Background
Recognition is growing with regard to the importance of reducing low-value care (LVC), i.e., “care that is unlikely to benefit the patient given the harms, cost, available alternatives, or preferences of the patient” [
1]. Common examples of LVC are non-indicated antibiotics, unnecessary imaging, potentially inappropriate medications for the elderly and unnecessary lab tests [
2]. LVC has become a pervasive problem in health care in high-income countries [
1,
3‐
5], with around 30% of care estimated to be of low value [
6]. Furthermore, estimations show that about 7% of the care considered to be best practice 1 year becomes LVC the next [
7]. Thus, the rapid development of new practices (e.g., diagnostics and treatments) not only calls for continuous implementation of new evidence but also requires the de-implementation of LVC. De-implementation entails a structured process with the purpose of reducing or ceasing the use of LVC [
8].
Similar to implementing evidence-based interventions, de-implementing LVC is a complex process influenced by multilevel factors [
9]. Determinants of LVC use and de-implementation include various patient characteristics such as age, gender, ethnicity, and socio-economic factors, although there are no consistent patterns as to their positive or negative influence on LVC [
2]. For instance, older age is usually associated with use of LVC [
10‐
12], but some studies have linked younger age with higher LVC use [
13,
14]. Patients’ health conditions—e.g., the severity of illness and characteristics of the disease [
15,
16]—often contribute to use of LVC. Patient expectations, e.g., patients who request non-indicated prescriptions, also tend to increase the occurrence of LVC [
17,
18]. Health professionals’ characteristics are also associated with LVC use. As with patient characteristics, the results are inconsistent regarding professionals’ age, gender, and length of experience [
2]. However, a lack of or inadequate training has consistently been linked to use of LVC [
17,
19]. Professionals’ knowledge of LVC contributes to and protects against the use of LVC [
2]. For example, a lack of knowledge about cost-effectiveness [
20] and poor cost-awareness [
21] are associated with use of LVC. Professionals’ expectations and attitudes also influence LVC use, e.g., their fear of malpractice and desire to meet patient requests [
2]. Interaction between patients and professionals can also impact the use of LVC, e.g., communication about unnecessary antibiotics or tests [
22].
Determinants of LVC also exist at the contextual level [
2]. Inner context, including setting characteristics, care processes (e.g., lack of care continuity), perceived lack of time and time pressure when performing work tasks, accessibility of decision support, staffing levels, and composition and organizational incentives for LVC use have been identified [
2]. Outer context determinants have included location of the health care organization (e.g., metropolitan, urban, suburban, or rural), financing and financial incentives (e.g., fee-for-service funding), policy and political support, and marketing initiatives such as promotion of screening directed to the population and direct-to-consumer advertising about drugs or treatments.
Yet, knowing the potential determinants of LVC is not sufficient for changing them. Strategies to address determinants constitute the “how-to” component of changing practice. Strategies are methods and techniques to facilitate implementation of evidence-based practices and/or de-implementation of LVC [
23]. Some implementation strategies are likely to be applicable for de-implementation, while other strategies may be unique or more applicable for de-implementation [
24]. However, the evidence for de-implementation strategies is dispersed across multiple clinical fields, which makes it difficult to document and survey findings [
2]. Studies investigating strategies to reduce LVC have been published in a broad range of journals, typically within specific clinical and medical care areas, from microbiological research on antimicrobial resistance to potentially inappropriate medication for the elderly [
2]. Studies on strategies for de-implementation have also focused on specific LVC within fields such as nursing [
25], low-value blood management techniques in primary hip and knee arthroplasty [
26], pharmacological prescriptions [
27], and cancer [
28]. An exception is a systematic review of de-implementation strategies covering a wide range of clinical areas [
29] that found promising results for clinical decision support and performance feedback, concluding that multicomponent strategies addressing both clinicians and patients had the greatest potential for reducing LVC. Another review [
28] focused specifically on cancer care similarly found that most de-implementation strategies were multifaceted. The most widely used strategies were audit and feedback, use of clinical champions, educating clinicians through developing and disseminating guidelines, and decision-support tools. Integrating a clinical decision-support tool in the electronic health record system for real-time alerts was the most effective strategy.
The de-implementation field suffers from a lack of established nomenclature for how strategies to de-implement LVC are named, defined, and organized. This makes it difficult to compare strategies across studies, hindering an accumulation of a generalizable body of knowledge related to effectively de-implementing LVC. A notable exception is a recent study in which researchers used the behavior-change techniques taxonomy to categorize de-implementation categories based on data from three systematic reviews [
30]. They also compared de-implementation strategies to implementation strategies and showed that
behavior substitution,
monitoring of behavior by others without feedback, and
restructuring social environment were more frequently used in de-implementation efforts than in implementation. However, this study was limited to strategies aiming to change individual clinicians’ behavior and did not cover strategies on a system or policy level.
In contrast, researchers studying implementation have developed taxonomies to guide the identification, selection, and reporting of strategies, thus making it easier to compare strategies across studies [
31‐
34]. One such taxonomy is the ERIC compilation [
31], which consists of 73 discrete implementation strategies belonging to nine categories [
35]. ERIC has been widely used in implementation science and is useful in evaluations of implementation strategies [
36‐
39]. However, it is unknown whether the same types of strategies are also used in de-implementation. It is likely that some of the 73 strategies and nine categories of the ERIC compilation are also relevant for de-implementation purposes. Still, findings from a recent study [
30] using behavior change taxonomy suggest that this might not be the case, since de-implementation differs in the primary behavior-change techniques utilized. Thus, it is necessary to investigate to what extent de-implementation and implementation strategies are the same.
This study addresses two important knowledge gaps in the literature. First, due to the dispersion of studies on strategies for de-implementation across many clinical fields, no overview of the strategies that can be used to support LVC de-implementation exists. This fragmentation inhibits the systematic development of knowledge about effective strategies for the de-implementation of LVC. Second, it is unknown whether and to what extent implementation strategies are also applicable for de-implementing LVC. Addressing these key knowledge gaps, the aim of this review is to evaluate the scope of the literature to (1) identify strategies for the de-implementation of LVC described in the scientific literature and (2) compare de-implementation strategies to implementation strategies as specified in ERIC and strategies added by Perry et al. [
36].
Discussion
This scoping review identified a total of 71 unique de-implementation strategies described in the literature. Of these, 62 strategies could be mapped onto ERIC compilation strategies, whereas four strategies could be mapped onto one of the added strategies [
36]. Thus, 87% of de-implementation strategies reported across various fields overlapped with strategies used in implementation, while four identified strategies could not be mapped onto any existing implementation strategy. Two of these strategies (i.e.,
policy and regulations and
international collaboration) are likely to be useful for both de-implementation and implementation, whereas
accountability tool, communication tool and
black box warning may be unique to de-implementation.
The most commonly used category of strategies was
train and educate stakeholders, ranging from
distribute educational materials to
make training dynamic. These types of strategies are also prevalent strategies for implementation [
100]. However, previous studies have suggested that education alone is insufficient for successful de-implementation [
29] and implementation [
101]. Colla et al. [
29] found that educational strategies, combined with patient information and/or audit and feedback (i.e., multicomponent strategies), were more effective at reducing LVC. Regardless, as many as 24 of the studies in our review were based on the
train and educate stakeholders category as the sole strategy. This suggests that de-implementation strategies may be chosen pragmatically, without much regard for research findings as to what is most effective.
Four de-implementation strategies were not possible to map onto ERIC or the additional strategies suggested by Perry et al. [
36]. Three of these strategies may be unique to de-implementation and thus differ from the implementation process of introducing a new practice. Of these, the strategy
accountability tool (
n = 22) was most common and had the purpose of holding clinicians accountable when prescribing an LVC practice. It serves to disrupt the habitual use of a practice and forces clinicians to stop and reflect on whether they should prescribe the practice. This could be considered a more important strategy for de-implementation than implementation. The other strategies that may be unique for de-implementation were communication
tool and
black box warning.
Communication tools consisted of a structured method for communicating with patients or next of kin about why a patient did not receive a practice, and
black box warning consisted of a clear written warning on the packaging for certain medications. The other two strategies identified in this scoping review and not captured by ERIC or the additions by Perry et al. were policy and regulation and international collaboration. These two strategies might be relevant for both implementation and de-implementation, which could suggest that the ERIC compilation should be extended. Thus, as suggested in the study by Perry et al. [
36], a potential limitation of the ERIC compilation is that all possible implementation strategies may not be covered. The original authors of ERIC similarly stated that the compilation should not be seen as the final word and welcomed comments and critique [
31].
We found that only half (50%) of the 73 ERIC strategies had been used in the included studies. However, it is unclear whether the remaining strategies lack relevance or applicability for de-implementation or whether they have not been used for other reasons. Strategies from the category labeled
adapt and tailor to context may be less applicable for de-implementation, where drift from protocols and guidelines may be the very reason for a practice becoming LVC. Examples include indication creep (when a practice is used for purposes for which it has not been proven efficient) and prevention creep (when a practice developed for symptomatic disease is used for asymptomatic individuals [
102]. Other exemplary strategies that were rare in this review included the category
develop stakeholder relationships. This might have considerable potential as a de-implementation strategy, since one determinant for the use of LVC is professionals’ expectations, attitudes, and behaviors [
2]. Strategies such as informing local opinion leaders, identifying early adopters, or conducting local consensus discussions can influence professionals’ expectations, attitudes, and behaviors to support de-implementation.
Several of the de-implementation strategies that matched ERIC strategies involved more than one inductive code. For instance, the ERIC strategy
develop educational materials involved development of information materials that comprised several codes in the inductive coding based on the material’s target: staff, providers, or patients. In fact, patient expectations have been found to be an important determinant for the use of LVC [
2,
103], which suggests that strategies involving information for patients may be more important for de-implementation. The de-implementation strategies within the ERIC strategy
audit and provide feedback entailed many types of audit and feedback. Some of these researchers used individual feedback, and others delivered group-level feedback. More innovative examples included quality-improvement contests and setting a goal for prescription and delivering rewards when reaching the goal. Finally, one frequent example was delivering feedback only to high prescribers, which seems to be a more specific de-implementation strategy because a small number of clinicians can have a large impact on the total amount of prescriptions [
104]. Both of these examples indicate that some strategies are more multifaceted and heterogeneous than others, making it challenging to compare the effectiveness of strategies across studies. For future research on the de-implementation of LVC, various components of the strategies must be reported transparently and in detail, preferably using guidelines for specifying and reporting strategies [
105].
Very few of the identified studies used any of the ERIC strategies that could be classified as pre-analysis approaches to assist in choosing the most suitable de-implementation strategies. The pre-analysis strategies found were
assess readiness and identify barriers and facilitators (5% of the studies),
stage implementation scale-up (1% of the studies),
conduct a local needs assessment (3% of the studies), and
tailor strategies (3% of the studies). This finding implies that the choice of de-implementation strategies is rarely tailored to the determinants of LVC use. In contrast, the importance of a comprehensive analysis of the current practice is considered crucial for successful implementation [
101].
It is noteworthy that we could not find any studies within the behavioral health field (i.e., all studies were related to medicine). This could be due to the fact that it may be easier to determine that a medical practice is of low value because the efficacy or effectiveness of such practices can often be tested in trials that produce more unequivocal evidence. For instance, the problem of overprescribing antibiotics was defined in parallel with the development of the medication [
106], whereas the side effects within psychotherapy research have only been investigated in recent years [
107].
Knowledge gaps and implications
The findings provide an overview of the most-used strategies within de-implementation. Most strategies could be found within ERIC, suggesting that the same type of strategies used for implementation purposes are also relevant for de-implementation. However, some new strategies were found that could be interpreted as more relevant to de-implementation than to implementation, including accountability and communication tool. The accountability tool provides a hurdle for routine use of LVC, and the communication tool helps the professional communicate their decision not to use LVC to patients or their families. Only half of the strategies described in ERIC was found in our review. This could be because some implementation strategies are irrelevant or under-utilized for de-implementation. Future studies could determine if some unused implementation strategies are also beneficial for de-implementation purposes.
Almost one-third of the studies in this review were focused on non-indicated antibiotics, implying that the most common strategies found in this review are a reflection of the most common strategies within the de-implementation of this type of LVC. One question that remains unanswered is whether different strategies could be beneficial for different types of LVC. For instance, patient centered care was suggested as an alternative to potentially inappropriate medications where individual strategies were described based on what caused anxiety to patients with dementia and how to best calm them without the unnecessary use of medications (e.g., [
108]). This strategy is perhaps most suitable for patients within nursing homes or similar facilities and perhaps not for other patient populations. However, the lack of studies within behavioral health makes it difficult to assess the generalizability of the findings to this field.
Methodological considerations
A considerable strength of this research is the number of studies included and the breadth of clinical fields covered. The study was also rigorous in its processes. We designed and performed the literature search in collaboration with the university library. Two reviewers screened all references independently. Two of the authors completed all coding and mapping and solved issues through discussions within the entire author group. However, there were also some limitations. The inconsistent terminology for LVC and de-implementation makes it plausible that studies may have been missed. Moreover, the review only covered literature written in English and published in peer-reviewed sources. We did not report the efficacy or effectiveness of various strategies because this was not the aim of the study. This allowed us to include a wider range of study designs including qualitative, process evaluations and cost-effectiveness studies and as an effect identify a wider range of strategies. Furthermore, the amount of information concerning the described strategies in the included studies varied, which may affect the trustworthiness of the inductive coding and the mapping onto the ERIC compilation. Data were charted individually, which may have influenced the information extracted from the articles. To ensure that relevant data were charted in a consistent way, an additional author piloted the data charting form, all individuals conducting the data charting were trained, and the charted data for a subset of articles were compared across individuals before starting the data charting. Finally, it is unknown if we would have received other results if we had coded the data based on another taxonomy. Future studies could investigate how de-implementation strategies differ depending on which taxonomy is used to code the strategies (e.g., [
109]).
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.