Drug-drug interactions (DDIs) are preventable adverse events that are responsible for 5% to 14% of adverse drug reactions in patients that are hospitalized [1,2], are a major risk factor for hospitalization [3], and occur in up to 13% of older adult ambulatory patients [4-6]. Exposure to life-threatening DDIs occurs despite the widespread use of clinical decision support. Alarmingly, up to 24% of patients on warfarin receive a prescription for a nonsteroidal anti-inflammatory drug (NSAID), which increases the risk of gastrointestinal bleeding up to twofold [7,8].

Most electronic health records (EHRs) implement DDI clinical decision support functionality with underlying logic provided by drug knowledge base vendors, but DDI alerts continue to be overridden at rates as high as 90% [9-11]. The current model for DDI decision-making is highly clinician-centric in spite of the fact that patients can play a substantial role in preventing potential harm due to DDIs. Studies that have explored different clinical decision support for DDIs [12,13] indicate that interactive decision dashboards have the potential to foster informed decision-making by patients [13]. These decision aids allow patients and clinicians to deliberate together about the advantages and disadvantages of different therapies and arrive at decisions that are concordant with the best available evidence, clinician knowledge, and patient preferences [14,15].

Accordingly, the overarching goal of this study is to incorporate patient-centered shared decision-making (SDM) for addressing DDIs, an advance from clinician-centric decision-making models. SDM is a conversation where patients share their values and preferences to choose a treatment that aligns with their goals [16,17]. Electronic decision aids can support this conversation; however, SDM is uncharted in the DDI domain [18].

The purpose of this study was to design and evaluate the usability of DDInteract, an SDM tool for the warfarin and NSAID DDI.

The design and usability assessment of DDInteract was guided by user-centered design principles and an SDM framework. The user-centered process included iterative and overlapping steps of prototyping (ie, low fidelity, stable, and high fidelity), stakeholder feedback, and usability heuristics and testing [19-24]. The SDM framework consists of five steps: (1) seek your patient’s participation, (2) help your patient explore and compare treatment options, (3) assess your patient’s values and preferences, (4) reach a decision with your patient, and (5) evaluate your patient’s decision [25]. Figure 1 depicts a summary of the design and usability process. This study was approved by the University of Utah Institutional Review Board.

The design team consisted of multidisciplinary experts in DDIs, clinical decision support, patient and provider communication, SDM, and pharmacotherapy outcomes. The process began with an artifact appraisal of SDM tools, DDI alerts, and clinical practice materials relevant to anticoagulants. Information from this appraisal was used to sketch low-fidelity feature prototypes, which were reviewed and discussed with the design team in weekly meetings. Features and functionality deemed important by the design team were retained for future iterations. Features were linked to the SDM steps and checklist items from the International Patient Decision Aid Standards Collaboration [26]. Once the design team coalesced on preliminary feature designs, these were combined into an initial protype of the complete user interface using Adobe XD (Adobe Inc).

Target users (ie, 2 physicians and 1 pharmacist) were individually shown the initial complete user interface prototype and asked to provide feedback on the usefulness, aesthetics, proposed functionality, and content. Several iterations were made in collaboration with these target users until no substantial feedback was provided. At this point the prototype was considered stable enough for a heuristic evaluation. The heuristic evaluation was based on knowledge of Nielsen's 10 Usability Heuristics for User Interface Design [27,28] and was performed by two experts with training and experience in human-centered design, psychology, and medical informatics. The goal of the heuristic evaluation was to identify design flaws that could be addressed prior to conducting resource-intensive testing. Specific feedback regarding design that might impede users’ goals were noted and shared in a team meeting along with a discussion of potential solutions for each flaw. Once the stable prototype was modified to address findings from the heuristic evaluation, it was considered high fidelity and ready for usability testing.

Usability assessments consisted of two parts: (1) patient interviews with simulated clinic visits and (2) clinician task performance assessments and usability surveys. DDInteract was designed to be used by clinicians at the point of care. Since patients would not use DDInteract without a clinician present, we did not test task completion success and efficiency with patients.

Although DDIs are a novel application for SDM, we did assess several relevant electronic decision aids such as those used for cardiovascular and diabetes management [31]. We compared a variety of relevant decision aids to the SDM steps and the International Patient Decision Aid Standards checklist [26]. Generally, decision aids lacked features to elicit patient values and preferences. Additionally, decision aids varied on the content provided to make treatment decisions including price, effectiveness, and side effects of treatment [32].

Several features from relevant decision aids were adapted to the DDI use case. Features included the icon array, personalized risk, and ability to simulate risk based on patient factors and treatments. Features and functionality evolved through multiple iterations. For example, icon arrays have shown promise in communicating risk to patients and clinicians [33-35]. Figure 2 depicts how the icon array changed over two iterations. Icon array features were based on findings from the literature and expert feedback [36,37]. Once features such as the icon array were acceptable from the design team’s perspective, they were adapted and placed into the complete user interface (ie, stable prototype). Figure 3 depicts the stable prototype used for the heuristic evaluation. Overall, 2 major, 13 moderate, and 14 minor issues were identified. One of the two major issues was associated with the number four in Figure 3. The evaluators thought that users may not correctly interpret “Substitute” and “Add” when selecting treatment options. In response, we removed the order entry context, which will avoid a user from referencing the warfarin or NSAID order in process.

The DDInteract prototype used for evaluating usability consisted of four sections. The first section in the top left corner of Figure 4 is a patient-specific risk profile. Substantiated risk factors for gastrointestinal bleeding are listed and stratified by risk and supporting evidence. When patient-specific risk factors are pulled from the EHR, the toggle is on (ie, blue). Toggles can be manually changed to account for data not in the EHR or for clinicians to test different scenarios such as adding an antidepressant. The second section in the bottom left is a dynamic risk icon array. The icon array changes with risk factors and treatment options. Absolute numeric risk is provided along the visualization. The third section on the bottom of Figure 4 provides patient education and supporting evidence. Succinct and image-oriented patient education is provided for DDIs and gastrointestinal bleeding. Since risk factors and estimates are evidence based, information is provided on how these aspects were derived. The fourth section is a decision tree with three questions that help structure the conversation and support treatment decision-making. The questions were created and validated by clinicians with the aim to streamline the SDM process in the context of patient care appointments. Question attributes were designed to elicit patient values and preferences associated with bleeding risk and pain treatment. Unobtrusive nudges to reduce risk were used through prepopulated NSAID dosing and default proton pump inhibitor (PPI) selection when an NSAID is chosen. Finally, functionality for generating documentation of the SDM discussion was provided.

This study designed and assessed the usability of a tool for SDM with DDIs (Figure 4). Overall, it appears that SDM can be enhanced by using a tool that displays risks of harm and alternatives. The process of designing DDInteract was rigorous, applying user-centered design principles through iterative prototyping (Figure 2). Target users found DDInteract easy to use and believed it could be useful for supporting SDM (Tables 3 and 4). Given that DDI clinical decision support has been traditionally clinician centric, this study may contribute to a major shift in the way certain medication alerts are developed and used. Through the process designing and evaluating DDInteract with clinicians and patients, lessons were learned regarding patient decision-making and their understanding of the warfarin-NSAID DDI. Furthermore, lessons were learned from clinicians about when in the workflow DDI alerts are addressed and implementing SDM in routine patient care.

Not all DDIs are amenable to SDM, and clinicians should use their judgement before opening a discussion about certain DDIs. Although patients appreciated discussing the warfarin-NSAID DDI with the provider, only the low-risk scenario seemed to be relevant. Certain high-risk DDIs should be avoided without SDM. Additionally, although these patients had been on warfarin for multiple years, knowledge about DDIs and treating pain while taking warfarin was limited. This aligns with what others have found on patient knowledge about anticoagulant therapy [38,39]. Knowledge about which medications are NSAIDs and symptoms of bleeding should not be assumed despite experience with warfarin. Regardless of the decision to avoid an NSAID, patient education about the DDI is needed. Finally, patients mentioned that after previous provider encounters it was difficult to recall information about treatment decisions. Consideration for allowing patients to reference the tool after an encounter may help with comprehension and adhering to decisions; however, DDInteract and other similar tools would need to be adapted to and tested with patients to understand decision-making without provider assistance.

Clinicians had questions about when and how DDInteract would be used. Medication prescribing often occurs at the end of an encounter without the patient. If DDInteract was triggered when an NSAID is ordered, the patient might not be available for discussion. Opportunities to use an SDM tool for DDIs earlier in the workflow may be needed. For example, triggering DDInteract for a patient who is having pain or starting warfarin are additional use cases. Triggering on pain is especially relevant for patients on warfarin due to frequent use of over-the-counter NSAIDs. Although DDInteract was designed to mitigate risk associated with DDIs, clinicians requested decision support for other aspects of anticoagulant therapy, such as deciding to start an anticoagulant or which anticoagulant to use. To achieve broad uptake of SDM for DDIs, the scope of DDInteract may need to be expanded to other decision-making and clinician workflow opportunities.

The next step is to conduct a formative evaluation of DDInteract to understand how it impacts measures of decision-making, satisfaction, and clinician workflow. To maximize dissemination and enable integration with EHR systems, we have developed an interoperable DDInteract app using emerging clinical decision support standards including Clinical Quality Language, Clinical Decision Support Hooks, and SMART on Fast Healthcare Interoperability Resources (FHIR) [40]. A SMART on FHIR prototype of DDInteract is available on Logica Sandbox, and we have successfully implemented DDInteract in an EHR test environment at the University of Utah. Further research is needed to understand how SDM with DDIs can be integrated with overarching decisions surrounding anticoagulant therapy. Finally, research is needed to explore how decision aids in the EHR can be adapted to clinical workflows to enable SDM in routine patient care.

This study describes the design and usability testing of DDInteract. The findings contribute to knowledge about implementing SDM in routine patient care and expand the use of SDM to DDIs. A multidisciplinary design team collaborated with patients, clinicians, and health information technology experts to design a tool that provides a patient-specific risk calculation, elicits patient preferences, and guides both the patient and clinician to a decision. The rigorous design process resulted in a usable and potentially useful tool. Through the design process and usability testing, key lessons were learned from the patient and clinician perspectives. The next step is to evaluate the utility of DDInteract in a clinical setting, and if successful, to implement it across EHRs using interoperable technology.