|Home | About | Journals | Submit | Contact Us | Français|
The field of improving health care has been achieving more significant results in outcomes at scale in recent years. This has raised legitimate questions regarding the rigor, attribution, generalizability and replicability of the results. This paper describes the issue and outlines questions to be addressed in order to develop an epistemological paradigm that responds to these questions.
We need to consider the following questions: (i) Did the improvements work? (ii) Why did they work? (iii) How do we know that the results can be attributed to the changes made? (iv) How can we replicate them? (Note, the goal is not to copy what was done, but to affect factors that can yield similar results in a different context.)
Answers to these questions will help improvers find ways to increase the rigor of their improvements, attribute the results to the changes made and better understand what is context specific and what is generalizable about the improvement.
This article raises an important issue in the field of improving healthcare today: ‘How do we learn about improving healthcare, so that we can make our improvement efforts more rigorous, attributable, generalizable and replicable?’ (See Fig. 1). Intended to outline what we know and what remains to be answered, the article represents a synthesis and analysis of the knowledge and experiences of the authors and reviewers. Although we acknowledge that the improvement community's opinions differ regarding the definition of certain terms, this article is framed by the working definitions found in Table 1. The paper concludes with key questions that need to be addressed in order to advance the field of healthcare improvement.
The World Health Organization (WHO) describes quality care as care that is effective, efficient, accessible, acceptable, patient-centered, equitable and safe . Yet, much of the care received in high-, middle- and low-income settings does not meet the WHO criteria, often due to the complexity of healthcare―which is why we need to improve healthcare . For the purposes of this paper, we will use the following working definition for improving healthcare: ‘The actions taken to ensure that interventions established to be efficacious are implemented effectively every time they are needed.’
The multi-level structure found in healthcare settings exists as interconnected sets of autonomous healthcare providers, teams and units within healthcare organizations, nested within health systems. The dynamic human interactions between and among healthcare workers, patients, managers, payers and other actors and the variety of social, cultural, economic, historical, political and other factors within this multi-level structure put improving health care squarely in the arena of complex adaptive systems . Complex adaptive systems are comprised of individuals who learn, self-organize and evolve in response to changes in their internal and external environment, and inter-relate in a non-linear fashion to accomplish their work and tasks [3, 4]. These complex adaptive systems, in which implementers operate, present unique challenges. They cannot be reduced to their component parts without the risk of negating the original intention of the intervention, therefore interventions must be implemented in a contextually adaptive manner in order to effectively work in any given system.
The field of improving health care has evolved to address the complex, interdependent, systemic nature of healthcare challenges, particularly through the use of adaptive, iterative testing and implementation of changes, and by empowering teams to use data in real time to do so. Teams and individuals who learn in these complex, adaptive systems use real-time data to assess whether introduced changes lead to an improvement in the outcome of interest—after which, they institute, adapt or discard these changes as needed, in a continuous cycle of testing and learning. This process is demonstrated in the example in Fig. 2.
Improvements like these have generally relied on the use of time-series charts, which plot indicators of the improvement sought against frequent (daily or weekly) time intervals before, during and for a time period following the introduction of changes. At times, they have also used control limits, but as a rule, improvement has relied on analytic statistics, as opposed to enumerative statistics, to establish significance .
Improving healthcare engages all key stakeholders involved in the outcome of interest, including healthcare providers directly involved in delivering care, together with their supervisory and support structures, as required. This process can also include patients, families, communities and other stakeholders. By meaningfully engaging all key stakeholders, health teams are able to address systems and micro- and macro-political human-factors issues  and build ownership through restructuring processes. In addition, the adaptive, iterative nature of testing and learning allows the teams to implement changes appropriate to the local context and responsive to emergent and shifting dynamics among different actors and at various levels of the system.
Over the past two decades, the application of modern improvement methods has expanded beyond the administrative processes in facilities, where it was first applied (for example, to reduce waiting times) to include clinical improvements (like reducing the incidence of hospital-acquired infections) and more significant improvements in health care (like reducing secondary complications and decreases in mortality); and, moreover, achieving these outcomes and results at scale [7, 8]. This evolution has generated greater interest in improvement and simultaneously raised more questions regarding the validity, rigor, attribution, generalizability and replicability of the results.
In response to these legitimate questions, there have emerged two schools of thought. One school of thought calls for continuing to accept analytic statistics as sufficient evidence for improvement . The other calls for subjecting improvement to the same enumerative statistical methods used in clinical research, such as the use of randomized control trials.
The use of randomized controlled trials, the gold standard for clinical research, has been limited in the field of improving health care. In part, this is because randomized controlled trials often presume a linear, mechanistic system; however, improvement takes place within complex adaptive systems, which do not lend themselves neatly to this type of study. Furthermore, as illustrated in Fig. 1, our interventions must necessarily adapt to the context, which is often at odds with the conceptualization of improving health care as a fixed set of activities which can be studied through controlled trials . Other constraints for the use of randomized, controlled trials include donor and host-country government constraints and ethical issues regarding randomization of interventions with proven efficacy, particularly, for those in low- and middle-income country settings [10, 11].
Improving health care is the act of taking an efficacious intervention from one setting and effectively implementing it in different contexts. It is this key element of adapting what works to new settings that sets improvement in contrast to clinical research [12, 13]. The study of these complex systems will therefore require different methods of inquiry. Such methods may include, but are not limited to, stepped-wedge designs, comparison groups with calculations of difference-in-difference, qualitative evaluations to understand why and how the interventions worked, and the use of mixed-methods’ approaches, including randomization and observation. Toward this end, there is a need not only for evidence-based practice, but also for ‘practice-based evidence’—which is relevant to the stakeholders responsible for implementation—and for systems that allow for rapid learning in order to build this knowledge. This knowledge may be generalizable or may need further adaption for a different context .
A 2015 literature review by Portela et al.  provided a useful overview of various methods that can be used to learn about improving health care and describes the strengths and weakness of various approaches that range from more traditional, experimental designs to quasi-experimental designs, as well as systematic reviews, program and process evaluations, qualitative methods and economic evaluations. The authors note that the dichotomy between designs classified as practical (‘aimed at producing change’) and those classified as scientific (‘aimed at producing new knowledge’) may be a false one, and that the field should find ways to optimize rigor and generalizability of studies, without compromising the importance of adaptability and context.
On this note, Davidoff et al. and other authors call for the demystification of the use of theory in improving health care , including more use of theory a priori to better understand how and why an improvement occurs and provide insight into the so-called ‘black box’ of improvement. Parry et al. provide a guide for a formative, theory-driven approach to evaluating improvement initiatives by defining three improvement phases for initiatives (innovation, testing, and scale up and spread), each defined by the degree of belief in the intervention, and each requiring a different evaluation approach .
Campbell et al.  also emphasize the importance of phasing; for example, in the use of randomized controlled trials for complex interventions. The authors describe a more flexible approach to the 2000 Medical Research Council framework  on this topic by considering an iterative, stepwise approach to building understanding of the context of the problem, intervention and evaluation, in order to obtain meaningful information from randomized, controlled trials of complex interventions.
Braithwaite et al. urge improvement teams to learn from the way clinicians adjust their behavior with their concept of Safety II: ‘[w]e must understand how frontline staff facilitate and manage their work flexibility and safely, instead of insisting on blind compliance or the standardization of their work’ .
Others have called attention to different ways to integrate adaptability with the fixed concept of randomized, controlled, trial interventions (delineating between standardization of, but not the form of, the intervention)  and the need for earlier pilot testing with iterative learning and non-linear evaluation processes, in order to more fully understand complex adaptive systems .
Furthermore, as a social intervention, improving health care is complex and, therefore, difficult to understand, design, implement, reproduce, describe and report. Frameworks, such as the Consolidated Framework for Implementation Research, developed by Damschroder et al. , provide guidance for ways to evaluate complex adaptive systems by classifying them into various domains, which describe various intervention and contextual factors relevant to the evaluation. The Template for Intervention Description and Replication (TIDieR) guidelines developed by Hoffmann et al.  and the Standards for Quality-Improvement Reporting Excellence (SQUIRE) Guidelines  provide a guide for improved reporting of interventions in order to improve the completeness of reporting and aid replicability .
The work presented by the above-mentioned authors is rooted in the ideas of showing that improvement has occurred, why it has occurred, and how to better learn about improvement to make it happen more effectively and in other contexts. There is evidence to support these objectives and we must build upon this to move improvement forward. These are also the themes that lead us to the next level of questions for improvement.
In order to learn about improving health care, there is a need for thought leaders in improving health care, as well as from related fields that also use complex adaptive systems; researchers; and others to come together to co-develop a robust framework that has widespread support and that reflects the diverse, nuanced ways we learn about improving health care. Many questions remain to be answered. These include, but are not limited to:
Answers to these questions will necessarily lead to multiple study types, depending on what we want to learn, in what context, and for what purpose.
A phenomenon that is relevant to learning in general which also equally applies to improving health care, is that successful work which produces results tend to get published. Learning comes from not only what has worked and why, but also what has not worked and why not. In order to enhance learning we need to be deliberate about studying not only success, but also failures. The changes which failed, why they did not work and under what circumstance [10, 27].
A new framework for how we learn about improvement will help in the design, implementation and evaluation of improving health care to strengthen attribution and better understand variations in effectiveness through reproducible findings in different contexts. This will in turn allow us to understand which activities, under which conditions, are most effective at achieving sustained results in health outcomes.
The complexity of health care requires a more rigorous approach to advance our understanding of methods for learning about improving health care. Additionally, the greater use of robust qualitative, quantitate and mixed methods is needed to assess effectiveness—not merely to demonstrate if an intervention works, but why and how it works—and to explore the factors underlying success or failure.
Key questions to examine further include how to strengthen the rigor of the improvement; increase attribution of results to the changes tested; provide better balance to the often opposing needs of improving fidelity of the intervention, versus allowing for adaptation; make conclusions that are generalizable, but that also respond to the local context; and account for political considerations in improvement activities. This can lead to an improved epistemological paradigm for improvement.
The authors would like to acknowledge the contributions of a number of reviewers who guided this manuscript: Bruce Agins, Brian Austin, Don Goldmann, Frank Davidoff, Jim Heiby, Lani Marquez, Michael Marx, John Øvretveit, Alex Rowe and Alexia Zurkuhlen.