Real-World Data and Real-World Evidence: Shaping the Future of Clinical Research

Two Hundred and Fifty Devices, Thirty-Five Drugs

American healthcare generates a staggering volume of patient data every day — electronic health records, claims, pharmacy fills, laboratory results, registries and, increasingly, sensor streams. Congress recognised the opportunity in 2016, directing the FDA through the 21st Century Cures Act to build a formal programme for using real-world evidence in regulatory decision-making. Nine years later, the results are lopsided. More than 250 medical device premarket authorizations have incorporated real-world evidence. Drug and biologic applications have done so roughly 35 times.

Figure 1: The bottleneck in pharmaceutical real-world evidence has not been regulatory appetite.

That gap is not explained by regulatory reluctance. The FDA has published a framework, a stack of guidances covering data sources, study design and submission mechanics, and has repeatedly signalled willingness to engage. The gap is explained by something less comfortable for pharmaceutical leadership: most organisations hold real-world data. Comparatively few have built real-world evidence.

For US pharma and biotech executives, that distinction is now a competitive variable. It determines which questions a company can answer without a randomized trial, how quickly it can defend a label, and how credibly it can negotiate with payers who are running their own analyses of the same claims data.

What Separates Real-World Data From Real-World Evidence

The terms are used interchangeably in boardrooms and are not interchangeable at all. Real-world data is information about patient health status and care delivery collected routinely, outside a research protocol. Real-world evidence is clinical evidence about the use, benefits or risks of a medical product derived from analysis of that data. One is raw material. The other is a manufactured product, and the manufacturing process is where most programmes fail.

Figure 2: Provenance, linkage, validation and pre-specification are the conversion steps.

Extracting records from an EHR system and calling the output real-world evidence does not meet any regulator's standard. What does is a dataset with documented provenance, validated variable definitions, transparent handling of missingness, and an analysis specified in a protocol written before the data were examined — the same discipline a sponsor would apply to a prospective trial, applied retrospectively to data it did not design.

“Most organisations hold real-world data. Comparatively few have built real-world evidence. The difference is a manufacturing process, not a database.”

What the FDA Has Actually Committed To

The agency's position is more specific than the industry conversation suggests. Its 2018 framework was followed by guidance on assessing electronic health record and claims data, on registries, on data standards for submissions containing real-world data, on the mechanics of submitting such data, on the design and conduct of externally controlled trials, and on non-interventional studies. Taken together, these documents describe a regulator that has decided how it wants to be persuaded.

They also describe a regulator that remains sceptical in specific places. On externally controlled trials, the FDA has stated plainly that the likelihood of credibly demonstrating effectiveness with an external control is often low, and that sponsors should consult the relevant review division early to establish whether the design is reasonable at all. That caution is not an obstacle to plan around; it is a description of where the evidentiary burden actually sits.

The direction of travel is nonetheless expansionary. In late 2025 the agency relaxed identification requirements for real-world evidence supporting certain device submissions, making large de-identified datasets more usable — a practical constraint that had suppressed adoption more than any principled objection ever did.

CASE STUDY

Where Real-World Evidence Carries Weight — and Where It Does Not

Executives frequently ask whether the FDA accepts real-world evidence. It is the wrong question. The agency accepts it routinely for some purposes and almost never for others, and the distance between those poles is where investment decisions should be made.

Figure 3: Real-world evidence is not one regulatory proposition; it is several, with different odds.

What moves a use case up that scale is rarely a new dataset. It is the strength of the causal argument available around the data: whether a plausible confounder can be measured rather than assumed away, whether the outcome can be ascertained the same way it would have been in a trial, and whether the comparator population was treated in the same era under the same standard of care. Sponsors who invest in those three things before selecting a data vendor tend to arrive at scientific advice meetings with a defensible position rather than a hopeful one.

Fit for Purpose: The Test That Decides Every Submission

Regulators assess real-world data on two axes. Relevance asks whether the dataset actually contains the exposures, outcomes, covariates and follow-up the question requires, in a population resembling the one the label will cover. Reliability asks whether those values are accurate, complete and traceable to their source. A dataset can be enormous and fail both.

The failure modes are well catalogued and rarely surprising in hindsight: outcomes defined by billing codes that were never designed to measure clinical response; index dates chosen in ways that build immortal time bias into the survival curve; confounding by indication that no amount of propensity weighting fully resolves; and analyses refined after the results were seen. Disciplined teams pre-specify the protocol, emulate a target trial explicitly, and register the analysis plan before touching outcome data.

Figure 4: Fund real-world evidence where randomization is hardest and the data asset is strongest.

Beyond the Label: Payer and Commercial Infrastructure

The regulatory use case is the most visible but rarely the largest. US payers, integrated delivery networks and pharmacy benefit managers already analyse claims data to challenge value claims, and federal price negotiation has raised the stakes on demonstrating real-world effectiveness in populations that trials underrepresented — older patients, patients with comorbidities, patients on polypharmacy.

Real-world evidence also increasingly shapes how a product is launched rather than merely how it is defended. Treatment-pattern analyses reveal where a therapy is actually being used and where prescribing has stalled; linkage between claims and clinical data exposes which patient segments discontinue early and why. These are commercial questions answered with the same infrastructure that supports a regulatory submission.

The strategic implication is that a single curated data asset can serve regulatory, health economics, medical affairs and market access at once, provided it is built to the higher standard from the start. Companies that build separate datasets for each function pay several times for the same evidence and defend inconsistent numbers across audiences.

The C-Suite Agenda for Real-World Evidence

Four commitments distinguish organisations generating regulatory-grade evidence from those generating slide decks. Assign single-point accountability: real-world evidence spread across epidemiology, HEOR, medical affairs and commercial analytics produces overlapping activity and no owner. Contract for data rights and portability rather than for reports, so the asset accrues to the company. Engage the review division before the protocol is final, because the questions the agency will ask are cheaper to answer prospectively. And hire genuine causal-inference capability — the analytical bar for observational evidence is higher, not lower, than for a randomized trial.

Governance belongs on the same list. Patient privacy obligations, an expanding patchwork of state health-data laws and the mechanics of tokenized linkage all sit upstream of any analysis, and are far harder to retrofit than to design in.

Conclusion: 

Evidence Generation Becomes a Continuous Function

The randomized controlled trial is not being displaced. It remains the most reliable instrument medicine has for establishing causal effect, and nothing in the real-world evidence agenda changes that. What is changing is the assumption that evidence generation stops at approval. Increasingly, a product's evidence base is assembled continuously — before the trial to define the population, alongside it to contextualise results, and long after launch to defend effectiveness, safety and value in the patients who actually receive it.

For American pharmaceutical and biotechnology leadership, the practical conclusion is unglamorous. The companies that will convert real-world data into regulatory and commercial advantage are the ones investing now in curated data assets, causal-inference talent and early regulatory dialogue — well before a specific question makes the investment urgent. By the time the question arrives, the asset either exists or it does not.