A qualitative systematic review of studies using the normalization process theory to research implementation processes

The Normalization Process Model (NPM) was initially developed as an applied theoretical model to assist clinicians and researchers to understand and evaluate the factors that inhibit and promote the routine incorporation of complex healthcare interventions in practice [12‐14]. Much of the early work was related to implementation of e-health applications.

The further empirical applications of the NPM showed that while it could explain factors that promote and inhibit ‘collective action’ (i.e., the distribution of work required among stakeholders and the resources to support that), it did not address how participants understood and came to engage and support a new practice and how they reflected on and evaluated it. Through the development of further constructs (see NPT theoretical constructs from Finch, Mair, et al. [15]), accounting for how people understand and make sense of a practice (i.e., Coherence), engage and participate with it (i.e., Cognitive Participation), and reflect or appraise its effects (i.e., Reflexive Monitoring), the model became a theory, i.e., NPT. For the most part, the term NPT is used through out this paper, unless otherwise stated.

Since then, the theory’s development has focused on building a middle-range theory that explains how material practices (the things that people do when they implement complex healthcare interventions) become routinely embedded in their social contexts as the result of people working, individually and collectively, to enact them [13, 14].

Given its sociological origins, NPT is not focused on the relationship between individual attitudes and intentions and behavioral outcomes, which is the concern of psychological theories such as the Theory of Planned Behavior [16]. Like the sociological theory of Communities of Practice [17], NPT does pay attention to how knowledge is held, transferred, and created within and across professional groups, but it also seeks to understand the work that actors (clinicians, implementers, and patients alike) have to engage in to implement new knowledge in practice [18, 19]. Similar to theories of actor networks and diffusion of innovation [20, 21], NPT pays attention to the legitimacy of the intervention and the role of opinion leaders; it is concerned with understanding trust and interpersonal relationships within social networks as they impact on the introduction of innovation [22, 23]. However, NPT extends beyond the initial introduction of innovation to investigate the processes by which innovation may become embedded and routinized in practice, so much so that it becomes regarded as a normal and taken-for-granted way of working. Among NPT’s distinctive features is the attention to all stakeholders’ involvement in implementation processes, the work that they have to do individually and collectively, and the subtle and gradual processes from embedding and integrating to normalization [13]. The theory is centered on understanding social phenomena defined by four theoretical constructs, which characterize mechanisms that are energized by investments made by individuals and organizations (see Figure 1).

The development of NPT [14] focused on addressing two key criteria for theory to be ‘useful’: that it must offer adequate description and be fit for purpose. Thus, as outlined by Finch, Mair et al.[15], the theory has been developed to offer transparent and transferable explanations for the phenomena of interest (processes of embedding new practice and ways of working) revealed by empirical investigation.

There is a growing community of researchers who have made an a priori choice to use NPT in their research. Given the aforementioned calls to investigate new theories, we consider it timely to review how this particular new theory is being operationalized and, importantly, explore what benefits, if any, are seen to derive from its utilization. This review of NPT contributes to a body of knowledge about how theory may benefit implementation research, and if so what should the research agenda now be for NPT and other relevant theories in the field. Therefore, in this article our research objectives are to review the literature on NPT (see Additional file 1: PICO Table) in order to understand what interventions NPT is being used to analyze, how NPT is being operationalized and the reported benefits, if any, of using NPT.

Utilizing the interface engine EBSCOhost, the first author (RM) completed an advanced search incorporating the following electronic bibliographic databases: Academic Search Complete; AMED—The Allied and Complementary Medicine Database; Biomedical Reference Collection: Expanded; CINAHL Plus with Full Text; MEDLINE; OmniFile Full Text Mega (H.W.Wilson); PsycARTICLES; PsycINFO; Social Sciences Full Text (H.W. Wilson); UK and Ireland Reference Center (see Additional file 2). A search of both Embase and Pubmed was also carried out. In the search process, the terms ‘Normalization Process Theory’ or ‘Normalization Process Model’ were used. The first author (RM) also screened all citations linked to six key concept papers that were specific to the development of the theory/model [10, 12‐14, 24, 25]. Citations were also solicited from academics involved in the development of NPT. Google Scholar alerts were further activated during the search process, which commenced in February 2012 and ended on June 1st 2012. The final list of articles was then circulated to a number of experts in the field of NPT for review and no additional papers were noted. The issue of discrepancies did not arise.

Papers on the subject of ‘Normalization Process Theory’ or ‘Normalization Process Model’ in a health-related field that met all the following criteria were included: Published peer-reviewed empirical papers or papers that may be in press or accepted for publication; and an explicit reference to NPT or NPM in the article heading, abstract or keywords.

Given the decision to review the application of NPT in completed empirical work, study protocols, editorials, discussion/briefing papers, conference papers, or core concept papers that describe the theory and its development were not included (e.g., [13]).

Citations were downloaded into EndNote reference manager software, and full documents were imported into NVivo 10 software for analysis.

Given our interest in identifying a priori themes (i.e., the four NPT constructs), we adopted a framework analysis approach [26]. We also sought out emergent issues (e.g., country of origin, research focus, stakeholder involvement) that were considered relevant to the objectives of our review.

This framework analysis approach incorporated the following five key stages: familiarization, identifying a thematic framework, indexing, charting, and mapping and interpretation. For indexing and charting, the first author extracted material from any part of the paper that was relevant to an a priori or emergent theme. For the a priori themes, only data that were explicitly connected to NPT by the authors of the identified papers were coded.

The final stage of mapping and interpretation focused primarily on our a priori themes and had two objectives. First, to examine stability of the constructs across studies, we explored how NPT was being operationalized across settings. For this we analyzed whether data coded explicitly by authors as a particular construct had resonance with our understanding of that construct as per the core NPT papers. We also compared authors’ accounts of each construct as applied to their study setting and any relevant corresponding data. Given the subjectivity involved in interpretive analysis processes, our aim was to see if we could understand authors’ coding decisions rather than judging them to be ‘correct’ or ‘incorrect’ per se. Second, to determine whether there were any benefits to using NPT we focused on the authors’ reflections, usually located in the discussion section of the papers.

The mapping and interpretation of the data, and recommendations, were discussed with all co-authors and refined until consensus was reached.

From the 383 records screened, 354 were excluded and 29 full-text articles that met the inclusion criteria were retrieved (see Figure 1) (see Additional file 1: PICO Table). Due to the qualitative nature of the literature being reviewed, PICO as a search strategy did not neatly fit with our research questions, so we adapted the PICO Table to include the following criteria: participants, study design and collection approach, interventions, analysis, aims/discussion, and outcomes.

Drawing on Hawker, Payne, et al.[27] (see Additional file 3: Quality Appraisal Tool), each of the 29 papers was subjected to a quality appraisal process undertaken by three of the authors (RM, AMacF and LB). This process enabled us to make an informed decision about the quality of the reported research. All 29 papers were included, with scores ranging from 18 to 36 (maximum score being 36) and a mean score of 29.06. Study-specific appraisal results are included in Additional file 3. Overall, these were good quality studies.

To ensure consistency during the data abstraction and framework analysis phase of work, one-fifth of the articles were double-coded independently. Coding agreement was close to 100%.

NPT, and its four related constructs, were utilized in over one-third of the papers reviewed (n = 11) [15, 19, 28, 30‐32, 34, 35, 37, 44],[45]. One paper [41] focused solely on the construct of Coherence. The remaining papers (n = 17) [5, 18, 29, 33, 36, 38‐40, 42, 43],[46‐52] focused solely on the construct of Collective Action, or NPM as it was earlier known (see Table 2, column 2).

Our analysis of the application of the constructs across settings indicates that authors attributed meanings to each construct that, in general, resonated with our understanding of the constructs and had veracity in terms of their reported analysis and interpretation of data from their specific study setting.

For Coherence, there was a clear emphasis on understanding and conceptualization of interventions and their work (n = 8) [15, 19, 32, 34, 35, 37, 41, 45];

For Cognitive Participation, the emphasis was on notions of legitimation and buy-in, both in terms of the individuals involved and involving others (n = 8) [15, 19, 28, 32, 34, 35, 37, 45];

For Collective Action, the emphasis was on organizational resources, training and divisions of labor, confidence and expertise as well as the workability of the intervention in clinical interactions (n = 25) [5, 15, 18, 19, 28, 29, 32‐40, 42],[43, 45‐52];

For Reflexive Monitoring, the emphasis across studies was on appraising and monitoring implementation work (n = 9) [15, 19, 28, 32, 34, 36, 44, 48],[52].

There were, however, exceptions to this general finding, which typically related to the overlap between constructs. For example, in Gunn et al.’s [45] paper we considered that data about the doubts practice nurses and receptionists had regarding their role in delivering depression care could fit with Cognitive Participation’s subcomponent legitimation, and could also fit with Collective Action’s subcomponent skill set workability, which relates to the division of labor and the allocation of tasks. Sanders et al.[41] also reported all their data about a new system for treating back pain under Coherence, although for us much of the data appeared to relate to Collective Action (because the data were based on data generated with study participants based on their experiences of doing implementation work). However, the main point is that irrespective of whether the issue fitted best in one construct or another, the framework helped to raise important issues.

In almost all cases NPT was used as an organizing framework for analyses and reporting findings. It was also used to inform study/intervention design [5, 40, 41, 44, 46], to generate research questions for fieldwork [39, 41, 45], and to create tools for investigating and supporting implementation (TARS and eHIT) [15, 31] (see Table 2, column 3).

As presented within Table 2, column 4, while almost half of the studies (n = 13) [5, 30, 32, 38‐40, 42, 46‐50, 52] were multi-perspectival with involvement of professionals and service users, 12 of the remaining studies [15, 18, 29, 31, 33‐36, 41, 43‐45] focused on the perspectives of healthcare professionals only. However, it is evident that within the latter studies, different perspectives were often sought from within the healthcare profession (i.e., GPs, nurses, allied health professionals, senior management).

It is interesting to note that Gallacher et al.’s study [19] is the only one to focus solely on the patient’s perspective. While not multi-perspectival, it does reinforce the message that the implementation of complex health interventions owes as much to the work of patients as it does to service providers and other personnel in health and social care agencies [31].

The majority of papers reviewed [5, 15, 18, 19, 28, 30, 34, 35],[37‐40, 43, 45‐47, 49‐52] provided data about their experiences of using NPT, and the various challenges and benefits in using it.

MacFarlane and O’Reilly-de Brún [40] reflected on the challenge, for example, of overcoming tensions around using a predetermined conceptual framework and not wishing to ‘force data into predetermined codes or categories.’ They also described their concerns about understanding the constructs and ‘getting it right,’ stating that ‘it was sometimes difficult to know exactly what CI was about in this specific setting, and how it differed from SSW, and so on.’ The concern was that if the authors misunderstood the intended conceptual meaning of the constructs, their analysis would not be congruent with the NPM, and this would reduce the benefits of using the theory in the first place. Atkins et al.[38] and Franx et al.[35] shared similar views in that the application of the NPT constructs was problematic due to the aforementioned overlap and difficulty of discerning the differences between the constructs. Gunn et al.[45] remarked on the efforts required in developing each of the constructs within the complexity of current organizational practice.

Some authors commented on data falling outside of the NPT/NPM coding frame [19, 28, 50]. Mair et al.[28] noted that:

‘only 6% of issues fell outside of their coding framework, either because they were strictly technical and attitudinal or because they were so generic and vague, with accompanying contextual data, that it was not possible to determine whether the concept really lay outside the model or was simply too general to be coded.’

Gallacher et al.[19] commented on the fact that while very few data fell outside the NPT coding frame, those that did consisted of emotional work. Mair et al.[50] reflected on ‘attitudes’ and comment that this particular theme did not map onto the constructs of NPM. However, they acknowledged that unlike other theories such as the Theory of Planned Behavior, NPM relates to the work being done in interactions rather than focusing on how attitudes or intentions will affect work.

Overall, there was strong endorsement from researchers for the theory across a range of disciplines. Of the 20 papers that provided data about their experiences of using NPT, 15 [5, 18, 19, 28, 35, 37‐39, 43, 46],[47, 49‐52] commented that NPT was beneficial because it provided an explanatory theoretical framework for helping to identify factors that promote and inhibit implementation of complex interventions:

‘Our findings suggest that NPT provides a useful framework for understanding the processes that affect the implementation, embedding, and integration of new technologies into healthcare systems’ [18].

‘The findings suggest that NPT is a theoretical framework that facilitates understanding of experiences of health care work at the individual, as well as the organizational, level’ [19].

It was also emphasized that NPT had assisted in making clear recommendations for future implementation (n = 11) [5, 18, 28, 35, 38, 39, 45‐47, 49],[51]:

‘The model also assists in making clear recommendations for future implementation. This was important as this program was a pilot with a view to inform service decisions on whether and how to scale up the program across the province. We anticipated that the model could provide insights regarding the factors that would lead to normalization of the program’ [38].

Finally, some authors discussed the positive impact of NPT on trial design and development of an intervention (n = 3) [5, 46, 52], for example:

‘The NPM provides a useful structure for both guiding and analysing the process by which an intervention is optimized for testing in a larger scale trial or for subsequent wide-scale implementation’ [46].

At the same time, authors suggest further development of NPT [15, 18, 38, 50] relating to: the determination of its value to guide the development of interventions for use in routine healthcare [18]; the need for study-specific measure/application of NPT [15]; the development of tools and methods to assist in the use of the NPM [38]; its potential to be used as a tool to assess the likelihood of future normalization of a complex intervention [50].

First, this review did not include non-English language papers and therefore we cannot comment on the use and perceived value of NPT in non-English speaking settings. However, it would be valuable to do so in future reviews and to explore its use and stability across cultural settings, given that our search identified a number of foreign language papers using NPT from countries such as Sweden and Italy and the use of NPT in three European-wide research projects: REsearch into implementation STrategies to support patients of different ORigins (RESTORE) [6], Self-care Support for People with Long Term Conditions, Diabetes and Heart Disease: A Whole System Approach (EU-Wise) [57], and INnovative, Midlife INtervention for Dementia Deterrence (In-MINDD) [58].

Second, reviews are one step removed from the primary data, and therefore we rely on the authors’ reports of benefits and limitations of NPT usage, which could be limited or sanitized versions of their experience. No attempts were made to contact authors for additional information.

In terms of gaps in the information provided by the studies, we noted that in the quality appraisal process the lowest scoring domain tended to be in terms of ethics and bias, primarily because there was very limited discussion of either of these issues within the papers. As an example, while authors reported whether they had ethical approval or not, they did not elaborate on ethical issues in the research process. This limited reporting about ethics and bias is likely to be a function of the word count restrictions in journal articles.

As this was a qualitative review of predominantly qualitative empirical studies, some aspects of the PRISMA [59] statement were not applicable (see Additional file 4).