Protocols get most of the attention when a registry or real-world evidence study is planned, but it is the readiness of the data at each site that usually decides whether timelines, amendments and budgets hold.
Registries and real-world evidence studies are among the most valuable tools a medical affairs team has. They can answer questions trials were never designed to answer, build relationships with investigators, and generate local evidence that payers and clinicians trust. They are also among the easiest projects to underestimate.
The reason is rarely the science. Protocols are reviewed carefully, endpoints are debated and statistical plans are signed off. What receives far less scrutiny is whether the information each site holds can actually be turned into the dataset the protocol assumes. When that gap is discovered after sites are activated, the fix is slower and more expensive: amendments, retraining, extra abstraction time and, sometimes, variables that simply cannot be collected. The checks below bring those questions forward to the point where they cost the least to answer.
When a data problem is found matters as much as the problem itself. Illustrative, not to scale.
Checks 1 to 3
For every variable in your draft case report form, find out where it sits at each site today: a structured field in the hospital system, the laboratory system, a pharmacy record, a scanned referral letter, or a clinician's note. The same variable may live in a different place at every site. This mapping is the single most useful piece of feasibility work you can do, because it shows which variables will flow easily and which will need to be found, read and interpreted by a person.
Ask your sites
For each variable on our list, where is it recorded today, and in what form?
A feasibility questionnaire tells you what sites believe they hold. A small sample of real, de-identified records tells you what they actually hold. Reviewing even a handful of patient files per site, before the protocol and case report forms are locked, often reveals missing fields, unexpected formats and inconsistent documentation that would otherwise surface months later. It is one of the cheapest ways to protect a study timeline.
Ask your sites
Could we review a small number of de-identified records for eligible patients, exactly as they are stored today?
Much of the clinical story sits outside structured fields: in clinical notes, discharge summaries, imaging reports, histopathology reports and referral letters, sometimes in more than one language. Disease stage, reasons for switching treatment, adverse events and response assessments are common examples. If a large share of your variables live in unstructured text, the study needs a clear plan and budget for extracting them reliably.
Ask your sites
Which of our variables would need to be read from notes, reports or scanned documents rather than pulled from a system?
Checks 4 to 6
Sites record the same clinical fact in different ways. Test names vary, units differ, reference ranges are local, and some institutions maintain their own code sets. Medication names may appear as brands in one site and generics in another. None of this is a flaw in the sites; it is how care is documented. But it means data from different sites cannot simply be combined, and the work of making it consistent needs clinical judgement.
Three records, one clinical fact. Multiply this across every variable and every site.
Ask your sites
Which local names, units, reference ranges or code sets do you use for the variables on our list?
In many studies, site staff abstract data by hand alongside their clinical work. Research coordinators and nurses are often stretched across several studies, and manual abstraction is slow, tiring and prone to variation between people. Understanding who will do this work, how long each patient record is likely to take, and how that time is supported helps you set realistic recruitment and data timelines, and it shows respect for the sites you depend on.
Ask your sites
Who would abstract the data, roughly how long would one patient record take, and how many other studies are they supporting?
Data quality is easier to design in than to fix later. Decide upfront how data will be checked: which fields are verified against source, how discrepancies are raised and resolved, and who with clinical training reviews the information that needs interpretation. A clear quality plan also makes your results easier to defend with reviewers, payers and regulators, who will want to know how reliable the underlying data is.
Ask your study partner
How will data be checked against source, who performs clinical review, and how is accuracy measured and reported?
Checks 7 and 8
An endpoint that is precise on paper can be ambiguous in practice. Response, progression, remission, exacerbation and treatment failure are often documented in narrative form, using different criteria at different sites, or not documented explicitly at all. Before finalising endpoints, check how clinicians at your sites record them in routine care, and decide how the study will handle cases where the record is incomplete or open to interpretation.
Ask your sites
How do your clinicians record this outcome in routine care, and which criteria do they use?
Many registries combine historical records with new data collected going forward. The two behave differently. Historical records were captured for care, not for your study, so they are more likely to be incomplete, unstructured and inconsistent. Prospective data can be shaped by the study, but only if sites are supported to capture it well. Treating both the same way usually means the historical data is under-resourced.
Ask your sites
How far back do usable records go, and what changes in systems or documentation practice have happened in that time?
Checks 9 and 10
Consent models, ethics requirements and data access approvals vary between countries and often between institutions in the same country. Some sites will need patient consent for retrospective data; others may allow it under a waiver or an approved research framework. Mapping these requirements site by site, before activation, avoids a situation where part of the dataset cannot be used as planned.
Ask your sites
What approvals and consent are required for retrospective and prospective data at your institution, and how long do they usually take?
Trace the path the data will take from each site to the final analysis dataset: who extracts it, where it is stored, where it is processed, who can see record-level data, and whether it crosses borders. Data protection laws in the region, and the expectations of some institutions, may require processing in country or limit who can access identifiable information. Designing the flow with this in mind from the start is far simpler than retrofitting it.
Ask your study partner
Where will data be stored and processed at each step, and can processing stay in country where required?
Checks 11 and 12
Much of the effort in a registry or RWE study goes into making site data usable. If that work is done only for one analysis, much of its value is lost when the study closes. Using consistent definitions, keeping a clear record of how data was transformed, and agreeing permissions that allow appropriate follow-on research can turn a single study into a foundation for future evidence, with the right governance in place.
Ask your study partner
Will the prepared dataset, definitions and transformation records be reusable for future questions, and under what permissions?
Study budgets often cover site fees, monitoring and analysis in detail, while the work of extracting, cleaning, harmonising and clinically reviewing site data is folded into general costs or assumed to be small. When it turns out to be larger, the money has to come from somewhere, and timelines slip. Naming data preparation as its own line item, based on what the record sample showed, makes the budget more honest and the study more likely to deliver.
Ask your study partner
What proportion of the budget is set aside for data preparation, and what is that estimate based on?
Use this table in feasibility discussions and study planning meetings. A study does not need a perfect score to go ahead, but each red flag should have an owner and a plan before sites are activated.
The gap between what a protocol assumes and what a site's records hold is where much of our work sits. You do not need to have every answer before talking to us. There are two ways we can support your registry or real-world evidence study.
