Presentation (CTI)

OBJECTIVES: Manual surveillance of clinical trial registries is time-consuming. We developed an automated, rule-based tool, generating structured trial intelligence from ClinicalTrials.gov, and validated the output in polycythemia vera as a case study.
METHODS: The tool retrieves records via the ClinicalTrials.gov application programming interface and tabulates phase, recruitment status, sponsor, design, interventions, comparators, and endpoints. Three deterministic procedures were implemented: case-insensitive, normalisation of intervention names to canonical entities; rule-based classification of arms as active-comparator, placebo-controlled, standard-of-care, or single-arm; and criterion-based selection of landmark trials (enrolment, head-to-head design, recency, phase III, quality-of-life endpoints). Active interventions were ranked by a momentum score summing per-trial weights (recruiting +3, active or not yet recruiting +2, started <3 years +1, phase II +1, III +2, IV +1). Retrieval and normalisation were checked against the registry export; comparator classification was validated against independent human review.
RESULTS: A "polycythemia vera" query on 21 June 2026 returned 361 trials, of which 87 were active, 188 were completed, and 50 were phase-III, consistent with source registry. Intervention normalisation regrouped 424 distinct raw names to 380 canonical entities across 863 mentions. Benchmarked against a human reference standard, per-arm drug-set classification by the tool achieved 95% agreement (precision 0.86, recall 0.93), correctly categorising dose-ranging, combination, double-dummy, and mis-coded arms. Interventions with the highest momentum scores, were ruxolitinib (57), hydroxyurea (23), pacritinib (17), momelotinib (15), and ropeginterferon alfa-2b (14). Spleen volume/response and MPN-SAF total symptom score were the most assessed endpoints across trials.
CONCLUSIONS: The Trial Landscape Scanner converts weeks of manual registry searching into an automated workflow, generating the landscape in under one minute. It can support HTA evidence planning, network meta-analysis feasibility, competitive intelligence, and evidence-gap identification. Deterministic normalisation and label-aware classification improve interpretability and robustness to inconsistent coding. Further studies are warranted to test the generalisability of these findings across multiple indications and interventions.