Executive summary

Real-world health data (RWHD) comprises information generated outside conventional, protocol-driven trials: electronic health records, claims, prescribing, registries, laboratory and imaging systems, genomics, mortality records, patient-reported outcomes and data from connected devices. Real-world evidence (RWE) is the clinical or economic inference produced from those data. This distinction matters: a large dataset is not evidence, and evidence is not automatically decision-grade.

The literature supports a broad and increasingly institutionalised role for RWE. Regulators now publish detailed frameworks; the FDA uses distributed data for active safety surveillance, the European Medicines Agency is scaling DARWIN EU, NICE maintains a living RWE framework, and the European Health Data Space (EHDS) creates a future legal architecture for secondary use.[1–7] Yet evidence of financial return is less mature than evidence of utility. Most published value claims describe mechanisms—lower-cost data collection, faster feasibility, better recruitment, external controls, safety surveillance and market-access support—rather than audited return on investment.[8–12]

Six conclusions

• The highest-value asset is not the biggest database; it is the smallest reliable dataset that can resolve an expensive uncertainty.

• RWE usually complements rather than replaces randomised trials. Its comparative advantage is scale, longitudinal follow-up, representativeness and observation of routine care; its weakness is causal ambiguity.

• Financial value is mainly risk reduction, time advantage and option value: avoiding a failed study, accelerating a launch, preserving reimbursement or identifying the right population.

• Patient and commercial value often align—especially in faster recruitment, rare-disease natural history, safety detection and treatment targeting—but can diverge when monetisation encourages exclusion, opacity or extractive data practices.

• Accessibility is a system property, not a permission checkbox. Discoverability, metadata, linkage, data quality, contracting, secure analysis and the ability to publish all determine whether nominally available data are usable.

• Trust is productive infrastructure. Transparent governance, patient involvement, credible public benefit and enforceable controls increase the durable supply of usable data.

Where value concentrates

1. What counts as real-world health data?

Regulators define real-world data as routinely collected information relating to patient health status or the delivery of healthcare; RWE is the clinical evidence derived from analysis of those data.[1] The analytical question—not the source label—determines fitness for purpose.

A hierarchy of usefulness

The same dataset can be adequate for counting and inadequate for causal inference. A useful maturity ladder is:

• Descriptive: what happened and to whom?

• Predictive: who is likely to experience an outcome?

• Comparative: what would have happened under a different intervention?

• Operational: can the result trigger a workflow at the right time?

• Decision-grade: is the provenance, method and governance sufficient for a regulator, HTA body, payer or clinician?

The value equation

A practical strategic model—not an accounting identity—is:

This model explains why exclusive raw data can be less valuable than non-exclusive, well-curated data embedded in a repeatable decision workflow. It also prevents teams from treating record count as a proxy for value.

2. What the evidence genuinely supports

Strongly supported

• Safety and performance after launch. Large, longitudinal networks can detect uncommon events and study heterogeneous populations that trials are rarely powered to represent. FDA Sentinel already supports routine post-market drug-safety analyses.[1,13]

• Natural history and external context. Rare diseases, oncology and devices benefit from longitudinal untreated or standard-of-care cohorts, although external controls remain vulnerable to endpoint and population mismatch.[9,14]

• Comparative effectiveness and health economics. Routine data can quantify treatment pathways, resource use, adherence and outcomes under ordinary practice—inputs central to HTA, formulary and pricing discussions.[3,10]

• Trial design and operations. Feasibility counts, eligibility simulation, endpoint selection, recruitment and pragmatic follow-up can make studies more executable and inclusive.[8,15]

• Lifecycle evidence. RWE can support label changes, post-authorisation commitments, device surveillance and evaluation of AI-enabled products whose performance may drift.[2,6,16]

Supported, but conditional

• Replacing a randomised control arm. Credibility depends on prespecification, exchangeability, endpoint alignment and sensitivity analysis. A review of seven oncology external-control cases found decisive impact in only a minority of regulatory decision contexts.[14]

• Causal treatment comparisons. Sophisticated methods cannot repair unmeasured confounding, poorly measured exposure or missing outcomes. Target-trial emulation clarifies the estimand but does not create absent data.

• Generalisability. Broader coverage can improve external validity, but routinely collected datasets may still underrepresent uninsured people, minorities, rural populations or people who disengage from care.

• Digital biomarkers and patient-generated data. They can capture outcomes continuously and closer to lived experience, but validation, adherence, device changes and equitable access remain material constraints.[16,17]

Weakly supported or commonly overstated

• A universal monetary value per record. No defensible cross-sector number exists; value is purpose-, rights- and context-specific.

• A general percentage reduction in trial cost. Published ranges are mostly scenario estimates, interviews or vendor claims. They should be treated as hypotheses for a business case, not benchmarks.[11]

• Automatic regulatory acceptance. Guidance creates a route for use, not a presumption of adequacy.

• Bigger is always better. Adding low-quality records can increase bias, cost and false confidence.

3. Opportunity by industry

Pharma

The strategic shift

The opportunity is not a separate “RWE department”. It is an evidence spine from target-product profile through reimbursement and pharmacovigilance. Reuse matters: the phenotype and data pipeline created for feasibility should, where methodologically appropriate, support recruitment, follow-up and post-launch learning.

Biotech

Biotech’s greatest constraint is often not data volume but capital and decision time. RWHD can improve patient identification, rare-disease natural history, biomarker prevalence, trial feasibility and early payer narratives. The most valuable use may be a negative one: stopping an implausible indication before a costly trial. Small companies should buy decision-specific access and analytical capability rather than accumulate undifferentiated data licences.

Diagnostics

Diagnostics create value when a test changes management, not merely when it predicts an endpoint. Linked laboratory, treatment and outcome data can establish clinical validity, reveal testing gaps, quantify the diagnostic odyssey, identify cut-offs and demonstrate downstream utility. The FDA and Medical Device Innovation Consortium have developed pathways for using RWE in diagnostics and in-vitro devices.[6,18]

Medical devices and digital health

Devices generate iterative, operator-dependent and increasingly software-mediated performance. Registries and connected-product data can support post-market surveillance, comparative performance, learning curves, durability, rare failures and algorithm drift.[6,16,19] The commercial prize is a lifecycle evidence system; the patient prize is earlier identification of deterioration and inequitable performance.

Health-data businesses

Sustainable businesses move up the value stack: from access, to linkage and curation, to validated phenotypes, to reproducible evidence, to workflows that change decisions. Defensibility comes less from hoarding rows than from lawful rights, provenance, longitudinal linkage, rare modalities, quality controls, transparent methods, trusted relationships and integration into customer decisions.

4. Accessibility: available is not usable

Accessibility has at least seven layers. Failure at any one can erase the apparent advantage of a dataset.

UK, EU and US: comparative view

Sources: EHDS, EMA, HDR UK, NHS England, OECD, FDA Sentinel and PCORnet.[4,5,7,13,20–25]

What is changing

The EHDS Regulation entered into force on 26 March 2025. Most secondary-use provisions for major data categories are scheduled to apply from March 2029, so it is a strategic platform under construction rather than a current pan-European data lake.[4] DARWIN EU already operates as a federated regulatory network: partners retain protected data and return standardised analyses.[5] In England, the Health Data Research Service is intended to provide a clearer, secure route to linked national data, but implementation quality and public legitimacy will determine whether it reduces rather than merely relabels friction.[21,22]

The US demonstrates a different model. FDA Sentinel analyses distributed claims and clinical data without centralising all source records, while PCORnet provides standardised access through clinical research networks.[13,24,25] This provides scale but does not solve insurance discontinuity, fragmented care or commercial opacity.

5. How financial value is created

Seven value mechanisms

A disciplined business case

• Name the decision, owner and deadline. “Generate insights” is not a value proposition.

• Write the counterfactual: what happens without the data? Include the current error rate and decision delay.

• Specify the estimand and minimum evidentiary standard. Regulatory-grade evidence costs more than exploration.

• Model value as a range and probability-weight it. Separate time savings, cost savings, revenue and risk.

• Include full friction costs: licence, linkage, curation, governance, secure compute, validation, expert review and refresh.

• Predefine a kill criterion. If the data cannot measure the critical exposure, outcome or confounders, stop.

• Measure realised value after the decision. Build an internal evidence base rather than recycling industry anecdotes.

Illustrative calculation

Suppose an evidence programme costing £1.2m has a 35% chance of advancing a portfolio decision by six months, and each month is worth £2m only if the product ultimately succeeds (25% probability). Expected time value is £1.05m (0.35 × 6 × £2m × 0.25), before considering avoided trial cost, failure risk or patient benefit. On time value alone the case is negative. This is precisely why probability, counterfactual and full cost belong in the model.

6. How patient value is created—and lost

Benefit pathways

• Earlier access: faster feasibility, recruitment and evidence generation can shorten the route to effective interventions.

• Safer care: broader surveillance detects rare harms, device failures and performance drift.

• More relevant evidence: routine-care populations reveal effects in older, multimorbid and underrepresented people.

• Better targeting: linked clinical and molecular data can identify who benefits, who is harmed and who is being missed.

• Less burden: routine outcomes and remote measures can reduce visits, duplicate data entry and control-arm exposure.

• More meaningful outcomes: patient-reported symptoms, function and quality of life can complement laboratory and utilisation endpoints.[17,26]

• Learning systems: evidence feeds back into pathways, guidelines and product design rather than ending at publication.

Where value can be destroyed

The social licence

Public acceptance is conditional, not absent. UK evidence suggests broad willingness to support research when safeguards, transparency, public involvement and benefit to the health service are credible.[23] Consent alone cannot carry the full ethical burden: patients also need understandable purpose limitation, independent oversight, security, avenues for challenge and evidence that benefits are shared.

7. Barriers and practical mitigations

Why common data models help—but do not solve everything

Common models such as OMOP and network-specific schemas reduce repeated mapping and make distributed analysis possible. They improve syntactic consistency; they do not guarantee semantic equivalence. A diagnosis recorded to justify billing, one entered by a specialist and one inferred from medication are not interchangeable merely because they share a code.

Federation versus centralisation

Federated analysis can keep identifiable data with custodians and distribute standardised queries, as in DARWIN EU and Sentinel.[5,13] This can improve control and political feasibility, but it shifts difficulty into common specifications, local execution, quality assurance and disclosure review. Centralised linkage may support richer analysis but increases concentration and breach risk. Architecture should follow the question and risk—not fashion.

8. Strategic opportunities

1. Evidence operating systems

Create reusable phenotypes, source mappings, quality tests, protocols and analytical components around priority decisions. The moat is a shorter path from question to defensible answer, not a catalogue of disconnected datasets.

2. Trial–care convergence

Link trial data to routine records for feasibility, recruitment, long-term outcomes and pragmatic follow-up. This can reduce duplicate collection and reveal real-world performance while preserving randomisation where causal certainty matters.

3. Regulatory-grade data products

Package lineage, quality metrics, validation, refresh cadence, representativeness, permissible uses and analysis environments as part of the product. Buyers should be able to answer “fit for which decision?” before contracting.

4. Rare-disease and precision cohorts

Federate specialist clinical, genomic and patient-reported data to create natural-history cohorts and trial-ready populations. Value depends on longitudinal depth and expert phenotyping more than raw scale.

5. Device and algorithm observatories

Continuously evaluate safety, subgroup performance, calibration, drift and workflow effects. This turns post-market surveillance from a compliance cost into a product-improvement loop.

6. Diagnostic utility platforms

Connect tests to downstream treatment and outcomes. The key asset is the ability to show whether diagnosis changed care and improved outcomes—not only analytical accuracy.

7. Privacy-enhancing collaboration

Trusted research environments, federated analytics, secure linkage and, selectively, privacy-enhancing technologies can enable collaboration without routine export of person-level data. They reduce some risks but do not replace governance or method.

8. Patient benefit contracts

Tie access to transparent benefit commitments: publication, clinically relevant outputs, capability transfer, service improvement, return of aggregate findings and fair terms. This strengthens legitimacy and may improve long-run data availability.

9. Practical playbooks

For pharma and biotech leaders

• Start with the three most capital-intensive uncertainties in the portfolio. Map data only after naming them.

• Create one cross-functional evidence roadmap spanning clinical, regulatory, safety, access and medical affairs.

• Seek early scientific advice when RWE may support a pivotal regulatory or HTA claim.

• Fund validation and sensitivity analysis as product work, not academic embellishment.

• Track time-to-decision, avoided work and evidence reuse—not the number of datasets acquired.

For diagnostics and device companies

• Design identifiers, versioning and post-market measurement into the product from the outset.

• Measure clinical utility and workflow effects, not only technical performance.

• Monitor performance by site, operator, demographic group and software/hardware version.

• Predefine drift thresholds and corrective actions.

• Use registries or coordinated networks where single-centre data cannot capture rare failures.

For health-data businesses

• Publish a decision-specific data sheet: population, provenance, linkage, missingness, refresh, rights and known failure modes.

• Sell reproducibility and service levels alongside access.

• Make customer code and derived outputs portable; compete on continuous value rather than captivity.

• Build patient and clinician governance into product design.

• Avoid promising “regulatory grade” without an explicit use case and documented validation.

For data custodians and health systems

• Treat metadata, linkage and reproducible access as core infrastructure.

• Standardise contracts, review criteria and turnaround reporting.

• Use tiered access and safe outputs rather than blanket refusal or uncontrolled release.

• Publish who used the data, for what purpose, with what findings and benefit.

• Retain bargaining power for patient and system benefit while avoiding prohibitive transaction costs.

For investors and boards

• Ask what decision the asset improves and what the counterfactual is.

• Discount revenue claims that rest on record count, exclusivity or generic AI narratives.

• Test rights, provenance, refresh dependence, customer concentration and regulatory exposure.

• Examine whether scientific publication and external validation are contractually possible.

• Treat trust failure and inaccessible data as balance-sheet risks even when accounting does not.

10. Due-diligence scorecard

A 90-day action agenda

Conclusion: from data accumulation to decision advantage

Real-world health data can create substantial value for life sciences, but it does not do so by existing. Value is produced through a chain: lawful access, reliable linkage, fit-for-purpose measurement, credible design, timely analysis, acceptance by a decision-maker and action. Every broken link converts a celebrated asset into cost.

The commercial opportunity is therefore larger—and harder—than data brokerage. It lies in building durable evidence capabilities that reduce uncertainty across the product lifecycle. Pharma and biotech can allocate capital and develop products more intelligently; diagnostics can demonstrate utility; device companies can operate continuous performance systems; data businesses can turn fragmented records into trustworthy, reproducible decision infrastructure.

Patient value is not a secondary dividend. It is the strongest long-run test of the system. Faster access, safer products, more representative evidence and better targeting justify the use of sensitive information. Conversely, opaque monetisation, biased datasets and weak accountability consume trust and eventually restrict access.

Research limitations

This independently written review synthesises regulatory materials, infrastructure descriptions, peer-reviewed reviews and selected industry perspectives. The literature is heterogeneous and financial outcomes are rarely reported with transparent counterfactuals. Consultancy and trade-body sources were used for strategic context, not treated as independent proof. The legal and infrastructure landscape is evolving; specific programmes and access routes should be verified at the point of use. All tables, organising frameworks, comparisons, recommendations and the value equation are original analyses produced for this report; they should not be attributed to the cited organisations.