When complexity science meets implementation science: a theoretical and empirical analysis of systems change

In what now seems to us like the distant past, yet, in reality, was merely a decade or so ago, medical scientists believed that the translation of research evidence into practice followed a prescribed set of research steps, moving from test tube to needle, or bench to bedside. It was common to apply the concept of a ‘pipeline’ as a heuristic for understanding research uptake. Adherents to this view frequently diagrammed the process as a linear one, conceptualizing interventions through a series of stages starting from the laboratory, into the randomized trial environment, and then across real-world settings.

Such models implicitly assumed that those on the clinical frontlines would naturally provide new types of care, such as novel pharmaceuticals, practices, or innovative technologies, based on the latest evidence, and all heavily informed by upstream research. While various research pipeline models were proposed over the years, all were similar in that research evidence was assumed to advance in a rational, step-wise manner. One influential model, described in the Cooksey report [3] (Fig. 1), was developed following a review of health research funding in the UK examining the critical pathways to successful research translation; it is often referred to, and equivalent models have been developed in other countries [4, 5].

However, the linear, rational way in which such a model assumes that evidence is converted into practice masks the complexity of the research–practice ecosystem [6, 7]. It hides much of what is important in trying to accomplish evidence-based medicine, namely, that basic research is fundamentally risky and often does not produce any useable breakthrough; that some ideas never even reach the prototype stage, let alone pre-clinical development; that even if developments progress to a trial, this may prove unsuccessful; that health services research is relatively poorly funded and thus implementers often fall short of truly understanding how socio-professional systems work in practice; and that the ‘translation gaps’ (more like chasms) between research findings and their use in practice often cannot be bridged [8–10].

This traditional manner of thinking about research pathways was founded on a Newtonian-style, clockwork universe paradigm, representing a mechanistic and reductionist view of the way the world works, dominated by the randomized clinical trial and precision measurement. In reality, when we deal with non-mechanical human systems, this view has serious limitations [11]. To extend the metaphor, in contrast to a Newtonian view, the health system is more quantum mechanical than classically clockwork, and is characterized by uncertainty, emergence, and embedded unpredictability. Participants exert effects on the system; sometimes, the system appears wave-like (akin to group behaviors), sometimes particle-like (with individual agents’ efforts having influence), and it changes once measured or observed, because measurers and observers are entangled within the system and each other. The health system is probabilistic and stochastic rather than deterministic and causal.

Some 10 to 15 years ago, several thinkers began to realize the limitations inherent in the pipeline idea [12] as it became increasingly obvious that getting evidence into practice was much harder than earlier proponents believed. This recognition came from the knowledge and understanding of human systems that had been accumulating in sociology, ecology, and evolutionary biology ever since the 1940s, and with antecedents even earlier, which we can loosely call ‘systems thinking’.

The systems view is based on several fundamental ideas, essentially, that all systems are composed of a set of seemingly discrete but actually interdependent components, defined not just by their inter-relations but by the permeable and shifting boundaries between them. The components (people, technology, artefacts, equipment) are combined haphazardly and in unexpected ways, aggregating to be more than the sum of their parts, and are characterized by eddying, recurring patterns of behavior. Key moments in the path of articulating a systems view of the world arose through the work of many theorists, but management scientist Peter Checkland [13], biologist Ludwig von Bertalanffy [14], and organizational theorist Andrew Van de Ven [15] can be used as proxy exemplars.

Checkland’s pioneering work [13], beginning in the 1960s, was encapsulated under the title ‘soft systems methodology’. This approach differentiated between hard systems, represented by relatively rigid techniques, technology, artefacts, and equipment, and soft systems, which involve the learning that occurs in fuzzy, ill-defined circumstances as people navigate across time in messy ecosystems.

Von Bertalanffy’s ideas date decades earlier, and his development of ‘General System Theory’ laid the platform for much of the later work. He, in turn, drew on even earlier sociological, mathematical, and biological research and theories, and by approximately 1946 he had assembled General System Theory, applying universal principles and espousing the ontological underpinnings for the interactive and dynamic nature of social organization and structuring [14, 16].

For Van de Ven and his intellectual successors, the trajectories to an innovative outcome always have several variations, multiple pathways, unanticipated processes and results, and exhibit conflict between stakeholders. People flex and adjust, accommodating to local conditions, and always deviate from idealized pathways.

Innovation processes for Van de Ven are neither stable and predictable nor stochastic and chaotic. Being an innovator implies working with inherent unpredictability, sometimes with random effects, and dealing with the multiplicity of internal and external forces that impinge on and are intrinsic to the journey. Sometimes, innovators need to run with the pack, and at other timess do so in opposition. Persistence in the face of setbacks and an ability to work with, or simply just understand, multiple agents who inhabit indistinct, orthogonal, or oppositional cultures and subcultures, facing sometimes destructive and sometimes constructive politics, and experiencing periods of inactivity, are all features of the innovation journey.

From 2004, this rich theorizing and new-fashioned conceptualizing of the ontology of improvement pathways began to be applied more concertedly to healthcare. Many of these ideas converged in Greenhalgh’s work on the diffusion of innovation, where she and her colleagues brought together disparate research in an influential paper that provided an extended systems model articulating the intricacies, problematics, and minutiae of getting evidence into practice (Fig. 2) [12]. The Greenhalgh model suggested four pivotal systems factors important for innovation, namely the innovation itself, and its characteristics; the system’s propensity, or its readiness, for change; the journey or implementation process; and the external or outer context. For ease of access and readability, we have streamlined this model by rationalizing the number of variables that Greenhalgh et al. [12] stipulated in their original work. Of course, all models are simplifications of reality and even one that acknowledges a very large number of variables is, nevertheless, merely a model that reduces real-world complexity for the purposes of explication.

This is not to deny that there are, at the broadest levels, iterative roadmaps from bench to bedside or test tube to needle. However, this assessment does illuminate the reality that there are many components, moving parts, and shifting relationships, and that innovative journeys are much more convoluted, imprecise, uncertain, ambiguous, and deceptive than the pipeline proponents realized or hoped for. Social science had been waiting in the wings, eager to point this out, and have the mechanistic pipeline view excised. It brings to mind the English poet, David Whyte, who aphoristically said, “Stop trying to change reality by attempting to eliminate complexity” [17], and Abdus Salam, the Pakistani theoretical physicist and Nobel Prize winner, who once remarked, “From time immemorial, man has desired to comprehend the complexity of nature in terms of as few elementary concepts as possible” [18]. Yet, more mechanistic, simplified views of the world cannot wish away its complexity.

That said, there are some today who still persistently hold a traditional pipeline view, even in the face of experience with its shortcomings. At bedrock, this most likely has something to do with the architecture of the human mind, which often sees things in cause-and-effect terms [11, 19]. The brain has evolved to compose a narrative, linear account of events that unfold with a past–present–future representation of how things work [11, 19]; this forms part of the executive function of the brain responsible for planning, organizing, and reasoning [20]. Of course, the mind is also capable of out-of-the-box creative thinking, but straight-line rationalizing frequently trumps other ways of imagining how the world works.

Complexity theory can be applied at multiple scales, from the very smallest, ranging from quantum foam to quarks, to the minutiae of the chemical and biological underpinnings of matter, to the behavior of molecules and cells, up to the macro interactions of humans, their groups, and even entire civilizations [21]. Complexity science has more recently been utilized in healthcare in order to apprehend, for example, the management, safety, and organization of clinical services [22, 23], as well as the implementation of interventions and the translation of evidence into practice [24].

Complexity science challenges conventional wisdom and an unduly straight-line approach to implementation on a number of fronts. Traditionally, people have studied parts of a system (the people, the intervention, the outcomes) as distinct variables, assuming the influences on one another to be straightforward [25], or at least knowable. These effects were conceived as additive, where the sum of the parts equaled the whole and a predictable relationship existed; that is, causes were identifiable because they preceded effects, and led to them. In designing interventions, people in this mode have aimed for reduction and control, removing the influence of, or controlling, ‘extraneous’ or ‘confounding’ variables [26]. Researchers and implementers then inferred the ability to generalize findings derived from this approach across contexts. Thus, an effect observed through well-controlled experimentation in one environment would be assumed to occur similarly in other situations; this may have worked in some cases, but by no means always.

In contradistinction, in complexity science, while the components of a system, namely the agents and their artefacts, are important, they are often secondary to the relationships between these components [27]. In such systems, agents communicate and learn from each other and from their environment, and adjust their behavior accordingly. However, there are many cross-cutting interconnections and influences. As such, the system is best described as a CAS, meaning that it has the capability to self-organize, accommodate to behaviors and events, learn from experience, and dynamically evolve [28], but not necessarily in ways anyone can forecast with any degree of confidence.

The self-organizing, iterative, reverberating interactions among agents, which in the healthcare CAS includes stakeholder groups such as doctors, allied health, patients, nurses, managers, and policymakers, as well as many other subgroups, give rise to unpredictability and nonlinearity, with causes and effects often disconnected or disproportional to one another [19, 25]. CASs are distributed in space and behave dynamically across time, with idiosyncratic interactions among agents at the local level determining the context, and the present and future behaviors of the system [24]. Through the interactions among the system’s components, global system patterns emerge and new factors (e.g., technology, policy, novel relationships, practices) eventuate.

These patterns are influenced by feedback loops, where different system inputs at different points in time perpetuate their own outputs, either dampening or enhancing them. Feedback helps to explain how responses to interventions, which might be positive at first, are often not sustained. The relatively loosely or tightly coupled interconnections between agents within a CAS, and their changeability over time, suggest there is much propensity for unintended consequences of an intervention in addition to the improvements agents hope for [29]. Borrowing from Gould and Eldridge’s famous distinction in evolutionary biology [30], health system progress in such circumstances resonates much more with the idea of punctuated equilibrium than that of morphologic gradualism.

More recently, the efforts to study methods and mobilize knowledge, designed to enhance the ways in which we acquire and use evidence in healthcare, have been termed ‘implementation science’. For convenience, we can date this idea from the first issue of Implementation Science in 2006, although some scholars had been working on the development of this field before then. Implementation science is not a unified approach to getting evidence into practice, but rather comprises diverse perspectives, frameworks, and methods. However, broadly, implementation science is characterized by three aims, namely (1) to describe the process of translating research into practice (process models), (2) to understand what influences implementation outcomes (determinant frameworks, classic theories, implementation theories), and (3) to evaluate the implementation of interventions (evaluation frameworks) [31].

Despite their differences, the two theoretical paradigms can be used together to the benefit of both theory building and healthcare practice and systems improvement. The complexity lens can help illuminate the scope of the implementation problem to be tackled and the dynamics of change and inertia. The translation of evidence into new clinical or organizational practices does not unfold in a static and controlled environment awaiting the attention of top-down change agents; it takes place in settings comprised of diverse actors with varying levels of interest, capacity, and time, interacting in ways that are culturally deeply sedimented, and have often solidified [32, 33]. In other words, the complex patterns by which healthcare is delivered, and the enmeshed social structures inherent within the system, are already established and entrenched. In such a networked, at times tightly and at others loosely coupled ecosystem, already teeming with activity and relationships, knowledge uptake is rarely simple or straightforward, and has to find a place in an intricate, pre-existing milieu.

Going further, spread is closely related to uptake. The patterns of interaction between agents and their environment are locally specific, and although they share features with other CASs, they also exhibit remarkable variation from one site to the next. The notion, then, that a new practice can be adopted equally well and in the same manner across a whole health system, is untenable. Thus, standardization of an intervention, and assuming its generalizability, can be the downfall of successful implementation [34].

However, implementation scientists, or at least those working within implementation science with pluralistic conceptualizations of the world, have not been standing still. The need to factor in context is being increasingly recognized by scholars in implementation science, as is the identification of barriers and facilitators to an intervention [35]. For example, the Promoting Action on Research Implementation in Health Services formula [36] sees successful implementation as a function of the explicit interrelations among evidence, context, and facilitation. Nevertheless, these contextual characteristics of the environment are often viewed as ‘confounders’ in implementation research, rather than the normal conditions of practice in healthcare. Complexity science, in highlighting the dynamic properties of every CAS and the local nature of each system’s culture, suggests that what operates as a ‘barrier’ to implementation in one site may not do so in another, and could even be facilitative [24].

In complexity-informed approaches to implementation it is not enough to leverage facilitators or eliminate barriers; the focus of implementation shifts from the fidelity of the intervention to its effective adaptation [37, 38]. Thus, Hawe et al. [34] argue that, rather than standardizing aspects of an intervention, despite some essential functions being replicable, the form of an intervention should be varied as required by context [39]. This type of CAS-oriented approach is particularly important when attempting to scale-up or spread interventions previously found to be effective in one, or a limited number of sites, to the whole system. Improvement structures may thus involve tailoring to context and harnessing the self-organizing and sense-making capacities of local agents [38]. Indeed, working with bottom-up local stakeholders is paramount to adapting an intervention to their practices, facilitating ways to get them onboard with the intervention, in piloting it, in reflecting on progress amongst stakeholders, and in providing feedback to participants to help them embrace implementation iteratively over time. In such a messy, complex set of circumstances, it makes less and less sense to think of ‘knowledge producers’ as conceptually distinct from ‘knowledge users’ [40] when indeed they are inter-related.

Chambers et al. [41] suggest that a further consideration is the sustainability of an intervention. Sustainable change requires the ongoing adaptation of an intervention to multilevel contexts, with expectations for lasting improvement rather than diminishing outcomes over time. In this regard, implementation in the hands of complexity theorists is increasingly recognized as an iterative and recursive, long-term process rather than a linear one [35]. Complexity science thereby encourages not only attention to the context of an intervention, but also to the interactions between elements and the consequences of this intervention for the system. The implementation method of choice will not necessarily be the randomized clinical trial or experimental design, but will be the iterative and responsive, more ecology-aware, social science-informed approaches such as those envisaged by longer term realist designs or process evaluation of implementation efforts [32, 42].

Despite the potential utility in harnessing complexity science for implementation, until now, not much conjoining of the two, either theoretically or empirically, has occurred. There have been intermittent examples of using a complex systems framework to inform clinical transformation, as when Best et al. [43] applied complexity thinking in the implementation of new clinical guidelines in British Columbia, Canada. They noted that the implementation of the guidelines required the ability to tailor system-level recommendations to local context. In another promising turn, there have been more recent attempts to explicitly challenge the pipeline view of knowledge translation, with Kitson et al. [40] undergoing an iterative process to develop a complexity-informed model that highlighted the connections between phases previously conceptualized as discrete such as problem identification and knowledge synthesis. This model (Fig. 3) in essence highlights the key issues to be considered, including the distinctions and connections between knowledge users and knowledge generators, the importance of arriving at good definitions for the gaps, and co-producing new knowledge and contextualizing it, as well as implementation and evaluation.

That said, a recent systematic review by Brainard et al. [29] found that health interventions using complexity science approaches have done so inconsistently, for example, often not incorporating an evaluation component or failing to analyze the potential, unintended consequences of the intervention. Nevertheless, this recent work has suggested the value of complexity science in creating large-scale system transformation, including sensitizing stakeholders to the natural properties of CAS that might then be leveraged by emphasizing distributed leadership, networks, sense-making, and feedback loops [38, 42, 44].

Thus, thinking is altering, at least amongst some leading theorists and researchers, and we are now more advanced in understanding systems change, with new models replacing the pipeline approach. Having established the juxtaposition of complexity and implementation, we now examine how some of these ideas have been leveraged to accomplish large-scale system transformations in Australia, exploiting the combined complexity–implementation paradigm.

Case 2: New nation-wide safety and quality standards

In 2013, systems-level reform of the Australian accreditation model occurred with the implementation of the Australian Health Service Safety and Quality Accreditation Scheme. A critical component of the scheme, overseen by the Australian Commission on Safety and Quality in Health Care (ACSQHC), has been the development and application of new National Safety and Quality Health Service Standards (NSQHSS). The development of the 10 standards represented an important element in the safety and quality of care architecture of the health system. The standards cover areas including governance arrangements, partnerships with consumers, and eight key clinical areas of health service operation (Box 1).

Each standard has a set of criteria, and for each criterion, a series of actions are required to be fulfilled. To achieve accreditation status, all core actions for health services must be demonstrated. The work has drawn international interest and is informing efforts to improve the safety and quality of healthcare in other countries [54].

The Australian Health Service Safety and Quality Accreditation Scheme has been enacted with an appreciation of the CAS features of healthcare, and the implementation process was dynamically modified in response to the multifarious and interlinked institutions, groups, and structural arrangements that can hinder or facilitate implementation, and must ultimately adopt the model. International experience shows that the inherent complexity of healthcare and in-built resistance, regardless of country, can be an impediment to adoption of such systems-level reforms [55–58].

To respond to this challenging environment, the ACSQHC undertook extensive consultation activities with the aim of determining appropriate methods of utilizing existing government legislative powers to support the reform measures, to align the views and actions of diverse groups, and to foster distributed leadership across reform elements [59–61]. In total, the ACSQHC arranged 227 separate consultation activities involving over 1000 stakeholders spanning the breadth of the Australian health system. The perceived importance of these activities for maximizing the effectiveness of the scheme reinforces the fundamental role of continued stakeholder engagement as a necessary facilitator of national reform [54]. The need for effective stakeholder engagement has also been identified in relation to other systems-level healthcare reforms internationally [62, 63]. The ACSQHC continues to undertake consultation with health services to facilitate effective implementation of the scheme and further revisions have been made to the standards over time (in 2016 and again in 2017), assuring their continued relevance [59–61].

Despite the nature of the standards’ implementation as seemingly a top-down, government-sponsored, homogeneous model, NSQHSS have been well received by the system due to the clinical focus of most of the standards. This was considered crucial for increasing the engagement of health professionals and board members in health and quality improvement activities [54]. Participants proposed that the NSQHSS provided, for the first time, a clearly evidenced-oriented, coherent, and integrated national framework. The scheme separated and clarified responsibilities of different actors for accreditation standards development, surveying processes and decisions, and regulation and policy matters. As a result, the initiative has been seen to mobilize expectations, integrate roles and responsibilities, and promote transparency [54].

From the outset, two potential risks to the credibility of and satisfaction with the scheme at the health system level were raised, namely the application of the NSQHSS across varied settings and the reliability of assessments by different accrediting agencies. The application of the NSQHSS across settings was discussed in the consultations as a point of credibility – that the same expectations would be applied to different health services, in different settings, was considered vital to the government’s interests in equity [54].

Four strategies to facilitate implementation, to reinforce the potential benefits, and to overcome the substantial challenges facing the scheme emerged (Fig. 4). The widespread ACSQHC consultation activities were seen to facilitate implementation by providing a common platform for knowledge transfer, encouraging widespread stakeholder engagement. At those meetings, high-quality, accessible educational activities and materials were provided. Feedback loops in the form of regular review of the program and updates to the system using progress data helped maintain momentum.

Pipeline models sprang initially from those adhering to a linear worldview of the path from knowledge creation, through knowledge products, to knowledge use. The task was to get evidence into practice, and this was seen by many as a simple, staged activity, following recipe-style models such as the one expressed by Cooksey [3]. In the minds of many scholars and practitioners, including some who self-define as implementation scientists, the process of bench to bedside has, by and large, continued to be conceptualized in a largely mechanical frame, although some researchers and theorists have introduced complexity ideas to it [7, 40, 64]. Complexity science offers a radically different set of considerations to those interested in systems change. As a paradigm, it denies over-simplification, and is conceptually transformative, adding a much richer set of understandings to the task of systems improvement.

The two traditions of implementation science and complexity science can be drawn together, and culminate in more textured, multi-dimensional, complexity-informed models. Paradigm-shifting exemplars that have achieved this include those offered by Greenhalgh et al. [12] on innovation (Fig. 2) and Kitson et al. [40] on knowledge transfer (Fig. 3).

The key is to harness such understanding to strengthen progress with other multifaceted health systems interventions. Based on these examples, the portents are for future change agents to conjoin complexity science and implementation science approaches for the benefit of systems-level change.

Notwithstanding this analysis and these case exemplars, we conclude with a word of warning. Complexity thinking adds a real-world, multidimensional appreciation of the system and its density and dynamics, but it does not make it easier to effect change; in fact, the opposite is true. We can no longer assume to solve health systems issues by pretending or conspiring to imagine that they have Newtonian properties, and pipeline models should be seen for what they always were – idealistic, normative renderings of the world. Even though this makes our ambitions to improve healthcare infuriatingly more difficult, we must grapple with the world we actually inhabit, not the one we wish we did.