What are the challenges for building real world comparative cohorts in oncology?

Abstract

e19285 Background: There has been a proliferation of single arm trials in oncology being presented to regulators. An external comparator using real world data (RWD) can help establish context to trial results and direct comparison to the treatment arm by mirroring the inclusion/exclusion criteria for the trial and examining trial outcomes in an RWD external comparator cohort. RWD has successfully supported some regulatory submissions, although there have been failures as well, with comparability of patients and endpoints under scrutiny. Methods: IQVIA has partnered with life sciences companies on > 20 external comparator projects from 2017-present using a mix of data acquisition strategies (e.g., chart review, database extraction) and methods (e.g., real-world cohort vs. real-world benchmark, propensity matching vs. inverse probability treatment weighting). We collated, reviewed, and synthesised the challenges and risk mitigations documented throughout study execution in aggressive and indolent lymphomas, multiple myeloma, Merkel cell carcinoma, and synovial sarcoma. Results: Patient characteristics used to assess trial eligibility (e.g., biomarker expression, International Prognostic Index, ECOG status) and outcomes (e.g., response rate) are not often captured in a structured format in RWD or recorded in routine clinical practice. As a result, there is a greater reliance on chart reviews today over database extractions to fulfil external comparator requirements, though in some diseases, databases have proven to be a faster and more cost-effective solution. Conclusions: To improve the success of external comparator studies, trial design should be informed by existing RWD so that relevant real-world endpoints and outcome evaluation criteria are included alongside RCT standards. Advances are needed to facilitate RWD capture that matches trial data more closely. This could include bespoke data collaborations, biobank linkage, and natural language processing to drive study execution in databases in the future.