Journal of Thrombosis and Haemostasis, 15: 1913–1922 DOI: 10.1111/jth.

13783

ORIGINAL ARTICLE

Effect of extended-duration thromboprophylaxis on venous

thromboembolism and major bleeding among acutely ill
hospitalized medical patients: a bivariate analysis
G. CHI,* S. Z. GOLDHABER,† J. M. KITTELSON,‡ A. G. G. TURPIE,§ A. F. HERNANDEZ,¶ R. D.
H U L L , * * A . G O L D , † † J . T . C U R N U T T E , † † A . T . C O H E N , § § R . A . H A R R I N G T O N ¶ ¶ and C . M . G I B S O N *
*Cardiovascular Division, Departments of Medicine, Beth Israel Deaconess Medical Center, Harvard Medical School; †Cardiovascular
Division, Brigham and Women’s Hospital, Harvard Medical School, Boston, MA; ‡Department of Biostatistics and Informatics, Colorado
School of Public Health, University of Colorado Anschutz Medical Campus, Aurora, CO, USA; §Department of Medicine, Hamilton Health
Sciences, General Division, Hamilton, Ontario, Canada; ¶Duke University and Duke Clinical Research Institute, Durham, NC, USA; **Division
of Cardiology, R. A. H. Faculty of Medicine, University of Calgary, Calgary, Alberta, Canada; ††Portola Pharmaceuticals Inc., South San
Francisco, CA, USA; §§Department of Haematological Medicine, Guy’s and St Thomas’ Hospitals, King’s College, London, UK; and ¶¶Division
of Cardiovascular Medicine, Department of Medicine, Stanford University, Stanford, CA, USA

To cite this article: Chi G, Goldhaber SZ, Kittelson JM, Turpie AGG, Hernandez AF, Hull RD, Gold A, Curnutte JT, Cohen AT, Harrington RA,
Gibson CM. Effect of extended-duration thromboprophylaxis on venous thromboembolism and major bleeding among acutely ill hospitalized
medical patients: a bivariate analysis. J Thromb Haemost 2017; 15: 1913–22.

described bivariate method that does not assume a linear

Essentials
• Anticoagulants prevent venous thromboembolism but
may be associated with greater bleeding risks.
• Bivariate analysis assumes a non-linear relationship
between efficacy and safety outcomes.
• Extended full-dose betrixaban is favorable over stan-
dard enoxaparin in bivariate endpoint.
• Clinicians must weigh efficacy and safety outcomes in
decision-making on thromboprophylaxis.
Summary. Background: Among acutely ill hospitalized
medical patients, extended-duration thromboprophylaxis
reduces the risk of venous thromboembolism (VTE), but
some pharmacologic strategies have been associated with
greater risks of major bleeding, thereby offsetting the net
clinical benefit (NCB). Methods: To assess the risk–bene-
fit profile of anticoagulation regimens, a previously
Correspondence: C. Michael Gibson, PERFUSE Study Group, Car-
diovascular Division, Department of Medicine, Beth Israel Dea-
coness Medical Center, Harvard Medical School, 330 Brookline
Avenue, Suite OV-540, Boston, MA 02215, USA
Tel.: +1 617 975 9950
E-mail: mgibson@bidmc.harvard.edu
Clinical Trial Registration: URL: http://www.clinicaltrials.gov.
Unique identifier: NCT00077753 (EXCLAIM); NCT00457002
(ADOPT); NCT00571649 (MAGELLAN); NCT01583218 (APEX).
Received: 9 May 2017
Manuscript handled by: J. Douketis
Final decision: F. R. Rosendaal, 14 July 2017
risk–benefit tradeoff and can accommodate different mar-
gins for efficacy and safety was performed to simultane-
ously assess efficacy (symptomatic VTE) and safety
(major bleeding) on the basis of data from four random-
ized controlled trials of extended-duration (30–46 days)
versus standard-duration (6–14 days) thromboprophylaxis
among 28 227 patients (EXCLAIM, ADOPT, MAGEL-
LAN and APEX trials). Results: Extended thrombopro-
phylaxis with full-dose betrixaban (80 mg once daily) was
superior in efficacy and non-inferior in safety to stan-
dard-duration enoxaparin, and showed a significantly
favorable NCB, with a risk difference of
0.51%
(
0.89% to
0.10%) in the bivariate outcome.
Extended enoxaparin was superior in efficacy and inferior
in safety (bivariate outcome: 0.03% [
0.37% to 0.43%]),
whereas apixaban and rivaroxaban were non-inferior in
efficacy and inferior in safety (
0.20% [
0.49% to
0.17%] and 0.23% [
0.16% to 0.69%], respectively).
Reduced-dose betrixaban did not show a significant dif-
ference in either efficacy or safety (0.41% [
0.85% to
1.94%]). Conclusions: In a bivariate analysis that assumes
non-linear risk–benefit tradeoffs, extended prophylaxis
with full-dose betrixaban was superior to standard-dura-
tion enoxaparin, whereas other regimens failed to simulta-
neously achieve both superiority and non-inferiority with
respect to symptomatic VTE and major bleeding in the
management of acutely ill hospitalized medical patients.
Keywords: adverse drug reaction; anticoagulant; clinical
efficacy; hemorrhage; venous thromboembolism.

© 2017 International Society on Thrombosis and Haemostasis

1914 G. Chi et al
Introduction
Among hospitalized medically ill patients at risk of
venous thromboembolism (VTE), standard pharmacologic
thromboprophylaxis is recommended during the period of
immobilization or acute hospital stay [1]. The risk of
VTE may, however, extend well beyond this brief stan-
dard course of anticoagulation, and may remain elevated
for weeks to months after hospital discharge [2]. As com-
pared with standard-duration thromboprophylaxis,
extended-duration therapy with a novel oral anticoagu-
lant may reduce the risk of symptomatic deep vein throm-
bosis (DVT) or pulmonary embolism (PE), but may also
increase the risk of major bleeding, offsetting the net clin-
ical benefit (NCB) of such a strategy [3].
Clinical decisions regarding the utility of thrombopro-
phylaxis require the simultaneous consideration of both
efficacy and safety outcome. The conventional approach
to net clinical benefit makes three assumptions: (i) it
assumes a linear relationship between safety and efficacy;
(ii) it assumes that there is no upper limit or boundary to
the rate of safety outcome that would be acceptable; and
(iii) it assumes that both efficacy and safety will be sub-
jected to the same statistical margins in the testing of
either superiority or non-inferiority.
To overcome the limitations of this linear NCB frame-
work, a novel bivariate approach that considers a multi-
plicative definition of NCB has been proposed [4]. A
bivariate model does not assume a static, linear relation-
ship between safety and efficacy: the magnitude of effi-
cacy required to offset a safety signal may increase
dramatically and non-linearly as the absolute risk of
bleeding rises, for example. Whereas a conventional NCB
would use either a simple composite endpoint where effi-
cacy and safety endpoints are weighed 1 : 1 (Fig. S1) or
at a ratio of 2 : 1, for instance (Fig. S2), both of these
ratios are constant over any given range of efficacy or
safety [5]. Under the bivariate framework, this tradeoff is
not fixed, and varies non-linearly according to the abso-
lute risk difference in efficacy and safety (Fig. S3).
Although Figs S1 and S2 show different slopes for the
tradeoff between efficacy and safety, both are straight
lines. In contrast, Fig. S3 shows that, in a bivariate analy-
sis, there is a curvilinear tradeoff between efficacy and
safety. Stated differently, the acceptable tradeoff between
efficacy and safety varies widely according to the rates of
efficacy and safety (more specifically the risk difference in
efficacy and safety).
Second, in contrast to the conventional method, in a
bivariate analysis a boundary is set for the difference in
bleeding between the two strategies that would be clini-
cally acceptable. These are shown by solid circles at the
limits of the curves in Fig. S3. Simply stated, these solid
circles indicate the upper limit of excess bleeding and
excess efficacy events that would be clinically acceptable
(Fig. S3).
Finally, not only is the relationship between efficacy
and safety curvilinear, as stated above, but this curvilin-
ear relationship is, in fact, asymmetric. As shown in
Fig. S3, the distance from the x-axis to the curve (the
length of the green arrow; the amount of acceptable dif-
ference in efficacy) is different from the distance from the
y-axis to the curve (the length of the red arrow; the
amount of acceptable difference in safety). Stated simply,
there is asymmetry in the amount of efficacy versus safety
that is clinically acceptable. In a bivariate analysis, the
margins for efficacy and safety can be tailored in an
asymmetric fashion to reflect the clinical relevance of the
respective endpoints. The researcher can decide to set
parameters to assess whether a new therapy is superior in
efficacy and simultaneously non-inferior with respect to
safety, or alternatively whether a regimen is significantly
safer (superior in safety) and simultaneously non-inferior
with respect to efficacy.
The bivariate method has been employed to evaluate the
risk–benefit profile of the anticoagulation strategy in vari-
ous settings [6–8]. In this study, the bivariate approach was
applied to simultaneously analyze the risk and benefit of
extended-duration thromboprophylaxis among acutely ill
hospitalized medical patients on the basis of the published
results of four phase III/IV randomized controlled trials
(EXCLAIM, ADOPT, MAGELLAN, and APEX) [9–12].
The effects of extended-duration enoxaparin, apixaban,
rivaroxaban and betrixaban on symptomatic VTE and
major bleeding were compared in a bivariate model to
assess their risk–benefit profile.
Methods
Data extraction
Randomized controlled trials were selected if they fulfilled
all of the following criteria: (i) the study population com-
prised patients hospitalized for an acute medical illness,
including heart failure, respiratory failure, infectious dis-
ease, rheumatic disorder, or ischemic stroke; (ii) the study
design entailed comparison between extended-duration
(30–46 days) and standard-duration (6–14 days) pharma-
cologic thromboprophylaxis; and (iii) the primary study
endpoints included symptomatic DVT, PE, VTE-related
mortality, and major bleeding. Four phase III/IV ran-
domized controlled trials met these criteria and were
included in the study: EXCLAIM [9], ADOPT [10],
MAGELLAN [11], and APEX [12].
Study interventions and endpoints
The EXCLAIM trial Following open-label subcuta-
neous enoxaparin 40 mg once daily for 10  4 days, 6085
patients were randomized in a 1 : 1 ratio to receive either
enoxaparin or placebo for an additional 28  4 days [9].
The primary efficacy endpoint was VTE, defined as a
© 2017 International Society on Thrombosis and Haemostasis

composite of symptomatic or asymptomatic proximal
DVT, symptomatic PE or fatal PE during the double-
blind treatment period (28  4 days after random assign-
ment). The primary safety endpoint was the incidence of
major hemorrhagic complications during and up to 48 h
after the double-blind treatment period.
The ADOPT trial A total of 6528 patients were ran-
domized in a 1 : 1 ratio to receive either oral apixaban
2.5 mg twice daily for 30 days or subcutaneous enoxa-
parin 40 mg once daily during their hospital stay, for a
minimum of 6 days [10]. The primary efficacy endpoint
was a composite of asymptomatic proximal DVT, symp-
tomatic DVT, fatal or non-fatal PE or death related to
VTE in the 30-day treatment period. Major bleeding dur-
ing the 30-day treatment period was included as one of
the main safety endpoints.
The MAGELLAN trial A total of 8101 patients were
randomized in a 1 : 1 ratio to receive either oral rivaroxa-
ban 10 mg once daily for 35  4 days or subcutaneous
enoxaparin 40 mg once daily for 10  4 days [11]. The
primary efficacy endpoint was a composite of asymp-
tomatic proximal DVT, symptomatic proximal or distal
DVT, symptomatic non-fatal PE, or death related to
VTE. Major bleeding was included in the safety outcome
measures. Data on efficacy and safety endpoints up to
35 days were extracted for the purpose of homogeneity.
The APEX trial A total of 7513 patients were random-
ized in a 1 : 1 ratio to receive either oral betrixaban
80 mg once daily (full-dose regimen) for 35–42 days or
subcutaneous betrixaban 40 mg once daily for
10  4 days [12]. Patients with severe renal insufficiency
(creatinine clearance of < 30 mL min
1) or being treated
with a concomitant strong P-glycoprotein inhibitor
received a reduced-dose regimen of betrixaban (40 mg
once daily). The primary efficacy endpoint was a compos-
ite of asymptomatic proximal DVT between day 32 and
day 47, symptomatic proximal or distal DVT, symp-
tomatic non-fatal PE, or death from VTE between day 1
to day 42. The principal safety outcome was the occur-
rence of major bleeding at any point until 7 days after
discontinuation of study medications.
Statistical analysis
The analysis assessed the efficacy–safety tradeoff between
symptomatic VTE (a composite of symptomatic DVT,
PE, and VTE-related mortality) and major bleeding on
the basis of on study-level data from the publications for
the EXCLAIM, ADOPT, MAGELLAN and APEX tri-
als. The detailed methodology of bivariate analysis has
been previously described [8]. In brief, the risk differences
in efficacy (RDE) and safety (RDS) were calculated by
subtracting the event rate for standard-duration
© 2017 International Society on Thrombosis and Haemostasis
Bivariate analysis and net clinical benefit 1915
enoxaparin from the event rate for extended-duration
thromboprophylaxis. A bivariate two-dimensional plane
was thus defined by RDE and RDS, with the lower left
quadrant representing reductions in both symptomatic
VTE and major bleeding risks with extended-duration
treatment (Fig. 1). The 95% confidence interval (CI)
of risk difference on the efficacy and safety endpoint for
each trial was summarized as a rectangle on the bivari-
ate plane. A multiplicative definition of NCB was
considered as:
  
RDE
AE RDS
AS
NCB ¼
WE WS
where:
AE ¼ horizontal asymptote for the efficacy endpoint
AS ¼ vertical asymptote for the safety endpoint
WE ¼ weighting factor for the efficacy endpoint
WS ¼ weighting factor for the safety endpoint
Derivation of the NCB curve
A clinically important risk difference was set at 2.00%
for both efficacy and safety events to approximate the
maximum effect size among the extended-duration throm-
boprophylactic agents. The non-inferiority efficacy margin
of 9% (equivalent to an absolute risk difference of
0.18%, i.e. from 2.00% to 2.18%) was derived from pre-
serving 50% of the treatment effect of standard-duration
enoxaparin dosed at 40 mg once daily against placebo
based on previous studies [13–15] in the random-effects
model [16] with the fixed margin method [14] (Data S1).
The non-inferiority safety margin was set at 25% (equiva-
lent to an absolute risk difference of 0.50%, i.e. from
2.00% to 2.50%) in accordance with the original publica-
tion [8]. The parameters AE, AS, WE, and WS were subse-
quently solved to derive the NCB curve representing the
null hypothesis.
Qualitative and quantitative assessment of the bivariate
outcome
The NCB curve divided the bivariate plane into two
regions. The lack-of-benefit region was defined as the part
above the curve. The risk–benefit profile was deemed to
be favorable against standard-of-care enoxaparin if the
95% CI rectangle did not contain the lack-of-benefit
region. The qualitative inference obtained with the multi-
plicative NCB definition was compared with the conven-
tional NCB definition. Furthermore, bivariate outcomes
were quantitatively assessed according to the minimum
distance from the NCB curve to three reference points: (i)
center of the confidence rectangle; (ii) southwest corner of
the confidence rectangle representing the lower boundary
of the 95% CI; and (iii) northeast corner of the
1916 G. Chi et al
Risk reduction
Superiority in safety
inferiority in efficacy
Efficacy endpoint
Superiority in both
efficacy and safety
Safety endpoint
Fig. 1. Bivariate plane divided into four quadrants based on superiority or inferiority with respect to efficacy and safety endpoints. [Color fig-
ure can be viewed at wileyonlinelibrary.com]
confidence rectangle representing the upper boundary of
the 95% CI (Fig. 2). Accordingly, the bivariate outcome
(the collective treatment effect on symptomatic VTE and
major bleeding) was expressed as an observed risk differ-
ence along with a range of corner distance (CD). These
metrics were analogous to reporting the point estimate
with a 95% CI; that is, positive values indicate an
increased risk and negative values indicate a decreased
risk.
Sensitivity analysis of the non-inferiority margin
The NCB was compared in three sensitivity analyses to
test a spectrum of non-inferiority margins ranging from
10% to 50%. In the first sensitivity analysis, the non-
inferiority margin for safety was maintained at 25%, and
the efficacy margin ranged from 10% to 50%. In the sec-
ond sensitivity analysis, the non-inferiority margin for
efficacy was maintained at 9%, and the safety margin
ranged from 10% to 50%. In the third sensitivity analy-
sis, paired margins for efficacy and safety ranging from
10% to 50% were used.
Sensitivity analysis of weighing VTE against major bleeding
The four included trials universally included asymp-
tomatic DVT as one of the components of the primary
Inferiority in both
efficacy and safety
Risk reduction
Superiority in efficacy
inferiority in safety
efficacy outcome. In view of this, a sensitivity analysis
weighing VTE (a composite of asymptomatic and symp-
tomatic DVT, PE, and VTE-related mortality) against
major bleeding was also performed.
Results
Summary of trial results
Four double-blind, double-dummy, randomized con-
trolled trials were included. The analyzed data encom-
passed 28 227 patients randomized to receive extended-
duration thromboprophylaxis with enoxaparin, apixaban,
rivaroxaban or betrixaban versus standard-duration
enoxaparin for VTE prevention. The efficacy and safety
event rates are summarized in Table 1.
Qualitative assessment of the bivariate outcome
Trial results were expressed as a rectangle defined by the
95% CI of risk difference in symptomatic VTE and in
major bleeding in the bivariate plane (Fig. 3). Rectangles
for the EXCLAIM, ADOPT and MAGELLAN trials
and the full-dose regimen from the APEX trial clustered
within the lower right quadrant, suggesting that the
reduction in VTE risk generally came at the cost of an
increase in the major bleeding risk. The 0.00% horizontal
© 2017 International Society on Thrombosis and Haemostasis
 Bivariate analysis and net clinical benefit 1917

Risk reduction

Lack-of-benefit region
Efficacy endpoint

NCB curve

Risk reduction
Northeast corner
ound
Lower b

Center
e
Point estimat

Upper bound
Southwest corner

Safety endpoint

Fig. 2. Interpretation of the bivariate outcome. Qualitatively, a favorable net clinical benefit (NCB) is established if the rectangle defined by
the 95% confidence interval of risk difference does not include the lack-of-benefit region. Quantitatively, the bivariate outcome is measured
according to the minimum distance from the NCB curve to the center and opposing corners of the confidence rectangle. The minimum dis-
tances to the center, northeast corner and southwest corner represent the point estimate, lower bound and upper bound of the bivariate out-
come.

Table 1 Effect of extended-duration thromboprophylaxis on symptomatic venous thromboembolism (VTE) and major bleeding among acutely
ill hospitalized medical patients

Active Control Risk difference (%) NNT or

Trial n/N (%) n/N (%) (95% Wald CI) P-value NNH*

Efficacy endpoint: symptomatic DVT, PE, and VTE-related mortality

EXCLAIM (ED enoxaparin versus SD enoxaparin) 5/2485 (0.20) 24/2510 (0.96)
0.75 (
1.17 to
0.34) < 0.001 134
ADOPT (ED apixaban versus SD enoxaparin) 14/2211 (0.63) 27/2284 (1.18)
0.55 (
1.10 to 0.00) 0.05 NI
MAGELLAN (ED rivaroxaban versus SD enoxaparin) 42/2967 (1.42) 59/3057 (1.93)
0.51 (
1.16 to 0.13) 0.12 NI
APEX (ED betrixaban versus SD enoxaparin)*
Full-dose 24/2987 (0.80) 45/2990 (1.51)
0.70 (
1.24 to
0.16) 0.011 143
Reduced-dose 10/730 (1.37) 8/725 (1.10) 0.27 (
0.87 to 1.40) 0.65 NI
Safety endpoint: major bleeding
EXCLAIM (ED enoxaparin versus SD enoxaparin) 25/2975 (0.84) 10/2988 (0.33) 0.51 (0.12–0.89) 0.011 197
ADOPT (ED apixaban versus SD enoxaparin) 15/3184 (0.47) 6/3217 (0.19) 0.28 (0.00–0.57) 0.047 358
MAGELLAN (ED rivaroxaban versus SD enoxaparin) 43/3997 (1.08) 15/4001 (0.37) 0.70 (0.33–1.07) < 0.001 143
APEX (ED betrixaban versus SD enoxaparin)†
Full-dose 15/2986 (0.50) 16/2991 (0.53)
0.03 (
0.40 to 0.33) 0.86 NI
Reduced-dose 10/730 (1.37) 5/725 (0.69) 0.68 (
0.36 to 1.72) 0.20 NI

CI, confidence interval; DVT, deep vein thrombosis; ED, extended-duration; NI, non-interpretable; NNH, number needed to harm; NNT,
number needed to treat; PE, pulmonary embolism; SD, standard-duration. *NNT or NNH was deemed to be NI for regimens that did not
show a significant risk difference. †Although the majority of APEX study participants did not require dose adjustment, and received full-dose
betrixaban (80 mg once daily), patients with severe renal insufficiency or being treated with a concomitant strong P-glycoprotein inhibitor
received reduced-dose betrixaban (40 mg once daily).

line and 0.00% vertical line represent the superiority betrixaban lay below the horizontal line, indicating that
boundary for efficacy and safety, respectively. The rectan- these agents were superior to standard-duration enoxa-
gles for extended-duration enoxaparin and full-dose parin with respect to VTE risk reduction. The rectangle

© 2017 International Society on Thrombosis and Haemostasis

1918 G. Chi et al
for apixaban, rivaroxaban and reduced-dose betrixaban
crossed the horizontal line, indicating that superiority in
VTE risk reduction was not achieved. All extended-dura-
tion regimens except for reduced-dose betrixaban
achieved non-inferiority with respect to VTE risk reduc-
tion.
With respect to safety, the rectangles for all four trials
either crossed or were on the right of the vertical line,
indicating that these agents failed to achieve superiority
in major bleeding risk reduction. Indeed, extended-dura-
tion enoxaparin, apixaban and rivaroxaban were signifi-
cantly inferior to standard prophylaxis with enoxaparin
with respect to major bleeding risk reduction (the
95% CIs did not overlap and lay to the right of the verti-
cal line). Similarly, the 0.18% horizontal asymptote and
0.50% vertical asymptote represented the non-inferiority
margins for efficacy and safety, respectively. Only full-
dose betrixaban achieved non-inferiority with respect to
major bleeding risk reduction.
In terms of the multiplicative NCB, only the rectangle
for full-dose betrixaban did not include the lack-of-benefit
region, indicating that only extended-duration betrixaban
at 80 mg daily had a favorable risk-benefit profile versus
standard-of-care enoxaparin. When they were evaluated
under the conventional NCB framework, which assumes
a 1 : 1 tradeoff between symptomatic VTE and major
bleeding, none of the extended-duration regimens was
preferable to standard-duration enoxaparin (Fig. S4).
3.00%
2.00%
Venous thromboembolism
1.00%
0.00%
–1.00%
–2.00%
–3.00%
–3.00%
Fig. 3. Bivariate outcome of symptomatic venous thromboembolism and major bleeding among acutely ill hospitalized medical patients. The
part above the dotted curve (multiplicative net clinical benefit) represents the lack-of-benefit region. [Color figure can be viewed at wileyonline-
library.com]
Quantitative assessment of the bivariate outcome
Quantitatively, bivariate outcomes were assessed accord-
ing to the minimum distance from the NCB curve to the
center and opposing corners of the rectangle (Fig. 4).
Extended-duration enoxaparin (0.03%; CD
0.37% to
0.43%), apixaban (
0.20%; CD
0.49% to 0.17%),
rivaroxaban (0.23%; CD
0.16% to 0.69%) and
reduced-dose betrixaban (0.41%; CD
0.85% to 1.94%)
did not significantly improve NCB. Only full-dose betrix-
aban (
0.51%; CD
0.89% to
0.10%) showed a sig-
nificantly favorable risk–benefit profile relative to
standard-duration enoxaparin.
Sensitivity analysis of the non-inferiority margin
Sensitivity analyses with a range of non-inferiority mar-
gins are shown in Figs S5–S20. In the first sensitivity
analysis, in which the safety margin was fixed at 25%,
only full-dose betrixaban showed a favorable NCB pro-
file. In the second sensitivity analysis, in which the effi-
cacy margin was fixed at 9%, extended-duration
enoxaparin, apixaban and full-dose betrixaban showed a
favorable NCB when the non-inferiority safety margin
reached 50%, 40%, and 20%, respectively. In the third
sensitivity analysis, apixaban and full-dose betrixaban
showed a favorable NCB with efficacy and safety margins
of 40% and 20%, respectively.
Risk reduction
Risk reduction
EXCLAIM
ADOPT
MAGELLAN
APEX (Full-dose)
APEX (Reduced-dose)
–2.00% –1.00% 0.00% 1.00% 2.00% 3.00%
Major bleeding
© 2017 International Society on Thrombosis and Haemostasis

EXCLAIM
ADOPT
MAGELLAN
APEX (Full-dose)
APEX (Reduced-dose)
–1.50 –1.00 –0.50 0.00 0.50 1.00 1.50 2.00
Favors
extended-duration
thromboprophylaxis
Fig. 4. Forest plot of the bivariate outcomes (expressed as minimum distance to center [southwest corner and northeast corner]) from the net
clinical benefit curve to the rectangle defined by the 95% confidence interval of risk difference in symptomatic venous thromboembolism and in
major bleeding. [Color figure can be viewed at wileyonlinelibrary.com]
Sensitivity analysis of weighing VTE against major bleeding
A sensitivity analysis evaluating the tradeoff between
VTE (a composite of asymptomatic and symptomatic
DVT, PE, and VTE-related mortality) and major bleeding
is shown in Figs S21 and S22. Extended-duration enoxa-
parin and rivaroxaban were both superior to standard-
duration enoxaparin with respect to efficacy, but did not
achieve non-inferiority with respect to safety, and there-
fore were not favorable with respect to NCB (respectively:

0.21%, CD
0.62% to 0.23%; and
0.01%,
CD
0.41% to 0.45%). Apixaban was non-inferior in
efficacy, was inferior in safety, and was not favorable
with respect to NCB (
0.35%; CD
0.70% to 0.26%).
Betrixaban was superior in efficacy, was non-inferior in
safety, and showed a favorable NCB (
0.61%;
CD
1.00% to
0.20%).
Discussion
Although the risk of VTE may remain elevated for weeks
to months after hospital discharge, the optimal duration
and regimen of thromboprophylaxis for acutely ill hospi-
talized medical patients continue to be debated.
Extended-duration thromboprophylaxis refers to prophy-
laxis that extends beyond the initial course of
10  5 days through approximately 35 days. The risk–
benefit profile of extended-duration treatment with the
novel oral anticoagulants versus standard-of-care enoxa-
parin in the prevention of VTE in medically ill patients
has not been comprehensively evaluated. The present
study excluded asymptomatic VTE events and weighed
symptomatic VTE events against major bleeding events in
order to compare the NCBs of extended-duration enoxa-
parin, apixaban, rivaroxaban, and betrixaban. Extended
prophylaxis with full-dose betrixaban showed a favorable
NCB than standard-duration enoxaparin, with a signifi-
cant favorable risk difference of
0.51% (
0.89% to
© 2017 International Society on Thrombosis and Haemostasis
Bivariate analysis and net clinical benefit 1919
Bivariate outcome
0.03 (–0.37 to 0.43)
–0.20 (–0.49 to 0.17)
0.23 (–0.16 to 0.69)
–0.51 (–0.89 to –0.10)
0.41 (–0.85 to 1.94)
Favors standard-duration
enoxaparin

0.10%) in a bivariate analysis. Among the regimens
that failed to achieve a favorable NCB, extended-duration
enoxaparin was superior in efficacy but inferior in safety,
whereas apixaban and rivaroxaban were both non-inferior
in efficacy and inferior in safety. Reduced-dose betrixaban
was administered to a subgroup of patients with severe
renal insufficiency or being treated with a concomitant
strong P-glycoprotein inhibitor, and did not show a sig-
nificant difference in efficacy or safety.
Bivariate analysis is a novel statistical method that
allows simultaneous assessment of efficacy and safety end-
points based upon clinical relevant and potentially asym-
metric statistical margins for efficacy and safety. The
output includes a qualitative comparison of the efficacy–
safety rectangle of the 95% CI with the NCB curve dis-
played on the risk–benefit plane based on a structured
definition of non-inferiority, and a quantitative compar-
ison of risk reduction in the bivariate outcome illustrated
in a forest plot based on the minimum distance from the
NCB curve to the rectangle. The qualitative and quantita-
tive inferences may be affected by the choice of non-infer-
iority margin. Therefore, a standardized method for
determining the appropriate margin is crucial for estab-
lishing the NCB [14]. A sensitivity analysis in which the
non-inferiority bound is varied can be performed to
examine the robustness of the bivariate outcome. In the
present sensitivity analyses, in which a clinically impor-
tant risk difference was set at 2.00% for both symp-
tomatic VTE and major bleeding, full-dose betrixaban
consistently achieved a favorable NCB when the safety
margin was set at ≥ 20%. In other words, if a 0.40%
threshold of risk difference in major bleeding is clinically
acceptable, full-dose betrixaban for 35–42 days is pre-
ferred over enoxaparin for 10  4 days with respect to
NCB.
For the currently available extended-duration thrombo-
prophylaxis to reach a favorable NCB would require
reduction in symptomatic VTE risk while an acceptable
1920 G. Chi et al
rate of major bleeding is maintained. The advantage of a
bivariate analysis is that it does not assume a linear or
constant tradeoff in efficacy versus bleeding across a wide
range of efficacy and bleeding event rates, and allows for
asymmetry in the statistical margins for efficacy and
bleeding. It yields one number that can be used to com-
pare regimens across trials (the bivariate outcome). Given
the inherent risk of bleeding associated with anticoagu-
lants, it is difficult at this time to believe that an anticoag-
ulation regimen would be superior to the standard of care
with respect to not only efficacy but also safety outcomes.
Indeed, extended-duration enoxaparin, apixaban and
rivaroxaban were significantly inferior to standard-dura-
tion enoxaparin with respect to major bleeding. Betrixa-
ban, however, has unique pharmacokinetic properties that
differentiate it from these agents, including a longer half-
life with a lower peak-to-trough plasma concentration
ratio and the lowest renal clearance among any of these
agents, which may improve the safety profile in an elderly
population such as acutely ill hospitalized medical
patients [17,18]. Perhaps it is, at least in part, because of
these unique pharmacokinetic characteristics that full-
dose betrixaban was non-inferior to standard-duration
enoxaparin with respect to major bleeding.
Although this study was performed after trial comple-
tion as a post hoc exploratory analysis, the bivariate
method has been applied in clinical trial design and
interim analyses. In these circumstances, the efficacy and
safety components of the bivariate outcome are cus-
tomized according to their clinical impact. For instance,
in the Kids-DOTT trial (ClinicalTrials.gov identifier:
NCT00687882), which aims to evaluate the duration of
anticoagulation in children with first-episode acute venous
thrombosis, a bivariate endpoint is used as the primary
outcome measure, in which the efficacy component is the
risk of symptomatic, radiologically confirmed recurrent
VTE, and the safety component is clinically relevant
bleeding. In addition, the bivariate method has been used
in devising the stopping criteria for the interim analyses
[6]. In the MARINER trial (ClinicalTrials.gov identifier:
NCT02111564), which compares the efficacy and safety of
rivaroxaban versus placebo for VTE prevention in high-
risk medically ill patients, the prespecified analyses
include the bivariate approach to assess the risk–benefit
profile with symptomatic VTE or VTE-related death as
the efficacy component and major bleeding as the safety
component [7]. In the present context of acutely ill hospi-
talized medical patients, although the primary efficacy
endpoint of recent trials was the composite of asymp-
tomatic and symptomatic DVT, PE, and VTE-related
death, the clinical relevance of the asymptomatic DVT
remains controversial [19]. It is for this reason that only
symptomatic events were included in the present analysis,
and the treatment benefit of reducing symptomatic VTE
events was weighed against major bleeding in light of
their comparable clinical significance and mortality rates.
Future studies are warranted to systematically assess the
clinical impact of both asymptomatic DVT and non-
major bleeding events in this population.
Conclusions
Bivariate analysis is a novel method that allows collective
assessment of efficacy and safety endpoints with consider-
ations on the non-inferiority margins. In the management
of acutely ill hospitalized medical patients, extended pro-
phylaxis with full-dose betrixaban was superior to stan-
dard-duration enoxaparin, whereas other regimens failed
to simultaneously achieve both superiority and non-infer-
iority with respect to symptomatic VTE and major bleed-
ing.
Addendum
G. Chi, A. T. Cohen, R. A. Harrington, and C. M. Gib-
son designed the study. G. Chi performed the analysis. G.
Chi, S. Z. Goldhaber, A. T. Cohen, R. A. Harrington,
and C. M. Gibson interpreted results and wrote the
manuscript. J. M. Kittelson, A. G. G. Turpie, A. F. Her-
nandez, R. D. Hull, A. Gold, and J. T. Curnutte critically
revised the manuscript. All authors approved the final
manuscript.
Disclosure of Conflict of Interests
The study was funded by Portola Pharmaceuticals Inc. G.
Chi has received research grant support paid to the Beth
Israel Deaconess Medical Center from Portola, Bayer,
and Janssen Research. S. Z. Goldhaber has provided con-
sulting for Boehringer Ingelheim, Bayer, Portola, Daiichi-
Sankyo, Janssen, BiO2 Medical, EKOS/BTG, BMS, and
Zafgen. J. M. Kittelson receives partial salary support
from the University of Colorado through grants from the
National Institutes of Health. He has received consulting
fees from Bayer Healthcare for advisory and steering
committee activities. He has received honoraria for partic-
ipation in US Food and Drug Administration activities,
and has received consulting fees for work on data and
safety monitoring committees from Cystic Fibrosis Foun-
dation Therapeutics, Novo Nordisk, Genentech, and Bio-
Marin pharmaceuticals. He receives honoraria from CPC
Clinical Research (a non-profit academic research organi-
zation affiliated with the University of Colorado) as a
member of the Antithrombotic Trials Leadership and
Steering (ATLAS) Group (atlasresearch.org). A. G. G.
Turpie has received speaker’s honoraria and consultancy
fees from, and participated in scientific advisory boards
for, Bayer and Janssen Research & Development, LLC.
A. F. Hernandez reports receipt of grant support from
Amgen, AstraZeneca, Bayer, Bristol-Myers Squibb,
GlaxoSmithKline, Luitpold, Merck, and Novartis, and
personal fees from Amgen, AstraZeneca, Bayer,
© 2017 International Society on Thrombosis and Haemostasis

Bristol-Myers Squibb, Boston Scientific, Luitpold, and
Novartis outside the submitted work. R. D. Hull reports
receiving grant support from Portola Pharmaceuticals
during the conduct of the study, and grant support and
personal fees from Leo Pharma outside the submitted
work. A. Gold and J. T. Curnutte report receiving per-
sonal fees from Portola Pharmaceuticals outside the sub-
mitted work. A. T. Cohen reports receiving grant
support, personal fees and non-financial support from
Portola Pharmaceuticals during the conduct of the study,
and grant support, personal fees and non-financial sup-
port from Daiichi-Sankyo, Bristol-Myers Squibb, Pfizer,
Janssen, and Bayer Pharmaceuticals, personal fees from
Boehringer Ingelheim and Sanofi, and personal fees and
non-financial support from Johnson & Johnson and
Aspen Pharmaceuticals outside the submitted work. R. A.
Harrington reports receiving grant support from Portola
Pharma during the conduct of the study, grant support
from CSL Behring, AstraZeneca, GlaxoSmithKline,
Regado, and Sanofi Aventis, grant support and personal
fees from Merck and The Medicines Company, personal
fees from Amgen, Gilead Sciences, MyoKardia, and
WebMD, and other support from Scanadu, SignalPath,
Element Science, Vida Health, and Adverse Events out-
side the submitted work. C. M. Gibson reports receiving
grant support from Portola Pharmaceuticals during the
conduct of the study, and grant support from Johnson &
Johnson and Bayer outside the submitted work.
Supporting Information
Additional Supporting Information may be found in the
online version of this article:
Fig. S1. Conventional net clinical benefit with a linear,
symmetric, and unlimited risk–benefit tradeoff of 1 : 1
ratio (for example, a 1-unit improvement in efficacy end-
point is offset by a 1-unit worsening in safety endpoint).
Fig. S2. Conventional net clinical benefit with a linear,
symmetric, and unlimited risk–benefit tradeoff of 2 : 1
ratio (for example, a 1-unit improvement in efficacy end-
point is offset by a 2-unit worsening in safety endpoint).
Fig. S3. Net clinical benefit with a non-linear and asym-
metric risk–benefit tradeoff defined by the acceptable
worsening in safety/efficacy endpoint when clinically
meaningful improvement in efficacy/safety endpoint is
achieved.
Fig. S4. Conventional net clinical benefit with a 1 : 1
tradeoff between symptomatic VTE and major bleeding
among acutely ill hospitalized medical patients.
Fig. S5. Sensitivity analysis of the non-inferiority margin.
Fig. S6. Sensitivity analysis 1: EXCLAIM.
Fig. S7. Sensitivity analysis 2: EXCLAIM.
Fig. S8. Sensitivity analysis 3: EXCLAIM.
Fig. S9. Sensitivity analysis 1: ADOPT.
Fig. S10. Sensitivity analysis 2: ADOPT.
© 2017 International Society on Thrombosis and Haemostasis
Bivariate analysis and net clinical benefit 1921
Fig. S11. Sensitivity analysis 3: ADOPT.
Fig. S12. Sensitivity analysis 1: MAGELLAN.
Fig. S13. Sensitivity analysis 2: MAGELLAN.
Fig. S14. Sensitivity analysis 3: MAGELLAN.
Fig. S15. Sensitivity analysis 1: APEX (full-dose).
Fig. S16. Sensitivity analysis 2: APEX (full-dose).
Fig. S17. Sensitivity analysis 3: APEX (full-dose).
Fig. S18. Sensitivity analysis 1: APEX (reduced-dose).
Fig. S19. Sensitivity analysis 2: APEX (reduced-dose).
Fig. S20. Sensitivity analysis 3: APEX (reduced-dose).
Fig. S21. Bivariate outcome of VTE and major bleeding
among acutely ill hospitalized medical patients.
Fig. S22. Forest plot of the bivariate outcomes (expressed
as the minimum distance from the net clinical benefit
curve to the rectangle defined by the 95% confidence
interval of risk difference in VTE and in major bleeding).
Data S1. Supplementary data.
References
1 Kahn SR, Lim W, Dunn AS, Cushman M, Dentali F, Akl EA,
Cook DJ, Balekian AA, Klein RC, Le H, Schulman S, Murad
MH, American College of Chest Physicians. Prevention of VTE
in nonsurgical patients: Antithrombotic Therapy and Prevention
of Thrombosis, 9th ed: American College of Chest Physicians
Evidence-Based Clinical Practice Guidelines. Chest 2012; 141:
e195S–226S.
2 Spencer FA, Lessard D, Emery C, Reed G, Goldberg RJ.
Venous thromboembolism in the outpatient setting. Arch Intern
Med 2007; 167: 1471–5.
3 Liew AY, Piran S, Eikelboom JW, Douketis JD. Extended-dura-
tion versus short-duration pharmacological thromboprophylaxis
in acutely ill hospitalized medical patients: a systematic review
and meta-analysis of randomized controlled trials. J Thromb
Thrombolysis 2016; 43: 291–301.
4 Kittelson JM, Spyropoulos AC, Halperin JL, Kessler CM, Schul-
man S, Steg G, Turpie AG, Cutler NR, Hiatt WR, Goldenberg
NA. Balancing risk and benefit in venous thromboembolism tri-
als: concept for a bivariate endpoint trial design and analytic
approach. J Thromb Haemost 2013; 11: 1443–8.
5 Singer DE, Chang Y, Fang MC, Borowsky LH, Pomernacki
NK, Udaltsova N, Go AS. The net clinical benefit of warfarin
anticoagulation in atrial fibrillation. Ann Intern Med 2009; 151:
297–305.
6 Goldenberg NA, Abshire T, Blatchford PJ, Fenton LZ, Halperin
JL, Hiatt WR, Kessler CM, Kittelson JM, Manco-Johnson MJ,
Spyropoulos AC, Steg PG, Stence NV, Turpie AG, Schulman S,
Kids DTI. Multicenter randomized controlled trial on Duration
of Therapy for Thrombosis in Children and Young Adults (the
Kids-DOTT trial): pilot/feasibility phase findings. J Thromb Hae-
most 2015; 13: 1597–605.
7 Raskob GE, Spyropoulos AC, Zrubek J, Ageno W, Albers G,
Elliott CG, Halperin J, Haskell L, Hiatt WR, Maynard GA,
Peters G, Spiro T, Steg PG, Suh EY, Weitz JI. The MARINER
trial of rivaroxaban after hospital discharge for medical patients
at high risk of VTE. Design, rationale, and clinical implications.
Thromb Haemost 2016; 115: 1240–8.
8 Kittelson JM, Steg PG, Halperin JL, Goldenberg NA, Schulman
S, Spyropoulos AC, Kessler CM, Turpie AG, Cutler NR, Hiatt
WR. Bivariate evaluation of thromboembolism and bleeding in
clinical trials of anticoagulants in patients with atrial fibrillation.
Thromb Haemost 2016; 116: 544–53.
1922 G. Chi et al
9 Hull RD, Schellong SM, Tapson VF, Monreal M, Samama
MM, Nicol P, Vicaut E, Turpie AG, Yusen RD. Extended-dura-
tion venous thromboembolism prophylaxis in acutely ill medical
patients with recently reduced mobility: a randomized trial. Ann
Intern Med 2010; 153: 8–18.
10 Goldhaber SZ, Leizorovicz A, Kakkar AK, Haas SK, Merli G,
Knabb RM, Weitz JI. Apixaban versus enoxaparin for thrombo-
prophylaxis in medically ill patients. N Engl J Med 2011; 365:
2167–77.
11 Cohen AT, Spiro TE, B€ uller HR, Haskell L, Hu D, Hull R,
Mebazaa A, Merli G, Schellong S, Spyropoulos AC, Tapson V.
Rivaroxaban for thromboprophylaxis in acutely ill medical
patients. N Engl J Med 2013; 368: 513–23.
12 Cohen AT, Harrington RA, Goldhaber SZ, Hull RD, Wiens BL,
Gold A, Hernandez AF, Gibson CM. Extended thromboprophy-
laxis with betrixaban in acutely ill medical patients. N Engl J
Med 2016; 375: 534–44.
13 Dahan R, Houlbert D, Caulin C, Cuzin E, Viltart C, Woler M,
Segrestaa JM. Prevention of deep vein thrombosis in elderly
medical in-patients by a low molecular weight heparin: a ran-
domized double-blind trial. Haemostasis 1986; 16: 159–64.
14 Samama MM, Cohen AT, Darmon JY, Desjardins L, Eldor A,
Janbon C, Leizorovicz A, Nguyen H, Olsson CG, Turpie AG,
Weisslinger N. A comparison of enoxaparin with placebo for the
prevention of venous thromboembolism in acutely ill medical
patients. Prophylaxis in Medical Patients with Enoxaparin Study
Group. N Engl J Med 1999; 341: 793–800.
15 Lederle FA, Sacks JM, Fiore L, Landefeld CS, Steinberg N,
Peters RW, Eid AA, Sebastian J, Stasek JE Jr, Fye CL. The pro-
phylaxis of medical patients for thromboembolism pilot study.
Am J Med 2006; 119: 54–9.
16 DerSimonian R, Laird N. Meta-analysis in clinical trials. Control
Clin Trials 1986; 7: 177–88.
17 Connolly SJ, Eikelboom J, Dorian P, Hohnloser SH, Gretler
DD, Sinha U, Ezekowitz MD. Betrixaban compared with war-
farin in patients with atrial fibrillation: results of a phase 2, ran-
domized, dose-ranging study (Explore-Xa). Eur Heart J 2013; 34:
1498–505.
18 Chan NC, Bhagirath V, Eikelboom JW. Profile of betrixaban
and its potential in the prevention and treatment of venous
thromboembolism. Vasc Health Risk Manag 2015; 11: 343–
51.
19 Lloyd NS, Douketis JD, Moinuddin I, Lim W, Crowther MA.
Anticoagulant prophylaxis to prevent asymptomatic deep vein
thrombosis in hospitalized medical patients: a systematic review
and meta-analysis. J Thromb Haemost 2008; 6: 405–14.
© 2017 International Society on Thrombosis and Haemostasis