Pal, Sumanta K.

5108 Background: We previously proposed a hybrid framework combining LLMs and rule based algorithm (RBA) for automating risk stratification in locPCa. Herein, we aim to validate the risk stratification agent (RSA) in a prospective cohort, develop & evaluate the treatment recommendation agent (TRA) based on NCCN guidelines, and develop an interactive interface to facilitate clinicians for accurate risk stratification and treatment recommendations at the point of care. Methods: This study included pts with locPCa (2004-2024) presenting at Mayo Clinic with at least 1 positive prostate biopsy and MRI report available. For RSA prospective validation, GPT4 extracted key phenotypic variables (PSA, T stage, prostate volume, number of cores, Gleason patterns, grade group) from unstructured MRI and biopsy reports using a zeroshot prompt. An RBA then classified pts into NCCN risk groups. The agent performance was compared with the treating clinician's documentation and evaluated against gold-standard labels manually annotated by two independent clinicians. For development of TRA, two experiments were performed using GPT4 with and without retrieval-augmented generation (RAG) to generate treatment plans. Generated treatment plans were evaluated using NCCN guideline-informed treatment decision tree based algorithm (DTA). Evaluation metrics, weighted – accuracy (acc) and F1, were computed. A clinician facing interface (lisr.org/risk) was developed to provide accurate risk stratification and treatment plans. Results: A total of 858 pts were included (500 for prospective validation, 358 for treatment recommendations). Prospective validation for RSA demonstrated a higher F1 score of 0.89 compared to the treating clinician (F1: 0.58). Treatment recommendation experiments showed that GPT4 with RAG achieved higher acc (64% full – all correct treatment options and 36% partial – at least 1 correct treatment option) compared to GPT4 alone (35% and 65%, respectively). Sensitivity analysis using DTA-informed GPT4 note generation achieved 94% full acc and 6% partial acc. GPT4 alone generated hallucinated treatment options in 71% of cases, while GPT4 with RAG reduced this to 32%. Conclusions: The multi-agent framework based on LLM and RBA achieves high accuracy in risk stratification and treatment recommendation for locPCa. A multi-agent framework with an interactive interface holds high promise to enable efficient, accurate decision-making, and improve locPCa management at the point of care Performance evaluation. RSA(acc; F1) % TRA(full acc; partial acc) % Clinician RSA* GPT4 RAG+GPT4 DTA+GPT4 Overall 60; 58 89; 89 29; 71 64; 36 94; 6 V. High 94; 33 96; 96 55; 45 69; 31 93; 7 High 46; 52 92; 89 - - - Int. Unf 80; 61 91; 90 5; 95 38; 62 93; 7 Int. Fav 57; 64 84; 87 0; 100 98; 2 100; 0 Low 57; 65 87; 80 100; 0 95; 5 90; 10 V. Low 8; 13 50; 67^ - - - *Statistical significance (p<0.01) except for ^V. Low.

PURPOSE The COSMIC-021 study assessed the safety and efficacy of cabozantinib plus atezolizumab in advanced solid tumors. Presented here are results from the expansion cohorts with advanced urothelial carcinoma (UC). METHODS This phase Ib study (ClinicalTrials.gov identifier: NCT03170960 ) enrolled patients with inoperable locally advanced/metastatic UC into four tumor cohorts: first-line cisplatin-eligible (cis-eligible), first-line cisplatin-ineligible (cis-ineligible), previous platinum-containing chemotherapy (previous chemotherapy-treated), and previous immune checkpoint inhibitor (ICI)–treated. Patients received oral cabozantinib 40 mg once daily and intravenous atezolizumab 1,200 mg once every 3 weeks. The primary end point was objective response rate (ORR), as assessed by the investigator per RECIST v1.1 every 6 weeks for 12 months and every 12 weeks thereafter; the secondary end point was safety. RESULTS A total of 121 patients (previous ICI-treated cohort, n = 31, and each of the other cohorts, n = 30) received study treatment from March 2018 to November 2021. The ORR (95% CI) was 30% (15 to 49) for cis-eligible, 20% (8 to 39) for cis-ineligible, 27% (12 to 46) for previous chemotherapy-treated, and 10% (2 to 26) for previous ICI-treated cohorts. The median progression-free survival (95% CI) was 5.5 (1.6 to 11.6), 5.6 (3.1 to 11.1), 5.4 (1.6 to 7.6), and 3.0 (1.8 to 5.5) months, respectively. Grade 3 or 4 treatment-related adverse events (TRAEs) were experienced by 43%, 67%, 57%, and 45% of patients, respectively. TRAEs led to discontinuation of all treatment components in 17%, 13%, 3%, and 19%, respectively. No grade 5 TRAEs were reported in any cohort. CONCLUSION The novel combination of cabozantinib plus atezolizumab exhibited clinical activity in advanced UC that is cis-eligible, cis-ineligible, or previously treated with platinum-containing chemotherapy; clinical activity in previous ICI-treated UC was modest. The toxicity profile was consistent with that previously reported for the combination.

341 Background: Specialist clinicians (oncologists, urologists, radiation oncologists) read through free text reports of prostate MRI and biopsy reports to risk stratify patients during initial consultation of localized PCa. This laborious and time-consuming process is repeated for each day during the clinic. Herein, we propose a hybrid stratification framework utilizing a combination of large language model (LLM) and rule-based programming to automate this process. Methods: This retrospective study included patients with localized PCa (2004-2024) presenting to care at Mayo Clinic with at least one positive prostate biopsy and MRI report available. The state-of-the-art LLM – GPT4 – was utilized using a structured zeroshot prompt to extract relevant phenotypic variables (clinical T stage, Gleason patterns, grade group, extent of specimen/cores involved, prostate size/volume) from unstructured MRI and biopsy reports. Using a rule-based algorithm, the extracted variables and PSA at PCa diagnosis were used to categorize each patient into one of the six NCCN risk categories (very low [VL], low [L], favorable intermediate [FInt], unfavorable intermediate [UFInt], high risk (HR), very high risk [VHR]). Prompts were iteratively developed using ~5% of the total dataset and validated on ~10% of the dataset. Final performance was assessed against a held-out expert annotated test dataset using evaluation metrics (accuracy, F1-score, precision, recall). Additionally, manual annotation was conducted by two independent novice reviewers to compare machine annotated and novice-human annotated risk categorization. Results: A total of 397 patients were included in the evaluation. The median age at diagnosis was 64.8 (IQR: 59.7-68.5); majority of the men were White (n: 366; 92%) and non-Hispanic (n: 371; 94%). The most prevalent risk category was UFint (n: 151; 38%) followed by FInt (n: 80; 20.2%), HR (n: 76; 19.1%), VHR (n: 64; 16.1%), L (21: 5.3%), and VL (n: 5; 1.3%). Novice reviewers achieved an accuracy of 79%, precision of 80%, recall of 82% and a F1-score of 0.81. The hybrid stratification framework using GPT4 achieved an accuracy of 89%, precision of 89%, recall of 88% and a F1-score of 0.88. Among 43 errors by GPT4 risk stratification, majority of the errors were due to miscategorization of clinical T-stage (n: 21; 49%) followed by erroneous number of positive cores (n: 13; 30%). However, additional evaluation showed that GPT4 achieved numerically higher performance (accuracy: 90%; precision: 91%; recall: 90%; F1: 0.90) compared to novice reviewers (accuracy: 77%; precision: 89%; recall: 77%; F1: 0.82) for ascertaining clinical T-stage. Conclusions: Large language models exhibited superior performance than novice clinicians for risk stratification in patients with localized PCa. As the next step, we intend to validate our framework for automated risk classification in a prospective manner.

PURPOSE Patients with no evidence of disease (NED) after metastasectomy for renal cell carcinoma are at high risk of recurrence. Pazopanib is an inhibitor of vascular endothelial growth factor receptor and other kinases that improves progression-free survival in patients with metastatic RCC (mRCC). We conducted a randomized, double-blind, placebo-controlled multicenter study to test whether pazopanib would improve disease-free survival (DFS) in patients with mRCC rendered NED after metastasectomy. PATIENTS AND METHODS Patients with NED after metastasectomy were randomly assigned 1:1 to receive pazopanib 800 mg once daily versus placebo for 52 weeks. The study was designed to observe an improvement in DFS from 25% to 45% with pazopanib at 3 years, corresponding to 42% reduction in the DFS event rate. RESULTS From August 2012 to July 2017, 129 patients were enrolled. The study was unblinded after 83 DFS events (92% information). The study did not meet its primary end point. An updated analysis at 60.5-month median follow-up from random assignment (95% CI, 59.3 to 71.0) showed that the 3-year DFS was 27.4% (95% CI, 17.9 to 41.7) for pazopanib and 21.9% (95% CI, 13.3 to 36.2) for placebo. Hazard ratio (HR) for DFS was 0.90 ([95% CI, 0.60 to 1.34]; Pone-sided = .29) in favor of pazopanib. Three-year overall survival (OS) was 81.9% (95% CI, 72.7 to 92.2) for pazopanib and 91.4% (95% CI, 84.4 to 98.9) for placebo. The HR for OS was 2.55 (95% CI, 1.23 to 5.27) in favor of placebo ( Ptwo-sided = .012). Health-related quality-of-life measures deteriorated in the pazopanib group during the treatment period. CONCLUSION Pazopanib did not improve DFS as the primary end point compared with blinded placebo in patients with mRCC with NED after metastasectomy. In addition, there was a concerning trend favoring placebo in OS.

4504 Background: C, a multitargeted receptor tyrosine kinase inhibitor (TKI), promotes an immune-permissive environment that may enhance response to immune checkpoint inhibitors (ICIs). COSMIC-021, a multicenter phase 1b study, is evaluating C + A (anti‒PD-L1 therapy) in various solid tumors (NCT03170960). C + A demonstrated encouraging clinical activity in cohort 2 of COSMIC-021 in patients (pts) with UC previously treated with platinum-containing chemotherapy (chemo) (Pal S et al. ASCO 2020. Abstract 5013). Outcomes of C + A from 3 other UC cohorts (C3, C4, C5) are presented. Methods: Pts with inoperable locally advanced/metastatic UC with transitional cell histology and ECOG PS 0‒1 were eligible. Pts enrolled in C3 and C4 had no prior therapy and were cisplatin-based chemo ineligible (C3) or eligible (C4). C5 enrolled pts with one prior ICI and no prior VEGFR-TKI therapy. Pts received C 40 mg PO QD and A 1200 mg IV Q3W. CT/MRI scans were performed Q6W for first year and Q12W thereafter. The primary endpoint is objective response rate (ORR) per RECIST v1.1 by investigator. Other endpoints: safety, duration of response (DOR), PFS, and OS. Results: Thirty pts each were enrolled in C3 and C4, and 31 in C5. Baseline characteristics for C3, C4, and C5, respectively: median age, 74 y, 66 y, 68 y; male, 67%, 73%, 55%; ECOG PS 1, 63%, 57%, 74%; lung/liver metastasis; 33%/17%, 40%/20%, 58%/23%; ≥3 tumor sites, 30%, 43%, 45%; bladder as primary site, 67%, 70%, 71%. As of Nov 30, 2021, the median follow-up for C3, C4, and C5 was 27.9, 19.1, and 32.9 mo, respectively, with 1, 6, and 1 pts on treatment. C + A demonstrated clinical benefit across all cohorts (Table). Most common treatment-related adverse events (TRAEs) of any grade across C3, C4, and C5, respectively, were diarrhea (43%, 33%, 35%), nausea (27%, 17%, 26%), fatigue (27%, 27%, 48%), and decreased appetite (33%, 27%, 39%); grade 3/4 TRAEs occurred in 63%, 43%, and 45%, and there was no grade 5 TRAE. Conclusions: C + A demonstrated encouraging clinical activity with manageable toxicity in inoperable locally advanced/metastatic UC as first-line systemic therapy in cisplatin-based chemo eligible/ineligible pts and as second- or later line in pts who received prior ICI. Clinical trial information: NCT03170960. [Table: see text]

PURPOSE COSMIC-021 is evaluating cabozantinib plus atezolizumab in patients with solid tumors. We report results from patients with advanced clear cell (cc) and non–clear cell (ncc) renal cell carcinoma (RCC). METHODS This phase Ib study ( NCT03170960 ) enrolled patients age ≥ 18 years with advanced RCC. A dose-escalation stage was followed by expansion cohorts. For cohort expansion, prior systemic therapy was not permitted for ccRCC but allowed for nccRCC. Patients received oral cabozantinib 40 mg once a day (ccRCC and nccRCC) or 60 mg once a day (ccRCC only) plus atezolizumab (1,200 mg intravenously, once every 3 weeks). The primary end point was investigator-assessed objective response rate (ORR) per RECIST v1.1; the secondary end point was safety. RESULTS A total of 102 patients were enrolled. Median follow-up was 25.8, 15.3, and 13.3 months for the 40-mg ccRCC, 60-mg ccRCC, and nccRCC groups, respectively. ORR was 53% (80% CI, 41 to 65) in the 40-mg ccRCC group (n = 34) and 58% (80% CI, 46 to 70) in the 60-mg ccRCC group (n = 36), 3% and 11%, respectively, with complete response; median progression-free survival (exploratory end point) was 19.5 and 15.1 months, respectively. In nccRCC (n = 32), ORR was 31% (80% CI, 20 to 44), all partial responses; median progression-free survival was 9.5 months. Grade 3 or 4 treatment-related adverse events (TRAEs) were reported by 71% of patients in the 40-mg ccRCC group, 67% in the 60-mg ccRCC group, and 38% in the nccRCC group; TRAEs leading to discontinuation of both agents occurred in 15%, 6%, and 3% of patients, respectively. There were no grade 5 TRAEs. CONCLUSION The novel combination of cabozantinib plus atezolizumab demonstrated encouraging clinical activity and acceptable tolerability in patients with advanced ccRCC and nccRCC. Disease control was observed across dose levels and histologic subtypes.

Abstract Lessons Learned Background Endoglin is an angiogenic receptor expressed on proliferating tumor vessels and renal cell carcinoma (RCC) stem cells that is implicated as a mechanism of resistance to vascular endothelial growth factor receptor (VEGFR) inhibitors. This study evaluated an antiendoglin monoclonal antibody (carotuximab, TRC105) combined with axitinib in patients with advanced or metastatic clear cell renal cell carcinoma (mccRCC) who had progressed following one or more prior VEGF inhibitors. Methods TRAXAR was a multicenter, international randomized 1:1 (stratified by ECOG, 0 vs. 1), phase II study of carotuximab combined with axitinib versus axitinib alone in mccRCC patients who had progressed following one or more prior VEGF inhibitors. The primary endpoint was progression-free survival (PFS) assessed by independent central review (ICR) per RECIST 1.1 Results A total of 150 patients were randomized. The combination therapy resulted in shorter median PFS by RECIST 1.1 than axitinib monotherapy (6.7 vs. 11.4 months). The combination was tolerated similarly to axitinib monotherapy, and there were no treatment related deaths. Exploratory evaluation of pretreatment circulating biomarkers suggested the combination might benefit patients who have low baseline VEGF levels. Conclusion The combination of carotuximab with axitinib did not demonstrate additional efficacy over single agent axitinib in patients with mccRCC who progressed following one or more prior VEGF inhibitor treatment.

BackgroundSystemic therapy (ST) can be deferred in patients who have metastatic renal cell carcinoma (mRCC) and slow‐growing metastases. Currently, this subset of patients managed with active surveillance (AS) is not well described in the literature.MethodsThis was a prospective observational study of patients with mRCC across 46 US community and academic centers. The objective was to describe baseline characteristics and demographics of patients with mRCC initially managed by AS, reasons for AS, and patient outcomes. Descriptive statistics were used to characterize demographics, baseline characteristics, and patient‐related outcomes. Wilcoxon 2‐sample rank‐sum tests and χ2 tests were used to assess differences between ST and AS cohorts in continuous and categorical variables, respectively. Kaplan‐Meier survival curves were used to assess survival.ResultsOf 504 patients, mRCC was initially managed by AS (n = 143) or ST (n = 305); 56 patients were excluded from the analysis. Disease was present in 69% of patients who received AS, whereas the remaining 31% had no evidence of disease. At data cutoff, 72 of 143 patients (50%) in the AS cohort had not received ST. The median overall survival was not reached (95% CI, 122 months to not estimable) in patients who received AS versus 30 months (95% CI, 25‐44 months) in those who received ST. Quality of life at baseline was significantly better in patients who were managed with AS versus ST.ConclusionsAS occurs frequently (32%) in real‐world clinical practice and appears to be a safe and appropriate alternative to immediate ST in selected patients.

256 Background: PARP inhibitors, olaparib and rucaparib recently improved survival outcomes (in the Profound and Triton-2 trials) in pts with aPC harboring HRRm. Notably, ~35% of pts screened for the PROfound trial did not meet eligibility solely due to the lack of adequate quality/quantity of tumor tissue required for CGP (De Bono, NEJM, 2020). cfDNA analysis non-invasively assesses tumor-derived genomic alterations. We evaluated the prevalence of HRRm in a real-world aPC population, who had commercially available cfDNA assay results available. Methods: cfDNA based CGP (using a clinically-validated 73- to 74-gene panel i.e. Guardant360 or G360) from consecutive aPC pts between 11/2016–8/2020 were used for detection of HRRm. Frameshift and nonsense mutations were included as pathogenic. Variants of unknown significance were excluded. In this unmatched study, the prevalence of each HRRm was compared with those reported in literature using the chi square test. Results: cfDNA CGP from 7701 aPC pts were available for interrogation of BRCA1/BRCA2 mutations, & from 4671 aPC pts for ATM mutations. Prevalence of BRCA1 and 2 as detected by cfDNA was similar to historical CGP of primary tissue. Prevalence of BRCA1 was higher than historical CGP of metastatic tissue. Prevalence of ATM in cfDNA was higher than historic tumor tissue CGP but similar to prior reports from cfDNA testing (Table). Conclusions: In this large real-world population of pts with aPC undergoing routine cfDNA CGP by a CLIA certified lab, prevalence of HRRm was similar (or higher for BRCA1 and ATM) to what has been previously reported from the tumor tissue. These data provide the rationale for utilizing cfDNA CGP as a routine test for detection of HRRm in men with aPC to identify men who are candidates for PARPi. [Table: see text]

5564 Background: Cabozantinib (C) may enhance response to immune checkpoint inhibitors (ICIs) by promoting an immune-permissive microenvironment and has shown encouraging activity in combination with ICIs in tumor types including RCC and HCC. C and atezolizumab (A) have shown low objective response rates as monotherapy in metastatic castration-resistant prostate cancer (mCRPC) (Smith JCO 2012; Kim JCO 2018). COSMIC-021 (NCT03170960), a multinational phase 1b study, is evaluating the combination of C + A in various solid tumors. We report results for Cohort 6 in mCRPC. Methods: Eligible patients (pts) were required to have radiographic progression in soft tissue after enzalutamide and/or abiraterone, measurable disease, and an ECOG PS of 0 or 1. Prior chemotherapy for mCSPC was permitted. Pts received C 40 mg PO QD and A 1200 mg IV Q3W. CT/MRI scans were performed Q6W for the first year and Q12W thereafter. The primary endpoint is ORR per RECIST 1.1. Other endpoints include safety, ORR per irRECIST, duration of response (DOR), PFS, and OS. Results are presented for the first 44 pts enrolled. Results: Median follow-up as of Dec 20, 2019 was 12.6 mo (range 5, 20) for the 44 mCRPC pts. Median age was 70 y (range 49, 90), 50% had ECOG PS 1, 34% had visceral metastases, and 61% had extrapelvic lymph node metastases. 27% had prior docetaxel and 52% had 2 prior novel hormonal therapies. The most common any grade treatment-related adverse events (TRAEs) were fatigue (50%), nausea (43%), decreased appetite (39%), diarrhea (39%), dysgeusia (34%), and PPE (32%). One grade 5 TRAE of dehydration was reported in a 90 y/o. Median duration of treatment was 6.3 mo. ORR per RECIST 1.1 among all 44 pts was 32% (2 CRs [4.5%] and 12 PRs [27%]); 21 (48%) pts had SD resulting in a disease control rate of 80% in all pts. One pt with PD per RECIST 1.1 had an irPR per irRECIST. ORR per RECIST 1.1 was 33% in 36 pts with high-risk disease (visceral and/or extrapelvic lymph node metastases). Median DOR for all pts with response per RECIST 1.1 was 8.3 mo (range 2.8, 9.8+). 17 (50%) of 34 pts with post-baseline PSA evaluation had a decrease in PSA. In 12 responders with post-baseline PSA evaluation, 8 (67%) had a PSA decrease ≥50%. Tumor PD-L1 expression will also be reported. Conclusions: The combination of C + A had a tolerable safety profile and demonstrated clinically meaningful activity with durable responses in men with mCRPC. Given the encouraging activity in these pts, especially in those with high-risk disease, further evaluation of C + A in men with mCRPC is being pursued. Clinical trial information: NCT03170960 .