On patient record systems in Swedish healthcare. We should build our own. We cannot afford not to.

S Seven deaths in which the patient record system Cosmic is identified as a factor, though not the sole cause. That was reported by SVT’s investigative programme Uppdrag granskning in April 2026, after mapping incident reports since autumn 2024. Of 3,210 reports mentioning Cosmic, 646 were judged to have a clear connection to the system. The material describes, among other things, missing referrals and difficulties finding notes in patient records. During a cardiac arrest in Falun, staff could not log in and read the patient record. The incident report judged that the patient’s life probably could not have been saved anyway, but the loss of access to the record was serious. Cambio stated that it had received no information identifying faults in Cosmic as the cause of a death. Doctors interviewed in the investigation also estimated that they were seeing 15–50 per cent fewer patients. Those are their estimates, not an overall measure of healthcare activity across all regions.

In November 2025, Region Stockholm signed an agreement with Cambio to introduce Cosmic in Stockholm and on Gotland. At the contract award decision in February 2025, Stockholm stated a contract value of SEK 2,375 million for the initial twelve-year term. When the agreement was signed, the region described a five-year implementation phase followed by eleven years of use, with a gradual introduction within five years. CGM and Systematic had challenged the award in court, but the legal proceedings ultimately did not prevent the agreement being signed. Despite the criticism, Stockholm is proceeding with Cosmic.

I am a psychologist licensed in Sweden. I have worked in the TakeCare environment in Stockholm healthcare for almost a decade, from first-line mental healthcare in a primary care centre to specialist psychiatry and inpatient care, and today I run my own practice for talking therapy. I am therefore writing this as a clinician, not a technician, one of many thousands who have sat with these systems between themselves and the patient. And my point is not that TakeCare should be defended (it has been in use since the 1990s, still impressive). My point is that the whole model is wrong. What honestly surprises me most is that this is not a much bigger debate, that the conversation is still about which of the existing systems, read: dead ones, should be the hub of Swedish healthcare, when the real question ought to be why we are choosing between them at all.

A system from another time

TakeCare has been used in Stockholm healthcare since 1995. It needs saying, because it is not unimportant: in my experience it is a reliable system that does the basics. But I see it as built on another era’s assumptions about what a patient record is. My objection concerns the workflow, an interface where the clinician has to adapt to the system rather than the other way round, and where it is difficult to get the overview the work requires. What ought to take seconds takes minutes, over and over again. That friction is not a detail to write off in a footnote; it is working conditions, patient safety and healthcare capacity all at once.

And TakeCare is not owned by the healthcare service. It is owned by CGM (CompuGroup Medical, a German group).

The replacement is dead on arrival

Cosmic, the replacement, is not new. The system began being introduced into Swedish healthcare more than twenty years ago, and as a piece of technology it is about as alive as a grilled chicken in the summer heat: it can be served, but no one wants to use it. This is what Stockholm is now to pay billions to introduce, a system that, in regions already running it, has prompted doctors to describe their clinics being halved, swollen the administration and literally brought healthcare staff out in protest. As noted, Uppdrag granskning counted seven deaths in which the system was identified as a factor, though not the sole cause. One might ask how a supplier can retain trust after such an investigation, and why the next major customer continues the introduction. Another problem is the inertia of not also owning the development ourselves.

Cambio argues that a new patient record system can temporarily affect capacity, but disputes the picture of a general reduction in capacity. That defence would have been easier to take seriously had the GP Knut Bodin, who was there when Uppsala introduced Cosmic more than twenty years ago, not described in the same investigation how, in his judgement, the loss of capacity was still there. He estimated that he had seen fourteen to seventeen patients a day before the system. Now, he told SVT, he could not even manage ten.

The ownership does not help. Cambio was founded in Linköping in 1993, and Investcorp, based in Bahrain, became the majority owner in 2019. The year before, five regions had carried out the Sussa procurement, which a further four regions later joined. Sveriges Radio’s P4 stations have investigated Investcorp’s links to undemocratic regimes on the Arabian Peninsula. This does not mean the investors own the patient records or automatically have access to them. Healthcare providers are responsible for the processing of personal data, and Cambio states that the owners have no access to patient data. But for me, the question remains of who ultimately controls the supplier and the long-term development of such sensitive infrastructure. Ownership ought to carry considerably more weight than it does. Uppdrag granskning has also shown how scores were changed in Cosmic’s favour in the Sussa procurement.

Millennium in Västra Götaland became another failure. The programme had an estimated cost of around SEK 5.5 billion, but implementation in the region’s southern area was paused after three days in November 2024 when patient safety risks were identified. In February 2026, the regional assembly decided to move towards a modular health information environment. The contractual issue with Oracle was still under discussion at that point. The SEK 5.5–11 billion appearing in later material concerns additional costs for different options, not an established final bill for Millennium. For me, this is more than an individual procurement mistake. It shows a systemic fault in how Sweden obtains critical infrastructure. Germany, Bahrain or the USA: that is the ownership picture behind the major systems I discuss here. Where is the alternative that healthcare itself controls?

The third (only) alternative: build it ourselves!

Here is what has changed, and what the debate has not quite taken in: AI can change how we develop software. In Anthropic’s presentation of Fable 5, Stripe describes, for example, a code migration completed in one day that was otherwise estimated to require more than two months of manual work by a team. This is an example reported by a supplier, not a calculation for Swedish healthcare. But such changes ought to make us reconsider what we can build ourselves and what it would cost.

I should be clear that I am not the right person to calculate the cost of a national patient record system; there are people who do this properly and they should do it. But let me still do the sums out loud, as a thought sketch. A development team of a hundred people – systems architects, AI engineers, security experts and, just as importantly, clinicians designing the workflows – at an assumed full cost of two million kronor per person per year, for five years. That makes one billion kronor for the team. It does not show that a finished national patient record system would cost one billion, or that a hundred people and five years would suffice. Operations, implementation, integrations, migration, training and long-term maintenance also need to be included. But calculate the alternative! Continuing with today’s model is also an economic decision, and that comparison ought to be made with today’s development tools.

Of course I know that implementation, migration and training also cost a fortune. That is why a system we own ourselves must be built for gradual introduction and testing in real healthcare workflows. This requires both the right architecture and the right organisation. Ownership alone does not solve it. And no serious engineer promises a system without bugs. Established interoperability standards such as FHIR can provide a basis for information exchange, but access control, information security and patient safety must be designed and reviewed separately. AI-assisted testing can support that work. It does not replace the responsibility to demonstrate that the system works where it will be used.

The sketch is not difficult to formulate. A state-owned technical core for patient records, with shared solutions for information management, access and security. The division of responsibility between the state and healthcare providers must be clear, and patients’ rights must come with it. On top of that, interchangeable modules for notes, prescribing, referrals, laboratory services, booking and decision support. The ambition is for each module to be procured, replaced, improved or rewritten separately, by Swedish companies or healthcare’s own developers, without tearing up the whole. That is the infrastructure I want: a system built to be improved rather than replaced. The patient record becomes like the road network: the state owns the ground and many can build on it. It would also be honourable, in the old sense of the word, and a worthy farewell to TakeCare after decades in the service of healthcare. A country building its most important infrastructure itself, well and on its own account.

And flexibility is not a bonus in that sketch; it is the point itself. AI development cannot be seen as a one-off event we adapt to and finish with; the tools change rapidly. Anthropic’s launch of Fable 5 and Mythos 5 on 9 June 2026 is an example of how quickly the conditions change. It means the healthcare systems of the future must be built to respond continually to new possibilities and new security threats, rather than to be finished on delivery. We needed to start rethinking yesterday. The next best thing is to begin now, from the ground up, ourselves, with the support of the best models available.

Passing the ball to the state

Sweden has done this before. BankID, Swish and the Swedish Tax Agency’s digital services show that Sweden can build world-class digital infrastructure when banks, public authorities or both decide to, and we have the engineers, AI expertise and public agencies that could carry the task within the country’s borders. What is missing is not ability but a decision.

Minister for Social Affairs Jakob Forssmed has called Cosmic a failure, and the government has instructed the Swedish eHealth Agency to analyse the market, procurement and introduction of health information systems. That is good. The state is also working on national digital infrastructure and shared conditions for exchanging health data, which is more than simply admitting more suppliers. But I want to go further, to a shared patient record platform that the public sector owns and whose development it can direct. More systems on the market do not in themselves settle who should control the underlying infrastructure.

So, Jakob Forssmed, Elisabet Lann and Erik Slottner, the question lies on all three of your desks: social policy, healthcare and digitalisation. Why should the most sensitive thing we have, the story of our bodies and minds from birth records to palliative care, depend on systems whose long-term development healthcare does not itself control? I want the same seriousness about healthcare’s digital self-determination as about defence. Give a coordinated national mandate to build a Swedish, open, modular patient record platform using the AI-assisted development capacity already in the country. Let a proper costing and testing of security guide implementation. We should build our own, taking responsibility both for the cost and for what the systems will carry.

Afterword · 6 September 2026

Two things have happened since the essay was published. On 18 June, the Government Offices presented proposals for laws and regulations as a first step towards national digital infrastructure for health data. This is proposed legislation, not legislation already in force.

On 28 August, VGR and Oracle signed an agreement to terminate the Millennium contract. The termination payment is SEK 657 million. That announcement belongs to developments after the essay’s publication on 12 June.