Data linkage is a vital process in healthcare research that allows us to combine information from different sources while ensuring patient privacy. At Barts Health, we focus on making this process straightforward and secure.

How We Link Data
We use various methods for data linkage:
Data Transfers: We can either send Barts Health data to a partner's secure environment or receive data from partners to link with our datasets.
Federated Analysis: In some cases, we analyse datasets separately without moving them, then combine the results using statistical methods.
Information Governance

Before linking data, we ensure that all necessary approvals are in place. This includes checking whether the identifiers (like NHS numbers) are already authorised for use. We assess the sensitivity of the data to determine how safely it can be shared. For example, sharing a list of NHS numbers is generally low risk, while sharing names and addresses is much more sensitive.

Linkage Methods
We have several approaches to linking data:
Partner Matching: One partner matches identifiers based on existing approvals, which is straightforward and low-risk.
Third-Party Matching: A trusted third party can perform the matching, ensuring that neither partner sees the other's full dataset.
Cryptographic Matching: Instead of sharing identifiable information directly, we use cryptographic hashes. This means that sensitive identifiers are transformed into a secure format that cannot be reversed, enhancing privacy.

Data Transfer Options
Once data is linked, we can transfer it in different ways:
Manual Transfers: Researchers can download and upload data manually when they have the necessary approvals and permissions.
Team Transfers: Our Barts Health Data Platform team can handle transfers directly for researchers.
• API Integration: For more advanced setups, we can automate data transfers through APIs, making the process seamless.

By continuously improving our data linkage strategies, we aim to support impactful research while safeguarding patient confidentiality.