The data is there. So why aren’t we using it?

Reflections from HLTH Europe 2026, where I chaired a panel on pharma’s most valuable, and most wasted asset

Only about 5% of health AI tools are validated on real patient-record data (Bedi). Only about 5% of UK clinical trials use (ABPI). Set those two numbers against the volume of data that health systems generate every day, and you have the central paradox of healthcare’s AI moment: the asset is everywhere, and we barely touch it.

I put that paradox to four people who spend their working lives on it: Karina Morley of AstraZeneca, Dr Kimberly Noel of Roche and the newly launched Human in the Loop, Dr Fernanda Polubriaginof of Memorial Sloan Kettering, and Dr Steffen Hess of Germany’s Health Data Lab at BfArM. What came back was not a story about missing data. It was a story about access, trust, and what you do once you finally get in.

The need for connected data

Karina Morley made the sharpest observation of the session: in a fragmented system, the patient is the connective tissue. They are the ones carrying their own history from one clinician to the next. Around 40% of the time, she said, a patient seeing a new provider has to explain that history again from scratch. The information exists. It is simply trapped in silos, and the patient is the only thing holding it together.

When that data is connected it enables the full loop of enabling patient centred care, supporting proactive management of conditions, and enabling research.This matters because the largest prize is the patient already diagnosed but not optimised: the person with hypertension, COPD or chronic kidney disease whose condition is known to the system but whose care has stalled. CF’s own analysis of analysis of unmet care gaps found drop-off of between 25% and 75% as patients move from prevalence, to diagnosis, to treatment, to optimisation. Connected data closes that loop.

Denmark shows the same principle at the level of research. Through Trial Nation, a standing partnership between academia, the health ministry and industry, Denmark recruits into clinical studies at one of the highest rates per capita in Europe. The data is connected, the partnership is permanent, and the results follow. Research must be embedded in care, not bolted on, since research-active settings deliver better care even to patients outside the study, and real-world data should decide where trials run, going to where patients are.

Enabling access to data

The problem isn’t that the data doesn’t exist. It’s access, and access splits three ways. In the US, you can buy data on 170 million people, but it’s fragmented and rarely follows a patient over time. In the UK, you can’t buy it at all, yet cradle-to-grave records exist for the whole population. And Europe, which shares the UK’s reluctance to sell, is quietly opening up, with Germany the biggest surprise of all

For years I was told that the kind of population-scale access I wanted was impossible in Germany. The legal culture would never allow it. That is no longer true, and it was the most important shift I heard all week.

Steffen Hess runs the Health Data Lab at BfArM. He now holds statutory-insurance data on 75 million Germans, roughly 90% of the population, going back to 2009. In its first half-year of operation the lab took close to 100 data applications. If your application is in order, he said, you get an answer in four to seven weeks. Not months. Weeks. And, tellingly, the lab has not yet refused a single application outright; it works with applicants to reshape them.

A law five years ago set the mission: data already collected for administration must be made available for research, inside a secure environment, with as much freedom as the law allows. Germany is now broadly aligned with the European Health Data Space, due to be operational by 2029.

The instinct across pharma has been to treat GDPR as the enemy. The opposite is closer to the truth. Strong data-protection regimes, GDPR in Europe and HIPAA in the US, build the public trust without which none of this is usable. Karina was blunt about it: patients do not trust how their data is used, and if you do not solve that, the infrastructure is irrelevant. Trust is the key to unlock data. Regulation, done well, is how you earn it.

What you do once you are in

Access is necessary but not sufficient. The richest data is unstructured, buried in notes, PDFs and faxes, and turning it into something usable has always been the bottleneck.

Fernanda Polubriaginof described the scale of the problem with one figure. For a single cancer case, her teams curate around 1,000 data concepts by hand, with a lag of six to nine months. By the time the data is ready, the treatment decisions it should have informed have already been made. MSK now runs a combination of large language models across the entire patient chart to flag trial eligibility a week before the patient’s visit, so the clinician walks in already knowing. The work is in evaluation, measured task by task against human abstractors, and Fernanda was clear that for clinical decisions the human stays in the loop. That is the right instinct. AI that supports the clinician scales; AI that tries to replace the clinician’s judgement does not get adopted.

But there is a trap here, and Karina named it. A large language model let loose on a secondary-care record gives you what is in secondary care. The chronic-disease patients who represent the biggest prize are not there. They are in primary care, where the data is least connected. Solve the problem only where the data is easy and you miss the people who need it most.

The frontier: insight beyond patient data

The most provocative thread came from Kimberly Noel, who used the panel to launch Human in the Loop and who leads AI advocacy at Roche. Her argument: data is no longer the problem; insight is the frontier. And the most valuable signals increasingly sit outside the health system altogether.

She is right, and the examples are striking. The earliest indicator of an epidemic is not a clinical report but over-the-counter sales of cough and anti-diarrhoeal medicines, people treating themselves before they ever reach a doctor. New diagnostics can flag a predisposition to conditions such as Alzheimer’s a decade before any symptom. Fernanda made it personal: she has worn a device for ten years, and not once has that data informed her own care. The gap between visits is where disease actually progresses, and it is the gap we are blindest to.

The implication is uncomfortable for incumbents. The traditional data holders, providers, payers and pharma, are no longer the only ones with the asset. Patients hold it too, and armed with tools like ChatGPT they are moving faster than the institutions around them.

So what should leaders do?

First, treat data access as a strategic asset, and go where the patients actually are rather than where the prestige sits. Second, embed research in care instead of bolting it on; the evidence that research-active hospitals deliver better care even to patients outside the study is now hard to ignore. Third, build for trust from the first line of code, because compliance designed in is an enabler and compliance bolted on is a tax.

And stop waiting. Steffen’s closing line is the one that really struck me: the data is there, it will only get better, and if you can produce an actionable insight today, start now.

Ben Richardson is Co-founder and Managing Partner of Carnall Farrar. He chaired “Pharma’s greatest asset: unlocking the value of data” at HLTH Europe, Amsterdam, 16 June 2026. The unmet-care-gap figures draw on CF’s Value in Health analysis (2025).