Trusted Research Conversations series 1 is a wrap

Earlier this year we launched Trusted Research Conversations, a podcast from Aridhia for the people navigating the changes in health data research. With the first series now complete, this post looks back at all six episodes.

Why we started the podcast

Much of the most useful thinking on these questions does not sit in specifications or press releases. It sits with the people building federated networks, writing standards, harmonising data and deploying models in hospitals. The podcast gives us a way to have longer conversations with them about the practical realities of secure data research, federated analysis, AI governance, and the standards and regulations that underpin it all.

The episodes

We opened the series with Federated Learning and Trusted Research Environments, in which Chong Shen Ng of Flower Labs explained how federated learning moves computation to the data rather than data to the computation. He also explained the challenges of using this in a healthcare setting, where each hospital has its own policies and infrastructure. Solving this problem is the basis of our own work, described in Introducing Flower on the Aridhia DRE.

In episode 2, Dr Christian Cole (University of Dundee) and Tim Machin (University College London) described how community input steered the Standard Architecture for Trusted Research Environments (SATRE) towards risk and process rather than technology, and discussed its adoption and the open question of accreditation. Since recording, Data and Analytics Research Environments UK (DARE UK) has published SATRE 2.0 with a new Federation pillar.

Episode 3 turned to identity. Antti Kettunen of Veridient explained the European Digital Identity (EUDI) Wallet, expected in most member states between December 2026 and January 2027, and how a common, high-assurance identity layer could ease cross-border access under the European Health Data Space (EHDS).

In episode 4, Wouter Franke and Liam Glück of The Hyve walked through mapping data to the Observational Medical Outcomes Partnership (OMOP) Common Data Model with the Observational Health Data Sciences and Informatics (OHDSI) tools. They were candid that semantic mapping and data quality, not the tooling, are the hard parts.

In episode 5, Max Salmi of Veil.AI distinguished anonymised data, transformed from real records, from synthetic data, sampled from a model of a population. He argued that what matters is evidence of negligible re-identification risk and enough utility for the intended use, illustrated by a study in a Finnish Secure Processing Environment (SPE).

We closed the series with Johnny Sheldon and Cato Pauling of the Great Ormond Street Hospital (GOSH) DRIVE unit, who gave a frank account of building an assistive triage model for fetal skeletal surveys, including a shortcut the model learned from differences in scanner technology.

Common themes

The six conversations covered very different ground, but several themes recurred often enough that we think they are worth drawing out.

• Move the analysis, not the data. Federation was the explicit subject of episode 1, the focus of SATRE 2.0's new pillar in episode 2. Our own approach is set out in Building a Federated Trusted Research Environment and the TREvolution project update.
• The hard problems are rarely purely technical. Flower works, OMOP tooling is mature, and anonymisation methods are well researched. What slowed projects down was harmonisation, governance, permits, consistent definitions and getting secure infrastructure in place quickly enough.
• Standards need a community behind them. SATRE, OMOP/OHDSI, eIDAS and Flower's open source framework all depend on broad participation to work. Several guests stressed openness and transparency, including with the public, as a reason for adoption rather than a constraint on it.
• Quality and trust have to be demonstrated. Whether through a SATRE score, an OHDSI data quality report, evidence of low re-identification risk or a clinician reviewing model outputs, guests repeatedly returned to the need to show, not assert, that data and processes are fit for purpose.
• Rare conditions make collaboration essential. Rare diseases in children (PHEMS), rare abnormalities in fetal imaging (GOSH DRIVE) and small hospital cohorts (Veil.AI) all came up. Small numbers at any single site are a recurring reason why federation, harmonisation and privacy enhancing techniques are needed together rather than as alternatives.

No single technology or standard addresses all of this. What the series made clear to us is that TREs are increasingly the place where these approaches meet, and that the most useful work happens when platform providers, standards bodies, researchers and clinicians work on them together.

Thank you, and Series 2

We would like to thank all of our guests for giving their time and expertise so generously: Chong Shen Ng (Flower Labs), Dr Christian Cole (University of Dundee), Tim Machin (UCL), Antti Kettunen (Veridient), Wouter Franke and Liam Glück (The Hyve), Max Salmi (Veil.AI), and Johnny Sheldon and Cato Pauling (GOSH DRIVE). Each conversation taught us something new.

We are pleased to confirm that Series 2 of Trusted Research Conversations will begin in March 2027. We will continue to speak with people across the research, clinical and technology communities about the practical realities of secure data research, and we will share the line-up nearer the time. All Series 1 episodes are available on our podcast page.

If you would like to suggest a topic or guest for Series 2, or would like to know more about anything discussed in the series, please get in touch.