Global life sciences AI · Published 16 November 2026 · HERMS Consultancy DMCC
Explore the HERMS approach and our solutions portfolio.
Real-world evidence can support regulatory and market access decisions when the research question, data, design and analysis fit the decision being made. Start with the uncertainty that matters, select data with adequate relevance and reliability, address bias and confounding, and preserve a traceable path from source records to conclusions. Large datasets alone do not establish treatment effects or payer value. Credible studies make their assumptions, limitations and remaining uncertainty clear enough for independent assessment. Real-World Evidence Considerations for the Use of Real-World Data and Real-World Evidence To Support Regulatory Decision-Making for Drug and Biological Products NICE real-world evidence framework
What is real-world evidence—and how is it different from real-world data?
Real-world data, or RWD, describe health status or healthcare delivery captured in routine sources such as electronic health records, claims and registries. Real-world evidence, or RWE, is clinical evidence derived from analyzing those data. A database is therefore an input, not a study conclusion. RWE can arise from more than retrospective observational research; randomized studies can also use routinely collected data. Real-World Evidence
The distinction matters commercially and scientifically. A dataset may describe treatment patterns well but lack disease severity needed for comparative effectiveness. Claims may measure billed utilization more directly than symptoms or quality of life. Study suitability is specific to the question, population and intended use—not a permanent badge attached to a data vendor. NICE real-world evidence framework
Start with the decision, not the database
Write a decision-first evidence brief before negotiating access or running analyses. Identify who will use the evidence, what decision is pending, which uncertainty could change that decision, and what claim the study could legitimately support. Separate a descriptive question from a causal claim.
| Decision context | Example question | Relevant evidence | Key limitation to resolve |
|---|---|---|---|
| Regulatory effectiveness | Does initiating A rather than B improve a defined outcome? | Comparative design with aligned cohorts and follow-up | Confounding, endpoint validity and applicable evidentiary standards |
| Postmarketing safety | Is exposure associated with a specified adverse outcome? | Appropriate pharmacoepidemiologic study | Exposure timing, detection and confounding |
| HTA comparative value | How does the technology perform against relevant local care? | Comparative effects plus applicability assessment | Comparator and population relevance |
| Economic evaluation | What resource use, utilities or long-term outcomes inform the model? | Context-specific estimates with uncertainty | Transferability and extrapolation |
| Budget impact | How many eligible patients may receive treatment and at what cost? | Population, uptake and utilization inputs | Local pathways, prices and assumptions |
| Evidence development | Which uncertainty should future collection reduce? | Feasibility assessment and prospective evidence plan | Follow-up, governance and measurable outcomes |
Regulatory approval and reimbursement are different decisions. Evidence relevant to safety or effectiveness may not answer a payer’s question about the local comparator, resource use or affordability. NICE’s framework supports using RWD to characterize care and inform economic models, but it does not set universal minimum acceptance standards. Considerations for the Use of Real-World Data and Real-World Evidence To Support Regulatory Decision-Making for Drug and Biological Products NICE real-world evidence framework
Build fit-for-purpose data around relevance and reliability
Assess relevance: are the population, exposures, outcomes, confounders, care setting and calendar period appropriate? Then assess reliability: how were records captured, curated, linked and checked? Document accuracy, completeness and traceability in relation to the variables that drive inference. A single overall completeness score can conceal a missing essential confounder. Considerations for the Use of Real-World Data and Real-World Evidence To Support Regulatory Decision-Making for Drug and Biological Products M14 General Principles on Planning, Designing, Analyzing, and Reporting of Non-interventional Studies That Utilize Real-World Data for Safety Assessment of Medicines NICE real-world evidence framework
| Data source | Potential contribution | Question to resolve before use |
|---|---|---|
| Electronic health records | Clinical measurements, diagnoses and care context | Are severity, timing and out-of-network outcomes captured? |
| Administrative claims | Utilization, dispensing and reimbursed services | Do codes represent the clinical concepts and payer perspective? |
| Disease or product registries | Disease-specific detail and longitudinal outcomes | How were participants selected and followed? |
| Mortality or other linked records | Additional outcome ascertainment | What linkage error, delay and permissions apply? |
| Patient-reported information | Symptoms, function and quality of life | Are measures appropriate and missing responses informative? |
| Digital health sources | Repeated measurements outside visits | Are devices, algorithms and missingness understood? |
Create a variable-level feasibility table before committing to a full study. Include coding algorithms, observation windows, validation evidence, missingness, linkage success and known changes over time. If a critical endpoint is absent, consider another source or additional collection; sophistication in analysis cannot manufacture the missing measurement. NICE real-world evidence framework Methods for real-world studies of comparative effects
Design the study before examining comparative outcomes
For causal questions, specify the hypothetical trial the observational study seeks to emulate. Define eligibility, treatment strategies, assignment, follow-up, outcomes, causal effect of interest and analysis. This target trial approach makes design choices explicit and exposes gaps between the ideal question and the available data. It does not turn observational treatment assignment into randomization. Methods for real-world studies of comparative effects
Where suitable, an active-comparator, new-user cohort can improve comparability. Compare patients initiating clinically relevant alternatives at the same treatment stage rather than mixing incident and long-standing users. Align eligibility, treatment initiation and the start of follow-up to avoid giving one group an artificial period in which the outcome could not have occurred. Methods for real-world studies of comparative effects
Define the estimand: which population, treatment contrast, outcome, time horizon and handling of events such as switching or discontinuation are intended? A treatment-initiation effect and an on-treatment effect answer different questions. Censoring people when they switch may create bias if switching relates to prognosis; state the assumptions and methods rather than treating censoring as administrative housekeeping. Methods for real-world studies of comparative effects
Control confounding without claiming it has disappeared
Confounding arises when factors influence both treatment selection and the outcome. Identify plausible confounders from clinical knowledge, literature and explicit causal reasoning. Avoid selecting adjustment variables solely because they are statistically significant. A directed acyclic graph can help distinguish confounders from mediators and colliders. Methods for real-world studies of comparative effects
Choose methods aligned to the estimand and data: restriction, matching, stratification, regression or propensity-score approaches may be appropriate. Assess covariate balance and treatment overlap, and report the consequences of exclusions or extreme weights. A propensity score balances measured information under assumptions; it cannot guarantee balance in unmeasured severity or clinician judgment. Methods for real-world studies of comparative effects
Time-varying confounding may require methods beyond standard baseline adjustment. Obtain specialist statistical input where previous treatment affects a confounder that also influences subsequent treatment and outcomes. Do not choose a complex method merely because it produces a more favorable result. Methods for real-world studies of comparative effects
Address missing data, bias and endpoint validity
| Threat | How it can undermine inference | Proposed response |
|---|---|---|
| Confounding by indication | Treatment reflects disease severity or prognosis | Clinically appropriate comparator; measured severity; sensitivity analysis |
| Immortal time bias | Exposure definition requires surviving a period | Align time zero and treatment strategies |
| Outcome misclassification | Codes do not represent the intended event | Validate algorithms; assess differential measurement |
| Informative censoring | Loss of follow-up relates to prognosis | Define reasons; use suitable methods and sensitivity analysis |
| Missing data | Recording depends on care or health status | Describe patterns and mechanisms; justify handling assumptions |
| Calendar/site differences | Care changes across time or institutions | Assess comparability; restrict, adjust or stratify as appropriate |
| Poor overlap | Groups lack comparable patients | Report support, exclusions and changed target population |
| Selective analysis/reporting | Methods follow favorable findings | Prespecified protocol/SAP; label deviations and exploratory work |
Report missingness by treatment group and over time where relevant. Multiple imputation is not a universal repair: its validity depends on assumptions and the model. Consider sensitivity or quantitative bias analyses when missingness, unmeasured confounding or misclassification could materially alter interpretation. Methods for real-world studies of comparative effects
Present effect estimates with uncertainty and clinically meaningful absolute quantities where appropriate, not only p-values. If sensitivity analyses yield different conclusions, explain which assumptions drive the difference. Statistical precision does not establish freedom from systematic bias. NICE real-world evidence framework Methods for real-world studies of comparative effects
Use external controls cautiously, especially in rare disease and advanced therapies
External controls can help address evidence gaps when suitable randomized comparisons are unavailable, but similarity of diagnosis alone is insufficient. Assess eligibility, disease stage, prior treatment, index date, supportive care, endpoint definitions, assessment frequency and calendar period. Differences between trial participants and routine-care patients can generate apparent effects unrelated to the technology. Methods for real-world studies of comparative effects
For cell and gene therapies, sparse populations and long-term uncertainty make careful natural-history and follow-up planning important. State where data cannot establish a credible comparison. A single-arm study plus a large external dataset does not automatically demonstrate a causal benefit; feasibility may show that additional collection is necessary. NICE real-world evidence framework Methods for real-world studies of comparative effects
Translate RWE into market access evidence—not unsupported value claims
Connect each proposed payer claim to a specified analysis and decision context. Distinguish clinical effectiveness, healthcare resource utilization, costs, health-state utilities and budget impact. Lower observed spending may reflect case mix, shorter follow-up or incomplete capture rather than a treatment effect. NICE real-world evidence framework Methods for real-world studies of comparative effects
For economic models, document the role of every RWE input: baseline risk, event rates, treatment duration, utilization or long-term outcomes. Explain how it relates to trial evidence, how uncertainty propagates and why it applies to the target health system. Budget-impact scenarios should expose eligible population and uptake assumptions rather than presenting a single estimate as inevitable. These are proposed implementation practices grounded in transparent evidence generation. NICE real-world evidence framework
Transferability deserves its own assessment. A study from one country may inform another but not directly supply local prices, care pathways or baseline event rates. Separate internal validity from applicability; reweighting cannot correct every unmeasured difference between settings. Methods for real-world studies of comparative effects
Preserve a traceable analysis from source to claim
An independent reviewer should be able to understand how the source became the analytical dataset and how the dataset became the result. Retain source descriptions and extraction dates; data dictionaries and code lists; linkage and curation rules; cohort attrition; protocol and statistical analysis plan versions; executable code and environment details; quality checks; deviations; and output-to-claim mappings. NICE real-world evidence framework Methods for real-world studies of comparative effects
Plan governance for lawful access, ethics review where applicable, privacy, permitted linkage, retention and secure analysis. Transparency does not mean publishing identifiable patient records. Where raw data cannot be shared, explain restrictions and provide enough methods, metadata and executable specifications for meaningful scrutiny. NICE real-world evidence framework
AI may assist literature triage, coding proposals or documentation, but generated citations, mappings and extracted fields require verification proportionate to their use. Keep study-design authority, access permissions and scientific sign-off under accountable human control. This is a recommended operating safeguard, not a claim of a universally mandated AI review procedure.
A hypothetical study: one dataset, two decision questions
Suppose a company wants to compare hospitalization after initiation of Therapy A versus Therapy B and also inform an HTA submission. A claims source captures dispensing and admissions, but disease severity is incompletely recorded. The team first specifies treatment stage, new-user eligibility, baseline observation, time zero, endpoint and follow-up.
Feasibility shows that a linked clinical source may improve severity measurement. The protocol prespecifies confounding adjustment, overlap assessment, missing-data handling and sensitivity analysis. If adequate comparability cannot be achieved, the team narrows the claim or changes the evidence plan rather than advertising a causal advantage.
For the payer question, it separately assesses local comparator relevance, unit costs, eligible population and model uncertainty. The regulatory analysis and economic assessment share documented evidence, but neither is relabelled as proof of the other. This scenario is illustrative; it reports no treatment result or HERMS deployment.
Regulatory and HTA landscape: what the sources actually say
| Source | Status / context | Practical implication |
|---|---|---|
| FDA RWD/RWE considerations | August 2023 final nonbinding drug/biologic guidance | Assess regulatory purpose, study feasibility and suitable data |
| FDA non-interventional studies | March 2024 draft; not for implementation | Do not relabel proposed recommendations as final requirements |
| ICH M14 at FDA | March 2026 final guidance; safety focus | Plan, design, analyze and report fit-for-use non-interventional studies |
| EMA RWE / DARWIN EU | EU evidence-generation infrastructure and explanatory resources | Data discovery and studies support regulatory questions; no automatic study acceptance |
| NICE RWE framework | Living methods framework; published June 2022 | Use relevant data and transparent methods; acceptability remains contextual |
FDA’s March 2026 M14 guidance replaces its July 2024 M14 draft; this is distinct from the separate March 2024 FDA non-interventional-study draft. M14 focuses on safety, while FDA notes that its recommendations can also inform effectiveness studies. International harmonization does not mean identical national implementation or guaranteed approval. Considerations for the Use of Real-World Data and Real-World Evidence To Support Regulatory Decision-Making for Drug and Biological Products Real-World Evidence: Considerations Regarding Non-Interventional Studies for Drug and Biological Products M14 General Principles on Planning, Designing, Analyzing, and Reporting of Non-interventional Studies That Utilize Real-World Data for Safety Assessment of Medicines
EMA’s HMA-EMA catalogues help identify data sources and studies. DARWIN EU generates evidence addressing medicines-regulation questions through participating data partners. Catalogue inclusion is not proof that a dataset fits a particular protocol or that a sponsor’s conclusion is accepted. Real-world evidence Data Analysis and Real World Interrogation Network (DARWIN EU)
A practical evidence-readiness checklist
- Decision and claim: identify the user, uncertainty and permissible conclusion.
- Design: define the population, comparator, time zero, estimand and follow-up.
- Data: confirm essential variables, provenance, permissions and feasibility.
- Bias: document confounders, overlap, missingness and measurement limitations.
- Analysis: prespecify primary methods, sensitivity analyses and interpretation rules.
- Traceability: retain code, versions, attrition, deviations and quality checks.
- Applicability: assess target-country care pathways and decision relevance.
- Engagement: seek suitable regulatory or HTA advice before locking consequential design choices.
Use these items as proposed project gates, not a claim that every authority requires the identical checklist. NICE encourages early advice for companies planning to use RWD in submissions. NICE real-world evidence framework
How HERMS connects evidence generation with decision readiness
HERMS describes VALTRIX as a clinical and value-evidence solution area, VALTIRA as pricing, HTA and budget-impact intelligence, and VERIQIS as evidence integrity and assurance. These are company positioning statements, not independently validated performance claims. The useful evaluation question is whether a proposed deployment can demonstrate source-to-claim traceability, appropriate methods and accountable review for its intended use. AI Life Sciences Software | 25 HERMS Solutions
Explore HERMS clinical, value-evidence and market-access solutions to discuss a demonstration around your evidence gap, study-readiness needs and decision requirements.
Frequently asked questions
What is the difference between RWD and RWE?
RWD are the underlying routine-care data. RWE is the clinical evidence derived from analysis. Collecting records or purchasing a dataset does not itself create credible evidence; the research question, study design, analytical assumptions and data suitability determine what can be concluded. Real-World Evidence
Can RWE support regulatory approval?
It can support specified regulatory uses when the evidence satisfies applicable standards and addresses the question. Acceptance is not automatic and depends on the submission context. Engage the relevant authority on consequential design choices; do not assume RWE replaces randomized evidence in every setting. Real-World Evidence Considerations for the Use of Real-World Data and Real-World Evidence To Support Regulatory Decision-Making for Drug and Biological Products
How does RWE support market access and HTA?
It can inform disease burden, treatment pathways, comparative evidence, economic model inputs and applicability to routine care. A payer still needs relevant comparators, credible methods and uncertainty assessment. Regulatory acceptance of a study does not guarantee reimbursement or establish affordability. NICE real-world evidence framework Methods for real-world studies of comparative effects
What makes data fit for purpose?
The data must contain appropriate patients, treatments, outcomes and covariates, with adequate measurement, follow-up and traceability for the specific question. Evaluate both relevance and reliability. A source suitable for utilization research may be unsuitable for a causal effectiveness comparison. M14 General Principles on Planning, Designing, Analyzing, and Reporting of Non-interventional Studies That Utilize Real-World Data for Safety Assessment of Medicines NICE real-world evidence framework
Does propensity-score matching remove confounding?
It can address measured confounding under assumptions, provided suitable covariates and overlap exist. It does not guarantee control of unmeasured factors or correct poor endpoint definitions. Report balance, exclusions, residual limitations and sensitivity analyses instead of treating matching as proof of causality. Methods for real-world studies of comparative effects
What is target trial emulation?
It is the explicit design of an observational study around the randomized trial one would ideally conduct. Specify eligibility, strategies, assignment, follow-up, outcomes, causal effect and analysis. The approach helps avoid design errors but cannot supply missing variables or make observational treatment allocation random. Methods for real-world studies of comparative effects
When are external control arms credible?
Credibility depends on comparability of populations, care, time periods, data capture and outcomes, plus an appropriate design and analysis. Assess what cannot be aligned and whether remaining uncertainty undermines the claim. An external control is not automatically suitable because the disease is rare. Methods for real-world studies of comparative effects
Can AI generate a decision-ready RWE study?
AI may support selected tasks, but decision readiness requires suitable data, sound causal reasoning, reproducible analysis and qualified review. Verify generated references, extraction and code. No architecture or vendor label independently establishes scientific validity, regulator acceptance or payer relevance.
A useful RWE study does not begin with the largest dataset or end with the most favorable estimate. It begins with a decision and ends with a transparent account of what the evidence supports, what it does not, and which uncertainty remains.
Discuss your next evidence-generation or market-access question through HERMS Solutions , or explore related HERMS life-sciences insights .
Relevant HERMS solutions
Explore HERMS solutions relevant to the workflows discussed in this article: VALTRIX — clinical and value evidence; VALTIRA — pricing, HTA and budget impact; EXIRIX — clinical experiment design; VERIQIS — evidence integrity and assurance. Evaluate each solution against your intended use, evidence and approval requirements.
Sources and further reading
- Real-World Evidence
- Considerations for the Use of Real-World Data and Real-World Evidence To Support Regulatory Decision-Making for Drug and Biological Products
- NICE real-world evidence framework
- M14 General Principles on Planning, Designing, Analyzing, and Reporting of Non-interventional Studies That Utilize Real-World Data for Safety Assessment of Medicines
- Methods for real-world studies of comparative effects
- Real-World Evidence: Considerations Regarding Non-Interventional Studies for Drug and Biological Products
- Real-world evidence
- Data Analysis and Real World Interrogation Network (DARWIN EU)
Explore the pharmaceutical AI series
- AI decision authority and human oversight
- Agentic versus generative AI in pharmaceuticals
- AI governance in life sciences
- Human approval controls for pharmaceutical AI
- AI medical device validation and change control
- Global pharmaceutical market access, HTA and pricing evidence
Explore HERMS life sciences solutions and AI governance and traceability, or request a workflow-focused demonstration.
