The ReInforcement of adherence Via self-monitoring app Orchestrating biosignals and medication of RivaroXaban in patients with Atrial Fibrillation and co-morbidities (RIVOX-AF) study is a prospective, randomized, open-label, nationwide, multicenter trial to evaluate the efficacy of smartphone apps (MEDI-app) in improving drug adherence to NOACs in patients with AF. Thirteen tertiary university hospitals in South Korea will participate in this study. Enrollment began in March 2022, is currently ongoing, and is expected to be completed in late 2023. The study design has been registered at ClinicalTrials.gov (NCT05557123).

We will enroll patients with AF aged 19 years or older with one or more comorbidities, including heart failure, myocardial infarction, stable angina, hypertension, or diabetes mellitus. Patients can be enrolled 3 months after myocardial infarction or percutaneous coronary intervention. Rivaroxaban (Rivoxaban, Samjin Pharm, Seoul, Korea) will be administered to the participants at an open-label dosage of 20 mg once daily. Patients using other NOACs can be enrolled after swiching to Rivoxaban. The dosage will be reduced to 15 mg or 10 mg once daily if creatinine clearance is 15–49 ml/min or at the physician's discretion. Since smartphone usage is essential in this study, participants should be proficient in the use of smartphones and be able to understand and follow Korean instructions for using the program. Patients with severe chronic kidney disease (creatinine clearance <15 ml/min), moderate or severe mitral valve stenosis, or a history of mitral valve replacement or repair will be excluded. The detailed inclusion and exclusion criteria are listed in Table 1.

At baseline, sex, age, and comorbidities will be evaluated in all participants. The CHA₂DS₂-VASc score for stroke risk prediction will be calculated by the summation of all assigned points: one point each for congestive heart failure (C), hypertension (H), age between 65 and 74 years (A), diabetes mellitus (D), vascular disease (V), and female sex (Sc) and 2 points each for a history of stroke/transient ischemic attack/thromboembolism (S₂) or age ≥75 years (A₂) (1, 33).

Any patient who comes to the clinic for AF management will be considered a potential participant. After a comprehensive interview, eligible participants will be asked to provide written consent for participation in the trial. A baseline survey covering demographics and cardiovascular comorbidities will be given to eligible participants (Table 2). These data will be collected by a research nurse, who will re-interview the patient and refer to the doctor's notes. Then, using a web-based central randomization service (http://matrixmdr.com), eligible participants will then be randomly assigned to either the intervention group (MEDI-app) or the conventional treatment group in a 1:1 ratio. It will not be possible to blind the participants to group allocation owing to the nature of the intervention.

We developed a mobile application (MEDI-app) to improve NOAC adherence in patients with AF. Participants in the intervention group will install the study app provided by a research nurse on their smartphones (Figure 1). Because the app is not an open application available to the public, the control group cannot access the application. The workflow, operating system, and display of the smartphone application are shown in Figures 1, 2. The patient can use facial recognition to log in to the app (Supplementary Figure S1). The design, development, and utilization details of the physician-oriented web and patient-oriented smartphone app are described in the online Supplementary material.

The research coordinator will enter the information on rivaroxaban administered to the patient into the mobile application and will set the alarm initially. The participants will be able to change the time of the alarm for rivaroxaban. When the alarm sounds at a pre-specified time, the patient will be reminded to take rivaroxaban. The app also determines whether the patient takes the medication through medication and action recognition software (Supplementary Figure S2). To elaborate, the first step is medication recognition, which determines whether the patient is taking the correct pill. A deep learning model of medication recognition will be exported after the completion of the training phase to perform real-time inference on the server from a picture captured by the camera on the patient's smartphone to match the name, shape, and color of the pill. In the second step, the app will visually confirm if the patient has taken the pill through action recognition via the smartphone camera by watching the patient put the medicine into their mouth through the camera.

If the patients do not confirm taking the medication at the scheduled time, the app will send an alert message to ensure that the patients take the medicine until 2 h after the set drug intake time. If the patient does not comply with the pill intake procedure within these 2 h, it is assumed that the patient did not take the pill, and the app will send non-intake data to the server accordingly. As an alternative, patients can also confirm that they have taken the medicine by pressing the intake button within a day, rather than using action recognition via the camera. This information can be updated on the server. Data regarding the time of rivaroxaban intake will be stored daily, and patients can check the list of drug adherence per week or month on a calendar through the app. The dashboard system is designed to show the physicians the history and number of days that the participants confirmed taking their medication (Supplementary Figure S3). In this study, the app will be initially configured only for the rivaroxaban dose, frequency, and pill supply.

The control group will receive only standard care which is recommended in the guidelines, including education on the disease and the importance of medication compliance by physicians at each scheduled visit. No other specific interventions will be performed. Follow-up visits will be scheduled at 12 (visit two) and 24 (visit three) weeks after randomization (Figure 3).

The primary endpoint of the study is adherence to rivaroxaban at 12 and 24 weeks. Drug adherence will be evaluated with the “pill count” measurements. The patients brought the remaining tablets to each scheduled visit, and trained and certified research nurses counted the number of returned drugs and calculated drug adherence as follows:Drugadherence=numberofpillsdispensed−numberofpillsreturnednumberofdaysbetweendispensingdateandfollow−updateIn addition, we will evaluate the concordance of drug adherence by pill count measurements and app-calculated indices.

The key secondary endpoints are clinical composite endpoints, including systemic embolic events, stroke, major bleeding requiring transfusion or hospitalization, or death during the 24 weeks of follow-up. Other secondary endpoints include (1) major adverse cardiac and cerebrovascular events and hospitalization, (2) thromboembolism (stroke, transient ischemic attack, pulmonary embolism), (3) major bleeding requiring hospitalization or transfusion, or (4) minor bleeding. To evaluate the feasibility of the smartphone app, additional analysis will be conducted regarding the frequency of smartphone app use.

To date, no large studies have evaluated the use of a smartphone app to improve adherence to NOACs using pill count measurements. Thus, precise sample size calculation is not possible. We assumed that drug adherence would be 91% in the control group and 95% in the intervention group with a standard deviation of 17%, considering previous studies (30, 34, 35). With a two-tailed alpha of 0.05, power of 0.95, and a dropout rate of 10%, 1,042 patients (521 patients in the intervention and 521 patients in the control groups) are required.

Categorical variables will be reported as frequencies (percentages), and continuous variables will be expressed as means ± standard deviations or medians with interquartile ranges. Categorical variables will be compared using Pearson's chi-square test or Fisher's exact test, and continuous variables will be compared using Student's t-test or the Mann–Whitney U test.

Intention-to-treat analyses will include all randomized patients. Efficacy endpoints will be mainly analyzed using the full analysis set, which includes all randomly assigned participants who underwent at least one assessment of the primary endpoint. We will also perform a per-protocol sensitivity analysis, including all patients who completed the study protocol. In the subgroup analyses, we will evaluate the differential effects of the intervention on the primary outcomes with respect to sex, age and other comorbidities.

All tests will be two-tailed, and a P-value <0.05 will be considered statistically significant. Statistical analyses will be performed using R version 4.2.0 (The R Foundation for Statistical Computing, Vienna, Austria).

This study has several limitations. First, the results of our trial may not be applicable to patients who are not capable of using smartphones, such as the elderly, because we will only enroll participants who have smartphones and are able to use them. As a large proportion of patients with AF are elderly and have difficulties using smartphones, excluding them from participation may lead to bias and study limitations. Second, the participants who consent to participate in the study may be more concerned with their own health. As a result, their medication adherence may be greater than that of the general AF population. Third, our sample size calculation may not be accurate because no prior large-scale trial has evaluated the use of smartphone apps to improve NOAC adherence. It is also possible that our estimates of drug adherence may overestimate the actual adherence, considering the findings of a previous study (17). In addition, the sample size has been calculated to determine differences in drug adherence; it is not powered to reveal any differences in adverse clinical events. Fourth, the follow-up period of this study was limited to 24 weeks. Although longer follow-up periods may provide a more reliable data on adherence, the follow-up period in our study was decided on the basis of manpower and cost.

Despite these limitations, this study is a nationwide, multicenter randomized control trial with a relatively large number of participants compared with previous studies (30, 31). Additionally, in our study, drug adherence will be evaluated by the “pill count” method, in which the number of returned drugs will be counted. Drug adherence by pill count may be more accurate than the drug score (37, 38), and this is a strength of our study compared with previous studies. In addition, we will evaluate the concordance of drug adherence by pill count measurements or app-calculated indices.