De-implementation of low-value care: contrasting theories, models, and frameworks to generate methodological guidance

Abstract

Background

Low-value care remains pervasive across healthcare systems and consumes scarce resources while exposing patients to avoidable harms. De-implementation, the purposeful process of reducing, restricting, replacing, or discontinuing low-value practices, has gained momentum, yet the field still lacks clear methodological guidance grounded in theory. De-implementation is often treated as “reverse implementation”, despite accumulating evidence that stopping practices can activate distinct mechanisms and constraints.

Main body

In this paper, we synthesize and contrast theories, process models, and frameworks relevant to de-implementation of low-value care to generate actionable methodological guidance. Drawing on a theory-informed narrative review and constant-comparative synthesis, we identify where de-implementation converges with implementation (e.g., staged processes; multilevel determinants; use of established determinant, strategy, and outcome frameworks) and where it diverges in ways that matter for design and evaluation. Across sources, three recurring lenses structure these divergences: (i) the psychology of stopping (habit disruption, loss aversion, cognitive biases, professional identity threats), (ii) multi-level constraints and politics (incentives, regulation, professional norms, stakeholder interests), and (iii) the nature of the low-value practice and endpoint (reduction vs. restriction vs. elimination; replacement vs. disenchantment discontinuance). We translate these contrasts into ten streamlined methodological recommendations that specify what investigators should state and report in practice, including configuration (stand-alone vs. embedded/paired), low-value classification, explicit determinant-to-strategy-to-mechanism logic, inclusion of patient experience and unintended consequences, and dual-trajectory evaluation when substitution is involved.

Conclusion

De-implementation is not simply implementation in reverse. Methodological rigor in de-implementation research requires explicitly specifying configuration and endpoint, aligning strategies with stopping-specific mechanisms and multilevel constraints, and evaluating beyond utilization to include mechanisms, patient experience, equity-relevant impacts, and unintended consequences. This paper provides a practical, theory-grounded set of recommendations to strengthen the design, evaluation, and reporting of future de-implementation studies.

Contributions to the literature

• By mobilizing established theories, process models, and frameworks, this paper shows how they can guide de-implementation research and strengthens the argument that stopping differs from starting.

• De-implementation diverges from implementation across three lenses: stopping triggers distinct psychological barriers (habit loops, loss aversion, professional identity), operates under multilevel constraints and politics (incentives, regulations, norms), and depends on the nature of the low-value practice and the intended endpoint (reduction vs. restriction vs. elimination; replacement vs. disenchantment discontinuance).

• From this debate paper, ten practical theory-informed recommendations are derived to guide study design, determinants analysis, strategy selection, and outcome evaluation in future de-implementation studies, directly addressing a methodological gap.

Background

In healthcare systems globally, at least 30% of clinical interventions provided as part of routine care are considered low-value because they have no proven therapeutic benefit or may even lead to significant harm for patients [1]. These low-value interventions use limit material and human resources at a time when healthcare systems are under strain [2]. To improve the quality and safety of care, several initiatives have been developed in recent years to reduce or even eliminate low-value care across different clinical domains [3, 4]. In support of these objectives, implementation science has increasingly turned to de-implementation, defined as a purposeful, systematic process of reducing, restricting, replacing, or discontinuing low-value, ineffective, or potentially harmful clinical interventions [2, 5].

A major initiative is the Choosing Wisely campaign, launched in 2012, which encourages physicians to discuss the necessity of certain treatments or tests with their patients [6]. In the following years, numerous studies identified areas of low-value care [7, 8]. However, their focus was mainly on stating which care should be reduced or eliminated, without addressing how to achieve it. Later, the literature began to focus more on the de-implementation process itself, by conceptualizing it and highlighting that a planned and structured sequence of steps is required to carry it out [9–11].

Although initiatives such as Choosing Wisely have identified hundreds of recommendations targeting low-value care, many such practices remain embedded in clinical guidelines and routine care [12]. This highlights the extent to which these practices are sustained through routinization and entrenchment [13]. This also underscores the need for a clearer understanding of what de-implementation entails, and how it differs from implementing evidence-based interventions, to enhance healthcare and protect patient safety. Many studies suggest that de-implementation is far more complex than simply asking clinicians to stop a practice, and that it cannot be understood as the mere reverse of implementation [2, 14–16]. Although the broad steps involved in de-implementation are similar to those in an implementation process, the literature, primarily reviews mapping the theories, models, and frameworks used in de-implementation studies, highlights key differences between the two, particularly in the determinants and strategies needed [2, 3, 10, 17]. The persistent problem of low-value care calls for moving beyond these initial insights in the literature and articulating the distinct theoretical foundations that can guide the design, evaluation, and reporting of de-implementation initiatives. In this paper, we map where theories, models and frameworks converge or diverge, and we translate these distinctions into methodological guidance.

Aim and objectives

Our overall aim is to advance conceptual clarity by contrasting implementation and de-implementation at the level of theoretical foundations, in order to generate actionable methodological guidance. Our specific objectives were to:

1. Identify and synthesise key theories, models, and frameworks that underpin de-implementation and compare them with those traditionally used in implementation science by highlighting points of convergence (shared assumptions) and divergence (unique constructs or mechanisms).

2. Translate these theoretical insights into methodological recommendations for future de-implementation studies.

Methods

This debate paper is grounded in a theory-informed unsystematic narrative review designed to generate methodological guidance. A search strategy was developed using database-specific descriptors and keywords targeting three core concepts: (1) low-value care, (2) de-implementation, and (3) implementation science. This search strategy was applied to the following databases: CINAHL Complete (EBSCOhost), MEDLINE (Ovid), and Web of Science. In addition, a grey literature search was conducted using combinations of keywords such as inappropriate, obsolete, low-value care, de-implement, de-adopt, and disinvest within ProQuest and Google Scholar. Articles were included if they made a conceptual contribution to understanding how de-implementation is framed as a process and how determinants, strategies, and outcomes are theorized. No limits were applied during the selection process with respect to study design, publication date, or setting. We purposively sampled foundational conceptual papers, as well as theoretical and empirical studies and influential reviews [3, 9, 11, 15, 16, 18], that explicitly discussed de-implementation (or closely related constructs such as de-adoption, disinvestment, exnovation, and discontinuance) in the context of low-value care. To conclude the selection process and to ensure conceptual coverage, additional sources were identified iteratively through forward and backward citation tracking and targeted searching around key terms and seminal authors based on team discussions. We extracted how each theory, process model, or framework: (i) defines the object of change (what is being reduced/removed and why it is low-value), (ii) positions de-implementation relative to implementation (stand-alone vs. embedded), and (iii) specifies assumptions about mechanisms, levels (individual, organizational, policy), and evaluation. We then conducted an interpretive, constant-comparative synthesis across sources, contrasting de-implementation and implementation assumptions to identify points of convergence and divergence and translating these into methodological recommendations.

Roadmap and framing lenses

The paper advances a single argument: de-implementation is not simply “reverse implementation”; it mobilizes overlapping stages and tools, but often activates distinct mechanisms and constraints that matter for study design, strategy selection, and evaluation. We proceed in three steps. First, we clarify how de-implementation is positioned relative to implementation through process models (embedded vs. stand-alone). Second, we synthesize theories that explain why stopping differs from starting (or when it may not). Our synthesis yielded three recurring lenses across theories, models, and frameworks: (i) the psychology of stopping: reducing or withdrawing a familiar practice often triggers loss-framed judgments [19], cognitive biases [20], professional identity threats [5, 20], and habit disruption [5, 21], shaping acceptability, feasibility, and patient–provider dynamics; (ii) multi-level constraints and politics: de-implementation is often conditioned earlier and more strongly than in implementation by outer setting constraints including political agenda embedded in payer incentives, professional standards, media attention, or public expectations that “more care is better” [14, 15, 22, 23] and inner-setting priorities, as low-value care is structurally reinforced through routine workflows and guidelines, making de-implementation contingent on organizational unlearning [13]; and (iii) the nature of the low-value practice: how “low-value” is established (strength of evidence, inefficiency, patient preferences) [5, 24] and the intended endpoint (reduction, restriction, elimination) [5, 25] shape which strategies are plausible and what outcomes must be tracked, including unintended consequences. These three main ideas underpin the subsequent sections. Third, we examine frameworks that structure determinants, strategies, and outcomes, culminating in methodological recommendations.

Process models: locating de-implementation relative to implementation

Many of the de-implementation process models identified, which describe the key steps involved in de-implementation, are theoretically grounded in well-established implementation process models [2, 9]. It reflects the idea that both change processes involve broadly similar stages (e.g., identifying the target practice, assessing determinants, selecting and tailoring strategies, evaluating outcomes, and sustaining change). However, when authors explicitly consider the relationship between de-implementation and implementation, differences emerge in the logic of the process and in the endpoint being pursued.

A recurring point of divergence in the literature is whether de-implementation is conceptualized as a stand-alone change process or as embedded within implementation (prior to, alongside, or as a consequence of adopting an alternative). This distinction is not semantic; it shapes what constitutes success, which mechanisms are expected to operate, and what should be measured. In particular, stand-alone de-implementation can often be framed around a single behavioral endpoint (e.g., reduction, restriction, or elimination of a low-value practice), whereas embedded de-implementation frequently implies a dual target (i.e., decreasing low-value care while increasing uptake of an alternative) and therefore requires evaluation of both trajectories.

Nilsen, Ingvarsson [10]’s scoping review identified a de-adoption process articulated by Niven, Mrklas [9], which adapts the Knowledge-to-Action model by Straus, Tetroe [26] to de-implementation. This model highlights a series of steps to be undertaken: (1) identify and prioritize low-value clinical practices, (2) select, tailor, and implement de-adoption intervention, (3) evaluate de-adoption process and outcomes, (4) sustain de-adoption, (5) assess current use of low-value practice, (6) adapt knowledge to local context, and (7) assess barriers and facilitators to de-adoption. The term de-adoption used by these authors is considered a synonym for de-implementation. Similarly, Grimshaw, Patey [2] proposed a process model in the context of the Choosing Wisely campaign, based on the four steps initially proposed by French, Green [27]: (1) identifying potential areas of low-value health care, (2) identifying local priorities for the implementation of Choosing Wisely recommendations, (3) identifying barriers and potential interventions to implement Choosing Wisely recommendations, (4) evaluating Choosing Wisely implementation programs, and (5) spreading effective Choosing Wisely implementation programs. Notably, these models describe de-implementation steps without requiring an explicit paired implementation effort; the primary focus is the reduction of low-value care and the conditions needed to initiate and sustain it.

By contrast, other authors argue that de-implementation often occurs before or alongside implementation, particularly when discontinuing low-value care is necessary to create space (time, resources, workflow capacity) for higher-value alternatives. Davidson, Ye [28] propose a cyclical model that begins with identifying the practice to de-implement and documenting current patterns, then investigating reinforcements and beliefs that maintain the practice, selecting “extinction” methods matched to those maintaining factors, conducting a de-implementation experiment, and evaluating consequences, including time and resources saved, before proposing the next practice to implement. In this view, de-implementation is not merely parallel to implementation but is frequently coupled to it, consistent with Rogers [25]’ discussion of discontinuance as a phenomenon embedded in the life cycle of innovations.

Overall, some authors conceptualize de-implementation as interdependent with implementation [25, 28], while others present it as a theoretically independent process [2, 29]. These perspectives are not necessarily in opposition; they can be understood as complementary lenses. However, because they imply different target behaviors (single vs. dual), different transition challenges (e.g., sequencing and workflow redesign when substituting practices), and different evaluation requirements (including substitution and unintended consequences), de-implementation studies should explicitly specify whether the initiative is framed as stand-alone de-implementation or as embedded within an implementation effort, and align determinants, strategies, and outcomes accordingly.

Theories: unpacking mechanisms that explain differences between stopping and starting practices

Following Nilsen [29]’s definitions, theories are “a set of analytical principles or statements designed to structure our observation, understanding and explanation of the world”. Here, we organize them by function: the specific explanatory job they perform in de-implementation studies. This matters because de-implementation typically requires layering multiple explanations: why reducing or stopping a practice may operate through different mechanisms than starting one; why withdrawal is often experienced as loss (by clinicians, patients, and organizations), amplifying resistance; and how organizational and policy environments enable or constrain discontinuation, often earlier and more forcefully than in many implementation efforts [14, 15, 22, 23].

First, some theories are most useful for explaining how behaviour is reduced (and why “more” and “less” can require different techniques). Operant Learning Theory and the Deterrence Theory emphasize stimulus–response learning and the role of reinforcement and sanctions [15]. Their central implication for de-implementation is not that “punishment is the answer,” but that behaviour reduction has its own mechanism set, often involving removal of cues/rewards, increased friction, accountability requirements, or consequences that make continued low-value care less likely. These theories help justify why certain strategy families (e.g., accountability tools, restrictions, hard stops, audit-and-feedback with consequences) may be more potent for discontinuation than for adoption [15].

A closely related functional strand explains habit disruption and automaticity, which is often central when low-value care is routinized. The Model of Habits suggested by Wood and Neal [21] highlights that once behaviours become habitual, they can be triggered by contextual cues with limited deliberate intention. This is particularly relevant to de-implementation because discontinuation frequently requires breaking cue–response loops rather than persuading individuals to form a new intention. As suggested by Norton and Chambers [5], de-implementation strategies can target the infrastructure of habit: remove prompts, redesign workflows, change defaults, and weaken the reinforcers that sustain the practice. This also clarifies why a determinant analysis should pay special attention to “where the habit lives” (electronic health record, team routines, patient scripts, billing, time pressure).

A third functional grouping explains judgment and decision dynamics under loss and uncertainty, helping to account for the recurrent observation that stopping can feel harder than starting. Prospect Theory and the notion of loss aversion highlight that discontinuing a familiar practice may be experienced as a loss of safety, autonomy, revenue, or “doing something,” even when evidence indicates the practice is low-value [19]. This loss framing can apply to clinicians (e.g., fear of missing diagnoses, medico-legal anxiety), patients (e.g., fear of being undertreated, “more care is better”), and organizations (e.g., perceived risk, reputational costs) [5, 30]. As outlined by Roman and Asch [20], related cognitive biases (e.g., confirmation bias, optimism bias, affect heuristics) help explain why contradictory evidence, particularly “medical reversals,” can be difficult to translate into practice change. Functionally, these theories from behavioral economics sharpen what to measure (e.g., perceived losses, trust, anxiety, legitimacy) and what kinds of strategies may be required (e.g., reframing, credible messengers, patient-facing communication supports, norm-based messaging), especially when de-implementation threatens professional identity or patient expectations [5].

Not all theories, however, imply that de-implementation and implementation are fundamentally different. Intention-based theories such as the Theory of Planned Behavior (Ajzen; applied by Powell, Bloomfield [31] to overuse) can be functionally useful because they explain how beliefs, norms, and perceived control shape intentions that drive both overuse and underuse. In practice, this means some strategy families (education, persuasion, social influence, skill-building) may be relevant across implementation and de-implementation, particularly when the low-value practice is not deeply habitual and when actors experience meaningful choice. At the same time, a functional limitation of intention-centred theories is that they may under-explain discontinuation when behaviour is primarily cue-driven, structurally incentivized, or politically protected, precisely the conditions often observed in low-value care.

A fifth functional grouping accounts for organizational and system conditions that make stopping possible. Lewin’s unfreeze–move–freeze model as cited by Crosby [32] and the Organizational Unlearning Theory developed by Hedberg and later applied by Nystrom and Starbuck [33] are useful here because they foreground the preparatory work required to destabilize the status quo. In de-implementation, “unfreezing” often requires more than presenting evidence: it may require legitimacy building, psychological safety, alignment with leadership priorities, and removal of organizational reinforcers that keep low-value care in place. Organizational unlearning emphasizes that discontinuation can require actively letting go of prior assumptions and routines to create cognitive and operational space for change, especially during crises or when entrenched practices are tied to identity, training, or organizational memory. Functionally, these theories point de-implementation researchers toward measuring readiness to let go, perceived legitimacy, and whether organizational conditions (incentives, workflows, norms, accountability) were changed, not just whether individual attitudes improved.

A sixth functional grouping captures system and policy mechanisms that shape whether stopping is feasible and legitimate, beyond individual motivation or organizational readiness. Bauer and Knill [22] policy dismantling framework foregrounds the political costs of removal. Discontinuing an established policy or publicly visible practice can trigger blame, backlash, and mobilization of stakeholders who benefit from the status quo. In de-implementation, these dynamics often show up as early constraints, through regulation, payer incentives, professional standards, media attention, or public expectations that “more care is better.” Functionally, this perspective directs de-implementation researchers to specify the decision arena (who can enable or block stopping), anticipate issue visibility and reputational risk, and attend to system-level determinants such as stakeholder coalitions, policy signals, and public trust, rather than assuming that evidence and local behavior change strategies will be sufficient [5].

The Diffusion of Innovations Theory contributes a complementary system-oriented mechanism by clarifying pathways of discontinuation. Rogers distinguishes discontinuance that occurs through replacement (stopping because an alternative is adopted) from discontinuance driven by disenchantment (stopping because performance, legitimacy, or fit erodes). This distinction matters for both design and evaluation. As previously discussed, replacement discontinuance implies a dual behavioral target, decreasing low-value care while increasing uptake of a substitute. Disenchantment discontinuance, by contrast, might point to sustainability-related mechanisms (e.g., perceived harms, disappointment, or mismatch with context) that may lead to stopping. Interestingly, van Bodegom-Vos, Davidoff [34] highlight that the early adopters involved in change are not necessarily the same when it comes to discontinuing a practice versus implementing an innovation. For example, early adopters in a de-implementation context tend to be cautious and concerned about the potential risks associated with certain practices. These individuals would more likely be considered laggards in an implementation process, as their skepticism may prevent them from quickly adopting new practices. There are thus key similarities between laggards in the context of innovation diffusion and high discontinuers in the context of discontinuance, both often characterized by lower levels of formal education, lower socioeconomic status, and less frequent contact with change agents [25].

Taken together, as synthesized in Table 1, these theories show that de-implementation is rarely just “implementation in reverse.” Some theories predict mechanistic asymmetries between starting and stopping (e.g., reinforcement/sanction pathways, habit disruption, loss-framed judgment, policy costs), while others help explain shared determinants that can matter in both directions (e.g., beliefs, norms, perceived control). The practical implication is methodological: de-implementation studies should state which mechanism(s) they are targeting, align strategy families accordingly (e.g., cue removal vs. persuasion; accountability vs. education; policy levers vs. local workflow redesign), and measure intermediate outcomes that correspond to the hypothesized pathway (e.g., habit cues and reinforcers, perceived losses and legitimacy, readiness to let go, political feasibility), not only changes in utilization rates.

Table 1.

Theories by function: mechanisms that shape de-implementation (stopping) vs. implementation (starting)

Functional grouping (what the theory explains)	Theory (key source)	Core mechanism	Why it matters for de-implementation (vs. implementation)	Typical strategy implications	What to measure (intermediate outcomes)
1) Behaviour reduction mechanisms (how “less” happens)	Operant Learning Theory (Skinner; Thorndike “law of effect”)	Behaviour changes through reinforcement or sanctions; contingencies shape repetition.	Stopping often requires different levers than starting: removing rewards/cues, increasing friction, or making continuation costly.	Accountability tools, restrictions or hard stops, consequences linked to audit & feedback; remove reinforcers.	Reinforcers present or removed; friction or effort; perceived consequences; continued use after contingencies change.
	Deterrence Theory (via Patey, Hurt [15])	Behaviour decreases when perceived certainty or severity of sanctions rises.	Useful where low-value care persists despite knowledge, suggests changing the “cost” of continuation.	Policies with enforcement; justification requirements; monitoring and consequences.	Perceived sanction risk; compliance; avoidance; “work-arounds”; acceptability and ethics perceptions.
2) Habit disruption and automaticity (how routinized practices persist)	Habit Theory / Model of Habits (Wood and Neal [21])	Cue-driven behaviour can bypass intention; contexts trigger automatic responses.	Low-value care is often embedded in workflows (defaults, order sets, routines). Persuasion alone may fail if cues remain.	Change defaults and order sets; redesign workflows; remove prompts; alter team routines.	Cue exposure; default use; habit strength and automaticity; workflow triggers; substitution patterns.
3) Loss-framed judgment and cognitive biases (why stopping “feels harder”)	Prospect Theory / Loss aversion (Kahneman and Tversky [19])	Losses loom larger than gains; risk preferences shift under loss framing.	De-implementation can feel like loss (safety, autonomy, revenue, “doing something”), driving stronger resistance than adoption.	Reframing (safety and benefit), credible messengers, shared decision supports, risk communication.	Perceived losses and gains; anxiety; trust; perceived safety; legitimacy of change.
	Cognitive biases (confirmation, optimism, affect heuristic) (Roman and Asch [20])	People overweight confirming and positive info; first impressions persist.	Helps explain why “medical reversals” and contradictory evidence don’t translate into stopping.	Narrative and social norm messaging; peer comparison; patient scripts; addressing medico-legal fears.	Belief updating; perceived evidence credibility; perceived legal and reputational risk; patient expectations.
4) Intention-based choice mechanisms (when stopping resembles starting)	Theory of Planned Behavior (Ajzen; Powell, Bloomfield [31])	Intentions shaped by attitudes, norms, perceived control.	When low-value care is not deeply habitual and clinicians have real choice, shared determinants may operate for both underuse and overuse.	Education and persuasion; social influence; skills/support; local champions.	Attitudes and norms; perceived control; intention; perceived appropriateness; deliberative decision frequency.
5) Organizational readiness and “letting go” (conditions that make stopping possible)	Unfreeze–Move–Freeze (Lewin; Crosby [32])	Change requires destabilizing status quo, then moving and stabilizing new norm.	In de-implementation, “unfreezing” may require legitimacy, psychological safety, leadership alignment, and removal of reinforcers, often more than evidence.	Leadership signalling; role modelling; workflow and incentive redesign; protected time; supportive climate.	Readiness to let go; psychological safety; leadership commitment; alignment with priorities; norms shift.
	Organizational Unlearning (Hedberg; Nystrom and Starbuck [33])	Letting go of old assumptions and routines creates space for new ones.	Stopping can require active unlearning (identity, training, memory). Particularly relevant in crises or entrenched practices.	Sensemaking; debriefs; learning forums; revising protocols and training; removing legacy routines.	Perceived legitimacy; unlearning indicators; routinization of new standard; “work-arounds”.
6) System, political feasibility and discontinuation pathways (who enables/blocks stopping)	Policy Dismantling (Bauer and Knill [22])	Dismantling triggers political costs, blame, backlash; visibility matters.	De-implementation often constrained by regulation, payers, professional bodies, media and public expectations, constraints may appear earlier than in implementation.	Stakeholder coalition building; policy signals; managing visibility; public communication strategies.	Decision arena mapping; issue visibility; stakeholder positions and power; public trust; policy alignment.
	Diffusion of Innovations: discontinuance (Kimberly & Evanisko; Rogers [25]; van Bodegom-Vos, Davidoff [34])	Discontinuance can be replacement (swap) or disenchantment (erosion of fit or legitimacy).	Clarifies whether de-implementation is a single target (stop) or dual target (stop + adopt substitute). Also suggests “de-adopters” may differ from adopters.	If replacement: coordinate paired implementation; If disenchantment: focus on fit/harms/legitimacy and sustainability mechanisms.	Stop and substitute trajectories; substitution effects; who discontinues vs. who adopts; perceived fit/legitimacy over time.

Frameworks: structuring determinants, strategies, and outcomes for de-implementation

Frameworks are more context- and phenomenon-specific than theories and process models. Rather than explaining why de-implementation succeeds or fails, they help specify what to assess and organize, the determinants that shape stopping, the strategy options available, and the outcomes that should be tracked.

Determinant frameworks

Qualitative syntheses and interview studies show that determinants of de-implementation are multifactorial and span individual, organizational, and policy/system levels [35, 36]. This mirrors implementation science’s long-standing multilevel view and supports the continued use of established determinant frameworks, such as the Theoretical Domains Framework (TDF) and the Tailored Implementation for Chronic Diseases (TICD) framework [10, 11] to explore factors influencing the de-implementation process. At the same time, de-implementation occurs within a broader contextual landscape encompassing outer-setting constraints (political agendas, financial stakeholders, media/public expectations) and inner-setting priorities (departmental incentives, opportunity costs, organizational legitimacy), which may shape change earlier and more strongly than in many implementation efforts [14, 23].

De-implementation also benefits from determinant framing that characterizes the low-value practice itself. Two complementary approaches are useful here. First, Norton and Chambers [5] classification based on strength of evidence, ineffective, contradicted/medical reversal, mixed, and untested, helps specify what “low-value” means from an evidence standpoint and anticipates the kinds of resistance that may emerge when evidence is ambiguous or evolving. Second, Verkerk, Huisman-de Waal [37] propose organizing low-value care across medical, societal, and patient perspectives: an intervention may be low-value because it is ineffective (medical), inefficient through inappropriate intensity/duration or delivery (societal), or misaligned with patient goals and preferences (patient). Together, these classifications foreground three de-implementation-relevant determinants, evidence strength, inefficiency, and patient preference-fit, that help define the de-implementation target. Finally, some determinants apply strongly in both implementation and de-implementation. For instance, complexity, whether the target is a single discrete practice versus a bundle of interdependent routines [5], has clear implications for strategy intensity and resourcing, and is explicitly captured within widely used determinant frameworks such as the Consolidated Framework for Implementation Research [38].

Strategy frameworks

Strategy frameworks and taxonomies such as the Behavior Change Wheel [39] and Implementation Mapping [40], provide a shared vocabulary for selecting and reporting “what was done” to drive practice change. In this paper we focus on the Expert Recommendations for Implementing Change (ERIC) taxonomy (73 strategies in nine clusters) developed by Powell, Waltz [41] because it remains the dominant reference point and is commonly used to label strategies in de-implementation studies, consistent with evidence that most de-implementation strategies map onto ERIC categories, as noted by Ingvarsson, Hasson [16]. According to their findings, 87% of the de-implementation strategies identified correspond to strategies already included in the ERIC taxonomy. However, four strategies were identified that are not included in the ERIC taxonomy and could be specific to de-implementation: (1) accountability tools, such as requiring clinicians to justify each decision to use a low-value intervention in the medical record, (2) black box warnings on medication packaging to alert users and clinicians to associated risks, (3) policies and regulations including directives instructing clinicians to avoid using low-value care; and (4) communication tools, such as a “written script describing a process for communicating with patients about why they are not receiving a low-value practice”. However, the fact that most strategy labels overlap does not imply that strategy mechanisms or ethical constraints are the same in stopping versus starting. As highlighted by the theoretical mechanisms presented, discontinuation often requires cue removal, friction, accountability, and legitimacy work that may be less central in adoption-focused initiatives [5, 16, 20, 21].

Two refinements help align strategy frameworks with de-implementation. First, according to Norton and Chambers [5], strategy selection should be explicitly linked to the de-implementation aim, commonly framed as reduction, restriction, or elimination of low-value care. These aims imply different degrees of structural change (e.g., soft nudges versus hard stops) and different risk profiles (e.g., restriction may be more acceptable than elimination in uncertain evidence categories). Second, substitution is best treated as a configuration rather than a purely “de-implementation strategy”: it creates a dual behavioural target (decrease the low-value practice and increase the alternative), and therefore requires paired measurement of both trajectories, attention to sequencing/workflow redesign, and assessment of substitution effects and unintended consequences [5, 25]. This framing keeps substitution consistent with how de-implementation is defined (reduce, replace, discontinue) while preserving its distinctive methodological implications in terms of dual targets rather than a single endpoint.

Empirically, reviews such as the one conducted by Kien, Daxenbichler [3] suggest that de-implementation studies most often draw on familiar strategy clusters (train/educate, clinician support, evaluative/iterative approaches, infrastructure/workflow change), with multifaceted strategies tending to show stronger effects than single-component approaches. Fontaine, Vinette [42] reached a similar conclusion in the context of implementation. Still, a recurring gap is that relatively few de-implementation initiatives report strategy development that is clearly and transparently tailored to a prior determinant analysis, despite repeated calls for determinant-informed tailoring in de-implementation reviews [3, 5, 16, 43].

Beyond the development of determinant-informed strategies, additional considerations include the alignment of strategies with theory-informed mechanisms of change. These mechanisms represent the active ingredients through which strategies operate, explaining how a given strategy mitigates barriers or leverages facilitators [27, 40]. As highlighted in the literature, de-implementation strategies may target mechanisms that differ from those emphasized in implementation, such as habit disruption, accountability, friction, loss-framed judgment, and political feasibility. Explicitly articulating and reporting the links between strategies and their underlying theory-informed mechanisms of change is essential to advance understanding of how de-implementation strategies can be designed, operationalized, and differentiated from implementation strategies.

Outcome frameworks

Outcome frameworks clarify which outcomes to report and distinguish between (i) process success (implementation/de-implementation outcomes) and (ii) practice impact (implementation/de-implementation effects). Proctor, Silmere [44]’s outcomes (acceptability, appropriateness, feasibility, fidelity, penetration, sustainability, etc.) remain highly relevant, and Prusaczyk, Swindle [45]’ adaptation provides useful conceptual guidance for applying these outcomes in a de-implementation context, where the intended direction is reduction rather than increase.

However, de-implementation outcome frameworks often need to extend beyond utilization. In many studies, the primary endpoint is simply reduced use of low-value care, yet patient experience and broader system consequences can be central, and sometimes decisive, for feasibility and legitimacy [9]. Verkerk and Riganti [46] argue that de-implementation evaluations should more routinely capture: (i) patient-level outcomes (acceptability, anxiety/fear, perceived abandonment, trust, impacts on the therapeutic relationship), (ii) unintended consequences (e.g., substitution to other low-value care, inequitable impacts, delayed diagnosis concerns), and (iii) system-facing outcomes that reflect decision-maker priorities without reducing the justification for de-implementation to cost-savings alone. Where substitution is present, outcome frameworks should explicitly require measurement of both the de-implemented practice and the adopted alternative, as well as net clinical and experiential impacts.

Beyond evaluating processes and practice-level impacts, de-implementation outcome frameworks should incorporate intermediate outcomes that assess whether strategy-driven mechanisms of change effectively influence targeted determinants. The literature remains limited in explicating how theory-driven mechanisms specific to de-implementation strategies, such as habit disruption, accountability, and friction, function and translate into outcomes.

In sum, frameworks in de-implementation research should be selected and reported in ways that preserve implementation science’s strengths (multilevel determinant assessment; standardized strategy labels; established outcome constructs) while making de-implementation-specific needs explicit: the characterization of “low-value,” early attention to outer-setting constraints, alignment of strategies to the de-implementation aim (reduction/restriction/elimination), development of strategies based on theory-driven mechanisms of change, and outcome sets that include patient experience, substitution effects, and unintended consequences. Table 2 aims to synthesize these nuances by mapping the similarities and differences between de-implementation and implementation, focusing on their processes, determinants, strategies, and outcomes.

Table 2.

Frameworks by function in de-implementation: what they structure and what to adapt

Framework function	Exemplar framework(s)	What it structures	De-implementation-specific additions/decisions to state	Minimum reporting elements (practical)
Determinant frameworks	TDF; TICD; CFIR (complexity) [10, 11, 38]	Multilevel barriers/facilitators (individual, inner setting, outer setting)	Explicitly assess outer-setting forces early (incentives, regulation, public expectations, media, stakeholder interests) [5, 14, 23].	Levels assessed; key determinants prioritized; rationale for “why now”; determinant-to-strategy linkage.
Low-value care classifications (upstream framing)	Evidence-strength categories (ineffective, contradicted, mixed, untested) [5]; medical, societal, patient perspectives [37]	Defines what “low-value” means and why the target is low-value	State whether the target is low-value due to evidence strength (note uncertainty when evidence is mixed or untested), inefficiency, or preference misfit [5, 37].	Classification used; evidence status; whose perspective(s) define low value; implications for risk/acceptability.
Strategy frameworks / taxonomies	ERIC taxonomy [41]; de-implementation strategy mapping; strategy cluster patterns [3, 16]	Naming and categorizing strategies; comparability across studies	State the de-implementation aim (reduction, restriction, elimination); if substitution, treat as dual target (stop + start) and plan evaluation accordingly [5, 25, 43].	Strategy labels; components/dose/actors/timing; tailoring logic; aim type; dual-target plan if substitution.
Outcome frameworks	Proctor outcomes [44]; de-implementation outcomes distinctions [45]	What “success” includes beyond utilization	Extend beyond utilization to include patient experience, trust or legitimacy, unintended consequences, substitution effects (when relevant) [9, 46].	Outcomes vs. effects clearly separated; patient-level outcomes included; unintended consequences plan; dual trajectory measurement if substitution.

Methodological recommendations

To address Objective 2, we translate the above contrasts into methodological recommendations intended to improve the design, evaluation, and reporting of de-implementation initiatives (see Table 3). These recommendations are organized around the core building blocks of de-implementation research: process specification, determinant assessment, strategy selection and mechanism alignment, and outcome measurement.

Table 3.

Methodological recommendations for future de-implementation studies

Domain	Recommendations	What to report (minimum expectations)
1. Framing and configuration	Define the low-value practice and why it is low-value, and specify whether de-implementation is stand-alone or embedded/paired with implementation	Classification used (evidence strength; societal inefficiency; preference-fit), and the intended endpoint [5, 37, 46]. Single vs. dual behavioural target; sequencing plan; if paired, explicit alternative practice and expected substitution dynamics [25, 28].
2. Process	Mobilize existing implementation process models, adapted for de-implementation.	Stages followed; where adaptation occurred (e.g., prioritization, stakeholder legitimacy work, sustainment for “less”) [9, 10, 29].
3. Determinants	Conduct multilevel determinant assessment; treat outer-setting constraints as first-order.	Determinants by level (individual/inner/outer); decision arena mapping; incentives/regulation/professional norms/media/public expectations [5, 14, 23, 38].
4. Strategy development	Tailor strategies to determinants and make mechanisms explicit.	Determinant to mechanism hypothesis to strategy components; actor(s), dose/intensity, timing, and operationalization (including workflow defaults and cue structure) [3, 5, 16, 43]
5. Strategy classification	Use ERIC where possible, but annotate “stopping-specific” levers; contribute to taxonomy development.	ERIC labels plus de-implementation annotations (accountability/friction/restriction/policy levers/communication scripts); ethical considerations for punitive levers [16, 41].
6. Outcomes vs. effects	Measure de-implementation outcomes and effects, plus mechanism indicators.	Outcomes (acceptability/feasibility/fidelity/sustainability etc.) plus effects (utilization change) plus mechanism indicators aligned with theory; heterogeneity by context where feasible [2, 5, 44, 45].
7. Patient and unintended consequences	Routinely include patient experience and unintended consequences.	Patient-level outcomes (trust, anxiety/fear, relationship impacts), equity implications, substitution to other low-value practices, delayed diagnosis concerns [5, 37, 46].
8. Participation	Prioritize participatory/co-design approaches when legitimacy and preference-fit are central.	Partner roles (patients/clinicians/decision-makers), co-design outputs (communication tools/scripts), and how experiential knowledge contributed to strategy design [5, 14, 36, 46]
9. Initiation toolkit	Develop and validate toolkits to identify/prioritize targets for de-implementation.	Integrated criteria (evidence strength, inefficiency, preference-fit, feasibility, political/organizational readiness) and decision support outputs [2, 5, 37]
10. Reporting standards	Develop de-implementation reporting standards to ensure reproducibility.	Required items: aim type; configuration; low-value classification; determinant–strategy–mechanism map; patient/unintended consequences; dual trajectories when paired [5, 25, 45, 46].

Recommendation 1: Specify the low-value practice and the de-implementation configuration upfront (stand-alone vs. embedded; single vs. dual target)

Because de-implementation may occur as a stand-alone initiative or as embedded or paired with implementation, for instance when substitution is required, studies should explicitly state the configuration and the implied target behaviour(s) [25, 28]. In embedded and paired initiatives, the appropriate unit of analysis is often two trajectories, i.e., decrease low-value practice and increase alternative practice, and evaluation should anticipate substitution effects and other unintended consequences [5, 25].

Recommendation 2: Use implementation process models, but adapt what must be explicit in de-implementation

Given the strong overlap in staged change logic, it is reasonable to mobilize established implementation process models (e.g., KTA-derived approaches) to guide de-implementation [9, 10, 29]. However, de-implementation requires explicit reporting of elements that are often implicit in implementation studies: the de-implementation aim (i.e., reduction, restriction or elimination), how “low-value” is defined (i.e., evidence strength, inefficiency or preference-fit), and the decision arena (i.e., who can enable, block the stopping across levels) [5, 24, 37].

Recommendation 3: Make determinant assessment visibly multilevel and treat outer-setting constraints as first-order

Determinant frameworks developed for implementation (TDF, TICD, CFIR) remain appropriate, but de-implementation studies should more consistently foreground outer-setting determinants (incentives, regulation, professional norms, public/media expectations, stakeholder interests), which often shape feasibility and legitimacy early [5, 14, 23, 38].

Recommendation 4: Develop strategies that are determinant-informed and mechanism-explicit

A recurring gap is weak transparency around how strategies were derived from determinant assessment [3, 16, 43]. De-implementation studies should explicitly map determinants to hypothesized mechanism(s) to strategy components, especially when targeting mechanisms that differ from implementation such as habit disruption, accountability, friction, loss-framed judgments, and political feasibility [5, 22, 47].

Recommendation 5: Extend strategy taxonomies (or annotate ERIC) for de-implementation-relevant strategy families

Because most de-implementation strategies map onto ERIC yet some appear under-specified or missing (e.g., accountability tools; warning/communication tools; policy/regulatory levers), the field would benefit from either (i) an ERIC extension/annotation for de-implementation or (ii) a dedicated de-implementation taxonomy that preserves comparability while capturing “stopping-specific” levers and ethical considerations [16, 41].

Recommendation 6: Measure both de-implementation outcomes and de-implementation effects and include mechanism and unintended consequences

Most studies over-focus on utilization change (effects) [3, 16, 43]. De-implementation evaluations should distinguish (i) process outcomes (adapted from Proctor/Prusaczyk: acceptability, appropriateness, feasibility, fidelity, penetration, sustainability) from (ii) effects (change in low-value practice use), and (iii) mechanism indicators (intermediate outcomes) aligned with the chosen theory (e.g., cue exposure, perceived losses, legitimacy, sanction certainty) [2, 5, 44, 45]. Study designs commonly used in implementation research including experimental and quasi-experimental designs, qualitative, and mixed-methods studies can be leveraged to test whether hypothesized mechanisms (e.g., habit disruption, accountability, friction, etc.) are activated and whether they mediate changes in factors influencing low-value care use [2].

Recommendation 7: Integrate patient perspective as a key outcome domain, not an add-on

Because stopping care can trigger fear, mistrust, and relationship strain, patient-level outcomes such as acceptability, anxiety, fear, trust, perceived abandonment, shared decision experience and equity-relevant unintended consequences should be routinely included, particularly when the low-value classification includes a patient-preference dimension [5, 37, 46].

Recommendation 8: Prefer participatory and co-design approaches when legitimacy and preference-fit are central

Participatory action research, co-production, and co-design are especially valuable in de-implementation because legitimacy, patient expectations, and professional identity threats can be decisive barriers. These approaches can also improve the tailoring of communication tools and shared decision supports [5, 14, 36, 46].

Recommendation 9: Develop upstream tools to initiate de-implementation (identification/prioritization toolkits)

To our knowledge, there is no widely adopted toolkit that supports identifying and prioritizing low-value care for de-implementation in a way that integrates evidence strength, societal inefficiency, and patient preference-fit while aligning with local priorities and feasibility constraints [2, 5, 37]. A practical research priority is to develop and validate such a toolkit.

Recommendation 10: Establish reporting standards specific to de-implementation

To strengthen reproducibility, reporting standards tailored to de-implementation should minimally require: the de-implementation aim (i.e., reduction, restriction, elimination), configuration (i.e., stand-alone vs. embedded/paired), low-value classification used, determinant-to-strategy mapping, mechanism indicators, patient/unintended consequences outcomes, and when paired, dual-trajectory measurement (stop and start) [5, 25, 45, 46].

Conclusion

This paper advances conceptual clarity by contrasting implementation and de‑implementation at the level of their theoretical foundations to generate actionable methodological guidance. Grounded in the argument that de‑implementation is not simply “implementation in reverse”, our theory-driven narrative review highlights three cross‑cutting insights: (i) stopping a practice activates distinct psychological mechanisms, including cognitive biases, habit disruption, professional identity threats, and loss‑framed judgments; (ii) de‑implementation is shaped by multilevel constraints and more strongly by higher-level determinants and politics such as incentives, regulation, professional norms, and public expectations; (iii) the nature of low‑value care, its strength of evidence, inefficiency, or misalignment with patient preferences, together with the intended endpoint, informs strategy selection and the outcomes to assess. These insights underpin ten methodological recommendations that span key domains of de‑implementation research, from study design elements such as framing, configuration, and participatory approaches to broader considerations including reporting standards, and the determinants, strategies, and outcomes that shape de‑implementation efforts. Future research should use this guidance to sharpen methodological rigor and continue refining the theoretical assumptions surfaced by this analysis. Given the implications of low‑value care for patient safety and quality, advancing understanding of de‑implementation is essential and strengthening how we study “stopping” is central to improving the value and integrity of healthcare systems.

Author contributions

SSR: conception and design of the paper; writing the article. GF: collaboration to the conception and design of the paper; writing the article; critical revision of the article. MHG: collaboration to the conception and design of the paper; critical revision of the article.

Funding

No funders were explicitly involved in this article. SSR is currently funded for her doctoral studies by the Canadian Institutes of Health Research (Canada) and the Fonds de recherche du Québec – Santé (FRQS, Canada). GF and MHG are holder of a Junior 1 salary award from the FRQS.

Data availability

No datasets were generated or analysed during the current study.

Declarations

Ethics approval and consent to participate

Not applicable.

Consent for publication

Not applicable.

Competing interests

GF is an Associate Editor for Implementation Science Communications.