What Makes Real-World Data “Research-Ready”?: A Fitness-for-Use Framework

Every real-world data (RWD) source looks strong on a slide. The patient counts are large. The therapeutic areas are broad. The logos are familiar. On paper, one dataset looks much like another.

Then the study begins. A key outcome variable turns out to be missing from most records. A cohort that looked national skews to a handful of health systems. Follow-up windows are too short to observe what the protocol set out to measure. The data was real. It was just not right for the question.

This is the gap the industry rarely names. Real-world data is the input. Real-world evidence is the output. The distance between them is where studies succeed or fail. Closing that distance is a matter of finding data that fits the question.

“Fit” has a formal name. Regulators and researchers call it fitness-for-use, and it is the single most useful lens a sponsor or CRO can bring to an RWD evaluation. This piece lays out what fitness-for-use means, why patient volume is a poor proxy for it, and the specific questions that separate a research-ready dataset from a large one.

Patient Count Is the Wrong First Question

Volume is the metric that markets itself. It is easy to state, easy to compare, and easy to believe. Size answers only one question. It tells you how many. It does not tell you whether those patients match your population, whether the records contain your variables, or whether you can trust what those records say. A large dataset that lacks your primary endpoint is just the wrong dataset at a larger scale.

The right first question is whether the data can answer the question you’re asking. Patient count is secondary. That reframe is the whole of fitness-for-use, and it has a structure behind it.

The FDA’s Framework for its Real-World Evidence Program organizes that structure around two tests. Data must be relevant to the question, and it must be reliable in what it reports. The Duke-Margolis Institute for Health Policy built on the same two-part frame to define when real-world data is fit for a given use. Peer-reviewed work has since turned it into step-by-step processes for selecting fit-for-purpose data. What follows applies it to the evaluation a buyer actually has to make.

Relevance: Does the Data Contain What Your Study Needs?

Relevance asks whether the dataset holds the information your research question requires. Three dimensions carry most of the weight.

Depth of the clinical record comes first. The variables that matter most for a study are often the hardest to find. Outcomes, exposures, disease stage, biomarker status, and the covariates that control for confounding do not always live in structured fields. They live in clinician notes, pathology reports, and discharge summaries. A dataset built only on structured entries captures a fraction of the clinical picture. It can tell you a diagnosis code was recorded. It often cannot tell you severity, progression, or response. Depth is what turns a record from a billing artifact into a research asset.

Population fit comes second. The right patients must be present, in sufficient number, and representative enough of the population you intend to study. A dataset can be large in total and thin where you need it. Concentration in a few sites or regions can quietly limit how far your conclusions generalize.

Temporality comes third. Evidence generation depends on time. Longitudinal continuity lets you observe what happens before and after an exposure. Adequate follow-up lets you see outcomes that take months or years to appear. A snapshot cannot support a question about change.

Reliability: Can You Trust What the Data Says?

Reliability asks a different question. It asks whether you can trust the values the data reports. Four dimensions matter most.

Provenance and traceability come first. Every data point should trace back to a source record. When evidence faces regulatory or scientific scrutiny, you have to show where each value came from. Auditability makes that possible. A value you cannot trace is a value you cannot defend.

Accuracy comes second. Data that has been curated or abstracted should validate against its source. Ask how that accuracy is measured, and what the validated rate actually is. A claim of accuracy means little without the method and the number behind it.

Completeness comes third. Missing data is not neutral. Gaps that occur at random weaken a study. Gaps that occur systematically bias it. A variable that is missing more often for sicker patients distorts the conclusion in ways volume cannot fix.

Conformance comes fourth. Data must be standardized to be analyzable. Normalized terminology and alignment to a recognized common data model are what let a dataset be studied rather than merely stored. Data that is present but not conformed is data you cannot yet use.

The Questions to Ask Any Real-World Data Source

Fitness-for-use becomes practical when it turns into questions. The following apply to any RWD source, from any provider, including the one you may already use.

1. How were my key variables captured? Structured extraction alone, or clinical abstraction that reaches unstructured records?
2. Can you trace a given value back to its source record?
3. How is abstraction accuracy measured, and what is the validated rate?
4. What are the completeness rates for the specific variables my study depends on?
5. Is the data conformed to a recognized common data model and normalized terminology?
6. Does the population contain enough of my target patients to support my analysis?
7. Is the follow-up window long enough to observe my outcomes?

Clear answers to these questions let you judge whether the data fits. Deflection to patient counts is a warning sign.

Where Clinical Abstraction Changes the Equation

Most of the fitness-for-use gap traces back to a single point. The depth of the clinical record determines both relevance and reliability, and depth is exactly what structured-only data leaves behind.

This is the problem clinical abstraction is built to solve. When the detail buried in notes and reports is recovered through validated abstraction and tied back to the source record, relevance improves because more of the needed variables are present, and reliability improves because each value remains traceable. BEKhealth is one example of an RWD approach organized around this principle. BEKplatform applies AI-driven chart abstraction to EHR data, structuring the unstructured record while preserving the link to source, across a footprint of more than thirty million patients. The lesson holds beyond any one provider. Depth of record is the lever, and abstraction is how that lever gets pulled.

Your Data is Only as Ready as Your Question

Research-ready’ describes a relationship between a dataset and a question. The same data can be research-ready for one study and unfit for another. The same data can be research-ready for one study and unfit for another. A safety signal analysis, a comparative effectiveness study, and an external control arm each demand different variables, different populations, and different follow-up.

That is why fitness-for-use is a discipline rather than a checkbox. It starts with the research question, not the data source. It evaluates relevance and reliability against that question. It treats patient count as one fact among many, not the headline.

The next time an RWD source is presented on a slide, the sharpest thing you can bring is a better question rather than a bigger number. Ask what the data must contain to answer yours, and whether this data does.

Frequently Asked Questions

What does “research-ready” real-world data mean?
Research-ready real-world data is data that is fit for a specific research question, meaning it is both relevant and reliable for that use. Relevance covers whether the data contains the needed variables, population, and time span. Reliability covers whether the values can be traced, validated, and trusted. Research-ready is defined in relation to a question rather than as a fixed property of a dataset.

What is the difference between real-world data and real-world evidence?

Real-world data, or RWD, is the raw information collected during routine care, such as electronic health records, claims, and registries. Real-world evidence, or RWE, is the clinical conclusion derived from analyzing that data. RWD is the input. RWE is the output. Fitness-for-use determines whether a given RWD source can support the intended evidence.

What is fitness-for-use for real-world data?

Fitness-for-use is the assessment of whether a real-world data source can answer a particular research question. It is built on two tests defined by the FDA’s Real-World Evidence Program: relevance, meaning the data reflects the population, variables, and timeframe of interest, and reliability, meaning the data is accurate, complete, traceable, and standardized. A dataset is evaluated as fit or unfit for a specific use, not as universally good or bad.

Why is patient count not enough when evaluating real-world data?

Patient count measures scale, not suitability. A large dataset can still lack the outcome variables, population representation, or follow-up a study requires. Fitness-for-use evaluates whether the data can answer the research question, which volume alone does not indicate.

Why does unstructured EHR data matter for real-world evidence?

Many of the variables most important to research, including disease severity, progression, and treatment response, are recorded in clinical notes rather than structured fields. Data limited to structured entries captures only part of the clinical picture. Recovering detail from unstructured records through validated abstraction increases the relevance and completeness of the resulting dataset.