Real World Evidence AI Platform: What to Require Before You Buy

FDA defines real-world data as "data relating to patient health status and/or the delivery of health care routinely collected from a variety of sources," with examples that include electronic health records, medical claims, and product or disease registries. Real-world evidence, in FDA's words, is "the clinical evidence about the usage and potential benefits or risks of a medical product derived from analysis of RWD."

The distinction matters when you buy. A platform that stores and harmonizes data is handling RWD. Only the analysis, with its design, methods, and documentation, produces RWE. FDA also notes that it has "a long history of using" this kind of data "to monitor and evaluate the postmarket safety of approved drugs," while use to support effectiveness has been "on a more limited basis." The bar for each use is different, and a platform should state which one it is built for.

FDA says it is committed to realizing the potential of "fit-for-purpose RWD to generate RWE." Fit for purpose is a property of the data and the question together. The same claims dataset can be adequate for describing treatment patterns and inadequate for estimating an effect. A platform demo that shows a polished dashboard tells you nothing about whether the underlying data answers your question.

You should be able to trace every analytic dataset back to source systems, transformations, and versions. FDA's RWE program includes guidance on assessing electronic health records and medical claims, and on assessing registries, to support regulatory decisions. A platform that cannot show how a record moved from source to analysis will struggle to support those assessments.

Any model that extracts, links, or classifies data should have a stated context of use, measured performance on your data, a change-control process, and an audit trail. Treat a vendor's model update like any other change to a validated system. Our guidance on 21 CFR Part 11 and GxP AI validation covers the controls that apply.

FDA's draft guidance on non-interventional studies, issued in March 2024 under its RWE Program, gives recommendations to sponsors considering submitting an observational study "to contribute to a demonstration of substantial evidence of effectiveness and/or evidence of safety of a drug." The platform should support pre-specified protocols and analysis plans, not only ad hoc exploration.

Patient-level data brings HIPAA and contractual limits. Require role-based access, de-identification methods you can document, and the right to export your analytic datasets and code, so the evidence outlives the contract.