Is your data fit for purpose? Designing research for regulatory-grade evidence

Overlooked outcomes, unrepresented patients and gaps in the audit trail can prevent real-world data from supporting a regulatory decision.
November 15, 2026
7 min read

A dataset can be technically sound and still fail a regulatory decision. Its suitability depends on the question, the affected population, the required variables and outcomes, and the traceability of the path from source to analysis. No universal completeness, linkage or quality threshold certifies a dataset for every use. A later question can expose a variable, patient group, follow-up period, linkage or provenance that the original source never captured. Research teams need to assess what a source contains for the current protocol and which gaps may emerge as evidence needs change over time.

What is regulatory-grade data?

There is no universal regulatory-grade threshold. “Regulatory-grade” is often used as shorthand for high-quality data, but that description is incomplete.

The MHRA defines regulatory-grade data as “data of sufficient quality and integrity to support regulatory decision-making”. Its analysis acknowledges that there isn’t currently a single consensus definition for what makes data “regulatory-grade”, but points to data quality and integrity, fitness for intended purpose, and representativeness of its target population, as determinants of data suitability.

The FDA’s guidance on electronic health record and medical claims data makes a similar distinction - necessitating ‘reliability’ and ‘relevance’. The guidance goes on to define reliability as including accuracy, completeness and traceability. Relevance, it says, depends on whether the study considers key study variables and a sufficient representative patient cohort.

The European Medicines Agency’s 2026 data quality framework does not set universal thresholds or acceptance criteria, but takes a contextual approach, assessing the fitness-for-use of a dataset against specific research objectives.

Across these frameworks, regulatory-grade data combines sufficient quality, integrity, relevance, representativeness and traceability with fitness for a specific purpose. Research designed to generate regulatory-grade evidence must therefore ask, “as the research evolves, which decision-critical gap is most likely to constrain it: what the source does not observe, whom it does not reach or what it cannot trace?”

Your evidence is constrained by its weakest dimension

Three sets of questions help establish whether a source can support the proposed decision:

  • Coverage of variables:
    ‍
    Does the source capture the exposure, endpoints, covariates and follow-up required by the protocol? Are those elements defined and measured with enough consistency for the intended analysis?
  • Cohort reach and representativeness:
    ‍
    Does the source represent the people affected by the decision? Who enters the dataset, who remains outside it and how might that selection affect interpretation?
  • Integrity and provenance:
    ‍
    Can reviewers trace the data from collection through transformation and analysis? Do the source records and documentation preserve definitions, changes and quality controls?

A decision-critical weakness in any one area can limit the study’s regulatory use, regardless of the source’s strengths elsewhere. Adding patients cannot recover an endpoint the source never measured, and richer variables cannot show whether findings apply to patients who never entered the cohort. If the audit trail does not allow reviewers to reconstruct how the data were collected and transformed, neither strength makes the evidence verifiable. Where provenance is incomplete, reviewers may be unable to reconstruct how the evidence was produced. Overall data quality cannot compensate for a requirement on which the regulatory question depends.

An analysis-ready dataset is not the same as a review-ready evidence package.

While the final dataset contains the values used for analysis, regulatory review may also require the underlying source records, consistent endpoint derivation, documented cohort selection and an auditable record of how data were transcribed or transformed. A 2025 peer-reviewed analysis by FDA staff examined recurring problems identified during inspection of studies using real-world data.

Inspection finding What reviewers could not verify
Source records were initially unavailable for inspection; a proposed remote review of a redacted subset was insufficient. The pivotal real-world data could not be fully verified until access and consent arrangements were resolved.
Clinicians in the external control arm assessed and scored the efficacy outcome differently from investigators in the interventional arm. The inconsistency affected the study results, and the external-control data were deemed not fit for use.
Medical-record data were manually transcribed into spreadsheets and emailed to a registry without an audit trail or sponsor monitoring. Inspectors found significant discrepancies in the primary endpoint between source records and submitted data.
Records needed to show how the external control population was identified were unavailable. Inspectors could not adequately evaluate selection bias, creating a risk to the evidentiary package.

These findings affected different parts of the evidence chain, but produced the same constraint: reviewers could not reconstruct or verify how the analytical evidence had been produced. Regulatory readiness therefore depends on the records, methods and controls behind the final dataset, not only the data submitted for analysis.

The FDA authors recommend engaging the agency and relevant third-party real-world data providers early when planning studies intended to support a marketing application. For evidence teams, this makes provider capability part of research design: can the provider preserve and explain the source records, cohort-selection logic, derivations and controls behind the analytical dataset?

Why the evidence gap often predates analysis

Post-hoc methods are an important analytical tool. Under explicit assumptions, analysts may be able to characterise missingness, adjust for measured imbalance, quantify linkage error or test the sensitivity of findings. However, analysis cannot directly observe a variable that was never captured, or verify whether findings hold in a patient population that was never reached. The EMA’s 2026 framework is plain on this point: if collected elements do not correspond to the population characteristics, exposures, outcomes and covariates required by the research question, the source is not fit.

Question to ask before source selection:

If the regulatory question changes, can this evidence base be extended—or will the new variable, outcome, follow-up period or patient group remain permanently outside it?

Evidence design determines which gaps remain addressable

Persistent, consented and recontactable cohorts offer a different evidence architecture from a fixed extract. Existing longitudinal health records can be combined with prospectively collected patient-reported outcomes, questionnaires, samples or other study-specific measures.

uMed ACCESS Cohorts combine longitudinal electronic health records with consented patient recontact and prospective data collection. Subject to the relevant protocol, consent and governance, research teams can collect newly required patient-reported outcomes, questionnaires, samples or follow-up and link them to the existing clinical record.

COMTEX-PD leveraged uMed’s AccessPD registry for a comparative-effectiveness study of opicapone in people with Parkinson’s disease. An existing consented cohort and ethics approvals enabled the first patient in eight days after contract signature. The study delivered its planned real-world evidence dataset in under six weeks, linking electronic health records from 1,100 patients with 3,200 patient-reported outcomes (PROs). Real-time EMR + PRO enabled faster insights, continuous retargeting and rapid iteration of study instruments. Longitudinal patient engagement allowed the team to collect purpose-specific measures and connect them with historical clinical data without recruiting a new cohort. Evidence teams cannot anticipate every future requirement, but longitudinal patient registries such as AccessPD preserve the ability to investigate these as they emerge.

‍