Atrial fibrillation (AF) and atrial flutter occur in epidemic proportions in the United States, affecting approximately 5 million people [1-5]. As AF is a major risk factor for stroke, professional societies recommend anticoagulation (AC) in most patients; however, some patients are reluctant to start taking it. Limited health literacy, inaccurate perception of the risk of AF, and lack of trust in physicians contribute to patient refusal, with up to 39.3% (257,415/655,000) of patients remaining off AC across the country [6-8]. Among the patients who start AC, many discontinue treatment, especially after the onset of bleeding or other setbacks. Providers struggle to evaluate the risks and benefits of AC [8]. The American Heart Association and other professional societies recommend shared decision-making as a way of arriving at the optimal decision about AC for each patient; however, the feasibility of integrating shared decision-making into routine clinical care is unclear [9-11].

The AFib 2gether mobile app [12] was developed by Pfizer Inc in consultation with a cardiologist (DM) as a potential approach for operationalizing shared decision-making around AC for AF. Specifically, Pfizer Inc convened a hackathon over a 2-day period that included physicians, pharmacists, app developers, and legal and patient education professionals at its headquarters. During this meeting, analysts programmed and improved the app using an iterative design methodology. Using the app, we aim to promote collaboration between patients and providers. Specifically, by using the app, patients first identify their stroke risk factors and later receive a stroke risk score with the projected yearly stroke risk. Patients may then select (from a list of commonly asked examples developed during the hackathon) the questions that they would like their provider to answer at their next visit. The provider can review patient entries on the same app (but with different landing pages based on their role as provider) at the time of each visit to help the patient make an informed decision about AC. The app has not been previously tested with patients and providers for usability, perceived usefulness, frequency of AC starts occurring after visits in which providers and patients use the app, or evidence of shared decision-making. We recently published a protocol to evaluate the usability, perceived usefulness, and feasibility of the app for actual clinical encounters between patients with AF and their cardiology providers [13]. This paper reports the results from the completed study and their interpretation.

We previously published a protocol for integrating the AFib 2gether app in encounters between patients and their cardiology providers [13]. We briefly summarize the methods below. The sponsor of this work, Pfizer Inc, reviewed and edited the drafts of our study design and manuscript interpreting the results.

The study included patients with AF and with a CHA₂DS₂-VASc score ≥2 who were not currently on AC and were visiting cardiology providers at the University of Massachusetts Memorial Medical Center. The CHA₂DS₂-VASc score assigns 1 point for congestive heart failure, hypertension, age 65-74 years, diabetes, vascular disease history such as myocardial infarction, and female sex and 2 points for age ≥75 years or prior stroke or transient ischemic attack.

We obtained consent from the cardiology providers. After enrollment, each cardiology provider completed questionnaires regarding their knowledge and confidence in AF management. Specifically, we asked providers to report their confidence in using information about stroke risk and bleeding risk to determine appropriate antithrombotic therapy and familiarity with the American College of Cardiology/American Heart Association/Heart Rhythm Society (ACC/AHA/HRS) guidelines for AF management. We also asked 2 questions requiring providers to accurately calculate the CHA₂DS₂-VASc score and make a decision to prescribe AC based on it for a patient presented in a clinical vignette. Next, we moved to recruit the patients of these providers by mailing introductory letters with a fact sheet to facilitate verbal consent. We then telephoned the patients to obtain their consent and enroll them. During the enrollment telephone call, we asked patients to download the app if they had a smartphone and provided a brief orientation (lasting 2-3 minutes, although we did not record the exact duration). For those without a smartphone, we provided a study smartphone when the patient arrived for their in-person visit with their cardiology provider.

After the app orientation, patients answered questions in the app to determine their stroke risk score based on the CHA₂DS₂-VASc risk score. Once the participants answered the questions, the app displayed their stroke risk factors and allowed them to select questions from a list of commonly asked questions that they wanted to ask their providers during their visits based on this score. The app then sent the patients' completed inventory of stroke risk factors and questions for discussion to their providers to review. The participants then completed the Mobile App Rating Scale (MARS) questionnaire [14]. The providers then reviewed their patients’ stroke risk answers and compared them to what they knew about the patients’ risk. After the providers made any corrections to the stroke risk factors, the patient scores were made available to the patients and providers to review and discuss.

At the visit, we reminded the providers to review the app information completed by the patients (if they had not done so before the start of the visit). We then audio recorded the conversation for those who had in-person visits (ie, those who were seen before restrictions were put in place during the COVID-19 pandemic of 2019-2020). At the conclusion of the visit, both the patients and providers completed a 3-question survey on the perceived usefulness of the app.

We transcribed the audio recordings and reviewed them for evidence of shared decision-making using a prespecified list of items modified from an established instrument for measuring shared decision-making (ie, Observing Patient Involvement in Decision Making Scale) [15] as well as novel items specifically related to AF management (Table 1).

We collected demographic information and information on comorbid conditions from electronic capture of patient information from our institution’s electronic health record (EHR). Specifically, we worked with an experienced EHR analyst who interrogated the Clarity database associated with Epic Systems EHR to identify demographic information. Through the use of ICD codes that we previously validated, the analyst was also able to identify the presence of comorbid conditions [3,16]. From chart reviews, we also captured the reason why a patient was not on AC. In this review, we grouped patients into the following categories: low AF burden, refused AC, fall risk, gastrointestinal bleeding, other bleedings, or unspecified reasons. During our review, we also captured the number of years elapsed since AF onset. For providers, we administered a provider knowledge survey, as described in the protocol.

For each domain of the usability of the app or the MARS items, we calculated the mean and SD. By contrast, for perceived usefulness, we grouped patients into 3 categories, combining strongly agree with agree and strongly disagree with disagree. For the start of AC, we tabulated the frequency and calculated the simple percentage of the number of starts divided by the total number of patients.

We received approval for our study from the University of Massachusetts Chan Medical School Institutional Review Board. We conducted our study ethically by obtaining informed consent, with no change in treatment for patients who declined to participate.

We sent letters to or approached in person a total of 165 eligible patients. From this pool of 165 patients, we conducted visits using our app with 37 (22.4%) patients who were seeing 13 cardiology providers. Of the 128 patients who were not included, we were unable to reach 10 (7.8%). Another 23.4% (30/128) declined, with not being interested as the most common reason, followed by not feeling well enough to participate as the next most common reason. For the remaining 68.8% (88/128) of patients, no attempt was made, as the study staff were not available or did not need to recruit additional patients at the time of the patient’s visit. Nearly all patients (36/37, 97%) were White, and approximately half of the enrolled participants were aged ≥75 years. This age distribution followed the typical epidemiology of AF, including those in our previous studies [3,16,19]. Most patients were men (26/37, 70%) and had a CHA₂DS₂-VASc score between 2 and 4 (28/37, 76%). A chart review of the reasons why patients were not on AC before their appointment revealed that, of the 37 patients, 16 (43%) patients had a low AF burden, whereas 10 (27%) patients refused without further explanation clearly documented. In terms of the length of AF diagnosis, 41% (15/37) of patients had a history of AF in the past 1 to 5 years (Table 2). All patients entered enough information in the app to compute their CHA₂DS₂-VASc scores. Of the 37 patients, 21 (57%) asked at least one question. Of the 37 patients, 21 (57%) did not have their own smartphone and therefore required the use of our study phone.

Most enrolled providers had been in practice for ≥10 years. Of the 13 providers, 9 (69%) were physicians, 2 (15%) were physician assistants, and 2 (15%) were nurse practitioners. The providers’ self-rated confidence was high in our sample. More specifically, 85% (11/13) of providers were very confident in assessing antithrombotic therapy for patients who had stroke and selecting the appropriate AC, and 77% (10/13) felt very confident in using oral AC therapies for reducing stroke risk. The HAS-BLED bleeding risk score assigns 1 point for hypertension, abnormal renal function, abnormal liver function, tendency for bleeding, labile international normalized ratio, age >65 years, history of alcohol or drug usage, and medication usage predisposing to bleeding. Confidence in using the HAS-BLED bleeding risk calculator was more variable. Only 31% (4/13) of providers felt very confident in this skill. For a 2-part knowledge inquiry based on a clinical vignette presented, 77% (10/13) of providers accurately calculated a CHA₂DS₂-VASc score and based anticoagulant decision-making on it. Of the 13 providers, 7 (54%) providers reported that <25% of their patients had a diagnosis of AF, whereas 6 (46%) reported that >25% had a diagnosis of AF (Table 3).

Patients rated the AFib 2gether app highly on all domains of usability. The MARS combined average functionality score for patients was 4.51 (SD 0.61) out of a possible score of 5. Patients also rated the app highly in the MARS esthetics category with a score of 4.26 (SD 0.51) out of 5. Patients rated the app usability highly; the mean MARS star usability rating was 4.24 (SD 0.89) out of a possible 5. There was no sizable difference in the rating of patients who asked questions and those who did not (overall rating 4.30 vs 4.17).

Patients also gave high perceived usefulness ratings. Specifically, 40% (15/37) of patients agreed with the statement “the app improved my knowledge regarding anticoagulation.” Similarly, 62% (23/37) of patients agreed with the statement “the app helped me clarify my provider preferences regarding anticoagulation” (Table 4).

Provider ratings for usability were also high. Out of a maximum of 5, the mean of providers’ ratings for functionality was 4.19 (SD 0.50), for esthetics was 4.04 (SD 0.50), and for overall star-based quality was 3.76 (SD 0.44).

Providers also reported high perceived usefulness. Specifically, 79% (27/34) somewhat or strongly agreed that the app helped clarify the preferences of their patients, and 82% (28/34) agreed that the app saved them time. Slightly fewer providers (20/34, 59%) agreed that the app helped their patients make a decision about AC (Table 1).

Approximately 32% (12/37) of patients started AC after their appointment. This included 56% (9/16) of patients who were previously not on AC because of the low AF burden. Of the 10 patients who had previously refused to be on AC, 2 (20%) started AC after the visit. Of the 5 patients without a specified reason for not being on AC before the visit, 1 (20%) started AC after it.

We were able to collect audio recordings from the first 68% (25/37) of patients we recruited. After that point, our institution restricted in-person recruitment to limit the spread of COVID-19 in 2019. From the available encounters, we noted that AC for AF was mentioned 84% (21/25) of the time. We also noted the discussion of multiple options of AC in 44% (11/25) of patient encounters. In 72% (18/25) of the encounters, the provider made a recommendation regarding AC for the patient. The recommendations included whether the provider believed that the patient should be anticoagulated, as well as which anticoagulant the provider believed would be best for the patient. We identified that in 48% (12/25) of the patient encounters, there was evidence of patient involvement in the discussion (Table 5).

Most patients in this study gave high usability ratings to our shared decision-making app across 3 separate domains. They also reported that the app helped clarify their preferences to their providers and improved their knowledge of AC. One-third of these patients started AC after their appointment with their providers. Approximately half of the patients demonstrated involvement in AC decision-making.

To understand the importance of our findings, we compared them with findings of other intervention studies in the AC management of AF. Man-Son-Hing et al [20] developed a decision aid based on a risk stratification scheme that helped patients with AF and their providers make informed decisions about whether to use warfarin or aspirin. They then tested this in a randomized trial. They found that patients in the group assigned to the decision aid were more able to make definite choices regarding antithrombotic therapy (99% vs 94%; P=.02). More recently, Kunneman et al [21] tested a shared decision-making tool that provided individualized risk estimates of stroke in various anticoagulant treatment options. Although they did not find a significant effect on treatment decisions, more clinicians were satisfied with the encounter in the intervention arm compared with the standard arm. Neither of the 2 studies specifically studied the usability or perceived usefulness of their shared decision-making tool.

Our app compared favorably with the ratings published for other health mobile technology tools evaluated using MARS. In particular, authors of a systematic review documented an average score (averaged across the same 3 domains of usability that we analyzed) ranging from 2.40 to 2.63 for blood pressure apps deployed on smartphone platforms [22]. We found that our app performed significantly better than those included in the review of blood pressure apps, although our particular cohort of patients did not evaluate other apps’ limiting comparisons.

Our findings have several implications. Our app appears to be usable by both patients and providers. Moreover, 32% (12/37) of patients started AC after having used the app for their clinical visit. To better understand the effectiveness of the app, we will require a controlled study, given that the participants who agreed to participate in our study may have been more educated or activated than typical patients in routine clinical practice. In terms of other implications, we also found that there was a moderately high level of patient involvement as measured through audio recordings. However, some elements of shared decision-making occurred infrequently. Further enhancements of the app, for example, prompts to encourage the use of more shared decision-making elements, may stimulate even greater encouragement for patient involvement. Other enhancements to the app may also better prepare patients to participate in AC discussions. Of the 37 patients, 16 (43%) were not on AC because of the isolated and low burden of AF. However, the current ACC/AHA/HRS guidelines for AC do not take the AF burden into account when recommending treatment [10]. Further clarification of guidelines by its authors for prescribing AC in the setting of isolated and low burden of AF and education of providers may help overcome some of the gaps in AC use beyond patient hesitancy or refusal.

There are a number of limitations to this study. One limitation is the absence of a control group. Nevertheless, we are not aware of other single-arm studies that demonstrated a 32% increase in AC with a single-encounter intervention. Expanding our testing to other centers and including a control group would provide better information on the potential benefits of the app. Another limitation of our work is that we were only able to record encounters for 68% (25/37) of patients, given the interruption in our study caused by the restrictions against in-person recruitment at the time of the COVID-19 pandemic. Despite this limitation, we demonstrated a moderately high level of patient involvement in the visits that we recorded. Another limitation was the absence of custom information for patients with discrete reasons, such as bleeding, for not being on AC at the time of the visit. Future iterations of the app may want to include custom information. Finally, our population was heterogeneous in terms of the reason for not being on AC, potentially introducing bias into the measurements we recorded.

In conclusion, patients and providers found the AFib 2gether app usable, and there was a high level of perceived usefulness that facilitated an informed discussion with the provider, leading to increased guideline-based AC management. We await further testing at other centers and in a controlled study design to assess the potential benefit of the app and its ability to increase AC use, consistent with prevailing professional society guidelines designed to prevent stroke in patients with AF.