Lina Palaiodimou, MD
The ARISTOPHANES study is a large retrospective observational study with real-world data pooled from 5 data sources in the United States, in order to compare stroke/systemic embolism (SE) and major bleeding (MB) among nonvalvular atrial fibrillation patients treated with either non-vitamin K antagonist oral anticoagulants (NOACs) or warfarin.
The researchers of this study collected data about a total of 321182 patients with a diagnosis of AF, who had an active medical and pharmacy health plan enrollment and were prescribed treatment with either NOAC or warfarin according to pharmacy claims. Exclusion criteria were the following: evidence of valvular heart disease, venous thromboembolism, previous oral anticoagulant treatment, transient AF due to treatable causes (pericarditis, hyperthyroidism, thyrotoxicity), heart valve replacement or transplant, pregnancy and recent surgical operations. Demographic data, Charlson Comorbidity Index score, baseline bleeding and stroke/SE history, comorbidities, baseline comedication and dose of NOAC were recorded about all patients. In contrast, reason for lose-dose NOAC prescription, creatinine clearance, international normalized ratio (INR) measurements in warfarin-treatment group and patient adherence data were not available. The outcome measures were time to stroke (either ischemic or hemorrhagic) or systemic embolism and time to major bleeding, either gastrointestinal or intracranial or at other key sites (eyes, pericardium, urinary tract, joints). Identification of these events were based just on hospitalization incidence with stroke/SE or MB as the principal diagnosis according to International Classification of Diseases, Ninth Revision (ICD-9). Mortality due to all-causes was also evaluated, but just for the patients enlisted in only one out of five data sources. Patients were followed up each for a different period, according to drug discontinuation date, switch to another drug date, death, end of medical and pharmacy health plan enrollment, or end of study period, whichever occurred first. In the conducted sensitivity analysis, the follow-up period was restricted to 1 year, to better achieve balance between the cohorts.
The total patient group was quite heterogeneous. Warfarin patients were older and had the highest baseline CHA2DS2-VASc and HAS-BLED scores, followed by apixaban, rivaroxaban and dabigatran patients. The unadjusted incidence rates of stroke/SE (stroke was defined as ischemic as well as hemorrhagic stroke) was 1.3, 1.5, 1.4 and 2.2 per 100 person-years for apixaban, dabigatran, rivaroxaban and warfarin, respectively. Additionally, the unadjusted incidence rate of major bleeding was 3.5, 3.5, 5.3 and 6.3 per 100 person-years for apixaban, dabigatran, rivaroxaban and warfarin, respectively. But for a direct comparison between NOACs and warfarin, as well as among different NOACs, to be achieved, one-to-one propensity score matching was conducted according to multiple covariates. A total of 285292 patients were included, and six matched cohorts were generated: apixaban versus warfarin, dabigatran vs warfarin, rivaroxaban vs warfarin, apixaban vs dabigatran, apixaban vs rivaroxaban, dabigatran vs rivaroxaban. Cox models were used to evaluate the risk of stroke/SE and MB across matched cohorts.
The ARISTOPHANES study demonstrated the following results, regarding the comparison between NOACs and warfarin. Apixaban was superior to warfarin in preventing stroke/SE and in reducing the risk of MB and intracranial hemorrhage (similar to the results of the ARISTOTLE trial). Dabigatran was associated with a lower risk of both stroke/SE and MB compared with warfarin. These findings were consistent with the RE-LY trial, with the exception of MB rates of dabigatran 150mg dose that were comparable with warfarin. Rivaroxaban was associated with lower risk of stroke/SE compared with warfarin, but it had statistically significant higher bleeding rates, although ROCKET-AF trial had demonstrated non-inferiority of rivaroxaban versus warfarin for both stroke/SE and MB.
Given the fact that there are no randomized controlled studies comparing head-to-head different NOACs, indirect comparisons generated from real-world data should be carefully evaluated. In the ARISTOPHANES study, apixaban demonstrated a lower rate of stroke/SE and MB compared to dabigatran and rivaroxaban, while dabigatran was associated with similar risk of stroke/SE and a lower risk of MB and intracranial hemorrhage compared to rivaroxaban.
Subgroup analyses was also conducted, according to age, sex, baseline CHA2DS2-VASc and HAS-BLED scores, congestive heart failure, coronary artery disease, peripheral arterial disease, diabetes mellitus, renal disease, prior stroke/SE and lower dose of NOACs. Some interactions, such as even lower rates of stroke/SE for apixaban versus warfarin in older patients, were noticed. Additionally, after propensity score matching between patients receiving the standard dose of NOAC versus those with the lower dose, the same associations were partly reproduced.
When evaluating mortality, the data for which were drawn from only one out of five data sources, it was indicated that all NOACs were associated with lower rates of mortality compared with warfarin, whereas apixaban was associated with lower rates of mortality compared to dabigatran and rivaroxaban.
Several limitations exist in the ARISTOPHANES study and are well stated by the authors. It must be emphasized, though, that it is an observational retrospective study of patients that are quite heterogeneous. Even though propensity score matching was done, there may be several cofounders that lacked inclusion in the logistic regression. Despite, real-world data have an important role to play in the evaluation of safety and effectiveness of oral anticoagulants, caution is needed in case of result interpretation. Such data should be used as additional to the results of large retrospective controlled trials, as in the case of comparison between NOACs and warfarin. Regarding comparisons among NOACs, the resulting conclusion that apixaban is more effective and safer than dabigatran and rivaroxaban seems rather simplistic and should be used for hypothesis generating only. The most important methodological shortcoming was related to the fact that the majority of co-authors were employees of Bristol-Meyer Squibb and Pfizer that co-market apixaban. Another important methodological consideration that should be acknowledged was the substantial number of patients (n=5123) treated with an off-label dose (10mg) of rivaroxaban for stroke prevention. Furthermore, the matching process needs to be scrutinized extensively, since it is striking that out of 27571 patients treated with dabigatran, nearly 100% (99,9%) were matched to 27538 patients receiving rivaroxaban. Large RCTs about NOAC versus NOAC comparison are required and expected to provide more robust results, which afterwards can be compared with real-world data evidence.
Dear Dr. Palaiodimou,
On behalf of all authors of the article published in Stroke based on ARISTOPHANES study (NCT03087487), I would like to thank you for your thorough review and helpful comments about our article. Please see below for our responses to some questions you raised in the blog post.
Although 10 mg is not an approved dose for rivaroxaban’s AF indication in US, we included those patients in our study as long as they did not have any evidence suggesting the use of rivaroxaban for VTE treatment or VTE prophylaxis. A similar small percentage of patients (~5%) were observed to be on the 10 mg dose in rivaroxaban cohort of our study as were in a post-marketing safety surveillance study required by FDA for rivaroxaban’s AF indication in US (Tamayo 2015, PMID: 25588595). The patients on 10 mg rivaroxaban had similar incidence rates for stroke/SE and major bleeding compared to those on 15 mg rivaroxaban. After we excluded patients on 10 mg rivaroxaban and their matched counterparts from the cohorts after propensity score matching, results for any stroke/SE and major bleeding comparisons remained similar to what were reported in the article. Results from this post-hoc sensitivity analysis are listed below:
Propensity score-matched NOAC vs warfarin and NOAC vs NOAC comparisons (any dose):
Apixaban (5 mg/2.5 mg) vs rivaroxaban (20 mg/15 mg): stroke/SE HR (95% CI): 0.81 (0.71-0.92), major bleeding HR (95% CI): 0.55 (0.51-0.59)
Dabigatran (150 mg/75 mg) vs rivaroxaban (20 mg/15 mg): stroke/SE HR (95% CI): 1.13 (0.94-1.36), major bleeding HR (95% CI): 0.70 (0.63-0.77))
Rivaroxaban (20 mg/15 mg) vs warfarin: stroke/SE HR (95% CI): 0.76 (0.69-0.83), major bleeding HR (95% CI): 1.07 (1.02-1.13)).
Propensity score-matched lower dose NOAC vs warfarin and lower dose NOAC vs lower dose NOAC comparisons:
Apixaban (2.5 mg) vs rivaroxaban (15 mg): stroke/SE HR (95% CI): 0.75 (0.57-0.98), major bleeding HR (95% CI): 0.52 (0.45-0.60)
Dabigatran (75 mg) vs rivaroxaban (15 mg): stroke/SE HR (95% CI): 1.63 (1.08-2.46), major bleeding HR (95% CI): 0.83 (0.66-1.04))
Rivaroxaban (15 mg) vs warfarin: stroke/SE HR (95% CI): 0.89 (0.75-1.05), major bleeding HR (95% CI): 1.16 (1.06-1.27)).
Over 97% of dabigatran patients could be matched to rivaroxaban patients due to the very similar risk profiles of the two cohorts. They had similar values in mean age (71.4 years vs 71.6 years), mean CHA2DS2-VASc score (3.3 vs 3.3), and mean HAS-BLED score (2.6 vs 2.7)(see Supplemental Table II of the article). Further, the big difference in the size of the two cohorts (27,571 dabigatran patients vs 103,477 rivaroxaban patients) meant that each dabigatran patient had on average 3-4 potential rivaroxaban patients as potential candidates for matching.
All-cause mortality was only examined in CMS Medicare database in our study, because it is the only one among the five data sources which includes comprehensive mortality information and has been used for mortality comparisons in other published NOAC studies (eg, Graham 2015, PMID: 25359164; Graham 2016, PMID: 27695821; Graham 2019, PMID: 30639551).
We appreciate your blog post and would be happy to answer any other questions from you and any readers.
Best regards,
Dr. Steve Deitelzweig