Funt, Samuel A.

1617 Background: Large language models (LLMs) are increasingly evaluated for oncology decision support, yet guideline concordance and hallucinations remain variable across cancer types and tasks. Retrieval-augmented generation (RAG) may improve safety by constraining the evidence base used during generation. Prior oncology LLM studies report discordance with guideline-based care, but few describe a reproducible health-system “vetting mechanism” that couples guideline-only retrieval with clinician-scored safety rubrics across multiple oncology domains. Methods: We assembled 316 de-identified oncology cases in a tumor-board template: medical oncology (breast, GI, GU, leukemia; n=216), gynecologic oncology (n=50), and neuro-oncology/CNS metastatic cancer (n=50). Each case was run on three systems: baseline GPT-4 (O1), NCCN-guideline anchored RAG (O2; guideline-constrained retrieval), and a literature-anchored system (OpenEvidence; O3). Oncologists independently scored outputs using a modified Generative Performance Score (mGPS; −1 to +1) combining guideline concordance with hallucination penalties; readability/rationality was scored on a 1–5 Likert scale. Inter-rater reliability was assessed (κ =.63). This workflow was designed as a pre-deployment model credentialing step for health-system governance (source policy → case bank → clinician-scored acceptance criteria). Results: Across all 316 cases (case-averaged), O2 achieved the highest mean total mGPS (0.75) compared with O1 (0.43) and O3 (0.59). In the medical oncology cohort (n=216), O2 outperformed O1 and O3 (mean mGPS 0.72 vs 0.35 and 0.54), with higher guideline concordance (0.85 vs 0.68 and 0.75) and lower hallucination penalties (−0.09 vs −0.25 and −0.19); readability was also highest for O2 (4.17 vs 3.22 and 3.86). In gynecologic oncology (n=50), O2 again performed best (mGPS 0.827) relative to O3 (0.697) and O1 (0.647). In the CNS cohort (n=50), mean composite mGPS favored O2 (0.82) over O3 (0.68) and O1 (0.55), while readability remained stable across models (3.55–3.73). Severe hallucination/unsafe recommendation rate: O1 [26%], O2 [5%], O3 [17%] (pre-specified definition; e.g., major fabricated citation or guideline-discordant harmful recommendation). Conclusions: In a 316-case, multi-domain clinician-scored evaluation, limiting retrieval to oncology guideline sources improved safety-aligned performance relative to unconstrained generation or literature-only anchoring. A guideline-constrained RAG configuration paired with clinician-scored rubrics and tail-risk reporting can serve as a practical health-system vetting mechanism for governance, re-validation, and change control of LLM decision support, including in resource-limited settings where reliable guideline-consistent recommendations are essential.

BACKGROUND: Immune checkpoint inhibitors are an important therapeutic option for urothelial carcinoma, but durable responses are achieved in a minority of patients. Identifying pre-treatment biomarkers that may predict response to these therapies or who exhibit intrinsic resistance, is of paramount importance. OBJECTIVE: To explore the prevalence of PD-L1 copy number alteration in urothelial carcinoma and correlate with response to immune checkpoint inhibitors. METHODS: We analyzed a cohort of 1050 carcinomas of the bladder and upper urinary tract that underwent targeted next generation sequencing, prospectively. We assessed PD-L1 protein expression, copy number status (next generation sequencing/FISH), and detailed treatment response. RESULTS: We identified 9 tumors with PD-L1 amplification and 9 tumors with PD-L1 deletion. PD-L1 protein expression was the highest in PD-L1 amplified tumors. Of the 9 patients whose tumors harbored PD-L1 amplification, 6 received immunotherapy with 4 deriving clinical benefit, and 2 achieving durable response. Of the 9 patients whose tumors had PD-L1 copy number losses, 4 received immunotherapy with 3 experiencing disease progression. CONCLUSIONS: PD-L1 copy number alterations may serve as potential biomarkers of response to immunotherapy in urothelial carcinoma patients, if validated in larger cohorts.

5017 Background: HDCT/ASCT represents a curative salvage treatment for patients with GCT but is rarely used for other solid tumors. Patients undergoing HDCT/ASCT for hematologic neoplasms require revaccination for their childhood immunizations. Whether this is necessary in patients with GCT is unknown. Methods: In this prospective longitudinal study, patients with GCT undergoing HDCT-ASCT from 11/2010 to 5/2018 had serologies for Measles, Mumps, Rubella, Diphtheria, Tetanus, Polio, and Varicella Zoster measured before HDCT and at 3, 6, and in a subset, 12+ months after the last HDCT with results at these timepoints compared using descriptive statistics. In addition, titer levels at ≥6 months post-transplant were matched 1:1 for age and gender with HL patients who underwent HDCT/ASCT during the same time period. Immunity was compared between cohorts using the Cochran-Mantel-Haenszel test. Results: Of 80 patients with GCT (median age 30, 84% nonseminoma), 91% received 3 sequential transplants and 68 had repeat titers at ≥6 months. Immunity at baseline was >95% for Diphtheria, Tetanus and Polio and 89% for Varicella Zoster but lower for Measles (74%), Mumps (85%), and Rubella (83%) (Table). Compared to baseline, proportional immunity for all infections was similar at 3, 6, and 12 months post-transplant in the GCT population (≥6 months shown in Table). Matching resulted in 58 GCT-HL pairs. One-year immunity was numerically lower for most infections in the HL vs. GCT patients and significantly decreased for Measles and Rubella (Table). Conclusions: To our knowledge, this is the first study to assess vaccine titers following HDCT/ASCT for GCT. We demonstrate that HDCT/ASCT does not result in loss of immunity to childhood vaccines and that GCT patients retain protective titers more frequently than those with HL. However, 15-31% of GCT patients lack MMR immunity at baseline and at 1-year post-ASCT. Therefore, we recommend checking MMR titers at 1-year post-ASCT with revaccination of those lacking immunity. Titer evaluation and revaccination is not necessary for other childhood immunizations.[Table: see text]

4563 Background: We previously reported no difference in favorable response rate (FRR) or PFS for TIP vs BEP. Here we present results of a pre-planned analysis of biomarkers of outcome. Methods: HCG and AFP were drawn on days 1 and 15 of each cycle and rates of decline classified as satisfactory [S] or unsatisfactory [US] by MSK (Motzer JCO 2007) and GETUG (Fizazi Lancet Oncol 2014) methods. IHC for ERCC1, RAD51, PARP1, HER-2, and p-AKT was performed on pre-treatment tumor samples. An H-score (0 – 300) was calculated for each stain (H = stain intensity [0 – 3] x % positive cells [0-100]). H-score and marker decline category were correlated with FRR (PR + CR) and PFS. Patients (pts) who received disease-stabilizing chemotherapy were excluded from marker analyses. Results: Of 91 pts, 80 did not receive disease-stabilizing treatment with 79 having sufficient marker values for analysis by the MSK method and 75 by GETUG. By MSK, 49 had S decline vs 30 US; by GETUG, 34 S vs 41 US. FRR and PFS were improved for pts with S vs US decline by both methods and remained significant by the MSK method when stratified by IGCCCG group (Table). IHC (n=77) quality was adequate in 71 to 73 pts (varied by stain) and was positive (H >0) for PARP in 68/73, ERCC1 in 54/71, RAD51 in 54/73, p-AKT in 5/72, and HER2 in 4/72. Only PARP1 was associated with outcome with worse PFS for the lowest expression tertile (H < 180; p=0.013). Conclusions: PARP1 expression and tumor marker decline rates, particularly by MSK method, were significantly associated with outcome to initial chemotherapy in int/poor risk GCT. Future trials incorporating marker decline into treatment allocation and validating the prognostic effect of PARP1 expression are warranted. Clinical trial information: NCT01873326. [Table: see text]