From innovation to evidence: Trends in AI/ML clinical trials in GU oncology, 2010-2025.

Abstract

337 Background: Genitourinary (GU) cancers generate complex multimodal data — imaging, pathology, genomics, and clinical variables — that challenge clinical decision-making. Artificial intelligence/machine learning (AI/ML) may enhance care by synthesizing these data points. Yet, the scope, design, and real-world impact of AI/ML GU trials remain unclear. Methods: We queried ClinicalTrials.gov (as of 8/7/2025) for adult GU oncology trials between 2010-2025 using a predefined set of AI/ML keywords. Trials using AI/ML were identified by manual review and classified as “AI/ML primary” (i.e., AI/ML tool development or validation) and/or “therapeutic” (delivering cancer therapy). All classifications — including each trial’s primary AI/ML purpose and data modality — were guided by a standardized codebook and independently coded by two reviewers (Cohen’s κ = 0.48-0.69), with discrepancies adjudicated by a third. PubMed (searched 10/6/2025) was queried by NCT ID to identify peer-reviewed publications and assess dissemination lag. Descriptive statistics were used to summarize trial characteristics, and logistic regression analyses were performed in Stata to assess temporal trends. Results: Of 127 AI/ML GU trials, 107 (84%) were initiated between 2020 and 2025, comprising 2.8% of total GU trials during this period (Table). From 2020–2025, the odds of a GU trial involving AI/ML increased annually (OR 1.28; 95% CI 1.11-1.48; p<0.001). Most trials focused on prostate (n=68, 54%) or bladder (n=25, 20%) cancers and were conducted in China (n=37, 29%) or the USA (n=29, 23%). Common AI/ML applications included detection/diagnosis (n=64, 50%), risk stratification (n=24, 19%), and treatment planning/decision support (n=12, 9%). Radiology imaging was the predominant data modality (n=63, 50%), followed by multimodal (n=32, 25%) and clinical data/text (n=10, 8%). Among 102 AI/ML primary trials, only 33 (32%) were interventional, 12 (12%) were randomized, and 7 (7%) were therapeutic. Just 14 (14%) had a PubMed-indexed publication, of which one was randomized, and none were therapeutic. Conclusions: AI/ML trials in GU oncology are growing rapidly but remain a small fraction of the overall trial activity. Most focus on diagnostics rather than therapeutic integration, and publications remain rare. To realize the promise of AI/ML, future efforts must emphasize trial design rigor, translation into therapeutic applications, and timely dissemination — paralleling the evolution of immunotherapy a decade ago. Temporal trends in GU AI/ML trials (2020-2025). Year Total AI/ML trials (n) Therapeutic AI/ML trials (n) Total GU trials (n) % of GU trials that were AI/ML 2020 5 0 655 0.8 2021 13 0 679 1.9 2022 20 2 605 3.3 2023 21 0 728 2.9 2024 30 2 717 4.2 2025* 18 2 465 3.9 Total 107 6 3849 2.8 *Cutoff 8/7/2025.