An API instead of a portal: the 2015 meaningful use proposals drew health IT’s next architecture

The Stage 3 and 2015 Edition proposals, released March 20, 2015, were debated for their thresholds. Their lasting content was four architectural layers: API access, a common data set, exchange that incorporates, and certification that discloses. The layers outlived the program.

Looking Back revisits past weeks one at a time, up to the most recent, and asks what it meant for the banks, health systems and public agencies we serve. The series was written in 2026, with the benefit of hindsight, and each article says which week it looks back on.

Two-minute summaryThe dek and the first sentence of each section, quoted from the article

The Stage 3 and 2015 Edition proposals, released March 20, 2015, were debated for their thresholds. Their lasting content was four architectural layers: API access, a common data set, exchange that incorporates, and certification that discloses. The layers outlived the program.

  1. The week On Friday, March 20, 2015, the Department of Health and Human Services released two proposed rules together.
  2. The access layer: an API in place of the portal Under Stage 2, patient access meant a portal.
  3. The data layer: a common set every interface had to carry An API is only as useful as the data it returns.
  4. The exchange layer: from sending records to incorporating them Stage 2 had required providers to send summaries of care electronically at transitions.
  5. The governance layer: who decided which software counted The fourth layer was the certification program itself.
  6. The strongest objections, and what they missed The March 2015 proposals drew serious criticism.
  7. Reading the architecture for a health system today The four layers described in March 2015 still organize a health system’s interoperability architecture.
  8. With hindsight CMS and ONC finalized both rules on October 16, 2015.

The week

On Friday, March 20, 2015, the Department of Health and Human Services released two proposed rules together. The Centers for Medicare & Medicaid Services proposed Stage 3 of the Medicare and Medicaid EHR Incentive Programs, the program widely known as meaningful use. The Office of the National Coordinator for Health Information Technology proposed a new edition of certification criteria for health IT, the 2015 Edition. The announcement said more than 433,000 eligible professionals and hospitals had received incentive payments since the programs began in 2011 [7].

The week of March 23 was the first full week in which hospitals, physician groups and vendors worked through the proposals. Together they ran to more than 180 pages of the Federal Register. Both were published there the following Monday, March 30, with comments due by May 29, 2015 [1] [2].

The CMS proposal described Stage 3 as expected to be the final stage of meaningful use. After an optional year in 2017, all providers would report on a single Stage 3 definition from 2018, whatever their prior stage. Every provider would report on a full calendar year, with a limited exception for some first-time Medicaid participants. The rule proposed eight objectives. They were protecting patient health information, electronic prescribing, clinical decision support and computerized provider order entry. The others were patient electronic access, coordination of care through patient engagement, health information exchange, and public health and clinical data registry reporting [1].

The patient access objective contained a new option. Providers would give more than 80 percent of patients access to their information within 24 hours of its availability to the provider. They could do it through the familiar view, download and transmit function, or through an ONC-certified application programming interface that third-party applications or devices could use. CMS wrote that a provider using the API option would not be required to separately purchase or implement a patient portal [1].

The patient engagement objective proposed three measures, of which providers would have to meet two. One counted patients who actively engaged with their electronic record, either through view, download and transmit or through an API. The health information exchange objective proposed that providers send an electronic summary of care for more than 50 percent of transitions and referrals. They would also incorporate summaries received from outside for more than 40 percent of incoming transitions and new patients. For more than 80 percent, they would reconcile medications, medication allergies and problems [1].

ONC’s companion proposal set out the technology providers would need. It proposed a Common Clinical Data Set, a defined list of data elements with specified standards. It proposed an “application access” criterion: health IT would have to demonstrate an API that responded to requests for any category of the Common Clinical Data Set, or all of it at once. ONC specified outcomes rather than a single standard. The API needed a trusted connection with registration, authorization and logging, a way to select a patient, and responses in XML or JSON, or a C-CDA document for the full set [2].

ONC said it intended the criterion to allow, but not require, the use of HL7’s FHIR standard. It also proposed that developers publish technical documentation and terms of use for the API. Certification bodies would make a link to them available, so outside developers could see the requirements before building [2].

The ONC proposal went beyond meaningful use. It proposed to make the certification program open to more types of health IT and more care settings, including long-term and post-acute care, behavioral health and pediatrics. It replaced references to “EHR technology” with “health IT.” It proposed more rigorous testing of C-CDA documents and randomized surveillance of certified products in the field. It also proposed to expand the disclosures developers already had to make about the costs and limitations of certified capabilities [2].

The timing put the proposals beside other pressure on the program. Many providers were struggling with Stage 2, and CMS said it would address changes for 2015 through 2017 in separate rulemaking. Congress was also moving on legislation to replace the Medicare physician payment formula. The House passed that bill, the Medicare Access and CHIP Reauthorization Act, on Thursday, March 26. It would reshape the incentive program for clinicians within weeks.

Commentary on the March 2015 proposals focused on thresholds and burden. Physician groups argued that the percentages were too high and the reporting too heavy. Hospitals worried about timelines. Those concerns were reasonable, and many of the thresholds changed in the final rules and in the years after.

The more durable content was architectural. Read together, the two proposals described how health IT should be layered. They defined who could reach a patient’s data and how, what the data was, how it moved between providers, and how the market for the software would be governed. Most of the program’s thresholds are gone. The layers remain, in revised form, in the rules health systems follow in 2026.

We take the proposals in four layers, from the patient outward. Together they cover the architecture the rules described:

  • The access layer: how patients and the applications they chose would reach their data, and the proposal that an API could stand in for a portal.
  • The data layer: the Common Clinical Data Set, which defined the payload every interface had to carry.
  • The exchange layer: how records moved between providers at transitions of care, and the new expectation that incoming data be incorporated and reconciled.
  • The governance layer: how the certification program would decide which software counted, and what developers had to disclose.

After the four layers, we test the reading against its strongest objections and set out what it means for a health system’s architecture today.

The access layer: an API in place of the portal

Under Stage 2, patient access meant a portal. Providers bought or built a website where patients could view, download and transmit their records. Most did so through their EHR vendor’s portal, because it was certified and connected to the record. The portal was the front door, and the EHR vendor owned it.

What the proposal changed

Stage 3 proposed a second door. A provider could meet the access measure by giving patients access to an ONC-certified API that third-party applications could use. CMS stated the consequence directly: a provider using the API option would not need to buy or implement a separate portal [1].

That sentence moved the patient’s point of access out of the provider’s control. Under a portal model, the provider chose the interface, the features and the vendor. Under an API model, the patient could choose an application, and the provider’s system would answer the application’s requests. The provider still controlled authentication and the data, but not the experience.

Why it mattered for architecture

For a health system’s technology leaders, the change implied several new components. The organization would need an API gateway or equivalent layer in front of the EHR, able to register applications, authorize them on behalf of patients and log every request. It would need an identity process that linked the patient’s credentials to the right record. It would need policies on which applications to allow and how to respond if an application misused data.

ONC’s proposed criterion described exactly those functions. The API had to establish a trusted connection, let an application register and be authorized, and log all interactions. It had to let the application find the patient’s record and then request data by category or in full [2]. Those were the building blocks of what later became standard patient-facing API architecture.

Identity at the front door

An API that answers on a patient’s behalf is only as safe as the process that proves the patient is who they claim to be. A portal had the same need, but its identity process sat inside one vendor’s product. With an API, the provider had to issue credentials that outside applications could use through an authorization step. It also had to decide what level of proofing was enough before a person could connect an application to a record.

Weak proofing let the wrong person connect to a record. Strong proofing, such as in-person verification, kept patients out. Most health systems settled on a mix: online proofing against existing records for most patients, with in-person or assisted routes for those who failed. The proxy problem added a layer. Parents of minors, caregivers of older adults and legal representatives all needed access on someone else’s behalf, with rules that varied by age and state law.

Patient matching was the other half of identity. ONC’s proposal listed enhanced patient matching among its aims [2]. A request through an API, or a query from another provider, needed to land on the right record. Health systems with duplicate records, or with patients registered under different name spellings, found those errors exposed when outside parties began to ask for data. Cleaning the master patient index became an interoperability project.

The provider’s responsibility

The proposal raised hard questions about responsibility that the 2015 documents only began to address. HIPAA gives patients a right of access and lets them direct their information to a third party. A provider that sends data to an application the patient chose is generally acting on the patient’s right. The application, if it is not a covered entity or business associate, is often outside HIPAA altogether.

That gap worried privacy officers in 2015 and still does. A patient could connect an application with weak security or aggressive data practices. The provider could not refuse simply because it disliked the application’s business model. Health systems had to learn to distinguish legitimate security concerns, which could justify limits, from preferences, which could not. Organizations should check with counsel on specific situations, because the boundaries were later refined by regulation.

What the access layer cost

For a large health system with a major EHR, the API layer would arrive largely through the vendor, once certified products existed. The cost lay in configuration, identity proofing, application review processes, monitoring and patient support. For a small practice, the API would come with its certified EHR, and the practice would rely on the vendor’s default settings. Few small practices had the staff to review applications or monitor API traffic.

The economics of the portal also shifted. Portals had been a selling point for EHR vendors and a cost for providers. If patients could use any application, the portal became one option among several. Providers that had invested in portal engagement campaigns faced a choice between promoting their own portal and supporting outside applications. Most did both, which meant supporting two access paths with different identity, support and security needs.

Measuring patient engagement through an API

The proposed engagement measure counted patients who actively engaged with their record, through view, download and transmit or through an API [1]. Counting API use required the provider to know when a patient’s chosen application had retrieved data. That required logging that tied each API call to a patient and a provider, and reporting that rolled those logs into a measure. A measure that looked simple in the rule implied a data pipeline in the provider’s systems.

It also put part of the provider’s performance in the hands of application developers. If no useful applications existed, patients would not use the API, and the provider could not meet the measure that way. That dependency was one of the strongest criticisms of the proposal, as the objections section below discusses.

The data layer: a common set every interface had to carry

An API is only as useful as the data it returns. ONC’s answer was the Common Clinical Data Set, a list of data elements with the standards each had to use. It included patient name, sex, date of birth, race and ethnicity, preferred language, smoking status, problems, medications, medication allergies, laboratory tests and results, and vital signs. It also covered immunizations, the assessment and plan of treatment, goals and health concerns, among other elements [2].

Why a defined payload mattered

Before 2015, each certification criterion carried its own list of data. The transition of care document, the portal export and the data portability function each had slightly different requirements. A developer building to one did not automatically satisfy another. For a provider, that meant the same patient’s medication list could look different depending on which function produced it.

The Common Clinical Data Set gave every function the same core. The API had to return it. The C-CDA documents used at transitions had to contain it. The data portability function had to export it. In architectural terms, it was a canonical data model at the boundary of the EHR: whatever happened inside, this was what came out.

Standards inside the set

The set also tied each element to a vocabulary. Problems were to be coded in SNOMED CT. Medications used RxNorm. Laboratory tests used LOINC. Those standards were not new, but naming them inside one definition meant that every outbound interface had to produce coded data. Free text that had been acceptable in a document became a gap.

For health systems, that requirement exposed the state of their internal data. Problem lists maintained as free text, local medication codes and laboratory tests without LOINC mappings all needed work. The work fell on clinical informatics teams and laboratory staff more than on IT. Mapping laboratory tests to LOINC, for instance, required knowledge of each test’s method and specimen as well as its name.

The limits of the set

The Common Clinical Data Set was a floor. ONC said additional data was permitted and encouraged in API responses [2]. In practice, most implementations returned what certification required and little more. Clinical notes, imaging, and much of the record that clinicians actually read were outside the set. A patient’s application could show medications and lab results, but not the discharge summary that explained them.

That limit shaped what third-party applications could do. Applications built on the 2015 data set tended toward medication lists, results and immunizations. More ambitious uses, such as decision support using notes, needed data the set did not guarantee. The later move from the Common Clinical Data Set to the United States Core Data for Interoperability, with versioned expansions, was a direct response.

Provenance and trust in incoming data

A common data set also travels inward. When a provider receives a medication list from another organization, it needs to know where the list came from, when it was last updated and who recorded it. Without that provenance, clinicians cannot judge whether to trust it. Data that looks authoritative because it is coded and structured can still be stale.

The 2015 proposals said little about provenance, and the gap showed in practice. Records assembled from several sources could display a medication that one source had stopped months earlier. A health system’s design for incoming data needed to keep source and date visible to the clinician, rather than blending outside data into the local record without attribution.

Ownership of the data layer

A canonical data model needs an owner inside the organization. Someone must decide how local codes map to standard ones, how mappings are maintained when new tests or drugs appear, and how errors found by outside users get corrected. In many health systems in 2015, those decisions were spread among the laboratory, pharmacy, health information management and IT. The data layer proposed in March 2015 made the absence of an owner visible, because errors now appeared outside the organization, in other providers’ systems and patients’ applications.

A practical model assigned terminology ownership to a clinical informatics function, with defined service levels for new mappings and a process for receiving and fixing reports of bad data. It measured the share of outbound data that was coded to standard vocabularies, by data type. Those measures turned an abstract interoperability goal into work that could be staffed and tracked.

The exchange layer: from sending records to incorporating them

Stage 2 had required providers to send summaries of care electronically at transitions. Stage 3 proposed to measure what happened at the other end. The receiving provider would have to incorporate an electronic summary from outside for more than 40 percent of incoming transitions and new patients. For more than 80 percent, it would reconcile medications, medication allergies and problems [1].

Push, pull and incorporate

The objective’s wording contained three verbs. A provider sends a summary when transitioning a patient, retrieves a summary on first encounter with a new patient, and incorporates summaries from other providers [1]. Each implied a different integration pattern.

Sending was a push: the referring provider created a C-CDA document and transmitted it, often through Direct secure messaging, to a known recipient. Retrieving was a pull: the receiving provider queried a network or exchange for records about a patient it had never seen. Incorporating was internal: the receiving system took discrete data from the incoming document and merged it into its own record, with clinician review.

Most health systems in 2015 had invested in the first pattern. Few had working query-based exchange at scale outside their own vendor’s network or a regional health information exchange. Fewer still had reliable incorporation, because incoming documents varied widely in quality. The proposal effectively asked providers to build the second and third patterns.

Why incorporation was the hard part

Incorporation depended on the quality of documents created by other organizations’ systems. A C-CDA that met the standard on paper could still be hard to use: sections empty, codes missing, medications listed as text. ONC’s proposal to test C-CDA creation more rigorously was aimed at exactly that problem [2]. Better testing on the sending side would make incorporation possible on the receiving side.

Reconciliation added clinical workflow. A clinician had to compare the incoming medication list with the local one, decide which entries were current and record the result. That took time in every visit with a new patient or after every admission. A system that presented incoming data clearly, flagged duplicates and conflicts, and let the clinician accept or reject items quickly made the measure feasible. A system that dumped a document into the chart made it a burden.

Architecture choices for exchange

For a health system, the exchange layer posed several choices. It could rely on its EHR vendor’s network for exchange with other customers of the same vendor. It could join a regional health information exchange for exchange with local providers on other systems. It could connect to national networks that were forming at the time. Each choice had different costs, governance and coverage.

The proposal did not pick among them. It set outcomes and left the networks to the market. That allowed flexibility but meant that a provider’s ability to meet the measure depended on whether the organizations its patients came from were on a compatible network. A rural hospital whose referral partners used different vendors and no common exchange faced a harder task than an urban system whose partners shared its vendor.

Who paid for the networks

Exchange had a business model problem that the proposals did not settle. Regional health information exchanges charged participation fees. Direct messaging required accounts with a service provider, often bundled with the EHR. Vendor networks were included in some contracts and priced separately in others. A provider asked to exchange more records with more partners faced costs that varied by partner and network.

Those costs fell unevenly. A large system could absorb them and negotiate. A small practice might find that reaching a particular hospital required a fee it could not justify for a handful of referrals a year. Interface fees charged by vendors to connect with outside systems became a recurring complaint, and later a focus of the information blocking rules. For a CIO in 2015, the practical step was to inventory exchange costs by partner and to raise them in vendor negotiations, rather than accept them as fixed.

Clinical safety in incorporation

Incorporating outside data into the local record carries safety risk. An allergy imported from another system could be wrong. A medication that a patient stopped could reappear as active. A problem recorded for a different patient with a similar name could enter the chart if matching failed. The proposals required reconciliation partly for that reason: a clinician had to review what came in before it became part of the record.

That made the design of the reconciliation screen a patient safety question. Health systems that involved pharmacists and nurses in designing it, and that monitored errors introduced through incorporation, managed the risk better than those that treated it as an interface configuration task.

Differences by type of organization

An academic medical center received patients from many referring providers, on many systems. Its incorporation measure depended on the quality of documents from all of them. An ambulatory network that referred mostly to one affiliated hospital system could build tight integration with that system and meet the measure more easily. Critical access hospitals, often with smaller EHR vendors and limited staff, faced the steepest climb. Behavioral health providers and long-term care facilities, which were not eligible for incentive payments, were often the missing partners in the transitions the measure counted.

ONC’s proposal to open certification to those settings addressed that last gap in principle [2]. Certification did not fund adoption, though. A skilled nursing facility had no incentive payment to buy certified technology, so the exchange partner most needed at hospital discharge was often the least equipped.

The governance layer: who decided which software counted

The fourth layer was the certification program itself. ONC’s proposal changed what the program covered and how it watched products after certification. Those changes mattered to health systems because certification was the main assurance they had that a product did what the rules required.

Opening the program

ONC proposed to make certification available to more types of health IT, including modules that supported settings outside the incentive program. It replaced “EHR” with “health IT” throughout, signaling that certification was no longer only about the core record [2]. That allowed specialized products, such as an API platform or a care coordination tool, to be certified for specific criteria without being a full EHR.

Architecturally, that supported a modular approach. A health system could combine certified modules from different developers, as long as together they met the definition of certified EHR technology. In theory, that reduced dependence on a single vendor. In practice, the integration work of combining modules fell on the provider, and most large systems continued to rely on their primary EHR vendor for most criteria.

Watching products in the field

ONC proposed that certification bodies conduct randomized surveillance of certified products, in addition to reacting to complaints [2]. That addressed a real problem. A product could pass testing in a lab and then behave differently in a customer’s configuration. Surveillance in the field gave providers a way to report that gap and a body with authority to act on it.

For a health system, surveillance had a practical meaning. When a certified function did not work as certified, the provider could report it to the certification body instead of only negotiating with the vendor. Few providers used that route in 2015. Knowing it existed changed the balance in vendor discussions.

Disclosure of limitations and costs

Developers already had to disclose additional types of costs that providers would pay to use certified capabilities. The proposal built on that requirement and had certification bodies check disclosure during surveillance. For the API criterion specifically, developers had to publish technical documentation and terms of use, with a public link [2]. That gave outside developers, and providers evaluating products, information that had often been available only under nondisclosure agreements.

Disclosure mattered most for the access layer. An API that existed but cost a third-party developer an unknown fee, or required a contract with restrictive terms, would not produce a market of applications. Publishing documentation and terms let developers judge the cost of building against an API before starting. It also let providers compare vendors on openness, which few procurement processes had measured before.

Privacy and security moved to the developer

One less visible change in the ONC proposal shifted responsibility for privacy and security capabilities. Under the 2014 Edition, providers had to make sure their combination of certified technology met all the privacy and security criteria. ONC proposed instead that each certified module meet the privacy and security criteria that applied to it, as a condition of its own certification [2].

For architecture, that change mattered most in a modular environment. A provider combining an EHR with a separate certified API platform no longer had to work out which product supplied audit logging or authentication for which function. Each module carried its own. The provider still had to configure and operate those capabilities and meet its HIPAA Security Rule obligations, but certification now said more about each piece.

Certification as a reference for other programs

ONC also proposed to make certification easier for other HHS programs, other public programs and private organizations to reference [2]. That turned certification into a general-purpose assurance rather than a feature of one incentive program. Payers, state Medicaid agencies and grant programs could require certified technology for their own purposes.

Over time, that is what kept certification relevant after the incentive payments ended. Programs that had nothing to do with meaningful use continued to rely on the certified criteria as a definition of what a capable system could do.

What it meant for procurement

For health system procurement teams, the governance layer offered new tools. Contracts could require that the vendor maintain certification to named criteria, cooperate with surveillance, and disclose costs for API access by outside applications. They could require service levels for the API as well as for the core EHR. They could address who paid when an outside application’s use increased load on the provider’s systems. In 2015 those terms were uncommon. The proposal’s disclosure requirements gave procurement a basis to ask.

The strongest objections, and what they missed

The March 2015 proposals drew serious criticism. Four objections were strongest, and each had merit.

Providers were being measured on what others did

The engagement and exchange measures counted actions by patients and other providers. A physician could not make a patient download a record or make a referring hospital send a usable document. Critics argued that providers would be penalized for behavior outside their control.

That objection was largely right as a matter of measurement, and the thresholds were later reduced and then replaced. As an architectural signal, it missed the point of the design. The measures pushed providers to build the infrastructure for engagement and exchange even where use was low at first. Once the infrastructure existed, later rules could require it to be available without measuring use. That is roughly what happened.

The API criterion was too vague

ONC specified outcomes, not a standard. Critics, including some developers, argued that without a mandated standard each vendor would build a different API, and applications would have to integrate with each separately. That would recreate the interface problem the API was meant to solve.

The criticism was accurate for the short term. APIs certified under the 2015 Edition varied, and outside developers faced real integration costs. ONC’s stated intent to allow FHIR without requiring it reflected the standard’s immaturity in early 2015 [2]. The later decision to require a standardized FHIR-based API answered the objection once the standard had matured. Requiring an immature standard in 2015 might have locked in a worse result.

Security and privacy were not ready

Privacy officers argued that opening records to third-party applications created risks HIPAA did not cover. Patients might not understand what an application would do with their data. Providers might be blamed for breaches at applications they had no control over.

That concern remains open. Later rules limited providers’ ability to block applications on grounds other than genuine security risks, while consumer protection agencies took on some oversight of health applications. Patients gained access, and some of the risk moved to them. A health system’s best response was to inform patients clearly about the choices they were making, and to secure its own side of the API well.

The API would entrench the largest vendors

A fourth objection came from the opposite direction. If APIs were controlled by EHR vendors, the vendors would set the terms on which outside applications could reach data. The largest vendors could favor their own applications, charge outside developers, or slow approvals. An API mandate might strengthen incumbents rather than open the market.

The proposal’s documentation and terms-of-use disclosure was a partial answer [2]. It made terms visible without limiting them. The fuller answer came later, when the Cures Act and its rules addressed information blocking and set conditions on certified API developers, including limits on fees. In 2015, providers who wanted an open access layer had to negotiate for it in their own contracts.

Reading the architecture for a health system today

The four layers described in March 2015 still organize a health system’s interoperability architecture. The names and standards have changed. The checks a chief information officer or chief architect can run on each layer are similar to those the proposals implied.

Access

Every route by which outside parties reach patient data should be listed, with its identity assurance level and its owner. A register of connected applications needs a review process and a documented response to misuse. API logs deserve the same monitoring for abnormal patterns as any other external interface, including alerts on unusual volumes and on applications that request far more data than their stated purpose needs.

Data

A named function should own terminology mappings and data quality at the boundary. The organization should know what share of its outbound data is coded to standard vocabularies, by data type. It should also know how long it takes to fix an error at the source once an outside user reports it.

Exchange

The organization should know which networks it relies on for push and query exchange, and what share of its referral partners it can reach through them. It should measure how much incoming data is incorporated as discrete data and how much clinician time reconciliation takes. Those two numbers show whether exchange helps clinicians or adds to their load.

Governance

Vendor contracts should require maintained certification, cooperation with surveillance and disclosure of API costs and terms. There should be a known route for reporting certified functions that do not work as certified. Procurement should score openness alongside features and price.

Sequencing the work

The layers also suggest an order. Data quality comes first, because an API or exchange connection that carries poor data spreads errors faster. Access controls come next, because opening an interface without identity proofing, logging and an application review process creates risk that is hard to retract. Exchange and incorporation follow, with clinician workflow designed before the volume arrives. Governance runs through all of it, in contracts and in the monitoring of certified functions.

Organizations rarely get to follow that order cleanly. Regulatory deadlines usually arrive for access first, because patient access is the most visible requirement. A health system that must open an API on a fixed date can still run data quality work in parallel and limit the initial scope to data it trusts.

How it differs by organization

For a large integrated system with one major EHR, most of the layers arrive through the vendor, and the work is configuration, governance and monitoring. For an ambulatory network running several EHRs after acquisitions, the data layer is the hardest, because each system codes data differently and patients appear in more than one. For a critical access hospital, the constraint is staff: the technology may be certified, but no one may have time to review applications or reconcile incoming records. Each type of organization needs a plan for the layer where it is weakest, rather than a generic interoperability program.

Those checks are more useful than the thresholds that dominated debate in 2015. The thresholds measured adoption in a particular year. The layers describe a system that has to keep working as standards, networks and applications change.

With hindsight

CMS and ONC finalized both rules on October 16, 2015. The CMS rule, issued as a final rule with comment period, kept Stage 3 optional in 2017 and required for all participants beginning in 2018. It also modified meaningful use for 2015 through 2017 [3]. The 2015 Edition final rule took effect on January 14, 2016 and adopted the application access criteria, under which certified health IT had to offer APIs for the Common Clinical Data Set [4].

The program around those rules changed quickly. The Medicare Access and CHIP Reauthorization Act, signed in April 2015, folded the Medicare meaningful use program for clinicians into a new Merit-based Incentive Payment System. CMS renamed the hospital program Promoting Interoperability in 2018. The single-stage design survived, but as a component of broader payment programs rather than a standalone incentive.

Stage 3 itself arrived later and lighter than proposed. The required use of 2015 Edition technology slipped, and thresholds for patient engagement and exchange were reduced or reworked. Measures that counted patient use of portals and APIs gave way to measures that required access to be provided. Several of the specific numbers debated in March 2015 never took effect as written.

The architecture moved the other way. The 21st Century Cures Act, enacted in December 2016, called for APIs that allowed access to health information without special effort, and made information blocking a statutory concern. ONC’s final rule implementing those provisions was published on May 1, 2020 and took effect on June 30, 2020. It replaced the Common Clinical Data Set with the United States Core Data for Interoperability. It established API requirements for patients’ access to their information without special effort, and it set the information blocking framework [5].

The new standardized API criterion required certified health IT to support HL7 FHIR-based APIs for single patients and populations, with developers given until the end of 2022 to update their products. The 2015 intent, to allow FHIR without requiring it, had become a requirement once the standard matured. Payers were brought into the access layer through a separate CMS rule in 2020 requiring patient access APIs for many health plans in federal programs.

The governance layer evolved too. ONC’s HTI-1 final rule, effective February 8, 2024, discontinued year-themed editions in favor of incremental updates to the ONC Certification Criteria for Health IT. It adopted USCDI version 3 as the baseline and added transparency requirements for decision support interventions, including predictive tools [6]. The 2015 proposal’s logic, that certification should disclose what products did and cost, extended to algorithms.

What held up from March 2015 is the four-layer structure. Patient access through APIs, a defined and growing data set, exchange that expects incorporation and reconciliation, and a certification program that discloses and surveils are all features of the 2026 environment. Consumer applications began pulling records through FHIR APIs from 2018 onward. National exchange frameworks, including TEFCA, now handle query-based exchange that most providers lacked in 2015.

What did not hold up is the measurement approach. The idea of paying or penalizing providers based on patients’ use of their records, and on other providers’ document quality, proved hard to sustain. Later rules shifted toward requiring capabilities and prohibiting interference, and away from counting outcomes that providers could not control.

The open problem from 2015 also remains. Applications outside HIPAA receive health data at patients’ direction. Federal agencies have addressed parts of that gap, but a health system that sends data to an application still has limited say over what happens next. Its responsibility is to secure its side, verify identity well and give patients clear information about the choice they are making.

Spotted an error? Tell us and we will check it and correct the article where needed.

  • meaningful use
  • Stage 3
  • 2015 Edition
  • APIs
  • interoperability
  • ONC certification