3
PRIORITIES FOR ACCELERATING CDS PROGRESS

Through the project working groups, various implementation challenges were assessed related to widespread adoption and use of equitable, scalable, sustainable, and accountable CDS that can be deployed in large systems as well as single physician practices. Four primary priorities were identified for realizing the vision for CDS that will ultimately make a difference at the level of individual patient-centered care, and also contribute to a learning health system and improved population health:

• Development of CDS content that distills the wealth of information and clinical guidelines into a few action items that will have the biggest impact on patient-centered care.
• Learning from CDS implementing experience, including that related to incorporation into the EHR and delivery to the practitioner in a way that provides optimal support for the recommended clinical decisions.
• Practical strategies for embedding CDS in real-world environments that considers change management, people management, measurement of use, and usability considerations.
• Explication of the value proposition that fosters scale and spread of CDS through the development of clearinghouses and web-based repositories of CDS artifacts that can be shared, evaluated, and continuously improved through feedback from clinicians and patients.

Reports from the workgroups assigned to these challenges, delivered by the workgroup leaders, and the subsequent discussions among all of the workshop attendees were the major focus of the project’s second meeting, and the contents of those reports are discussed in this chapter.

CREATING, MANAGING, AND CURATING CONTENT⁸

One important barrier to widespread CDS use is the relative lack of effective and shareable reference CDS content that can be easily adopted across health care organizations and health IT systems. Beyond medication CDS, there is relatively little effective CDS content that has been disseminated in a widespread manner. There are vendors for CDS interventions including rules, order sets, and documentation templates but these interventions often take significant effort to implement within health care systems because different EHR systems and health care systems utilize different underlying patient data models and CDS integration mechanisms.

Consequently, CDS content creation and implementation usually involves at each health care organization either a laborious configuration of external licensed content or a laborious reinvention of the wheel as the organization creates its own content. Additionally, maintenance and curation of the CDS content usually takes a lower priority compared with meeting requests for new content creation, especially when content is developed and managed locally; as such, existing content often becomes outdated, with corrective action only taken if users identify and report a problem. While the creation of CDS content in-house is an expensive and resource-intensive endeavor, sharing CDS content, either with peers or through the licensing of vendor content, is presently perceived to be equally or more expensive; thus this duplication of effort at each site has persisted.

On the subject of CDS content standardization, there are at least four important technical challenges to sharing CDS content: insufficient standardization of patient data representation; insufficient standardization of CDS knowledge representation; insufficient standardization of CDS integration mechanisms; and a need to align with broader standardization initiatives. With regard to patient data representation, different EHR systems, and in many cases different health care organizations using the same EHR system, differ in how they represent patient data. Because CDS relies on inferencing using patient data, this heterogeneity in patient data representation poses an immense obstacle to sharing CDS. With regard to CDS knowledge representation and CDS integration mechanisms, different EHR systems generally use different approaches, making it extremely difficult to use CDS content developed in one EHR system in another. Moreover, such sharing is often quite difficult even for different health care organizations

using the same EHR system. Current CDS knowledge representation approaches typically have limited capacity to efficiently manage local differences in clinician preferences and workflows. Finally, even if standardization along these technical axes was achieved for CDS, there must be strong alignment with standardization efforts in other aspects of health IT, such as electronic clinical quality measurement or data exchange. Otherwise, the required implementation effort will be increased, and vendor buy-in is likely to be diminished.

At the second meeting, Kensaku Kawamoto began his presentation by reminding the attendees that the Arden Syntax for medical logic modules was developed in 1990 (Hripcsak, 1991) and the Roadmap for National Action on Clinical Decision Support was published, as noted in the previous chapter, in 2006. Even this many years later, sharing of effective CDS is still limited. In the workgroup’s view, there is a great deal of work to build from to reach the desired state of widespread sharing of effective CDS content. The field is at a tipping point.

Given that health care is a business and investment decisions are made based on financial calculations, the workgroup called for a strong business case for CDS content creation. Fee-for-service reimbursement policies, Kawamoto said, are a significant barrier in this regard because they provide no incentive to improve the quality of care through the use of CDS. As an example, he cited the common practice of ordering lumbar magnetic resonance imaging (MRI) for patients with lower back pain, the number of which would likely be reduced with effective CDS. The potential game changers, he added, could be the Centers for Medicare and Medicaid Services’ (CMS) merit-based incentive payment system, alternative payment models, and physician-focused payment models that would make investments in CDS an “existential imperative.” Creating that existential imperative will involve developing a business case for CDS content using incentives, such as sponsored projects and challenges.

Kawamoto and his colleagues recently demonstrated that implementing CDS for sepsis management in the inpatient setting reduced length of stay and produced an average savings of $5,000 per patient, which at his institution would equal savings of $500,000 per year (Lee et al., 2016). Demonstrating this type of return on investment, with hard numbers obtained from demonstrations at multiple institutions using multiple EHR platforms, would produce the impetus for change. While such efforts are likely to focus on demonstrations conducted under the auspices of CMS programs, Kawamoto added that it makes sense to look at instituting financial incentives for the health care system at large.

Brian Alper, founder of DynaMed and vice president for innovations and evidence-based medicine development at EBSCO Health, noted that it should be possible to create a business case even in a fee-for-service payment model.

This could be possible if CDS is viewed as not just a technological feature but also as a service that helps patients make informed decisions about their health care. In his opinion, CMS’s recently approved physician-focused payment model for Medicare may enable this type of valuation for CDS. Joshua Mandel, research scientist in biomedical informatics at Harvard Medical School and in the Department of Biomedical Informatics at Boston Children’s Hospital, added that the incentive for using CDS should focus on the outcomes that CDS can help providers achieve, rather than for simply using CDS. Kawamoto agreed with this idea because it eliminates the argument that such incentives are for a process measure rather than appropriate care and outcomes. Blackford Middleton noted that return on investment should also include some metric for the social goods that accrue with knowledge sharing, which he acknowledged will require new modeling work.

Roberto Rocha, clinical informatics director at Partners Healthcare, remarked that at least some of the return has to accrue to frontline clinicians. Too often, he said, the clinician is making a big investment in terms of the time it takes to input the information that is needed to produce decision support. All the while, this investment may or may not produce a return for that clinician. Kawamoto noted that, at his institution, it is the physicians who make the ultimate yes or no decision when it comes to adding new CDS to the EHR. Usability, added Jeff Cohn, a physician with Broadlands Family Practice, has to be a primary focus for CDS content to reduce alert fatigue and provider resistance. In that regard, he asked if it might be necessary to explore whether there would be a benefit to targeting different clinicians differently. “A primary care physician might need a different type of alert than say a nurse or a specialist, for example,” he said.

Moving on, Kawamoto discussed the second key issue this workgroup addressed: the need for efficient, standards-based CDS content sharing. The main barrier here, according to the workgroup, is that sharing CDS content is either more expensive or perceived to be more expensive than creating content de novo at each institution, with the one exception being CDS relevant to knowledge-based medication-related CDS content. The concern here is that while ONC has sponsored several efforts that have worked as pilot demonstrations, the demonstrations were not as effective when scaled because of a lack of specificity on the information model. Examples of past and present efforts include the Health eDecisions initiative that resulted in the development of standards for knowledge artifacts and CDS services, the ONC- and CMS-sponsored Clinical Quality Framework initiative to harmonize Health eDecision standards with measurements, and the development of the Fast Healthcare Interoperability Resources (FHIR) specification for exchanging health care information electronically. Using FHIR as an

example, Kawamato said that it is enabling promising interoperability projects, but each project requires so many decisions on specific actions to take that “we are not going to get true interoperability unless we solve that issue.”

As is the case with the need to demonstrate a return on investment for CDS deployed across multiple platforms at multiple institutions, CDS developers need to create content based on clear standards. What happens today, said Kawamoto, is that standards are not always defined clearly enough, so a developer will make a decision that enables content to work within systems at the test institutions but are not scalable nationally. The HL7 Clinical Quality Information Work Group initiative is harmonizing decision support standards and those for electronic clinical quality measures. The workgroup recognized that while this effort has made significant strides toward standardization, this work has not yet achieved the necessary level of detail in the standards and how they are applied to clinical decisions, said Kawamoto. He noted that the Clinical Quality Language standard for logic expression has been well-received by the community and CMS is moving forward with an initiative to promulgate this quality measurement program.

Another effort, called OpenCDS, is being implemented by the Department of Veterans Affairs in its eHMP program, but this open-source standard will be useful, said Kawamoto, only for specific use cases or where adding another tab to the EHR makes sense. “In cases where it does, I think this is very powerful, and vendors are adopting it, which makes it very attractive for implementing,” he said. A second program, CDS Hooks, provides decision support services for specific use cases, and there is an effort to harmonize this approach with FHIR.

In Kawamoto’s opinion, the most promising approach is to provide EHR support for CDS. Epic, for example, has a capability called the Best Practice Advisory Web Service that operates within Epic’s native rules authoring environment. His hope is that every vendor could provide a similar web-based CDS content delivery service supported by the native architecture of that vendor’s EHR. The main challenges here, said Kawamoto, relate to performance issues—it currently takes six or seven seconds for the EHR to retrieve and package patient data, send it to the web service, and retrieve and answer.

Middleton noted that while the focus on accessing web services has been on how to enable EHRs to reach out to access those services, research on the EHR context is also needed to define insertion points and hooks. Such research would create spaces in the EHR into which CDS information displays would fit naturally within the context of the provider’s workflow and in a team-based care model. Creating these insertion points and hooks will require additional dialog with EHR vendors, he said. One challenge will be to develop secure application program interfaces (APIs), the set of routines, protocols, and tools

that will enable CDS content developers to create applications that will interact with EHRs, in much the same way that computer operating systems have APIs that enable third-party developers to create external programs that work on top of those operating systems.

With regard to action priorities, the workgroup concluded there is still a need to develop, validate, and adopt standards and tools for CDS sharing, including important building blocks such as value sets and mapping tools. As an example, Kawamoto described a hypothetical situation in which a sepsis decision algorithm requires information on a patient’s white blood cell count. A mapping tool needs to know which internal code denotes white blood cell count.

The second action priority is to seed the marketplace with useful CDS content. Kawamoto explained that justifying the decision to deploy CDS at an institution would be easier if a significant starter set of CDS content, such as for the U.S. Preventive Services Task Force A and B recommendations, existed for immediate inclusion in the EHR. Sumi Sexton, a physician whose practice joined the Privia Medical Group, noted that the expense of acquiring CDS content can be too high for a small practice such as hers.

The content development workgroup’s third key issue was the need to discover and disseminate CDS best practices, with the main barriers being insufficient guidance on the creation of effective CDS and the fact that providers view much of the current CDS as a nuisance because of “alert fatigue.” Kawamoto said that the field has learned a great deal about effective CDS from clinical trials and various meta-analyses, and particularly from the AHRQ Patient-Centered Outcomes Research Decision Support Learning Network described in chapter 2. Nonetheless, he said, more work remains to discover and disseminate best practices. The workgroup called for the nation to make an investment in CDS research, particularly for multisite randomized, controlled trials, and to establish a robust, interoperable CDS marketplace within the context of business incentives to improve care quality. A potential initial place to start addressing the challenges of CDS content development, the workgroup concluded, would be to integrate standards of CDS Web services within EHR vendors’ rule authoring platforms.

LEARNING FROM CDS IMPLEMENTATION

Several comments focused on CDS technical implementation challenges. Currently, there are various implementation approaches across health systems and vendors, resulting in different implementation of each CDS system. With the variability in approaches, there is also little crosscutting evidence on how:

• CDS performs in real-world environments (e.g., whether CDS users provide better care than control groups);
• CDS implementation designs accommodate actual workflow patterns; or how
• CDS implementation models are validated and compared.

Nor are there mechanisms that enable health systems and researchers to regularly monitor and study system use and usability over time to determine what is working and what is missing (e.g., lack of sufficient data to support CDS logic).

While traditionally CDS, including rules engines and interaction software, has been embedded natively as part of the EHR, hybrid systems that utilize cloud-based services in real time are becoming increasingly more common. The development of implementation standards, especially for the cloud, has the potential to reduce the vendor implementation workload and provide common base content for every vendor. However, standards development efforts, especially for standards developed based on real-world experimentation and trial use, are insufficient.

As Scott Weingarten, chair of the technical implementation workgroup reported, this workgroup primarily took an industry or vendor perspective on the issues it was charged to address. The reason for this, he explained, is that every EHR vendor has a product roadmap, and if the vendors are being asked to implement standards that will enable CDS integration, they may have to alter the product roadmap and make a decision about whether to do so, absent a federal mandate. He noted, too, that the workgroup had representatives from six EHR vendors and five content developers accounting for the major shares of their respective markets.

The technical implementation workgroup concluded that the large number of EHR native rule engines, each with its own approach and workflow, creates a difficult environment in which to develop scalable CDS content. Mapping across these different systems is difficult and expensive, and making content changes to reflect new knowledge is perhaps no less difficult. The workgroup then concluded that implementing CDS in a cloud-based environment offers the best opportunity to achieving the desired outcome of scale and spread. Toward that end, using a standards-based, Web API approach makes sense in that it will reduce EHR vendor work, CDS content vendor work, and implementation costs.

Marc Overhage, chief medical informatics officer at Cerner, added that even with a web- or cloud-based CDS service, it will still be necessary to establish local EHR mechanisms for integration. These mechanisms can help access the necessary data, rule sets, and care plans that have to interact with CDS for it to be effective and fit within the workflow of the provider. Todd Rothenhaus, senior

vice president and chief medical officer at athenahealth noted the challenge of keeping web-based systems synchronized with primary sources of information. Based on what he has seen at various health IT conferences, Howard Strasberg, vice president for medical informatics at UpToDate/Wolters Kluwer Health, said it appears that a web-services approach is rapidly gaining traction among EHR vendors.

When it comes to deploying a web-based system, standards will be essential for success on a large scale, the workgroup concluded. In addition to enabling scale and spread across multiple EHR and technology platforms, standards are necessary to create systems that are sustainable, maintainable, and updatable. They will also reduce maintenance costs, Weingarten noted. Members of the workgroup agreed on the importance of standards but also noted that standards can be constraining. At least one member of the workgroup was concerned that standards are not yet mature enough for implementation.

James Tcheng wondered if there is a need for what he called an “uber authority,” a single source of truth from which all CDS derive. From his perspective as chair of the digital steering committee of the American College of Cardiology, Tcheng noted that this organization, like other specialty colleges in the medical profession, is responsible for authoring clinical guidelines. However, in his opinion, the College has struggled to convert these paper guidelines into computable guidelines, in large part because there is no standards-based pathway for creating computable guidelines that could be used in any EHR context. “If I was on the authoring side of trying to create CDS, I do not want to work with 35 different vendors and 35 additional institutions or enterprises in the CDS field. I would like to create a knowledge representation that others can then consume,” said Tcheng.

Jonathan Teich, a practicing emergency medicine physician at Brigham and Women’s Hospital, noted that when he led the AHRQ Roadmap for National Action on Clinical Decision Support project, vendors supported the idea of a standard framework set by a governmental authority but were concerned with how it would survive over time. He added that one advantage of FHIR is that it has taken off by itself and become a de facto standard. David Bates, senior vice president and chief innovation officer at Brigham and Women’s Hospital, suggested that a central authority for standards to act as a clearinghouse could be in order, but not for content. In fact, he said, having the specialty societies set guidelines is not ideal because they inevitably develop recommendations beneficial to their members, not necessarily patients.

Vindell Washington, who at the time of the meeting was the National Coordinator for Health Information Technology at ONC, said that ONC

established the Interoperability Standards Advisory as a public-private partnership rather than create a set of standards itself. He noted, though, that the Advisory has come under pressure to be more declarative and to develop a rating system for these standards.

In considering priorities for action, the implementation workgroup noted the importance of expanding the evidence base for implementation science around CDS. It is critical, the workgroup concluded, to conduct research and evaluation of real-world deployments of CDS in clinical environments spanning the gamut from small physician practices to large health systems and across a variety of workflows and to determine if there are models of implementation that are more efficient and successful than others. Research is also needed to understand usability, physician satisfaction, and the effect of CDS deployment on physician burnout over time, as well as whether each individual CDS intervention improves care outcomes. Some CDS, for example, may not produce measure-able benefits over time and should therefore be removed from cloud-based (and local) repositories.

Another priority action the workgroup identified was the need for the development of standards related to cloud-based implementation. This effort should involve federal and industry partners. Examples of such standards include the SMART on FHIR^®, an open-source set of specifications for integrating apps within EHRs and health information exchanges, and CDS Hooks.⁹ Additional priorities included sharing best practices about implementation and integration, and considering how the field can make the case that CDS will help reach incentives already in place for providing high quality and less costly or more cost-effective care. As Weingarten noted, there was disagreement between the workgroup participants on whether or not CDS implementation should be driven by the market or by incentives for using CDS.

PRACTICAL STRATEGIES FOR EMBEDDING CDS

To be effective, CDS must be integrated into, and supportive of, the clinical work environment. This includes attending to timing the presentation of interventions, providing concise yet sufficient information to drive action clearly on the screen, and making it easy for the user to understand the recommendations and take optimal actions, all without overly interrupting clinical workflow. There has been much discussion about usability principles, but not enough practical guidance and examples. Nonetheless, evidence suggests that providers are more

likely to follow suggestions when they are delivered in ways that have better usability. Currently, CDS works well in some institutions but has low or even negative impact in others. Factors that need to be carefully addressed during implementation include:

1. Design/workflow factors
   • CDS that is too nonspecific and generates too many low-impact alerts;
   • Failure to show the clinical user the underlying data and rationale leading to a clinical recommendation; and
   • CDS that excessively interrupts workflow and demands a change in plan, as opposed to providing information earlier in the care process before decisions have been made.
2. Lack of standard/repeatable implementation across institutions
   • No easily-accessed store of good exemplars of CDS design that can be used as reference for new implementations;
   • Basic operational components—including triggers, notifications, and action items—are known to some experts, but not widely known and not easily available as standard sets and templates; and
   • No agreement across EHR vendor systems about where to place available CDS trigger points; some systems, for example, provide drug dose checking as soon as an order is entered, while others wait until a full session of multiple orders is ready to be signed.
3. Communication/implementation factors
   • Failure to understand the range of stakeholders whose workflow could be affected by an intervention; and
   • Failure to include users and stakeholders (clinicians, quality leaders, and often patients/consumers) in design and configuration early in the process.
4. Measurement limitations
   • EHRs and CDS systems not providing sufficient and easily accessible data to allow measurement of whether given CDS interventions are being used, are triggering consistently and appropriately, are generating recommendations, and are being accepted or excepted appropriately; and
   • Difficulty in determining whether a given CDS intervention has impacted health decision making, outcomes, and patient experience; thus, it is difficult to distinguish effective CDS interventions from ineffective ones.

5. Priority-setting factors
   • Providing optimal CDS has often not been a priority for vendors, compared to supporting documentation requirements for the current regulatory and reimbursement environment; and
   • Patient-facing CDS and shared decision-making tools have not been prioritized, which further slows the understanding and development of effective CDS of this type.

While there have been many efforts to codify CDS logic, the field has paid less attention to making it easy, shareable, implementable, usable, trackable, and measurable, said Jonathan Teich, who chaired the operations workgroup. The development of vendor-specific APIs represent a good step in the right direction, he said, and this workgroup focused on actions needed to make the CDS more universal and practical, he explained.

Before presenting the operations workgroup’s findings, Teich briefly described the typical CDS process (Figure 3–1) that starts with a trigger, some piece of data that starts the logic process resulting in either no action or the decision to present something to the user and perhaps notify the user if the user is offline at the time. The CDS presentation can be an alert, an order set, a care plan facilitator, or decision tool. The EHR supplies data that informs the CDS.

With regard to operational problems and priorities, the workgroup noted that triggers need to be more precise to avoid alert fatigue, and presentations need to not only suggest an action but also state the reasons for a recommended intervention, including relevant data and information supporting the action. According to the workgroup, action items need to be understandable and presented in a form that can be readily absorbed by the user. In addition, there should be an effort to collect and codify examples of good practices to educate CDS developers, and the field needs to publish usability evaluations of EHR and CDS systems.

The workgroup referenced the recommendations of the National Quality Forum (NQF) expert panel, which concluded there are approximately a dozen trigger points in the standard workflow that are appropriate places to initiate live CDS processing (National Quality Forum, 2010). Some EHR vendors, said Teich, have provided trigger points at some of those places, but there is little consistency across vendors. “It is clear that if all of the major vendors of EHRs had a consistent set of trigger points, it would be easier to write both embedded and cloud-based CDS to support that,” said Teich. “This is something that I would suggest could be standardized that would facilitate CDS without actually controlling what is in the CDS itself.”

The ability to extract discrete data from an EHR and send it to CDS is fairly well developed, the workgroup noted, but there is still work to be done on the reverse process (i.e., CDS returning a recommended action to the EHR). Vendors, Teich explained, need more proof that these actions are supported by evidence and have been validated. They have also expressed concern that CDS input does not adversely affect the fundamentals of the EHR. The workgroup suggested that just as with trigger points, the field could develop a set of standard actions that CDS systems would ask the EHR to perform. Having services in the EHR to support such standardized actions could make it easier to reuse and spread both embedded and cloud-based CDS. The workgroup also suggested that the field should create model CDS built on core elements that would include an order set, several different alerts, and a clinical pathway that users could modify for specific clinical conditions. CDS needs to be specifically tested in an electronic environment, as paper-based systems invariably require some degree of judgment in application, whereas CDS, by definition, is triggered not by judgment but by data.

In terms of exemplary CDS, the workgroup concluded that much of the success results from high-caliber implementation and communication. Too often, stakeholders are not involved early in the implementation process, governance is inconsistent, and patients are not involved where appropriate. It is important, said Teich, “to make sure that the computer is not making policy before people know what the policy is and that people have a reasonable place to come back

with feedback from something that is or is not working well.” Collaboration among CDS proponents, users, and vendors must become routine, the workgroup noted, and the field should collect best practices to educate and facilitate implementation.

Going forward, measurement is essential. “We want to make sure that the CDS is working,” said Teich. To optimize CDS and increase adoption and acceptance, it will be critical to determine which interventions are firing at the appropriate times and are then accepted by the clinical care team and patients and changing care for the better. This capability will be important at both the local and national scale if the goal is to reduce the burden on providers and health systems to each identify important lessons on their own.

The operations workgroup noted that while developers are relying more on usability science, they do not have specific examples of what works best in the context of how to reduce alert fatigue, how to trigger alerts at appropriate times, and how to create alerts that are not so full of information as to be unreadable on the computer monitor. Another deficit currently is that some systems lack the ability to follow through easily on a recommended action, such as ordering a particular test or prescribing a specific medication. A potential solution, the workgroup noted, would be to develop a resource of good practices that systems and providers could use when starting to implement their own CDS systems. Another solution would be to evaluate the usability of EHR and CDS systems in enough depth to enable users and developers to identify which specific tasks their systems do well and which ones need improvement. One workgroup member, David Bates, noted that even when he accepts an alert, he cannot tell if the system actually accepted and followed through on any actions he might take as a result of the alert, something he counts as a usability issue.

These standards would not dictate specifics of a CDS but act as templates that would provide a common starting place and format. The example Teich gave was how all iPhone apps have a similar look to them. Certification may be an avenue to encourage or facilitate standardization. Standardization would help enable reusable CDS in that way that the Pyxis medication dispensing system enables a new drug to be added in a simple and straightforward process. Currently, Teich said, a new CDS intervention requires a year’s worth of committee work to implement in an EHR. “It is very hard to leverage good work done elsewhere,” he said. “It is very hard for hospital A to make an intervention that actually works and for hospital B to pick it up.” The idea is not so much to have one repository of CDS but one standard for any repository.

With regard to implementation, the workgroup stressed the importance of communication and getting all stakeholders involved early in the implementation

process. It may be useful, the workgroup noted, to convene vendor-user group meetings to share concerns and create priority lists for vendors. The workgroup also suggested that EHR vendors include logging and analysis tools in EHRs to determine how often and under what circumstances CDS fired and what the response of the clinical staff was to an alert. As a final note, this workgroup agreed with the other workgroups on the need for payment and regulatory reform to change the financial incentives

During the ensuing discussion, Middleton noted that a missing operational piece in most systems is knowledge management—providing information about the provenance of the knowledge going into CDS and having the ability to update that information when needed. He also suggested that one approach to standardization of CDS would be for EHR vendors to create a style guide so that CDS developers can create alerts that make it easier to navigate within a given EHR. Sexton suggested that alerts could be tied to continuing medical education credits and board recertification, an idea that several meeting participants also endorsed.

EXPLICATING THE CDS VALUE PROPOSITION¹⁰

As was apparent throughout the other presentations, despite a substantial national investment in health IT, there continues to be a struggle to articulate and demonstrate the health IT value proposition. Looking at the issue of CDS use from a macro level, the workgroup focused on the CDS marketplace noted that the articulation, dissemination, and adoption of an industry-scale value case for CDS is critical to its long-term use. Some prior experience in this respect is offered in the development of a tool developed by the National Academy of Medicine’s Digital Learning Collaborative¹¹ in conjunction with the Healthcare Financial Management Association¹² (HFMA) to provide a systematic mechanism for assessing the costs and benefits of EHR implementation, alternate approaches to implementation, and models of use (Adler-Milstein et. al, 2014).

There are a number of challenges to creating a competitive market for CDS. A 2015 analysis of the CDS marketplace (Figure 3–2) conducted for ONC (Discern Health, 2015) found there was a good supply of new knowledge generated by

the nation’s biomedical research enterprise but a lack of structured formats and processes to extract a structured output that could represent core knowledge elements. As a result, the knowledge that would go into CDS needs to be extracted manually from the literature and inputted into CDS manually, an expensive and time-consuming process that must be repeated every time research generates new knowledge. This analysis also found that demand was weak because of cost of customization and maintenance, as well as lack of awareness and mistrust of CDS. The remedy for the disconnect between the supply of information and demand for that information in the form of CDS was to create a two-sided exchange or marketplace for buyers and sellers of CDS. In the ideal state, standards and automation will support efficient translation of knowledge into CDS, reducing costs. On the demand side, efforts to raise awareness of the benefits of CDS and increase trust, coupled with the lower cost of creating CDS and the availability of standards and automation to reduce the costs of customizing and maintaining CDS, will stimulate demand. At the same time, competition in the marketplace would lead to better products and multiple options, further lowering costs, which together with feedback that contributes to a continuously learning system, would further drive demand.

When presenting the scaling and spreading the value proposition workgroup’s findings, Middleton noted that because of its operational and technical issues, CDS is at times still viewed as a hindrance to clinical care instead of as a tool

for improved quality of care. Improvements in implementation, content, and operations could increase the value of CDS tools, as would increased interoperability with EHRs and other workflow systems. The hassles of adoption and sharing must also be reduced. Clinicians and institutions may favor robust CDS that is also integrated and bundled with EHR systems, particularly when they meet the “five rights” of effective CDS discussed in chapter 2. An additional challenge is that a sales model problem exists. According to Overhage at the third meeting, vendors do not know how to sell $100,000 items (the cost of standalone CDS); for health systems, it is easier to buy an entire health IT system that includes knowledge instead of only purchasing one CDS resource. A clear business case is needed in order to get a toolset or vision in the company to create an infrastructure.

However, according to Middleton, a major barrier to CDS adoption (even if interoperability reduced hassles in adoption, and trust in CDS resources are achieved) is that the health care industry is not fundamentally designed to focus on or reward optimal decision-making focused on quality and safety. Health care financing is heavily weighted toward documentation of observations as the basis for determining and auditing payments. The result is that EHRs and health care transactions are developed to support documentation of specific observations and not support documentation or facilitation of decision-making. Observations useful for supporting payment but not useful or less useful for decision-making or patient care are valued and prioritized, while observations that are most relevant to decision-making and documentation of the decision-making are not valued or prioritized. Although national attention is gravitating toward payment incentives that reward greater effectiveness and efficiency in outcomes—the aims of CDS—actual practices are still heavily oriented to fee-for-service.

According to the scaling and spreading the value proposition workgroup, health care service provision is more optimally done with a focus on decision-making. Patients seek guidance in health care decision-making. Though sometimes fully delegating to the health care professional, patients often actively participate in their own health care decision-making. Aligning decision-making with individual values and preferences is important for patient satisfaction, personalized health care, patient engagement, patient adherence, and quality of health care in the view of the recipient.

To address the numerous challenges to distribution, the workgroup developed guiding principles for creating a marketplace supportive of CDS being widely adopted by health systems and providers. In such a “post-EHR world,” said Middleton, vendors need access to externalized data to keep their customers happy

and enable them to provide appropriate care and to externalized knowledge-based tools and services. The guiding principles articulated by the workgroup are:

• CDS distribution should be anchored in the basic principles of being actionable/reportable, integrated into the workflow, interoperable, and available as a web service.
• Each CDS intervention should have a value proposition for each of the different purposes and variety of sizes and types of health care organization, across care settings, for which it is targeted. This value statement must include how CDS will benefit the care of a given patient both at the time of care and extending over time through the accrual of secondary benefits.
• Quality reporting as a byproduct of CDS tools should be expanded and measurement should be embedded as a tracer and transparent byproduct of CDS technology.
• CDS should support providers’ success as health care delivery and payment models increasingly emphasize outcomes as opposed to volume of services performed.
• Health care financing should be reoriented to reward providers for documenting decision-making based on CDS, in addition to clinical observations, the impact of those decisions on patient outcomes, and the value patients place on those outcomes.
• Industry should work closely with federal partners, patients and families, and representatives from professional societies to advance awareness, understanding, and application of CDS strategies and address legal barriers to CDS use and knowledge sharing.
• Industry, in collaboration with multiple partners, should take a lead in developing industry regulation and certification efforts as they relate to assessing and defining an appropriate regulatory framework for CDS.

With regard to this last item, Middleton expressed his concern about efforts at the Food and Drug Administration (FDA) to consider software as a medical device and regulate it as such. In his opinion, CDS is not autonomous, that is, there is a learned intermediary who has to decide whether to act on CDS-generated alert. He acknowledged the argument that a clinician may not always have time to exercise his or her judgment about a given alert, and said that careful thought needs to be given about the appropriate boundary conditions for when a regulated CDS service should be considered a medical device. In this respect, the 21st Century Cures Act provides clarification that software/data used within the context of an electronic health record does not necessarily

constitute an FDA-regulated device, thereby supporting use of the electronic health record as a vehicle for decision support.

REFERENCES