During the final session of the workshop individual participants reflected on the day’s presentations and discussions and discussed actions they felt were important to progress in the areas discussed. The suggestions made by individual participants covered six broad thematic areas: improving awareness and gap assessment of existing data sources; improving the quality, patient orientation, and utility of data input; improving the access, tools, and capacity for data analysis; ramping up the involvement and engagement of the patients and the public for improved clinical data; building a clinical data learning utility; and developing clarity on the governance needed.
CURRENT DATA SOURCES: BETTER AWARENESS AND ASSESSMENT
A number of speakers raised the issue of the need for a better understanding of what data sources exist, their characteristics, their relationships to each other, and the implications of these details on the uses of the data (see Chapters 2-5).
Resource mapping. Discussion during the final session yielded several suggestions from individual attendees on how to work toward getting a better understanding of digital health data sources. Assembling a taxonomy of digital health data sources with descriptors to better understand what sources are out there and their specific characteristics was suggested by a few participants as a potential first step toward this goal.
Utility mapping. It was suggested that the mapping process should also
include not just taxonomy and inventory of data sources, but assessment of how high-priority questions and issues map to existing sources and methods, including annotation of sources used in research studies.
DATA INPUT: IMPROVE PATIENT ORIENTATION, QUALITY, AND UTILITY
As emphasized in Crossing the Quality Chasm in 2001, patient-centered care has been highlighted as a central component of quality health care (IOM, 2001). Extending this notion to digital health information was a frequent theme in workshop discussions.
Information patients care about. Several participants and speakers emphasized collecting information that patients care about, and including information on wellness and productivity, as a first step toward this goal. Similarly, increasing the inclusion of patient-reported information in the digital health data utility was highlighted as a priority. Development, validation, and encouragement of the use of patient-reported preferences, symptoms, care-process measures, and outcomes were called out as potential important components of this strategy.
Usability. Improving the usability of health and biomedical information technology and prioritizing information collection were strategies suggested to minimize the burden imposed on data collectors. Identifying and eliminating the collection of low-value data, as well as automating data collection, whenever appropriate, were suggested as potential approaches.
Contextual tagging. Maintenance of the provenance and context of the data was also highlighted as an important issue, particularly when data is used for a purpose other than that for which it was collected. The use of metadata tagging and strategies to enable access to full original context (e.g., on place, time, person/SES) were called out as potential approaches.
Core elements. In order to make progress on the goal of improving data quality, the development of more standardized digital health data definitions and representations was highlighted for attention. In particular, several participants emphasized the need for development of a set of core minimum standardized data elements to provide timely essential information on cost, quality, and health status and trends, available across institutions and geographic areas and designed to harmonize funder data set requirements. Some participants felt that inclusion of these core elements as part of the certification process was an effective way to further progress.
DATA ANALYSIS: IMPROVE ACCESS, TOOLS, AND CAPACITY
Only through analysis and use of digital health data will its full potential be realized (IOM, 2012). Improving the analytic tools and capacity necessary for learning were common themes in workshop discussions.
Toolsets. Specifically, the creation of toolsets that would expand access to tools and applications beyond the traditional research community and open opportunities for analysis and learning was cited as a potential approach with some precedence in other areas of science.
Curation. Strategies and methods to curate data sources in an ongoing way were also suggested. The need for better metrics to measure data quality and utility, in context-appropriate ways, was also discussed. Some participants suggested that these metrics could focus on the impact of information collection and input processes on the data; for example, data collection in the course of routine care through an EHR versus as part of a clinical trial.
Data integration. Several participants asserted that putting patients at the center of digital health data also included facilitating the integration of their data across the several facets of health, including with public health information and other sources, some of which may be external to health care. Strategies for data integration, in particular, including public health data, were suggested as a necessary first step toward this goal. A related concept of triangulating several data sources to improve predictive accuracy was also mentioned by several speakers as an important advantage to having large amounts of diverse data.
PUBLIC AND PATIENT ENGAGEMENT: RAMP UP INVOLVEMENT
Many participants stressed that successfully engaging stakeholders is crucial for fully realizing the learning and improvement potential of digital health data. Whether a data donor, collector, or user, a patient, clinician, public health official, or researcher, all stakeholders have unique, and changing, roles to play.
Patient voice. Drawing further from the notion of collecting and including information patients care about, many participants cited the need for a strong strategy for building the capacity for direct patient engagement. Specific approaches included the development and refinement of portals, and the inclusion of patient preferences and other patient-sourced data.
Trust. Building trust among stakeholders was a common denominator in issues identified to take advantage of expanding capacity for continuously learning health care. Several discussants noted that in order to create and nurture this trust, stakeholders must feel that their participation in data
collection and use processes are respectful of their efforts, their privacy, and responsive to their needs.
Regulatory reform. Development of mutual understandings of expectations for confidentiality, privacy, and security were highlighted as key to building and maintaining strong stakeholder support in the rapidly evolving environment of social media, increased availability of information online, and the growing integration of genomics into clinical care and diagnostics.
Presentation. Increasing the usefulness of data to patients, and other stakeholders, through the use of user-appropriate data presentation techniques, including visualization, was suggested by several workshop participants.
Health literacy. A few participants cautioned that efforts to improve understanding through raising awareness and targeted strategies at different health literacy levels will be necessary to facilitate these discussions.
Culture of participation. Given this changing environment, the suggestion was made to empower potential data donors (notably patients) with the option and ability to donate their information for use. Along similar lines, the idea of studying the benefits and risks of patient-requested portable identifiers was suggested as a way to make progress on the issue of identity resolution and data linkage, and a first step toward developing a strategy for their development and application.
BUILDING A CLINICAL DATA LEARNING UTILITY
Throughout workshop presentations and discussions, some speakers and participants stressed the need to harness the potential for learning from the digital health data utility. The challenges and opportunities afforded by the increasing scale of data available for learning informed many of these discussions.
Innovative methods. The development of methods using EHRs as a data source and performing observational studies on big data were highlighted as specific needs. In particular, the development, validation, and use of predictive models to inform health-data uses, including risk interpretation by individuals, was singled out as holding great promise. Noting that most digital health data is in unstructured formats, the potential for learning from this data through natural language processing (e.g., IBM’s Watson) was highlighted by several workshop participants. An emphasis on the need for the development and application of reasoning and inference tools was highlighted as a potential priority going forward.
Distributed approaches. Given the importance of privacy and security in the collection and use of patient health data, presentations and discussions frequently touched on the advantages of distributed data approaches
and the need to further develop and pilot the policies, analytic methods, and technologies associated with their use.
Engaging bias. Several challenges and barriers to learning from the digital health data utility were cited, including uncertainty about the completeness and reliability of many data sources, as well as the presence of multiple forms of bias. The need for detailed expert assessments of the implications of bias on analyses, as well as in the new context of the very large datasets now emerging, was suggested by several workshop participants.
Core elements. The identification and application of a set of minimum data elements to provide information on cost, quality, health status, and health trends was suggested, by several discussants, as a critical component to accelerating progress on learning from the health data utility. Reform of regulatory frameworks to encourage structured collection, assessment, and use of routinely collected data, in order to facilitate and support this learning, was highlighted by some participants as an important first step.
Greater clarity on governance, both in terms of what it would look like and the issues for engagement, specifically in terms of access and sustainability, was a theme echoed in many workshop discussions.
Domains. Some participants pointed to a need to identify key domains for which governance structures are necessary to accelerate the evolution of the digital data utility, and begin to catalyze their engagement.
Access and ownership. Suggested approaches to ensuring participation included enabling broader access to data sources and ensuring that the flow of information is multidirectional. This democratization of roles could facilitate the engagement of the issue of data ownership, broaden sources of input, exhibit the potential of information use to meet stakeholder needs, and demonstrate the value of the collection and use of the data.
Business model. There is a need for a better understanding of both the costs and benefits associated with the uses of digital health data for learning and continuous improvement. Quantitative and qualitative approaches to insights on how information might be leveraged to increase health benefits and minimize associated costs from the perspectives of the many diverse stakeholders were highlighted by some participants as an important first step on this count. Additionally, the application of analytics for patient panel management and to support pay for performance payment initiatives such as ACOs were cited by individual participants as examples of areas of promise for establishing sustainable efforts.