Barriers and facilitators influencing medication-related CDSS acceptance according to clinicians: A systematic review

A


Introduction
Medication-related problems are responsible for approximately 3-5% of all hospital admissions and approximately 20 % of all readmissions [1,2]. Various aspects, such as relevant patient characteristics and drug-drug interactions, need to be considered by the clinician during medication-related processes (e.g. prescribing, medication review etc.) [3,4]. Errors made during these processes can cause preventable injuries, negatively affecting patient safety and leading to unnecessary health care costs [3]. Therefore, it is of great importance to diminish the number of adverse drug events.
Clinical decision support systems (CDSSs) are systems that link patient health data with health knowledge (e.g. computer-interpretable guidelines) to guide the clinical decision making process [5]. CDSSs can support many aspects of care, such as preventative care, diagnosis, or therapy, including medication [6,7]. A medication-related CDSS is a system that supports medication-related decisions and processes, such as prescribing, administration, and monitoring for effectiveness and adverse effects.
Research shows that medication-related CDSSs offering advice to clinicians can prevent medication errors and thereby improve patient safety and healthcare quality [8][9][10]. However, clinicians override 49-96% of all drug-safety alerts [11]. This raises the question of which factors influence the acceptance of these systems. Acceptance entails not just accepting the CDSS (i.e. purchasing or using it), but also accepting its guidance and advice. Barriers are considered factors that negatively influence the acceptance of the CDSS. Facilitators, on the other hand, encourage acceptance of a CDSS. Insight into barriers and facilitators is needed to improve the acceptance of CDSSs.
In particular, it is useful to know what the clinicians themselves, as the primary users of CDSSs, see as barriers and facilitators for CDSS acceptance. Recently, Van Dort and colleagues [12] systematically reviewed 13 qualitative studies regarding barriers and facilitators for using medication-related CDSSs. They found that perceived threats to clinical autonomy, mistrust of information and irrelevant alerts were important barriers, while perceived improvements to patient safety and efficiency and ease of use were important facilitators. In our review, we aim to expand these findings by creating a broader overview of barriers and facilitators for medication-related CDSS acceptance. Qualitative studies provide an in-depth view from a small sample which is typically not representative of a larger population, while surveys provide a less detailed view, but of a larger and more representative section of the population [13]. Combining knowledge from qualitative as well as survey studies will result in an overview of the full continuum of research in the field. This systematic review thus aims to create a complete overview of barriers and facilitators of medication-related CDSS acceptance as reported by clinicians.

Search strategy
The systematic review was pre-registered in Prospero (Ref. CRD42020171318). The literature search was performed on 14 April 2020 in four databases: Medline, Scopus, Embase and Web of Science Conference Proceedings. The search was developed with the aid of a medical librarian (JD), and customized for each of the databases (Appendix A). To verify the search strategy, we confirmed that 21 relevant articles obtained by citation analysis and similarity tracking appeared in the search results. Furthermore, the references of included papers were manually checked for possible additional relevant papers.

Study selection
We applied the following inclusion criteria: (1) The study was a qualitative or survey study investigating barriers and facilitators for CDSS acceptance. Investigating barriers and facilitators did not have to be an aim of the study. (2) The study reported the results of primary data collection (thus not a review or other secondary data). (3) The study focused on medication-related CDSSs. This entails patient-specific advice about medication-related care processes (e.g. prescriptions, drug alerts). Thus, an e-prescribing system or Computerized Provider Order Entry without drug-related alerts was not considered a CDSS. However, a study reporting on clinicians' opinions of drug-drug interaction-or allergy alerts within such as system would be eligible for inclusion. (4) Study participants were clinicians who were (potential) users of the CDSS, with barriers and facilitators gathered from these clinicians. We excluded usability studies, as these look at more systemspecific usability problems rather than general barriers and facilitators, and two systematic reviews on usability aspects have been conducted recently [14,15]. While screening the titles and abstracts, articles that did not allow a clear inclusion/exclusion decision were retained for the full paper phase. Subsequently, during full text screening, all of the inclusion criteria had to be fully met.
To ensure reliability in the screening process, all records were screened in Rayyan [16] by at least two reviewers (LW, KP, KV, SG, LS). To ensure a common understanding of definitions and inclusion criteria, decisions on the first 200 titles screened by any two reviewers were compared and discussed before completing the rest of the screening. All full texts were screened by two reviewers (LW, KP, KV, SG). In both phases, any disagreement was resolved through discussion between the two reviewers, if any uncertainty remained, a third reviewer was consulted (SM).

Data extraction
Data were extracted from all studies by one author (LW) using a data extraction sheet tested independently by four authors (LW, GB, JW, SM). Title, authors, year of publication, and journal were extracted from each article. Furthermore, for each study we extracted the setting, year of data collection, type of study, type of questions asked, country, aim of the study, number, age and work experience of participants, and type of clinicians participating. Relevant information regarding the CDSS and its target users, and the barriers and facilitators for CDSS acceptance as mentioned by clinicians were extracted. We classified an item as a barrier or facilitator according to the classification used in the source study; no attempt was made to reclassify related items (e.g. "takes time" as a barrier and "saves time" as a facilitator).

Quality assessment measure
To assess the methodological quality of each of the included studies, the validated QualSyst tool [17] was used, as it allows scoring of both qualitative and quantitative studies. Quality assessment was done concurrently with data extraction by one author (LW). A summary score for each study was calculated by dividing the total score by the total possible score, resulting in a score between 0.0 and 1.0. Studies with a score of 0.5 or higher were considered of sufficient or good quality.

Data analysis
The extracted barriers and facilitators were categorized using the Human, Organization and Technology-fit (HOT-fit) model, intended for evaluation of health information systems, such as CDSSs [18] (Fig. 1). HOT-fit extends the IS success model's [19] constructs of Use, User Satisfaction, and Information, System, and Service Quality with the organizational factors and concept of "fit" from the IT-organization fit model [20]. Barriers and facilitators to acceptance among clinicians (Human), technical problems with the software (Technology) and the extent to which the system can be integrated in the organizational environment (Organization) all affect CDSS usage [10]. For each of these dimensions, Yusof and colleagues provide "evaluation measures" [18] which constitute sub-categories of the eight components (e.g. System Quality, Information Quality, Structure etc.), and were used to categorize the extracted barriers and facilitators.
Barriers and facilitators that did not fall into this classification were placed in an "other" category, and then grouped into emergent themes. There seemed to be partial overlap between the component User Satisfaction and other components (which sometimes included indirect Fig. 1. The HOT-fit framework [18]. indications of satisfaction). Therefore, we defined User Satisfaction as remarks specifically about satisfaction, and not remarks from other categories that imply satisfaction. Categorization was carried out by one author (LW). Any ambiguities were thoroughly discussed by a team of four of the authors (LW, GB, JW, SM). The total number of barriers and facilitators related to each HOT-fit component and underlying evaluation measure was counted, resulting in an overview of the mostfrequently-mentioned barriers and facilitators.

Search results
Our search strategy resulted in 6816 records. After removing duplicates (n = 1412), 5404 records remained for title and abstract screening, during which, 5264 records were excluded. Subsequently, the remaining 140 full-text papers were screened. Agreement on independent full-text screening was 86 %, all disagreements were resolved after discussion. In total 63 articles were included. Fig. 2 summarizes the complete screening process in a PRISMA flow diagram.
The review includes a mixture of qualitative (n = 42), survey (n = 16) and mixed methods (n = 5) studies. The studies were performed at various sites, with hospitals (n = 31) and general practices (n = 11) being represented most frequently. The studies include data collected in 23 different countries, with the most common being the USA (n = 20) and Australia (n = 12). Information on each included paper can also be found in Table 1. Quality assessment of the included papers yielded an average score of 0.71 (range 0.42− 0.89), with two studies having a score below 0.5. We checked to see if these two studies influenced our results. However, the themes presented in our results section below still predominated when excluding these two studies. We therefore retained them in our analysis.

Barriers and facilitators
In the 63 included studies , 327 barriers and 291 facilitators were identified. Barriers or facilitators named in more than one study were consolidated, resulting in 195 unique barriers and 174 unique facilitators. Barriers and facilitators were categorized into HOT-fit's evaluation measures, which are sub-categories of the dimensions Information Quality, Service Quality etc. An overview of the most-frequently-encountered evaluation measures with example barriers and facilitators can be found in Table 2. In this table, the total amount of barriers and facilitators is reported for each evaluation measure. In the text below, we report the unique number of barriers and facilitators for the most frequently encountered evaluation measures. The complete list of barriers and facilitators can be found in Appendix B.

Technology
Barriers and facilitators from all three categories in the Technology component of HOT-fit were encountered. Information Quality was represented in more barriers than facilitators (n total barriers = 97, n total facilitators = 67), while System Quality (n total barriers = 97, n total facilitators = 104) and Service Quality (n total barriers = 2, n total facilitators = 4) were recognized in more facilitators. The most-oftenencountered evaluation measures related to Information Quality were usefulness (n unique barriers = 12, n unique facilitators = 16), relevance (n unique barriers = 14, n unique facilitators = 2), format (n unique barriers = 4, n unique facilitators = 17), conciseness (n unique barriers = 5, n unique facilitators = 6), reliability (n unique barriers = 4, n unique

Human
Both factors regarding the Human component of HOT-fit were present in the included articles. System Use was mostly encountered in barriers (n total barriers = 72, n total facilitators = 12), while User Satisfaction was mostly visible in reported facilitators (n total barriers = 5, n total facilitators = 11). The most-often-reported evaluation measures related to system use are expectation/belief (n unique barriers = 16, n unique facilitators = 3), training (n unique barriers = 4, n unique facilitators = 2) and reluctance/resistance (n unique barriers = 8, n unique facilitators = 0). Examples included dependence on the system and not receiving adequate training.

Organization
Reported barriers and facilitators were related to both categories in the Organization component of HOT-fit: both Structure (n total barriers = 24, n total facilitators = 5), and Environment (n total barriers = 6, n total facilitators = 2) were most often encountered in barriers. Related to Structure, the evaluation measure clinical process (n unique barriers = 7, n unique facilitators = 3) was most often represented, concerning, for instance, the workflow of the clinician.
= 0, n unique facilitators = 4) and clinical practice (n unique barriers = 0, n unique facilitators = 10), with barriers and facilitators such as patient safety and prescription errors.

Outside of HOT-fit
Besides the aforementioned classification of barriers and facilitators within the HOT-fit model, the included papers also yielded some barriers and facilitators that did not fit into the HOT-fit model (n total barriers = 22, n total facilitators = 24). A total of 8 barriers and 5 facilitators were related to the context in which the CDSS was used. For instance, clinicians mentioned as a barrier that the CDSS was not useful on a specific ward such as the intensive care unit or the emergency department [83]. Furthermore, 6 facilitators were related to the system having an educational role; this was a facilitator because the system has the potential to increase clinicians' knowledge [25,33,57,77,79,83].

Main findings
This systematic review aimed to identify barriers and facilitators for medication-related CDSS acceptance as indicated by clinicians. Our review revealed that the included studies mostly focused on barriers and facilitators related to the Technology component from HOT-fit, and more specifically to Information Quality and System Quality. Barriers and facilitators about efficiency and ease of use of the system, and usefulness and relevance of the information were most often reported. Systems being time consuming was the most often encountered barrier, and ease of use was the most often encountered facilitator. The other HOT-fit components, Human, Organization and Net Benefits, were encountered less often. Context was identified as an important new factor.

Interpretations, implications and impact
The evaluation measure efficiency was most often encountered in all of the barriers and facilitators. Clinicians often mentioned time constraints as an impeding factor and the system saving them time as a facilitator. While designing a CDSS, developers should keep in mind that system usage should not be time-consuming or ideally even time-saving. Ease of use was also a frequently occurring theme and should be closely monitored while developing new CDSSs. Clinicians often indicated that a simple, easy-to-use system facilitates usage. Complex systems with hard-to-find information inhibited usage according to clinicians.
Furthermore, usefulness and relevance of the information presented were also often mentioned. If clinicians perceived the advice as useful, this facilitated usage. On the other hand, redundant alerts, e.g. if the presented information is already well known, were indicated as a barrier. Similarly, irrelevant alerts were seen as a major barrier, for instance if an alert regarding pregnancy was shown for a male patient. In practice, this means that the content of the alerts should be critically evaluated. Only truly useful and relevant information should be presented to the clinician.
A common theme in the factors discussed above is that clinicians agree that certain factors inhibit or facilitate usage, but have different views on how to achieve this. Clinicians for instance agreed that useful information facilitates usage, but had different visions on which information is useful. Therefore, during development, clinicians from the system's target group should be involved. User-centered design could be a suitable method for this, as research has shown its ability to make CDSSs more effective and easy to use [84]. This will allow system developers to make sure that the CDSS's features overcome barriers and facilitate acceptance of the system before implementation.
Some aspects of the HOT-fit model were found less frequently in our results; specifically, User Satisfaction, Service Quality, and Environment. We defined User Satisfaction as remarks specifically about satisfaction, and not remarks from other categories implying satisfaction, thus it not surprising that this is rarely used. Likewise, it is logical that when discussing their interaction with a CDSS, clinicians may not spontaneously discuss the organization-and government-level factors related to Environment, or mention Service Quality unless asked directly about those factors. Further research is needed to determine if these factors nonetheless play an important role in CDSS acceptance.
From the barriers and facilitators that could not be categorized into HOT-fit, one clear theme arose: context. Clinicians indicated that a CDSS's success was dependent on the context in which it was used; some systems were only considered useful in certain specialties. Context is relevant for non-medication-related CDSSs as well; for example, systems that aid in the diagnosis of disease are inherently specialty-specific, as well as preventative care reminders etc. Therefore, context is a theme that should be taken into account when developing a CDSS. However, we also see this in studies of other health information systems. Electronic Health Record (EHR) adoption has been shown to significantly correlate with physician specialty, suggesting context is also important in these systems [85]. Therefore, adding context to the HOT-fit model can make it more complete and useful for future studies.
HOT-fit provides a useful lens for viewing the results of this review, but they could be interpreted in light of other models as well. The Two-Stream Model views a CDSS's interaction with a user as two decisions: deciding what advice to present and deciding how to present it [86]. In this model, barriers such as "Guidelines did not always pertain to all of the patients" (usefulness) [58] would be grouped under Clinical Knowledge, while "Skepticism of the accuracy of described data and information" (data accuracy) [45] would be seen as reflecting problems with data and/or the encoded logic of the CDSS. Other models could provide different views, and thus different insights, on the data identified by our review.

Comparison to other studies
The results of this systematic literature review are complementary to previously-conducted reviews. Kilsdonk and colleagues [87] conducted a review of factors influencing CDSSs' success, using the HOT-fit framework. Similar to our results, they found that organizational HOT-fit factors were underrepresented in the included literature. However, Kilsdonk and colleagues focused on "mid-level guideline-based CDSSs" (thus not focusing on medication and excluding "simple" CDSSs such as drug dose, allergy, interaction). Their review also differed methodologically; our review focuses on opinions of end users (clinicians), while the Kilsdonk review did not differentiate between opinions of the system users and opinions of the authors or development team.
In 2013, Roshanov and colleagues [88] performed a meta-analysis of factors reported by the authors of published randomized controlled trials and their association with the efficacy of the systems under study. The findings of this study have surprisingly little overlap with our findings. Unsurprisingly (but in contrast to Roshanov et al.), the studies in our review report integration with the workflow and at the time of decision-making as a facilitator, e.g. "Alerts during medication prescribing were generally viewed as more helpful" [43]. Likewise, although Roshanov showed a negative association with success for CDSSs integrated in the EHR, lack of integration with other systems is reported as a barrier. Roshanov reported positive findings for systems that offered advice to patients and systems that require professionals to enter a reason for overriding advice; these were not mentioned in our studies, although data entry in general is seen as a barrier. These contrasting findings illustrate that qualitative and quantitative techniques are complimentary, and offer different insights on the same problem.
Lastly, our findings are in line with the prior review by Van Dort and colleagues [12]. However, our review includes far more studies and therefore provides a more complete overview of barriers and facilitators.

Strengths and limitations
There are several methodological strengths of this systematic review. Firstly, the search strategy was made in collaboration with an expert in systematic literature searches and performed in multiple databases. Reference lists were cross-checked to ensure no papers were missed. Furthermore, every record was screened by at least two coders in title/ abstract and full text inclusion. This reduces the chance of missing eligible papers. Lastly, the data extraction sheet was created and tested in an iterative process with multiple authors (LW, GB, JW, SM), ensuring reliability of the data extraction process.
The HOT-fit model was considered a useful model for categorization of the barriers and facilitators, and this systematic categorization is a strength of our review. However, the authors of the HOT-fit model provide explanations regarding its main components (e.g. Information Quality, System Use etc.), but the underlying evaluation measures are merely named and not defined. Other systematic reviews using HOT-fit have also reported this issue [87,12]. Additional clarification about how to use and interpret the evaluation measures would make categorization more reliable. Furthermore, even though the categorization was thoroughly discussed by a team of four authors (LW, GB, JW, SM), the entire list was not independently coded by a second person. Applying this approach in the future might also enhance reliability of the categorization.
Furthermore, the results of this review showed some gaps in the literature, as some of the HOT-fit dimensions were barely present in the extracted data. However, our results only show which dimensions are missing, but not why they are missing. The question remains whether these dimensions are understudied, or whether they are simply not as relevant in a CDSS context. Whether a barrier or facilitator emerges in the results of a study depends on what questions were asked, particularly in closed-question survey studies. A limitation of closed-question survey studies is that they direct respondents to select or report items. Surveys with open questions and qualitative studies allow more freedom in responses. Thus a barrier or facilitator might not be mentioned frequently because it is not important to the participants, or because they were focused on other themes. It is important to keep this limitation in mind when interpreting the results of this study. Although a list of common barriers and facilitators is valuable for the design and implementation of future systems, it is not a substitute for the involvement of users in the development process.

Future research
This study focused on medication-related CDSSs; future research could compare overlap in barriers and facilitators between this and other domains, and assess the reasons behind these differences. It can also be assessed whether some barriers or facilitators are universal for all CDSSs or if there are domain-dependent patterns. Furthermore, it could be interesting to look at how barriers and facilitators might vary per user group and per context. Splitting up the results for different groups in this way can create valuable additional insights. This could eventually lead to guidelines for developing specific kinds of CDSSs.
Overall, the findings of the current review are especially relevant for medication-related CDSS developers. More specifically, in the near future the findings will be used in the development of a medicationrelated CDSS for general practitioners to reduce older patients' medication-related fall risk. During the development of this system, and also of other future CDSSs, the barriers and facilitators found in this review can suggest specific features and functions that will help overcome barriers and facilitate usage.

Conclusion
In short, this review provides valuable insight into a broad range of barriers and facilitators for accepting a medication-related CDSS as perceived by clinicians. The Technological HOT-fit component predominated, and clinicians named many barriers and facilitators related to System Quality and Information Quality, for instance regarding efficiency, usefulness and ease of use. We also found context to be an important additional factor. To our knowledge, the current review is the first large systematic review of barriers and facilitators for medicationrelated CDSSs. The barriers and facilitators identified by this study can be used as a starting point for designing high-quality CDSSs, although they should not be considered a substitute for involvement of end-users during the development. Furthermore, future research should explore similar overviews of barriers and facilitators for usage in different CDSS domains. Eventually this will contribute to the development of more effective CDSSs and ultimately improve patient care. Summary table What was already known on the topic • Medication-related Clinical Decision Support Systems (CDSS) offering advice to clinicians can prevent medication errors and thereby improve patient safety and healthcare quality. • Even though evidence of its effectiveness exists, a high percentage of CDSS systems are not used, and alerts are overridden or ignored by clinicians. • Individual studies have investigated clinicians' reasons for accepting or not accepting medication-related CDSSs, but no attempt has been made to systematically summarize the evidence base of reported barriers and facilitators of medication-related CDSS acceptance.

What this study added to our knowledge
• This review provides a valuable, systematic insight into a broad range of barriers and facilitators for medication-related CDSS acceptance as perceived by clinicians. • Data categorization provides a clear overview of frequently found themes and gaps in the literature. • The common barriers and facilitators identified by this study can be used as a starting point for the design of high-quality CDSSs, although they should not be considered a substitute for involvement of end-users, preferably from the start of the design process. ;1;

Declaration of Competing Interest
The authors report no declarations of interest.