Safety of Switching From a Vitamin K Antagonist to a Non–Vitamin K Antagonist Oral Anticoagulant in Frail Older Patients With Atrial Fibrillation: Results of the FRAIL-AF Randomized Controlled Trial

The data that support the findings of this study are available from the corresponding author on reasonable request.

From January 10, 2018, through April 25, 2022, a total of 2621 patients were screened for eligibility. Most of these patients were not included because they were considered nonfrail. A total of 1396 patients provided informed consent. In these patients. before randomization, renal function was assessed, and an additional 66 patients were excluded from randomization because of an eGFR <30 mL·min^–1·1.73 m^–2. Thus, a total of 1330 underwent randomization (Figure 1). After randomization, 7 patients (0.5% of the trial population) were excluded from analysis, because 5 patients were in hindsight wrongly registered as having AF by the participating thrombosis service, 1 patient had an eGFR <30 mL·min^–1·1.73 m^–2, and valvular AF was present in 1 patient; these were all a priori defined exclusion criteria for participating in our trial. Thus, the ITT population included 662 patients that switched from a VKA to an NOAC and 661 patients that continued with INR-guided VKA management. This ITT population was used for all further analyses, both for our primary and secondary outcomes. Of note, all ITT analyses were also repeated, including these 7 excluded patients, yielding similar findings (data not shown). Mean age was 83±5.1 years and the median score on the GFI was 4. Other characteristics of these patients, comorbidities, and renal function are presented in Table 1. The median duration from randomization to start with the NOAC in the intervention arm was 52 days (interquartile range, 35–72 days). A total of 22 patients did not switch to an NOAC despite being allocated to switching (3.3%), 57 (8.6%) patients were switched to dabigatran, 332 (50.2%) to rivaroxaban, 115 (17.4%) to apixaban, and 109 (16.5%) to edoxaban. In the remaining 3 patients (0.5%), information on the prescribed NOAC was missing. In patients randomly assigned to switch from a VKA to an NOAC, dosing followed the market-authorized dosing in most patients, except for 44 patients (6.6%) in whom off-label dose reduction occurred. The mean duration of follow-up was 344 days, and 90 patients died during follow-up (44 in the intervention arm [6.6%] and 46 in the control arm [7.0%]). Of those patients that died, a total number of 31 deaths were cardiovascular-related deaths: in the intervention arm, 12 cardiovascular deaths (1.8%; 8 terminal heart failure, 4 fatal myocardial infarction); and in the control arm, 19 cardiovascular deaths (2.9%; 14 terminal heart failure, 5 fatal myocardial infarction). A total of 10 deaths were 5 fetal bleedings in both the intervention (0.8%) and control arm (0.8%). In total, 8 patients were lost to follow-up (3 patients in the VKA group and 5 patients in the NOAC group); in the remaining 1269 patients who did not withdraw consent (99.4%), occurrence of the primary outcomes was ascertained.

In our pragmatic randomized controlled trial among frail older patients with AF, switching INR-guided VKA management to an NOAC-based treatment strategy was associated with a 69% increase in major and CRNM bleeding complications. Event rates for thromboembolic events, major bleeding in isolation, hemorrhagic stroke, or the composite of hemorrhagic and ischemic stroke were low in both treatment arms, preventing us from drawing firm conclusions on these clinically relevant outcomes. There was no clear signal for either a reduced or improved efficacy for these outcomes in patients switching from a VKA to an NOAC.

Our trial strengthens the evidence by studying at the complete domain of frailty (surpassing individual domains) in a large pragmatic trial in older patients with AF, accounting for the downfalls of observational studies such as confounding bias. Even more so, we aimed to extend (ie, “stretch the tails” of) the trial evidence to the most vulnerable (and increasing) AF population, a population that has previously been largely excluded from clinical trials.

To elaborate on this, before our trial, trial evidence on the effect of ageing and frailty on clinical outcomes in NOAC- or VKA-treated individuals with AF was limited to subgroup analyses from either individual or aggregated data from the pivotal 4 NOAC trials.^17–20 However, it is difficult to compare these studies with our trial, given that frail older patients were underrepresented in the 4 NOAC trials, because these patients were either not eligible (eg, due to a high anticipated bleeding risk) or physicians were hesitant to include these vulnerable older patients in clinical trials. Moreover, in these subgroup analyses, apart from the effect of ageing, frailty was predominantly quantified as a cumulative deficit of an increasing number of comorbidities and increasing polypharmacy. Albeit ageing, multimorbidity, and polypharmacy are important drivers of the concept of frailty, this clinical syndrome is broader, including for instance weight loss, communication difficulties, loneliness, dependency on others, cognition, mental condition, and overall physical fitness, all items that are likely related to drug availability in the human body, and thus bleeding and thromboembolic risk. Nevertheless, some interesting comparisons with our findings can be drawn to put our trial into perspective.

First, data from the COMBINE-AF consortium (Collaboration Between Multiple Institutions to Better Investigate Non-Vitamin K Antagonist Oral Anticoagulant Use in Atrial Fibrillation) that pooled individual patient data from all 4 pivotal NOAC trials (n=71 683 patients) revealed that, compared with warfarin, NOAC treatment was associated with a lower risk of major or clinically relevant nonmajor bleeding in patients regardless of age: overall HR for standard-dose NOAC treatment was 0.87 (95% CI, 0.75–1.02) and for reduced-dose NOAC treatment 0.70 (95% CI, 0.59–0.82).^21,22 Although overall effects remained similar, the authors showed that the better efficacy of standard-dose NOAC treatment over VKA treatment was mainly driven by the results in patients who are VKA naive. Moreover, an interaction of ageing on safety outcomes was observed: for standard-dose NOAC treatment, every 10-year increase in age led to a 10.2% increase in HR for major bleeding (P_interaction 0.02), and for reduced-dose NOAC treatment every 10-year increase in age led to a 17.6% increase in the HR for major bleeding (P_interaction 0.01). In addition to these results of the COMBINE-AF study, the ROCKET-AF trial (Rivaroxaban Once Daily Oral Direct Factor Xa Inhibition Compared With Vitamin K Antagonism for Prevention of Stroke and Embolism Trial in Atrial Fibrillation) and the ARISTOTLE trial (Apixaban for Reduction in Stroke and Other Thromboembolic Events in Atrial Fibrillation) both found a statistically significant interaction for the effect of polypharmacy on major bleeding with a waning (and in some analyses a reversed) advantage of NOACs over VKAs on this safety outcome when using more drugs.^22,23 Last, in the ENGAGE AF-TIMI 48 trial (Effective Anticoagulation With Factor Xa Next Generation in Atrial Fibrillation-Thrombolysis in Myocardial Infarction 48), edoxaban was associated with a significant lower rate of bleeding compared with warfarin, at different levels of frailty, except in those at the most severe end of the frailty spectrum. Here, the HR for major bleeding no longer reached statistical significance; the HR for edoxaban 30 mg was 0.74 (95% CI, 0.36–1.52), and the HR for edoxaban 60 mg was 0.60 (95% CI, 0.29–1.26).²⁴ Hence, given that at the current end of the trial tails from the pivotal NOAC trials, a waning (and in some analyses a reversed) advantage of NOACs over VKAs in the oldest and most comorbid trial participants had already been observed, our findings of an increased risk of major or clinically relevant nonmajor bleeding associated with switching VKA treatment to an NOAC in a trial with patients who are even older and more frail may be less unexpected than a priori foreseen.

In addition to this trial evidence, observational studies looked at the effect of ageing and frailty in real-world patients with AF treated with a VKA or an NOAC. With respect to ageing, findings from these observational studies are largely in line with the above-described trial evidence. For instance, a systematic review in 444 281 included older patients with AF found that the HR for hemorrhagic stroke was lower in older patients treated with an NOAC compared with VKAs (HR, 0.61 [95% CI, 0.48–0.79]).²⁵ Similar to what we observed in our trial, the HR for gastrointestinal bleeding was higher in NOAC recipients compared with INR-guided VKA (HR, 1.46 [95% CI, 1.30–1.65]). It is important to note, however, that observational studies exploring the effect of frailty are more scarce and also more difficult to perform given that, in the context of frailty, residual confounding bias remains problematic.²⁶

For full appreciation, a number of topics need to be discussed. First, our population included patients who were tolerant to VKA treatment. Switching from a treatment that most patients tolerate to a newer drug (NOAC) could have resulted in a higher tendency to report bleeding complications in the group that switched. Previous reports using both aggregated or pooled individual patient data from the pivotal NOAC trials also revealed that the efficacy and safety differences favored NOACs over warfarin most strongly in patients with AF that were VKA naive.^2,21 However, including patients that currently use INR-guided VKA was the clinically relevant population for the research question addressed in this trial, which was to study whether these patients, provided they were old and frail, should switch from a VKA to an NOAC. Also, inherent to this switching design, slightly more crossover was observed in the switch-to-NOAC arm of our trial (n=73) than in the continue-with-VKA arm (n=51). Nevertheless, adherence to the protocol was still relatively high, certainly for this older frail population: 89% adherence in the intervention arm versus 92% adherence in the control arm of our trial.

Second, one could postulate that the infrastructure of INR-guided VKA management is adequate in the Netherlands, which may positively affect the time in therapeutic range (TTR) positively in the VKA arm of our trial. Levels of TTR were not an inclusion criterion in our trial nor were individual participants’ TTR levels registered. Monitoring of the INR levels at the 8 study sites in FRAIL-AF was done according to current Dutch clinical practice in this pragmatic trial. The range of TTR levels in Dutch clinical practice for the participating thrombosis services in this trial, specifically for the older individuals that are visited at home for their INR measurements (thus the frailest individuals), during the study years of our FRAIL-AF trial, was between 65.3% and 74.0% (measured as part of yearly quality reports, see https://www.fnt.nl/algemeen/jaarverslagen). As a comparison, the effect of TTR on efficacy and safety of apixaban versus warfarin was studied in the ARISTOTLE trial population and resulted in a TTR from patients recruited from the Netherlands around the median study average of 66.4%, which is similar to countries like the United States, the United Kingdom, Italy, Germany, and Canada, for example.²⁷ At that TTR level, apixaban still was associated with a lower rate of major bleeding, compared with warfarin, in a nonfrail population with a median age of 70 years. Hence, we believe that levels of TTR did not influence our findings significantly or hamper generalizability to the substantial population of older patients living with frailty in many countries, and we consider our findings to be generalizable to patients currently receiving adequate INR-guided VKA management. Our findings should lead to a careful consideration whether or not to switch a patient, who is stable on INR-guided VKA management (TTR ≈70%) to an NOAC, given our finding of a higher risk of major or clinically relevant nonmajor bleeding. Our trial does not allow us to draw conclusions for patients with a low TTR, for whom switching to an NOAC may certainly be considered appropriate.

Third, the choice of the NOAC was at the discretion of treating physicians. Albeit this would mimic (future) clinical practice, it could have affected our results. In observational studies, rivaroxaban (the most prescribed NOAC in our trial) is associated with more bleeding complications than other NOAC types, notably gastrointestinal bleeding, with apixaban having the best safety profile in older individuals.^26,28–30 In our trial, a post hoc analysis per NOAC type showed that rivaroxaban and apixaban had a similar HR for our primary outcome. Nevertheless, because the type of NOAC prescribed was nonrandomized, our trial cannot answer whether one NOAC should be preferred over the other in this frail population.

Fourth, our trial was not powered to show differences in clinical outcomes in isolation such as hemorrhagic stroke. Due to the small number of events, we cannot draw any conclusions on possible differences between treatment arms.

Last, rather than comparing 2 types of anticoagulant molecules, it is important to acknowledge that our open-label pragmatic trial allows us to draw conclusions from the comparison of 2 health care anticoagulation strategies in older patients living with frailty, namely switching from INR-guided VKA therapy to an NOAC or continue with VKAs. This was done deliberately, because it answers the clinically relevant question on whether this particular patient with AF living with frailty should switch from a VKA to an NOAC or not. For this pragmatic clinical question, we decided an open-label design was most appropriate, because this would mimic future clinical care as much as possible. Nevertheless, by design, study procedures were not blinded, and, moreover, some bleeding events in the NOAC group occurred while the patient was (still) taking a VKA, and vice versa. However, the proportion of these bleeding events occurring not on the anticoagulant strategy they were randomly allocated to was small in both treatment arms: 7 of 101 (6.9%) bleeding events in the NOAC arm and 5 of 62 (8.1%) bleeding events in the VKA arm (see Table S1).

In conclusion, our FRAIL-AF pragmatic trial showed that switching INR-guided VKA treatment to an NOAC in frail older patients with nonvalvular AF is associated with more bleeding complications than continuing INR-guided VKA treatment. Albeit our trial was not powered to demonstrate differences in thromboembolic events, major bleeding in isolation, hemorrhagic stroke, or the composite of hemorrhagic and ischemic stroke, there was no clear signal that switching results in reduction of these outcomes in our trial population. Hence, we believe our trial indicates that careful consideration should be applied when choosing between continuing VKA or switching from a VKA to an NOAC in older patients living with frailty.