The answer is rarely available in the pivotal study. Randomized controlled trials are designed to isolate treatment effects under controlled conditions. Routine healthcare is not. Patients arrive with comorbidities, adherence problems, fragmented records, changing treatment pathways and the occasional inconvenient refusal to behave like a study population. This is precisely why a medical affairs real-world data generation strategy matters — not as a fashionable supplement to the evidence plan, but as the mechanism for finding out what the intervention means once it meets ordinary care.
I have watched medical teams treat RWD as both a miracle ingredient and a vague data lake. Neither interpretation survives contact with a regulator, a health technology assessment body or a skeptical clinician. RWD is not evidence by itself. It is raw material. The scientific method applied to that material produces real-world evidence; without the method, there is only a larger spreadsheet and a more expensive slide deck.
The regulatory landscape is becoming structured — which is less exciting than saying it is being transformed
The old industry story was simple: randomized controlled trials establish truth, while real-world data provides colour around the edges. It was tidy, reassuring and increasingly inadequate.
The FDA’s Framework for Real-World Evidence Program, established in 2021, formalized a more practical approach to using routine healthcare data. The framework supports applications that extend well beyond post-marketing description, including disease natural history, trial participant selection, non-interventional studies and external control arms. That does not mean RWE has replaced randomized trials as the regulatory standard for initial efficacy approval. It means the regulatory conversation has become more granular.
The distinction matters. A medical affairs team that says RWD can answer every question is selling optimism. A team that says RWD can answer nothing that matters is defending an outdated operating model. The useful position sits between those two forms of institutional theatre.
The European landscape has moved in a similar direction. The European Medicines Agency’s DARWIN EU network provides a formal structure for coordinating analysis of healthcare data across participating sources. Its significance is not simply that more data may become available. The real shift is toward defined data quality expectations, methodological discipline and a more operational route for turning routine healthcare information into usable evidence.
For medical affairs leaders, this changes the planning question. It is no longer enough to ask whether a dataset exists. The more serious questions are:
- Is the data provenance clear enough to support the intended claim?
- Does the dataset capture the clinical variables needed to address confounding?
- Can treatment exposure, outcomes and follow-up be defined consistently?
- Is the population relevant to the regulatory, clinical or payer decision?
- Can the analysis be reproduced, challenged and explained without hiding behind the phrase “advanced analytics”?
The last question is not decorative. The more complex the model, the greater the temptation to make methodological opacity look like scientific sophistication. That tends to work until somebody asks what the missing data mean.
RWD is not a shortcut around evidence standards. It is a test of whether the organization actually understands them.
A robust medical affairs real-world data generation strategy therefore begins with the decision, not the database. The intended decision determines the evidence question; the evidence question determines the design; the design determines whether the data are fit for purpose. Reversing that sequence produces the familiar corporate ritual: acquire an impressive dataset, run several analyses, then search for a strategic question that can be attached to the output.
From clinical evidence to a real-world evidence strategy
Medical affairs teams often inherit an evidence plan organized around publication milestones, congresses and launch dates. Those are delivery events, not strategy. A real-world evidence programme should instead be connected to the uncertainties that remain after clinical development.
A pivotal trial may establish efficacy and safety in a selected population. It may not explain:
- how treatment is used across lines of therapy;
- how clinicians adapt dosing in routine practice;
- which patients discontinue and why;
- how outcomes differ in patients excluded from the trial;
- how the intervention compares with the treatments actually used in practice;
- what resource consequences emerge across the care pathway;
- whether guideline-directed therapy is being delivered consistently.
These are not secondary questions simply because they were not answered in the registrational programme. They are often the questions that determine whether a therapy changes care or merely occupies a place in the product catalogue.
The first practical step is to create an evidence gap map that links each uncertainty to its audience and decision point. A regulator may require a post-marketing safety analysis. A payer may need comparative effectiveness or resource-use evidence. A clinical society may need information on treatment sequencing. A medical information team may need to answer recurring questions about subgroups that were underrepresented in the trial. A field medical team may need a credible explanation of how trial findings translate into local practice.
These are related needs, but they are not interchangeable. Treating them as one universal RWE question is how evidence plans become broad, expensive and strangely unhelpful.
A more disciplined evidence architecture
A useful planning model separates four layers:
1. The clinical uncertainty
What remains unknown about disease progression, treatment response, safety, sequencing or patient experience?
2. The decision-maker
Who needs the answer — regulator, payer, clinician, guideline group, investigator or patient community?
3. The estimand and comparison
What treatment effect or association is being estimated, in which population, over what period and against which comparator?
4. The data and design
Which source and analytical approach can answer that question with acceptable validity?
This sounds obvious. In practice, organizations regularly begin at layer four because a vendor has already presented an attractive dataset. That is not evidence generation planning; it is procurement with a scientific accent.
A stronger RWE generation planning process for medical leads also distinguishes exploratory work from confirmatory work. Exploratory analyses can identify patterns, generate hypotheses and reveal where the data are weak. Confirmatory studies need a prespecified design, defensible definitions, appropriate controls and transparent handling of missingness and bias. The two activities can coexist, but they should not be dressed in the same language.
RWD integration in medical affairs: the operational problem nobody solves with a dashboard
Real-world data comes from electronic health records, claims databases, patient registries and digital health devices. Each source observes healthcare differently.
Claims data may provide broad visibility into utilization and reimbursement events, while offering limited clinical detail. EHRs can capture diagnoses, laboratory values, clinician notes and treatment decisions, but data quality may vary across sites and workflows. Registries can be clinically rich and disease-specific, yet they may represent a selected population and require sustained operational support. Digital devices can produce frequent measurements, although frequency is not the same as clinical relevance.
The source is not the study. It is one component of the study design.
This is where many RWD programmes become quietly unconvincing. Teams discuss “integration” as if linking sources automatically produces a complete patient journey. It does not. Linkage introduces its own questions: whether records can be matched reliably, which variables are harmonized, where duplicates exist, and how differences in coding practice affect the analysis. A dataset can be large, technically sophisticated and still fail to capture the variable that determines treatment choice.
The operational model should therefore be built around the evidence question rather than the prestige of the source. A practical comparison might look like this:
| Evidence need | EHR data | Claims data | Disease registry | Digital health data |
|---|---|---|---|---|
| Clinical characteristics | Often detailed, but dependent on documentation quality | Usually limited | Typically rich and disease-specific | Variable; may rely on patient-entered or device-generated information |
| Treatment exposure | Can capture prescribing and administration, with gaps between intention and delivery | Useful for reimbursed events | Often well characterized within the registry protocol | Usually indirect unless connected to clinical records |
| Healthcare utilization | Partial visibility | Strong for reimbursed services and encounters | Variable | Generally limited without linkage |
| Longitudinal follow-up | May be fragmented across providers | Broad where coverage remains continuous | Strong within participating sites or programmes | Potentially frequent, but attrition can be substantial |
| Best use case | Clinical outcomes, care patterns, subgroup analysis | Utilization, treatment pathways, economic outcomes | Disease-specific effectiveness and natural history | Symptoms, function, adherence proxies and patient-generated measures |
The point is not to crown one source as superior. There is no universal winner, only a better or worse match between the question and the observation system.
The field medical connection
RWD integration also changes the role of the medical science liaison. Field medical teams are often the first to hear where evidence fails in practice. They hear clinicians describe patients who do not fit trial eligibility criteria, treatment decisions shaped by local access, and outcomes that are not visible in standard publications.
That feedback should not be treated as anecdotal decoration. Nor should it be converted directly into evidence. It is a source of hypothesis generation and contextual interpretation. A mature medical affairs organization creates a controlled loop:
1. Field teams identify recurring evidence questions and care-pattern differences.
2. Medical affairs assesses whether the question is clinically meaningful and answerable.
3. Epidemiology and data science define the appropriate design and source.
4. The resulting evidence returns to the field in a form that supports scientific exchange rather than promotional repetition.
The loop breaks when every field observation becomes a request for a study, or when every study is designed without listening to the field. One produces research sprawl; the other produces elegant irrelevance.
Beyond randomized controlled trials — without pretending they are obsolete
The phrase “bridging clinical trials and real-world data” is useful, provided nobody turns it into a declaration of war against randomized evidence.
RCTs remain powerful because randomization reduces confounding and supports causal inference under the conditions of the study. Their limitations are equally familiar: selective eligibility, controlled follow-up, limited treatment duration, protocol-driven visits and a population that may not resemble routine care in all clinically relevant ways.
RWD addresses different dimensions of uncertainty. It can illuminate treatment use in broader populations, longer-term outcomes, care pathways and resource utilization. It may support external control arms in appropriate settings, but external controls are not a magical substitute for randomization. Differences in baseline characteristics, treatment assignment, calendar time, site practice and outcome measurement can produce a result that looks precise while answering the wrong question.
This is the uncomfortable part of RWE: statistical adjustment does not repair every design flaw. Propensity scores, weighting, matching and regression can help address measured differences. They cannot reliably neutralize unmeasured confounding simply because the model is elaborate. A forest plot remains a forest plot even when accompanied by a more expensive logo.
For each study, the team should be explicit about what the design can and cannot establish. That discipline protects the evidence programme from two common failures:
- Overclaiming: presenting an observational association as if it were equivalent to a randomized treatment effect.
- Underusing: dismissing clinically useful evidence because it does not imitate an RCT.
A sensible programme may use several designs across the product lifecycle:
- retrospective cohort studies to describe treatment pathways and outcomes;
- prospective observational studies where key variables need to be collected deliberately;
- registries for disease-specific follow-up and subgroup characterization;
- external control analyses when the clinical and data context supports them;
- non-interventional safety studies for post-marketing surveillance;
- linked datasets for healthcare utilization and outcome assessment.
The design should follow the decision. It should not be selected because it is currently fashionable at a congress.
HTA and reimbursement: the evidence story gets more demanding
Regulatory acceptance is not the same as reimbursement relevance. A dataset that supports a post-marketing safety question may be poorly suited to a payer’s comparative value assessment. Payers generally care about outcomes in the population they cover, under treatment patterns that resemble their system, with consequences for resource use and budget that can be understood in context.
This is where real-world evidence for post-launch value becomes more than a communications phrase. After launch, the evidence task often expands from “does it work?” to “for whom, compared with what, under which conditions, and at what cost to the health system?”
The UK’s National Institute for Health and Care Excellence has been highlighted as one of the health technology assessment bodies publishing dedicated guidance on evaluating RWE for reimbursement decision-making. That development reflects a broader direction of travel: HTA organizations are becoming more explicit about data quality, relevance, bias and applicability rather than accepting the existence of RWD as a badge of modernity.
Not every HTA body has adopted one unified, legally binding approach. There is no universal European RWE rulebook waiting patiently in a drawer. Requirements remain jurisdiction-specific, and the evidentiary burden depends on the decision, the disease area, the comparator landscape and the maturity of the available data.
For medical affairs, the implication is straightforward: payer evidence should not be bolted onto the end of the programme after launch. If the product value story depends on reduced hospitalization, improved adherence, treatment persistence or outcomes in a broader population, those endpoints need to be considered early. Otherwise, the organization may discover that it has excellent evidence for a clinical endpoint and very little evidence for the decision anyone needs to make.
The difference between an evidence story and a slogan
An evidence story has a chain of reasoning:
- the unmet need is defined in clinical terms;
- the target population reflects actual treatment decisions;
- the comparator is relevant to routine practice;
- outcomes matter to patients and decision-makers;
- the analytical method matches the causal question;
- limitations are visible rather than buried in an appendix;
- the findings can be translated into clinical and economic context.
A slogan has “value,” “impact” and “patient-centricity” in the title.
I have sat through enough strategic alignment sessions to know the difference. Alignment is useful when it connects medical, commercial, market access and clinical development teams around a legitimate evidence question. It becomes an echo chamber when every function repeats the same product narrative and nobody is allowed to ask whether the data can support it.
Building the evidence story across the product lifecycle
The strongest medical affairs RWD programmes are not assembled as a series of disconnected publications. They are designed as an evolving evidence system.
Early in development, routine healthcare data can help characterize disease natural history, identify treatment patterns and understand the population that may ultimately receive the intervention. It may also inform trial feasibility and participant selection. During launch preparation, RWD can clarify baseline care, unmet need and relevant comparators. After launch, the focus may shift toward effectiveness, safety, persistence, treatment sequencing, subgroups and healthcare utilization.
This lifecycle perspective is consistent with the direction represented by Clinical Evidence 2030 and Medical Affairs 2030 visions: medical affairs is being asked to generate evidence that supports better care, not merely more content. An example from AstraZeneca Spain described a structured, stakeholder-driven evidence generation framework aligned with those long-term visions, with the aim of improving guideline-directed medical therapy and building strategic product value stories. The important lesson is less about the brand than the operating principle: evidence planning should connect stakeholder needs, care delivery and measurable clinical outcomes.
That connection requires governance. A credible programme should define:
- who owns the evidence question;
- who approves the protocol and analytical plan;
- how scientific independence is protected;
- how patient privacy and data access are managed;
- how publication decisions are made;
- how field insights are incorporated without contaminating study design;
- how findings are communicated when they do not flatter the product.
The final point is where culture becomes methodology. If only positive findings are welcomed, no amount of statistical sophistication will create trustworthy medical affairs.
A working sequence for medical affairs leaders
When I build or review an RWD strategy, I prefer a sequence that is deliberately less glamorous than the usual transformation vocabulary:
1. Start with the unresolved decision.
Name the clinical, regulatory, payer or operational decision that lacks adequate evidence.
2. Define the population in real care.
Avoid relying solely on trial eligibility criteria when the intended question concerns routine practice.
3. Specify the comparison and outcome before selecting the source.
A database should not dictate the estimand simply because it is available.
4. Map data limitations early.
Identify missing variables, inconsistent coding, incomplete follow-up and changes in treatment practice.
5. Use the least complicated design that can answer the question credibly.
Complexity is not a quality marker. Sometimes it is just camouflage.
6. Separate signal generation from claim generation.
Exploratory patterns can guide further research; they should not automatically become statements in a value dossier.
7. Plan dissemination for scientific use.
A publication, congress abstract, medical information response and field tool have different purposes. Copy-pasting the same conclusion into all four is not a communications strategy.
8. Review the evidence programme as care changes.
New comparators, guidelines, diagnostics and access conditions can make yesterday’s question strategically irrelevant.
This sequence is not a substitute for epidemiological expertise. It is a way to prevent the strategy from being swallowed by procurement, platform demonstrations and internal optics.
The real gap is often organizational
Medical affairs teams do not usually suffer from a complete absence of data. They suffer from fragmented ownership of the questions.
Clinical development owns one part of the evidence lifecycle. Market access owns another. Commercial teams may hold important insights into treatment patterns, while field medical teams understand local clinical friction. Medical information sees recurring questions. Pharmacovigilance sees safety signals. Data science sees the architecture. Nobody sees the full evidence pathway unless the organization deliberately builds that view.
This is why RWD strategy cannot be reduced to hiring a vendor or licensing a dataset. The hard work is alignment — genuine alignment, not the conference-room version in which everyone agrees to a common slide template.
A functioning model gives medical affairs the authority to connect evidence generation with clinical practice while preserving scientific boundaries. It also recognizes that not every useful question needs a large study. Some require better data definitions. Some require a targeted chart review. Some require prospective collection. Others should be answered by acknowledging that the evidence is not available yet.
That final answer is often the most professional one. It is also the least popular in organizations trained to convert uncertainty into a launch asset.
The mature evidence strategy does not eliminate uncertainty; it identifies which uncertainty is worth reducing, and refuses to decorate the rest.
Conclusion: stop treating real-world evidence as the after-party
A medical affairs real-world data generation strategy should not be an appendix to the clinical programme, a post-launch rescue mission or a polished response to the latest payer question. It should connect the evidence created in trials with the care delivered after the protocol ends.
The FDA framework and the DARWIN EU network signal a more structured environment for using RWD. HTA bodies are becoming more explicit about how they assess relevance and quality. The technical options are expanding. None of this removes the central responsibility of medical affairs: defining a question that matters, selecting data that can answer it, and communicating the result without confusing confidence with proof.
The practical takeaway is not to collect more data. It is to stop collecting data without a decision attached.
Start with the gap. Define the population. State the comparison. Admit what the dataset cannot see. Then build the study — and let the evidence, rather than the launch narrative, decide what the organization is entitled to say.