A rational model for assessing and evaluating complex interventions in health care

The processes by which 'innovations' can be diffused across healthcare systems have been intensively described, both in theory [2], and as Greenhalgh et al's review has shown, in practice [3]. Diffusion of innovation models do not, however, provide a framework for assessing the conditions in which such interventions become practically workable in healthcare settings. This article responds to the evident need to understand the conditions in which new technologies, techniques, working practices, and organizational interventions – complex interventions – can become embedded as routine elements of clinical and organizational work in health care. This reflects a key problem for decision-makers in health care. They need to do two things in the face of a potential complex intervention. In parallel, they must consider: (a) its workability, clinical and cost effectiveness (the focus of Health Technology Assessment research), and (b) evaluate its capacity for successful integration into existing or new configurations of health services (the focus of research on Service Delivery and Organization), and professional practice (the focus of Quality Improvement research). Given these requirements, the normalization process model provides a rational framework that enables practical understanding of the conditions in which complex interventions can become embedded within clinical work.

Normalization is defined as the embedding of a technique, technology or organizational change as a routine and taken-for-granted element of clinical practice [4, 5]. This is different from market and management decisions about diffusion or adoption – the focus of much research around medical innovation – because it focuses on the conditions of use and the behavior of everyday users rather than the special champions and early adopters so important to diffusion theories. It reflects the importance of stability, order, and practicability in professional and organizational behavior in healthcare. These are important even when radical innovations are in play – a point at which the search for stability often becomes more urgent. New techniques and technologies are often locally invented or re-invented, and externally defined innovations are often subject to local modification and reconfiguration [6]. Because of this, normalization is a more flexible concept than diffusion. There are two reasons for this. First, the concept of normalization acknowledges that technological and organizational change in health care settings is often imposed, and is thus not always the product of the kinds of processes set out in diffusion of innovations models. Second, it acknowledges that irrespective of source of change, clinicians (and patients) often creatively work to flexibly configure practices in ways that meet specific local situations and requirements [7]. When they cannot do so, for example, when working within the protocol of a clinical trial, problems of workability and integration ensue. This means that how to understand normalization processes is an important question across a variety of domains of healthcare R&D: from health informatics and health technology assessment, to evidence-based clinical practice and other fields of implementation research. The complexities of implementation are well illustrated in recent research on technical innovation in medicine [8] and information technologies in health care [9, 10]. For the purposes of this article, modified or new technologies, techniques or organizational forms are all treated as members of the same category – complex interventions.

In their review of innovations studies and their application to health care, Greenhalgh et al [3] have called for a 'whole systems' approach to understanding the potential for implementation of new modes of practice. This follows from a very large body of studies that have sought to develop theories and models of the capacity of organizations to innovate and deploy new systems of practice. These range from the analysis of technological 'regimes' [11], through diffusion models [2], studies of the dynamics of organizational performance [12], analyses of the role of networks [13], and studies of technological and organizational integration [14].

Accounting for change at a whole systems level is highly complex. Theories that engage with complexity at a systems level – such as Actor-Network Theory [15], Complex Adaptive Systems Theory [16], or Structuration Theory [17] – give prominence to the behavior of organizations, networks or collectives as basic units of analysis. This fits well with analysis at a macro-level, but does little to assist in understanding the everyday micro-level components of clinical practice. Indeed, one of the results of the application of complexity theories may be paralyzing uncertainty about the unpredictable consequences of intervening in a complex system. Macro-level theories struggle with the business of accounting for action at a micro-level [18], a task that is usually delegated to theories of interaction processes or to psychological models of intention and volition. Although these produce sophisticated explanations and important results, they are sometimes difficult to apply to practical problem solving. Nor do these approaches always lend themselves to the kinds of conceptual models that researchers, especially trialists, seek in framing the evaluation of complex interventions.

It is not surprising, therefore, that sponsors of trials of complex interventions are concerned with problems of accounting for their very complexity [19]. Nor is it surprising that trialists in health technology assessment or quality improvement often frame their work through atheoretical constructs of observed barriers and facilitators to change [20]; or through psychological constructs that can be shown to predict certain aspects of individual behavior [21], even though the latter can be difficult to apply to collective action.

The model is derived from formative (secondary) and summative (tertiary) analyses within four groups of qualitative (interview and ethnographic) studies (n = 23) undertaken between 1995 and 2005. All studies were of physicians, nurses, other professionals, and patients, in primary care and associated settings at the interface between primary and secondary care in the UK National Health Service. The four groups of studies and their associated formative analyses were:

Re-analysis demands caution. For example, collaborative data clinics are theory-led, but also risk being confirmatory rather than critical in their approach to the data because they rely on social processes of inter-subjective consensus building. Formal recoding of restricted bodies of data risks the application of inappropriate coding frames, or missed deviant cases in the data. Both techniques risk the over-interpretation of data. This arises either: (i) because of inappropriate confirmation in collaborative data clinics; or (ii) because of selective objectification and missed opportunities for disconfirmation in the formal recoding of randomly or purposively sampled data. Formative analyses therefore involved actively searching for deviant or disconfirming cases. Although such analysis has limitations and risks, the sheer volume of qualitative data collected in the various studies comprising the four groups of studies research meant that complete re-analysis of these data sets would have been impractical even if it were desirable.

The second stage of model building followed from the methods developed in the production of formative analyses. These had focused on developing analytic propositions that led from studies of clinical practice, technological development, and organizational change. The objective of the second stage was to build on these and to develop a general set of summative constructs and propositions. This process consisted of four interpretive theory-building activities.

The objective of stage 2 was to develop summative analytic propositions that drew together the results of formative analyses, and stepped beyond the confines of individual studies. Although the literature of qualitative research is replete with claims about its capacity to make theoretically generalizable propositions [57], in fact there are few examples of such work where the process of generating such propositions is clearly described. Glaser and Strauss's [58] account of the development of substantive 'grounded' theory remains the most convincing, and it was this approach that informed the interpretive work of stage 2, leading to the production of four theoretically generalizable constructs and their associated propositions.

Once again, there are reasons to be cautious about the claims made about such work. The most obvious is that the shift from formative analysis to summative model building involves a move from the collaborative use of methods for the secondary analysis of qualitative data to the individual interpretation of its outcomes. There are no obvious mechanisms through which the products of such interpretation can be verified while they are under construction. This is because such work is explicitly aimed at the production of theory rather than the application of methods for data analysis. What follows is therefore a limited model, from which propositions spring, rather than a general theory intended to encompass all aspects of the normalization of complex interventions.

Endogenous processes comprise elements of professional-patient relations and their associated material practices in the clinical encounter, and they can be defined in terms of the interpersonal context for normalization. Formative analyses revealed key properties of these interactions in the management of patients with chronic illness. They also revealed how new technologies of practice decouple the traditional individual authority of the professional from management and decision-making about clinical responses to illnesses with long temporal horizons. Since the 1930s, sociological studies have shown that the clinical encounter proves remarkably stable across different kinds of health services and in different cultural and historical contexts [59‐61]. Such studies have pointed to the important role of asymmetries of power and knowledge in the clinical encounter and their effects. While these are important elements of social relations in health care, from the perspective of this article it is also important to remember that clinical encounters are also governed by social norms about co-operative conduct and are goal-oriented. This raises questions about what interactional work is necessary to bring a complex intervention into play within the clinical encounter, and leads to the first construct and proposition of the model.

Looking at endogenous processes enables an analysis of the work that is done in chronic disease management and how it is achieved. Nevertheless, this is of little use without an analysis of how work is arranged and attributed to particular categories of professional and of the operational contexts in which they are located. Exogenous processes comprise the ways that work is organized, its division of labor, and the institutional structures and organizational processes in which it is located.

Because much of the literature about organizational and technological change is derived from studies of large corporations and small firms and focuses on competitiveness in the market place it emphasizes the dynamic fluidity of business processes and the focus of managers on specific types of goals as they respond strategically to market conditions. There is now a very large body of literature that points the way to methods of securing the diffusion, adoption, and implementation of innovations in healthcare has been reviewed by Greenhalgh et al [3]. An equally large body of literature that points to methods of change management in healthcare organizations has been reviewed by Iles and Sutherland [66]. This literature helps us to understand how diffusion occurs and is managed, but is less successful in accounting for normalization. Its focus on dynamic change does not fit well with healthcare organizations' search for stability, order and predictability in organizing and delivering services. The epidemiological landscape in which they are set is characterized by the increasing prevalence of chronic diseases that require care rather than cure, and where 'management' is a central operating concept. Furthermore, policy is mediated through increasingly active management and regulation by government agencies while professional self-regulation and self-organization declines. Dynamic change within health services is constrained by competition for limited resources. Macro-level active management of healthcare processes increasingly shapes professional action at the micro-level by defining quality standards for practice [67]; by information processes that routinely measure performance, and are mediated through local management interventions that increasingly regulate professional action [68]; and by guidelines for practice that seek to standardize care against 'best evidence' [43]. The capacity of health service organizations, both large and small, to effectively implement complex interventions is dependent on their capacity to integrate these in divisions of labor, and specific organizational settings. These settings comprise elements of the contexts – the 'exogenous processes' – in which professional work is located and in which complex interventions are ultimately enacted.

Summative analysis suggested ways that these exogenous processes reflect the changing structures of clinical work, in particular the extension of an increasingly detailed division of labor and its associated distribution of expert and routinized tasks in the management of different kinds of chronic illness. It also described the modes of policy and activity necessary to enact complex interventions. Two further paired constructs and propositions that stem from them can be expressed, and these are set out below.