(374) Data linkage and tokenization in action: A systematic review of North American clinical trials utilizing linkage to real world data

Background: : Data linkage and tokenization are increasingly being adopted to address challenges in clinical trials, such as higher costs, evaluating unique endpoints, providing long-term surveillance of patient outcomes, and more.

Objective: : The objective of this systematic review was to quantify examples of published clinical trials that have used data linkage and evaluate the uses of linked data in trial analyses.

Methods: : Relevant articles were identified through PubMed and ClinicalTrials.gov searches implemented on Nested Knowledge, an artificial intelligence-assisted systematic literature review software platform, for articles published from 2014-2024. Grey literature was identified on congress websites and via Google Scholar searches. Articles were included if the clinical study population was in North America (USA and Canada) and the study had a medical intervention. Data on study sponsor, objective, patient disease state, type of linked data, linked data elements, and linkage methods were extracted from each study.

Results: : A total of 698 abstracts were screened, and 20 articles reporting trials with linkage were included in this review. There were 8 interventional trials, 1 phase II, 8 phase III, and 3 phase IV trials The key objectives in data linkage were safety and efficacy (10), cost (3), feasibility (3), survival (3), and prior medical history (1). The disease state of the trial populations were Cancer/Tumors (6), Cardiovascular Risk (3), Aortic Stenosis (2), Kidney Disease (2), and 1 each: Perianal, Rheumatoid Arthritis, Alzheimer's, Joint Replacement, Depression, COVID-19, and Spinal Muscular Atrophy. The studies were sponsored by industry (7), academic (7) and government institutions (6). Studies linked trial data with real-world datasets, including Medicare claims, registries, and administrative health records. The methods used included probabilistic matching, deterministic matching, tokenization, secure data transfer to link patient data across different datasets, unique identifiers for longitudinal tracking, and integrating registry data, and/or manual review for validation. The percentage of the study population that was successfully linked was 74.1% on average (range: 13%-100%).

Conclusions: : The small number of clinical trial publications involving data linkage indicates that this practice is still in the early stages of adoption. This adoption is seen broadly across private and public sectors and generally targets registries, insurance, and health records for safety and efficacy data.

Sponsorship:: Novo Nordisk Inc.