In August 2018, the head of the US Centers for Medicare and Medicaid Services, Seema Verma, stood up at a federal interoperability forum and challenged developers to make doctors’ offices “a fax free zone by 2020.”
It is worth sitting with how that sounds today. The fax machine is still a working part of health care in many countries, the United States included. Referral letters, discharge summaries and prior authorisation forms still move through it daily.
The standard explanation is that interoperability is hard engineering. Data formats differ, vocabularies clash, legacy systems are brittle. All true. But the engineering has largely been solved on paper for years. What remains unsolved is the thing nobody puts in the architecture diagram: nobody who controls the data is paid to share it well.
The numbers look better than the experience
On the surface, progress is real. The US Office of the National Coordinator for Health IT reported that by 2023, 70% of non-federal acute care hospitals were at least sometimes engaging in all four domains of interoperability: sending, receiving, finding and integrating patient data. In 2018 that figure was 46%.
Look one layer down, though, and the picture changes. Only 43% of hospitals reported doing all four routinely. The gap between “sometimes” and “routinely” is the gap between a capability that exists and one that clinicians can rely on at three in the morning. ONC itself has argued that “sometimes” is not good enough.
And “integrate” is where the value actually lies. A PDF of an outside discharge summary sitting in a document tab is technically interoperable. It is also, in practice, something a busy clinician may never open. Data that arrives but cannot be reconciled with the local medication list or problem list does not change care. It just adds another place to look.
The uncomfortable incentive underneath
Every party in the chain has a rational reason to under-invest.
Hospitals and health systems compete for patients. A complete, portable record makes it easier for a patient, or a referring doctor, to go elsewhere. Nobody writes that in a strategy document, but it shapes how much effort goes into making outbound data clean and usable.
EHR vendors are paid to serve their customer, the health system, not the competitor across town. Connections between installations of the same vendor tend to work far better than connections across vendors, which is a commercial feature as much as a technical one.
Clinicians carry the cost of poor exchange in time and repeated tests, but have little say in procurement. Patients bear the risk of missing information and have the least leverage of all.
Legislators have noticed. The 21st Century Cures Act of 2016 made “information blocking” by providers, developers and exchanges a regulatory offence, with rules phased in over the following years. That is a meaningful signal that policymakers see this as a behaviour problem rather than a standards problem.
Standards are necessary, not sufficient
The technical toolkit has matured. HL7’s FHIR standard gives developers a modern, web-style way to request specific pieces of data. In the United States, the Trusted Exchange Framework and Common Agreement went live in December 2023, when the first Qualified Health Information Networks were designated and data began flowing between them.
These are real advances. But a standard only defines how data can move. It does not make anyone move it, check it, or map it into a form the receiving clinician can use. Much of the remaining work is unglamorous: consistent patient matching, cleaning up local codes, deduplicating medication lists, deciding who is responsible when an incoming allergy conflicts with a local one.
That work does not show up in product launches. It shows up as a better referral a year later.
There is a trap here for low- and middle-income countries building digital systems now. It is tempting to buy a national platform and assume interoperability comes with it. It does not. The contracts signed today, including who owns the data, what export formats are guaranteed and what it costs to connect a new system, will decide whether the country ends up with one record per patient or a dozen islands. Writing those terms in at procurement is cheap. Retrofitting them later is not.
What actually moves the needle
Countries and systems that have made progress tend to combine several things rather than rely on any single standard:
- Pay for exchange that is used, not just possible. Incentives tied to data being integrated into the receiving record, not merely transmitted, change behaviour.
- Make blocking costly. Rules against information blocking only work if they are enforced and penalties are material.
- Invest in identity. Without reliable patient matching, even perfect data exchange produces errors or duplicates.
- Give clinicians a seat in procurement. The people who suffer from poor integration should help decide what “working” means.
- Measure the clinical experience. Survey clinicians on whether outside information was available and usable at the point of care, not just whether a connection exists.
A worked example: one patient, three systems
Consider an older woman with heart failure and diabetes who is admitted through the emergency department of a hospital that is not her usual one.
Her primary care practice uses one EHR. Her cardiologist, at a different health system, uses another. The admitting hospital uses a third. She takes eleven medications, two of which were changed by the cardiologist last month.
In a well-functioning system, the emergency physician queries a network, retrieves her current medication list, recent echocardiogram and last set of kidney results, and sees them in a structured form that can be reconciled against what she brought in her handbag. The discharge summary then flows back to both her GP and her cardiologist, landing in a place they will actually see.
In the system many clinicians describe, the emergency physician receives a bundle of documents from one source, nothing from the second, and a fax from the third several hours later. The medication list in the documents is from before the cardiologist’s changes. The echocardiogram report is a scanned PDF. The admitting team rebuilds the medication list from scratch by asking the patient and her daughter. The discharge summary is sent to the GP, but it goes into a general documents folder and is not seen until the patient returns a month later with a problem that the summary would have anticipated.
Every system in that story may be certified as interoperable. Every connection may technically exist. The failure lies in the last mile: whether the right information arrives, in a usable form, to the right person, in time. That last mile is where most of the harm happens and where least of the investment goes.
The mechanism: why “sometimes” persists
The distance between hospitals that can exchange data sometimes and those that do it routinely is not mainly about technology. It reflects the cost structure of exchange.
Setting up a connection is a one-off project with a visible budget. Keeping it working day to day is an operating cost that nobody specifically funds. Interfaces break when either side upgrades software. Patient matching errors require someone to resolve them. Codes drift as local practices change. Clinicians need training to find incoming data. Each of these jobs falls to informatics teams that are usually stretched and whose performance is measured by other things.
So connections tend to decay into “sometimes.” They work for the patients and data types that happen to be easy and fail quietly for the rest. Nobody is accountable for the difference, because nobody is paid for routine success.
The strongest counterargument
Some would argue that progress is real and the pessimism is overdone. National networks now move very large volumes of data. Patients in many countries can download their own records to their phones. Standards like FHIR have given developers a common language that did not exist a decade ago. Regulation against information blocking has changed vendor behaviour. On this view, the remaining problems are the long tail of any large infrastructure project and will be solved with time.
This argument partly holds. The direction of travel is positive, and it would be wrong to ignore how much has changed. Exchange that was impossible in 2015 is routine for some organisations today.
But “given time” is not a strategy. The history of interoperability shows that progress happens where incentives and enforcement line up, and stalls where they do not. Long tails can last decades. The question is not whether things will improve on their own but how fast, and for whom. Rural hospitals, small practices, long-term care facilities and behavioural health providers have often been left out of earlier waves of investment, and they tend to care for patients with complex needs who move between settings the most.
Second-order effects to watch
More data, not more insight. As exchange volumes grow, clinicians can be buried in incoming documents. Without good summarisation and reconciliation tools, better interoperability can increase workload rather than reduce it.
Privacy trade-offs. The easier data moves, the more important consent, access controls and audit become. A scandal involving misuse of exchanged data could set back public trust quickly.
New gatekeepers. As networks consolidate, the organisations that run them gain influence over who can connect and on what terms. Interoperability can create new chokepoints even as it removes old ones.
AI dependence. AI tools that summarise records depend entirely on the completeness of the data they receive. Poor interoperability will limit what they can do, and may lead them to produce confident summaries of incomplete records.
What this means for different players
Ministries and regulators should shift from certifying capabilities to measuring use: how often data is retrieved, integrated and viewed at the point of care. Public reporting of those measures by organisation would create pressure that certification does not.
Hospital leaders should fund the operating side of exchange, not just the projects that set it up. A named owner for data quality and connection reliability, with a budget, is a small investment relative to the cost of repeated tests and medication errors.
Vendors should be expected to make exchange across competitors as smooth as exchange within their own customer base. Where regulation allows, contracts should include performance commitments on external exchange.
Clinicians can help by reporting when outside data is missing or unusable, so that failures become visible. Informal workarounds, such as phoning a colleague or asking the patient, keep care moving but hide the size of the problem.
Funders working in lower-income countries should prioritise shared infrastructure, such as patient identifiers, facility registries and terminology services, over standalone applications. Those shared foundations make every later project cheaper and more likely to connect.
The takeaway
The fax machine did not survive because engineers could not replace it. It survived because it works across every organisational boundary, requires no agreement between competitors, and puts the burden of reading and re-typing on the receiver, who has no power to refuse it.
Interoperability will be solved when sharing data well becomes the easiest way for health systems and vendors to get paid. Until then, expect more standards and more pilots, and expect the fax to outlast a few more deadlines.
What is the single piece of outside information you most often wish you had at the point of care, and why doesn’t it reach you? We’d like to hear from clinicians and IT teams alike.
