PURPOSE:A multimodal artificial intelligence (MMAI) biomarker was developed using clinical trial data from North American men with localized prostate cancer treated with definitive radiation, using biopsy digital pathology images and key clinical information (age, PSA, and T-stage) to generate prognostic scores. This study externally validates the biomarker in a prospective, real-world dataset of men who underwent radical prostatectomy (RP) for localized prostate cancer at a tertiary referral center in Sweden. EXPERIMENTAL DESIGN:Association between the MMAI scores (continuous and categorical) and endpoints of interest was assessed with Fine-Gray and cumulative incidence analyses for biochemical recurrence (BCR) and logistic regression for adverse pathology (AP) at RP. RESULTS:The analysis included 143 patients with evaluable biopsy pathology images and complete clinical data to generate MMAI scores. The median follow-up was 8.8 years. At diagnosis, the median PSA was 7.5 ng/mL, the median age was 64 years, 29% had a Gleason grade group ≥3, and 88 men were evaluable for AP at RP. MMAI was significantly associated with BCR [subdistribution HR, 2.45; 95% confidence interval (CI), 1.77-3.38; P < 0.001] and AP at RP (OR, 4.85; 95% CI, 2.54-10.78; P < 0.001). Estimated 5-year BCR rates for MMAI intermediate to high versus low were 25% (95% CI, 16%-36%) versus 4% (95% CI, 1%-11%), respectively. CONCLUSIONS:The MMAI biomarker, previously shown to be prognostic for distant metastasis and prostate cancer-specific mortality in men receiving definitive radiation, was prognostic for post-RP endpoints: BCR and AP. This biomarker validation study further supports the use of MMAI biomarkers in men with prostate cancer outside North America and those treated with RP.

BACKGROUND AND OBJECTIVE:Biochemical recurrence (BCR) after radical prostatectomy (RP) is a heterogeneous disease state in prostate cancer with multiple treatment options. Improved risk stratification could enable more personalized decision-making. We developed and validated a digital pathology-based multimodal artificial intelligence (MMAI) model to predict outcomes in post-RP BCR patients undergoing salvage therapy. METHODS:An MMAI model was trained to predict distant metastasis (DM) using prostate histopathology image features and clinical variables (pathologic grade group, pathologic T stage, prostate-specific antigen level before salvage radiotherapy [SRT], age, and surgical margin). The locked model was validated in 533 patients from NRG/RTOG 9601 and 0534 treated with SRT ± hormone therapy (HT), using Cox regression and time-dependent area under the receiver operating characteristic curve. KEY FINDINGS AND LIMITATIONS:With a median follow-up of 9.3 yrs, MMAI score was significantly associated with DM (subdistribution hazard ratio = 2.17 per standard deviation [95% confidence interval 1.65-2.85]; p < 0.001) and remained independently prognostic after adjusting for clinical variables and treatment. The 10-yr time-dependent area under the receiver operating characteristic curve for MMAI was 0.74 compared with 0.68 for a clinical nomogram. Binary risk categorization demonstrated higher 10-yr DM incidence in the MMAI high-risk (25%) than in the low-risk (8.8%) group. The absolute reduction in 10-yr DM incidence with HT plus SRT versus SRT alone was 21% in the high-risk group versus 2.5% in the low-risk group. Limitations include the use of archived trial cohorts. CONCLUSIONS AND CLINICAL IMPLICATIONS:The post-RP MMAI model provides individualized risk estimates after SRT ± HT and may support shared decision-making about salvage treatment. External and prospective validation are ongoing.

Abstract Purpose: Prostate cancer is a heterogeneous disease ranging from indolent localized to metastatic castration-resistance. Efforts to generate prognostic and predictive biomarkers to understand disease trajectory beyond clinical variables alone include the Decipher Prostate Genomic Classifier (GC) and Artera Multimodal AI (MMAI). Both are validated prognostic biomarkers within localized prostate cancer and are currently being evaluated in the metastatic setting. It is unknown if these biomarkers are reporting on similar biology through different means (gene expression vs digital pathology) or if they are complementary and provide orthogonal insights. Herein, we aim to correlate GC and MMAI scores in patients with metastatic prostate cancer. Methods: We conducted a retrospective review of patients with oligometastatic castration-sensitive prostate cancer (omCSPC) with available transcriptome and digital H&E images from prostate biopsy tissue. GC scores were calculated from RNA sequencing data using the same coefficients but scores were re-scaled to a reference cohort from GRID registry while missing features were imputed as 0. Following digitization of H&E slides, an AI-detected, 128 image feature vector (IFV) was generated per patient which was subsequently combined with Gleason score, PSA, and T stage for final MMAI scoring (Artera, Inc). The primary endpoint was to assess correlations between these biomarkers as continuous variables with linear regression. Given the MMAI score is composed of both AI-detected digital pathology features and clinical features, we evaluated any associations between the GC and AI-detected image features. Uniform Manifold Approximation and Projection (UMAP) was performed on the 128 IFV to generate digital pathology clusters which were then associated with GC both as a continuous and categorical variable using ANOVA and chi-square test, respectively. Results: 85 patients (Metachronous n=74; Synchronous n=11) were included in the analysis. The median GC and MMAI scores were 0.60 and 0.52, respectively. Linear regression identified a very weak positive association between scores (R2=0.08, 95%CI 0.00-0.20). UMAP identified 4 digital pathology clusters. No cluster was found to be enriched with higher GC scores with median scores of 0.64, 0.67, 0.58, and 0.5 for clusters 1-4 respectively (p=0.138). Additionally, no cluster was enriched with either low (GC <.45, p=0.87), intermediate (GC ≥ 0.45-<0.6, p=0.73), or high (GC≥0.6, p=0.12) GC risk groups. Conclusions: We demonstrate for the first time that Decipher GC and Artera MMAI scores do not strongly correlate in a population of patients with omCSPC. This suggests these biomarkers may be complementary, identifying independently prognostic disease biology. Further work validating these findings is warranted. Citation Format: Philip A. Sutera, Yang Song, Amol Shetty, Jarey Wang, Kim Van der Eecken, Alex Hakansson, Yang Liu, Adrianna Mendes, Xiaolei Shi, Elai Davicioni, Emmalyn Chen, Rikiya Yamashita, Timothy Showalter, Tamara Lotan, Theodore DeWeese, Ana Kiess, Daniel Song, Matthew Deek, Piet Ost, Phuoc Tran.Evaluating associations between genomic classifier and digital pathology based mutli-modal AI biomarkers in oligometastatic castration-sensitive prostate cancer.[abstract]. In: Proceedings of the AACR Special Conference in Cancer Research: Translating Targeted Therapies in Combination with Radiotherapy; 2025 Jan 26-29; San Diego, CA. Philadelphia (PA): AACR; Clin Cancer Res 2025;31(2_Suppl):Abstract nr P008

BACKGROUND AND OBJECTIVE:Owing to the expansion of treatment options for metastatic hormone-sensitive prostate cancer (mHSPC) and an appreciation of clinical subgroups with differential prognosis and treatment responses, prognostic and predictive biomarkers are needed to personalize care in this setting. Our aim was to evaluate a multimodal artificial intelligence (MMAI) biomarker for prognostic ability in mHSPC. METHODS:We used data from the phase 3 CHAARTED trial; 456/790 patients with mHSPC had evaluable digital histopathology images and requisite clinical variables to generate MMAI scores for inclusion in our analysis. We assessed the association of MMAI score with overall survival (OS), clinical progression (CP), and castration-resistant PC (CRPC) via univariable Cox proportional-hazards and Fine-Gray models. KEY FINDINGS AND LIMITATIONS:In the analysis cohort, 370 patients (81.1%) were classified as MMAI-high and 86 (18.9%) as MMAI-intermediate/low risk. Estimated 5-yr OS was 39% for the MMAI-high, 58% for the MMAI-intermediate, and 83% for the MMAI-low groups (log-rank p < 0.001). The MMAI score was associated with OS (hazard ratio [HR] 1.51, 95% confidence interval [CI] 1.33-1.73; p < 0.001), CP (subdistribution HR 1.54, 95% CI 1.36-1.74; p < 0.001), and CRPC (subdistribution HR 1.63, 95% CI 1.45-1.83; p < 0.001). The proportion of MMAI-high cases was 50.0%, 83.7%, 66.7%, and 92.1% in the subgroups with low-volume metachronous (n = 74), low-volume synchronous (n = 80), high-volume metachronous (n = 48), and high-volume synchronous (n = 254) mHSPC, respectively. The MMAI biomarker remained prognostic after adjustment for treatment, volume status, and diagnosis stage. CONCLUSIONS AND CLINICAL IMPLICATIONS:Our findings show that the MMAI biomarker is prognostic for OS, CP, and CRPC among patients with mHSPC, regardless of clinical subgroup or treatment received. Further investigations of MMAI biomarkers in advanced PC are warranted. PATIENT SUMMARY:We looked at the performance of an artificial intelligence (AI) tool that interprets images of samples of prostate cancer tissue in a group of men whose cancer had spread beyond the prostate. The AI tool was able to identify patients at higher risk of worse outcomes. These results show the potential benefit of AI tools in helping patients and their health care team in making treatment decisions.

5080 Background: Oligometastatic castration-sensitive prostate cancer (omCSPC) is a state of limited metastatic disease. Randomized trials have demonstrated improvements in progression-free survival in patients with omCSPC treated with metastasis-directed therapy (MDT). However, clinical outcomes remain heterogenous and response to MDT is variable, raising the need for prognostic/predictive biomarkers. A multimodal artificial intelligence (MMAI) biomarker (ArteraAI Prostate Test) was recently trained using data from patients with localized prostate cancer and found to be prognostic. Here, we evaluated this biomarker in omCSPC. Methods: We performed an international multi-institution retrospective review of 221 men with omCSPC who were evaluated with MMAI scoring. The primary objective was to compare overall survival (OS) between patients with high- and low-MMAI score (stratified by median score). OS was defined as the time from diagnosis of omCSPC to death of any cause, calculated with the Kaplan-Meier method and compared using the log-rank test and Cox regression. The secondary objective was to evaluate MMAI score as predictive for MDT treatment effect in a subset of patients enrolled in the STOMP and ORIOLE randomized clinical trials. Given too few OS events for this subset, we evaluated MMAI for metastasis-free survival (MFS), defined as time of randomization to development of a new metastasis or death of any cause and analyzed using Cox regression. An interaction test was performed between treatment arm and MMAI score. Results: Median follow-up of the surviving patients was 38.0 months. Patients with high MMAI (>0.527) were found to have higher PSA (5.8 vs 2.9, p=0.005), higher Gleason score (68.2% vs 38.8% Grade Group ≥4, p<0.001), more likely to have de novo metastatic disease (28.2% vs 8.1%, p<0.001), and more likely to have bone metastases (55.5% vs 39.6%, p=0.019). Patients with a high MMAI had a significantly worse OS (HR=4.38, 95% CI=1.24-15.56; p=0.022) with a median OS of 108.4 mo. vs “not reached” (p=0.012). In the STOMP and ORIOLE subset (N=51; median follow-up=61 months), MMAI was not prognostic for MFS (HR=1.24, 95% CI=0.64-2.43, p=0.52). MMAI however was predictive for MDT benefit as patients with high (HR=0.32, 95% CI=0.12-0.90; p=0.03), but not low (HR=1.59, 95% CI=0.63-4.04; p=0.33) MMAI demonstrated improvement in MFS when treated with MDT (p-interaction=0.02). Conclusions: We have shown for the first time that the ArteraAI MMAI biomarker is prognostic for OS in patients with omCSPC. Further, MMAI appears to be predict benefit of MDT with high MMAI demonstrating a greater improvement in MFS following MDT over observation. Further work in validating these findings is warranted to allow for greater personalization in the management of patients with omCSPC.

5077 Background: The ArteraAI MMAI prognostic biomarker leverages histopathology images and clinical data to risk stratify patients (pts) with localized prostate cancer. This study aimed to validate the MMAI biomarker in mHSPC. Methods: The validation cohort included a subset of pts from the CHAARTED phase III randomized trial with known age at diagnosis, baseline PSA, T-stage, and available histopathology images. Baseline characteristics were compared to examine the balance between included and excluded patients. Using the existing MMAI model, scores were generated, and prognostic ability was evaluated for overall survival (OS) in Cox proportional hazard model. Four defined prognostic groups based on tumor volume (high (HV) vs. low (LV)) and metastatic stage at diagnosis (M1 vs. M0) were included: 1. M0LV, 2. M0HV, 3. M1LV, 4. M1HV. These clinical groups were evaluated using multivariable analysis (MVA) which included MMAI risk group and treatment. Results: A subset of the CHAARTED patient population had evaluable digital histopathological images, allowing inclusion in this study (N=456/790, 57.7%), of which 370 (81.1%) were classified as MMAI-high and 86 (18.9%) as MMAI-intermediate/low risk. Data were available from 394/456 pts for classification into M0LV (N=57), M0HV (N=29), M1LV (N=66), or M1HV (N=242). MMAI-high proportion among groups were 56.1%, 69.0%, 86.4% and 92.6%, respectively. Median follow-up of the censored patients was 4.1 (IQR: 3.3-5.0) yrs. The estimated 5-yr OS across MMAI high, intermediate, and low groups was 39%, 58%, 83%, respectively (log rank p=<0.001). On MVA, the MMAI-high risk group model associated with OS (HR: 1.77 (95% CI: 1.10-2.84) p=0.02) adjusting for treatment arm, volume status, and stage at diagnosis (Table 1). When compared to M0LV, HV subgroups were also associated with OS on MVA. Conclusions: The ArteraAI MMAI model was found to be prognostic for OS among men with mHSPC in CHAARTED. This effect persisted when controlling for treatment, metastatic burden, and metastatic status at diagnosis. On MVA, M0HV and M1HV also maintained prognostic value. This analysis is exploratory in nature, to be followed by development of a model optimized for advanced prostate cancer.[Table: see text]

BACKGROUND:There is a need for clinically actionable prognostic and predictive tools to guide the management of oligometastatic castration-sensitive prostate cancer (omCSPC). METHODS:This is a multicenter retrospective study to assess the prognostic and predictive performance of a multimodal artificial intelligence biomarker (MMAI; the ArteraAI Prostate Test) in men with omCSPC (n = 222). The cohort also included 51 patients from the STOMP and ORIOLE phase 2 clinical trials which randomized patients to observation versus metastasis-directed therapy (MDT). MMAI scores were computed from digitized histopathology slides and clinical variables. Overall survival (OS) and time to castration-resistant prostate cancer (TTCRPC) were assessed for the entire cohort from time of diagnosis. Metastasis free survival (MFS) was assessed for the trial cohort from time of randomization. RESULTS:In the overall cohort, patients with a high MMAI score had significantly worse OS (HR = 6.46, 95 % CI = 1.44-28.9; p = 0.01) and shorter TTCRPC (HR = 2.07, 95 % CI = 1.15-3.72; p = 0.015). In a multivariable Cox model, MMAI score remained the only variable significantly associated with OS (HR = 6.51, 95 % CI = 1.32-32.2; p = 0.02). In the subset of patients randomized in the STOMP and ORIOLE trials, high MMAI score corresponded to improved MFS with MDT (p = 0.039) compared to patients with a low score, with pinteraction = 0.04. CONCLUSION:The ArteraAI MMAI biomarker is prognostic for OS and TTCRPC among patients with omCSPC and may predict for response to MDT. Further work is needed to validate the MMAI biomarker in a broader mCSPC cohort.

Background: Accurate risk stratification is critical to guide management decisions in localized prostate cancer (PCa). Previously, we had developed and validated a multimodal artificial intelligence (MMAI) model generated from digital histopathology and clinical features. Here, we externally validate this model on men with high-risk or locally advanced PCa treated and followed as part of a phase 3 randomized control trial. Objective: To externally validate the MMAI model on men with high-risk or locally advanced PCa treated and followed as part of a phase 3 randomized control trial. Design, setting, and participants: Our validation cohort included 318 localized high-risk PCa patients from NRG/RTOG 9902 with available histopathology (337 [85%] of the 397 patients enrolled into the trial had available slides, of which 19 [5.6%] failed due to poor image quality). Outcome measurements and statistical analysis: Two previously locked prognostic MMAI models were validated for their intended endpoint: distant metastasis (DM) and PCa-specific mortality (PCSM). Individual clinical factors and the number of National Comprehensive Cancer Network (NCCN) high-risk features served as comparators. Subdistribution hazard ratio (sHR) was reported per standard deviation increase of the score with corresponding 95% confidence interval (CI) using Fine-Gray or Cox proportional hazards models. Results and limitations: The DM and PCSM MMAI algorithms were significantly and independently associated with the risk of DM (sHR [95% CI] = 2.33 [1.60-3.38], p < 0.001) and PCSM, respectively (sHR [95% CI] = 3.54 [2.38-5.28], p < 0.001) when compared against other prognostic clinical factors and NCCN high-risk features. The lower 75% of patients by DM MMAI had estimated 5- and 10-yr DM rates of 4% and 7%, and the highest quartile had average 5- and 10-yr DM rates of 19% and 32%, respectively (p < 0.001). Similar results were observed for the PCSM MMAI algorithm. Conclusions: We externally validated the prognostic ability of MMAI models previously developed among men with localized high-risk disease. MMAI prognostic models further risk stratify beyond the clinical and pathological variables for DM and PCSM in a population of men already at a high risk for disease progression. This study provides evidence for consistent validation of our deep learning MMAI models to improve prognostication and enable more informed decision-making for patient care.

299 Background: Recently, an MMAI prognostic biomarker, ArteraAI Prostate, was trained and validated in localized prostate cancer to more accurately risk stratify patients for multiple endpoints compared to NCCN risk groups (Esteva et al., 2022). Prognostication within an NCCN risk group remains clinically important given the multiple treatment decisions required within each risk group (e.g., radiotherapy dose or hormone therapy use). Herein, we validated the MMAI biomarker in high-risk prostate cancer where an increasing number of therapeutic decisions is required. Methods: This study leveraged histopathology image and clinical data from patients with at least one high-risk feature (HRF; cT3-cT4, Gleason 8-10, PSA > 20 ng/mL, primary Gleason pattern 5) from six NRG/RTOG phase III randomized trials (n=1,088). Patients from two trials not part of the initial MMAI biomarker training/validation (RTOG 0521 [n=344] and 9902 [n=318]) and the MMAI validation cohort (RTOG 9202, 9408, 9413, and 9910 [n=426]) were included. Fine-Gray, cumulative incidence, and time dependent area under the curve (tdAUC) analyses were performed for time to distant metastasis (DM) and prostate cancer-specific mortality (PCSM) for standard clinicopathologic variables (age, PSA, Gleason score, T-stage, number of HRFs) and the MMAI model, as a continuous score (per standard deviation increase) and categorically by quartile. Death from other causes were treated as competing risks. Results: The analyzed cohort had a median follow-up of 10.4 years. Median PSA was 21 ng/mL, 60% had Gleason 8-10 disease, 37% had cT3-T4 disease, and 20% were African American. On univariable analysis, the MMAI model was significantly associated with DM (subdistribution hazard ratio [sHR] 2.05, 95% CI 1.74-2.43, p<0.001) and PCSM (sHR 2.04, 95% CI 1.73-2.42, <0.001). On multivariable analysis, the MMAI model, adjusting for either age, PSA, Gleason score, T-stage, or number of HRFs, was the only variable significantly associated with DM. TdAUC was highest for the MMAI biomarker for both 5-year DM (0.71), compared to PSA (0.56), Gleason score (0.61), T-stage (0.63), or number of HRFs (0.64), and for 5-year PCSM (0.75), compared to clinicopathologic variables (range 0.53-0.63). The estimated 10-year DM and 15-year PCSM rates for MMAI quartile 1 vs 4 were 8% vs 31% and 8% vs 34%, respectively. Conclusions: Our novel MMAI prognostic biomarker was successfully validated across six phase III randomized trials with long-term follow-up to be independently prognostic over standard clinical and pathologic variables for men with high-risk prostate cancer. Despite all patients having high-risk disease, the MMAI biomarker identified those with highly variable risks for DM and PCSM. This tool can help enable personalized, shared decision making for patients and providers.

Prostate cancer is the most frequent cancer in men and a leading cause of cancer death. Determining a patient's optimal therapy is a challenge, where oncologists must select a therapy with the highest likelihood of success and the lowest likelihood of toxicity. International standards for prognostication rely on non-specific and semi-quantitative tools, commonly leading to over- and under-treatment. Tissue-based molecular biomarkers have attempted to address this, but most have limited validation in prospective randomized trials and expensive processing costs, posing substantial barriers to widespread adoption. There remains a significant need for accurate and scalable tools to support therapy personalization. Here we demonstrate prostate cancer therapy personalization by predicting long-term, clinically relevant outcomes using a multimodal deep learning architecture and train models using clinical data and digital histopathology from prostate biopsies. We train and validate models using five phase III randomized trials conducted across hundreds of clinical centers. Histopathological data was available for 5654 of 7764 randomized patients (71%) with a median follow-up of 11.4 years. Compared to the most common risk-stratification tool-risk groups developed by the National Cancer Center Network (NCCN)-our models have superior discriminatory performance across all endpoints, ranging from 9.2% to 14.6% relative improvement in a held-out validation set. This artificial intelligence-based tool improves prognostication over standard tools and allows oncologists to computationally predict the likeliest outcomes of specific patients to determine optimal treatment. Outfitted with digital scanners and internet access, any clinic could offer such capabilities, enabling global access to therapy personalization.

BACKGROUND: Androgen deprivation therapy (ADT) with radiotherapy can benefit patients with localized prostate cancer. However, ADT can negatively impact quality of life, and there remain no validated predictive models to guide its use. METHODS: We used digital pathology images from pretreatment prostate tissue and clinical data from 5727 patients enrolled in five phase 3 randomized trials, in which treatment was radiotherapy with or without ADT, as our data source to develop and validate an artificial intelligence (AI)–derived predictive patient-specific model that would determine which patients would develop the primary end point of distant metastasis. The model used baseline data to provide a binary output that a given patient will likely benefit from ADT or not. After the model was locked, validation was performed using data from NRG Oncology/Radiation Therapy Oncology Group (RTOG) 9408 (n=1594), a trial that randomly assigned men to radiotherapy plus or minus 4 months of ADT. Fine–Gray regression and restricted mean survival times were used to assess the interaction between treatment and the predictive model and within predictive model–positive, i.e., benefited from ADT, and –negative subgroup treatment effects. RESULTS: Overall, in the NRG/RTOG 9408 validation cohort (14.9 years of median follow-up), ADT significantly improved time to distant metastasis. Of these enrolled patients, 543 (34%) were model positive, and ADT significantly reduced the risk of distant metastasis compared with radiotherapy alone. Of 1051 patients who were model negative, ADT did not provide benefit. CONCLUSIONS: Our AI-based predictive model was able to identify patients with a predominantly intermediate risk for prostate cancer likely to benefit from short-term ADT. (Supported by a grant [U10CA180822] from NRG Oncology Statistical and Data Management Center, a grant [UG1CA189867] from NCI Community Oncology Research Program, a grant [U10CA180868] from NRG Oncology Operations, and a grant [U24CA196067] from NRG Specimen Bank from the National Cancer Institute and by Artera, Inc. ClinicalTrials.gov numbers NCT00767286, NCT00002597, NCT00769548, NCT00005044, and NCT00033631.)

5001 Background: Androgen deprivation therapy (ADT) improves survival and reduces risk of metastasis in men with high-risk localized prostate cancer (PC) receiving radiotherapy (RT). Predictive biomarkers are needed to guide ADT duration to maximize benefits and minimize risks. We sought to train and validate the first predictive biomarker for long-term (LT) vs short-term (ST) ADT using multiple phase III NRG Oncology randomized trials. Methods: Pre-treatment prostate biopsy slides were digitized from six phase III NRG/RTOG randomized trials of men receiving RT +/- ADT. The artificial intelligence (AI)-derived clinical and histopathological predictive biomarker was trained on RTOG 9408, 9413, 9902, 9910, and 0521 to predict differential benefit of LTADT on distant metastasis (DM). After the AI biomarker was locked, it was validated on RTOG 9202, which randomized men to RT + STADT (4 mo) vs LTADT (28 mo). The predictive utility of the AI biomarker was evaluated for the primary and secondary endpoints of DM and PC-specific mortality (PCSM), respectively, for ADT duration with Fine-Gray interaction models. Event rates were estimated by the cumulative incidence method. Deaths from other causes were treated as competing risks. Results: The AI-derived biomarker was trained on 2,641 men (median follow-up of 9.8 years, IQR [8.2, 11.5]) and validated on 1,192 men from RTOG 9202 (median follow-up of 17.2 years, IQR [9.1, 19.6]), where 80% had at least one high/very high (H/VH) risk feature (cT3-4, Gleason 8-10, PSA > 20, or primary Gleason pattern 5). Consistent with published results, LTADT significantly improved DM (subdistribution HR [sHR] 0.64, 95% CI 0.50-0.82, p < 0.001) in the validation cohort. The AI biomarker was prognostic for DM (sHR 2.35, 95% CI 1.72-3.19, p < 0.001). A significant biomarker-treatment interaction was observed (p = 0.04), in which AI-biomarker (+) men (n = 785, 66%) had reduced DM with LTADT (sHR 0.55, 95% CI 0.41-0.73, p < 0.001), but no benefit was observed (sHR 1.06, 95% CI 0.61-1.84, p = 0.84) for AI-biomarker (-) men (n = 407, 34%). The 10-year DM rate difference between RT + LTADT vs RT + STADT was 13% in AI-biomarker (+) men vs 2% in AI-biomarker (-) men. Similar trends were observed for PCSM outcomes. Risk classification (NCCN intermediate [n = 221, 43% (+)] vs other H/VH risk [n = 954, 71% (+)]) was prognostic but not predictive of LTADT benefit. Conclusions: We have successfully validated the first predictive biomarker of LTADT benefit with RT in localized high-risk PC using an AI-derived digital pathology-based platform in the phase III NRG/RTOG 9202 trial. The predictive AI biomarker identified 34% of men that could derive similar benefit with STADT, avoiding the side effects of prolonged ADT, and 43% of intermediate risk men who would benefit from LTADT.

223 Background: The current standard of care for men with intermediate- and high-risk localized prostate cancer treated with radiotherapy (RT) is the addition of androgen deprivation therapy (ADT). Presently, there are no validated predictive biomarkers to guide ADT use or duration in such men. Herein, we train and validate the first predictive biomarker for ADT use in prostate cancer using multiple phase III NRG Oncology randomized trials. Methods: Pre-treatment biopsy slides were digitized from five phase III NRG Oncology randomized trials of men receiving RT with or without ADT. The training set to develop the artificial intelligence (AI)-derived predictive biomarker included NRG/RTOG 9202, 9413, 9910, and 0126, and was trained to predict distant metastasis (DM). A multimodal deep learning architecture was developed to learn from both clinicopathologic and digital imaging histopathology data and identify differential outcomes by treatment type. After the model was locked, an independent biostatistician performed validation on NRG/RTOG 9408, a phase III randomized trial of RT +/- 4 months of ADT. The DM rates were calculated using cumulative incidence functions in biomarker positive and negative groups, and biomarker-treatment interaction was assessed using Fine-Gray regression such that death without DM was treated as a competing event. Results: Clinical and histopathological data was available for 5,654 of 7,957 eligible patients (71.1%). The training cohort included 3,935 patients and had a median follow-up of 13.6 years (IQR [10.2, 17.7]). After the AI-derived predictive ADT classifier was trained, it was validated in NRG/RTOG 9408 (n = 1719, median follow-up 17.6 years, IQR [15.0, 19.7]). In the NRG/RTOG 9408 validation cohort that had digital histopathology data, ADT significantly improved DM (HR 0.62, 95% CI [0.44, 0.87], p = 0.006), consistent with the published trial results. The biomarker-treatment interaction was significant (p-value = 0.0021). In patients with AI-biomarker positive disease (n = 673, 39%), ADT had a greater benefit compared to RT alone (HR 0.33, 95% CI [0.19, 0.57], p < 0.001). In the biomarker negative subgroup (n = 1046, 61%), the addition of ADT did not improve outcomes over RT alone (HR 1.00, 95% CI [0.64, 1.57], p = 0.99). The 15-year DM rate difference between RT versus RT+ADT in the biomarker negative group was 0.3%, vs biomarker positive group 9.4%. Conclusions: We have successfully validated in a phase III randomized trial the first predictive biomarker of ADT benefit with RT in localized intermediate risk prostate cancer using a novel AI-derived digital pathology-based platform. This AI-derived predictive biomarker demonstrates that a majority of patients treated with RT on NRG/RTOG 9408 did not require ADT and could have avoided the associated costs and side effects of this treatment.

Prostate cancer is the most commonly diagnosed neoplasm in American men. Although existing biomarkers may detect localized prostate cancer, additional strategies are necessary for improving detection and identifying aggressive disease that may require further intervention. One promising, minimally invasive biomarker is cell-free DNA (cfDNA), which consist of short DNA fragments released into circulation by dying or lysed cells that may reflect underlying cancer. Here we investigated whether differences in cfDNA concentration and cfDNA fragment size could improve the sensitivity for detecting more advanced and aggressive prostate cancer. This study included 268 individuals: 34 healthy controls, 112 men with localized prostate cancer who underwent radical prostatectomy (RP), and 122 men with metastatic castration-resistant prostate cancer (mCRPC). Plasma cfDNA concentration and fragment size were quantified with the Qubit 3.0 and the 2100 Bioanalyzer. The potential relationship between cfDNA concentration or fragment size and localized or mCRPC prostate cancer was evaluated with descriptive statistics, logistic regression, and area under the curve analysis with cross-validation. Plasma cfDNA concentrations were elevated in mCRPC patients in comparison to localized disease (OR 5ng/mL = 1.34, P = 0.027) or to being a control (OR 5ng/mL = 1.69, P = 0.034). Decreased average fragment size was associated with an increased risk of localized disease compared to controls (OR 5bp = 0.77, P = 0.0008). This study suggests that while cfDNA concentration can identify mCRPC patients, it is unable to distinguish between healthy individuals and patients with localized prostate cancer. In addition to PSA, average cfDNA fragment size may be an alternative that can differentiate between healthy individuals and those with localized disease, but the low sensitivity and specificity results in an imperfect diagnostic marker. While quantification of cfDNA may provide a quick, cost-effective approach to help guide treatment decisions in advanced disease, its use is limited in the setting of localized prostate cancer.

PURPOSE Cell-free DNA (cfDNA) may allow for minimally invasive identification of biologically relevant genomic alterations and genetically distinct tumor subclones. Although existing biomarkers may detect localized prostate cancer, additional strategies interrogating genomic heterogeneity are necessary for identifying and monitoring aggressive disease. In this study, we aimed to evaluate whether circulating tumor DNA can detect genomic alterations present in multiple regions of localized prostate tumor tissue. METHODS Low-pass whole-genome and targeted sequencing with a machine-learning guided 2.5-Mb targeted panel were used to identify single nucleotide variants, small insertions and deletions (indels), and copy-number alterations in cfDNA. The majority of this study focuses on the subset of 21 patients with localized disease, although 45 total individuals were evaluated, including 15 healthy controls and nine men with metastatic castration-resistant prostate cancer. Plasma cfDNA was barcoded with duplex unique molecular identifiers. For localized cases, matched tumor tissue was collected from multiple regions (one to nine samples per patient) for comparison. RESULTS Somatic tumor variants present in heterogeneous tumor foci from patients with localized disease were detected in cfDNA, and cfDNA mutational burden was found to track with disease severity. Somatic tissue alterations were identified in cfDNA, including nonsynonymous variants in FOXA1, PTEN, MED12, and ATM. Detection of these overlapping variants was associated with seminal vesicle invasion (P = .019) and with the number of variants initially found in the matched tumor tissue samples (P = .0005). CONCLUSION Our findings demonstrate the potential of targeted cfDNA sequencing to detect somatic tissue alterations in heterogeneous, localized prostate cancer, especially in a setting where matched tumor tissue may be unavailable (ie, active surveillance or treatment monitoring).

Author(s): Chen, Emmalyn | Advisor(s): Witte, John S | Abstract: Motivation: Prostate cancer remains the most commonly diagnosed neoplasm in American men, with existing biomarkers (i.e. PSA, nomograms, MRI) having varying levels of sensitivity and specificity in identifying more advanced and potentially aggressive disease. Tumor tissue biopsies remain the gold standard for confirming the presence of prostate cancer, as well as evaluating the genomic heterogeneity and clonal architecture that may be predictive of poor outcomes (i.e. recurrence and metastasis). However, tissue biopsies are limited in their ability to comprehensively assess tumors, and may lead to underestimation of disease grade and stage. These hurdles may be overcome with cell-free DNA (cfDNA), which allows for minimally invasive, repeated sampling through blood draws. This is particularly important when tumor tissue is unavailable during active surveillance or disease monitoring for the detection of residual disease or progression. Additionally, genomic interrogation via cfDNA sequencing typically requires prior knowledge of existing mutations from a patient’s tumor. The work presented here leverages a number of methods to ensure broad, yet sensitive detection of cfDNA variants for patients with localized prostate cancer, including sequencing with a machine-learning guided 2.5Mb targeted panel. In this dissertation, I investigate the use of cfDNA concentration, fragment size, and sequencing to identify advanced prostate cancer, as well as detect somatic mutations present in patient-matched tumors.Methods: The patient cohort included in these studies are composed of 268 individuals: 34 healthy individuals, 112 men with localized prostate cancer who underwent radical prostatectomy (RP), and 122 men with metastatic castration-resistant prostate cancer (mCRPC). Plasma cfDNA concentration and fragment size were quantified with a Qubit fluorometer or Bioanalyzer utilizing a chip-based capillary electrophoresis method for nucleic acid analysis. Low-pass whole-genome and targeted sequencing were used to identify single nucleotide variants (SNVs), small insertions and deletions (indels), and copy number alterations (CNAs) for a subset of patients. Plasma cfDNA was barcoded with duplex Unique Molecular Identifiers (UMIs) to construct consensus reads and improve variant detection by leveraging duplicate reads and sequence complementarity of the two DNA strands. Extensive tissue sampling was used to capture tumor heterogeneity and provide a patient-specific gold standard for comparison of matched cfDNA.Results and Conclusions: Patients with advanced mCRPC had higher cfDNA concentration than men with localized disease or healthy controls, and those with localized disease had shorter average fragment sizes than controls. Importantly, cfDNA concentration and fragment size remained independent predictors after adjusting for age and PSA. We found that targeted sequencing of cfDNA—without a priori patient-specific tumor mutation information—identified somatic alterations found in matched tumor tissue from multiple regions, potentially allowing for dynamic monitoring of emerging resistant subclones throughout the course of disease. Detection of these concordant variants was associated with seminal vesicle invasion and the number of somatic variants found in the tumor tissue samples, predicating its use for patients with poor prognostic factors in a localized setting. Similar to cfDNA concentration, plasma cfDNA mutational burden was also found to increase with disease severity. The results from our studies demonstrate the ability of cfDNA to identify somatic variants in patients with heterogeneous, localized prostate cancer.

Early cancer diagnosis, especially while the disease is still localized and before symptoms appear, results in significantly higher survival rates compared to late-stage diagnosis. At the time of diagnosis, it is also common to find multiple foci within a single prostate gland in men with localized disease. Cell-free DNA (cfDNA) may not only reflect underlying disease biology, but also simultaneously allow for the identification of genetically distinct tumor subclones. The objectives of this study are to determine if 1) cfDNA levels are able to distinguish between healthy individuals from patients with localized or metastatic castration-resistant prostate cancer (mCRPC), and 2) if somatic mutations identified in tumor tissue are detectable in cfDNA and representative of the distribution observed in tumor tissue. This study includes samples from 130 individuals at UCSF: 21 healthy donors, 100 patients with localized prostate cancer who underwent radical prostatectomy (RP), and 9 mCRPC patients. Blood samples and matched tissue from adjacent normal seminal vesicles and multiple tumor regions (1-9 samples per patient) were collected from patients undergoing radical prostatectomy. CfDNA was extracted from plasma, and the concentration and fragment length distribution were measured with a Bioanalyzer 2100. Comparisons of cfDNA levels between groups were assessed with a Welch’s t-test due to the potential for unequal variances. Fifty-seven samples from nine patients have been subjected to whole exome sequencing, and 22 samples from five patients have been subjected to whole genome sequencing at ~40x coverage. Somatic variant calling was performed with Broad Institute’s Firecloud platform (GATK4/MuTect2) for tumor tissue and with the Curio platform for cfDNA to build consensus sequences leveraging unique molecular tags. CfDNA levels are able to distinguish between healthy and localized (p = 0.005), as well as healthy and metastatic groups (p = 0.043). Tumor foci within a patient’s prostate gland are genetically heterogeneous, with the majority of somatic mutations private to tumor regions and a subset of mutations at the intersection of these regions. Preliminary analyses result in an average of 72 somatic SNVs and indels per tumor tissue region, with ~10% overlap between regions in the same patient. Additional mCRPC and follow-up blood samples are being collected from all patients. The association between cfDNA levels prior to RP surgery and biochemical recurrence will be investigated when follow-up collection concludes. Further analysis of mutational concordance, clonality, and copy number variation between tumor tissue DNA and cfDNA, along with clinical data, will be performed. Citation Format: Emmalyn Chen, Clinton Cario, Lancelote Leong, Karen Lopez, Patricia Li, Erica Oropeza, Imelda Tenggara, Janet Cowan, Jeffry Simko, Daniel Wells, Robin Kageyama, June Chan, Terence Friedlander, Pamela Paris, Peter Carroll, John Witte. Assessing the utility of cell-free DNA in identifying prostate cancer and characterizing tumor heterogeneity via whole exome and whole genome, multi-region sequencing [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2019; 2019 Mar 29-Apr 3; Atlanta, GA. Philadelphia (PA): AACR; Cancer Res 2019;79(13 Suppl):Abstract nr 1371.

Background Cell-free DNA's (cfDNA) use as a biomarker in cancer is challenging due to genetic heterogeneity of malignancies and rarity of tumor-derived molecules. Here we describe and demonstrate a novel machine-learning guided panel design strategy for improving the detection of tumor variants in cfDNA. Using this approach, we first generated a model to classify and score candidate variants for inclusion on a prostate cancer targeted sequencing panel. We then used this panel to screen tumor variants from prostate cancer patients with localized disease in both in silico and hybrid capture settings. Methods Whole Genome Sequence (WGS) data from 550 prostate tumors was analyzed to build a targeted sequencing panel of single point and small (< 200 bp) indel mutations, which was subsequently screened in silico against prostate tumor sequences from 5 patients to assess performance against commonly used alternative panel designs. The panel's ability to detect tumor-derived cfDNA variants was then assessed using prospectively collected cfDNA and tumor foci from a test set 18 prostate cancer patients with localized disease undergoing radical proctectomy. Results The panel generated from this approach identified as top candidates mutations in known driver genes (e.g. HRAS) and prostate cancer related transcription factor binding sites (e.g. MYC, AR). It outperformed two commonly used designs in detecting somatic mutations found in the cfDNA of 5 prostate cancer patients when analyzed in an in silico setting. Additionally, hybrid capture and 2500X sequencing of cfDNA molecules using the panel resulted in detection of tumor variants in all 18 patients of a test set, where 15 of the 18 patients had detected variants found in multiple foci. Conclusion Machine learning-prioritized targeted sequencing panels may prove useful for broad and sensitive variant detection in the cfDNA of heterogeneous diseases. This strategy has implications for disease detection and monitoring when applied to the cfDNA isolated from prostate cancer patients.

At the time of diagnosis, it is common to find multiple spatially distinct foci within a single prostate gland in men with prostate cancer. Recent studies evaluating the genetic heterogeneity of localized, multifocal prostate cancers through whole-genome sequencing and whole-exome sequencing have only recently been described. These tumors were highly heterogeneous for single-nucleotide variants, copy number alterations, and genomic rearrangements. One exciting use of cfDNA is its potential ability to simultaneously capture all genetically distinct tumor subclones. Currently, the degree to which a cfDNA sample is representative of the entire genetic landscape of localized prostate cancer is unknown. The objective of this study is to determine if somatic mutations identified in tumor tissue are detectable in cfDNA. Tumor tissue and blood samples have been collected from twenty-nine patients undergoing radical prostatectomy. Samples from fourteen patients have already been subjected to whole-exome sequencing with a target sequencing depth of 200X (HiSeq 4000). Preliminary analyses indicate that tumor foci within a single prostate gland in an individual are genetically heterogeneous. Further analysis of mutational concordance between tumor tissue DNA and cfDNA along with clinical data will be performed. Citation Format: Emmalyn Chen, Clinton L. Cario, Lancelote Leong, Karen Lopez, Jeffry P. Simko, Peter R. Carroll, Caroline Tai, John S. Witte. Assessing the genetic heterogeneity of localized, multifocal prostate cancer via cell-free DNA [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2018; 2018 Apr 14-18; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2018;78(13 Suppl):Abstract nr 3669.