Intake reminders are effective in enhancing adherence to direct oral anticoagulants in stroke patients: a randomised cross-over trial (MAAESTRO study)

The MAAESTRO study (electronic Monitoring and improvement of Adherence to direct oral Anticoagulant treatment—a randomized crossover study of an Educational and reminder-based intervention in ischemic STROke patients under polypharmacy) [15] included patients who (i) were hospitalised in the University Hospital of Basel, Switzerland, for ischaemic stroke, (ii) were diagnosed with AF, (iii) were prescribed a DOAC and at least two other medicines, and (iv) self-administered their medication. Recruitment of participants was performed during hospitalisation (visit 0). Participants electronically monitored their daily DOAC intakes for 12 months following hospital discharge using the Time4Med™ Smart card (Adherence Innovations, Hong Kong) [16]. By pushing the button of the Smart card, time stamps are recorded on a built-in microchip. The first 6 months after hospital discharge were observational. A staggered intervention was delivered. First, during the follow-up visit at 6 months (visit 1), a counselling session and a pillbox were provided to all patients. Then and according to randomisation, patients obtained an additional reminder device that was used beginning either at visit 1 (group 1) or 3 months later (group 2, Fig. 1), for a total of 3 months in each group. Study investigators were not involved in DOAC prescription, which was up to the hospital physician, general practitioners and other healthcare providers as standard of care.

Clinical and neuroimaging parameters (Fazekas score for white matter lesions on neuroimaging [17], National Institutes of Health Stroke Scale (NIHSS) for stroke severity [18], and Montreal Cognitive Assessment (MoCA) scores) [19] and self-reported variables (satisfaction with reminder) were assessed at each visit, as described in the study protocol [15].

The reminder was the main intervention in this trial. It generates acoustic and visual alarms at time points chosen by the patient. In addition, all patients received a pillbox and a counselling session with a pharmacist during which they were shown a visualisation of their intake data from the observational phase (Fig. 2).

The primary outcome was 100%-timing adherence, defined as no dose taken outside the grace interval. The grace interval is defined as allowing an intake ± 6 h (for once-daily DOAC) and ± 3 h (for twice-daily DOAC) around the median intake time. We hypothesised that when patients use a reminder, they will reach 100%-timing adherence more often than without a reminder.

The secondary outcomes were:

We used secuTrial® (version 6.3.2.7) to capture electronic case report forms. The adherence monitoring data were cleaned, and metrics were calculated with Microsoft Excel® 2016 according to an internal standard operating procedure [20, 21]. Patients without sufficient electronic monitoring data records (less than one-third of the expected records was available) were excluded from the analysis.

The sample size calculation was based on the following assumptions, informed by our previous experience with this patient population [15, 22]: (i) a 20% adherence-enhancing effect of the reminder on 100%-timing adherence; and (ii) a 45% rate of timing adherence below 100% in the control group. The required sample size to achieve a power of 0.9 at an alpha error of 0.05 was 114 participants. In addition, a 12% drop-out rate was estimated. We therefore aimed to recruit 130 participants. Patients were randomly allocated to group 1 or group 2 in a 1:1 ratio. Randomisation was performed by a central computer during visit 1 over the web-based secuTrial® system with minimisation for sex, age, NIHSS and prior pillbox use at visit 0.

The predefined statistical analysis plan dictated the use of mixed logistic models to assess the effect of reminders on timing and taking adherence, with reminder use ("with reminder" vs "without reminder") as fixed and subject identifier as a random effect. Due to the prohibitively small number of participants reaching the prespecified primary outcome of 100%-timing adherence, we analysed the effect of reminders on the primary outcome using an exact McNemar’s test for paired nominal data instead and explored the alternative thresholds of 90% and 80%-timing and taking adherence using the planned mixed-effects analysis. For all analyses, we report two-sided p-values and present model-based odds ratio (OR) estimates with 95%-confidence intervals (CI).

Subgroup analyses were done by (i) DOAC type (once-daily vs twice daily), (ii) stroke location (right vs left hemisphere), (iii) Fazekas score (0–3) and (iv) MoCA score (0–30). Analyses were done on an intention-to-treat basis as prespecified in the statistical plan and were repeated in the per-protocol population (including only patients who used the reminder according to randomisation). We investigated time-trend and carry-over effects in sensitivity analyses. The effect of the counselling session and pillbox was assessed by comparing overall adherence during the observational phase and the first half of the interventional phase without the reminder (group 2). We present clinical events of interest and patient satisfaction with the reminder descriptively. All analyses were performed in R (version 4.1.2). Study outcomes were reported according to the CONSORT guideline [23].

The MAAESTRO study was approved by the Ethics Committee of Northwest/Central Switzerland (EKNZ 2017–01552) and registered at ClinicalTrials.gov (NCT03344146). All participants gave written informed consent; study participation was voluntary. For patients initially consenting to study participation but withdrawing their consent during the course of the study, the use of their study data up to the timepoint of consent withdrawal was allowed according to the signed informed consent. To accelerate the initiation of the interventional study phase, the study protocol was amended on April 28, 2020, to shorten the observational phase from 6 to 3 months for participants who entered the study after April 28, 2020.

From January 3, 2018, to March 14, 2022, 130 stroke patients were recruited, of whom 88 were randomised and 84 were eligible for analysis (Fig. 3). Mean [SD] age was 76.5 [9.1] years and 61% were male. Overall, stroke severity was mild (median [IQR] NIHSS 1 [0–2]). A twice-daily DOAC was prescribed in 74% of the patients and a once-daily DOAC in 26%. Participant characteristics were well-balanced between randomisation groups (Table 1; Supplement A).

A 100%-timing adherence was observed in 10 patients (12%) who were using the reminder, and in zero patients without the reminder (p = 0.002). A 100%-taking adherence was observed in 10 patients (12%) who were using the reminder, and in 2 patients without (p = 0.039; Table 2). Patients reached 90%-timing adherence (OR 2.65; 95% CI 1.05–6.69; p = 0.039) and 80%-timing adherence (OR 25.9; 95% CI 1.51–444; p = 0.025) significantly more often with a reminder than without. With the reminder, the participants' overall timing adherence increased by 4% (median [IQR] 92% [81–97%] without reminders, 96% [85–99%] with reminders, Fig. 4). The odds of taking a DOAC dose on time increased by 70% with the use of a reminder (OR 1.7; 95% CI 1.55–1.86; p < 0.01).

The use of the reminder increased the overall amount of DOAC doses taken, with significantly more patients reaching 90%-taking adherence with the reminder than without (OR 3.06; 95% CI 1.20–7.80; p = 0.019). This also applied for 80%-taking adherence (OR 27.4; 95% CI 1.56–480; p = 0.024). The overall taking adherence increased from a median [IQR] of 94% [83–98%] without reminders to 97% [90–99%] with reminders (Fig. 4). The odds of not missing a DOAC dose increased by 67% with the use of the reminder (Supplement B).

The adherence-enhancing effect of reminders on 90%-timing adherence did not differ in any of the predefined subgroups (Supplement C). The exclusion of 18 patients due to participants not using the reminder (9 participants per randomisation group) resulted in 66 participants eligible for the per-protocol analysis. Consistent with the intention-to-treat analysis, the effect of the reminder on 90%-timing adherence was preserved by the per-protocol analysis (OR 2.53; 95% CI 0.86–7.41; p = 0.091). No carry-over effect (p = 0.950) or time-trend (p = 0.366) was observed between both interventional phases. There was no evidence that the delivery of a counselling session plus pillbox affected the number of patients reaching 90%-timing adherence (Supplement C).

As a higher number of patients dropped out before randomisation, the evaluable sample in MAAESTRO fell short of the planned 114 full patient sets, despite the recruitment of the planned 130 patients. As a post hoc analysis, we calculated scenarios if our main outcome would have changed if 114 participants could have been included in the analysis. In the most likely scenario, our main outcome would remain unchanged (Supplement D.1).

A total of 21 clinical events of interest occurred throughout the study in 16 patients: recurrent ischaemic stroke (8), major extracranial haemorrhage (5), intracranial haemorrhage (4), myocardial infarction (2) and death (2). Fourteen clinical events occurred during the observational study phase (67%), five during the interventional phase without the reminder (24%), and two during the interventional phase with reminder (9%, see also Supplement D.2 and D.3).

A trend towards lower overall adherence was observed in patients with ischaemic events compared to patients without events (Table 3).

The satisfaction with the reminder differed considerably among participants (Fig. 5). About half of the participants found the reminder impractical and denied benefitting from it. Regarding the integration of the reminder into their daily routine, 81% of the participants felt no restriction (Supplement E).

This study has several strengths. First, the cross-over design of our study enabled each patient to serve as their own control. When investigating individual behaviours, this design delivers more robust results than inter-individual comparisons because the risk of confounders is negligible. We observed no time trends or carry-over effects between the cross-over phases, which supports the appropriateness of our design. Second, our study population was a well-described and homogeneous sample of patients treated with DOAC for secondary prevention due to a recent ischaemic stroke attributable to atrial fibrillation. Reminders increased adherence in this specific sample and should therefore be offered to these patients after suffering a stroke. Third, we used electronic monitoring to assess adherence, which is considered the gold standard when investigating the implementation of a medicine. The Smart cards delivered accurate information on intake times, which was crucial to judging the impact of reminders on timely DOAC intakes.

There were also limitations to this study. First, the primary outcome (100%-timing adherence) proved to be overambitious. When transposed into practice, this threshold means that one single delayed dose in 3 months suffices to disqualify for 100%-timing adherence. Thus, the use of a theory-based approach to developing our study yielded an unrealisable goal for individuals with some active life. It is likely that applying realist research would have provided a different primary outcome. In any case, evidence on a more appropriate threshold for timing adherence to DOAC was limited in the literature. Hence, the alternative outcome of 90%-timing adherence that we used in our secondary analysis is exploratory in nature. Second, the drop-out rate of 36% was considerably higher than the 12% that were estimated from the literature. This number must be put in light of the evolving character of a recent stroke. Drop-outs occurred mainly during the first weeks following study inclusion and concerned older participants with poorer cognitive and functional outcomes. Although this selection bias is difficult to prevent, it reduces the generalizability of our results. We assume that the recruitment timing was inadequate, as patients had just experienced a stroke, which can be considered a rather disturbing situation. Some patients might have overestimated their ability to pursue the study procedures in view of the unusual circumstances. In addition, some inclusion criteria for participation (e.g. self-administration of medication) were subject to changes after discharge, which increased the drop-out rate. However, the calculation of ‘worst case’ scenarios showed no change in the main outcome if the planned patient number would have completed the trial. Third, we experienced technical issues with the electronic devices. Participants regularly reported Smart cards that stopped beeping when registering the DOAC intake. We systematically documented these defects and excluded these periods from the calculation. Nevertheless, we cannot exclude that some adherence gaps were in reality device malfunctions.