Loading field map
How should health services researchers define, measure, and improve the quality and safety of care? This question has driven a half-century of conceptual evolution, producing three major frameworks that each reoriented the field's priorities, methods, and assumptions. The Donabedian Quality-of-Care Model (1966) made quality measurable through a simple triad. Patient Safety and Quality Improvement Science (1990) shifted attention from measuring attributes to preventing harm through systems thinking. The Learning Health System (2000) then proposed collapsing the gap between research and practice by embedding improvement into routine care. These frameworks did not simply replace one another; they coexist today, each offering a distinct lens on what quality means and how to pursue it.
Avedis Donabedian's structure-process-outcome framework, introduced in 1966, was the first systematic attempt to define health care quality in terms that could be studied empirically. Structure refers to the settings, qualifications, and organizational resources of care (e.g., hospital accreditation, staffing ratios). Process covers the actual delivery of care (e.g., whether a guideline-recommended test was ordered). Outcome captures the end results for patients (e.g., mortality, symptom relief, satisfaction). Donabedian argued that good structure increases the likelihood of good process, which in turn increases the likelihood of good outcomes.
The framework's distinctive contribution was to transform quality from a vague professional ideal into a set of observable, measurable variables. It authorized observational studies that linked structural features to processes and outcomes, and it became the dominant teaching tool for quality assessment across health services research. However, the model had a critical limitation: it described what quality looked like but offered no causal mechanism for why poor outcomes occurred. A hospital could score well on every structural measure yet still produce preventable harm. The model also focused on individual patient encounters, making it less suited to analyzing system-level failures.
The Patient Safety and Quality Improvement Science framework emerged around 1990 as a direct response to the gaps Donabedian left open. Where Donabedian's model measured quality attributes, patient safety science asked a different question: why do things go wrong? Drawing on human factors engineering, cognitive psychology, and high-reliability organization theory, this framework treated adverse events not as individual failures but as consequences of poorly designed systems.
Its core commitments included root cause analysis, failure mode and effects analysis, and iterative improvement cycles such as Plan-Do-Study-Act (PDSA). These methods shifted the unit of analysis from the patient encounter to the clinical microsystem—the team, the workflow, the equipment, the culture. The framework also introduced a new object of study: safety culture, or the shared attitudes and practices that determine whether staff feel able to report errors without blame.
Compared to Donabedian's model, patient safety science narrowed the focus from general quality to the specific problem of harm, but it also deepened the causal analysis. It preserved Donabedian's structure-process-outcome logic as a background taxonomy while adding a layer of systems thinking that the earlier model lacked. The framework's methods were more interventionist: rather than observing and measuring, researchers actively redesigned processes and tested changes in real clinical settings. This made it a natural ally of quality improvement (QI) science, which shared its pragmatic, iterative ethos.
The Learning Health System (LHS) framework, articulated around 2000, represents an attempt to absorb and transcend both of its predecessors. Its central insight is that the gap between generating evidence and applying it in practice—a problem that neither Donabedian's measurement nor patient safety's local improvement cycles fully solved—can be closed by making every clinical encounter a source of data for continuous learning.
An LHS is defined by its ability to produce knowledge as a natural byproduct of care delivery. Its methodological toolkit includes pragmatic trials embedded in electronic health records, rapid-cycle analytics, and real-world evidence generation. Where Donabedian's framework relied on periodic audits and patient safety science on project-based improvement, the LHS envisions a permanent infrastructure in which data flows automatically from clinical documentation to analysis and back to decision support at the point of care.
The LHS explicitly repurposes concepts from the earlier frameworks. It retains Donabedian's outcomes as the ultimate metric but embeds them in a dynamic feedback loop rather than a static assessment. It absorbs patient safety's emphasis on systems and culture but scales it from the microsystem to the entire organization or population. The framework also transforms the role of the researcher: instead of an external evaluator (Donabedian) or a facilitator of local change (patient safety), the researcher becomes a designer of learning cycles that operate continuously.
These three frameworks remain active today, and their coexistence is not always harmonious. Each authorizes a different methodological style. Donabedian's model still underpins most large-scale quality measurement and public reporting, such as hospital compare databases. Patient safety science dominates the study of adverse events, safety culture surveys, and high-reliability organizing. The LHS is the youngest and most ambitious, driving initiatives in embedded pragmatic trials, learning networks, and data-driven improvement collaboratives.
The leading frameworks agree on several points: that quality is multidimensional, that systems matter more than individual blame, and that improvement requires measurement. But they disagree sharply on the proper unit of analysis. Donabedian's model is fundamentally encounter-focused; it assesses quality one patient at a time. Patient safety science operates at the microsystem level—the team or unit. The LHS pushes toward the population or system level, where data aggregation enables pattern detection but may obscure the local context that patient safety researchers consider essential.
A second tension concerns methodological rigor. Donabedian's observational studies and patient safety's QI cycles have been criticized for weak causal inference. The LHS responds by advocating for embedded randomized designs, but these require data infrastructure and analytic capacity that many settings lack. A third debate centers on equity: all three frameworks have historically focused on average effects, and critics argue that quality improvement that does not explicitly address disparities may widen them. The LHS, with its capacity to analyze subgroup effects in real time, may offer tools for this problem, but the field has only begun to grapple with it.
The history of quality and safety research is not a story of one framework replacing another but of accumulating layers of insight. Donabedian gave the field a common language for measurement. Patient safety science added a causal theory of harm and a culture of improvement. The Learning Health System now proposes to make learning continuous and embedded. Each framework remains useful for specific purposes, and the most sophisticated research today draws on all three, matching the framework to the question at hand. The ongoing challenge is to integrate their strengths—measurement, systems thinking, and data-driven learning—while resolving the tensions between encounter-level and population-level perspectives, between local context and generalizable evidence, and between improving average quality and closing equity gaps.