tumor cells into mice.

Evaluate bone marrow metastasis initiated by PC3M cells via intracardiac injection of tumor cells into mice.

Purpose: There is an urgent need for developing new biomarker tools to accurately predict treatment response of breast cancer, especially the deadly triple-negative breast cancer. We aimed to develop gene-mutation-based machine learning (ML) algorithms as biomarker classifiers to predict treatment response of first-line chemotherapy with high precision. Methods: Random Forest ML was applied to screen the algorithms of various combinations of gene mutation profiles of primary tumors at diagnosis using a TCGA Cohort (n = 399) with up to 150 months follow-up as a training set and validated in a MSK Cohort (n = 807) with up to 220 months follow-up. Subtypes of breast cancer including triple-negative and luminal A (ER+, PR+ and HER2−) were also assessed. The predictive performance of the candidate algorithms as classifiers was further assessed using logistic regression, Kaplan–Meier progression-free survival (PFS) plot, and univariate/multivariate Cox proportional hazard regression analyses. Results: A novel algorithm termed the 12-Gene Algorithm based on mutation profiles of KRAS, PIK3CA, MAP3K1, MAP2K4, PTEN, TP53, CDH1, GATA3, KMT2C, ARID1A, RunX1, and ESR1, was identified. The performance of this algorithm to distinguish non-progressed (responder) vs. progressed (non-responder) to treatment in the TCGA Cohort as determined using AUC was 0.96 (95% CI 0.94–0.98). It predicted progression-free survival (PFS) with hazard ratio (HR) of 21.6 (95% CI 11.3–41.5) (p < 0.0001) in all patients. The algorithm predicted PFS in the triple-negative subgroup with HR of 19.3 (95% CI 3.7–101.3) (n = 42, p = 0.000). The 12-Gene Algorithm was validated in the MSK Cohort with a similar AUC of 0.97 (95% CI 0.96–0.98) to distinguish responder vs. non-responder patients, and had a HR of 18.6 (95% CI 4.4–79.2) to predict PFS in the triple-negative subgroup (n = 75, p < 0.0001). Conclusions: The novel 12-Gene algorithm based on multitude gene-mutation profiles identified through ML has a potential to predict breast cancer treatment response to therapies, especially in triple-negative subgroups patients, which may assist personalized therapies and reduce mortality.

Abstract Background [18F]PSMA-1007 is a promising tracer for integrated positron emission tomography and computed tomography (PET/CT). Objective Our aim was to assess the diagnostic accuracy of [18F]PSMA-1007 PET/CT for primary staging of lymph node metastasis before robotic-assisted laparoscopy (RALP) with extended lymph node dissection (ePLND). Design, Setting and Participants The study was a retrospective cohort in a tertiary referral center. Men with prostate cancer that underwent surgical treatment for intermediate- or high-risk prostate cancer between May 2019 and August 2021 were included. Interventions [18F]PSMA-1007 PET/CT for initial staging followed by RALP and ePLND. Outcome measurements and statistical analyses Sensitivity and specificity were calculated both for the entire cohort and for patients with lymph node metastasis ≥ 3 mm. Positive (PPV) and negative (NPV) predictive values were calculated. Results and limitations Among 104 patients included in the analyses, 26 patients had lymph node metastasis based on pathology reporting and metastases were ≥ 3 mm in size in 13 of the cases (50%). In the entire cohort, the sensitivity and specificity of [18F]PSMA-1007 were 26.9% (95% confidence interval (CI); 11.6–47.8) and 96.2% (95% CI; 89.2–99.2), respectively. The sensitivity and specificity of [18F]PSMA-1007 to detect a lymph node metastasis ≥ 3 mm on PET/CT were 53.8% (95% CI; 25.1–80.8) and 96.7% (95% CI; 90.7–99.3), respectively. PPV was 70% and NPV 93.6%. Conclusions In primary staging of intermediate- and high-risk prostate cancer, [18F]PSMA-1007 PET/CT is highly specific for prediction of lymph node metastases, but the sensitivity for detection of metastases smaller than 3 mm is limited. Based on our results, [18F]PSMA-1007 PET/CT cannot completely replace ePLND. Patient summary This study investigated the use of an imaging method based on a prostate antigen-specific radiopharmaceutical tracer to detect lymph node prostate cancer metastasis. We found that it is unreliable to discover small metastasis.

Purpose: Despite the high mortality of metastatic colorectal cancer (mCRC), no new biomarker tools are available for predicting treatment response. We developed gene-mutation-based algorithms as a biomarker classifier to predict treatment response with better precision than the current predictive factors. Methods: Random forest machine learning (ML) was applied to identify the candidate algorithms using the MSK Cohort (n = 471) as a training set and validated in the TCGA Cohort (n = 221). Logistic regression, progression-free survival (PFS), and univariate/multivariate Cox proportional hazard analyses were performed and the performance of the candidate algorithms was compared with the established risk parameters. Results: A novel 7-Gene Algorithm based on mutation profiles of seven KRAS-associated genes was identified. The algorithm was able to distinguish non-progressed (responder) vs. progressed (non-responder) patients with AUC of 0.97 and had predictive power for PFS with a hazard ratio (HR) of 16.9 (p < 0.001) in the MSK cohort. The predictive power of this algorithm for PFS was more pronounced in mCRC (HR = 16.9, p < 0.001, n = 388). Similarly, in the TCGA validation cohort, the algorithm had AUC of 0.98 and a significant predictive power for PFS (p < 0.001). Conclusion: The novel 7-Gene Algorithm can be further developed as a biomarker model for prediction of treatment response in mCRC patients to improve personalized therapies.

Low‐affinity immunoglobulin gamma Fc region receptor III‐A (FcγRIIIa) is a cell surface protein that belongs to a family of Fc receptors that facilitate the protective function of the immune system against pathogens. However, the role of FcγRIIIa in prostate cancer (PCa) progression remained unknown. In this study, we found that FcγRIIIa expression was present in PCa cells and its level was significantly higher in metastatic lesions than in primary tumors from the PCa cohort (P = 0.006). PCa patients with an elevated level of FcγRIIIa expression had poorer biochemical recurrence (BCR)‐free survival compared with those with lower FcγRIIIa expression, suggesting that FcγRIIIa is of clinical importance in PCa. We demonstrated that overexpression of FcγRIIIa increased the proliferative ability of PCa cell line C4‐2 cells, which was accompanied by the upregulation of androgen receptor (AR) and phosphatidylinositol‐4‐phosphate 5‐kinase alpha (PIP5Kα), which are the key players in controlling PCa progression. Conversely, targeted inhibition of FcγRIIIa via siRNA‐mediated knockdown or using its inhibitory antibody suppressed growth of xenograft PC‐3 and PC‐3M prostate tumors and reduced distant metastasis in xenograft mouse models. We further showed that elevated expression of AR enhanced FcγRIIIa expression, whereas inhibition of AR activity using enzalutamide led to a significant downregulation of FcγRIIIa protein expression. Similarly, inhibition of PIP5K1α decreased FcγRIIIa expression in PCa cells. FcγRIIIa physically interacted with PIP5K1α and AR via formation of protein–protein complexes, suggesting that FcγRIIIa is functionally associated with AR and PIP5K1α in PCa cells. Our study identified FcγRIIIa as an important factor in promoting PCa growth and invasion. Further, the elevated activation of FcγRIII and AR and PIP5K1α pathways may cooperatively promote PCa growth and invasion. Thus, FcγRIIIa may serve as a potential new target for improved treatment of metastatic and castration‐resistant PCa.

Dear Editor, Currently no accurate prognostic test is available to predict prostate cancer (PCa) biochemical recurrence (BCR) after treatment or cancer metastasis.1-7 To address the unmet medical need, we developed a novel 23-Gene Classifier urine test as the first accurate and noninvasive tool for PCa prognosis with potential to improve cancer treatment. We used previously identified biomarkers with differential gene expression in PCa and benign prostate as candidates for BCR prediction and metastasis.8-10 Discriminant analysis was used to assess the ability of various combinations of mRNA expression quantities of the biomarker candidates in prostate tissue specimens collected before prostatectomy with BCR information during follow-up as classifiers to distinguish BCR and non-BCR patients. A 23-Gene Classifier consisting of PTEN, PIP5K1A, CDK1, TMPRSS2, ANXA3, HIF1A, FGFR1, BIRC5, AMACR, CRISP3, PMP22, GOLPH2, EZH2, GSTP1, PCA3, VEGFA, CST3, CCNA1, CCND1, FN1, MYO6, KLK3, and PSCA was found to predict BCR with the highest accuracy. We followed STARD guidelines for biomarker validation. Detailed patient cohorts and study methods are described in Supplementary Methods. The prostate epithelial cells are released into the urine so urine can be used as a noninvasive liquid biopsy source to detect prostate-specific biomarkers for PCa prognosis. The 23-Gene Classifier was developed as a urine test for BCR prognosis using urines collected without digital rectal examination (DRE). Using BCR Urine Prediction Algorithm, the mRNA levels of the 23 genes were used to generate a classification score to predict the patients as having BCR or Non-BCR (Supplementary Methods). A multicenter study was designed prospectively using retrospectively collected urine samples without DRE from 520 patients before prostatectomy or other treatments (IND-CHTN cohort). Forty-six patients developed BCR during the follow-up period averaging 8 years (Table 1). A total of 105 patients from the cohort were randomly selected as a training set to test the 23-Gene Classifier urine test for BCR prediction and the resulting area under the receiver operating characteristic curve (AUC) was 0.94 (95% CI 0.87-1.01). The prognostic performance of the 23-Gene Classifier urine test to predict BCR-free survival was validated in the remaining patients (n = 414). The patients were divided into two risk groups based on diagnosis by the 23-Gene Classifier and Kaplan-Meier survival analysis showed statistically significant association of the 23-Gene Classifier Negative group with shorter BCR-free survival (∼60% BCR-free survival at 48 months) as compared with the 23-Gene Classifier Positive group (100% BCR-free survival at 120 months) (Figure 1A) (log rank P = 0.000). In contrast, the two groups segregated by cancer stage or Gleason score had much smaller difference in BCR-free survival (Figures 1B and C). Univariate and multivariate Cox regression analysis was performed and the 23-Gene Classifier had a hazard ratio (HR) of 1730.90 (95% CI 4.52-6.63E+5) in the univariate analysis (Table 2), which indicated that the patients with a positive 23-Gene Classifier score was 1731 times more likely to have BCR than patients with a negative 23-Gene Classifier score and the BCR prediction was statistically significant (P = 0.014). Its predictive power remained large and significant in multivariate regression after adjusting for cancer stage and Gleason score with HR of 1795.01 (95% CI 4.30-7.49E+5) (P = 0.015). In contrast, cancer stage and Gleason score had much lower HR and were statistically insignificant (Table 2). In addition, univariate and multivariate logistic regression and discriminant analysis were performed to measure the predictive accuracy of the 23-Gene Classifier. The result showed high accuracy with sensitivity of 100% (95% CI 100-100%), specificity of 86.29% (95% CI 82.80-89.79%), and AUC of 0.93 (95% CI 0.90-0.96) (P < 0.0001) (Tables S1 and 3, Figure 1G). Cross-validation of the 23-Gene Classifier showed similarly high accuracy in BCR prediction (Table 3). In contrast, cancer stage and Gleason score had much lower specificity and AUC (Table 3, Figures 1H and I). 100% (100-100%) 7.28% (4.63-9.92%) 10.88% (7.77-13.99%) 100% (100-100%) 100% (100-100%) 2.43% (0.86-3.99%) 10.40% (7.42-13.37%) 100% (100-100%) 100% (100-100%) 86.29% (82.80-89.79%) 45.16% (35.05-55.28%) 100% (100-100%) 23G classifier Cross-validation 100% (100-100%) 86.17% (82.14-90.20%) 45.07% (33.50-56.64%) 100% (100-100%) 100% (100-100%) 88.11% (84.81-91.41%) 48.84% (38.27-59.40%) 100% (100-100%) 16.67% (4.49-28.84%) 99.04% (97.16-100.91%) 85.71% (59.79-111.64%) 77.44% (70.34-84.55%) 48.57% (32.01-65.13%) 96.12% (92.39-99.85%) 80.95% (64.16-97.75%) 84.62% (78.08-91.15%) 86.11% (74.81-97.41%) 100% (100-100%) 100% (100-100%) 95.41% (91.49-99.34%) 23G classifier Cross-validation 87.50% (64.58-110.42%) 100% (100-100%) 100% (100-100%) 96.97% (91.12-102.82%) 85.71% (74.12-97.31%) 100% (100-100%) 100% (100-100%) 95.37% (91.41-99.33%) In silico validation study was conducted to test if the 23-Gene Classifier can also be used in prostate tissue specimens for BCR prognosis using a tissue cohort MSKCC (Table 1). Its similarly high prognostic performance (Tables 2 and 3, Figures 1D-F, K-N) validated the results from the urine study and confirmed the 23-Gene Classifier as a more accurate prognostic tool for BCR prediction than cancer stage and Gleason score. Accurate prediction of cancer metastasis at diagnosis is important for patients to be treated early with effective therapies to prevent development of castration-resistant metastatic cancer and reduce mortality. We tested if the 23-Gene Classifier urine test could be used for metastatic cancer prediction. We tested its performance in the multicenter, retrospective IND-CHTN Cohort (n = 520), a multicenter, prospective 7-HOSPITALS Cohort (n = 207), and a combination cohort combining the patients (n = 727) (Table 1). mRNA expression quantities of the 23 genes were used to classifier each sample as metastatic or nonmetastatic cancer using MET Urine Prediction Algorithm and such classification was compared with the metastatic cancer diagnosis by the imaging measurements to calculate the predictive performance (Supplementary Methods). The result showed that the 23-Gene Classifier urine test had similarly high accuracy in predicting metastatic cancer in the retrospective, prospective and combination cohorts (AUC of 0.92 [95% CI 0.79-1.05] for the retrospective cohort, 0.89 [95% CI 0.83-0.95] for the prospective cohort, and 0.98 [95% CI 0.96-1.01] for the combination cohort) (P < 0.0001). In contrast, Gleason score had much lower specificity and AUC (Table S2 and Figure 2). Development of accurate and actionable prognostic tests is important and urgently needed for PCa treatment. None of the clinicopathological parameters, nomograms, or biomarker panels used in clinic or reported in publications was capable of accurately predicting BCR or cancer metastasis with HR above 20 or AUC above 0.9.1-7 The 23-Gene Classifier had HR above 40 and AUC above 0.9 in all cohorts assessed, suggesting its higher accuracy and more robust performance for PCa prognosis. In addition, the 23-Gene Classifier can be used with prostate tissue specimens. In this study, we developed and validated a novel 23-Gene Classifier that can be used as a highly accurate and noninvasive urine test for prediction of BCR and cancer metastasis with great potential to improve PCa treatment and reduce mortality in clinical practice. The authors would like to thank C. Yun for excellent technical support and S. Liao for skillful assistance in urine collection. The retrospective urine study was approved by IRB at San Francisco General Hospital (IRB #: 15–15816) to use archived urine sediment samples acquired from Cooperative Human Tissue Network Southern Division and Indivumed GmbH. These organizations obtained ethical approval and patient consent prior to collection of patient urine samples. The prospective urine study was approved by IRB at Shenzhen People's Hospital (Study Number: P2014-006) to use urine samples collected from patients treated at the collaborating hospitals in the study with prior consent. All authors have agreed to publish the manuscript. The data supporting this study are available from the corresponding authors upon reasonable request or are publicly available in GEO. Heather Johnson is an employee of Olympia Diagnostics, Inc., and inventor of a pending patent application of prostate cancer diagnostic and prognostic biomarkers. No conflict of interest or financial interest was declared by the other authors. This study was supported by grants from Sanming Project of Medicine in Shenzhen (SZSM201412014), The Science and Technology Foundation of Shenzhen (JCYJ20170307095620828), The Science and Technology Foundation of Shenzhen (JCYJ20160422145718224), and The Shenzhen Urology Minimally Invasive Engineering Center (GCZX2015043016165448) (to Jinan Guo, and Kefeng Xiao); funds from Olympia Diagnostics, Inc. (to Heather Johnson); the Swedish Cancer Society (CAN2017/381), The Swedish Children Foundation (TJ2015-0097), H2020-MSCA-ITN-2018 GlycoImaging (721279), The Swedish National Research Council, the Malmö Cancer Foundation, the Government Health Innovation Grant, the Medical Faculty, Lund University, Kempestiftelserna, Umeå University, Medical Faculty Grants, the Norland Fund for Cancer Forskning, Insamlings Stiftelsen, Umeå University, Bioteknik medel, the Medical Faculty, Umeå University, Medical Faculty Grants, Umeå University, and grant from Umeå University Center for Microbiology Research (UCMR) and Biofilm Center at Malmö University (to Jenny Persson). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. HJ, CZ, LC, KX, and JLP contributed to study concept and design. HZ, JG, XF, CZ, KX, AHBW, and LC participated in study coordination and supervision. JG, TX, FL, and WT contributed to sample collection. XZ, JG, HJ, HZ, and XF contributed to sample processing and analysis. HZ, HJ, AJ, AS, ND, and JLP contributed to data collection and processing, and statistical analysis. HJ, PA, ND, LK, AS, and JLP contributed to data interpretation. XZ, HJ, and JLP contributed to literature search. JG, HJ, HZ, JLP, and CZ contributed to manuscript writing. Please note: The publisher is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.