In 2013, an expert workshop was convened, by the US National Science Foundation, to establish the research agenda for the Learning Healthcare System (Friedman, Rubin et al. 2015). It identified 106 research questions around four system level requirements that a Learning Healthcare System must satisfy:
1. A LHS trusted and valued by all stakeholders
2. An economically sustainable and governable LHS
3. An adaptable, self-improving, stable, certifiable and responsive LHS
4. A LHS capable of engendering a virtuous cycle of health improvement
The coming years are likely to see progress across each of these requirements. This section highlights some of the areas considered to be particularly important.
Much of the discussion in this report may seem like science fiction to clinicians who still work in organisations that use paper clinical notes. In the US, there has been significant adoption of electronic health records over the last five years. The Affordable Care Act and Meaningful Use incentives have meant that adoption has been more complete and that systems have enjoyed more functionality than in the UK (Bates 2015).
The UK enjoys world-leading use of primary care EHRs, but is behind in the adoption of such systems in secondary care. NHS England have set a goal for the NHS to be “paper-free at the point of care” by 2020 (National Information Board 2015). Progress against this goal is being embedded in the commissioning framework and the quality inspection regime, which will make compliance a high priority. There is likely to be significant progress on this prerequisite of Learning Healthcare Systems. This is expected to bring multiple near term benefits (Payne, Corley et al. 2015, Bates 2015):
- Improved medication safety
- Easier to identify patients who should be included on a more appropriate pathway
- More efficient tracking of care for patients with chronic conditions
- Easier collection of data on standards of care
- Potentially increased efficiency. For example, automation and prompts regarding tests due for preventative measures
These are necessary developments, but they are not sufficient to enable Learning Healthcare Systems. There is a risk that simply creating electronic versions of paper notes will be a costly diversion from creating true sociotechnical Learning Healthcare Systems that can support the use cases outlined in this report.
Participants noted significant uncertainty about what progress might be made in this field over time. They believe that the next five years will see, increasing efforts to learn from routine data, improvement of the related user interfaces, further development of outcome measures and outcomes based reimbursement, progress on interoperability and more acceptance of data sharing by patients (Akerman 2015, Etheredge 2015, Rubin 2015).
Professor Friedman defines the next five years as the “Institutional Stage”, when the following could be achieved (Friedman 2015):
- The maturation of inter-organisational networks that are now forming (e.g. PCORI CDRNs, CancerLinQ, Cancer Research Network): These will address problems requiring multiple institutions e.g. public health and rare diseases
- Intra-organisational networks or learning organisations will evolve: Large healthcare delivery systems will become learning entities
- Developments of the platform components for the LHS
o Standards
o Technical components
o Implementations
o Best practice
o A knowledge base of how to learn
Other participants believe that the next five years will see, increasing efforts to learn from routine data, improvement of the related user interfaces, further development of outcome measures and outcomes based reimbursement, progress on interoperability, more acceptance of data sharing from patients (Akerman 2015, Etheredge 2015, Rubin 2015).
Prof Friedman envisages the five to ten year period as being the “Consumer Stage”, potentially characterised by (Friedman 2015):
- The emergence of the consumer as the most important stakeholder
- The concept of the LHS creeping into consumer awareness. “The consumer will become the glue that will hold the system together”, potentially bringing data from a wider range of sources and seeking analysis from multiple providers
- Interoperability will be tackled
- LHSs will scale, becoming true networks of networks
- Governance will emerge
- Privacy concerns will continue, but best practice may emerge for managing this issue
- There may be a move towards a safety culture where errors are seen as a learning opportunity
Platform components are important because they will enable the rapid development and distribution of Learning Healthcare Systems. Currently, projects to implement large scale Learning Healthcare Systems, such as TRANSFoRm (TRANSFoRm 2015) and FDA Mini-Sentinel must first develop an expensive distributed network infrastructure and rules for its operation, before the Learning Healthcare System functions can be deployed. This is analogous to recreating a new version of the Internet every time a new website is launched. Currently this is only possible for large EHR vendors and well-funded academic collaborations. These organisations cannot, by themselves, develop all of the potential applications that could benefit healthcare.
Once the infrastructure is in place, for example, in the form of truly interoperable EHRs, individual organisations (e.g. governments, providers, research organisations, patient groups, technology companies) will be able to deploy and share innovative Learning Healthcare Systems at low cost. They will also be able to reuse components of systems developed by other organisations, rather than having to reinvent the wheel each time.
This would dramatically reduce the cost and time taken to build useful systems in a similar way to which the development of HTTP and HTML allowed anyone to develop and share their own website on the Internet. As more organisations begin developing systems, novel use cases may emerge that have not yet been considered and an ever-broader coverage of health care and other sectors may be achieved (Foley and Fairmichael 2015).
For example, an organisation might develop a new predictive model that enables providers to improve care or reduce costs. Currently, this model would need to be implemented separately in every provider, making uptake unacceptably slow. If the appropriate platform is in place, then the distributing organisation could trade or give away their solution, possibly through some brokering service, so that other organisations could deploy it on their system almost as consumer would download an app. The model could then run against every patient on that provider’s system, flagging up results for clinical consideration.
It is not yet clear who will control these platforms. Publicly funded projects, such as TRANSFoRm (TRANSFoRm 2015) and FDA Mini-Sentinel (Brown 2015) are currently developing open source systems. There are private companies, such as IBM (IBM 2015), who are encouraging developer ecosystems around their core technologies. There are also public-private, not for profit partnerships, such as QResearch (QResearch 2015). Large EHR vendors are also developing Learning Healthcare System components that can be rapidly deployed, but only to providers who use their particular system (Swindells 2015). It is not yet clear whether any of these platforms will become dominant or even widely accepted.
A variety of approaches will probably co-exist and the market structure in each country is likely to be influenced by the political and economic context. For example, innovation in the US has been more market driven than in the UK (Bates 2015). There may eventually be a role for market regulators, such as the Competition and Markets Authority in the UK, in ensuring that platform operators do not abuse their position.
Professor Friedman envisages the five to ten year period as being the “Consumer Stage”, potentially characterised by (Friedman 2015):
- The emergence of the consumer as the most important stakeholder
- The concept of the LHS creeping into consumer awareness. “The consumer will become the glue that will hold the system together”, potentially bringing data from a wider range of sources and seeking analysis from multiple providers
- Interoperability will be tackled
- LHSs will scale, becoming true networks of networks
- Governance will emerge
- Privacy concerns will continue, but best practice may emerge for managing this issue
- There may be a move towards a safety culture where errors are seen as a learning opportunity
There is also likely to be significant progress on outcomes measurement, for example, ICHOM have set the goal of covering half of all medical care with transparent medical data within 10 years (Stowell 2015). With this level of coverage, benchmarking would become wide spread and the public would likely demand transparency.
Given the nature of this field, no participants were prepared to speculate on precise developments beyond ten years, except to say that they are likely to be transformative.
Evidence