Clinical practice change is difficult. This is illustrated by the proverbial 17-year time lag from idea to practice and the lag in evidence use in practice [1]. Terms such as “creature of habit” and “resistance to change” have been used to describe humans in general and physicians in particular [2–4]. There are many models of physician behavior change but little agreement beyond the difficulty of accomplishing it [5, 6]. A variety of factors have been proposed to account for the difficulty, most focusing on barriers to learning and implementing new knowledge [7, 8]. Consequently, conceptual frameworks of implementation include behavior change at their core [9–11]. The Theoretical Domains Framework is an overarching framework of behavior change which integrates 128 constructs from 33 different theories [12, 13]. The recognition that there is considerable use of low-value care and even zero or negative value care (harm) has prompted national initiatives such as Choosing Wisely [14]. Addressing this issue requires not necessarily learning new knowledge, but rather unlearning old and outmoded knowledge [4, 15]. Many terms in addition to unlearning have been used to describe this process, e.g., de-implementation, de-adoption, and de-diffusion [16, 17]. Although physicians have a difficult time adopting new practices, they seem to have an even more difficult time “unlearning” outmoded ones [4].

Unlearning, which can occur at the individual and/or the organizational levels, has been defined in many ways, and the nature of unlearning itself is still contested. For example, the phenomenon of extinction describes a form of behavior change in which the strength or rate of a behavior declines when the reinforcing outcome is eliminated such that the behavior goes away, i.e., it is extinguished [18]. Some have considered this to be an example of unlearning while others contend that the original behavior is not erased, but remains, leaving the possibility of relapse [18–20]. Similarly, avoidance behavior and counterconditioning may be relevant to the general phenomenon of unlearning [19, 21, 22]. Notwithstanding these considerations, applied to the management context, Becker defined unlearning as “the process by which individuals and organizations acknowledge and release prior learning (including assumptions and mental frameworks) in order to incorporate new information and behaviors” [23]. Rushmer and Davies distinguished between routine unlearning (and subsequent re-learning) and deep unlearning which requires a substantive break with previous modes of understanding, doing, and being [24].

Most healthcare-related studies of unlearning have usually adapted models of change from the fields of business and management, focusing on identifying various drivers and barriers to the unlearning process faced by healthcare professionals, including those in management positions [17, 23, 25–34]. There have been studies that describe a variety of psychological and other phenomena that might affect a physician’s ability to change practice in general and unlearn specific clinical practices in particular [4, 8, 35–42]. Many of these studies use constructs from the Theoretical Domains Framework. Nevertheless, the topic of unlearning by physicians remains relatively unexplored.

The objective of this study is to elucidate the experience of trying to abandon an outmoded clinical practice and its relation to learning a new one. In so doing, this study aims to better understand the intricacies of individual physician unlearning in order to inform more successful de-implementation strategies for outmoded, ineffective clinical practices.

This qualitative study utilized a grounded-theory approach aligned with the development of grounded theory that flows from the transition of Strauss and Corbin and Charmaz, which allows investigators to listen open-mindedly to the experiences of participants in real-life situations and enable “continuous interplay between data collection and analysis” [43, 44]. For quality control, we utilized the consolidated criteria for reporting qualitative research (COREQ) [45].

We obtained our data from interviews and used an iterative process to refine the interview methodology. After conducting an interview, we reflected on which questions were most successful in eliciting responses that reflected the interviewees lived experience and which questions were less generative. Using a constant comparative approach, we made adjustments in our interview method, using theoretical sampling to guide the respondent selection process, seeking data to develop the properties of theoretical categories until the categories had been saturated, i.e., we were not surfacing any new dimensions or properties in subsequent interviews [43].

Data analysis

To analyze the data, one team member (DG) generated codes that both summarized and categorized each piece of data in the interview transcripts [43]. The other team members reviewed the coding. Differences were resolved by consensus. These codes assisted with the interpretation of the data and the identification of underlying themes. We used a three-step process described by Strauss and Corbin to code the data, starting with open codes, followed by axial codes, and ending with theoretical codes. We identified a total of 204 open codes, which were distilled into 52 axial codes, which were then further synthesized into 5 theoretical codes (or themes). All three team members applied abductive reasoning to those themes, asking “what core set of elements and relationships are implied by these themes that can explain the unlearning successes and failures that our subjects reported?” Our findings emerged from that multi-stage process. An example of this coding process can be seen in Fig. 1.

Three of the emergent themes ultimately identified by this coding process concern the following: disturbing the equilibrium with practice change, practicing with a purpose vs. cookbook medicine, and questioning “evidence-based medicine” and implications. As part of analyzing our interview data, we engaged in memo-writing [43]. Writing memos helped us to develop our conceptual framework and ensure that our emerging insights were truly grounded in the data. For each respondent, we wrote a memo that touched upon the themes we identified supported by specific data in the transcript that supported our interpretation. This process demonstrated that the themes we identified were supported by all 15 interviews and representative of the entire sample, as opposed to being heavily influenced by a few select interviews.

Our findings indicate that practice change disturbs the status quo equilibrium that establishing a new equilibrium that incorporates the change may be a struggle and that part of this struggle involves both unlearning the old evidence itself and tensions between evidence and context. Moreover, this process occurs in the context of the whole system in which physicians practice. Our findings are consistent with some of the models of change that refer to unlearning such as that of Kurt Lewin [29, 49].

Lewin’s change theory provides a linear three-stage model of change, starting with unfreezing of the status quo, change or transition, and finally refreezing. Manchester et al. applied Lewin’s theory to the implementation of evidence-based practice in clinical settings [29]. Our findings support the implication that unlearning, equivalent to the “unfreezing” stage of Lewin’s change theory, is a necessary prerequisite for successful practice change and ultimate “refreezing.” We noted that physicians who develop personal contingency theories and revert back to an outmoded practice to accommodate patient autonomy never fully unlearn or unfreeze the practice, hindering their ability to successfully change their practice. This may be consistent with the dual process model of decision-making, also extended to learning [50–52]. Some physicians may be stuck in a particular heuristic (“fast thinking”) and must change to a more rational approach (“slow thinking”) before the old heuristic is replaced by a new one. Our findings also provide further detail regarding the intricacies of the unfreezing and “change” stages of the change theory, such as the factors that inhibit and support successful implementation of evidence-based practice. For instance, the “Finding 2” section sheds light upon what physicians might consider as evidence-based medicine that initiates practice unfreezing in the clinical setting, ranging from RCTs to local evidence collected by physician observations of patients in the clinic.

An alternative model was developed by Becker in the context of business management [23]. When applied to health care, a modified Becker model of unlearning describes the process of physician unlearning as a product of the tension between prior knowledge—influenced by the prior mental model, characteristics of the clinician, physician’s beliefs about the consequences to patients, social influences, and organization context—and subsequent knowledge [53]. The “prior knowledge” and “tension” phases of the physician unlearning process are in line with the findings from our study. The different struggles physicians face when encountering practice change are manifestations of the tension in the modified Becker model between the previous and subsequent knowledge. Thus, our findings also complement the various factors identified by the model that hinder or help the unlearning process. For instance, in support of the “physician characteristics factor” in the model, our findings elaborated upon the various physician characteristics that might affect the unlearning process, such as personal biases, clinical experiences, and openness to change.

In addition to supporting the modified Becker model, our findings provide further detail regarding factors that both enable and inhibit physician unlearning. For example, not only are physicians’ beliefs about the consequences to patients an important aspect of unlearning, but also the actual resistance from patients and respect of patient autonomy are significant barriers of unlearning. Additionally, physicians’ opinions regarding the actual initiating agent of unlearning, such as the perceived strength and quality of the RCTs or guidelines, influence whether physicians follow through with the unlearning and are not included in the model. As a result, our findings support for the initial phases of unlearning described by the modified Becker model and elaborate upon these phases.

While our findings do support aspects of the modified Becker model and Lewin’s change theory, they are inconsistent with the linearity implied by the two models; the struggle to change is interactive. Thus, “static” models do not capture the dynamic nature of change in the wild; clinical practice is a process rather than a discrete event. When introduced with practice change, physicians often go through a process of trial-and-error with the new practice, developing their own method of implementation and building comfort with the new practice. Very rarely, however, do physicians completely unlearn the outmoded practice. Most physicians develop personalized contingency theories based on particular patient characteristics, as previously demonstrated by Respondent 13’s implementation of statin guidelines and Respondent 14’s respect of patient autonomy when recommending mammogram screening, that guide when they use the new practice and when they revert back to using the outmoded practice. As a result, the unidirectional linear process of change depicted by the Lewin and Becker models does not accurately depict the back-and-forth involved in physician unlearning. The ultimate step of refreezing or development of the subsequent mental model is not fixed; rather, physicians develop a dynamic equilibrium centered around the new practice as opposed to the outmoded practice. As a result, perceived “success” in unlearning may be redefined in this new context, determined by the new equilibrium that is established instead of the degree of relinquishment of the outmoded practice.

Additionally, both the Lewin and Becker empirical models allude to change as an isolated step in the process of transitioning from the prior mental model to the subsequent mental model. These models explain the process of change in the context of stability that the initiating agent disturbs the status quo and the incorporation of the new evidence-based practice into the status quo restores stability or “refreezes” the equilibrium. When physicians change medical practice, however, they find themselves practicing in a new context and new environment, with a new set of expectations and new understandings of what the best practice is. This new equilibrium is plagued with instability and uncertainty as physicians struggle to learn a new practice that they must adapt to their patients with little guidance or direction. The adaption of a new practice, far from restoring stability, introduces further change, as physicians are continually sensing, self-reflecting, discussing with colleagues, critically evaluating, and modifying the new practice. As a result, our findings support a model of medical practice that is a continuous spectrum of change. Each instance of unlearning an outmoded practice and learning a new practice is merely a punctuation in this spectrum of change.

In summary, our findings suggest that current models of physician unlearning be modified to incorporate the constantly changing nature of clinical practice. Such models should demonstrate an episode of unlearning and learning as a snapshot of the constant change that takes place every time physicians carry out a particular practice. Additionally, while unlearning is a necessary prerequisite for learning, rarely does there occur a single unidirectional linear progression from unlearning to learning of a new practice as depicted by both the Lewin and Becker models. Rather, these models should echo a multi-directional process of change, influenced by various structured and unstructured sources of data. This shift in paradigm regarding physician unlearning supported by our findings is also critical to the development of future unlearning models and application of such models to de-implementation strategies.

Our two findings include (1) practice change disturbs the status quo equilibrium. Establishing a new equilibrium that incorporates the change may be a struggle; and (2) part of the struggle to establish a new equilibrium incorporating a practice change involves both the evidence itself and tensions between evidence and context. Our findings provide evidence-based support for many of the empirical unlearning models that have been adapted to healthcare. Our findings differ from these empirical models in that they refute the static and unidirectional nature of change that previous models imply. Rather, our findings suggest that clinical practice is in a constant flux of change; each instance of unlearning and learning is merely a punctuation mark in this spectrum of change. We suggest that physician unlearning models be modified to reflect the constantly changing nature of clinical practice and to demonstrate that change is a multi-directional process.