How to REVISIT the increasing “real world” evidence for stroke prevention in non-valvular atrial fibrillation?

Randomized controlled trials (RCTs) are the best way to test the effectiveness and safety of a new drug or intervention in a controlled manner but, by design, RCTs are perhaps artificial scenarios with specific inclusion and exclusion criteria, a set protocol, regular follow-ups and enthusiastic compliant patients that are followed up by equally enthusiastic research teams. After the drug is approved and available for widespread clinical use, the drug is often used in a much broader patient population (sometimes including patients outside trial specific inclusion/exclusion criteria) in varied healthcare settings that may include diverse ethnic groups, multiple comorbidities and less compliant or infrequently followed up patients. Thus “real world” observational data would be rather different from the “unreal world” of RCT cohorts^1,2.

The REVISIT-US study

In this issue of the journal, Coleman et al.³ report on a retrospective "real world" study (REVISIT-US) of patients newly initiated on rivaroxaban, apixaban or warfarin, using MarketScan claims data from January 2012 to October 2014, propensity-score matched individually to warfarin users. Coleman et al.³ conclude that rivaroxaban treatment (n = 11411) was associated with a significant reduction of the combined endpoint of ischemic stroke or intracranial hemorrhage (ICH) versus warfarin (HR = 0.61, 95% CI = 0.45–0.82), which was essentially driven by a significant ICH reduction whilst ischemic stroke was nonsignificantly reduced. For the much smaller apixaban cohort (n = 4083), there was a nonsignificant reduction in the combined endpoint of ischemic stroke or ICH versus warfarin (HR = 0.63, 95% CI = 0.25–1.12) with ICH risk significantly reduced and ischemic stroke risk nonsignificantly higher with apixaban (HR = 1.13, 95% CI = 0.49–2.63) versus warfarin, with wide 95% confidence intervals.

As with many situations, the devil is in the detail. A combined endpoint of ischemic stroke or ICH and each endpoint individually were used by Coleman et al.³, which is inconsistent with the more conventional efficacy endpoint of RCTs (i.e. all stroke, thus including ischemic and hemorrhagic strokes, and systemic embolism). Not all ICH is hemorrhagic stroke, and Coleman et al. do not provide this subtype differentiation. Patients with a prior stroke, systemic embolism or intracranial hemorrhage (ICH) were excluded, although such high risk patients are often treated in everyday clinical practice. Mean duration of follow-up was not actually stated, but based on the patient-years reported, follow-up duration seems to be <1 year.

Perhaps an alternative (conservative) interpretation of the endpoints is that both rivaroxaban and apixaban were associated with a significant reduction in ICH, and there were no significant differences between either rivaroxaban or apixaban against warfarin for ischemic stroke, especially with small(er) apixaban sample size, short follow-up and the wide 95% CIs for the reported hazard ratio. Indeed, residual confounding could not be ruled out, especially with the use of administrative claims databases. Errors in diagnostic coding may occur, and important details such as quality of anticoagulation control in the warfarin users was uncertain. The latter is important as the performance of warfarin is highly dependent on the average individual time in therapeutic range, as well as many clinical variables^4,5.

Most importantly, Coleman et al.³ do not provide sensitivity analyses to separately show the comparative effectiveness of usual or reduced dose regimes of rivaroxaban and apixaban. Whilst 17.3% received the reduced dose of rivaroxaban 15 mg od, and 15.5% received the reduced dose of apixaban (2.5 mg bid), there is no information on whether these were appropriately prescribed based on guidelines or label recommendations. This is relevant as label (or guideline) adherent use is associated with better outcomes⁶. As with the "real world", poorer adherence and compliance may have contributed to the (modest) numerical differences in endpoints, especially with the short treatment follow-up period.

Perhaps another possible consideration is the time frame of the study (2012–2014). Rivaroxaban was approved in 2011, whilst apixaban was approved in 2012. Prescribers may be more “used” to rivaroxaban (being fairly established in the anticoagulant market by then), and early apixaban use may have included prior users of oral anticoagulants (and such patients seem to have more adverse outcomes⁷). Whilst the methodology of the paper by Coleman et al. reports that patients had to be oral anticoagulation (OAC) naïve in the 180 days prior to the first qualifying OAC dispensing, this did not necessarily mean they were new users or newly diagnosed.

Revisiting the evidence from the “real world”

The paper by Coleman et al.³ is one of a deluge of papers on real world observational data on the comparative effectiveness of the various non-Vitamin K antagonist oral anticoagulants (NOACs) against warfarin^8–11, or against each other^12–14. Many cohorts have had substantially larger sample sizes and, importantly, the present analysis provides no data on dabigatran despite the extensive real world data published for this drug¹⁵.

Some published cohorts are also independent analyses, with no sponsorship from industry, such as the large United States FDA Medicare analysis by Graham et al. comparing the relative effectiveness and safety of rivaroxaban and dabigatran¹³. In the latter study, treatment with rivaroxaban 20 mg once daily was associated with no significant difference in thromboembolic stroke but statistically significant increases in ICH and major extracranial bleeding, including major gastrointestinal bleeding, compared with dabigatran 150 mg twice daily. Mortality was higher on rivaroxaban compared with dabigatran, with borderline statistical significance. Other studies based on claims datasets have focused on early comparative safety differences between the NOACs¹⁶ (even with the same Marketscan claims dataset used by Coleman et al.³), but have not presented effectiveness data given the limitations of small numbers (and hence, underpower) and short follow-up period.

Quo vadis?

All the NOACs have excellent clinical RCT data, and have definitely changed the landscape of stroke prevention in atrial fibrillation (AF). The RCTs compared NOACs to warfarin, and how one NOAC would compare against another can only be definitively answered by a large prospective head to head RCT, which is unlikely to happen. Indirect comparisons have been published¹⁷, but are methodologically weak – however, the extensive published real world evidence now complements and augments the RCT and supplements the indirect comparisons, showing how these drugs might actually perform when used in clinical practice.

The patient perspective is also important, and patients place the highest priority on stroke prevention¹⁸ and all NOACs do fulfill this aspect. In contrast, bleeding seems to be more the concern of prescribers, and RCTs and real world data provide insights into how the different drugs may perform from this aspect. Finally, AF patients are clinically heterogeneous and thinking that "one drug fits all" is simply naïve and unrealistic. Given that apart from warfarin we now have four NOACs available, prescribers can now fit the drug to the patient profile and are perhaps spoilt for choice¹⁹.

Transparency

Declaration of funding

This editorial was not funded.

Declaration of financial/other relationships

G.Y.H.L. has disclosed that he has served as a consultant for Bayer/Janssen, BMS/Pfizer, Biotronik, Medtronic, Boehringer Ingelheim, Microlife and Daiichi-Sankyo and been a speaker for Bayer, BMS/Pfizer, Medtronic, Boehringer Ingelheim, Microlife, Roche and Daiichi-Sankyo.

The CMRO peer reviewer has disclosed that she/he received grant funding and consultancy fees from Janssen Pharmaceuticals, Bayer and Boehringer Ingelheim Pharmaceuticals.