How to design a post-authorisation study using registry data

How to design a post-authorisation study using registry data

Registry data does not become regulatory evidence by being collected. It becomes evidence by being designed for the question a post-authorisation obligation actually asks – and that design decision is made long before any analysis begins.

What a post-authorisation study obligation actually requires

A post-authorisation study is evidence generation conducted after a product receives regulatory approval, fulfilling ongoing obligations that EMA or FDA impose as a condition of market authorisation – commonly a Post-Authorisation Safety Study (PASS) or Post-Authorisation Efficacy Study (PAES). These obligations do not end when a product reaches the market. They extend across its commercial life, and registries are frequently the intended data source for meeting them.

Most registry-based post-authorisation submissions fall short of what regulators expect, and the reason is rarely the analysis. Regulators evaluate whether the evidence is reproducible, traceable to its clinical source, and fit for the specific decision it supports. When a submission fails that test, the cause sits upstream – in a registry that was never designed to answer the question the post-authorisation obligation actually asks.

This distinction matters because sponsors routinely treat registry selection as a data-sourcing exercise rather than a study design decision. As explored in what FDA and EMA actually require from real-world evidence, evidence must satisfy pre-specification, reproducibility, and traceability before a regulator will accept it. A registry assembled retroactively from whatever clinical data happens to exist cannot retrofit those properties once the safety signal or efficacy question has already been defined. The failure is structural, and it starts at the registry layer.

What regulators and HTA reviewers are actually testing

A regulator or HTA reviewer evaluating a registry-based post-authorisation submission is not simply checking that a dataset exists. They are testing four properties, and each one determines whether the evidence can support the conclusion the sponsor is asking them to accept.

Patient-level data completeness

The registry captures every relevant clinical variable for every enrolled patient – not a partial record supplemented by assumption.

Outcome traceability

Each recorded outcome – a hospitalisation, a disease progression event, a treatment discontinuation – can be traced to the specific clinical documentation that generated it.

Longitudinal follow-up depth

Patients are observed for a duration sufficient to capture the safety or efficacy signal the obligation was designed to detect, not truncated by a registry’s enrolment window.

Bias control

The patient population reflects the treated population as a whole, not the subset that happened to remain engaged with a manual data collection process.

A clinician reviewing this evidence asks a simpler version of the same question: does this dataset reflect what happened to these patients, or does it reflect what a data collection process managed to capture? Regulators ask the formal version of that question, and the answer depends entirely on decisions made before a single patient was enrolled.

Using registry data for a post-authorisation obligation is not a shortcut around primary data collection. It is a design choice that carries the same rigour primary collection would require.

The four design decisions that determine regulatory credibility

Four decisions determine whether a registry can support a post-authorisation submission, and each must be locked before data collection begins – not adjusted once the analysis reveals a gap.

Case definition and inclusion criteria determine which patients the registry represents. Defined loosely, the resulting population will not match the regulatory question being asked.
Outcome variable specification determines what the registry will be able to demonstrate. If the primary safety or efficacy endpoint was not defined at registry inception, no downstream analysis can recover it from data that was never structured to capture it.
Follow-up architecture determines whether patients are observed continuously or at scheduled intervals that may miss the clinical events the obligation requires.

The fourth decision – data extraction methodology – is where traditional registries most consistently fail post-authorisation requirements. As detailed in the gap between structured and unstructured EHR data, the clinical variables most relevant to safety and efficacy conclusions – disease staging, physician-assessed response, reason for treatment discontinuation – appear almost exclusively in free-text clinical narrative rather than coded fields.

A registry that extracts only structured data will systematically miss the evidence a post-authorisation obligation is designed to surface. Clinical NLP closes that gap by extracting these variables directly from the clinical record with the traceability a regulatory submission requires – not as a technical enhancement, but as the difference between a registry that can answer the question and one that cannot.

Why static registries cannot absorb a multi-year obligation

Post-authorisation obligations are rarely satisfied by a single submission. A PASS may extend across years of safety monitoring. A label update may require new comparative evidence years after the original approval. An HTA re-evaluation may demand outcomes data the original registry was never designed to track. Each is a distinct evidence question arriving on a timeline the sponsor does not control.

A registry built as a fixed-schema snapshot cannot absorb that sequence without being rebuilt. As argued in living registries versus traditional registries, each new evidence question that a static registry design cannot anticipate requires new site contracts, new data collection protocols, and a new multi-year timeline before usable data exists. For a sponsor facing a five-year PASS commitment with interim safety reporting requirements, that rebuild cycle is not a minor inconvenience. It is the difference between meeting a regulatory deadline and missing one.

A living registry architecture – one that continuously ingests clinical data and extends its scope as new evidence questions arise – is not an enhancement to post-authorisation evidence generation. It is what the obligation structurally requires, because the obligation itself does not arrive as a single, bounded question.

Governance-by-design: why audit-readiness cannot be retrofitted

Regulators do not evaluate registry data in isolation. They evaluate how it was governed: who accessed it, how it moved from clinical narrative into an analysis dataset, and whether an independent reviewer could reproduce the result. A dataset that is accurate but ungoverned is not defensible evidence – it is a claim without an audit trail.

Governance-by-design means data lineage is recorded automatically as clinical information moves from source document to structured variable – not reconstructed after the fact when a regulator asks. Access controls determine who touched the data and when. Version-controlled protocols document how extraction methodology evolved over the registry’s life. Audit trails survive scrutiny because they were built into the pipeline rather than assembled for the submission.

This is what Savana Data Space and Savana Manager Suite provide at the governance layer: audit-readiness that the system structurally guarantees by how the data moves through it – not something a compliance team manually assures after the fact. Treating governance this way is not compliance overhead. It is the condition under which registry-derived evidence becomes something a regulator can accept.

Network selection as a methodological decision, not a logistics exercise

The hospitals contributing data to a post-authorisation registry are not a logistical detail settled after the scientific design is finished. They are a methodological variable with direct consequences for the evidence’s validity.

Which institutions are included determines whether the patient population reflects the treated population as a whole or a narrower subset concentrated in academic centres with research infrastructure. Clinical pathway variation across institutions determines whether the registry captures real-world treatment diversity or a single institutional protocol. Data completeness rates vary by site, and regulatory jurisdiction determines which national frameworks govern the data a given institution contributes. Each of these factors shapes what the registry can credibly claim to represent – which means network selection must receive the same design rigour applied to patient inclusion criteria.

The Smart Health Alliance illustrates what this looks like when representativeness is achievable rather than aspirational: a federated network spanning more than 280 institutions across multiple countries, with harmonised data standards already in place. A post-authorisation registry built on a network of that depth can credibly claim population representativeness. A registry built on a handful of convenience sites cannot – regardless of how rigorously its case definitions were written.

Registry design as infrastructure investment, not a one-time deliverable

The decision a sponsor makes when choosing a registry for a post-authorisation obligation is not how to design this specific evidence generation effort. It is how to build evidence infrastructure that will serve this submission and the ones that follow – because post-authorisation obligations rarely arrive once.

A registry designed correctly at inception – with locked case definitions, extraction methodology that reaches unstructured clinical narrative, governance embedded from the point of data capture, and a network depth that supports representativeness – becomes a reusable asset. The same infrastructure that satisfies an initial PASS requirement extends to safety monitoring, supports a subsequent label extension, and feeds directly into HEOR and market access evidence generation without renegotiating site contracts or rebuilding data pipelines. Each subsequent evidence generation effort costs less than the one before it, because the infrastructure investment was made once, upstream, rather than repeated at every regulatory milestone.

This is what Savana NGR represents in practice: living evidence infrastructure built around the premise that post-authorisation evidence generation is not a single deliverable but a sustained capability. A sponsor evaluating registry data for a post-authorisation obligation is not choosing a data source. They are deciding what kind of evidence infrastructure their organisation will depend on for the next five years – and that decision determines the answer long before the first patient record is touched.

A post-authorisation study built on registry data is only as credible as the registry that generated it. Design determines validity before analysis ever begins.

Sources & further reading

  1. European Medicines Agency. Guideline on good pharmacovigilance practices (GVP) – Module VIII: Post-authorisation safety studies. EMA/813938/2011 Rev 3. ema.europa.eu
  2. U.S. Food & Drug Administration. Real-World Evidence: Framework for FDA’s Use of Real-World Evidence to Support Regulatory Decision-Making for Drugs and Biologics. December 2018. fda.gov
  3. European Medicines Agency. Guideline on registry-based studies. EMA/EFPIA/832808/2020. ema.europa.eu
  4. Savana. Regulatory-grade RWE: what FDA and EMA actually require. savanamed.com
  5. Savana. Structured vs unstructured EHR data: the gap that breaks RWE. savanamed.com
  6. Savana. Living registries vs traditional registries. savanamed.com
  7. Savana. What is a disease registry in clinical research? savanamed.com

Build the evidence infrastructure your PASS obligations actually require

Savana NGR is living evidence infrastructure designed for the full post-authorisation lifecycle – from initial PASS submission through safety monitoring, label extensions, and HEOR. Governance-by-design, clinical NLP extraction from unstructured EHRs, and a federated hospital network that supports genuine population representativeness, all within a single, expandable architecture.

Explore Savana NGR
ABOUT SAVANA

Savana is the only healthtech SaaS company delivering scientifically validated, multilingual Clinical NLP exclusively trained on EHRs, ensuring the deepest dRWE for hospitals, HTA bodies, pharma, and MedTech. It also powers the Smart Health Alliance, a global hospital network collaborating within Savana’s ecosystem to drive the most innovative and disruptive RWE use cases, including the market’s only AI-powered living registries.

Share it here:

Related Posts

The evidence behind the headlines.

Stay ahead of how real-world data is reshaping clinical practice, drug development, and healthcare policy.