Improve healthcare data capture at the source to build a learning health system
Studies estimate that only one in 10 recommendations made by major professional societies are supported by high-quality evidence. Medical care that is not evidence-based can result in unnecessary care that burdens public finances, harms patients, and damages trust in the medical profession. Clearly, we must do a better job of figuring out the right treatments, for the right patients, at the right time. To meet this challenge, it is essential to improve our ability to capture reusable data at the point of care that can be used to improve care, discover new treatments, and make healthcare more efficient. To achieve this vision, we will need to shift financial incentives to reward data generation, change how we deliver care using AI, and continue improving the technological standards powering healthcare.
The Challenge and Opportunity of health data
Many have hailed health data collected during everyday healthcare interactions as the solution to some of these challenges. Congress directed the U.S. Food and Drug Administration (FDA) to increase the use of real-world data (RWD) for making decisions about medical products. However, FDA’s own records show that in the most recent year for which data are available, only two out of over one hundred new drugs and biologics approved by FDA were approved based primarily on real-world data.
A major problem is that our current model in healthcare doesn’t allow us to generate reusable data at the point of care. This is even more frustrating because providers face a high burden of documentation, and patients report repetitive questions from providers and questionnaires.
To expand a bit: while large amounts of data are generated at the point of care, these data lack the quality, standardization, and interoperability to enable downstream functions such as clinical trials, quality improvement, and other ways of generating more knowledge about how to improve outcomes.
By better harnessing the power of data, including results of care, we could finally build a learning healthcare system where outcomes drive continuous improvement and where healthcare value leads the way. There are, however, countless barriers to such a transition. To achieve this vision, we need to develop new strategies for the capture of high-quality data in clinical environments, while reducing the burden of data entry on patients and providers.
Efforts to achieve this vision follow a few basic principles:
- Data should be entered only once– by the person or entity most qualified to do so – and be used many times.
- Data capture should be efficient, so as to minimize the burden on those entering the data, allowing them to focus their time on doing what actually matters, like providing patient care.
- Data generated at the point of care needs to be accessible for appropriate secondary uses (quality improvement, trials, registries), while respecting patient autonomy and obtaining informed consent where required. Data should not be stuck in any one system but should flow freely between systems, enabling linkages across different data sources.
- Data need to be used to provide real value to patients and physicians. This is achieved by developing data visualizations, automated data summaries, and decision support (e.g. care recommendations, trial matching) that allow data users to spend less time searching for data and more time on analysis, problem solving, and patient care– and help them see the value in entering data in the first place.
Barriers to capturing high-quality data at the point of care
- Incentives: Providers and health systems are paid for performing procedures or logging diagnoses. As a result, documentation is optimized for maximizing reimbursement, but not for maximizing the quality, completeness, and accuracy of data generated at the point of care.
- Workflows: Influenced by the prevailing incentives, clinical workflows are not currently optimized to enable data capture at the point of care. Patients are often asked the same questions at multiple stages, and providers document the care provided as part of free-text notes, which are frequently required for billing but can make it challenging to find information.
- Technology: Shaped by incentives and workflows, technology has evolved to capture information in formats that frequently lack standardization and interoperability.
Plan of Action
Recommendation 1. Incentivize generation of reusable data at the point of care
Financial incentives are needed to drive the development of workflows and technology to capture high-quality data at the point of care. There are several payment programs already in existence that could provide a template for how these incentives could be structured.
For example, the Centers for Medicare and Medicaid Services (CMS) recently announced the Enhancing Oncology Model (EOM), a voluntary model for oncology providers caring for patients with common cancer types. As part of the EOM, providers are required to report certain data fields to CMS, including staging information and hormone receptor status for certain cancer types. These data fields are essential for clinical care, research, quality improvement, and ongoing care observation involving cancer patients. Yet, at present, these data are rarely recorded in a way that makes it easy to exchange and reuse this information. To reduce the burden of reporting this data, CMS has collaborated with the HHS Assistant Secretary for Technology Policy (ASTP) to develop and implement technological tools that can facilitate automated reporting of these data fields.
CMS also has a long-standing program that requires participation in evidence generation as a prerequisite for coverage, known as coverage with evidence development (CED). For example, hospitals that would like to provide Transcatheter Aortic Valve Replacement (TAVR) are required to participate in a registry that records data on these procedures.
To incentivize evidence generation as part of routine care, CMS should refine these programs and expand their use. This would involve strengthening collaborations across the federal government to develop technological tools for data capture, and increasing the number of payment models that require generation of data at the point of care. Ideally, these models should evolve to reward 1) high-quality chart preparation (assembly of structured data) 2) establishing diagnoses and development of a care plan, and 3) tracking outcomes. These payment policies are powerful tools because they incentivize the generation of reusable infrastructure that can be deployed for many purposes.
Recommendation 2. Improve workflows to capture evidence at the point of care
With the right payment models, providers can be incentivized to capture reusable data at the point of care. However, providers are already reporting being crushed by the burden of documentation and patients are frequently filling out multiple questionnaires with the same information. To usher in the era of the learning health system (a system that includes continuous data collection to improve service delivery), without increasing the burden on providers and patients, we need to redesign how care is provided. Specifically, we must focus on approaches that integrate generation of reusable data into the provision of routine clinical care.
While the advent of AI is an opportunity to do just that, current uses of AI have mainly focused on drafting documentation in free-text formats, essentially replacing human scribes. Instead, we need to figure out how we can use AI to improve the usability of the resulting data. While it is not feasible to capture all data in a structured format on all patients, a core set of data are needed to provide high-quality and safe care. At a minimum, those should be structured and part of a basic core data set across disease types and health maintenance scenarios.
In order to accomplish this, NIH and the Advanced Research Projects Agency for Health (ARPA-H) should fund learning laboratories that develop, pilot, and implement new approaches for data capture at the point of care. These centers would leverage advances in human-centered design and artificial intelligence (AI) to revolutionize care delivery models for different types of care settings, ranging from outpatient to acute care and intensive care settings. Ideally, these centers would be linked to existing federally funded research sites that could implement the new care and discovery processes in ongoing clinical investigations.
The federal government already spends billions of dollars on grants for clinical research- why not use some of that funding to make clinical research more efficient, and improve the experience of patients and physicians in the process?
Recommendation 3. Enable technology systems to improve data standardization and interoperability
Capturing high-quality data at the point of care is of limited utility if the data remains stuck within individual electronic health record (EHR) installations. Closed systems hinder innovation and prevent us from making the most of the amazing trove of health data.
We must create a vibrant ecosystem where health data can travel seamlessly between different systems, while maintaining patient safety and privacy. This will enable an ecosystem of health data applications to flourish. HHS has recently made progress by agreeing to a unified approach to health data exchange, but several gaps remain. To address these we must
- Increase standardization of data elements: The federal government requires certain data elements to be standardized for electronic export from the EHR. However, this list of data elements, called the United States Core Data for Interoperability (USCDI) currently does not include enough data elements for many uses of health data. HHS could rapidly expand the USCDI by working with federal partners and professional societies to determine which data elements are critical for national priorities, like vaccine safety and use,, or protection from emerging pathogens.
- Enable writeback into the EHR: While current efforts focused on interoperability have focused on the ability to export EHR data, developing a vibrant ecosystem of health data applications that are available to patients, physicians, and other data users, requires the capability to write data back into the EHR. This would enable the development of a competitive ecosystem of applications that use health data generated in the EHR, much like the app store on our phones.
- Create widespread interoperability of data for multiple purposes: HHS has made great progress towards allowing health data to be exchanged between any two entities in our healthcare system, thanks to the Trusted Exchange Framework and Common Agreement (TEFCA). TEFCA could allow any two healthcare sites to exchange data, but unfortunately, participation remains spotty and TEFCA currently does not allow data exchange solely for research. HHS should work to close these gaps by allowing TEFCA to be used for research, and incentivizing participation in TEFCA, for example by making joining TEFCA a condition of participation in Medicare.
Conclusion
The treasure trove of health data generated during routine care has given us a huge opportunity to generate knowledge and improve health outcomes. These data should serve as a shared resource for clinical trials, registries, decision support, and outcome tracking to improve the quality of care. This is necessary for society to advance towards personalized medicine, where treatments are tailored to biology and patient preference. However, to make the most of these data, we must improve how we capture and exchange these data at the point of care.
Essential to this goal is evolving our current payment systems from rewarding documentation of complexity or time spent, to generation of data that supports learning and improvement. HHS should use its payment authorities to encourage data generation at the point of care and promote the tools that enable health data to flow seamlessly between systems, building on the success stories of existing programs like coverage with evidence development. To allow capture of this data without making the lives of providers and patients even more difficult, federal funding bodies need to invest in developing technologies and workflows that leverage AI to create usable data at the point of care. Finally, HHS must continue improving the standards that allow health data to travel seamlessly between systems. This is essential for creating a vibrant ecosystem of applications that leverage the benefits of AI to improve care.
This memo produced as part of the Federation of American Scientists and Good Science Project sprint. Find more ideas at Good Science Project x FAS
By better harnessing the power of data, we can build a learning healthcare system where outcomes drive continuous improvement and where healthcare value leads the way.
In this unprecedented inflection point (and time of difficult disruption) for higher education, science funding, and agency structure, we have an opportunity to move beyond incremental changes and advocate for bold, new ideas that envision a future of the scientific research enterprise that looks very different from the current system.
Assigning persistent digital identifiers (Digital Object Identifiers, or DOIs) and using ORCIDs (Open Researcher and Contributor IDs) for key personnel to track outputs for research grants will improve the accountability and transparency of federal investments in research and reduce reporting burden.
Research funding agencies should apply the content of grant applications to AI tools to predict the future of scientific and technological breakthroughs, enhance peer review, and encourage better research investment decisions by both the public and the private sector.