There are many potential sociotechnical configurations that would meet the definition of a Learning Healthcare System. The six use cases outlined below represent the systems that were most often cited by participants. They each rely on routinely collected data and medical knowledge that is assembled, analysed and interpreted, before being fed back into the healthcare system with the aim of promoting behaviour change.
Each use case has already been demonstrated at least at proof of concept stage. As these use cases are implemented throughout the healthcare system, it is expected that they will form an interrelated mesh of learning cycles that have the potential to promote rapid improvement in quality and cost effectiveness. It is however, important not to underestimate the challenges involved in implementing such systems at scale.
Intelligent automation will reduce the burden on clinicians and improve care by “making the right thing, the easy thing to do”. This may include automating routine aspects of care, prepopulating orders and clinic notes, and summarising case notes.
Learning Healthcare Systems have the potential to revolutionise Comparative Effectiveness Research (CER). There is scope for much greater use of observational studies and pragmatic RCTs that could help to fill gaps in the evidence base more quickly and at lower cost. Ultimately, insights could be generated into the likely effectiveness of different treatments in a particular patient. Traditional RCTs will continue and Learning Healthcare Systems could be used to identify eligible patients and ease data collection.
Improved data on outcomes will allow benchmarking between different providers. This will lead to improvement through a process known as positive deviance. Positive deviants (really good providers) will be identified, they will be studied to generate hypotheses about their performance that can be tested and evidence can be disseminated to other organisations.
Real-time surveillance systems are being developed to track epidemiological phenomena and adverse events related to new treatments. These use routinely recorded data and allow much more timely learning to take place.
Predictive models can be developed to identify instances where there is a high risk that low quality or unnecessarily expensive care will occur. Further impactibility modelling can help to identify which of those instances are most likely to respond to mitigation.
Clinical Decision Support Systems can be used to support clinicians in dealing with unfamiliar or high-risk situations. These systems can be powered by machine-readable guidelines and can be integrated into EHRs.