Keywords
venous thromboembolism - anticoagulants - cancer
Introduction
Patients with cancer-associated venous thromboembolism (CT) have a higher risk of
bleeding, venous thromboembolism (VTE) recurrences, and all-cause mortality compared
with those without cancer.[1]
[2]
[3] Guidelines list direct oral anticoagulants (DOACs) and low-molecular-weight heparins
(LMWHs) as options for the treatment of CT and the secondary prevention of VTE recurrences.[4]
[5]
[6] The strength of recommendation for DOACs is based on efficacy and safety data from
randomized controlled trials comparing DOAC to LMWH or vitamin K antagonists (VKAs).[7] Observational studies have also investigated the bleeding risk in patients treated
with DOACs for CT. Some of these studies lack a comparison with LMWHs, lack information
on cancer type, or include cancer types not recommended for DOAC treatment.[8]
[9] Overall, many studies had a lack of study power because of a small number of recurrent
VTE and bleeding events, affecting, amongst other things, the analysis of recurrent
VTE and bleeding events over time.
Our primary objective was to evaluate with an intention-to-treat (ITT) analysis the
effectiveness (recurrent VTE) and safety (significant bleeds and all-cause mortality)
of rivaroxaban versus LMWH for CT treatment in patients with active cancer excluding
those with a malignant neoplasm associated with a high risk of bleeding. The recognition
of patients with CT at high risk of bleeding was guided by expert interpretation of
the International Society on Thrombosis and Haemostasis (ISTH) guidance.[6] The Observational Studies in Cancer Associated Thrombosis for Rivaroxaban in the
United Kingdom Cohort (OSCAR-UK) is part of the OSCAR program with independent studies
in the United States,[10] United Kingdom,[11] and Sweden.[12]
Methods
Study Design and Data Source
This was a retrospective, observational cohort study of patients with active cancer
and incident CT subsequently treated with anticoagulants (AC).
Data were obtained from the UK Clinical Practice Research Datalink (CPRD) GOLD and
Aurum databases with linkage to inpatient and outpatient data from the Hospital Episodes
Statistics (HES) and the Office for National Statistics (ONS) mortality data. CPRD
includes patient demographics, lifestyle factors, medical diagnoses, and symptoms
recorded with primary care Read medical and Systematized Nomenclature of Medicine–Clinical
Terms (SNOMED CT) codes, referrals to secondary care, test results, and general practitioner
(GP) prescriptions. HES data include hospital admission and discharge dates, discharge
diagnoses recorded with ICD-10 codes (International Classification of Diseases, Tenth
Revision), and surgical operations and procedures (OPCS-4 codes). ONS mortality data
consist of date and cause of death (ICD-10 codes).
Selection of Participants
The study population consisted of all patients 18 years or older from English practices
in the CPRD that were eligible for linkage to HES and ONS data in the study period,
between January 1, 2013 and October 31, 2020. The study cohort included all eligible
patients in CPRD with an incident VTE and active cancer, and with evidence of therapeutic
rivaroxaban or LMWH use within 30 days following the VTE. VTE was identified according
to our previously developed, validated, and published algorithm.[11] VTE comprised pulmonary embolisms and deep vein thromboses (DVTs). DVT included
thromboses of the deep veins of the legs, calf vein thromboses, thromboses of pelvic
veins and vena cava as well as thromboses of the upper limb. Cerebral and abdominal
vein thrombi were not included.
Active cancer was defined as a cancer being diagnosed within 180 days prior to the
index CT, or associated with metastatic disease (regardless of time from initial cancer
diagnosis), or 180 days following the patient receiving anticancer therapy. To generate
a cohort of patients with incident CT and new users of ACs, we excluded patients with:
less than 1-year contribution to the CPRD-HES-ONS link before CT, and those with a
prior history of VTE (including cerebral and abdominal vein thrombi), insertion of
an inferior vena cava filter, prior therapeutic AC use, other indications for long-term
AC use (e.g., atrial fibrillation or artificial heart valves), thrombocytopenia, end-stage
kidney disease, recent hip or knee preplacement (35 days), active pregnancy, or a
recording indicative of palliative care initiation before the CT. Patients with VKA
use, a significant bleeding event, or a VTE recurrence between the initial CT and
the initiation of rivaroxaban/LMWH were also excluded. Cancer types for which use
of DOACs is endorsed by interpretation of the ISTH guidance were considered, thus
patients with the following cancer types were excluded from the study cohort: non-brain
central nervous system (i.e., spinal cord tumors), unresected colorectal/lower gastrointestinal
tract, hematologic (except lymphoma and myeloma), esophagus, stomach, and bladder.[6]
Observational Period
The day of the incident CT was designated as the cohort entry day and the day of the
first recording of therapeutic rivaroxaban or LMWH initiation within 30 days after
the acute CT was designated as the index day. When the rivaroxaban/LMWH initiation
was recorded during the initial CT hospitalization, the index day was shifted to the
first day after hospital discharge. As in-hospital pharmacy data, including group
and type of AC, are not systematically recorded in the linked HES database, we used
the first postdischarge prescription of an AC to determine the patient's initial type
of AC use. The 30-day period after hospital discharge was used to allow for any supply
of AC supplied by the hospital but not recorded by the GP.
The observational period started on the index day and ended on the first of the following
events: end of the study period (October 31, 2020), 1 year after CT, patient transferred
out of GP practice, end of data collection of GP practice, initiation of palliative
care, end of active cancer episode, cerebral or abdominal vein thrombus, first atrial
fibrillation recording or artificial heart valve insertion, patient became pregnant,
patient developed a study outcome.
Exposure
Exposure of interest was AC use with rivaroxaban or LMWH in a therapeutic dose. The
primary analysis was an ITT approach. In this approach switching and discontinuation
of AC treatment during the observational period was ignored. In a sensitivity analysis,
an on-treatment approach was performed whereby the observational period ended when
a patient discontinued the initial AC treatment or switched the AC. Patients switching
AC type before the start of the at-risk period from LMWH to VKA/other parenteral AC/a
DOAC other than rivaroxaban, or from rivaroxaban to LMWH/VKA/other parenteral AC/other
DOAC were excluded. Patients switching from LMWH to rivaroxaban within the first 7
days after initial LMWH treatment were allocated to the rivaroxaban group with the
day of rivaroxaban as start of at-risk period. Patients switching from LMWH to DOAC
other than rivaroxaban within 7 days after first LMWH record were removed from the
cohort.
Outcome
The outcomes of interest were VTE recurrences, significant bleeds defined as major
bleeds, or clinically relevant nonmajor bleeding requiring hospitalization (CRNMB-H),
and all-cause mortality.[12]
[13] Algorithms for the definition of recurrent VTE and of the bleeding events have previously
been developed and refined using all information available in CPRD, HES, and ONS.[2]
[11]
[14] All identified potential VTE recurrences and significant bleeds were manually reviewed
by three physicians (A.T.C., S.C., and C.M.). During the outcome reviews, the physicians
were blinded to AC treatment type. All-cause mortality was identified from ONS death
certificates. Duration of anticoagulation treatment (a secondary outcome) with rivaroxaban
and LMWH use was defined as time on continuous treatment with the respective medication
from initiation to discontinuation.
Covariates
Covariates included variables intended for description of the study cohort (including
cancer type and treatment), variables potentially related with choice of AC type in
CT patients (required for the determination of probability weights), and known or
suspected risk factors for VTE recurrences, significant bleeds, and death from any
cause (potential confounders).
Covariate groups were not mutually exclusive and consisted of demographics, comorbidities,
comedications, laboratory values, and vital signs. Clinical conditions were defined
from medical codes entered by GPs (Read and SNOMED CT codes), hospital discharge diagnoses
and procedures (ICD and OPCS codes), medication use derived from GP-issued prescriptions
(Gemscript and DM + D codes), and test results recorded by the GP. A full list of
baseline characteristics and covariates is included in the [Supplementary Material] (available in the online version).
Data Analysis
Baseline characteristics at cohort entry were described separately for rivaroxaban-
and LMWH-treated patients using numbers (proportions) for categorical variables and
mean (standard deviation) for continuous variables.
To adjust for potential confounding between the rivaroxaban and LMWH cohort, probabilities
for rivaroxaban initiation were estimated from multivariate logistic regression models
based on covariates identified at cohort entry (baseline). Covariates were only included
in the model when ≥3 patients were exposed to the covariate in each exposure group.
These probabilities were then used to assign weights to all individual patients in
the rivaroxaban and LMWH groups using the overlap weighting method, i.e., patients
were weighted with the probability of belonging to the opposite treatment group.[15]
[16] By design, overlap weighting resulted in the exact balance of all variables included
in the logistic regression model in the two exposure groups rivaroxaban and LMWH.
Crude incidence rates of recurrent VTE, significant bleeds, and death within 3, 6,
and 12 months following CT were calculated in the rivaroxaban and LMWH groups separately
before and after weighting. Univariate Fine and Gray regression models, accounting
for competing risks using AC exposure (i.e., LMWH) as the independent variable, were
used to estimate sub-distribution hazard ratios (SHRs) for VTE recurrence, significant
bleeds, and all-cause mortality at 3, 6, and 12 months following CT separately. Models
were performed with and separately without (unadjusted) overlap weighting. Competing
risks for each study outcome were the other two study outcomes, e.g., significant
bleeds and all-cause mortality for the analysis of VTE recurrences. The proportional
hazards assumption was investigated using Schoenfeld residuals.[17] Missing data were allocated to a category “unknown.”
In a sensitivity analysis, an on-treatment approach instead of an ITT approach was
used, i.e., patients who discontinued AC treatment or switched to a different AC type,
(e.g., from rivaroxaban to VKA) were censored. In addition, the duration of anticoagulation
use following initial CT was described for rivaroxaban and LMWH groups separately
using a competing risk approach and overlap weighting. In an exploratory analysis,
critical organ bleeds (a subgroup of major bleeds) were investigated as a separate
study outcome.
All statistical procedures were performed using Stata MP Version 14.2 (StataCorp LLC).
The study protocol was approved by CPRD's Research Data Governance process (Protocol
ID 21_000514). This study was registered at the European Union electronic Register
of Post-Authorisation Studies (EU PAS Register; EUPAS43329).
Results
Population Characteristics
A total of 5,642 anticoagulation-naïve adult patients with active cancer and CT, with
at least 1 year of history in CPRD/HES and treated with either rivaroxaban or LMWH,
were identified between 2013 and 2020. Of those, 3,383 patients were excluded due
to indications for anticoagulation use other than VTE, contraindications for rivaroxaban
use, initiation of palliative care before the start of the at-risk period, or an unknown
or non-ISTH-guided type of initial cancer. From the remaining 2,259 patients, 314
were treated with rivaroxaban, and 1,945 with LMWH ([Fig. 1]) within 30 days following the CT diagnosis.
Fig. 1 Ascertainment of rivaroxaban versus LMWH-treated CT cohort. AC, anticoagulant; CT,
cancer-associated VTE; CPRD, Clinical Practice Research Datalink; DOAC, direct oral
anticoagulant; DVT, deep vein thrombosis; HES, Hospital Episode Statistics; ISTH,
International Society on Thrombosis and Haemostasis; LMWH, low-molecular-weight heparin;
VTE, venous thromboembolism. aAtrial fibrillation, cardiac valve replacement, unusual site DVT (cerebral and abdominal
vein thrombi), or hip or knee replacement in the last month. bAt risk period: starts on day of first rivaroxaban/LMWH recording after CT but not
earlier than 1 day after VTE hospital discharge or 1 day after day of general practitioner
recording of VTE. cThrombocytopenia, active pregnancy, or end-stage kidney disease. dIncluding the following cancer types: non-brain central nervous system, unresected
colorectal, leukemia, other hematologic, esophagus, stomach, and bladder. ePatients switching from LMWH to rivaroxaban within the first 7 days after initial
LMWH treatment were allocated to the rivaroxaban group with the day of rivaroxaban
as start of at-risk period. Patients switching from LMWH to DOAC other than rivaroxaban
within 7 days after first LMWH record were removed from the cohort.
Rivaroxaban users were older, more likely to be males, less likely to be smokers,
and socioeconomically deprived. Cancer types varied in the two AC exposure cohorts.
Breast and prostate cancers were more prevalent in rivaroxaban users, while gastrointestinal
tract, lung, and cancers in “other” sites (other than the 12 specified sites) were
more prevalent in the LMWH users ([Table 1]).
Table 1
Baseline characteristics (prior to weighting) of study cohort by type of anticoagulant
|
Rivaroxaban, n (%)
|
LMWH, n (%)
|
|
Total
|
314
|
1,945
|
|
Age[a] [years]
|
|
Mean (SD)
|
72.4 (12.1)
|
66.9 (11.7)
|
|
Median (p25–p75)
|
73 (65–81)
|
68 (59–75)
|
|
<18
|
0 (0.0)
|
0 (0.0)
|
|
≥18 to <65
|
64 (20.4)
|
716 (36.8)
|
|
≥65 to <75
|
106 (33.8)
|
703 (36.1)
|
|
≥75 to <85
|
92 (29.3)
|
436 (22.4)
|
|
≥85
|
52 (16.6)
|
90 (4.6)
|
|
Gender
|
|
Male
|
148 (47.1)
|
802 (41.2)
|
|
Female
|
166 (52.9)
|
1,143 (58.8)
|
|
BMI[b] [kg/m2]
|
|
Known BMI
|
305 (97.1)
|
1,856 (95.4)
|
|
Mean (SD)
|
27.8 (5.4)
|
28.1 (5.7)
|
|
Median (p25–p75)
|
27 (24–31)
|
27 (24–31)
|
|
<18.5
|
7 (2.3)
|
40 (2.2)
|
|
≥18.5 to <25
|
83 (27.2)
|
545 (29.4)
|
|
≥25 to <30
|
123 (40.3)
|
690 (37.2)
|
|
≥30 to <35
|
67 (22.0)
|
376 (20.3)
|
|
≥35
|
25 (8.2)
|
205 (11.0)
|
|
Unknown BMI
|
9 (2.9)
|
89 (4.6)
|
|
Smoking status[b]
|
|
Known smoking status
|
313 (99.7)
|
1,929 (99.2)
|
|
Never
|
130 (41.5)
|
776 (40.2)
|
|
Ex
|
162 (51.8)
|
939 (48.7)
|
|
Current
|
21 (6.7)
|
214 (11.1)
|
|
Unknown smoking status
|
1 (0.3)
|
16 (0.8)
|
|
Socioeconomic status[c] [quintile]
|
|
1st (least deprived)
|
87 (27.7)
|
480 (24.7)
|
|
2nd
|
72 (22.9)
|
421 (21.6)
|
|
3rd
|
71 (22.6)
|
383 (19.7)
|
|
4th
|
47 (15.0)
|
330 (17.0)
|
|
5th
|
37 (11.8)
|
331 (17.0)
|
|
Type of first VTE
|
|
DVT only
|
99 (31.5)
|
612 (31.5)
|
|
PE only
|
200 (63.7)
|
1,235 (63.5)
|
|
PE and DVT
|
15 (4.8)
|
98 (5.0)
|
|
Cancer type
|
|
Breast
|
104 (33.1)
|
437 (22.5)
|
|
Brain
|
7 (2.2)
|
94 (4.8)
|
|
Gastrointestinal tract[d]
|
16 (5.1)
|
258 (13.3)
|
|
Lymphoma
|
13 (4.1)
|
91 (4.7)
|
|
Myeloma
|
5 (1.6)
|
37 (1.9)
|
|
Head or neck
|
3 (1.0)
|
32 (1.6)
|
|
Lung
|
29 (9.2)
|
301 (15.5)
|
|
Malignant melanoma
|
13 (4.1)
|
34 (1.7)
|
|
Ovarian
|
6 (1.9)
|
102 (5.2)
|
|
Prostate
|
88 (28.0)
|
197 (10.1)
|
|
Kidney/other urinary
|
5 (1.6)
|
53 (2.7)
|
|
Cervix/uterus
|
10 (3.2)
|
90 (4.6)
|
|
Other[e]
|
15 (4.8)
|
219 (11.3)
|
Abbreviations: BMI, body mass index; CT, cancer-associated VTE; DVT, deep vein thrombosis;
LMWH, low-molecular-weight heparin; p, percentile; PE, pulmonary embolism; SD, standard
deviation; VTE, venous thromboembolism.
a At the day of CT.
b Latest information available before the day of CT.
c Defined by index for multiple deprivation 2015 data.
d Resected colorectal, other lower gastrointestinal tract, or small intestine.
e Including, e.g., hepatobiliary, pancreas, and testicular.
Main Outcomes
Recurrent VTE
A total of 66 and 10 incident recurrent VTE events were identified with LMWH and rivaroxaban
use respectively in the first year after the initial CT ([Table 2]). Crude incidence rates of recurrent VTE in the first year after the initial CT
were 6.2 (95% confidence interval [CI]: 4.8–8.0) and 5.4 (95% CI: 2.6–10.0) per 100
person-years for LMWH and rivaroxaban use, respectively. The weighted SHR for VTE
recurrences in rivaroxaban compared with LMWH at 12 months was 0.80 (95% CI: 0.37–1.73).
At 3, 6, and 12 months after the CT, the incidence rates and SHRs are shown in [Table 2] and weighted survival probabilities in the first 12 months after CT are shown in
[Supplementary Fig. S1] (available in the online version).
Table 2
VTE recurrences, bleeding, and mortality at 3, 6, and 12 months after CT (sub-distribution
hazard ratios) in the intention-to-treat population
|
Time since CT
|
Events
|
Person-years
|
Incidence rate[a]
(0.95 CI)
|
Unweighted SHR[b]
(0.95 CI)
|
Overlap weighted[c] SHR[b] (0.95 CI)
|
|
VTE recurrences
|
|
3 months
|
|
LMWH
|
19
|
341
|
5.6 (3.3–8.8)
|
1
|
1
|
|
Rivaroxaban
|
3
|
57
|
5.3 (1.0–15.4)
|
0.95 (0.28–3.19)
|
0.96 (0.25–3.74)
|
|
6 months
|
|
LMWH
|
34
|
655
|
5.2 (3.5–7.3)
|
1
|
1
|
|
Rivaroxaban
|
6
|
111
|
5.4 (1.9–11.8)
|
1.04 (0.44–2.47)
|
1.31 (0.47–3.67)
|
|
12 months
|
|
LMWH
|
66
|
1,057
|
6.2 (4.8–8.0)
|
1
|
1
|
|
Rivaroxaban
|
10
|
184
|
5.4 (2.6–10.0)
|
0.90 (0.46–1.75)
|
0.80 (0.37–1.73)
|
|
All significant bleeds
|
|
3 months
|
|
LMWH
|
46
|
341
|
13.5 (9.8–18.1)
|
1
|
1
|
|
Rivaroxaban
|
10
|
57
|
17.5 (8.4–32.3)
|
1.30 (0.65–2.59)
|
1.03 (0.44–2.40)
|
|
6 months
|
|
LMWH
|
74
|
655
|
11.3 (8.8–14.2)
|
1
|
1
|
|
Rivaroxaban
|
14
|
111
|
12.6 (6.8–21.1)
|
1.13 (0.64–2.01)
|
0.85 (0.43–1.71)
|
|
12 months
|
|
LMWH
|
102
|
1,057
|
9.7 (7.8–11.8)
|
1
|
1
|
|
Rivaroxaban
|
20
|
184
|
10.9 (6.6–16.8)
|
1.17 (0.73–1.89)
|
1.01 (0.57–1.81)
|
|
Major bleeds
|
|
3 months
|
|
LMWH
|
23
|
341
|
6.8 (4.2–10.2)
|
1
|
1
|
|
Rivaroxaban
|
2
|
57
|
3.5 (0.4–12.7)
|
0.52 (0.12–2.20)
|
0.37 (0.08–1.76)
|
|
6 months
|
|
LMWH
|
31
|
655
|
4.7 (3.2–6.8)
|
1
|
1
|
|
Rivaroxaban
|
3
|
111
|
2.7 (0.5–7.9)
|
0.58 (0.18–1.90)
|
0.40 (0.11–1.44)
|
|
12 months
|
|
LMWH
|
39
|
1,057
|
3.7 (2.6–5.1)
|
1
|
1
|
|
Rivaroxaban
|
3
|
184
|
1.6 (0.3–4.8)
|
0.46 (0.14–1.50)
|
0.35 (0.10–1.24)
|
|
Critical organ bleeds[d]
|
|
3 months
|
|
LMWH
|
12
|
341
|
3.5 (1.8–6.2)
|
1
|
1
|
|
Rivaroxaban
|
1
|
57
|
1.8 (0.0–9.8)
|
0.50 (0.06–3.87)
|
0.49 (0.05–4.39)
|
|
6 months
|
|
LMWH
|
17
|
655
|
2.6 (1.5–4.2)
|
1
|
1
|
|
Rivaroxaban
|
2
|
111
|
1.8 (0.2–6.5)
|
0.71 (0.16–3.09)
|
0.52 (0.10–2.60)
|
|
12 months
|
|
LMWH
|
24
|
1,057
|
2.3 (1.4–3.4)
|
1
|
1
|
|
Rivaroxaban
|
2
|
184
|
1.1 (0.1–4.0)
|
0.50 (0.12–2.13)
|
0.42 (0.09–2.01)
|
|
CRNMB-H
|
|
3 months
|
|
LMWH
|
23
|
341
|
6.8 (4.2–10.2)
|
1
|
1
|
|
Rivaroxaban
|
8
|
57
|
14.0 (6.0–27.7)
|
2.09 (0.93–4.69)
|
2.02 (0.72–5.62)
|
|
6 months
|
|
LMWH
|
43
|
655
|
6.6 (4.7–8.9)
|
1
|
1
|
|
Rivaroxaban
|
11
|
111
|
9.9 (4.9–17.7)
|
1.53 (0.79–2.98)
|
1.30 (0.57–2.98)
|
|
12 months
|
|
LMWH
|
63
|
1,057
|
6.0 (4.5–7.7)
|
1
|
1
|
|
Rivaroxaban
|
17
|
184
|
9.2 (5.3–14.8)
|
1.61 (0.94–2.75)
|
1.57 (0.80–3.05)
|
|
All-cause mortality
|
|
3 months
|
|
LMWH
|
73
|
341
|
21.4 (16.7–27.0)
|
1
|
1
|
|
Rivaroxaban
|
7
|
57
|
12.3 (4.9–25.4)
|
0.57 (0.26–1.24)
|
0.63 (0.25–1.60)
|
|
6 months
|
|
LMWH
|
102
|
655
|
15.6 (12.7–19.0)
|
1
|
1
|
|
Rivaroxaban
|
9
|
111
|
8.1 (3.6–15.4)
|
0.51 (0.26–1.02)
|
0.59 (0.26–1.33)
|
|
12 months
|
|
LMWH
|
133
|
1,057
|
12.6 (10.5–15.0)
|
1
|
1
|
|
Rivaroxaban
|
10
|
184
|
5.4 (2.6–10.0)
|
0.44 (0.23–0.83)
|
0.49 (0.23–1.06)
|
Abbreviations: CT, cancer-associated venous thromboembolism; CI, confidence interval;
CRNMB-H, clinically relevant nonmajor bleeding requiring hospitalization; LMWH, low-molecular-weight
heparin; SHR, sub-distribution hazard ratio; VTE, venous thromboembolism.
a Incidence rate per 100 person-years.
b Sub-distribution hazard ratio estimated from univariate Fine & Gray regression accounting
for competing risks.
c Applying overlap weighting based on predicted rivaroxaban initiation probabilities.
d Critical organ bleeds are a subset of major bleeds including intracranial bleeds
and other critical organ bleeds.
Significant Bleeds
There were 102 and 20 significant bleeds in LMWH and rivaroxaban-treated patients,
respectively, in the first year after CT. There were 39 and 3 major bleeds in the
LMWH and rivaroxaban cohorts respectively, at 1 year. Of the major bleeds, 24 and
2 bleeds in LMWH and rivaroxaban users respectively were intracranial bleeds or bleeds
in another critical organ. There were 63 and 17 CRNMB-H in LMWH and rivaroxaban cohorts,
respectively, at 1 year of observation ([Table 2]). Crude incidence rates of significant bleeds in the first year after the initial
CT were 9.7 (95% CI: 7.8–11.8) and 10.9 (95% CI: 6.6–16.8) per 100 person-years of
LMWH and rivaroxaban use, respectively. The weighted SHR for significant bleeds in
rivaroxaban compared with LMWH at 12 months was 1.01 (95% CI: 0.57–1.81). Incidence
rates and weighted SHRs at 3, 6, and 12 months are shown in [Table 2] and weighted survival probabilities in the first 12 months after CT are shown in
[Supplementary Fig. S2] (available in the online version).
For major bleeds, the weighted SHRs at 3, 6, and 12 months after CT were decreased
for rivaroxaban compared with LMWH but were not statistically significant, 0.37 (95%
CI: 0.08–1.76), 0.40 (95% CI: 0.11–1.44), and 0.35 (95% CI: 0.10–1.24), respectively.
Due to the low number of events in rivaroxaban users, no further analyses for intracranial
bleeds or bleeds in another critical organ were performed.
For CRNMB-H the weighted SHRs at 3, 6, and 12 months after CT in rivaroxaban compared
with LMWH were increased but not statistically significant, 2.02 (95% CI: 0.72–5.62),
1.30 (95% CI: 0.57–2.98), and 1.57 (95% CI: 0.80–3.05), respectively ([Table 2] and [Supplementary Fig. S2] [available in the online version]).
All-Cause mortality
There were 133 and 10 deaths due to any cause in the LMWH and rivaroxaban-treated
patients within the first year after CT ([Table 2]). Cumulative crude mortality rates in the first year after the initial CT were 12.6
(95% CI: 10.5–15.0) and 5.4 (95% CI: 2.6–10.0) per 100 person-years of LMWH and rivaroxaban
use, respectively. The weighted SHR for death from any cause in rivaroxaban compared
with LMWH at 12 months was 0.49 (95% CI: 0.23–1.06). The weighted SHRs at 3 and 6
months after CT are shown in [Table 2] and weighted survival probabilities in the first 12 months after CT are shown in
[Supplementary Fig. S3] (available in the online version).
Sensitivity Analysis—Rivaroxaban Compared with LMWH, On-Treatment Analysis
The study cohort for the on-treatment analysis excluded 86 patients that switched
AC type before the start of the at-risk period. The definition of switching in these
86 patients is delineated in the Methods, in the exposure sub-section. These were
patients switching from LMWH and rivaroxaban to other ACs. The on-treatment analysis
consisted of a subset of 2,173 patients, 1,867 initially treated with LMWH (96% of
the ITT cohort), and 306 with rivaroxaban (97% of the respective ITT cohort; [Supplementary Fig. S4] [available in the online version]). At 1 year of observation, LMWH users cumulated
a total of 529 person-years (50.0% of the person-years in the ITT analysis) and rivaroxaban
users cumulated a total of 134 person-years (72.8% of the person-years in the ITT
analysis) of follow-up.
The duration of anticoagulation treatment in the first year of observation, as shown
by discontinuations of therapy over time, is illustrated in [Fig. 2]. Duration of anticoagulation treatment with LMWH use was less than the duration
of anticoagulation treatment with rivaroxaban throughout the complete year of observation
following the CT. The proportion of patients on anticoagulation treatment at 1 year
was 25.8% for LMWH users and 49.5% for rivaroxaban users. The main outcomes for the
on-treatment analysis are shown in [Table 3].
Fig. 2 Discontinuationa,b of anticoagulant treatment by anticoagulant and time since first treatment recording.
DOAC, direct oral anticoagulant; LMWH, low-molecular-weight heparin; VKA, vitamin
K antagonist; VTE, venous thromboembolism. aPersistence estimate accounting for death, significant bleeds, and VTE recurrence
as competing events and switching of anticoagulant (rivaroxaban, other DOAC, LMWH,
VKA, other parenteral) as censoring event. bApplying overlap weighting based on predicted rivaroxaban initiation probabilities.
Table 3
VTE recurrences, bleeding, and mortality at 3, 6, and 12 months after CT (sub-distribution
hazard ratios) in the on-treatment population
|
Time since CT
|
Events
|
Person-years
|
Incidence rate[a]
(0.95 CI)
|
Unweighted SHR[b]
(0.95 CI)
|
Overlap weighted[c] SHR[b] (0.95 CI)
|
|
VTE recurrences
|
|
3 months
|
|
LMWH
|
17
|
290
|
5.9 (3.4–9.5)
|
1
|
1
|
|
Rivaroxaban
|
2
|
53
|
3.8 (0.4–13.8)
|
0.63 (0.15–2.73)
|
0.56 (0.11–2.79)
|
|
6 months
|
|
LMWH
|
22
|
452
|
4.9 (3.0–7.4)
|
1
|
1
|
|
Rivaroxaban
|
4
|
96
|
4.2 (1.1–10.7)
|
0.89 (0.31–2.53)
|
0.86 (0.28–2.63)
|
|
12 months
|
|
LMWH
|
26
|
529
|
4.9 (3.2–7.3)
|
1
|
1
|
|
Rivaroxaban
|
6
|
134
|
4.5 (1.6–9.8)
|
1.00 (0.42–2.37)
|
0.82 (0.31–2.17)
|
|
All significant bleeds
|
|
3 months
|
|
LMWH
|
39
|
290
|
13.5 (9.5–18.5)
|
1
|
1
|
|
Rivaroxaban
|
9
|
53
|
17.1 (7.8–32.5)
|
1.31 (0.63–2.72)
|
1.23 (0.51–2.96)
|
|
6 months
|
|
LMWH
|
54
|
452
|
11.9 (8.9–15.6)
|
1
|
1
|
|
Rivaroxaban
|
11
|
96
|
11.4 (5.7–20.5)
|
1.06 (0.55–2.04)
|
1.04 (0.47–2.29)
|
|
12 months
|
|
LMWH
|
64
|
529
|
12.1 (9.3–15.5)
|
1
|
1
|
|
Rivaroxaban
|
17
|
134
|
12.7 (7.4–20.4)
|
1.10 (0.64–1.87)
|
1.08 (0.55–2.14)
|
|
Major bleeds
|
|
3 months
|
|
LMWH
|
19
|
290
|
6.6 (3.9–10.3)
|
1
|
1
|
|
Rivaroxaban
|
2
|
53
|
3.8 (0.4–13.8)
|
0.60 (0.14–2.55)
|
0.60 (0.13–2.84)
|
|
6 months
|
|
LMWH
|
24
|
452
|
5.3 (3.3–7.9)
|
1
|
1
|
|
Rivaroxaban
|
2
|
96
|
2.1 (0.2–7.6)
|
0.44 (0.10–1.88)
|
0.43 (0.09–2.05)
|
|
12 months
|
|
LMWH
|
29
|
529
|
5.5 (3.6–7.9)
|
1
|
1
|
|
Rivaroxaban
|
2
|
134
|
1.5 (0.1–5.5)
|
0.31 (0.07–1.34)
|
0.33 (0.07–1.67)
|
|
Critical organ bleeds[d]
|
|
3 months
|
|
LMWH
|
10
|
290
|
3.5 (1.6–6.4)
|
1
|
1
|
|
Rivaroxaban
|
1
|
53
|
1.9 (0.0–10.6)
|
0.57 (0.07–4.51)
|
0.62 (0.06–6.08)
|
|
6 months
|
|
LMWH
|
13
|
452
|
2.9 (1.5–5.0)
|
1
|
1
|
|
Rivaroxaban
|
1
|
96
|
1.0 (0.0–5.8)
|
0.41 (0.05–3.23)
|
0.40 (0.04–3.90)
|
|
12 months
|
|
LMWH
|
17
|
529
|
3.2 (1.8–5.2)
|
1
|
1
|
|
Rivaroxaban
|
1
|
134
|
0.7 (0.0–4.2)
|
0.24 (0.03–1.97)
|
0.27 (0.03–2.82)
|
|
CRNMB-H
|
|
3 months
|
|
LMWH
|
20
|
290
|
6.9 (4.2–10.7)
|
1
|
1
|
|
Rivaroxaban
|
7
|
53
|
13.3 (5.3–27.4)
|
2.00 (0.84–4.76)
|
1.85 (0.61–5.58)
|
|
6 months
|
|
LMWH
|
30
|
452
|
6.6 (4.4–9.5)
|
1
|
1
|
|
Rivaroxaban
|
9
|
96
|
9.4 (4.2–17.8)
|
1.54 (0.72–3.29)
|
1.60 (0.62–4.12)
|
|
12 months
|
|
LMWH
|
35
|
529
|
6.6 (4.6–9.3)
|
1
|
1
|
|
Rivaroxaban
|
15
|
134
|
11.2 (6.2–18.6)
|
1.71 (0.93–3.12)
|
1.80 (0.81–4.03)
|
|
All-cause mortality
|
|
3 months
|
|
LMWH
|
66
|
290
|
22.8 (17.6–29.1)
|
1
|
1
|
|
Rivaroxaban
|
7
|
53
|
13.3 (5.3–27.4)
|
0.59 (0.27–1.30)
|
0.66 (0.25–1.74)
|
|
6 months
|
|
LMWH
|
81
|
452
|
17.9 (14.2–22.3)
|
1
|
1
|
|
Rivaroxaban
|
9
|
96
|
9.4 (4.2–17.8)
|
0.57 (0.29–1.14)
|
0.71 (0.31–1.61)
|
|
12 months
|
|
LMWH
|
87
|
529
|
16.5 (13.1–20.3)
|
1
|
1
|
|
Rivaroxaban
|
10
|
134
|
7.5 (3.5–13.8)
|
0.55 (0.28–1.07)
|
0.67 (0.30–1.49)
|
Abbreviations: CT, cancer-associated venous thromboembolism; CI, confidence interval;
CRNMB-H, clinically relevant nonmajor bleeding requiring hospitalization; LMWH, low-molecular-weight
heparin; SHR, sub-distribution hazard ratio; VTE, venous thromboembolism.
a Incidence rate per 100 person-years.
b Sub-distribution hazard ratio estimated from univariate Fine & Gray regression accounting
for competing risks.
c Applying overlap weighting based on predicted rivaroxaban initiation probabilities.
d Critical organ bleeds are a subset of major bleeds including intracranial bleeds
and other critical organ bleeds.
Discussion
In this large cohort of patients with VTE and active cancer, not at high risk for
bleeding, we evaluated the effectiveness and safety at 3, 6, and 12 months of rivaroxaban
therapy compared with LMWH therapy. Treatment with rivaroxaban compared with LMWH
was associated with similar weighted (adjusted) SHR estimates for VTE recurrences
at 3, 6, and 12 months in all analyses. Treatment with rivaroxaban compared with LMWH
was associated with similar risk of all significant bleeds, the principal safety outcome.
In the bleeding subgroups, the results demonstrated that rivaroxaban had a consistently
lower but not statistically different risk of major bleeds and critical site bleeds
at each of the time points, and a higher but not statistically significant risk of
CRNMB-H. No significant differences were seen in all-cause mortality in the two treatment
cohorts, but in the rivaroxaban cohort all-cause mortality was consistently lower
in all analyses. Lower mortality in the rivaroxaban cohort may indicate unmeasured
prognostic differences in the cohorts.
These findings are consistent with the OSCAR-US study that demonstrated a reduced
risk of recurrent VTE and no differences in bleeding or mortality outcomes.[18] The outcomes for VTE recurrences, bleeding, and mortality are similar with the on-treatment
analyses to the ITT (main) analyses. Furthermore, these results comparing a DOAC (rivaroxaban)
with LMWH were consistent with the findings of other smaller observational studies,
clinical trials, and meta-analyses.[7]
[19]
[20]
[21]
[22]
[23]
The 1-year duration of anticoagulation treatment for patients receiving rivaroxaban
was approximately twofold greater compared with those receiving LMWH, a significant
difference. Different factors could have contributed to this finding including: (1)
better adherence/tolerability with rivaroxaban compared with parenteral ACs, (2) some
factors (covariates) that have an influence in the choice of rivaroxaban or LMWH might
have changed after treatment initiation, resulting in switching/discontinuation, and
(3) other burdens such as drug cost that favor persistent use of rivaroxaban compared
with LMWH. Despite the differences in duration of anticoagulation treatment with rivaroxaban
and LMWH, results of the on-treatment analyses were consistent with the results of
the ITT analyses.
Strengths and Weaknesses of the Study
The study cohort comprised a large heterogeneous cohort of 2,259 patients treated
with ACs for CT and followed up for 12 months, the period during which most patients
with CT are anticoagulated and which is associated with a high risk of recurrences,
bleeding events, and mortality.[2] The effectiveness and safety were assessed by validated outcomes that were verified
by clinicians who were blind to the therapy. These outcomes were recurrent VTE, significant
bleeds (major bleeds and CRNMB-H), and all-cause mortality. We used standard definitions
consistent with those used in clinical trials.[12]
[21]
[22]
[23] However, patients with a missing record of active cancer, VTE, or AC therapy did
not form part of our study cohort. Patients who had a recording of anticoagulation
later in the 30-day period post-CT diagnosis or anticoagulation treatment started
in hospital may have had early outcomes prior to the index day. Those potential cases
should have a similar distribution in the rivaroxaban and LMWH groups based on the
SELECT-D data, thus it is unlikely that this may have influenced the results.[23]
Although a vast set of covariates were used for adjustment, unmeasured confounding
cannot be ruled out as some covariates associated with a study outcome of interest
were not available in the database (such as cancer staging) and this may affect outcomes
such as mortality. Furthermore, covariate changes over time during the at-risk period
were not considered in the analyses of the different effectiveness and safety outcomes.
This could have resulted in differential/unbalanced risk sets for the comparison of
rivaroxaban with LMWH during the at-risk period and may have affected both the ITT
and the on-treatment analyses.
VTE and bleeding events were defined based on coded information rather than complete
clinical data. However, our VTE algorithm has previously been validated and showed
a sensitivity of 92.6% and a specificity of 98.8%.[11] Bleeding events were defined according to ISTH criteria and were validated by manual
review of all available patient records by three physicians who assessed all potential
bleeding events.[12] In-hospital pharmacy data including anticoagulation use are not available in the
CPRD-HES datalink. Medical diagnoses are recorded as hospital discharge diagnoses
but the day of occurrence during the hospitalization is either unknown or uncertain.
Consequently, in-hospital data were insufficient to establish the temporal relationship
between the status of anticoagulation treatment and the onset of an outcome event.
To avoid misclassification of anticoagulation exposure and of outcomes, we did not
consider outcome events that occurred during the same hospitalization as the initial
CT. To study an inception cohort, we excluded patients with a history of VTE or anticoagulation,
however, that meant we could not assess the risks in patients presenting with recurrent
VTE events. We also excluded patients with other indications for anticoagulation (10%),
and palliative care patients (as this affects management and data recording) (31%)
as well as cancer types that are associated with a high risk of bleeding or unknown
cancer type (12%). These exclusions were important for cohort definition, data quality,
and to be consistent with the guidelines; however, they impact the generalizability
of the findings.
We used the overlap weighting adjustment method based on propensity scores to make
the two AC exposure groups comparable with respect to baseline cohort differences
such as age, sex, and cancer site. The overlap weighting led to exact balance of all
measured baseline characteristics that were included in the regression model in the
two AC exposure groups. Outcome events were captured only if they were recorded according
to our outcome definitions based on previously developed and validated algorithms,
and manual review of all potential cases with reviewers blinded to the AC exposure
of interest. Missed outcome events were likely to be at random and independent of
the exposure of interest resulting in unaffected relative risk estimates but may lead
to underestimation of the absolute risk estimates. SHR estimates for outcomes with
small event numbers, such as critical organ bleeds in rivaroxaban, had low precision
as reflected by wide CIs.
In this cohort study of patients with CT treated with either rivaroxaban or LMWHs,
rivaroxaban was as effective as LMWH at preventing VTE recurrence and without differences
in the rates of significant bleeding (composite outcome), major bleeds, critical organ
bleeds, CRNMB-H, or all-cause mortality. Patients treated with rivaroxaban remained
on therapy for a longer period of time compared with LMWH. Our study findings support
the recommendation that rivaroxaban is a reasonable alternative to LMWH for the treatment
of CT when used in accordance with guidelines.
This study is part of the OSCAR program with independent studies in the United States,[10]
[20] United Kingdom,[13] and Sweden[24] that use consistent definitions of design, exposures of interest, covariates, and
data analyses. While the study in Sweden is being reported, the comparison of the
UK and U.S.[20] cohorts of the OSCAR program indicates that the study findings are generalizable
to patients with active cancer not including non-brain central nervous system, unresected
colorectal/lower gastrointestinal tract, hematologic (except lymphoma and myeloma),
esophagus, stomach, and bladder cancer and patients with conditions such as thrombocytopenia,
end-stage kidney disease, and current pregnancy.
Conclusion
Patients who are not at high risk of bleeding with cancer-associated thrombosis treated
with either rivaroxaban or LMWHs have comparable effectiveness and safety outcomes.
This finding supports the recommendation that rivaroxaban is a reasonable alternative
to LMWH for the treatment of CT when used in accordance with guidelines.
What is known about this topic?
-
Guidelines recommend the use of direct oral anticoagulants such as rivaroxaban, and
low-molecular-weight heparins (LMWHs), in treating patients with cancer-associated
venous thromboembolism (CT) based on clinical trials.
-
There is a paucity of population-based data allowing treatment outcome comparisons
in clinical practice.
What does this paper add?
-
Patients who are not at high risk of bleeding with CT treated with either rivaroxaban
or LMWH have comparable risk of recurrent venous thromboembolism.
-
The patients also have similar risk of significant bleeding and all-cause mortality.
-
These data support the use of rivaroxaban as an alternative to LMWH for the treatment
of CT.
