Development of a theory of implementation and integrationNormalization Process Theory

If you've ever tried to change how a clinic actually works—the way people talk to patients or the way a new technology gets used—you know the hard part isn't the idea. It's getting the practice to take root and show up in Tuesday afternoon routines and Friday morning handovers. It should still be there six months later when the champions are on leave. Implementation science has a vocabulary for that terrain. It separates three problems we tend to muddle: bringing a practice into action, embedding it so it becomes ordinary work, and integrating it so it keeps reproducing across the social structures of an organization. Think of implementation as getting the engine to turn over, embedding as the engine idling smoothly on its own, and integration as that engine being built into the whole fleet. May, Mair, Finch, MacFarlane and colleagues looked at those problems and said: the usual theories don't get us far enough. Diffusion models tell us how ideas spread, and learning theories tell us how individuals pick up skills. These are useful, but they leave out the social work—the coordinated doing—that makes practices stick. They start from action. Their premise is simple and a little unforgiving: a practice only exists because people keep investing in it, together, over time and across places. That makes this a sociological project about agency, not a story where technologies have their own momentum. The agency lives with people, and the question becomes: what work do those people have to do to make a practice routine? Their first step toward answering that was pragmatic. Between the early and mid-2000s, they went back through qualitative studies of healthcare work—telemedicine rollouts, chronic illness clinics, and the gritty details of how evidence is produced and used in consultations—and looked for regularities. From those secondary analyses, they wrote down empirical generalizations. These were not explanations yet. They were more like, "Here's what tends to happen when telemedicine is normalized" or "Here's a common pattern when chronic care gets reorganized." These were framed as propositions but were deliberately observational. The idea was to build a solid platform of what's regular before asking why it's regular. From that platform, they built an applied model called the Normalization Process Model, or NPM. Now we move from "what's regular" to "what helps or hinders the work." Using grounded theory methods—coding, constant comparison, and testing categories against new material—they boiled that messy reality into four constructs that, together, describe the collective work of embedding a practice. Interactional workability is about how the practice fits into person-to-person interactions. Relational integration is about whether people trust the knowledge and roles the practice depends on. Skill-set workability asks whether the right people with the right competencies are positioned to do the right parts. Contextual integration is the organizational piece—resources, policies, and structures lining up so the practice can operate. If you've ever tried to get a new order set used on a busy ward, you can probably feel all four of those domains tugging at you at once. They didn't rush to formal hypothesis testing. Instead, they "road tested" the NPM. They circulated manuscripts, argued in seminars, and checked whether the constructs travel. Then they plugged the model into live projects and saw if it helps you see. They did that in three big settings: implementing e-health technologies, integrating telecare systems, and the operational nuts and bolts of a large randomized controlled trial. Alongside qualitative synthesis, they pulled in quantitative analyses to do the three jobs any good theory should do: define the phenomena, explain the mechanisms, and support knowledge claims that could be tested later. By the end of 2006, the model had proved itself as a set of conceptual tools. It helped explain why some collective action coalesced, while other efforts fizzled. At the same time, the road tests showed what the model didn't do. Elwyn and colleagues pushed on exactly that point by taking the NPM into shared decision-making. They mapped its constructs onto what had been learned from primary studies and systematic reviews about decision aids in consultations. Could the model account for the collective work of making shared decision-making real? Their answer was yes—the NPM provided stable explanations of the teamwork involved in operationalizing those tools. Again, the boundary of the model came into focus. It told you a lot about the work of embedding but didn't really explain how practices come to hold together in the first place, how people get enrolled and stay committed, or how the work gets appraised and adjusted on the fly. Those weren't one-off exceptions; they showed up as systematic omissions across cases. That's the hinge. The same team then widened the lens and built a formal middle-range theory to sit on top of the applied model: Normalization Process Theory, or NPT. The idea wasn't to toss the NPM. It was to keep its collective-action engine and add the missing pieces that shape how practices form, how people buy in, and how they're judged and adapted over time. They introduced three additional mechanisms. Coherence is the sense-making work—how people figure out what the practice is, what it's for, and how it differs from what they already do. Cognitive participation is about enrollment and commitment—who initiates, who legitimates, and who keeps it going. Reflexive monitoring is the appraisal—how actors evaluate effects as they go, both formally and informally, and decide what to tweak or abandon. Put those alongside collective action, and you have four linked mechanisms that express human agency across the life of a practice: making sense, enrolling, doing, and appraising. What changed by making it a middle-range theory? Two things. First, scope. NPT moves beyond context-specific generalizations and asserts that these are generative mechanisms you should expect to see across different settings where people try to implement material practices in formally defined contexts. Second, testability. The constructs are specified tightly enough that you can derive propositions and go test them. It's abstract—on purpose—so it can apply outside healthcare, but it's also bounded. This is not a grand, all-of-society theory. It lives where people try to get practices into routine use and where social mechanisms drive that work. A practical bridge between the model and the theory was to connect macro and micro. If collective action is what you see at the system or service level—the practice becoming real in an organization—what are the micro ingredients? Here, the NPM's four constructs didn't get thrown out; they were nested. Collective action at the macro level maps down to interactional workability in the clinic room, relational integration across professional groups, skill-set workability in teams, and contextual integration in budgets and workflows. That mapping matters because it shows you how to look both ways: from strategic plans down into the minute-by-minute work, and back up again. They used a method they called analytical theorizing to stress-test the theory in development. They drew maps of processes and spelled out what counts as a mechanism, what counts as an investment of effort, and what components you should be able to observe. They then road tested for stability and usefulness, not p-values. Two kinds of tests stand out. One asked: do these constructs hold steady and make sense across very different contexts without us having to bolt on ad hoc patches? They looked at examples ranging from e-health implementations to a reconfiguration of mental health services in Victoria, Australia. The second compared the applied model and the formal theory head-to-head by coding two data sets both ways. Does Normalization Process Theory provide you analytic advantage—clearer mechanisms and sharper predictions—over what the Normalization Process Model already gives you? The answer they report is that the NPT constructs could be operationalized in ways that made mechanisms and investments visible prospectively, while preserving the practical value of the model's collective-action focus. Now, a small but important pause. Throughout this, they were careful about how far they claimed to have gone. Road testing is not formal testing. It tells you your constructs are stable and useful, not that you've proved the theory. So they framed NPT as a theory ready for empirical testing, with hypotheses that could be taken into trial designs, implementation studies, and comparative analyses. They also stressed range. Is it generalizable beyond healthcare? Yes, in principle. But it remains middle-range, by design, so it stays tight on the causal social mechanisms of implementation, embedding, and integration. What does all this mean for people doing the work? It means you can pick your tool to fit your question. If your focus is squarely on the collective action that embeds a practice—who does what work, with what resources, to make it routine—the NPM is still right there for you, and May and colleagues are explicit that nothing in NPT displaces it. But if your question extends to how people make sense of the practice, how they get enrolled and stay engaged, and how they appraise and adapt it over time, then you reach for NPT. It adds coherence, cognitive participation, and reflexive monitoring to that collective-action core, making the sequence of sense-making, enrolling, doing, and appraising visible and explainable. There's also an intellectual stance that comes through the whole arc, from the first secondary analyses to the formal theory. It’s skeptical in the best sense—don't assume, specify. Don't chase grand explanations when middle-range mechanisms will suffice. Define the range of the theory and say what would count as a test. Then go and test it. The early moves—turning qualitative regularities into observational propositions across telemedicine, chronic care work, and evidence-in-use—were there to keep the theory anchored. The middle move—building the NPM with constructs like interactional workability and contextual integration—was a way to give implementers handles. The later move—expanding into NPT with coherence, cognitive participation, and reflexive monitoring—was to cover the missing mechanisms that make practices cohere, recruit their people, and evolve. If you want to picture the whole thing in one breath, try this. To normalize a practice, people have to agree on what it is and why it matters—that's coherence. The right actors have to buy in and stay in—that's cognitive participation. The work has to happen reliably with the right skills, roles, and resources—that's collective action. As it happens, people have to watch what it's doing and recalibrate—that's reflexive monitoring. Those four mechanisms, acting through time, produce embedding and, if they keep running, integration. Where does the story go next? The team says formal testing is warranted and underway, which is the right kind of cautious promise. For you, whether you're studying a new telecare service or trying to rewire the flow of evidence in oncology clinics, the takeaway is practical. Be explicit about which part of the normalization problem you're addressing. Use the constructs to hypothesize what should happen and what would count as success or failure in the mechanisms, not just in the outcomes. Keep the skeptical posture the theory was built with: remain open to being surprised by the work people actually do. One last thought. It's tempting, in implementation work, to anthropomorphize technologies or policies. We talk as if an app "wants" to spread or a guideline "meets resistance." Normalization Process Theory is a gentle corrective. Practices don't normalize themselves. People do, through sense-making, engagement, doing, and appraisal, under real constraints and with real ingenuity. If you can see those mechanisms at work, you're not just describing a rollout. You're explaining why it takes hold, and what it would take to make it stick.