Author: Alyssa Ellis

By JOSHUA MANDEL, MD, DAN GOTTLIEB, MPA, and KENNETH D. MANDL, MD, MPH

June 1, 2023

Thank you for the opportunity to comment on the HTI-1 proposed rule! We understand that it takes time to see the effects of recent changes, as many of the effects of previous regulations are just now making their way into practice. It’s important to learn from experience and focus new regulations on demonstrated approaches. Given the design and demonstrated value of SMART on FHIR and SMART/HL7 Bulk FHIR Access API, we urge strong consideration of their use, or at least testing, in addressing use cases. The bandwidth of Health IT developers to reliably surface multiple APIs addressing minimally different use cases (e.g., quality metric reporting, eCR) may be better spent on ensuring performant SMART APIs with reliably mapped USCDI elements.

In the current landscape, we see many reasons for encouragement based on the roll-out of API access in 2023, especially for single-patient use cases (clinician and patient-facing apps). However, we recognize ongoing challenges that API users are experiencing with FHIR API performance and workflow for high-value population-level use cases at scale. More clarity on expectations from ONC and CMS is needed to ensure consistent and performant population-level APIs as are meaningful metrics.

USCDI Definitions and Expansions

We strongly support USCDI as a common basis of exchange. However, there are challenges in a one-size-fits-all model of “the core data set” — some elements are a stretch for EHRs, and other elements (such as, concepts common to payors systems) aren’t addressed deeply enough. We’ve also seen significant standards-development challenges around vague or ambiguous elements in USCDI. This leads to churn in the standards process, widespread optionality in the standard profiles, and, due to wide variation in formats, limited capabilities that may be unsuited for the use cases of the submitters who advocated for the inclusion of these elements in USCDI.

Our Recommendations

1. When adding elements to USCDI, ONC should:
   • Explain a set of intended use cases
   • Provide details about how and how frequently these data are being used to meet the intended use cases in real-world systems today, which will help the standardization effort to build on real-world experience and achieve ONC’s desired outcomes.
2. Within USCDI, ONC in its coordination role can pre-group data subsets, for example tagging them as related to demographics, clinical, financial / payment / insurance, or imaging.These tagged subsets could then be pulled in by other programs beyond the EHR, such as CMS regulation of  the APIs and data responsibilities of payors.

Decision Support Interventions

This is a significant and expanding area of great clinical import. Many predictive capabilities are expanding rapidly. Transparency and mitigation of bias are imperative. However, we believe that regulating a specific set of data and model attributes is premature, as there is not yet consensus on the set or structure of such attributes. Many techniques are evolving in parallel, and even a functional requirement that’s opinionated about (numerous!) specific data classes risks producing unintended consequences, such as information overload in the clinical UX and incomplete and unhelpful “box checking” data.

Targeting new regulations at EHRs here may also have unintended consequences. EHRs are integration points for diverse decision support, and choices about how to assess and manage risk are a core responsibility of the clinical providers deploying decision support. Imposing requirements on EHR vendors may lead to proliferation of decision support that route around the EHR rather than through it.

Our Recommendations

1. Add first-class support for CDS Hooks as an alternative to InfoButton
2. Do not regulate a specific set of attribute unless/until such attributes are successfully incorporated into some real-world deployments
3. Do not regulate “data review” UX requirements unless/until this is demonstrated to work in practice
4. Emphasize the importance of providing model cards and similar data to health systems at the time of selecting/configuring decision support, rather than overloading clinician-facing UX at runtime.
5. Emphasize that responsibility for risk assessment and mitigation should be shared with DSI providers, rather than falling onto EHRs.

Public Health: Electronic Case Reporting and Surveillance 

There is a substantial opportunity to empower Public Health with data from the care delivery system in far more robust ways than have been accomplished in the past. The most promising approach will be to ensure that Public Health does not rely on one-off solutions, even ones specified in rules and regulations, but rather derives data from common data infrastructures used by CMS and other entities.

Proposed case reporting and surveillance standards are complex for EHRs, with dynamic behavior required from every EHR, intricate temporal logic, and value sets that change over time without standard update protocols. Proposed case reporting standards are also brittle, with significant effort to support public-health-specific APIs, requirements that still vary by jurisdiction, and limited ability to predict future needs.

Our Recommendations

1. Focus on enabling public health use cases through existing SMART on FHIR and SMART/HL7 Bulk FHIR Access APIs rather than one-off public health-specific protocols
2. Leverage SMART on FHIR backend services and apps that:
   • Can be authored or sponsored by public health
   • Can manage alerting logic nimbly without per-EHR repeated work
   • Consume existing FHIR APIs and trigger from US Core data
3. Expand existing APIs to ensure support for cross-patient queries, such as:
   • GET /Encounter sorted by time, code, category
   • GET /Observation sorted by time, code, category
   • GET /Condition sorted by time, code, category
4. Invest in FHIR Subscriptions as a performance optimization, such as “US Core Event Feed” as a Subscription Topic, which supports filtering by resource type, patient, code, and category
5. Expand required parameter support on the FHIR bulk data API to:
   • Mandate support for the “_since” parameter, allowing apps to request only data modified after a specified date
   • Mandate support for “_typeFilter”, specifically to allow filtering of:
     • Patient resources by demographic information
     • Observations, Conditions, and other US Core resources by category (where applicable) and code (where applicable).

EHR Reporting Program

We appreciated the chance to participate in prior rounds of measure planning and applaud ONC’s introduction of specific measures. However, we are concerned about the complexity of the proposed measures, with multiple numerators, denominators, and stratifications, and unclear exactly what reporting data are being proposed. We are also concerned that many measures imply the collection of new types of data, such as app categories and intended audiences.

We are further concerned that some metrics propose to segregate by an unreliable concept of “two different endpoint types: patient-facing and non-patient-facing.” FHIR endpoints can serve multiple and broad audiences and need not have a “type” like “patient endpoint” or “provider endpoint”.

Our Recommendations

1. Start with a smaller reporting set and expand over time
2. Focus on data elements already known/available
3. Assess usage based on user types (e.g., patient vs provider — these are tracked and consistent), not “endpoint types” (which are not distinct in all deployments).
4. For bulk data requests, include at least one metric to track the real-world performance of implementations
5. Key metrics:
   • Consumer access (SMART on FHIR)
     • Count of distinct apps connected
     • Count of app connections per patient
     • Count of portal sessions per patient
   • Clinician access (SMART on FHIR)
     • Count of distinct apps connected
   • FHIR Bulk Data API
     • Count of export requests
     • Export time per resource (average)
     • Group size for successful exports (average)

FHIR Endpoint for Service Base URLs

We strongly support the ONC’s addition of specific metadata requirements for endpoint URL lists to help patients identify the correct endpoints for their healthcare providers. We are concerned about the specificity of the guidance around FHIR resources and data elements in the regulations and believe that introducing more detailed functional requirements instead would encourage the industry to refine and align on technical data profiles.

Our Recommendations

1. Enhance the existing functional requirements by mandating that published endpoint lists support: physical locations, organizational hierarchies, patient-facing brand names, and institution logos.
2. Encourage industry evaluation of detailed FHIR implementation guides to meet these functional requirements. A starting point is the Argonaut Project’s 2022 “SMART Patient Access Brands” profiles (available at https://build.fhir.org/ig/HL7/smart-app-launch/branches/pab/brands.html).

Standards Updates: SMART v2

We have been encouraged to see the breadth of adoption and are pleased to see the enumeration of SMART capabilities and clear rationale in HTI-1 — such considerations help justify industry investments. We agree with the features that HTI-1 proposes to require.

There has been ongoing confusion about consumer ability to connect browser-based apps persistently, without the need to re-authenticate. Previous ONC regulations guaranteed this ability for some apps but not others, drawing distinctions by client type (confidential vs public, or native vs browser-based). Previous ONC regulations also do not require support for Cross-Origin Resource Sharing, leading to a situation where some server configurations prevent browser-based apps from receiving an access token. SMART 2 introduced support for PKCE to reliably bind token issuance to a client’s authorization request, mitigating previous concerns. However, the HTI-1 proposal stil does not ensure long-term access for “public clients,” even though “public clients” can offer better data segregation than “confidential clients” by avoiding the need for consumer data to transit through a developer’s backend server. Such restrictions impose unnecessary risk by encouraging app developers to use a backend server and limit patient choice by artificially limiting the capabilities of browser-based local apps.

Our Recommendations

1. Adopt the current release — i.e., SMART 2.1 (not 2.0), which includes:
   • Improved FHIR Context management
   • App State capability

2. Mandate support for client-side browser-based apps:
   • Mandate that consumers can approve “offline_access” scope for any registered app, not just confidential or native clients
   • Mandate that servers “support purely browser-based apps” per https://hl7.org/fhir/smart-app-launch/app-launch.html#considerations-for-cross-origin-resource-sharing-cors-support

Information Blocking: TEFCA Condition for “Manner” exception

Organizations today have the right to negotiate the manner of access, however, HTI-1 notes that “it is reasonable and necessary for actors who have chosen to become part of the TEFCA ecosystem to prioritize use of these mechanisms”. We believe that artificially limiting this right for organizations that participate in TEFCA may have unintended consequences. TEFCA has not yet launched, and it is unknown what full-scale challenges may emerge. Some organizations may participate in TEFCA out of necessity, driven by certain use cases, but may not be well served by TEFCA for all use cases.

For single-patient use cases, there may be a significant delay before TEFCA’s draft FHIR roadmap is widely implemented. In the meantime, organizations should be able to negotiate for the manner of access that suits their requirements, including access to a certified EHR’s SMART on FHIR patient API endpoints. For population-level use cases, TEFCA may not offer any pragmatic alternative to bulk FHIR export, even if TEFCA supports access to all of the data elements and patients associated with a bulk request. Organizations should be able to negotiate for the manner of access that suits their requirements, including access to a certified EHR’s SMART on FHIR Population API endpoints.

Our Recommendations

1. Do not introduce a TEFCA Condition for the information blocking “Manner” exception
2. Instead, monitor TEFCA deployments closely with an eye to utility, completeness, timeliness, ease of access, security, privacy, transparency, and consumer participation. Base decisions on real-world experience.

Patient Right To Request a Restriction (proposed new criterion)

We strongly support individual rights, including the right to restrict disclosure of PHI. However, HTI-1’s proposal does not ensure patient rights and is very likely to lead to confusing UX, mistaken expectations, and violated trust. Consumers have the ability to request restrictions, but providers do not need to follow them, and provider technology does not offer a good way to enforce restrictions even if a provider wants to. Techniques to manage and enforce such restrictions are still an open research challenge.

Our Recommendations

1. Do not introduce a non-binding requirement for requesting restrictions to disclosures at this time.
2. Instead, adopt two pragmatic approaches that empower patients with controls over and insights into the use of their data:
   • Controls at the source: Let patients ensure data accuracy at the source, to prevent sharing of inaccurate information. Introduce a functional requirement aligning with the HIPAA right to request corrections and amendments to erroneous information. This would:
     • Ensure that patient portals and patient APIs provide patients an easy path to requesting corrections to their medical records, or to amending records in the case that providers decline to apply corrections.
     • Provide a clear market signal to drive participation in standardization efforts through HL7’s patient empowerment workgroup
   • Insights and controls for exchange: Let patients see who’s querying their data in TEFCA, and provide opt-out controls. Since most uses of TEFCA would fall under “treatment, payments, and operations,” TEFCA represents the largest-scale non-accountable PHI flow in history. See our perspective “The Patient Role in a Federal National-Scale Health Information Exchange” at https://www.jmir.org/2022/11/e41750 for background. ONC should prevent TEFCA from driving such “dark” traffic by providing a mechanism for patients to review:
     • Who made a query for my records?
     • When?
     • For what purpose of use?
     • Who responded to the query?

RFI Response: FHIR Subscriptions

FHIR has included preliminary support for subscriptions since R3 and has recently introduced support for “topic-based subscriptions” incorporating industry feedback to enable scalable deployment. Topic-based subscriptions are available starting with FHIR R4 based on the “Backport IG” at https://hl7.org/fhir/uv/subscriptions-backport, and are natively supported in FHIR R5.

Given ONC’s current investment in FHIR R4, we suggest that introducing topic-based subscriptions via the Backport IG is the most pragmatic and incremental way to expand capabilities of the current deployed base of FHIR implementations. Using this Backport approach, the certification program could require implementers to support a limited starter set of subscription topics, functionally defined to align with core regulatory priorities. A suggestion for initial use cases would be:

1. “Patient data updates.” Allow notifications when new/updated patient data are available in association with a specific patient ID. This could be used to allow monitoring by patient-facing apps that currently have to poll at regular intervals. It could also help enable public health applications that are monitoring patient encounters in the context of a reportable condition.
2. “Encounter data update.” Allow notifications when a new encounter is created or when an encounter is updated within the health system. This could be used to enable integration of public health triggering and reporting logic into a clinical system without requiring each clinical system vendor to independently implement support for detailed temporal triggering and reporting logic. This might also be the starting point for a FHIR-based encounter notifications alternative to ADT, but in our experience, it is easiest to drive adoption around new capabilities rather than reworking the mechanisms for already available capabilities.

Both of these use cases could be supported by a single “US Core Event Feed” topic, which would enable subscriptions to the resources defined in FHIR US Core with filtering available by:

• Resource Type
• Category (as applicable)
• Code (as applicable)
• Patient ID (for subscriptions at the single-patient level)

RFI Response: CDS Hooks

CDS Hooks is an HL7 standard for integrating external decision support into the clinical workflow. CDS Hooks represents a mature target for standards adoption. A suitable initial target is CDS Hooks 2.0, with mandatory support for the “patient-view” hook, mandatory support for prefetch of US Core data queries, and mandatory support for “fhirAuthorization” access tokens. With this functionality enabled, CDS Hooks can be permitted in regulations an alternative to InfoButton. Beyond this introduction, ONC should articulate use cases and desired outcomes to spur industry development and maturation of additional hook types.

RFI Response: SMART Health Cards and SMART Health Links

The SMART Health Cards (SHC) specification has been used to provide COVID-19 immunization and lab results to individuals at scale including deployments in at least 15 countries. Implementers include major EHRs, and pharmacy chains, as well as public health programs from over 24 US states. Immunization records are available as SMART Health Cards for over 225M people in the USA. By leveraging FHIR and W3C Verifiable Credentials, SMART Health Cards allows sharing of static data sets, with a focus on data that are small enough to fit in a single QR code.

Building on SHC’s format, signature scheme, and trust infrastructure, the SMART Health Links specification allows for more advanced usage including larger data sets (e.g., a full immunization history or patient summary) as well as dynamic data sets (e.g., an immunization history that can be augmented when a patients receives vaccine doses). The SHL specification underwent an initial round of design, prototyping, and feedback through the Argonaut standards accelerator in 2022. Early industry adoption of SHL in 2023 has focused on immunizations and health insurance coverage details, with additional prototypes demonstrating a workflow for sharing International Patient Summary documents.

We believe ONC can foster broader adoption of SHC and SHL by sponsoring programs to identify and address unmet needs in consumer mediated data exchange. In particular, SHCs demonstrate how reducing friction to health data access makes a tangible difference for patients, and SHLs enable this access pattern to be scaled to nearly any data type or use case. In the near-term, SHC and SHL provide a common pattern for sharing use-case-specific data bundles, expanding the reach of FHIR implementation guides for sharing with minimal coordination across organizations. Over time, we see a path for this technology to support more interactive sharing paradigms that could address some of the high-frequency, high-friction touch points that people have with healthcare. This could include prompted sharing (e.g., before a clinical visit, patients might receive an information sharing request that they can review and respond to using a personal health app or wallet) as well as assisted form-filling (e.g., when presented with a blank clipboard asking for clinical information, a personal health app could pre-fill many items and, when desired, could attach verifiable provenance to the submission).

RFI Response: Industry-Led Innovation Activities

Even as API access to clinical data has become routine in practice, API access to imaging data has remained challenge. This year the Argonaut Project has organized development and testing of a specification for SMART app access to Imaging data, building on the SMART on FHIR API capabilities of Certified EHR Technology. The Argonaut design leverages existing EHR app-registration and app-approval workflows, augmenting the available data from an EHR’s “clinical server” with an additional set of imaging data from an imaging endpoint. The imaging endpoint hosts a narrow subset of FHIR and DICOMweb functionality, authorized using SMART on FHIR token introspection to allow SMART apps to list the studies available for a patient and retrieve user-selected studies. We believe this architecture offers a promising and scalable approach to enhance patient and provider access to imaging studies alongside clinical data. Such access would facilitate diverse use cases for providers and patients including:

• For Providers
  • Support analysis w/ preferred tools, such as speciality-specific viewers, image-driven severity scoring, or image-driven risk calculations 
  • Streamline consultation workflows
• For Patients
  • Enable access, compilation of full data
  • Ensure data are available to specialists
  • Facilitate second opinions and 
  • Streamline data donations for research

ONC could identify opportunities to support Argonaut Imaging and functionality that expands on the API capabilities of Certified EHR Technology. Over time, projects like Argonaut Imaging Access can serve as a model for widespread application access to a growing swath of Electronic Health Information.

Sincerely,

The SMART Health IT Team

By DAN GOTTLIEB, MPA, JOSHUA MANDEL, MD, and KENNETH D. MANDL, MD, MPH

November 29, 2022

We applaud CMS’s vision for expanding availability and improving accuracy of healthcare directory information while reducing the provider burden associated with keeping directory listings up to date.

There is an opportunity to build on the existing, widely used, National Provider Plan and Enumeration System (NPPES) directory infrastructure to incrementally address these challenges. Crucially, improvements in incentives, user experience, and federation are needed to unlock the full potential of NPPES.

Overview. We propose that rather than developing a new centralized directory, enhanced NPPES listings could serve as a “primary key” for healthcare directory information, maintaining the unique id used to tie together data from multiple organizations and storing links to flat files in FHIR format with additional, frequently-updated, information maintained by payor and provider organizations (e.g., details about services, hours, plan participation, and facilities). The term “API” sometimes refers to fine-grained interfaces (e.g., supporting queries like “at which locations does Dr. Smith practice,” or “What time is Dr. Smith’s clinic open next Tuesday”), but we believe a simpler API might be a better starting point — beginning with the ability to host a set of “flat files” at a well-known URL. Hosting flat files that are updated on a frequent basis would satisfy many consumption use cases. Leveraging SMART/HL7 FHIR Bulk Data standards would take advantage of tools and infrastructure designed for working with FHIR data. Third-party aggregators could use the data in NPPES and the referenced flat files to create applications targeted at specific use cases (e.g., a user-friendly, patient-facing application for people to find new healthcare providers).

Incentives. The RFI notes that provider data in the NPPES directory is frequently inaccurate. Without strong incentives for providers to keep the data up to date, technology changes are unlikely to improve data quality. One approach to addressing this challenge would be attestation requirements, such as mandating that providers participating in CMS payment programs periodically attest to the accuracy of their information in the NPPES directory. Additionally, the more the NPPES data are actively used in CMS operations, the more likely providers are to keep it up to date, suggesting another avenue for incentives. For example, the contact information in the directory could be used for all communications around provider payment, necessitating up-to-date information and ongoing engagement with the system.

Improving User Experience. Anecdotally, we’ve heard from users that updating their information on the NPPES website can be cumbersome, particularly when managing listings for multiple providers. It seems common for organizations to be listed multiple times in NPPES, perhaps because creating a new entry is easier than gaining administrative control over a previous entry. CMS should launch a user centered design effort to understand how the site is being used and redesign it to better accommodate common use cases. For example, a simplified interface may be useful for solo practitioners, functionality to create or update listing by uploading a spreadsheet may be beneficial for users at mid-sized organizations, and the ability to integrate through an API may be preferred by users at large institutions.

Federation (hub and spoke). While a single repository for all healthcare directory data is an attractive notion, it would be costly to build and maintain, and would also risk a single point of failure for a system that has high uptime requirements. Furthermore, a single directory with multiple constituencies, varied data granularity, and a range of update frequencies for different data elements introduces substantial complexity in the data model. In a hub-and-spoke federated model, core metadata would be centrally hosted by NPPES while auxiliary information such as a provider’s hours at an institution and whether they are accepting new patients would be maintained in a file directly on the relevant organization’s website, and easily updated when the data changes. To aggregate this data, consuming systems would follow URLs to the files from the NPPES directory. Similarly, payer organizations could publish files listing plan information and participating providers, and the locations of the payer files could be published in a new section of the NPPES directory. By using providers’ NPPES ids to join the data in these files with those on provider sites, third parties can build up comprehensive databases to power specific apps without the need for novel, centrally maintained infrastructure. A similar approach has been successfully adopted with the SMART Scheduling Links project; working with the US Digital Service, participants enabled distributed discovery of COVID-19 immunization appointments across major pharmacies and healthcare providers (https://github.com/smart-on-fhir/smart-scheduling-links).

RFI: https://www.federalregister.gov/documents/2022/10/07/2022-21904/request-for-information-national-directory-of-healthcare-providers-and-services

Dear SMART Community:

10 years ago, the SMART project launched with a New England Journal of Medicine article proposing that EHRs could serve as a platform with a universal API supporting substitutable apps. Now, the 21st Century Cures Act requires the very API we envisioned and the draft rule, by the Office of the National Coordinator of Health Information Technology, following from Cures, is open for public comment.

The rule specifies that the SMART on FHIR API will be universally required so that an app written once can run anywhere in the healthcare system, and gain access to all of the elements of a patient’s electronic data, without special effort.

Getting the details right could not be more important to realizing a future healthcare system underpinned by a robust health information economy, driven by apps and real world information.

In advance of posting our public comments, we post a draft here for community feedback.

The input is from our perspective as founders and members of the SMART on FHIR team that developed the SMART API, defined with HL7 and ONC the Bulk Data Export API specification, and launched the CDS Hooks project.

Thank you for engaging with us on this.

Ken Mandl

---

Draft 

Public comments on the Proposed Rule for the 21st Century Cures Act: Interoperability, Information Blocking, and the ONC Health IT Certification Program

Dan Gottlieb, Josh Mandel, Kenneth Mandl

@gotdan @joshcmandel @mandl

Overview. Our first and overarching comment is that the ONC’s drafting of this rule reflects clear thought, responsiveness to the community, attention to detail, and tremendous skill in expressing critical desiderata for a robust health IT ecosystem. The product of these efforts, the Proposed Rule, is extraordinarily impressive to us. The Rule addresses and reinforces virtually all of the major underpinnings which are currently feasible and needed to produce an interoperable apps-based ecosystem.

We state clearly and emphatically that the Rule should be largely left intact in its spirit and in most of its details.

Priority Goals

1. Improve App Connectivity for Patients and Providers

Connectivity. We cannot overemphasize the importance of standardizing the capability for a SMART on FHIR app written one time, to run unmodified anywhere in the healthcare system, regardless of the EHR or database vendor or version, accessing all of a patient’s electronic health data.  The app should be substitutable–easily added to or deleted from an EHR or related database, as easily as apps are added to iPhones and Android mobile devices. The ONC has taken an enormous step forward with the certification requirement for the SMART application launch framework with the declaration of FHIR as a foundational standard, constrained by the Argonaut- or US Core-defined profiles.  

Weakening of any of the API requirements or information blocking provisions would potentially interfere with this objective.

Affordability. A patient’s access to all elements of her electronic medical record, across the SMART API, without cost, is well-supported in the Rule. The power of the provision is illustrated by the ease and speed with which Apple connected its Health App to more than 500 health systems across the SMART API, enabling patients at those institutions to download their data to iPhones for subsequent use in iOS apps. Under the proposed rule, any other app can connect to the same systems across the same API.

Equally important, however, is that when a data provider (e.g., a practice or a healthcare system) also acts as an API user, they must be able to connect any app of their choice to their EHR. This ability is well-supported in the Rule, including important guardrails around what an API provider can charge for cost recovery and on a per API call basis.  Affordability of this capability is essential. Data providers must be able to afford the enhanced health IT systems that have APIs to permit the fundamental interoperability that many believe should be native properties of any modern software system. The data providers must also be able to afford the cost of API access resulting from the use of apps that connect and provide value. There is a risk that the economics of paying for API calls to enable API provider cost recovery will be limiting.

In the API/Apps model, the EHR and related IT serve as a platform for multi-sided business. It will take some time to understand and support the economics underpinning a nascent apps-based economy. Getting it right will strongly promote American business innovation, job creation and improved infrastructure to deliver value-based healthcare.

Evaluation. We understand that ONC may be pursuing the cost-based pricing program out of necessity, but not that API Providers will not have an incentive to drive down their own costs. Additionally, current levels of revenue for API providers are not necessarily maintainable for a health system seeking value and growth in the diversity of app developers. API pricing structures should ideally level the playing field between first-party EHR functionality and third party app functionality for a modular, extensible IT infrastructure.

Therefore, we propose that a year after the implementation of the Rule, O NC conduct a study to evaluate the real-world cost of APIs being used by health systems for areas such as clinical decision support, payments, machine learning, and precision medicine, and use the results to drive future policy. Benchmarking these costs will be difficult, but potentially useful metrics could include:

• Comparing the cost of accessing data via APIs to that of accessing the same data with traditional EHR front end interfaces by amortized EHR licensing costs and adjusting for expected use.  
• Comparing the cost of accessing data via EHR APIs to that of similar APIs, including those from cloud providers such as Microsoft Azure API for FHIR or Google Cloud for Healthcare and those provided as part of software solutions in other industries.
• Examining the business models being used. For example, are API providers charging for system upgrades only, or are there substantial ongoing fees that limit app usage?

2. Embrace Community Driven API Standards

Population API Access. We are extremely pleased to see that the Rule supports Population API Access (including https://www.federalregister.gov/d/2019-02224/p-447 and https://www.federalregister.gov/d/2019-02224/p-837). This requirement should specifically mandate the  use of the FHIR Bulk Data Access Standard, currently being balloted through the HL7 standards organization (http://docs.smarthealthit.org/flat-fhir/ ).

Provide API Access for  Data Exports. On the subject of full EHI access, we are concerned about the gap between the proposed rule for EHI Export and the regulatory intent of the 21st Century Cures Act to achieve interoperability with APIs that provide “access to all data elements of a patient’s electronic health record to the extent permissible under applicable privacy laws.” In ONC’s proposal for certified API access, “all data elements” has been interpreted to mean “a limited set of data elements.” While not all data have been standardized, ONC should still include a functional requirement that the EHI Export capabilities must be available through a documented (if potentially vendor-specific) API, and accessible to patients. Specifically, to provide a usable consumer experience, the EHI Export capability must be:

1. Accessible directly to a patient (or authorized representative), rather than (or in addition to) being built as a feature for clinic staff. This is to ensure that patients don’t need to make a request by, e.g., phoning the medical records department, and waiting while the departmental staff hits the “export” button in response to a request — a friction-laden process.
2. Exposed end-to-end through an API, rather than being implemented exclusively as a button hidden deep within a patient portal experience, or being crippled by the exchange of physical media like CD-ROMs. This is critical because:
   • Portal experiences vary, making features difficult to find and correctly describe (e.g., if a third party is trying to guide patients toward the export functionality in a variety of portals). This was a clear challenge for anyone trying to identify the “Transmit to a third party” features of a patient portal in the MU2 timeframe.
   • Managing physical media, e.g., CDs,  would take access outside the realm of modern, convenient consumer experiences, and would violate the without special effort requirement.
   • Managing files may be challenging on many patient devices (e.g., mobile phones), and some files may be best suited for off-device storage (e.g., in the cloud). API connectivity ensures that patients can have a seamless experience for accessing all of their health data, not just a core data set.

Our key recommendation on EHI Export is that ONC should require certified EHRs to support full EHI export via patient-accessible API, even without standardizing the API or the data payloads. This will meet the Cures intent for API access.

API connectivity will ensure that patients can have a seamless experience for accessing all of their health data, not just a core data set and healthcare providers have the option to connect apps to data that is not currently available through the USCDI dataset. Providing access through open APIs, even for data that are not structured in open formats, will further this goal.

We have heard concerns about the overall scope of EHI access in a variety of exceptional circumstances (e.g., for data not stored in the EHR itself, or for data that represent purely internal system implementation details). As a baseline, we recommend that the data to be included in EHI Export should encompass the complete set of information that an EHR vendor currently makes available through their widely adopted data warehouse products, through their user interfaces, or through their reporting infrastructure.

SMART on FHIR.  The Rule should Specify which SMART on FHIR capabilities an API provider will need to support, at a minimum. The proposed rule requires mandatory support for “refresh tokens,” “Standalone Launch,” and “EHR Launch” requirements. However to ensure consistent implementation the Rule should also specify mandatory support for the following ten SMART capabilities: sso-openid-connect, launch-standalone, launch-ehr, client-public, client-confidential-symmetric, context-ehr-patient, context-standalone-patient, permission-patient, permission-user, permission-offline.

USCDI. The USCDI is a strong successor to the Common Clinical Dataset. To cover an expanded set of app use cases, we recommend the addition of data elements that cover  clinical encounters and clinical imaging metadata.

FHIR R4. The need to support multiple versions of standards can hinder interoperability.  ONC should adopt FHIR R4 as the only allowed version for standard API access moving forward, with an advancement process as described in the Proposed Rule.

ARCH. The ONC proposes the “API Resource Collection in Health” (ARCH) Version 1 implementation specification in § 170.215(a)(2), which would list a set of base FHIR resources that Health IT Modules certified to the proposed API criterion (§ 170.315(g)(10)) would need to support.  While we see value in providing a catalog, we strongly recommend that ONC populate it using community-developed standards by groups like Argonaut and HL7, who can take functional requirements from USCDI versions, and produce Implementation guides. If ONC maintains an ARCH definition, this definition should be limited to referencing implementation guidance from community-developed processes; ONC should not “get out ahead” of the community process. For example, the ARCH should not make unilateral determinations about which FHIR resource or data elements are needed to meet a given USCDI requirement. Instead, that determination should be made through a community-driven, iterative process with real world testing of use cases.

Immature Standards. The vast majority of referenced standards in the Rule are mature, in real world use, and widely embraced by the community.  Two standards, however, strike us as “not ready for prime time:” Data segmentation for privacy (DS4P) and Consent2Share.  While we recognize privacy maintenance and consenting as essential functions in health care, we are very concerned that a premature push for adoption of these immature standards would have unintended negative effects. ONC should omit these from the certification criteria (including voluntary certification) and focus on driving real-world implementation experience before pursuing regulations.

3. Move Beyond EHR Data

Clinical Imaging. We are concerned that sharing of images, an essential data type for patient care, may slip through the cracks. EHR systems often contain metadata around the available imaging studies, however, the imaging studies themselves are frequently stored in separate systems known as picture archiving and communication system (PACS).  Providers should be responsible for sharing this imaging data, regardless of the technology supplier they choose. We recommend making PACS vendors subject to EHR certification rules, specifically for API access requirements.

Laboratory Data. While some clinical laboratory result data are accessible to patients through EHR APIs, historical data may not be comprehensive. Recently, national laboratory companies have begun to make these data available through API access for select apps. To promote a robust ecosystem of clinical applications, guidance should be provided on how this access should be expanded to an open ecosystem of apps to comply with the information blocking restrictions in the Rule.

---

Kenneth D. Mandl, MD, MPH,^1,2 Dan Gottlieb, MPA,² Josh Mandel, MD,^1,2,3

1. Computational Health Informatics Program, Boston Children’s Hospital, Boston, MA

2. Department of Biomedical Informatics, Harvard Medical School, Boston, MA

3. Microsoft Healthcare, Redmond, WA

The opportunity has never been greater to, at long last, develop a flourishing health information economy based on apps which have full access to health system data–for both patients and populations–and liquid data that travels to where it is needed for care, management and population and public health. A provision in the 21st Century Cures Act could transform how patients and providers use health information technology. The 2016 law requires that certified health information technology products have an application programming interface (API) that allows health information to be accessed, exchanged, and used “without special effort” and that provides “access to all data elements of a patient’s electronic health record to the extent permissible under applicable privacy laws.”

After nearly two years of regulatory work, an important rule on this issue is now pending at the Office of Management and Budget (OMB), typically a late stop before a proposed rule is issued for public comment. It is our hope that this rule will contain provisions to create capabilities for patients to obtain complete copies of their EHR data and for providers and patients to easily integrate apps (web, iOS and Android) with EHRs and other clinical systems.

Modern software systems use APIs to interact with each other and exchange data. APIs are fundamental to software made familiar to all consumers by Google, Apple, Microsoft, Facebook, and Amazon. APIs could also offer turnkey access to population health data in a standard format, and interoperable approaches to exchange and aggregate data across sites of care.

The Office of the National Coordinator of Health IT (ONC)-funded SMART on FHIR API specification enables apps to connect with EHRs in a standards-based way, giving users a frictionless way to choose their favorite apps. This property of substitutability defines a new form of interoperability. SMART leverages the Health Level Seven (HL7) Fast Health Interoperability Resources (FHIR) standard and has been implemented by the major EHR products. The SMART app gallery, and EHR-specific app stores such as Epic’s App Orchard and Cerner’s code App gallery host scores of app that connect to EHRs.

Two particularly intriguing uses of SMART are (1) Apple’s use of the API to connect its health app to hundreds of health systems enabling users to download copies of their health records to their smartphones; and (2) the Centers for Medicare and Medicaid’s implementation of “Blue Button 2.0”, enabling beneficiaries to connect apps to their healthcare financial data.

Because the specifics of the final rule matter greatly, we strongly encourage policy makers to attend carefully to a few key requirements which derive from the phrases “without special effort” and “all data elements.”

Expanded data access. ONC has proposed a set of standardized clinical data that will grow over time from the 2015-era “Common Clinical Data Set” to a forward-looking “US Core Data for Interoperability”. This kind of consistent, standards-based data set holds tremendous promise for the ecosystem. At the same time, standards can lag behind clinical practice and cutting-edge technology development, so the Cures Act goal of “all data elements” would be challenging to achieve through detailed clinical modeling standards alone. We should not allow the perfect to be the enemy of the good. We propose a three-pronged approach to meeting the Cures provisions for “all data elements”:

1. Use standards that exist today. For example, FHIR “US Core” profiles cover the 2015 era Common Clinical Data Set, providing a common basis for communicating patient demographics, medidcations, conditions, lab results, vital signs, and more. These core data should be made available through APIs to provider- and patient-facing apps.
2. Continue developing these standards over time. For example, efforts like HL7’s Argonaut Project are driving common support for new data types like clinical notes as a fast-moving 2018 roadmap. We should start building a community-maintained “profile backlog” to articulate and prioritize the most valuable data that haven’t yet been standardized.
3. Enable flexible approaches to cover the gap between our well-standardized-and-growing “core data” definitions and the long tail implied by the Cures provision for “all data elements.” As one example to illustrate how EHR vendors could ensure that innovators have programmatic access to all of the clinical data accessible in the system: similar to the way vendors publicly document a subset of APIs today, they might expand this documentation to include database schema, tables, columns, and enumerations used to store complete clinical records.

This approach (use standards, develop standards, and cover the gap) would empower early adopters to develop cutting edge clinical integrations ahead of the standardization process, building experience to guide the standards process that follows.

Standard and ubiquitous APIs for patient facing apps, provider facing apps and population analytics. Our vision is that an app written once should run anywhere in the healthcare system. The availability of standardized APIs, ubiquitously implemented across care settings, is essential to driving down the “special effort” that is still typically required to create, distribute and use health apps.

1. Standardize APIs for apps. The SMART Health IT (a.k.a. SMART on FHIR) specification is sufficiently mature to be considered as an industry standard for launching and authorizing apps in an EHR or patient portal. It is in widespread use in clinical settings, has achieved consensus through the Argonaut process, is implemented in EHR products, and its core elements are being incorporated into the next release of the FHIR standard.

While the SMART-based app integration focuses on one‐patent‐at‐a‐time access to health system data, population level data export is critical for value‐based care, postmarket surveillance, quality improvement, and clinical research. The API should enable a user or an app to specify export of all EHR data or EHR data on defined cohorts at the discretion of the data owner. Under ONC funding, a standard for bulk data export in a FHIR‐formatted flat file has been proposed and the Argonaut implementation group is working to pilot it in 2018.

1. Allow multiple pathways to register apps for connection to EHRs and other HIT. As more EHR vendors build support for standards-based apps, developers are discovering that they need to independently register each new app with each vendor and complete a set of on-boarding, review, or “vetting” steps before users are able to install the app and authorize a data connection. The app registration and vetting landscape is evolving quickly as vendors create developer programs, launch partnerships, and build out their own app marketplaces. App vetting procedures review and assess critical aspects of integration including security, usability, and business/privacy practices and offer value to end-users, who expect a clean, safe, experience of choosing, installing, and running apps.

Nonetheless, we have observed that these vetting practices can cause friction for some use cases and believe it is too early to define a “one size fits all” standardized app vetting process.  As such, we propose an “escape hatch” in the form of an at your own risk principle, by which provider organizations and individual patients should be able to accept the risk of connecting an un-vetted app to their own data without vendor review. While many apps will follow a conventional path of registration and vetting, this option provides a route to ensure that all apps, even small-scale apps (e.g., one-offs produced by individual tinkerers, open-source developers, research efforts) can reach visibility and commercial viability within the real-world clinical landscape, and that providers have the opportunity to select any apps of their choice.

1. Ownership terms. App developers should have the option of retaining all intellectual property related to the app, regardless of how the app connects to the EHR and which underlying EHR APIs the app consumes.
2. Maintain free registration of apps for patients. As required now under Meaningful Use Stage 3, patients should always be able to connect apps of their choice, without cost.
3. App connections should be long lasting, when desired. In other words, the user should not need to reauthorize the app to the system each time data is accessed. This property will enable apps to perform functions on behalf of patients and providers, without special effort (for example, checking periodically for new lab results).

Summary. We are so pleased that ONC has and the OMB have gotten to the is stage in which a proposed rule is pending at OMB. We are on the precipice of creating a national-scale apps model for health, based on an API that promotes interoperability and data exchange via substitutable apps. The simple imperatives we enumerate above, could reshape the health IT industry by providing a channel for innovators to distribute and/or sell their software applications by enabling customers to select and integrate EHR-connected apps as easily as they do for smartphones. As the final proposed language implementing the 21st Century Cures Act API provisions is reviewed and prepared for release is decided, we encourage policy-makers to keep all eyes on this prize.

This blog has been cross posted at The Health Care Blog.

(A version of this blog was published by CNBC)

January 30, 2018

Ken Mandl (Twitter @mandl)

Despite spectacular advances in diagnostic imaging, non-invasive surgery, and gene editing, healthcare still faces a lackluster problem: many patients can only get health records from their doctor if the fax machine is working. Even when records are stored electronically, different chunks of every patient’s health information sit in the non-interoperable, inaccessible electronic record systems in different doctor’s offices.  

Anyone who needs her medical files gets them either printed or faxed, or has to log on into separate portals for each doctor and hospital, and even then getting view-only access. View-only apps can’t access data to help patients share information with family and healthcare providers, make decisions, monitor disease, stay on course with medications, or just stay well.

On the positive side, this is changing, sort of. Using the iPhone Health app, patients will soon be able to download and view health records on their phones. On the one hand, don’t get too excited–it will initially only work for patients at a handful of institutions, Android users are still out in the cold, and the data available will be limited. And, some dismiss the impact of Apple’s move because of others’ failures to give patients control of their records.

However, Apple’s move is a decisive and consequential advance in patients’ struggle to get a copy of their own health data. Apple wisely chose to use open, non-proprietary approaches that will float all boats–even for Android users.  

Every patient deserves a ‘bank account’ of her health data, under her control, with deposits made after every healthcare encounter. After my colleagues and I demonstrated an open, free version of a “bank account” to companies in 2006, Google and Microsoft launched similar personally controlled health records — GoogleHealth and Microsoft Healthvault. Walmart and other employers offered our version, Indivo, as an employee benefit. Unfortunately, even these industry giants couldn’t shake loose data from the proprietary computer systems in doctors’ offices, or make the case to patients that curating the data was worth the effort.

But 12 years later, Apple’s product enters healthcare under different circumstances.  A lot more patient data is electronic after a $48 billion federal investment in promoting the adoption of information technology to providers. But those products, mostly older software and purchased at enormous expense, still don’t promote record sharing with doctors or patients.

Recognizing this unacceptable limitation and having received a generous grant comprising a tiny fraction of that federal investment, our team created SMART on FHIR. SMART is an interface to make doctors’ electronic health records work like iPhones do. Apps can be added or deleted easily. The major electronic health record brands have built this interface into their products.

Apple uses SMART to connect the Health app to hospitals and doctors offices. The good news for patients, doctors, and innovators is that Apple chose a standardized, open connection over a proprietary, closed one. This approach lets any other app, whether running on the web,  iPhone, or Android, use that very same interface to connect.

So Apple will compete on value and customer satisfaction, rather than on an exclusive lock on the data. Does Apple’s approach help Americans trying to stay well or manage their conditions? Yes. But only with follow-through by Apple, health systems, technology companies, patient groups, policy makers, and government regulators. The emerging ecosystem’s nuances must be appreciated.

First of all, the floodgates for patient information are at least a crack open and will be very hard to close. As patients gain access to their data, they will recognize it is incomplete and feel frustrated it’s not available everywhere. But, patients in need will drive demand for data access in their role as health consumers.

Secondly, the government is effectively using law and regulations to compel an open interface. By selecting SMART on FHIR, Apple and its healthcare launch partners mark the importance of standardization. A uniform approach is critical for scale. Imagine if every electrical product required a differently shaped 120V outlet. Understanding this, Google, Quest Diagnostics, Eli Lily, Optum, and many other companies are using the same interface to plug into healthcare.

Thirdly, Apple’s first version of health records brings data onto the phone, but from there, like the portals many patients are already familiar with, the data are still “view-only.”  In 2009, I had the chance to meet with Apple’s rockstar Bud Tribble and talk about how the iPhone could serve healthcare. We concluded that crucial data–like the medication list–had to be as easy for iOS developers to use in their apps as contacts and location are now.  I would not be at all surprised if this is the next step in Apple’s journey–making the health records available to iPhone app developers. Here too is an opportunity to chose open interfaces, and to allow patients to export the data to another device.

Lastly, competition in healthcare IT is hot. Amazon, Google, Apple and Facebook all have healthcare divisions.  Apple’s extraordinary hardware, including sensors in the phone and watch, will monitor patients at home.  Google’s artificial intelligence will lead doctors and patients to diagnoses and decisions.  Amazon is rumored to be eying pharmacy management. Facebook has sifted through posts to detect and possibly intervene when users may be suicidal.

There are so many opportunities to compete. Locking up a patient’s data should never be one of them.  

Ken Mandl, MD, MPH directs the Boston Children’s Hospital Computational Health Informatics Program and is the Harvard Medical School Donald A.B. Lindberg Professor of Pediatrics and Biomedical Informatics.