Smart Health: Lessons from Seven Countries on Getting Patient Data Right

A woman in her seventies collapses outside a shopping centre on a Saturday afternoon. The ambulance takes her to the nearest emergency department, which is not the hospital she usually attends. She is conscious but confused. She knows she takes a blood thinner; she cannot remember which one, or the dose. She mentions a heart procedure a few years ago, in another state. Her regular doctor is closed until Monday. Her daughter is on a plane.

The registrar now has to make a decision about anticoagulation without knowing what she is already on. Get it wrong in one direction and there is a bleed. Get it wrong in the other and there is a clot. So the registrar does what any competent clinician does in that situation: orders a fresh set of tests, repeats an imaging study that was almost certainly done eighteen months ago, and manages conservatively until something more definite arrives. The woman is admitted overnight for observation that may well be unnecessary.

Nobody in this story did anything wrong. The failure is not clinical. It is that the information needed to treat her existed, in full, in at least four separate computer systems, none of which could show it to the person holding the chart.

The case for a unified patient view is this story, multiplied by every after-hours presentation in the country.

It is a good case. It is also not the whole case, because the same information that would have saved that registrar thirty minutes becomes something quite different when it is assembled for an entire population and made available to researchers, insurers, planners, vendors and whoever eventually acquires the contract. Several countries have now run that experiment far enough to produce results. The results are more interesting than either the promotional literature or the privacy critique suggests.

What a unified patient view actually is

The phrase gets used to mean three fairly different things, and the difference matters enormously for cost, risk and politics.

A central repository copies clinical information into one national store. Australia and England have both pursued versions of this. It is the easiest to query, the easiest to explain to a minister, and the single most attractive target ever assembled.

A federated model leaves the data where it was created and builds an index that says which systems hold records about a given person. Nothing moves until someone asks. Denmark and Estonia both work this way. It is harder to build, slower to query, and far less catastrophic to breach.

Document exchange does not attempt a view at all. It moves standard-format documents between systems on request, which is broadly what the United States has built at national scale. It delivers less, but it delivers it without anyone having to agree on a single architecture first.

When a government announces a single patient record, it is worth asking which of these three it actually means. England is instructive here: the Single Patient Record was committed to in the Ten Year Health Plan published in July 2025 and promised through the NHS App by 2028, but of the three candidate architectures set out the following September, none of them stores a patient record in one place. The name describes the user experience, not the engineering. That is a perfectly defensible design choice. It is a less defensible thing to leave unsaid.

The Australian push, in public terms

Australia has spent roughly fifteen years and several billion dollars on My Health Record, and for most of that time the criticism was straightforward: the thing existed but was largely empty, because uploading to it was voluntary and clinicians had no reason to bother.

Two pieces of public policy changed that. The National Healthcare Interoperability Plan 2023 to 2028, published under the title Connecting Australian Healthcare, set out ten principles and five priority areas, with identifiers, standards and terminology at the centre rather than at the margins. Then the Health Legislation Amendment (Modernising My Health Record — Sharing by Default) Act 2025 received Royal Assent in February 2025, and the sharing-by-default obligation commenced on 1 July 2026, beginning with pathology and diagnostic imaging.

That is a genuinely significant shift. It changes the record from an opt-in contribution to a default obligation on providers, and pathology and imaging are the right place to start, because they are high-volume, highly structured, and the single most duplicated category of test in the system.

The Australian Digital Health Agency reports consumer views of the record up 433 per cent to more than three million a week, and provider uploads up 105 per cent to more than five million a week. It also reports indicative savings of $146.3 million over two years. The usage figures are measurements. The savings figure is the programme reporting on itself, and it should be read the way any self-evaluation is read — as a claim awaiting independent replication, not as a finding. That caution is not a criticism of the agency. It is the standard that the international evidence, below, makes unavoidable.

Seven countries, one recurring pattern

Connected health is not an Australian idea, and the countries that got there first have already surfaced most of the problems.

Sources for each row are listed at the end. Figures cited in the text are drawn from official programme reporting, regulators, inquiries and court judgments.

Read those rows together and a pattern emerges that is hard to unsee. In England, in Taiwan and in India, the technical build ran ahead of the consent and oversight framework. In each case the correction arrived from outside the programme — as mass opt-out, as constitutional challenge, or as a civil-liberties group demonstrating that patient names were publicly searchable on the programme’s own website.

The engineering is rarely what fails. What fails is shipping the engineering before the rules that make it legitimate.

England’s numbers make the point quantitatively. When the GP Data for Planning and Research programme was announced in 2021, more than 1.3 million people opted out of data sharing in a single month, roughly doubling the national opt-out rate. The programme was paused and never implemented as designed. Its predecessor, care.data, had already been cancelled in 2016 after spending around £7.7 million. Meanwhile the NHS App, which gives people access to their own information rather than extracting it for others, passed 39 million registered users and recorded more than 62 million logins in a single month in late 2025.

The same population will adopt a digital health product enthusiastically and refuse a data programme emphatically. Those are not contradictory behaviours. They are a consistent preference about who the system is for.

What Denmark does that almost nobody else does

Nearly every connected health programme in the world promises that access is logged and misuse will be detected. Denmark is the one that has made the promise operational in a way patients can verify themselves.

A Danish citizen can log in and read the access log on their own record. If a clinician with no treatment relationship opens it, the system sends the patient a letter. Random audits are run twice a year. Prosecutions for improper access happen, in small numbers, every year. The cumulative effect is that the undertaking is not a policy document; it is a thing that visibly happens to people who breach it.

This is the most transferable lesson available, and it is cheap relative to the rest of the build. Logging is already there in every major system. What is usually missing is the decision to show the log to the person it belongs to, and the willingness to act on what it reveals. Those are governance choices, not technical ones, and they are the difference between consent that is informed and consent that is merely collected.

The technologies actually doing the work

The interesting technology in connected health is almost never the technology that gets announced.

Standards and identifiers are the unglamorous foundation. HL7 FHIR has become the common language for exchanging clinical data, with national profiles layered on top; SNOMED CT provides the clinical vocabulary; and a reliable way of knowing that two records refer to the same person and the same provider is the thing that makes everything else possible. Australia’s interoperability plan is unusual and creditable in putting identifiers and terminology first rather than treating them as plumbing. Programmes that skip this step spend the next decade reconciling duplicate patients by hand.

Artificial intelligence enters at two very different levels of maturity. Ambient documentation tools, which draft a clinical note from the consultation itself, address the single best-evidenced harm of digital records, which is the time they take from clinicians. That is a real and near-term opportunity. Diagnostic and predictive models are a longer and more contested story, and the honest position is that strong published performance on retrospective data has repeatedly failed to reproduce as improved outcomes in live deployment. Both things are AI in health. They deserve very different levels of confidence.

Remote monitoring and consumer devices generate a volume of data that no clinical workflow was designed to absorb. The constraint is not sensor accuracy; it is that someone has to be accountable for looking at the output, and funding models rarely pay anyone to do it.

Privacy-enhancing analytics may be the most underrated item on the list. Federated analysis, where a query travels to the data and only aggregate results return, offers a route out of the central-repository trap: planners and researchers get answers without anyone assembling a national copy of everyone’s medical history. It is slower, harder and less flexible than a warehouse. It also removes the single point of catastrophic failure, and the catastrophes are now well documented.

The cybersecurity question, stated honestly

This is the part of the argument where both sides tend to overclaim. Advocates treat breaches as an implementation detail. Critics treat them as proof that the whole enterprise is reckless. Neither reading survives the public record, which is by now extensive enough to reason from.

The scale is not in dispute. The February 2024 attack on Change Healthcare, a claims clearing house, affected 192.7 million individuals, making it the largest health data breach in United States history. Its parent company told investors the total 2024 cost was $3.09 billion, and its chief executive testified under oath that a $22 million ransom was paid. An American Hospital Association survey found 94 per cent of hospitals reporting financial impact and 74 per cent reporting direct impact on patient care — from an incident at a payments intermediary, not at a hospital.

That is the structural finding that matters most, and it recurs. Australia’s Medibank breach in October 2022 exposed data on 9.7 million customers, including health claims information for around half a million of them; the company refused to pay. The June 2024 attack on Synnovis, a pathology services partnership, delayed more than 11,000 appointments across London hospitals and prompted a national appeal for O-type blood; a patient safety investigation at King’s College Hospital later found the attack to have been a contributory factor in one patient’s unexpected death. Ireland’s health service was shut down for over four months in 2021, at a cost of €51 million in that year alone and an estimated €657 million to implement the review recommendations that followed.

In most of the largest health breaches on record, the data was taken from somewhere other than a hospital. Insurers, clearing houses, pathology partners, pharmacy chains.

This has a direct implication for how a unified patient view should be scoped. The argument that a national record concentrates risk is correct as far as it goes, but the measured risk has concentrated in the commercial layer around the clinical system rather than inside it. A programme that hardens the national record while leaving its supply chain unexamined has secured the wrong perimeter.

Two further things in the public record are worth putting side by side. IBM’s annual breach cost study has placed healthcare as the most expensive sector for well over a decade, but the figure has fallen two years running, from $9.77 million in 2024 to $7.42 million in 2025 to $6.64 million in 2026. And the FBI’s complaint data has named healthcare the most-targeted critical infrastructure sector in both 2024 and 2025. Most targeted, most expensive, and improving. All three are true, and the third one is the only one that tells you whether the investment is working.

The policy consequences have also been real rather than rhetorical. Singapore’s commission of inquiry into the 2018 SingHealth breach, which affected 1.5 million patients and specifically targeted the Prime Minister, concluded that the attacker’s success “was not inevitable” and issued sixteen recommendations; the regulator fined the two responsible bodies a combined S$1 million. Australia’s response to Medibank and the Optus breach fed directly into the Privacy and Other Legislation Amendment Act 2024 and a mandatory ransomware payment reporting obligation that commenced in 2025. The regulator’s civil penalty proceedings against Medibank, filed in June 2024, have not yet been decided, and nothing should be assumed about their outcome.

Does it save money? The evidence, in order

This is where the reader deserves the least comfortable section, because the gap between what was promised and what has been measured is the largest in the field, and it has a precise documentary history.

In 2005, a RAND study in Health Affairs projected that widespread adoption of health information technology could save more than $81 billion a year in the United States. It was explicitly a model, built on assumptions about adoption and effectiveness. It was also quoted for the next decade as though it were a finding, and it did substantial work in justifying the roughly $35 billion of federal incentives that followed.

In 2013, RAND published a reassessment in the same journal. Its authors wrote that the promised savings “hasn’t been achieved”, noted that annual United States health spending had risen by around $800 billion over the intervening seven years, and attributed the shortfall in significant part to systems that were “neither interoperable nor easy to use”.

The independent literature since has been consistent in a way that is easy to state and hard to accept. An analysis in the Journal of Health Economics in 2014 found health IT adoption associated with a 1.3 per cent increase in billed charges and no evidence of cost savings even five years after adoption. A systematic review of health information exchange published in 2015 identified 34 studies, of which only two were randomised trials — and both found no benefit; the strength of evidence was rated low. A 2021 survey of both the medical and economics literatures concluded that health IT does improve clinical quality and can lower costs, but that the effects are “modest so far”, take years to appear, vary enormously, and that costs tend to rise during early adoption.

Set against that, three findings are solid enough to build on.

●        Prescribing safety. A 2014 meta-analysis found computerised order entry associated with roughly half as many preventable adverse drug events and medication errors as paper ordering. A more sceptical 2024 rapid review for a United States federal agency qualified this considerably — decision support in addition to order entry did not reach statistical significance against usual care, and alert override rates ran between 46 and 96 per cent — but the direction of the evidence on prescribing is the most favourable in the field.

●        Duplicate testing. The best-supported cost-side finding is reduced repeat imaging and pathology in emergency departments specifically. In ambulatory settings the results are mixed, and one study found an increase. This is exactly the opening scene of this article, and it is the one place where the duplication argument has measured support.

●        Patient access. Tens of millions of people use national health apps where they exist. This is not a cost saving and should not be sold as one. It is a straightforward improvement in how people experience a system they fund.

And one finding deserves its own paragraph, because it reframes the whole debate. Electronic records demonstrably increase clinician documentation time: a 2016 time-and-motion study found physicians spending nearly two hours on records and desk work for every hour of direct patient contact, and a 2017 log study found just under six hours a day in the record, of which over an hour fell after clinic hours. But a 2021 comparison of 371 health systems using the same vendor’s software found United States clinicians spending about 50 per cent more time in it than clinicians outside the United States — 90 minutes a day against 59. The median American clinician was at roughly the ninety-ninth percentile of everyone else.

The burden is not a property of digital records. It is a property of what a given country requires clinicians to document, and why. That makes it a policy problem, which means it is fixable.

How the money flows, and why that decides what gets built

Connected health is usually discussed as a technology programme. It behaves much more like a payment reform, because the payment model determines which behaviours the technology is built to support.

Three mechanisms dominate, and each has an honest case and an honest cost.

The standard framing, as set out in OECD payment-reform work and in Australia’s own mid-term review of the National Health Reform Agreement.

Australia runs a deliberate mix. In 2021 to 2022, public hospital funding comprised about $50.0 billion of activity-based funding against the National Efficient Price, $8.7 billion of block funding against the National Efficient Cost, and around $0.5 billion of other payments. An independent authority sets the prices annually. There is a safety and quality adjustment that reduces funding where outcomes are poor, and a conditional payment mechanism tied to timely data submission — which is, in effect, money attached to interoperability.

The mid-term review of the National Health Reform Agreement, published in October 2023, was candid about the trade-off. It found that activity-based funding’s focus on acute activity “has detracted from allocating funds efficiently across the whole system” and “tends to drive care toward inpatient settings”, and that it “has responded to demand rather than shaping it”. It also recommended keeping it, as the foundation of transparency and comparability. Both judgments are correct, and that tension is the actual policy problem.

The obvious reply is value-based care: pay for outcomes rather than activity. The evidence there deserves a close look, because it is the clearest illustration available of how hard measurement is in this field.

Medicare’s shared savings programme in the United States reported gross savings of $6.5 billion for 2024, of which participating organisations kept $4.1 billion, leaving $2.4 billion net to government — a record. The independent literature has been less settled. A congressional advisory body concluded in 2019 that savings were “generally modest” and depended entirely on whether bonus payments came to less than the spending reductions. A 2023 analysis in JAMA Health Forum reported substantial net savings, was then retracted and replaced after an analytical error was identified, and the corrected version reported net losses of between $775 million and $2.1 billion. A further study publicised in early 2026, extending the window and using final rather than early performance data, reversed that again and reported net savings of between $4.3 billion and $13.4 billion.

Whether the largest value-based care programme in the world has saved money has been answered no, then yes, in the same journal, within three years, depending on the time window and the comparison group. That is not a scandal. It is what honest measurement of a complex intervention looks like, and it should calibrate how any digital health business case is read.

The British experience with pay-for-performance in primary care is the cleanest null result available. The Quality and Outcomes Framework was the largest such scheme in the world. A 2016 study in The Lancet compared United Kingdom mortality against a synthetic control drawn from 27 high-income countries that had not adopted large-scale pay-for-performance, and found no significant change in mortality for any targeted condition. By the 2025 to 2026 general practice contract, the framework had been substantially dismantled: 32 indicators worth around £298 million retired, roughly £100 million of it removed outright and reinvested, the rest redistributed into cardiovascular prevention. The stated reason was reducing bureaucracy.

None of this means outcome-based payment is a mistake. It means that measuring whether it worked is as difficult as the thing itself, and that any programme claiming savings before an independent evaluation exists is making a forecast, not a report.

Challenges and opportunities, side by side

What this means for a smart society

Health is where the smart society stops being a conference theme and starts having consequences, because health data is the most sensitive category most people will ever generate and health systems are the largest single thing most governments operate.

The positive case is strong and should be stated without hedging. A system that knows what medication a confused patient is on, that does not repeat a scan done eighteen months ago in another state, that can tell a planner where the gaps in cancer screening actually are, and that lets a person read their own record on their own phone, is a better system than the one we have. The gains are not speculative; they are documented, if narrower than the brochures. And the socio-economic return compounds in the places that are easiest to overlook — the rural patient who does not have to repeat their history to every new locum, the carer who can see the discharge summary, the public health response that can be aimed rather than broadcast.

But the distinctive risk of a smart society is not surveillance in the abstract. It is that ordinary politics encodes its trade-offs in legislation, which can be amended and repealed, while a smart society encodes them in standards, procurement contracts, data models and eligibility logic — which are considerably harder to reverse and usually invisible to the people living under them. Taiwan’s Constitutional Court spent three years ordering a legislature to create an opt-out right that would have cost almost nothing to design in at the start. England’s most contested data contract contains a break clause, and the existence of that clause is doing more work for public trust than any assurance in the announcement did.

A connected health system is a set of political judgments about privacy, autonomy and the public interest, cast in infrastructure. The judgments are legitimate. Casting them where they cannot be revisited is not.

Which is why the most useful question to ask of any connected health programme is not whether it is ambitious, or secure, or cost-effective. It is whether its builders have left themselves a way to be wrong about it.

Seven tests for a connected health programme

1.  Which of the three architectures is this, in plain words? Central store, federated index, or document exchange. If the name promises a single record and the design does not build one, say so at the outset rather than in a technical annex.

2.  Can a patient see who has looked at their record, without asking anyone’s permission? Denmark can answer yes. Most cannot. The logs almost always exist; the decision to show them is what is missing.

3.  Does an opt-out exist for secondary use, and was it designed in or litigated in? Taiwan’s court answered this question retrospectively and expensively. It is far cheaper as a design requirement than as a judgment.

4.  Is the supply chain in scope? Most large health breaches on record have taken data from insurers, clearing houses and pathology partners rather than from hospital clinical systems. A perimeter drawn around the national record alone is drawn around the wrong thing.

5.  Is the savings claim a measurement or a model? Both are legitimate; only one is evidence. The 2005 projection that became a decade of quoted fact is the cautionary case, and the correction came from the same institution that made it.

6.  Does it give clinicians time back, or take more? This is the best-evidenced harm in the field and the clearest near-term opportunity. If a programme cannot say which direction it moves clinician hours, it has not been designed around its primary users.

7.  What does reversing this look like in five years? Break clauses, sunset provisions, exportable data, more than one viable supplier. If there is no answer, the programme has encoded a political judgment somewhere it cannot be revisited — and that, not the technology, is the thing worth objecting to.

The woman in the opening scene does not need a national data strategy. She needs the registrar to know what she is taking. Everything above is about the difference between a programme that delivers that, and one that delivers a platform.

A note on sources

This article draws exclusively on publicly available material: official programme reporting, legislation, regulators, parliamentary and inquiry documents, court judgments, peer-reviewed literature and mainstream reporting. It contains no confidential, internal or privileged information about any organisation, programme or jurisdiction, and discusses security incidents only as publicly reported events and their policy consequences — not their technical mechanics. Views are the author’s own.

Sources

1. Australian Digital Health Agency, National Healthcare Interoperability Plan 2023–2028: Connecting Australian Healthcare — ten principles, five priority areas. https://www.digitalhealth.gov.au/healthcare-providers/initiatives-and-programs/national-healthcare-interoperability-plan

2. Health Legislation Amendment (Modernising My Health Record — Sharing by Default) Act 2025 (Cth) — Royal Assent 14 February 2025; sharing-by-default commenced 1 July 2026, beginning with pathology and diagnostic imaging. https://www.legislation.gov.au/C2025A00004/asmade

3. Australian Digital Health Agency annual reporting — consumer views up 433 per cent to more than three million weekly; provider uploads up 105 per cent to more than five million weekly; indicative savings of $146.3 million over two years. The savings figure is the programme’s own estimate. https://www.digitalhealth.gov.au/about-us/corporate-publications

4. sundhed.dk — Danish national health portal; patient-visible access logs, automatic notification where no treatment relationship exists, bi-annual random audits. https://www.sundhed.dk/borger/service/om-sundheddk/

5. Estonia, Health Information System (EHIS) and the X-Road data exchange layer. On the blockchain claim: KSI is hash-linked timestamping and predates Bitcoin; the Nordic Institute for Interoperability Solutions has stated X-Road is not blockchain-based. https://www.niis.org/blog/2021/2/11/is-x-road-a-blockchain

6. England: care.data cancelled July 2016; the GP Data for Planning and Research programme announced 2021, with more than 1.3 million opt-outs recorded in a single month and the national opt-out rate rising from 2.77 to 4.97 per cent. https://www.bmj.com/content/374/bmj.n1958

7. NHS England, Fit for the Future: 10 Year Health Plan for England, published 3 July 2025 — the Single Patient Record commitment, promised via the NHS App by 2028. https://www.gov.uk/government/publications/10-year-health-plan-for-england-fit-for-the-future

8. NHS App usage — more than 39 million registered users; 62.3 million logins in November 2025. https://digital.nhs.uk/services/nhs-app/nhs-app-dashboards

9. NHS England Federated Data Platform — contract awarded November 2023, seven years, approximately £330 million, with a break clause in February 2027. https://www.england.nhs.uk/digitaltechnology/digitising-connecting-and-transforming-health-and-care/federated-data-platform/

10. United States: HITECH Act incentives (approximately $35 billion in the standard framing); 96 per cent hospital adoption of certified electronic records by 2017; ONC data showing four-domain interoperability engagement flat between 2022 and 2023, with 71 per cent of hospitals having access to outside information but only 42 per cent of clinicians often using it. https://www.healthit.gov/data/quickstats

11. TEFCA — cumulative document exchange rising from approximately 10 million before 2025 to 464 million by the end of 2025. Information blocking rules effective April 2021; first enforcement notices February 2026. https://rce.sequoiaproject.org/

12. Singapore: Public Report of the Committee of Inquiry into the SingHealth cyber attack (2019), 1.5 million patients affected, sixteen recommendations, and the finding that the attacker’s success “was not inevitable”. Personal Data Protection Commission penalties of S$750,000 and S$250,000. https://www.mci.gov.sg/files/Public%20Report%20of%20the%20COI.pdf

13. Singapore, Health Information Bill — passed January 2026; NEHR contribution mandatory from early 2027; individual consent not required for contribution under Part 2. https://www.moh.gov.sg/newsroom/health-information-bill

14. Taiwan, Constitutional Court Judgment 111-Hsien-Pan-13 (12 August 2022) — Articles 79 and 80 of the National Health Insurance Act held unconstitutional for want of any opt-out from secondary use; remedial legislation passed 2 December 2025. https://cons.judicial.gov.tw/en/docdata.aspx?fid=100&id=310088

15. India, Ayushman Bharat Digital Mission — ABHA account and linked record counts as reported by the National Health Authority. On the exposure of searchable patient names on the Authority’s own site, see Internet Freedom Foundation. https://internetfreedom.in/

16. Change Healthcare breach, February 2024 — 192.7 million individuals affected; $3.09 billion total 2024 cost reported by UnitedHealth Group; $22 million ransom payment confirmed in sworn congressional testimony, 1 May 2024. American Hospital Association survey: 94 per cent of hospitals reporting financial impact, 74 per cent direct patient care impact. https://www.ocrportal.hhs.gov/ocr/breach/breach_report.jsf

17. Medibank breach, October 2022 — 9.7 million customers; health claims data for approximately 500,000. Office of the Australian Information Commissioner civil penalty proceedings filed 5 June 2024; undecided at the time of writing. https://www.oaic.gov.au/news/media-centre/oaic-takes-civil-penalty-action-against-medibank

18. Synnovis incident, June 2024 — more than 11,000 appointments delayed; national O-type blood appeal 10 June 2024; King’s College Hospital patient safety investigation finding the attack a contributory factor in one patient’s unexpected death; reported financial impact £32.7 million. https://www.bbc.co.uk/news/articles/cvg3z9v5e9no

19. Health Service Executive (Ireland) ransomware incident, May 2021 — recovery over four months; €51 million in 2021 costs; approximately €657 million projected to implement the subsequent review recommendations; no ransom paid. https://www.hse.ie/eng/services/publications/conti-cyber-attack-on-the-hse-full-report.pdf

20. IBM / Ponemon, Cost of a Data Breach — healthcare the highest-cost sector, with the average falling from $9.77 million (2024) to $7.42 million (2025) to $6.64 million (2026). FBI Internet Crime Complaint Center annual reports naming healthcare the most-targeted critical infrastructure sector in 2024 and 2025. https://www.ibm.com/reports/data-breach

21. Privacy and Other Legislation Amendment Act 2024 (Cth), and the mandatory ransomware payment reporting obligation commencing 30 May 2025 under the Cyber Security Act 2024 (Cth). https://www.legislation.gov.au/C2024A00106/asmade

22. Hillestad, R. et al. (2005), “Can Electronic Medical Record Systems Transform Health Care?”, Health Affairs 24(5):1103–1117 — the >$81 billion annual savings projection, explicitly a model. https://www.healthaffairs.org/doi/10.1377/hlthaff.24.5.1103

23. Kellermann, A. L. & Jones, S. S. (2013), “What It Will Take To Achieve The As-Yet-Unfulfilled Promises Of Health Information Technology”, Health Affairs 32(1):63–68 — RAND’s own reassessment. https://www.healthaffairs.org/doi/10.1377/hlthaff.2012.0693

24. Agha, L. (2014), “The effects of health information technology on the costs and quality of medical care”, Journal of Health Economics 34:19–30 — a 1.3 per cent increase in billed charges and no evidence of cost savings even five years after adoption. https://pubmed.ncbi.nlm.nih.gov/24411098/

25. Hersh, W. R. et al. (2015), “Outcomes From Health Information Exchange: Systematic Review and Future Research Needs”, JMIR Medical Informatics 3(4):e39 — 34 studies, two randomised trials, both null; strength of evidence rated low; reduced duplicate imaging and pathology in emergency settings the best-supported finding. https://medinform.jmir.org/2015/4/e39/

26. Bronsoler, A., Doyle, J. & Van Reenen, J. (2021), The Impact of Healthcare IT on Clinical Quality, Productivity and Workers, NBER Working Paper 29218 — effects “modest so far”, costs rising in early adoption. https://www.nber.org/papers/w29218

27. Jones, S. S. et al. (2014), “Health Information Technology: An Updated Systematic Review With a Focus on Meaningful Use”, Annals of Internal Medicine 160(1):48–54 — strong evidence for clinical decision support and computerised order entry; 61 per cent of studies gave no contextual detail. https://www.acpjournals.org/doi/10.7326/M13-1531

28. Nuckols, T. K. et al. (2014), “The effectiveness of computerized order entry at reducing preventable adverse drug events and medication errors in hospital settings”, Systematic Reviews 3:56 — pooled relative risk 0.47 for preventable adverse drug events and 0.46 for medication errors; the authors note weak study designs. https://link.springer.com/article/10.1186/2046-4053-3-56

29. Syrowatka, A. et al. (2024), Computerized Clinical Decision Support To Prevent Medication Errors and Adverse Drug Events, AHRQ Making Healthcare Safer IV — order entry with decision support not statistically significant against usual care; alert override rates of 46 to 96 per cent; burnout among documented harms. https://www.ncbi.nlm.nih.gov/books/NBK600580/

30. Sinsky, C. et al. (2016), “Allocation of Physician Time in Ambulatory Practice”, Annals of Internal Medicine 165(11):753–760 — 27.0 per cent direct clinical face time against 49.2 per cent records and desk work. https://pubmed.ncbi.nlm.nih.gov/27595430/

31. Arndt, B. G. et al. (2017), “Tethered to the EHR”, Annals of Family Medicine 15(5):419–426 — approximately 355 minutes a day in the record, of which 86 minutes after hours. https://www.annfammed.org/content/15/5/419

32. Holmgren, A. J. et al. (2021), “Assessment of Electronic Health Record Use Between US and Non-US Health Systems”, JAMA Internal Medicine 181(2):251–259 — 90.2 against 59.1 minutes a day in the same vendor’s software; the median US clinician at roughly the non-US 99th percentile. https://jamanetwork.com/journals/jamainternalmedicine/fullarticle/2774129

33. Laine, L. T. (2019), Digital Waste? Unintended Consequences of Health Information Technology, IZA Discussion Paper 12275 — Finland’s e-prescribing rollout, with increased repeat prescribing and worse outcomes among younger patients. https://schaeffer.usc.edu/research/digital-waste-unintended-consequences-of-health-information-technology/

34. Australian Government, Mid-Term Review of the National Health Reform Agreement Addendum 2020–2025, 24 October 2023 — activity-based funding’s acute focus, the recommendation to retain it, and the funding split. Prices are determined annually by the Independent Health and Aged Care Pricing Authority. https://www.health.gov.au/resources/publications/mid-term-review-of-the-national-health-reform-agreement-addendum-2020-25-final-report

35. OECD (2016), Better Ways to Pay for Health Care, OECD Health Policy Studies; and OECD Health Working Paper 154 (2023), Innovative providers’ payment models for promoting value-based health systems — the standard framing of activity-based, capitated and block funding trade-offs. https://www.oecd.org/en/publications/better-ways-to-pay-for-health-care_9789264258211-en.html

36. Centers for Medicare & Medicaid Services, Medicare Shared Savings Program 2024 performance year results — $6.5 billion gross savings, $4.1 billion retained by participants, $2.4 billion net. https://www.cms.gov/newsroom/fact-sheets

37. MedPAC (June 2019), Report to Congress, Chapter 6 — savings “generally modest” and contingent on bonus payments being smaller than spending reductions. https://www.medpac.gov/wp-content/uploads/import_data/scrape_files/docs/default-source/reports/jun19_ch6_medpac_reporttocongress_sec.pdf

38. Ryan, A. M. & Markovitz, A. A., “Estimated Savings From the Medicare Shared Savings Program”, JAMA Health Forum 2023 — retracted and replaced; the corrected version reports net losses of $775 million to $2.063 billion for 2013–2021. See the retraction notice. https://pmc.ncbi.nlm.nih.gov/articles/PMC11623111/

39. Bond, A., Khullar, D. et al., “Budgetary Impact of the Medicare Shared Savings Program on Traditional Medicare”, JAMA Health Forum (2026) — extending the window to 2012–2023 and reporting net savings of $4.3 billion to $13.4 billion, reversing the earlier net-loss finding. https://phs.weill.cornell.edu/news/budgetary-impact-medicare-shared-savings-program-traditional-medicare

40. Ryan, A. M., Krinsky, S., Kontopantelis, E. & Doran, T. (2016), “Long-term evidence for the effect of pay-for-performance in primary care on mortality in the UK: a population study”, The Lancet 387:268–274 — no significant mortality change under synthetic control against 27 comparison countries. https://www.thelancet.com/journals/lancet/article/PIIS0140-6736(16)00276-2/fulltext

41. NHS England, Changes to the GP contract in 2025/26 — 32 Quality and Outcomes Framework indicators worth 212 points (approximately £298 million) retired, with the stated aim of reducing bureaucracy. The framework has been substantially reduced rather than formally abolished. https://www.england.nhs.uk/long-read/changes-to-the-gp-contract-in-2025-26/

42. AIHW, Health expenditure Australia 2023–24 — total health spending of A$270.5 billion, approximately 10.1 per cent of GDP. ONS, UK Health Accounts (29 April 2026) — 11.4 per cent provisional for 2025. Note that Singapore’s frequently quoted 2.5 per cent is government health expenditure only; WHO’s comparable total measure is roughly double. https://www.aihw.gov.au/reports/health-welfare-expenditure/health-expenditure-australia-2023-24