Four distinct models of learning health systems: Strength through diversity

Abstract

The concept of a learning health system (LHS) was established nearly 20 years ago as a unifying commitment to speed the generation and use of evidence primarily by leveraging rapid advances in data and technologies, resulting in optimized care for each patient. In the ensuing decades, vanguard adopters of the LHS who have sought to move the LHS from conceptual to operational have done so in ways that fit with and reflect their organizational structure, mission, and culture—as well as their personal values and experiences. They have also extended the focus from health care to include individual and population health more broadly. This commentary describes four distinctive models that have evolved as learning health system activities have matured. Viewing this diversity as a strength, the features, commonalities, and unique differences of these models are described.

The concept of a learning health system was established nearly 20 years ago by the Institute of Medicine, now the National Academy of Medicine (NAM), as part of the charter for the Roundtable on Science and Value and Science‐Driven Health Care. ¹ It was envisioned as a unifying commitment to speed the generation and use of evidence by leveraging rapid advances in data and technologies, resulting in optimized care for each patient. The NAM's definition is intentionally broad, encompassing the foundational importance of science, informatics, incentives, and culture as necessary elements for learning; the central role of patients and families as active participants in their care; and a cycle of continuous improvement as the unwavering imperative. This broad conceptualization has provided useful grounding for those who aspire to build learning health systems, and it is not prescriptive as to how a system actualizes this definition in real‐world settings. Thus, in the ensuing decades, vanguard adopters of the LHS who have sought to move the LHS from conceptual to concrete have done so in ways that fit with and reflect their organizational structure, mission, and culture—as well as their personal values and experiences.

Given the well‐established challenges of ensuring better health outcomes and lower health care costs, interest in alternatives to the status quo remains high. ² The LHS—with its emphasis on leveraging evidence and data in a rigorous and rapid fashion—has been held out as a promising antidote to the current state. Proponents have shown that integrating evidence at the point of care—a key pillar of the LHS—is possible. ³ , ⁴ Many systems that have adopted the LHS as a cornerstone to their improvement efforts have published organizing frameworks ⁵ , ⁶ , ⁷ , ⁸ and common principles, ⁹ , ¹⁰ and the growth of the field has also yielded beneficial scoping reviews about various facets of the LHS. ¹¹ , ¹² , ¹³ , ¹⁴

The proliferative literature—and, indeed, the existence of this journal as a venue for publications on the growth and maturation of the LHS—is encouraging. To support continued growth of the LHS and respond to the need for more tactical descriptions to complement conceptual frameworks and experience reports, this commentary describes four distinctive models that have emerged as learning health system activities have matured. Features, commonalities, and unique differences are described to help others who wish to examine their own capabilities, and perhaps even align their own approach to one of the models described herein. Our aim is to codify the increasingly common approaches to building an LHS so that those seeking to better understand and progressively implement an LHS can examine their own capabilities and goals, and then identify and apply the model that is most aligned with their local context.

1. EMERGENCE OF FOUR DISTINCTIVE LHS MODELS

Our collective experience with Learning Health Systems, as a concept and a set of methods, dates back to 2009. Looking back to that date, one might have expected a singular vision of a Learning Health System to have evolved; but from our perspective, something much more divergent happened. To appreciate this, one need only compare the notion of a learning network ¹⁵ , ¹⁶ that originated from the work of ImproveCareNow ³ to the notion of a pragmatic trial pipeline ¹⁷ , ¹⁸ —both of which are portrayed in the literature as examples of Learning Health Systems—to see the extent of the divergence. Reports emerging from the federal government ¹⁹ and from the National Academy of Medicine ²⁰ portrayed yet another model of LHS stimulated by publicly funded initiatives; and another emergent model emphasizes the role of the so‐called embedded LHS researchers. ²¹ , ²²

The four models that have emerged are described below at a high level of abstraction, with Table 1 (introduced below) providing further detail. All models seek to improve health and health care through systematic self‐study and application of what is learned.

1. The participative model is inspired by the concepts of learning organizations ²³ and learning networks. ³ , ¹⁵ , ¹⁶ Participative LHSs carry out improvement cycles targeting specific health problems, supported by a social‐technical infrastructure and emphasizing co‐creation across all stakeholders. In a fully mature participative LHS, these activities pervade the entire organization or network and create a distinctive culture within it. Notable examples include networks such as ImproveCareNow, ³ , ¹⁶ PEDSNet, ²⁴ and institutionally based efforts such as those at the Geisinger Health System, ²⁵ the Universities of Michigan ²⁶ and Minnesota, ²⁷ and Wake Forest University. ⁵

2. The embedded researcher model emphasizes the role of individuals or small teams that are attached to programs for the purpose of generating real‐world evidence that will generate insights that promote the achievement of the program's health improvement goals. ⁵ , ²⁰ , ²⁸ An embedded researcher‐focused LHS grows as the number of embedded researchers within a host organization grows. Notable examples include the activities funded under the e‐STaR program, ²⁹ Kaiser Permanente's Care Improvement Research Team, ³⁰ the Texas Health Scholars Clinical Research Program, ³¹ and the embedded research partnership at Stanford Medicine. ³²

3. The pragmatic trial model is based on a primarily technical infrastructure that enables rapid and efficient collection and analysis of real‐world data comparing the success of intervention options. ³³ Improvement is achieved by widely implementing the option that demonstrates greater success, and a pragmatic trial‐focused LHS grows as the number and diversity of pragmatic trials increase. Notable examples include mature efforts at Vanderbilt University ¹⁷ and NYU Langone Health System. ¹⁸

4. The programmatic model seeks to promote LHS through the initiation of publicly funded programs, each of which supports data‐driven health improvement and/or the development of widely shared infrastructure. Governmental payers, such as the Center for Medicare and Medicaid Services, have taken interest in promoting LHS methods to improve care and care outcomes. ³⁴ Large‐scale employment of LHS discovery and implementation activities emerges through the aggregation of these funded programs on a national or provincial scale. ³⁵ , ³⁶ , ³⁷ Other notable examples include PCORnet®, ³⁸ programs of the VA Health System, ³⁹ as well as NIH‐funded LHS programs such as the Genomics‐Enabled‐LHS ⁴⁰ and EPInet. ⁴¹

TABLE 1.

Characteristics of four learning health system (LHS) models.

LHS model (examples)	Home discipline(s)	Change strategy	Morphology	Scale and visibility	Valuing of time urgency	What determines the intervention?	Patients' role	Required infrastructure	Organizational penetration potential	Determination of what gets addressed
Participative	Behavioral and system science	Iterative cycles by collaborative communities	Co‐occurring cycles with shared infrastructure	Meso	Low	What the community collectively decides	Co‐creators	Socio‐technical services supporting all learning cycle facets	High: End‐goal is to pervade the organization and its culture	Learning communities emerge organically or may be incentivized
Embedded Researcher	Health services research, health economics	Real‐world evidence generated by LHS scientist	Embedded in projects	Micro‐meso	Moderate	What the data show + what the project leadership decide	Recipients of care	Community of LHS scientists + shared tools	Limited: May not extend beyond the projects directly supported	Strategic needs of the organization and/or external funding
Pragmatic trial	Informatics, clinical research	Statistical comparison of alternatives	Self‐contained studies	Micro‐meso	High	Adoption of statistically superior alternative	Sources of data	Technology‐intensive pipeline	Limited: May not extend beyond the trials carried out	Pipeline governance oversees trial selection process
Programmatic	Public policy and organizational science	Embedded in the aims of funded programs	Funded programs	Macro	Moderate	What agency leadership and/or advisory boards decide	Beneficiaries of programs	Some programs build infrastructure; others require it as prerequisite	Moderate: May not spread beyond what is directly funded	Agency leadership and advisory board perceptions of population needs

2. DIFFERENTIATING THE FOUR MODELS

To promote a clearer comparison and contrast of the four models, Table 1 offers characteristics that differentiate them. We define each of these characteristics, comprising the column headers in Table 1, as follows:

Home discipline(s): The methods employed in each model rest on the theories, literature, and methods associated with specific and differing scientific disciplines. The disciplines called out in each cell can be thought of as identifying the literature one would consult to appreciate the origin of the model and learn in detail about the methods employed in it.

Change strategy: All LHS models seek to improve individual and population health with data‐driven interventions. Fundamental and unique to each model is a distinctive strategy for effecting changes targeted at health improvement.

Morphology: LHSs can take on different shapes and forms. This characteristic relates to what one would look for if one were trying to verify the existence of an LHS that is based on a particular model.

Scale and visibility: Tied closely to a model's morphology is a scale at which it typically operates and its consequent visibility. At micro‐scale, LHS activity exists within projects and is not readily visible. At meso‐scale, LHS activities, either within a single organization or through a network of organizations, have a scope and identity that makes them visible. At macro‐scale, LHS activities are highly visible through the descriptions of funded projects and initiatives.

Time urgency: Built into the culture of each model is a sense of the importance of producing rapid changes. Some models give high priority to rapidity, whereas others are willing to sacrifice speed for the potential to effect more systemic changes.

What determines the intervention: As noted above, all models employ data‐driven interventions to improve health. The models differ in the strategies and criteria used to decide the shape and focus of the intervention(s).

Patients' role: The models vary in how patients are perceived, and the role patients play in the improvement process.

Required Infrastructure: All systems, in varying ways and to varying extents, require technical and social infrastructure in order to function efficiently. Infrastructures provide centralized services that all facets of the system may use. The four LHS models vary in the infrastructural services that they require.

Organizational penetration potential: The LHS models vary, by virtue of their values and structure, in the extent to which they can pervade the environment in which they are based. The ability to influence and penetrate the organization is also tied to scale and visibility.

Determination of what gets addressed: Since virtually any health problem can be addressed by an LHS, the problems that make it “to the table” determine what is going to be achieved. The four models vary with respect to how these determinations are made.

3. DISCUSSION

It is beyond the scope and intent of this Commentary to take a deep excursion into each cell of Table 1. We recognize that the content of each cell is open to debate, and we would welcome these discussions. Nonetheless, we believe that the variability in the cells within each column of the table confirms our thesis that the four models of LHS propounded here differ significantly along multiple axes. We hope that our analysis will promote useful discussion among and across proponents of each model. Ideally, in our view, when the future developers and supporters of LHS describe their efforts, they will identify which model or models they are following. This will support ongoing efforts to build the field and the collective ability to solve complex challenges in health and health care.

While Table 1 illuminates differences across the models, it is important to be mindful of their similarities. Principally, all models share the end goal of improved individual and population health, and all undertake improvement through rigorous methods of discovery leading directly to the implementation of what is discovered. All seek to shorten the “17‐year” latency between discovery and implementation ⁴² and all seek to abide by the LHS Core Values ⁹ and Shared Commitments ¹⁰ that have been advanced in the literature. While the models differ in how teams are constituted and how decisions are made about priority interventions, interdisciplinary teamwork is fundamental to all four.

At this point in the evolution of Learning Health Systems, we believe the existence of distinctive models is a strength. Proponents and practitioners of different models can—and should, in our view—learn from each other and build bridges between them rather than silos or walls around them. While programs ⁴³ , ⁴⁴ , ⁴⁵ that train future “LHS scientists” might emphasize a particular model and/or area of specialization, such programs should prepare their graduates to apply their training across the range of models, such that they may adapt and contribute—with fluency and fluidity—to participatory research initiatives, embedded research institutes, academic health systems, or policy roles.

CONFLICT OF INTEREST STATEMENT

Sarah M. Greene reports consulting for the National Academy of Medicine.

Friedman CP, Greene SM. Four distinct models of learning health systems: Strength through diversity. Learn Health Sys. 2025;9(2):e70009. doi: 10.1002/lrh2.70009