5022 Background: Germline and somatic genomic testing in prostate cancer guide both targeted therapy selection and identification of heritable cancer risk. Although NCCN guidelines define testing eligibility based on disease risk and stage, real-world testing remains suboptimal. Population-level data indicate that only one-third of men with metastatic prostate cancer undergo genomic testing. As part of a larger effort to improve testing rates, we evaluated three consecutive years of prostate cancer care to quantify testing utilization and assessed artificial intelligence (AI) performance for risk stratification and testing eligibility. Methods: We conducted a retrospective chart review of patients with advanced prostate cancer treated at a large multi-site community oncology practice between 2023 and 2025. Clinical data were manually reviewed to assign disease stage, NCCN risk stratification, and eligibility for germline and somatic testing. In 2023, a rules-based AI applied explicit NCCN guideline logic (v1.2023). In 2024 and 2025, an untrained general-purpose AI model (ChatGPT 5.2) was provided NCCN guideline text (v1.2024 and v1.2025) and tasked with inferring risk category and testing eligibility. Primary outcomes included observed annual germline and somatic testing rates and AI concordance with manual review. Temporal trends in germline and somatic testing adoption were assessed using the Cochran-Armitage trend test. Results: Across 2023–2025, NCCN-indicated germline and somatic testing was inconsistently performed (Table 1), with significantly increasing rates from 2023 to 2025 (Cochran–Armitage trend test, germline p < 0.001, somatic p < 0.001). Testing outside of NCCN guidelines was ordered by clinicians in 10 (4%), 22 (9%), and 31 (18%) patients each year, respectively. Conclusions: In a large community oncology setting, NCCN-indicated germline and somatic testing for prostate cancer was frequently underutilized, despite significant improvement over time. Ongoing analyses seek to clarify which interventions contributed to increased testing rates and whether testing performed outside NCCN guidelines reflects broader clinical awareness of indications not consistently captured in structured data. AI-based guideline interpretation of NCCN guidelines demonstrated consistently high concordance with manual review and may offer a scalable approach to systematically identify patients eligible for genetic testing using structured data. Year Charts Reviewed Germline Eligible Germline Tested Somatic Eligible* Somatic Tested AI Risk Stratification AI Testing Recommendation Concordance 2023 259 194 54 (28%) 91 24 (26%) Not assessed Germline 97%; Somatic 100% 2024 255 195 82 (42%) 86 37 (43%) 235 / 255 (92%) Germline 100%; Somatic 100% 2025 177 167 98 (59%) 62 50 (81%) 177 / 177 (100%) Germline 100%; Somatic 100% *Somatic testing was “recommended” for patients with metastatic disease per NCCN 2023 -2025.
Singstock et al. (Wed,) studied this question.