A Practical Guide to Implementing Real‐World Evidence as Primary Basis for Approval and Label Expansion

Abstract

In June 2026, the FDA issued a revised draft guidance describing the circumstances in which one adequate and well‐controlled clinical investigation plus confirmatory evidence may satisfy the statutory substantial evidence standard for establishing efficacy and safety. Confirmatory evidence includes real‐world evidence (RWE). The evolution of the regulatory use of RWE builds on more than two decades of FDA experience with real‐world data (RWD), first for pharmacovigilance and later in effectiveness and risk/benefit determination. RWD/RWE guidances have been released since the 21st Century Cures Act was passed in 2016, but implementation remains operationally and methodologically challenging. This article provides practical guidance for teams considering RWE as part of a registrational program. Key topics include selection of fit‐for‐purpose data, study design, protocol development, study execution and governance, data transformation, analysis and assessment of residual bias, and submission. Recommendations emphasize early FDA alignment, filling gaps in procedural documentation and systems, multidisciplinary collaboration, and anticipating technological demands associated with large real‐world datasets.

On February 19, 2026, a New England Journal of Medicine Perspective by then‐FDA Commissioner Makary and CBER Director Prasad argued that one pivotal trial plus confirmatory evidence should become the default option for demonstrating effectiveness. ¹ While an important statement of regulatory direction, it was not itself an FDA guidance, rulemaking, or a change to the statutory substantial evidence standard.

In June 2026, the FDA issued the revised draft guidance: “Demonstrating Substantial Evidence of Effectiveness for Human Drug and Biological Products.” ² The draft guidance is nonbinding and currently undergoing public comment. This revision clarified the circumstances under which one adequate and well‐controlled clinical investigation plus confirmatory evidence may be sufficient for regulatory approval, while clearly stating that there is no single evidentiary formula. The strength of the pivotal trial, the source and strength of confirmatory evidence, the disease context, and the overall development program all matter. Importantly, a single‐trial effectiveness strategy should not be confused with a single‐study development program. Additional evidence may still be needed to characterize safety and support a favorable benefit–risk assessment. ²

The move toward greater use of a single adequate and well‐controlled trial is a matter of much debate. Advantages, such as efficient development, are being weighed against concerns such as truncated safety assessments in smaller pools of patients and the increased risk of Type I errors. The June 2026 draft guidance is a framework for evaluating the strength of the total evidence package, not a guarantee that one trial will be sufficient. ² Within that framework, RWE is one form of confirmatory evidence. The question for sponsors is not whether RWE is categorically acceptable. It is whether a specific data source, study design, and analysis can generate evidence sufficiently reliable and relevant to the regulatory question.

HISTORY OF REGULATORY USE OF REAL‐WORLD EVIDENCE

The FDA has long‐standing experience using real‐world evidence (RWE) to monitor drug safety. Since 1969, the FDA Adverse Event Reporting System (FAERS, now AEMS) has passively identified reported safety signals, which the FDA has used to rescind marketing authorization. ³ , ⁴ Based on the FDA Amendments Act of 2007, the FDA initiated the formalization of proactive pharmacovigilance with the Mini‐Sentinel Pilot, using data sources that included electronic health records (EHR) and insurance claims databases to search for potential safety signals. ⁵

The use of RWD as evidence of efficacy and its use in regulatory decision making is more recent. Lepirudin was approved in 1998 for anticoagulation in patients with heparin‐induced thrombocytopenia. ⁶ Registry data were used as historical controls for a single‐arm development program. Later examples included alglucosidase alfa for infantile‐onset Pompe disease, recombinant antithrombin for prevention of perioperative and peripartum thromboembolic events in hereditary antithrombin deficiency, and hepatitis B immune globulin for prevention of hepatitis B recurrence after liver transplantation. ⁶ These early cases involved serious or rare diseases in which randomized comparator data were limited or difficult to obtain. Beginning in the mid‐2010s, the FDA considered RWD more frequently, though still selectively, most often in oncology and rare disease. The purpose for which RWE was used varied across applications. In some instances, it was contextual or supportive; in others, it was part of the primary basis for approval. It is important when discussing RWE in a regulatory context to be specific about how it was used.

In 2016, Congress passed the 21st Century Cures Act, directing the FDA to evaluate the use of evidence from sources other than traditional clinical trials to support approval of new indications. ⁷ FDA released its initial RWE Framework in 2018 and subsequently issued a series of guidance documents addressing data relevance and reliability, data standards, regulatory considerations, externally controlled trials, noninterventional studies, and related topics. ⁸ , ⁹ , ¹⁰ , ¹¹ , ¹² , ¹³ , ¹⁴ , ¹⁵ , ¹⁶ These documents provide an increasingly detailed framework, but practical implementation challenges remain. Figure 1 descriptively summarizes FDA regulatory actions identified in published landscape reviews and FDA's public RWE examples.

Figure 1.

FDA regulatory actions in which real‐world evidence contributed to an approval or labeling decision. Cases were identified from the published landscape review by Alipour‐Haris et al. ⁷ and FDA's public compilation of RWE regulatory examples. ¹⁷ Each approval or label expansion was counted once in the calendar year of the regulatory action. Inclusion required that RWE contributed to the effectiveness or benefit–risk evidence considered in the decision; uses limited to postmarketing surveillance or unrelated contextual information were excluded. Because the contribution of RWE varies across applications and public reporting may be incomplete, the figure is descriptive.

Our objective is to build upon these guidances and regulatory precedent to provide practical advice for teams considering the use of RWE as part of a registrational evidence package. We discuss regulatory requirements, FDA recommendations, in conjunction with operational practices that reflect the authors' experience. As the guidances make clear, evidence packages are specific to the disease and product under investigation and not a cookie cutter formula that guarantees approval. The approaches we discuss here provide points of consideration and discussion with the FDA. This article does not address pragmatic randomized clinical trials in which data collection includes real‐world elements, such as passively collected outcomes from EHRs, as those designs are covered in separate FDA guidance.

STUDY PLANNING AND PREPARATION

The decision to use RWE in a development program depends on the history of that program, the disease, the investigational product, the feasibility of additional randomized trials, and the availability of a relevant and reliable data source. The June 2026 draft guidance outlines the conditions under which confirmatory evidence may be sufficient, including the strength of the pivotal trial results, the source and strength of confirmatory evidence, the seriousness of the disease and unmet need, and whether additional adequate and well‐controlled trials would be ethical and feasible. ² RWE should be considered when the proposed data source and study design can credibly answer the regulatory question as to efficacy and safety, either as supportive or as primary basis for approval as part of the total evidence package.

It is critical that, from the outset, the team include individuals with scientific expertise in RWE as well as the regulatory requirements of a registrational program. In many organizations, expertise and experience with RWE, clinical development, regulatory strategy, data standards, and trial operations reside in different individuals or functions. Over time, more epidemiologists and biostatisticians have begun to work across both RWE and trial environments, thus it is important to understand expertise and experience as opposed to fixed roles. The need is to assemble a multidisciplinary team with the requisite expertise in disease biology, causal inference, biostatistics, data provenance and curation, regulatory requirements, clinical operations, and submission standards.

STUDY DESIGN

The first question is what study design could provide sufficiently rigorous evidence for the proposed regulatory package? Broadly, these designs include randomized trials in which an internal control arm is supplemented with external RWD (a hybrid control); a single‐arm interventional study compared with an external control derived from RWD; a noninterventional comparative study or target‐trial emulation; and a natural history study from which efficacy and safety comparisons can be made. These approaches differ substantially in susceptibility to bias, feasibility, time, and cost, recognizing that even a natural history study, if data are collected prospectively, can be lengthy and costly. FDA's June 2026 draft guidance emphasizes strength of evidence rather than a preferred design hierarchy. ²

To produce evidence capable of supporting regulatory approval or label expansion, the design should structurally align with the RCT one would ideally conduct, a framework now commonly referred to as “Target Trial Emulation.” ¹⁸ The selection of the target population should follow standard clinical trial conventions, including patients with the disease and clinical characteristics consistent with the proposed indication, excluding patients with comorbidities that would significantly confound results. Once the target population is defined, a central challenge in this emulation is the precise identification of “Time Zero,” the moment when treatment decisions are made and clinical eligibility is met. ¹⁹ Establishing this “real‐world equivalent” of a randomization date is essential for ensuring that baseline variables are correctly characterized and that follow‐up for endpoints begins at the appropriate moment.

For studies comparing two active treatments, identifying Time Zero is relatively straightforward – this would be the initial receipt or fulfillment of the treatment. This approach is commonly referred to as a comparative new user design and has been a standard approach in pharmacoepidemiology for over two decades and has been used in many well‐known applications, such as the RCT DUPLICATE project. ²⁰ , ²¹ However, when a new treatment is being compared with no treatment or a background standard of care, the selection of an index date is significantly more complex because patients often meet inclusion and exclusion criteria at multiple points in their longitudinal history. Careful statistical handling for these multiple indexes is necessary to avoid bias in point estimates and confidence intervals. ²² Commonly, researchers select a random index from among all healthcare visits on which a patient meets eligibility criteria, using the first and last eligible dates in sensitivity analyses. ²³

DATA STRATEGY

One advantage of explicitly emulating a target trial is that the necessary variables and their time points for observation are predefined. Baseline variables should include important confounders, variables that could be related to either treatment assignment or outcomes. This requires subject matter experts and the development of a prespecified causal framework rather than data‐driven significance testing. Potential confounders include demographics, measures of disease severity, treatment history, comorbidities, concomitant medications, healthcare utilization, and calendar time.

Regardless of study design, the first and largest hurdle is identifying and acquiring a dataset that meets FDA expectations for fitness‐for‐use, meaning that the data can inform regulatory decision making by meeting standards for relevance and reliability, as well as fitness‐for‐purpose, with appropriate transparency and provenance to sufficiently address the question of interest. Primary sources include health insurance claims databases, electronic health records, and disease registries. Often these data are linked via tokenization or probabilistic approaches, which can present practical challenges including de‐identification under HIPAA's Safe Harbor or Expert Determination method, especially for rare disease. Sponsors should budget time and cost for this process. ²⁴

A first step is the reliable identification of eligible patients. A single diagnosis code is rarely sufficient, as it can reflect a rule‐out diagnosis, a coding error, historical conditions, billing practices, or conditions whose definition requires information not represented by the code itself. It is common practice to create algorithms that combine diagnosis with things like procedure codes, medications, laboratory values, pathology, imaging, and/or biomarkers. These algorithms should be validated against an appropriate reference standard ⁹ , ¹¹ , ¹⁴ and may involve medical record review or comparison with source documentation. Validation approaches include positive predictive value as well as assessments of sensitivity and specificity. When an endpoint would be centrally adjudicated in the corresponding trial, source data are often required to allow for blinded adjudication; for example, when RANO criteria are applied to assess progression‐free survival in solid tumors, source MRIs may be required. The level of source verification should be justified for the specific variable and regulatory use.

Once a candidate data source has been identified, a fit‐for‐purpose/feasibility assessment is conducted. Such assessments are generally iterative. An initial step determines whether the source contains the population, exposures, outcomes, adequate follow‐up, and key covariates needed to address the question. A second step entails a detailed assessment regarding completeness, timing, provenance, linkage quality, and potential misclassification. It also determines whether the sample size and overlap are sufficient for the proposed design. ⁸ , ⁹ , ¹¹ , ¹⁴ , ²⁵ Key questions include: does the database reliably capture requisite baseline variables, and are there important differences between patients with and without these assessments? How and how often are outcomes assessed? How much outcome information is missing at clinically relevant timepoints, and is missingness likely to be informative? Are key values implausible or inconsistently coded? If confounding or measurement error is anticipated, can its likely magnitude be characterized and addressed through statistical control? These assessments may lead to adaptation of the data source, design, endpoint, or estimand before the protocol is finalized, and may require protocol amendments as unexpected data challenges arise.

Safety evaluation in RWD requires more than a simple translation of diagnosis codes into MedDRA terms. The product and clinical setting determine the safety questions, including events of special interest, risk windows, ascertainment algorithms, validation needs, and availability of severity or grading information. All should be pre‐specified. For submission, FDA‐supported terminology and data standards may require mapping of source terminologies such as ICD‐10‐CM or SNOMED CT to standardized concepts. FDA recommends early discussion of planned transformations and mappings. ¹⁰ The WHO ICD‐10‐to‐MedDRA map can support this work but is incomplete and does not by itself establish event validity or severity. ²⁶ Clinical, coding, epidemiology, and data‐standard expertise are needed to define and validate safety outcomes.

Regulatory alignment should occur as soon as is feasible. FDA's June 2026 draft guidance recommends discussing the proposed approach early in development, such as at a pre‐IND meeting but no later than the end‐of‐phase II meeting. ² The specific meeting type and questions will depend on the program. For a RWE component, discussion should address (at a minimum) the proposed role of the RWE, data source fitness, study design and estimand, key sources of bias, prespecified analyses, and the planned relationship between the RWE and the pivotal trial. The goal is alignment, not a guarantee that the resulting evidence will ultimately be sufficient for approval.

PROTOCOL DEVELOPMENT

A protocol and statistical analysis plan should be finalized and registered before study initiation when possible, but certainly before outcome analyses begin. For transparency, sponsors can use ClinicalTrials.gov when applicable and may also register observational RWD studies in the HMA‐EMA Catalogue of RWD studies or another suitable public registry. Protocol amendments, their rationale, and timing relative to data access and analysis should be thoroughly documented. Study results and the final report should also be made publicly available.

As previously noted, the protocol and statistical analysis plan will require careful collaboration among clinical subject matter experts, epidemiologists specializing in RWE, and trial biostatisticians. The epidemiologists and data scientists will have the expertise to refine elements of the study design and apply statistical techniques to address confounding, censoring, and missing and mismeasured data. Decisions about confounding control, such as whether to use propensity score weighting vs. matching, can affect the validity and interpretation of the final results. Aspects of the design, such as whether to select a single random index date or use all eligible dates in a nested trial emulation, should be thoroughly considered.

Protocols will need to address methods to minimize missing data and implausible values; data curation (completeness trends over time for key variables, processes used to mine and evaluate unstructured data, accuracy of code mappings); and data transformation (e.g., data standardization efforts, linkage efforts and their subsequent quality, and de‐identification procedures).

Before study initiation, sponsors should assess whether existing standard operating procedures are adequately documented, including systems, controls and training, and that they adequately cover registrational use of RWD. If gaps are identified, they should be filled. Some existing clinical development procedures may apply directly to RWE. Others may need to be modified to address data provenance, linkage, validation, role‐based data access, reproducible programming, protocol/SAP amendments, independent governance, audit trails, and submission standards. Some SOPs may need to be created from scratch. Organizations should document how each relevant process is controlled and where new or revised procedures are needed.

FDA engagement should continue as the protocol and analysis plan are developed. Sponsors may consider applying to FDA's Advancing Real‐World Evidence Program, which provides an opportunity for enhanced interaction on selected RWE proposals. ²⁷ Participation in that program is distinct from an FDA demonstration project or other public case study activity. The advantages and transparency implications of any such pathway should be evaluated separately.

STUDY EXECUTION

Human‐subject protections depend on the nature of the data and study activities. Secondary research using de‐identified data may qualify for IRB exemption, but this should not be assumed categorically. Sponsors should determine and document the applicable IRB/ethics requirements and, when appropriate, obtain an IRB determination or exemption documentation. ¹⁴

Data cleaning, cohort construction, confounding adjustment, and efficacy and safety analyses should be conducted and documented to support reproducible processes appropriate to regulatory use. Role‐based segregation of data access by function and stage of analysis can help minimize potential investigator bias. For example, personnel assessing baseline balance or making analytic decisions not fully anticipated in the protocol/SAP can be restricted from outcome information until prespecified criteria are met. The objective is a transparent record of who had access to which data, when, and for what purpose, supported by version control, audit trails, and independent review where appropriate.

Endpoint adjudication is usually determined by the single registrational trial. If the interventional trial uses independent central adjudication, replicating that adjudication can strengthen an RWE study, provided the necessary source material is available. In the absence of a central committee, adjudication can be conducted by two independent reviewers, with ties going to a third for resolution. Blinding should be specified: adjudicators can be blinded to treatment/exposure status and other information that could reveal study group or subsequent outcomes. Investigators and programmers may require different access restrictions depending on their roles.

Should the study results be favorable, a central review question is whether residual bias or measurement error could plausibly explain the observed findings. FDA guidance on RWD/RWE, externally controlled trials, and noninterventional studies emphasizes prespecification of protocol and statistical analysis plan, data fitness, and evaluation of important sources of bias. ⁸ , ⁹ , ¹³ , ¹⁴ Best practices include examining covariate balance and overlap, assessing informative missingness and censoring, conducting sensitivity analyses using alternative eligibility or outcome definitions, evaluating robustness to analytic choices, and considering negative control outcomes or exposures when scientifically appropriate. Quantitative bias analysis can be used to assess the magnitude of unmeasured confounding or misclassification that would be required to materially alter the conclusion. These analyses do not “prove” the absence of bias; they make assumptions and residual uncertainty more explicit.

When a dataset is submitted to the FDA, the Agency requires data to conform to supported data standards. For RWD, sponsors should discuss mapping and transformation approaches with the review division as early as possible and describe them in the protocol, data management plan, and final study report. ¹⁰ Even if full CDISC conversion is deferred, a conversion or mapping plan should be developed early enough to inform data acquisition and study design. The source and analysis data can be used to conduct the study before conversion. Full transformation to applicable FDA‐supported standards only needs to be done if the study is included in the NDA/BLA. If the study is not submitted, for example, the program is discontinued or the sponsor elects not to file, full conversion may be unnecessary. When submitted, final analyses and tables should be reproducible from the standardized submission datasets. DEFINE.XML and related documentation can be labor‐intensive, and sponsors should plan adequate time for validation and reproducibility checks.

During the conduct of the study, an independent Clean Room Committee (CRC) or Data Review Committee can be useful to address unanticipated findings. ²⁸ Their purpose is to provide independent, documented review while limiting knowledge of treatment effects or other outcome information that could bias decision making. A CRC does not replace protocol/SAP amendments, sponsor governance, or FDA alignment for significant protocol changes. In a study of treatment effectiveness for newly diagnosed immune thrombocytopenia, independent review was used to evaluate issues arising during trial emulation and to inform documented analytic decisions. ²⁹ In another example, a composite outcome involving hospitalization for hepatic encephalopathy was revised after the original algorithm was found to be too narrow. ³⁰ Any change should be justified, dated, version‐controlled, and assessed for whether regulatory discussion or formal amendment is needed. All interactions with and actions of the CRC should be scrupulously documented.

Technological requirements should be assessed early, particularly for large longitudinal datasets. Sponsors should plan to retain sufficient source‐level and derivation information to reproduce cohort attrition, eligibility decisions, mappings, and analysis datasets. Submission data must conform to applicable FDA‐supported standards, but terminology mapping requirements depend on the source domain and submission standard. Sponsors should discuss data‐standard mappings, transformation, and submission logistics with FDA early. ¹⁰ Large RWD submissions can also strain data transfer, trial master file, validation, and eCTD workflows, so vendors and systems should be tested before filing.

FILING

A common question is how the CSR for an RWE study differs from a traditional trial CSR. The high‐level answer is: it does not. While there are some additional elements, such as the specific statistical methods described in analysis sections, the overall format of the CSR should not change. A template for a registrational trial with RWE should be created, and existing CSR SOPs should be appropriately modified.

After filing, sponsors should be prepared for an FDA audit. Data transfer, management, analysis, and reporting should follow appropriate guidances and be fully transparent to the Agency. Review existing SOPs and either modify or create SOPs specific to RWE study audits to ensure audit readiness (for practical advice, see Grandetti et al., 2025). ³¹

CONCLUSIONS

FDA's framework for substantial evidence and the use of RWD/RWE have evolved rapidly. The June 2026 revised draft guidance gives greater prominence to one adequate and well‐controlled clinical investigation plus confirmatory evidence, while preserving the statutory substantial evidence standard and emphasizing the strength of the total evidence package. ² RWE may serve as confirmatory or supportive evidence, but its acceptability depends on the specific data, design, conduct, analysis, investigational product, and clinical context. The practical challenge is to build an RWE component that is sufficiently transparent, reproducible, and fit for the regulatory question, recognizing both the strengths and shortcomings of RWE.

There are several key practical takeaways for implementing this change:

• Assemble multidisciplinary expertise early. Teams should include a combination of clinical, epidemiologic, biostatistical, regulatory, operational, safety, and data‐standard expertise required for the specific program.

• Assess procedural documentation and systems up front. Identify and fill gaps in data source evaluation and feasibility assessments, governance, data access, programming, validation, audit trails, training, and submission processes needed for registrational RWE.

• Discuss with the FDA the overall substantial‐evidence strategy and proposed role of RWE early in development and no later than the end‐of‐phase 2 meeting before protocols and analysis plans are finalized. ² Consider the Advancing RWE Program separately when its eligibility and interaction model fit the project. ²⁷

• Address technology issues, especially those concerning dataset size, up front. Do not wait until submission.

• Follow a three‐step process:

  • ◦
    Conduct data source and design feasibility work and prepare a sufficiently detailed synopsis to support early FDA discussion.
  • ◦
    After regulatory alignment, finalize and register the protocol/SAP and execute the study using documented controls appropriate to the intended regulatory use, including independent governance when warranted. The SAP should include plans to evaluate residual bias through prespecified sensitivity analyses.
  • ◦
    If the study will be submitted, complete the required data‐standard transformation before FDA submission. Ensure audit readiness appropriate to RWE.

The role of RWE in future approvals will depend on the quality of individual evidence packages, evolving FDA guidance, and accumulating regulatory experience. The June 2026 draft guidance creates additional clarity about how FDA will assess substantial evidence, but it does not establish that RWE will be accepted more frequently or that development programs will necessarily be shorter or less expensive. The practical recommendations in this article are intended to help teams evaluate when RWE is appropriate and, when it is used, to generate evidence that is transparent, reproducible, and aligned with regulatory expectations.

CONFLICT OF INTEREST

The authors declared no competing interests for this work.

FUNDING

No funding was received for this work.