Citations

78 Background: 177 Lu-PSMA-617 is currently under investigation in combination with an ARPI in those with metastatic hormone-sensitive prostate cancer (PSMAddition trial: NCT04720157). This observational study was performed to assess the effectiveness of 177 Lu-PSMA-617 in combination with an ARPI in patients with mCRPC in the real-world setting. Methods: This was a retrospective cohort study of adults with mCRPC in the PRECISION data platform, a comprehensive database of prostate cancer patients compiled from diverse clinical settings in the US. Patients who received 177 Lu-PSMA-617 during March 23, 2022–June 27, 2025 were identified, among whom two cohorts were selected for inclusion: those who had evidence of ARPI use (apalutamide, darolutamide, abiraterone, or enzalutamide) at any time during 177 Lu-PSMA-617 treatment (concomitant ARPI cohort), and those who had no other treatment for mCRPC during 177 Lu-PSMA-617 treatment (no concomitant therapy cohort). The index date was initiation of 177 Lu-PSMA-617. Patient characteristics and prostate-specific antigen (PSA) responses were evaluated descriptively. Progression-free survival (PFS), defined as the time from 177 Lu-PSMA-617 initiation to disease progression or death, was estimated using Kaplan–Meier analysis. Results: A total of 1,880 patients treated with 177 Lu-PSMA-617 were identified, 140 in the concomitant ARPI cohort and 1,740 in the no concomitant therapy cohort. The median age was 74 years. Demographic characteristics were balanced between the groups, except that median baseline PSA was lower in the concomitant ARPI cohort when compared with patients in the no concomitant therapy cohort (18 ng/mL vs. 32 ng/mL). Before 177 Lu-PSMA-617 initiation, 94.3% and 73.6% of patients in the concomitant ARPI and 80.9% and 56.7% in the no concomitant therapy cohort had received ≥1 prior ARPI and ≥1 prior taxane, respectively. PSA response rates were broadly comparable between cohorts. The median (95% confidence interval) PFS was 14.7 (11.7–21.3) months among patients in the concomitant ARPI cohort vs 12.5 (11.5–13.6) months in the no concomitant therapy cohort. Ongoing follow-up aims to evaluate the impact of combining ARPI with 177 Lu-PSMA-617 on long-term clinical outcomes, including overall survival. Conclusions: The results of this study demonstrate that 177 Lu-PSMA-617 delivers clinical benefits inpatients with mCRPC with or without ARPI; however, addition of an ARPI may potentially improve these benefits. Prospective randomized controlled studies are needed to confirm this.

776 Background: There continues to be a significant unmet need for safe, effective, and bladder sparing treatment options for patients with non-muscle invasive bladder cancer (NMIBC). TARA-002 is a lyophilized biological preparation for intravesical instillation containing inactivated cells of Streptococcus pyogenes (Group A, type 3) Su strain. TARA-002 rapidly enters cancer cells, activating TLR2 and NOD2 to trigger innate immunity, inflammation, and potential immunogenic cell death. ADVANCED-2 (NCT05951179) is an ongoing Phase 2, open-label study to evaluate the safety and efficacy of intravesical TARA-002 in adults ≥ 18 years with high-grade (HG)-NMIBC CIS (± Ta/T1). Preliminary results showed promising complete response (CR) rates and durable responses in patients with HG NMIBC treated with TARA-002 (Jayram et al. AUA 2025). This abstract presents pooled safety data of TARA-002 from the BCG-naïve cohort and BCG-unresponsive cohort of the ADVANCED-2 study. Methods: The ADVANCED-2 study includes 2 cohorts of BCG-naïve (Cohort A) and BCG-unresponsive (Cohort B) participants. Key exclusion criteria include penicillin allergy; history of ≥ T2 bladder cancer, nodal, or metastatic disease; and concomitant prostatic or upper tract urothelial involvement. Each participant is treated with TARA-002 to receive induction (6 weekly doses), reinduction (if persistent disease at 3 months), and maintenance (through 24 months). Response is assessed every 3 months for 2 years. Long-term follow-up is conducted up to 60 months. Safety is monitored throughout the study. Results: As of 07-October-2025, 60 participants (median age: 74; range: 45 to 92), have been enrolled and exposed to TARA-002: 31 BCG-naïve (Cohort A) and 29 BCG-unresponsive (Cohort B). In both cohorts, 18/60 (30.0 %) participants reported drug-related treatment emergent adverse events (TEAEs). Related TEAEs were Grade 1 (18/60; 30.0%) and Grade 2 (3/60; 5.0%), with no reported Grade 3-5 related TEAEs. Commonly reported related TEAEs (≥ 5%) included bladder spasm (10.0%), dysuria (13.3%), fatigue (6.7%), and micturition urgency (6.7%). Most TEAEs were mild and transient. In both cohorts, 10/60 (16.7%) experienced serious AEs (SAEs); 4/60 (6.7%) were Grade 2 and 9/60 (15.0%) were Grade 3. No participants experienced drug-related SAEs or drug-related TEAEs leading to withdrawal or death. Conclusions: TARA-002 monotherapy was well tolerated, thus demonstrating a favorable tolerability profile to date for the treatment of BCG-naïve and BCG-unresponsive HG NMIBC with CIS (±Ta/T1). TARA-002 has a promising safety profile as a bladder-sparing treatment in HG NMIBC across BCG exposure status. Updated safety data will be provided based on evaluable time points at the time of the presentation. Clinical trial information: NCT05951179 .

77 Background: The incidence of metastatic hormone-sensitive prostate cancer (mHSPC) is increasing in the US, likely due to a decrease in screening practices. There has also been a rapid increase in recent years in the number of treatment options that can improve progression-free and overall survival (OS) for these patients. The objective of the current study was to evaluate the clinical outcomes among patients with mHSPC treated in contemporary US clinical practice (2020 to 2025). Methods: This retrospective, observational cohort study used data from PRECISION, a harmonized dataset on advanced prostate cancer in patients in the US treated in a range of clinical settings. Patients aged ≥18 years who were diagnosed with mHSPC during 01/01/2020–01/01/2024 and who initiated ≥1 therapy (index date = first-line [1L] treatment initiation) were included. The study period was from 01/01/2010 to 06/30/2025. Patient characteristics at mHSPC diagnosis and treatment utilization were evaluated descriptively. Castration resistance-free survival (CRFS; defined as the time to castration-resistant prostate cancer or death) and OS were evaluated from index using Kaplan–Meier analysis. Results: A total of 27,708 patients were included (median age, 72 years); 20,015 (72%) were treated in community urology, 4994 (18%) in community oncology, and 2669 (10%) in academic oncology centers at index. Almost half of patients (47%) initiated 1L treatment with androgen-deprivation therapy (ADT) monotherapy, 44% initiated a 1L doublet regimen (36% ADT + androgen receptor pathway inhibitor [ARPI], 5% 1L ADT + radiotherapy, 2% 1L ADT + docetaxel), and 5% a 1L triplet regimen (ADT + radiotherapy + ARPI or docetaxel); the remaining patients received other regimens. The 5-year CRFS rate was 41%; this was significantly higher in patients with a 1L doublet (45%) or triplet (58%) regimen compared with patients with 1L ADT monotherapy (31%; p<0.001). Similarly, the 5-year OS rate was 59% overall and was significantly higher in patients with a 1L doublet (61%) or triplet (59%) regimen compared with patients with 1L ADT monotherapy (55%; p<0.001) (Table). Conclusions: Though both CRFS and OS rates appear to be increasing among real-world patients with mHSPC, >40% of patients in this study died within 5 years of initiating 1L treatment. Patients who received guideline-recommended combined systemic treatment regimens experienced improved CRFS and OS, confirming the importance of implementing therapies in practice that prolong survival. CRFS and OS rates from start of 1L therapy, by therapy type. CRFS rate (%) OS rate (%) 1 y 2 y 3 y 4 y 5 y 1 y 2 y 3 y 4 y 5 y Overall 68 58 51 46 41 93 84 76 67 59 ADT monotherapy 59 49 41 36 31 92 83 74 65 55 Doublet therapy 72 62 55 49 45 94 85 76 68 61 Triplet therapy 85 77 70 68 58 94 89 82 74 59

75 Background: Although the number of life-prolonging treatment options for pts with mHSPC has increased, initial real-world evidence has demonstrated slow uptake. This study utilized a large, contemporary, real-world dataset to evaluate the characteristics and treatment utilization among US pts with mHSPC in recent years. Methods: This retrospective, observational study used data from PRECISION, a harmonized dataset on pts with advanced prostate cancer. Adult men who were diagnosed with mHSPC during 01/01/2020–01/01/2024 (index date) and who initiated ≥1 therapy were included; the study period was 01/01/2010–06/30/2025. Pt characteristics were described at index. Treatment utilization was evaluated from index until progression to metastatic castration-resistant prostate cancer, death, or last follow-up. If an androgen receptor pathway inhibitor (ARPI), docetaxel, or radiotherapy was started within 4 months of index/androgen-deprivation therapy (ADT) start, it was considered a first-line (1L) combination; if it was started after >4 months, it was considered treatment intensification. All analyses were descriptive. Results: Among the 27,708 pts who met the inclusion criteria in PRECISION, 72%, 18%, and 10% were treated in community urology, community oncology, and academic oncology centers at index, respectively. Median age was 72 years; 62% were White; 43% were from the US South; 53% had synchronous and 54% had metachronous metastases (2% unknown). Median prostate-specific antigen level was 9 ng/mL (mean, 130 ng/mL). Prior to mHSPC diagnosis, 19% of pts had a prostatectomy and 17% had radiotherapy. Almost half of pts (47%) initiated 1L treatment with ADT monotherapy; 36% initiated 1L ADT + ARPI, 5% 1L ADT + radiotherapy, 2% 1L ADT + docetaxel, 5% 1L ADT + radiotherapy + ARPI or docetaxel, and 5% other regimens. From 2020 to 2023, 1L ADT monotherapy use decreased, while 1L combination use increased (Table). In addition, 19% of pts with 1L ADT monotherapy had evidence of subsequent treatment intensification, mainly the addition of an ARPI. Conclusions: In this study, use of 1L ADT monotherapy declined over time, accompanied by an increase in 1L combinations and treatment intensification. Nonetheless, approximately 1/3 of pts still did not receive a combination or intensification. Further research is needed to understand treatment trends in pts with mHSPC. Treatment patterns by index year. Pts, % 2020(n=5278) 2021(n=6203) 2022(n=7701) 2023(n=8526) 1L ADT monotherapy* 54 52 45 40 Subsequent treatment intensification* 18 19 20 18 1L ADT + ARPI* 29 33 38 42 1L ADT + radiotherapy 5 5 5 6 1L ADT + docetaxel* 3 2 2 1 1L ADT + radiotherapy + ARPI or docetaxel* 3 3 5 7 *p<0.001 for trend across years.

TPS290 Background: Landmark randomized trials established the survival benefit of adding D to ADT in mCSPC. Building on this foundation, subsequent trials (PEACE-1 and ARASENS) demonstrated further survival improvements when androgen receptor pathway inhibitors (ARPI) were added to the ADT + D backbone. However, the independent contribution of D within the triplet regimen remains unclear given lack of prospective evaluation. Additionally, tumor suppressor gene (TSG) alterations involving TP53, PTEN, and RB1 are frequently observed in PC and are associated with aggressive biology and poor outcomes, providing an opportunity to enhance prognostic accuracy and optimize selection of candidates for treatment intensification. We hypothesize that adding D to ADT plus A will improve survival compared to ADT + A in mCSPC and stratification by TSG status may identify different treatment effects across molecular subgroups. Methods: ASPIRE is an open-label, two-arm, randomized phase 3 study to assess the benefit of D added to ADT plus A for patients with mCSPC. Eligible patients include those with confirmed adenocarcinoma of the prostate with evidence of distant metastatic disease on conventional imaging and are appropriate to receive D. Patients with metachronous low-volume disease or PSMA-PET only disease are excluded. Candidates must have next-generation sequencing (NGS) results from a CLIA-certified tissue test available for TSG status stratification at registration. Eligible patients will receive ADT plus A (240 mg PO daily) +/- D for up to 6 cycles until progression of disease by PCWG3/RECIST version 1.1 criteria. Prior therapy with ADT (with or without ARPI) initiated within 120 days before registration is permitted. The primary endpoint is overall survival (OS). Secondary endpoints include OS by TSG status; radiographic progression-free survival (rPFS); time to castration resistance; symptomatic skeletal event–free survival; time to worsening of disease-related symptoms (NCCN-FACT FPSI-17); safety and tolerability (CTCAE v5.0); PSA90 response at 6 weeks and 6 months; time to PSA progression; and objective response rate in patients with measurable disease. The study will enroll 1200 patients and is currently actively accruing patients. Clinical trial information: NCT06931340 .

743 Background: Although patients with Bacillus Calmette-Guérin (BCG)-unresponsive non-muscle invasive bladder cancer (NMIBC) have several emerging therapeutic options, there remains a clinical unmet need to improve efficacy, durability of response, and tolerability while obviating the need for radical cystectomy. TARA-002 is a lyophilized biological preparation for intravesical instillation containing inactivated cells of Streptococcus pyogenes (Group A, type 3) Su strain. TARA-002 rapidly enters cancer cells, activating TLR2 and NOD2 to trigger the innate immune response, inflammation, and potential immunogenic cell death. Herein we present interim safety and efficacy data of TARA-002 from the BCG-unresponsive cohort of the ongoing ADVANCED-2 study (NCT05951179). Methods: ADVANCED-2 (Cohort B) is a Phase 2, open-label study to evaluate the safety and efficacy of intravesical TARA-002 in BCG-unresponsive adults ≥ 18 years with high-grade NMIBC CIS (± Ta/T1). Key exclusion criteria include penicillin allergy; history of ≥ T2 bladder cancer, nodal, or metastatic disease; or concomitant prostatic or upper tract urothelial involvement. TARA-002 is delivered intravesically for 2 hours and includes induction (6 weekly doses), reinduction (if persistent disease at 3 months), and maintenance (through 24 months). Response is assessed every 3 months for 2 years. Biopsy is mandated at Month 3. Long-term follow-up is conducted up to 60 months. Safety is monitored throughout the study. The primary endpoint is complete response (CR) at any time defined by the absence of any high-grade recurrence. The key secondary endpoint is duration of response. Results: As of 07-October-2025, 32 BCG-unresponsive participants have been enrolled. Majority of participants were White (84.4%, 27 of 32), non-Hispanic (93.8%, 30 of 32), and male (68.8%, 22 of 32). The median age was 74 years (range: 47 to 92). The majority of participants had a baseline diagnosis of CIS only ie, without concomitant Ta/T1 (75.0%, 24 of 32). TARA-002 demonstrated a 68.4% (13 of 19) CR at any time in evaluable participants. The 6-month CR was 75.0% (9 of 12), the 9-month CR was 50.0% (4 of 8), and the 12-month CR was 37.5% (3 of 8). The salvage rate was 75.0% (6 of 8) with reinduction therapy. Most treatment-emergent adverse events (TEAEs) were mild (CTCAE Grades 1-2) and transient; no participants experienced higher than Grade 3 TEAEs. No participants experienced drug-related serious adverse events (SAEs) or drug-related TEAEs leading to withdrawal or death. Conclusions: In participants with BCG-unresponsive high-risk NMIBC with CIS ± Ta/T1, TARA-002 appears to be well tolerated and demonstrated encouraging efficacy, supporting ongoing accrual of the ADVANCED-2 study. Further follow-up will inform durability of response, and data will be updated based on evaluable time points at the time of the presentation. Clinical trial information: NCT05951179 .

696 Background: EVP is the preferred 1st-line therapy for patients (pts) with aUC. Upfront EV dose reductions may be utilized to enhance tolerance and minimize treatment-related toxicities when intolerance is anticipated. Prior studies of EV monotherapy suggest that starting dose 1.25 mg/kg resulted in exposure that maximized chance of response. We hypothesize that upfront EV dose reductions would not compromise EVP efficacy. Methods: Pts with aUC treated with EVP in the multi-site UNITE database were analyzed. Upfront dose reduction was defined as starting dose < 1.25 mg/kg. Correlation between clinical features and EV dose reduction were assessed using Fisher’s Exact test. Endpoints include progression-free survival (PFS) and overall survival (OS) from EVP initiation, assessed by Kaplan-Meier analysis and Cox models. Multivariable Cox analysis (MVA) included age (< 75 vs ≥ 75), ECOG performance status (PS) (0-1 vs 2-3), hemoglobin level (< 10 vs ≥ 10), and liver metastases (present vs absent), chosen in accordance with Bellmunt criteria. Correlation with investigator-assessed observed response rate (ORR) was determined using logistic regressions. All tests were 2-sided with p < 0.05 considered significant. Results: A total of 456 pts across 17 sites were included (median age 72; 77% Caucasian; 75% male; 26% upper tract primary; 66% pure UC; 77% ECOG PS 0-1; 17% with liver metastasis; 89% 1 st line EVP). Upfront dose reduction occurred in 96 pts (21%); 76 pts (79%) started at 1.0 mg/kg and 20 pts (21%) at 0.75 mg/kg. Upfront dose reduction was significantly associated with age ≥ 75 (p = 0.002), ECOG PS 2/3 (p = 0.04), and hemoglobin < 10 (p = 0.01) but not with liver metastasis (p = 0.9). Lower starting EV dose correlated with shorter median PFS (5 vs 6 months) and OS (9 vs 10 months), and lower ORR (28% vs. 60%) when compared to full dose (Table). In MVA adjusting for prognostic factors for PFS and OS, these associations were no longer statistically significant. Conclusions: In this real-world analysis, upfront EV dose reductions were most frequently used in pts with factors associated with frailty and poor prognosis, e.g. older age and ECOG PS 2/3, and were associated with lower ORR. When adjusted for these factors, upfront EV dose reductions were not associated with shorter PFS/OS though a trend towards inferior outcomes persisted. These differences in outcomes may reflect baseline risk factors rather than dose intensity; however, prospective investigation is required to validate our hypothesis-generating results. Association of upfront EV dose reduction with outcomes. Outcome (Upfront dose reduction vs no upfront dose reduction) Univariable analysis Multivariable analysis PFS: HR (95% CI), p 1.50 (1.11 – 2.02), 0.008 1.33 (0.95-1.86), 0.09 OS: HR (95% CI), p 1.56 (1.11-2.20), 0.01 1.41 (0.96-2.06), 0.08 ORR: OR (95% CI), p 0.35 (0.21-0.60), 0.0001 0.38 (0.21-0.67), 0.0009

605 Background: Patients with stage I testicular cancer (TCa) have excellent survival, and surveillance after orchiectomy is often preferred. Limited data exist on survivorship outcomes of surveillance versus adjuvant therapy. We compared systemic health outcomes between patients managed with surveillance and those treated with upfront therapy. Methods: Claims data from the OptumLabs Data Warehouse (2007–2022) identified stage I TCa patients managed with surveillance, chemotherapy, or radiation. Adjuvant therapy was defined as chemotherapy or radiation within 4 months of diagnosis with no further therapy. Chemotherapy codes were required within 42 days to ensure multiple cycles were not given. Propensity score–matched cohorts were created using logistic regression. A 3-year analysis evaluated new diagnoses of cardiopulmonary complications, metabolic syndrome, men’s health, psychological, and miscellaneous systemic conditions. Outcomes were stratified by treatment group. Chi-square tests and Cox regression analyses compared survivorship outcomes between adjuvant therapy and surveillance. Results: Eight-hundred forty-two patients had 3 years of coverage: 531 surveillance, 159 chemotherapy, and 152 radiation. No significant differences were found in cardiopulmonary, metabolic, psychological, or other systemic outcomes between therapy and surveillance (Table 1). Men’s health diagnoses were more frequent with chemotherapy (35.2% vs 23.9%, p=0.027) compared to surveillance. However, Cox regression showed no significant time-to-event differences on time-to-event analysis between groups. Conclusions: Adjuvant chemotherapy may be associated with higher rates of men’s health complications (infertility, erectile dysfunction, hypogonadism), but over time the risk is not significantly different than surveillance. Propensity-score matched survivorship outcomes: A) surveillance vs chemotherapy; B) surveillance vs radiation. A) Surveillance (N=159) Chemotherapy (N=159) p value Cardiopulmonary Complications 0.157 Yes 14 (8.8%) 22 (13.8%) No 145 (91.2%) 137 (86.2%) Men’s Health Complications 0.027 Yes 38 (23.9%) 56 (35.2%) No 121 (76.1%) 103 (64.8%) Metabolic Syndrome Complications 0.631 Yes 53 (33.3%) 49 (30.8%) No 106 (66.7%) 110 (69.2%) Miscellaneous Systemic Disorders 0.879 Yes 25 (15.7%) 26 (16.4%) No 134 (84.3%) 133 (83.6%) Psychological Diagnoses 0.401 Yes 29 (18.2%) 35 (22.0%) No 130 (81.8%) 124 (78.0%) B) Surveillance (N=152) Radiation (N=152) p value Cardiopulmonary Complications 0.682 Yes 12 (7.9%) 14 (9.2%) No 140 (92.1%) 138 (90.8%) Men’s Health Complications 0.630 Yes 51 (33.6%) 55 (36.2%) No 101 (66.4%) 97 (63.8%) Metabolic Syndrome Complications 0.086 Yes 42 (27.6%) 56 (36.8%) No 110 (72.4%) 96 (63.2%) Miscellaneous Systemic Disorders 0.650 Yes 25 (16.4%) 28 (18.4%) No 127 (83.6%) 124 (81.6%) Psychological Diagnoses 0.650 Yes 28 (18.4%) 25 (16.4%) No 124 (81.6%) 127 (83.6%)

BACKGROUND: Left ventricular (LV) dysfunction is a potential cardioembolic source of ischemic stroke, but its role in recurrent stroke risk and treatment response remains unclear. We explore whether LV injury associates with the risk of recurrent stroke and modifies the association between anticoagulation and stroke recurrence using real-world data. METHODS: We performed a multicenter, retrospective study of Cardiac Abnormalities in Stroke Prevention and Recurrence cohort across 27 US sites. Patients with LV ejection fraction ≥20% were included. LV injury, defined as LV ejection fraction 20% to 40% and wall motion abnormality, was the primary exposure and treatment effect modifier. The treatment of interest was anticoagulant versus antiplatelet therapy. The composite outcome included recurrent stroke, major bleeding, or death. Outcomes were evaluated using unadjusted and inverse probability weighting adjusted Cox proportional hazards models, with treatment effect modification tested by LV injury status. RESULTS: Among 2685 patients enrolled, 2328 with complete data were analyzed (median age, 65 years; 49.8% female; median follow-up, 1.6 years). LV injury was present in 310 patients (13.3%). Overall, 535 events occurred: 258 recurrent ischemic strokes, 28 hemorrhagic strokes, 67 major hemorrhages, and 256 deaths. LV injury was associated with a higher unadjusted risk of the primary outcome (hazard ratio [HR], 1.51 [95% CI, 1.21–1.87]), though nonsignificant after inverse probability weighting adjustment (adjusted HR, 1.29 [95% CI, 0.97–1.70]). In the LV injury subgroup, anticoagulation versus antiplatelet therapy was associated with a lower risk of the primary outcome (adjusted HR, 0.24 [95% CI, 0.10–0.59]), relative to the non-LV injury subgroup (adjusted HR, 1.28 [95% CI, 0.83–1.95]; p[LV-interaction], 0.001). Similar interactions were seen for EF 20% to 40% (versus >40%; adjusted HR, 0.19 [95% CI, 0.04–0.86]; p[LV-interaction], 0.001) and wall motion abnormality (versus no wall motion abnormality; adjusted HR, 0.33 [95% CI, 0.15–0.73]). CONCLUSIONS: After cryptogenic stroke, anticoagulation in those with LV injury was associated with lower rates of recurrent stroke, major bleeding, and death. These findings warrant confirmation in a dedicated randomized controlled trial. REGISTRATION: URL: https://www.clinicaltrials.gov ; Unique identifier: NCT06398366.

Abstract Guideline-concordant breast cancer treatment selection is increasingly complex due to rapid evolution in biomarker-driven therapies. Large language models (LLMs), such as GPT-4, offer a potential tool to support clinical decision-making. However, hallucinations and deviation from evidence-based guidelines remain significant concerns. This study evaluated whether retrieval-augmented generation (RAG) using National Comprehensive Cancer Network (NCCN) guidelines improves GPT-4 performance in treatment recommendation tasks. We developed a retrieval-augmented GPT-4 agent (RAG-GPT) by indexing 2025 NCCN Breast Cancer Guidelines into a vector database using sentence embeddings. Forty clinical vignettes were created from published case reports to reflect diverse scenarios, including HER2-low, triple-negative, and BRCA-mutated disease across all stages. Two models—a baseline GPT-4 (without augmentation) and the RAG-GPT—were prompted with structured templates requesting treatment plans, rationale, clinical trial options, and citations.Outputs were independently scored by two blinded oncology-trained reviewers using the modified Generative AI Performance Score (mG-PS). The mG-PS scored outputs for 2 categories, guideline concordance (Gold standard = 1.0, Acceptable = 0.5, Non-concordant = 0.0) and hallucination errors (Severe = -1.0, Moderate = -0.5, Mild = -0.25, None = 0.0). Guideline Concordance was deemed based on NCCN guidelines and what an oncologist would due based on using the guidelines for each case. Hallucinations are a term used for falsehoods in Artificial Intelligence, like made-up facts or studies, incorrect information, improper doses, or inappropriate or wrong treatments. To ensure consistency and minimize inter-rater variability, reviewers were provided with comprehensive scoring guidelines and detailed instructions for each evaluation metric. Scores were normalized between -1.0 and 1.0 for improved evaluation of tools. Readability and rationale clarity were rated on a 5-point Likert scale. Inter-rater reliability was assessed using Cohen’s kappa. The RAG-GPT demonstrated a significantly higher mean mG-PS score (0.599 ± 0.102, 95% CI: 0.498-0.702) compared to the baseline GPT-4, which not only scored lower but had a negative mean (-0.21 ± 0.143, 95% CI: -0.353 to -0.067). For the mG-PS Cohen’s Kappa test resulted in a score of .69 with substantial aggreament. An independent sample t-test showed a difference (t(78) = 29.1, p &lt; 0.0001). Severe hallucinations were also absent in RAG-GPT outputs (0/40) but occurred in 8/40 baseline GPT-4 outputs. The evaluation of readability and rationality also showed notable differences, with the augmented LLM scoring significantly higher (M = 4.35, SD = 0.66) compared to the naïve LLM (M = 3.05, SD = 1.06). An independent samples t-test confirmed this difference t(65.3) = 6.58, p &lt; 0.0001, further supporting the benefit of augmented Large Language models. Cohen’s Kappa test resulted in a score of .76 for a substantial agreement for readability and rational clarity scale. Limitations include small sample size; further fine-tuning and larger real-world validation studies remain necessary. Linking GPT-4 to NCCN guidelines via retrieval-augmented generation significantly improves treatment accuracy and reduces hallucination severity in breast cancer care scenarios. While further refinement is needed, the reduced incidence of severe hallucinations, increase in guideline-concordant recommendations, and enhanced readability findings support further development and prospective validation of guideline-anchored LLMs for clinical use. Citation Format: C. Yost, A. Aseem, B. Callas, S. Monick, L. Bonilla, R. Nguyen, R. Galamaga, S. Osborn, Y. Kumar, N. Ertz-Archambault. Retrieval-augmented GPT-4 improves NCCN-concordant breast cancer treatment recommendations [abstract]. In: Proceedings of the San Antonio Breast Cancer Symposium 2025; 2025 Dec 9-12; San Antonio, TX. Philadelphia (PA): AACR; Clin Cancer Res 2026;32(4 Suppl):Abstract nr PS3-06-29.