Secondary Use, Registries and Real World Data

Part of Consent and Vulnerable Populations: Data, Samples and Secondary Use.

About 6 minutes. Free, in your language, no account needed.

Why this matters

A growing proportion of clinical evidence comes from data that was never collected for research: electronic health records, insurance claims, disease registries, and routine clinical information.

Nothing is done to anyone in this research. No visit, no intervention, no procedure. The traditional consent model is built around agreeing to have things done to you, and it simply does not describe the situation.

ICH E6(R3) Annex 2 addresses this territory, including consent and permission considerations for real world data, and its existence marks the field's recognition that the standard model has a gap here.

Why consent is often impracticable

Not a convenience argument in most cases.

Scale. A study using a decade of national records concerns millions of people.

The dead. A substantial proportion of people in historical datasets have died.

Loss of contact. People move, and a researcher has no route to them.

The selection problem, which is the strongest argument and the least understood. If consent is sought and only some people can be reached and agree, the resulting dataset is not a sample of the population. It is a sample of the reachable and willing, who differ systematically from everyone else in ways that bias the result.

That last point matters ethically as well as scientifically. A consented subset can produce a conclusion that is wrong for the population it will be applied to, and applying a wrong conclusion to people is a harm that a consent requirement caused.

What is required instead

Ethics committee approval, assessing whether the conditions for proceeding without consent are met: impracticability, minimal risk, that the research could not reasonably be done otherwise, and that the use does not conflict with what people were told.

A lawful basis under data protection law, which is separate and independent. Many frameworks provide a basis for scientific research that is not consent, with its own safeguards.

Safeguards proportionate to the risk: minimizing identifiability, access controls, defined retention, and restrictions on further sharing.

Transparency. People should be able to find out that their data may be used this way, through public information about the dataset and its uses. This is a weaker mechanism than consent and it is not nothing, and it is frequently the only mechanism available.

An objection route where feasible, allowing a person who does not want their data used to say so.

What "anonymous" does and does not mean

The most consequential misunderstanding in this area.

Data described as anonymous frequently is not, in the sense people assume. Removing names and identifiers does not make a record unidentifiable where the remaining information is distinctive enough. A rare condition, an unusual combination of characteristics, a precise date and location, or a genetic sequence can each identify a person.

The honest position is that identifiability is a spectrum and a property of a dataset in its environment, not a binary achieved by deleting a column. A dataset that is practically unidentifiable alone may be identifiable when linked to another.

Treat claims of anonymity as claims requiring assessment, and never tell participants their data is anonymous when what is true is that direct identifiers have been removed.

The consent people gave, and what it covers

Much secondary use concerns data collected under a consent given for something else.

A participant who consented to a trial in 2015 consented to that trial. Whether their data may be used in 2026 for a different question depends on what they were told, and most consent forms from that era said something general.

The disciplined approach is to read the actual consent and ask what the participant would reasonably have understood, rather than what the words could bear. Where the answer is unclear, an ethics committee decides rather than the researcher who wants the data.

In practice

A team proposes using data from three completed trials to answer a new question.

Rather than assuming the original consents cover it, they read all three. Two include language covering future research on related conditions, which reasonably covers the new question. One does not: it describes use for the trial and for regulatory purposes only.

The team's options are to exclude the third trial, or to seek approval to use it without consent, with an honest account of what participants were told.

They choose to seek approval, and the ethics committee grants it for the third trial with conditions: no attempt at re-identification, no linkage to external datasets, and destruction of the working dataset after the analysis.

The whole exercise takes three weeks. Assuming the consents covered it would have taken none, and would have used one trial's data outside what its participants agreed to.

What this looks like when it goes wrong

A registry collects clinical data with a consent describing use for improving care for the condition.

The registry's holder enters an agreement with a commercial partner, providing access to the de-identified dataset for product development.

The consent did not mention commercial partners, does not mention sharing outside the registry, and described a purpose participants would reasonably read as non-commercial.

When it becomes public, participants object, and some withdraw. The registry's recruitment falls and clinicians who referred patients stop doing so.

The legal analysis may well support the sharing. The consent describes a purpose, and participants understood that purpose in a way the arrangement did not match, and the registry's value depended entirely on a trust that this exhausted.

Key takeaways