INTRODUCTION
Learning health systems (subsequently referred to as LHS) are defined by the Institute of Medicine (IoM) as systems in which alignment of scientific and cultural tools lead to knowledge generation to improve healthcare as a result of daily practice.1 Since LHS was conceptualised in 2007, they have been the focus of increasing research attention.2–6 The opportunity and promise of LHS have resulted in texts presenting collections of LHS-specific research1,7 creation of a new journal, Learning Health Systems,8 and new courses of academic study.9,10 We believe it could be the most significant development in healthcare since the advent of evidence based medicine (EBM) and electronic health records (EHRs) that support EBM.14–16 EHR has existed for more than 40 years11–13 and organisations that implemented EHR discovered reductions in costs, clinical testing and patterns of repeated and sometimes unneeded prescriptions. Enhanced co-ordination and communication between clinicians were seen to improve the quality of patient care.12,17–20
Despite the benefits, early EHR systems were considered expensive, focused on information gathering rather than improving healthcare.20 Development lacked clinical input, existed as multiple stand-alone systems, experienced slow adoption, suffered from trust and data quality issues, claims that systems increase or exacerbate risk for errors, and concerns over patient privacy and security.18,20 All of these issues are still seen as unresolved barriers to adoption of EHR.18,21–27 Despite this, EHRs are the foundation for LHS. Efforts towards LHS, coupled with proposed changes to legislation, policy and the ethics of how clinicians engage with clinical datasets suggest an entirely new dimension to EHR. One in which they are used collectively as ‘big data’ and focused using individual patient’s attributes to identify causes and optimal treatments strategies for disease.
Descriptions developed in IoM reports are the basis of most author definitions and descriptions of LHS.28–33 Medical information systems that can be predictive, preventative, personalised and participatory represent the core principles of 4P medicine.34 According to the IoM, these systems have the potential to identify groups at greatest risk of complications for purposes of targeting interventions.7 In parallel, maturing technologies such as large datasets, machine learning, and enhanced processing power further enable the concept of LHS.2,34,35
The IoM promoted EBM as the primary driver for LHS,7 yet their definition fails to describe LHS attributes which contribute to quality, safety, efficiency and effectiveness of patient care.36 The four fundamental attributes37 listed in Figure 1 provide tangible metrics to compare and contrast LHS efforts. These attributes were not included in the IoM definition, but are widely found in LHS. They clarify the use of large EHR datasets as the source of knowledge for LHS achieving the goal of improving quality in individual patient care.
There are numerous examples of proposed benefits of LHS. For clinicians, these include assessing which laboratory or imaging tests may be more diagnostic given a patient’s presenting symptomology38; reducing risk from prescribing errors31 and increasing awareness of pharmacogenetics.39 It is claimed that patients would benefit from advanced knowledge developed from the experiences and diagnoses of past patients, which saves time and reduces costs.38,40,41 A learning healthcare organisation culture supports EBM, while successful integration of research into practice is what enables it.42 The ability to use technology to record, compare, contrast and present information in almost real-time enhances the input, analysis and decision phases of the learning lifecycle. Alternatively, it is said that the financial burden to implement and support health technology43–45 along with a persistent need for data and systems standardisation,45–48 interoperability46,49,50 and integration49,51 have all acted as barriers to broad LHS adoption.
In our group’s recent letter to the editor of this journal,52 we demonstrated the lack of awareness and barriers for researchers to appropriately identify their efforts as LHS solutions. We believe that this results from a number of significant problems in the domain. The lack of adequate classification and standardisation results in groups working on comparable systems not identifying their efforts as LHS, and may be the cause of unnecessary duplication of efforts and the observable lack of collaboration. The absence of a unifying framework means the domain is yet to generate a necessary critical mass, limiting efforts to resolve barriers and challenges to the adoption of LHS and constraining funding availability. We were only able to identify one primary article that attempted to consolidate and analyse the current state of knowledge in LHS.36 We extend that effort drawing on a larger collection of works to establish a unifying framework for LHS.