Background
Adjusted-dose unfractionated heparin or bodyweight-adjusted low molecular weight heparin overlapping with and followed by laboratory-titrated vitamin K antagonists (VKAs) is widely accepted as the standard treatment for patients presenting with acute deep vein thrombosis (DVT) and/or pulmonary embolism (PE) [
1]. Although highly efficacious and relatively safe, this treatment regimen is associated with a considerable number of disadvantages, including the need for parenteral administration of heparin and frequent laboratory monitoring of the pharmacodynamic effects of VKAs with subsequent dose adjustments.
Recently, small molecules have been developed that directly inhibit a single component of the coagulation cascade (such as Factor Xa or thrombin), can be taken orally and have a predictable pharmacokinetic/pharmacodynamic profile [
2‐
4]. Although this eliminates the need for parenteral administration and routine coagulation monitoring, it is essential that these new compounds are demonstrated to be clinically as efficacious as the current standard treatment. Demonstration of similar efficacy requires randomised trials involving patients with symptomatic DVT and/or PE that compare the incidences of recurrent venous thromboembolism (VTE) between the new compounds and the current standard treatment. To exclude a clinically relevant excess of recurrent VTE, the direct oral anticoagulants need to satisfy a non-inferiority margin that is based on the documented efficacy of current standard treatments [
5,
6].
In this paper, we systematically reviewed all studies that have evaluated the efficacy of the current standard treatment for DVT and/or PE and calculated accompanying non-inferiority margins from the meta-analysis.
Results
A total of 5388 publications were identified in our literature search, which used the seminal publication of Barritt and Jordan as a starting point [
7]. Review of the identified publications in accordance with the eligibility criteria resulted in the selection of a further 14 studies. One study was excluded because the ‘less intensive treatment’ regimen had been shown to be superior to placebo in another study [
22,
23]. Of the 14 studies included, nine evaluated anticoagulant therapy for a duration of up to 3 months [
7‐
15], three for a duration of up to 12 months, [
16,
17,
19] and two for longer durations [
18,
20].
Current standard of care was compared with placebo in five studies, with ‘no treatment’ in three studies and with ‘less intensive treatment’ in six studies. The design characteristics of these studies are summarised in Table
1. The results per study are summarised in Table
2.
Table 1
Summary of design characteristics for the 14 selected studies
| PE | Intravenous UFH q6h vs placebo | VKA. PT 2.0–3.0 times control vs placebo | 14 days | 3 months |
| DVT | Intravenous UFH | VKA started on day 10, PT 1.5–2.0 vs subcutaneous UFH bid | 6 weeks for calf DVT and 12 weeks for proximal DVT | 3 months |
| DVT | UFH | VKA started on day 1, TT 5–14% | 1 month vs 6 months | 3 months |
| Isolated calf DVT | Intravenous UFH | VKA started on day 1, INR 2.5–4.2 vs placebo | 3 months | 3 months |
| Proximal DVT | Intravenous UFH vs subcutaneous UFH | VKA started on day 6 or 7, INR 2.0–3.0 | 3 months | 3 months |
| Proximal DVT | Intravenous UFH vs placebo | VKA started on day 1, INR 2.0–3.0 | 6 months | 3 months |
| DVT or PE | Intravenous weight-based UFH vs ‘standard care’ UFH nomogram | VKA started after day 2 | Not specified or monitored | 3 months |
| Proximal DVT | Intravenous UFH | VKA INR 2.0–3.0 vs placebo | 4 weeks vs 12 weeks | 3 months |
| DVT or PE | Intravenous UFH or subcutaneous LMWH | VKA started on day 1, INR 2.00–2.85 | 6 weeks vs 6 months | 3 months |
| Idiopathic proximal DVT | Intravenous UFH or subcutaneous LMWH | VKA INR 2.0–3.0 vs no treatment | Months 4–12 vs no treatment | Months 4–12 |
| PE | Intravenous UFH or subcutaneous LMWH | VKA INR 2.0–3.0 vs no treatment | Months 4–12 vs no treatment | Months 4–12 |
| Idiopathic DVT or PE | Intravenous UFH or subcutaneous LMWH | VKA INR 2.0–3.0 vs placebo | Months 4–24 vs no treatment | Months 4–24 |
| DVT or PE | Intravenous UFH or subcutaneous LMWH | VKA INR 2.0–3.0 | Months 4–6 vs no treatment | Months 4–6 |
| First recurrent DVT or PE | Intravenous UFH or subcutaneous LMWH | VKA INR 2.00–2.85 vs no treatment | Indefinitely vs none | Month 6 onwards |
Table 2
Recurrent venous thromboembolic events during the comparison period
| 0/16 (0.0) | 11/19 (57.9) |
| 0/33 (0.0) | 6/35 (17.1) |
| 3/66 (4.5) | 6/69 (8.7) |
| 0/23 (0.0) | 7/28 (25.0) |
| 3/58 (5.2) | 11/57 (19.3) |
| 2/60 (3.3) | 10/60 (16.7) |
| 2/41 (4.9) | 8/32 (25.0) |
| 1/109 (0.9) | 9/105 (8.6) |
| 4/454 (0.9) | 26/443 (5.9) |
| 4/134 (3.0) | 11/133 (8.3) |
| 1/165 (0.6) | 6/161 (3.7) |
| 1/79 (1.3) | 17/83 (20.5) |
| 1/361 (0.3) | 6/375 (1.6) |
| 3/116 (2.6) | 23/111 (20.7) |
Total number of recurrent events across studies | 25/1715 (1.5) | 157/1711 (9.2) |
Establishing the non-inferiority margin
Recurrent VTE occurred in 25 (1.5%) of 1715 patients who received current standard of care and in 157 (9.2%) of 1711 patients who received placebo, ‘no treatment’ or ‘less intensive treatment’ (Table
2). Therefore, the observed treatment effect for prevention of recurrent VTE was associated with an odds ratio of 0.18 (95% CI 0.14−0.25; test for heterogeneity, p=0.87) and a risk ratio of 0.19 (95% CI 0.12−0.28; test for heterogeneity, p=0.85). Conversely, in the absence of current standard treatment, the risk of recurrent VTE increased with an odds ratio of 5.56 (95% CI 4.00−7.14) or a risk ratio of 5.26 (95% CI 3.57−8.33) (Table
3). Hence, the lower limits of the 95% CIs of the effect needed to be maintained relevant for the calculation of the non-inferiority thresholds were 4.00 for the odds ratio and 3.57 for the risk ratio (Table
3). Thus, to preserve 50% of the established treatment effect, the threshold for non-inferiority equalled 2.50 if calculated using an odds ratio and 2.29 if calculated using a risk ratio on a linear scale. Likewise, to preserve 75% of the treatment effect, these thresholds should be 1.75 and 1.64, respectively (Table
3). In Table
3, the limits on a geometric scale are also presented.
Table 3
Calculation of non-inferiority margins based on 14 eligible studies
Point estimate of current standard of care vs placebo, ‘no treatment’ or ‘less intensive treatment’ | 0.18 (0.14−0.25) | 0.19 (0.12−0.28) |
Reversed point estimate of current standard of care vs placebo, ‘no treatment’ or ‘less intensive treatment’ | 5.56 (4.00−7.14) | 5.26 (3.57−8.33) |
Non-inferiority margins
|
Odds ratio based on arithmetic/geometric scale
|
Risk ratio based on arithmetic/geometric scale
|
Limit for no preserved effect | 4.00/4.00 | 3.57/3.57 |
Threshold for preservation of 50% of effect | 2.50/2.00 | 2.29/1.89 |
Threshold for preservation of 66% of effect | 2.00/1.60 | 1.86/1.54 |
Threshold for preservation of 75% of effect | 1.75/1.41 | 1.64/1.37 |
Threshold for preservation of 100% of effect* | 1.00/1.00 | 1.00/1.00 |
The point estimates of the relative treatment effects between studies with a treatment duration of 3 months and studies with a treatment duration of longer than 3 months were consistent with odds ratios of 0.18 (95% CI 0.11−0.32) and of 0.18 (95% CI 0.12−0.28), respectively.
Discussion
Our systematic review of the medical literature demonstrates that the current standard treatment for VTE is highly efficacious, with a pooled odds ratio of 0.18 and a pooled risk ratio of 0.19, corresponding to a relative risk reduction of over 80%. Based on this review, non-inferiority margins of 2.50, 2.00 and 1.75 could be calculated using odd ratios, corresponding to maintenance of documented efficacy of current standard treatment of 50%, 66% and 75%, respectively, on a linear scale. In general, if the non-inferiority margins were based on a risk ratio, the margins would be lower than that based on the odds ratio, which would lead to somewhat larger sample sizes.
The meta-analysis on which these margins are based included a large variety of studies undertaken over a long time period (i.e. between 1960 and 2013). Our search yielded more than 5000 potential references for review. However, attempts to narrow the search resulted in the loss of pertinent articles known to the authors. Even using these broad search criteria, the pertinent reference of Barritt and Jordan was not found. Therefore, the approach was supplemented by an extensive manual search to identify all relevant studies. All 14 included studies had a randomised design, reported clinical outcomes and documented the incidence of recurrent VTE. Although all studies included a group of patients that received the current standard of care, the comparator varied between placebo and ‘no treatment’ or some form of anticoagulant treatment that lacked evidence of efficacy. Despite the lack of evidence of efficacy, it cannot be excluded that these ‘less intensive’ anticoagulant regimens offered some protection against recurrent VTE, which would make our estimates of effect and accompanying non-inferiority margins relatively conservative. No statistical heterogeneity was identified when combining the results of the 14 identified studies. Based on the consistency of the observed effect in all studies and the high number of patients in the meta- analysis, the 95% CIs around the observed summary treatment effect are relatively narrow. With the observed large risk reduction, the calculated non-inferiority margin to demonstrate maintenance of 50% of the treatment effect, a commonly accepted approach, is 2.50 on a linear scale, which might be viewed as unacceptable because it involves clinically important events.
On a geometric scale, the comparable number for the non-inferiority margin would be 2.00. From our calculations, it is apparent that calculating a summary relative risk and then applying a geometric scale results in more stringent non-inferiority margins (Table
3). Part of this effect is caused by the wider CI around the summary effect estimate when using risk ratios rather than odds ratios, as a result of the different underlying statistical models. The other important choice is using an arithmetic (linear) or geometric scale. When reasoning from the event rate in a ‘no treatment’ situation, the logical choice would be to use a geometric scale (relative reduction of outcome rates upon relative reduction). However, when clinically evaluating the allowed excess of outcome events versus current standard treatment, the starting point is the observed incidence on current standard treatment and the logical choice would be to use a linear scale.
Recent published studies that evaluated direct oral anticoagulants in patients with symptomatic DVT and/or PE used 1.80, 2.00, 2.75 and 2.85 as non-inferiority margins to calculate their sample size. For the EINSTEIN DVT and EINSTEIN PE studies that evaluated the direct Factor Xa inhibitor rivaroxaban, it was calculated that a total of 88 events were needed for each study to demonstrate non-inferiority versus standard of care with a margin of 2.0 [
24,
25]. The observed upper limits of the 95% CIs around the relative treatment effect were 1.04 in the EINSTEIN DVT study, 1.68 in the EINSTEIN PE study, and 1.19 for all patients in the EINSTEIN DVT and EINSTEIN PE studies, corresponding to a retention of treatment effect of 99%, 77% and 94%, respectively, on a linear scale. Using a geometric scale, these values become 97%, 63% and 87%, respectively. In the RE-COVER and RE-MEDY studies that evaluated the direct thrombin inhibitor dabigatran versus current standard treatment in patients with DVT and/or PE, it was calculated that 46 and 40 events were needed for each of the studies to demonstrate non-inferiority versus standard of care with a margin of 2.75 and 2.85, respectively [
26,
27]. The observed upper limits of the 95% CIs around the relative treatment effect were 1.84 in the RE-COVER trial and 2.64 in RE-MEDY trial, corresponding to retention of treatment effect of 72% and 45%, respectively, on a linear scale. Using a geometric scale, these values become 56% and 30%, respectively. For the AMPLIFY study that evaluated the direct Factor Xa inhibitor apixaban in patients with DVT and/or PE, it was calculated that a total of 123 events were needed to demonstrate non-inferiority versus standard of care with a margin of 1.80 [
28]. The observed upper limits of the 95% CIs around the relative treatment effect were 1.26 in patients with DVT, 1.61 in patients with PE, and 1.18 for all patients, corresponding to a retention of treatment effect of 91%, 79% and 94%, respectively, on a linear scale. Using a geometric scale, these values become 83%, 65% and 88%, respectively. In the HOKUSAI-VTE study, evaluating the direct Factor Xa inhibitor edoxaban in patients with symptomatic DVT and/or PE, a non-inferiority margin of 1.50 has been adopted (clinicaltrials.gov: NCT00986154) [
29].
In summary, the available studies allow the calculation of non-inferiority margins using a statistical approach that are appropriate for the evaluation of direct oral anticoagulants in patients with DVT and/or PE. Studies on direct oral anticoagulants that were conducted or are ongoing have adopted various margins and differ in the amount of retention of treatment effect that they have demonstrated.
Competing interests
MHP: Has acted as a consultant for Bayer HealthCare, Sanofi-aventis, Boehringer Ingelheim, GlaxoSmithKline, Daiichi Sankyo, Leo Pharma, Thrombogenics, Pfizer. AWAL is an employee of Bayer Pharma AG.
Authors’ contributions
MHP and AWAL performed the analysis and interpreted the data, were involved in drafting and revising the manuscript, and provided final approval for publication of the manuscript.