Professional Documents
Culture Documents
by
FENGLONG XIE
A DISSERTATION
BIRMINGHAM, ALABAMA
2018
ProQuest Number: 10750692
All rights reserved
INFORMATION TO ALL USERS
The quality of this reproduction is dependent upon the quality of the copy submitted.
In the unlikely event that the author did not send a complete manuscript
and there are missing pages, these will be noted. Also, if material had to be removed,
a note will indicate the deletion.
ProQuest 10750692
Published by ProQuest LLC (2018 ). Copyright of the Dissertation is held by the Author.
All rights reserved.
This work is protected against unauthorized copying under Title 17, United States Code
Microform Edition © ProQuest LLC.
ProQuest LLC.
789 East Eisenhower Parkway
P.O. Box 1346
Ann Arbor, MI 48106 - 1346
Copyright by
FENGLONG XIE
2018
ASSESSING THE COMPARATIVE EFFECT OF TOCILIZUMAB ON RISK OF
CARDIOVASCULAR DISEASE AMONG RHEUMATOID ARTHRITIS PATIENTS
USING CLAIMS DATA: A DIRECT COMPARISON AMONG BIOLOGIC
DISEASE-MODIFYING ANTIRHEUMATIC DRUGS
FENGLONG XIE
EPIDEIMOLOGY
ABSTRACT
receptor and is effective in the treatment of rheumatoid arthritis (RA) [1]. Multiple
burden and conduct comparative effectiveness studies with much larger sample
sizes than cohorts, which require primary data collection. However, Medicare
iii
data using The Reasons for Geographic and Racial Difference in Stroke
(REGARDS) data linked to Medicare claims. CVD events and cause of death in
the REGARDS study were adjudicated by two experts. Our algorithm can identify
Our study confirmed that tocilizumab was not associated with increased or
decreased CVD risk compared to etanercept: the hazard ratio for tocilizumab
compared to etanercept was 0.91 (95%CI: 0.66, 1.25). However, unlike the
clinical trial, which enrolled only high-risk patients, we extended this finding to
“low CVD risk” RA patients. We also showed that tocilizumab was not associated
further showed that tocilizumab was associated with reduced CVD risk when
compared to a pooled TNFi group. This is most likely due to slightly increased
assess the effect of methotrexate on CVD risk among bDMARDS users and
found that methotrexate was associated with an overall 27% (95%CI: 18-35%)
reduction in CVD risk. The hazard ratio for tocilizumab concomitantly with
algorithm, methotrexate
iv
DEDICATION
To my two sweet and beautiful daughters, Yuxi (Lucy) Xie and Allison (Ally) Xie
v
ACKNOWLEDGMENTS
I want to thank all the people whose love, help, and encouragement supported
Paul Muntner, for his instruction and sage advice. I want to express special
thanks to Dr. Jeffrey Curtis for his advice and encouragement, his generous
support, and for allowing me access to needed resources. Special thanks to Dr.
Elizabeth Delzell and Dr. Meredith Kilgore for their unlimited encouragement.
Nengjun Yi and Dr. Huifeng Yun, for their excellent suggestions and instructive
Yang, and all members of the PEER group. I am very lucky to work alongside
you.
Most of all, I want to thank my wife, Dongmei Sun, whose love and support are
vi
TABLE OF CONTENTS
DEDICATION............................................................................................................................... v
ACKNOWLEDGEMENTS ...................................................................................................... vi
INTRODUCTION ....................................................................................................................... 1
Treatment for rheumatoid arthritis and the risk for cardiovascular disease 3
Significance .................................................................................................................... 5
vii
GENERAL LIST OF REFERENCES ............................................................................... 127
viii
LIST OF TABLES
Table Page
ix
S6 Example of the calculation of the predicted probability for having a fatal
stroke……………………………………………………………..………………………………….…………………...…47
…………………………………………………………………………………………………………………..………………..79
x
METHOTREXATE USE AND THE RISK FOR CARDIOVASCULAR DISEASE
AMONG RHEUMATOID ARTHRITIS PATIENTS INITIATING BIOLOGIC
DISEASE-MODIFYING ANTIRHEMATIC DRUGS
2 Incidence rate and hazard ratio of composite CVD and its component
by associated with methotrexate use………………………………………………....….……….118
3 Incidence rates and hazard ratios for composite CVD and its component
associated with time varying methotrexate, by biologic DMARDS…………..119
S1 Incidence rate and hazard ratio for CVD associated with methotrexate
use, by biologic DMARDS for patients exposed to methotrexate before
initiation of biologic DMARDS.............................................................................. ..121
S2 Incidence rate and hazard ratio for CVD associated with time varying
methotrexate use with 90 days extension, by biologic DMARDS............. ..122
xi
LIST OF FIGURES
Figure Page
3 Receiver operating characteristic curve for fatal CVD, fatal CHD and fatal
stroke………………………………………………………………………………………………………………………….39
4 Steps to define fatal stroke, fatal CHD and fatal CVD using the claims-
based algorithms………………………………………………………….…………………………………………40
2 Forest plot for adjusted hazard ratio for primary outcome by data source.84
3 Forest plot for adjusted hazard ratio for primary outcome by history of
other cardiovascular disease and data source………………………………………………..85
S1 Forest plot for adjusted hazard ratio for myocardial infarction and stroke by
data source………………………………..…………………………………………………..……………..…………91
S2 Forest plot for adjusted hazard ratio for primary outcome for biologic
DMARDS experienced by data source………………………………………………..……………92
xii
METHOTREXATE USE AND THE RISK FOR CARDIOVASCULAR DISEASE
AMONG RHEUMATOID ARTHRITIS PATIENTS INITIATING BIOLOGIC
DISEASE-MODIFYING ANTIRHEMATIC DRUGS
xiii
1
INTRODUCTION
and loss of function, disability, poor quality of life, and shortened life expectancy
is believed to be both genetic and environmental. More than 100 genes, such as
is associated with higher rheumatoid factor titer and later disease severity [13].
patients [14]. The elevated risk of CVD death in RA patients is observed in both
men and women [8, 9], and traditional risk factors do not fully explain the
interleukin-1 and –6 (IL-1, IL-6), C-reactive protein (CRP), and tumor necrosis
factor (TNF). These cytokines and inflammatory markers play an important role in
Although the FDA has approved more than 50 drugs to treat RA, RA is not
DMARDS fall into three categories based on whether they target specific steps in
the inflammatory process, and whether they are biologic or small molecule.
system without a specific target and slow disease activity by modifying the
interferes with T cell costimulation [19]; tumor necrosis factor inhibitors (TNFi),
response by binding to and depleting CD20 B cells [25]. RA patients who do not
taken orally.
Treatment for rheumatoid arthritis and the risk for cardiovascular disease
associated with a reduced risk for CVD [27-36]. The underlining mechanisms for
this reduced risk are controversial and poorly understood, but most studies have
Many studies have suggested that treatment with TNFi reduces the risk of CVD
in RA patients, but results from those investigations are not consistent [37].
in lipid profile in RA patients treated with tocilizumab [39-41] and these results
raise the possibility that tocilizumab may increase the risk of CVD events. Few
studies have compared tocilizumab with other biologic DMARDS for the risk of
compared to etanercept among RA patients with risk factors for CVD [42]. Two
tocilizumab compared to TNFi or abatacept [43, 44]. All of these studies had
major limitations: the clinical trial only compared tocilizumab to etanercept but no
observational studies did not include fatal CVD as an outcome, since most
administrative claims data lack information on cause of death; and all three
studies had relatively small sample sizes and limited numbers of events in the
estimate the incidence of true CVD-related events. This may be even more
higher proportion of sudden coronary heart disease (CHD) death [45]. Due to
which could be an issue, since each of the medications in this class may have
different drug-specific risks for CVD. Finally, the observational studies did not
assess the impact of disease activity among biologic users to address the
5
While a clinical trial is the best way to assess the causality of exposure and
≥65 years of age and younger adults who are disabled or who have end-stage
through the Medicare program [48]. Medicare claims data provide the opportunity
safety studies with much larger sample sizes than cohort studies that require
non-fatal outcomes and all-cause mortality can provide useful information, cause
Significance
Through specific aim 1, we designed algorithms to identify fatal CVD, CHD, and
stroke in Medicare claims data with high sensitivity and specificity. These
algorithms address the major limitations of Medicare claims data, and can be
used to identify clinical factors associated with fatal CVD, CHD, and stroke
6
these algorithms, to non-fatal CVD events identified through claims data should
associations.
more economical way and provided better precision: tocilizumab was not
associated with increased CVD risk compared to etanercept. However, unlike the
clinical trial, which enrolled only high-risk patients, we extended this finding to
“low CVD risk” RA patients as well. In addition, we showed that tocilizumab did
not increase CVD risk when compared to abatacept and rituximab. Further, we
showed that tocilizumab lowered CVD risk when compared to a pooled TNFi
exposure, mostly attributable to slightly increased CVD risk associated with TNF.
Our results should remove the concern for tocilizumab treated RA patients.
associated with an overall 27% reduction in CVD risk. The strength of association
varies among concomitant biologic DMARDS. This finding has two clinical
to tocilizumab initiators may help to reduce the CVD risk; 2) Because RA patients
and biologic DMARDS like tocilizumab are effective with and without
reduce CVD risk among non-RA patients with high CVD risk.
7
by
Copyright
2018
By
Wiley
Used by permission
Corresponding Author:
Paul Muntner, PhD
Department of Epidemiology
School of Public Health
University of Alabama at Birmingham
1700 University Boulevard, Suite 450
Birmingham, AL 35294
pmuntner@uab.edu
Phone: 205-975-8077
Fax: 205-975-7058
Word count
Abstract: 249
Manuscript: 2,960
Tables: 4
Figures: 4
9
Abstract
cardiovascular disease (CVD; i.e., fatal coronary heart disease [CHD] or stroke),
study data were linked with Medicare claims to develop the algorithms. Events
Stepwise selection was used to choose predictors from Medicare data for
value (PPV) and negative predictive value (NPV) were used to assess algorithm
Results Data from 2,685 REGARDS participants with linkage to Medicare, who
died between 2003 and 2013, were analyzed. The C-index for discriminating fatal
CVD from other causes of death was 0.87. Using a cut-point that provided the
0.64, specificity 0.90, PPV 0.65 and NPV 0.90. The algorithms resulted in
discriminating fatal CHD and fatal stroke performed similarly to fatal CVD.
10
events from other causes of death performed better than the method of using
Medicare is a health insurance program for US adults ≥65 years of age and
younger adults who are disabled or who have end-stage renal disease [1]. In
program [2]. Medicare claims data provide the opportunity to estimate disease
burden and conduct comparative effectiveness research and safety studies with
much larger sample sizes than cohorts that involve primary data collection. A
While assessing the association of exposures with non-fatal outcomes and all-
cause mortality can provide useful information, cause of death is relevant for
stroke, is the leading cause of death in the US [3]. Many analyses of Medicare
claims data have investigated CVD as an outcome, but have relied on hospital
approach which fails to capture out-of-hospital CVD deaths and some in-hospital
CVD deaths [4-6]. In 2013, 84% of Medicare beneficiaries were ≥65 years of age,
and mortality is high in this population [7, 8]. Not having cause of death for
event rates, as many events are fatal [9], and in biased exposure-outcome
deaths that are fatal CVD, fatal CHD and fatal stroke versus deaths from other
causes.
Methods
Data source
We used data from the Reasons for Geographic and Racial Differences in Stroke
Hispanic white and black adults ≥45 years of age from all 48 contiguous US
states and Washington DC between January 1, 2003 and October 31, 2007 [10].
Data from REGARDS study participants were linked to Medicare claims using
social security numbers, with linkages confirmed using birthdate and sex [11].
The REGARDS study was approved by the institutional review board at the
informed consent.
Study population
We included REGARDS participants who had their study data linked to Medicare
claims and died between January 1, 2003 and December 31, 2013. We excluded
participants whose death date recorded in REGARDS study did not match with
that recorded in the Medicare beneficiary summary file; those who died before
the age of 65.5 years; and those who lacked 182 consecutive days of Medicare
fee-for-service coverage (Medicare Parts A and B but not Part C) prior to their
death, referred to as the “look-back” period, to identify claims that were used in
the algorithms. We excluded Medicare beneficiaries <65 years of age at the start
of the look-back period because they represent a select population that qualifies
myocardial infarction (MI) or stroke, as defined below, within 28 days prior to their
death for whom medical records could not be retrieved for adjudication by the
Living REGARDS study participants or their proxies are contacted every six
months via telephone to confirm vital status [9, 10]. Deaths are also identified
through searches in the National Death Index. When deaths are identified,
interviews with proxies are conducted, and death certificates and medical records
for hospitalizations in the last year of life are retrieved. Causes of death are
of fatal CVD, fatal CHD and fatal stroke events in the REGARDS study.
One approach to define fatal CVD, CHD, and stroke in Medicare is to use
discharge diagnosis codes for CVD, CHD and stroke from inpatient claims within
claim with a hospital discharge diagnosis code for MI (Supplemental Table 1).
inpatient claim with a hospital discharge diagnosis code for stroke (Supplemental
Table 1). Fatal CVD was defined as fatal CHD or fatal stroke. We chose to
include inpatient claims within 28 days prior to the death date to match the
Statistical methods
Candidate predictors
demographic data (age at death, sex, race) from the Medicare beneficiary
summary file and diagnosis and procedure codes, and current procedure
terminology codes from Medicare claims within 28 days and, separately, at any
time prior to death (Supplemental Table 1). Also, we evaluated a second set of
In the REGARDS study, fatal CVD was defined as fatal MI, CHD death or fatal
stroke and fatal CHD was defined as fatal MI or CHD death [12]. We evaluated
three approaches for developing algorithms for the prediction of these composite
Direct approach
Logistic regression models, with three rounds of variable selection, were used to
develop algorithms for having a fatal CVD event. For each algorithm, all pre-
for the first round of stepwise selection and those with a p-value ≤0.05 were
retained for the second round of stepwise selection. During the second round of
variable selection, the AHRQ CCS categories were added to a logistic regression
model with pre-specified candidate predictors retained following the first round of
selection. Given the large number of AHRQ CCS categories, these were added
to the model one at a time to avoid overfitting the model [14]. Pre-specified
candidate predictors with p-values ≤0.05 in the first round and AHRQ CCS
categories with p-values ≤0.05 in the second round of selection were included in
a single logistic regression model for the third round of stepwise selection. Only
candidate predictors with a p-value ≤0.05 in the third round of selection were
included in the final algorithm. A separate algorithm was developed for fatal CHD
Indirect approach 1
The first indirect approach used separate logistic regression models for fatal MI,
CHD death and fatal stroke, with three rounds of selection to identify candidate
predictors to retain for each outcome, as described above for the direct
cut-point for fatal MI, CHD death or fatal stroke were defined as having a high
higher than a pre-specified cut-point for fatal MI or CHD death were defined as
having a high predicted probability for fatal CHD. A description of how the pre-
Indirect approach 2
The second indirect approach modeled the outcomes of fatal MI, CHD death and
predicted probabilities for each outcome from the three separate models were
We identified the cut-point of predicted probability for each algorithm that resulted
in the closest percentage of the population with predicted and observed events.
17
having a high probability of that outcome (i.e., a positive test result). To compare
the direct and both indirect approaches for developing algorithms for the
value (PPV), and negative predictive value (NPV). The C-index was calculated
for the direct approach and the second indirect approach. Confidence intervals
for the C-index were calculated using a bootstrap method with 1,000 replications
[15]. The C-index cannot be calculated for the first indirect approach as this
approach does not provide predicted probabilities for the composite outcomes.
NPV and PPV and plotted the receiver operating characteristic curve. We
with hospital discharge diagnosis codes using the net reclassification index (NRI)
[16, 17]. Choosing the cut-point of predicted probability that provides the closest
outcome may not meet the needs of all researchers. Therefore, we calculated
test characteristics and discrimination using three additional cut-points for each
outcome. These included cut-points that provided (1) the maximum specificity
with a sensitivity ≥0.80; (2) the maximum NPV with a PPV ≥0.80, and (3) the
maximum sum of sensitivity and specificity. All analyses were conducted using
Results
Among 20,403 REGARDS participants with data linked to Medicare, 4,327 died
between January 1, 2003 and December 31, 2013 of whom 2,685 were included
mean age 79.6 years, 41.3% women, 32.8% black, 3.5% and 4.1 had a hospital
discharge diagnosis code for MI and stroke, respectively, within 28 days prior to
their death (Table 2). Based on adjudicated REGARDS study data, 608 (22.6%)
deaths were classified as fatal CVD, 479 (17.8%) as fatal CHD and 142 (5.3%)
as fatal stroke, with 13 participants having both a fatal CHD and a fatal stroke.
development
The predictors with the strongest associations with fatal CVD when using the
direct approach were hospital discharge diagnosis codes for stroke and MI in any
position within 28 days of death date, and for fatal CHD was hospital discharge
(Supplemental Table 2). When each component of fatal CVD was modeled
separately, the strongest predictors for fatal stroke and fatal MI were hospital
discharge diagnosis codes for stroke and MI in any position within 28 days of
death date respectively. The C-index for developing the fatal CVD and fatal CHD
algorithms were similar using the direct approach and the second indirect
approach (Table 3). For fatal CVD, sensitivity and PPV were higher when using
19
the indirect approaches compared with the direct approach. For fatal CHD,
sensitivity and PPV were within 0.02 when using the indirect approaches and the
direct approach. Based on these results, the second indirect approach was used
Performance of final algorithms for fatal CVD, fatal CHD, and fatal stroke
The C-index was >0.85 for fatal CVD, fatal CHD and fatal stroke (Figure 3). For
each cut-point evaluated, the sensitivity was higher using the algorithms
compared with hospital discharge diagnosis codes (Table 4). The specificity was
lower using the algorithms versus hospital discharge diagnosis codes when
predicted and observed events, maximum specificity with sensitivity ≥0.80 or the
probability were chosen to achieve the highest NPV with PPV ≥0.80, the
specificity was similar using the algorithms and hospital discharge diagnosis
codes. For each cut-point used to define a high probability of fatal CVD and fatal
diagnosis codes (Table 4 and Supplemental Tables 3 and 4). When cut-points
were chosen to provide the closest percentage of predicted and observed fatal
CVD events, maximum specificity with sensitivity ≥0.80 or maximize the sum of
sensitivity and specificity, the algorithm for fatal stroke resulted in a positive NRI
maximum NPV while maintaining a PPV ≥0.80, the algorithm for fatal stroke
Discussion
and fatal stroke, separately, versus other causes of death. Each algorithm
showed good discrimination. The algorithms had a positive NRI compared with
probability of fatal CVD, CHD or stroke based on the algorithms to obtain good
calibration (i.e., the same percentage of observed and predicted events), high
sensitivity, high PPV and maximizing the sum of sensitivity and specificity. Figure
4 and Supplemental Tables 6-8 provide details on how to use the algorithms.
The majority of Medicare beneficiaries are ≥65 years of age, consequently the
mortality rate is high in this population [7, 8]. One-third of CVD events among
Medicare beneficiaries are fatal [18]. Causes of death are available only for
Medicare beneficiaries who died between 2006 and 2008 through a linkage with
the National Death Index. Using these causes of death is not ideal as the
risk in Medicare often analyze CVD hospitalizations as outcomes [4, 5, 21, 22].
This approach will underestimate the rate of CVD as not all deaths are preceded
by a claim with an inpatient diagnosis code for CVD. In the current study, only
28.5% of participants with a fatal CVD event had a hospital discharge diagnosis
21
code for MI or stroke within 28 days of their death. In addition, risk factors may
show different associations with fatal and nonfatal CVD events [9, 23].
The hospital discharge diagnosis method has been used previously to define
fatal MI and stroke [24]. The sensitivity of this approach was low in the current
compared with this approach, while maintaining high specificity, and improving
CVD events, we recommend using the algorithms developed herein rather than
Those interested in using the algorithms that we developed may have different
and PPV and balancing sensitivity and specificity for defining a high probability of
fatal CVD, fatal CHD, and fatal stroke based on the algorithms. Multiple cut-
points can be applied to determine fatal CVD outcomes based on the algorithms
Data from cohorts including adjudicated fatal and nonfatal CVD events have
been used to develop the Pooled Cohort risk equations [25]. The Pooled Cohort
risk equations estimate 10-year predicted risk for fatal and non-fatal CVD events
and have been incorporated into clinical practice guidelines [25, 26]. Although
using the outcomes from the Pooled Cohort risk equations, this has not been
22
the algorithms developed in the current study can be combined with non-fatal
outcomes to calculate CVD rates that resemble the outcome of the Pooled
Cohort equations.
conflicting death date in the REGARDS database and the Medicare beneficiary
summary file. For 40 of these 130 participants, the death date only differed by
one day; for 38 participants, Medicare recorded the death date as the last day of
the same month that REGARDS adjudicated as the death date; for 17
summary file; and for the remaining 35 participants, there was no clear
relationship between the death dates recorded in the REGARDS database and
Medicare. Given that these 130 deaths only represented 3% of the total 4,327
with deaths linked in REGARDS and Medicare, these exclusions are unlikely to
The current study has several strengths including a relatively large sample size,
including 2,685 deaths, of which 608 were fatal CVD events. Fatal events in the
enrolled adults from across the US and participants with Medicare fee-for-service
coverage are comparable to all US adults aged 65 years or older with this
23
insurance coverage [11]. The current study also has some limitations. The
REGARDS study was restricted to blacks and whites and the algorithm needs to
adults ≥65 years of age. Although most deaths in the US occur among adults in
this age group [27], the validity of the algorithms in populations <65 years of age
datasets.
fatal CVD, CHD, and stroke events with high sensitivity and specificity. These
algorithms can be used to investigate factors associated with fatal CVD, CHD,
and stroke among Medicare beneficiaries. Adding fatal events with a high
identified through claims data should provide a more accurate event rate and
Acknowledgments
The work for this manuscript was funded by an industry /academic collaboration
provided by grants from the Agency for Healthcare Research and Quality (R01-
K24HL111154-K24).
Conflict of interest
Safford, and P. Muntner receive research grant support from Amgen. J. R. Curtis
receives research grant support from Amgen not related to this project. M. M.
Safford and M. Woodward receive consulting fees from Amgen. E. B. Levitan has
served on advisory boards for Amgen. E. B. Levitan has received consulting fees
References
http://www.aoa.gov/Aging_Statistics/Profile/2013/index.aspx. (Accessed at
http://kff.org/medicare/state-indicator/total-medicare-beneficiaries/?
currentTimeframe=0&sortModel= %7B%22colId%22:%22Location%22,
4. Serban MC, Colantonio LD, Manthripragada AD, Monda KL, Bittner VA,
69(11):1386-1395.
5. Zhang J, Xie F, Yun H, Chen L, Muntner P, Levitan EB, Safford MM, Kent
6. Spatz ES, Beckman AL, Wang Y, Desai NR, Krumholz HM: Geographic
8. Roth DL, Skarupski KA, Crews DC, Howard VJ, Locher JL: Distinct age
Evidence from a national cohort study. Social Science & Medicine 2016,
156:12-20.
9. Safford MM, Brown TM, Muntner PM, Durant RW, Glasser S, Halanych
race and sex with risk of incident acute coronary heart disease events.
308(17):1768-1774.
10. Howard VJ, Cushman M, Pulley L, Gomez CR, Go RC, Prineas RJ,
Graham A, Moy CS, Howard G: The reasons for geographic and racial
2005, 25(3):135-143.
11. Xie F, Colantonio LD, Curtis JR, Safford MM, Levitan EB, Howard G,
184(7):532-544.
27
12. Luepker RV, Apple FS, Christenson RH, Crow RS, Fortmann SP, Goff D,
Goldberg RJ, Hand MM, Jaffe AS, Julian DG et al: Case definitions for
Control and Prevention; and the National Heart, Lung, and Blood Institute.
at January 2017).
16. Maarten LJG MV, Jacqueline WCM, Michael PJ , and Ewout SW: Net
17. Pencina MJ, D'Agostino RB, Sr., D'Agostino RB, Jr., Vasan RS:
Evaluating the added predictive ability of a new marker: from area under
18. Pearte CA, Furberg CD, O'Meara ES, Psaty BM, Kuller L, Powe NR,
19. Halanych JH, Shuaib F, Parmar G, Tanikella R, Howard VJ, Roth DL,
20. Olubowale OT, Safford MM, Brown TM, Durant RW, Howard VJ, Gamboa
National Death Index: Results From the REasons for Geographic And
21. Brown TM, Deng L, Becker DJ, Bittner V, Levitan EB, Rosenson RS,
Kelman JA: Risk of acute myocardial infarction, stroke, heart failure, and
2010, 304(4):411-418.
23. Soliman EZ, Prineas RJ, Case LD, Russell G, Rosamond W, Rea T,
sudden cardiac death from incident coronary heart disease. Heart 2011,
97(19):1597-1601.
25. Goff DC, Jr., Lloyd-Jones DM, Bennett G, Coady S, D'Agostino RB, Sr.,
26. Stone NJ, Robinson JG, Lichtenstein AH, Goff DC, Jr., Lloyd-Jones DM,
Smith SC, Jr., Blum C, Schwartz JS, Panel AACG: Treatment of blood
30
160(5):339-343.
27. Kenneth D. Kochanek MA, Sherry L. Murphy, B. S., Jiaquan Xu, M.D.,
and therapy. Report of the WHO Task Force on Stroke and other
20(10):1407-1431.
31
Outcome Description
Fatal CVD Includes fatal CHD or fatal stroke
Fatal stroke Was defined as adjudicated stroke followed by death within
28 days. A stroke was adjudicated following the World
Health Organization (WHO) definition[28]. Events not
meeting the WHO definition but characterized by stroke-
related symptoms lasting less than 24 hours with
neuroimaging consistent with acute infarct or hemorrhage
were also classified as strokes
Fatal CHD Includes a fatal MI or CHD death, in accordance to the 2003
“Case definitions for acute coronary heart disease in
epidemiology and clinical research studies” [12].
Fatal MI Was defined as adjudicated definite or probable MI followed
by death within 28 days, following the 2003 “Case
definitions for acute coronary heart disease in epidemiology
and clinical research studies” [12]. A definite MI was defined
by the presence of diagnostic enzyme elevations or
diagnostic changes on the electrocardiogram. A probable MI
was defined by the presence of non-diagnostic enzyme
elevation and positive but non-diagnostic changes on an
electrocardiogram. If enzymes were missing, events were
adjudicated as a probable MI if an electrocardiogram has
positive but non-diagnostic changes in addition to clinical
signs or symptoms of ischemia.
CHD death Was defined following the 2003 “Case definitions for acute
CHD in epidemiology and clinical research studies” [12].
Specifically, CHD death was defined as a death from MI not
meeting the criteria for definite or probable MI (e.g., death
within six hours of hospital admission with cardiac
symptoms and/or signs, but with biomarkers or
electrocardiogram absent or not diagnostic), postmortem
findings consistent with the occurrence of coronary
occlusion within 28 days, or sudden death preceded by
cardiac symptoms or signs without evidence of non-
coronary causes).
CHD: coronary heart disease; CVD: cardiovascular disease; MI: myocardial infarction;
REGARDS: Reasons for Geographic and Racial Differences in Stroke.
32
Age at death date in years, 79.6 (7.6) 79.7 (7.6) 79.1 (7.4) 79.7 (7.6) 78.9 (7.6) 79.6 (7.6) 80.1 (6.9)
Mean (STD)
Female 41.3% 41.3% 40.5% 41.5% 39.5% 40.9% 45.8%
Black race 32.8% 32.8% 34.5% 33.0% 34.0% 33.0% 35.9%
Hospital discharge diagnosis code (any position), within 28 days of death
Myocardial infarction 3.5% 0.4% 14.0% 0.4% 17.85% 3.5% 2.1%
Stroke 4.1% 1.0% 15.0% 4.4% 2.9% 1.0% 59.9%
Physician visit diagnosis code, within 28 days of death
Myocardial infarction 4.5% 1.5% 14.6% 1.5% 18.6% 4.6% 2.1%
Stroke 9.7% 6.4% 21.1% 10.3% 6.9% 6.1% 73.2%
Hospital discharge diagnosis code (any position), beyond 28 days of death
Myocardial infarction 13.6% 12.0% 19.2% 11.7% 22.3% 14.0% 7.0%
Stroke 13.1% 12.3% 15.8% 13.5% 11.3% 12.1% 31.7%
Physician visit diagnosis code, beyond 28 days of death
Myocardial infarction 15.6% 14.0% 21.2% 13.6% 24.6% 16.0% 7.8%
Stroke 29.5% 28.9% 31.4% 30.2% 25.9% 28.2% 51.4%
Hospital discharge diagnosis code (any position), any time prior to death
Abdominal aortic aneurism or 11.7% 11.2% 13.3% 11.4% 12.9% 11.6% 13.4%
Peripheral arterial disease
Transient cerebral ischemia 6.6% 6.6% 6.4% 6.9% 5.0% 6.3% 11.3%
Viral or opportunistic infection 8.9% 9.3% 7.6% 9.3% 7.1% 9.0% 8.5%
End stage renal disease 7.4% 6.8% 9.2% 6.8% 10.2% 7.4% 6.3%
Chronic obstructive pulmonary 44.4% 45.3% 41.5% 44.8% 42.6% 44.9% 36.6%
disease
Organ transplant 0.8% 0.8% 1.0% 0.8% 1.0% 0.8% 0.7%
Gastrointestinal perforation 1.0% 1.0% 0.8% 1.0% 1.0% 1.0% 0.0%
Interstitial lung disease 6.3% 6.6% 5.4% 6.5% 5.2% 6.3% 6.3%
Hernia, Intestinal obstruction, 28.8% 30.2% 23.9% 29.7% 24.6% 29.2% 21.1%
Crohn’s disease
Liver, gall and pancreas 15.9% 17.1% 11.8% 16.8% 12.1% 16.2% 10.6%
disease
General symptoms 38.6% 39.3% 36.0% 39.7% 33.4% 38.2% 45.1%
Symptom involving respiratory 23.5% 23.8% 22.2% 23.7% 22.6% 23.7% 19.7%
and other chest symptoms
Other symptoms involving 9.2% 10.2% 5.6% 9.8% 6.1% 9.5% 3.5%
abdomen and pelvis
Diagnosis code from primary hospital discharge or emergency room visit, any time prior to death
Sudden death 5.3% 2.9% 13.5% 3.0% 16.3% 5.5% 2.8%
Physician visit diagnosis code, any time prior to death
Abdominal aortic aneurism or 26.2% 25.7% 28.1% 25.9% 27.8% 26.1% 28.9%
peripheral arterial disease
Malignancy 45.9% 51.4% 27.3% 50.1% 26.7% 46.9% 28.2%
Other heart disease 73.6% 72.2% 78.5% 72.6% 78.3% 73.4% 78.2%
33
Table 2: Continuation.
All Fatal CVD Fatal CHD Fatal stroke
No Yes No Yes No Yes
N=2,685 N=2,077 N=608 N=2,206 N=479 N=2,543 N=142
Physician visit diagnosis code, any time prior to death
Atrial fibrillation 40.3% 39.6% 42.8% 39.8% 42.6% 40.0% 45.1%
Sudden death or ventricular 21.3% 17.1% 35.7% 17.0% 41.3% 21.6% 15.5%
arrhythmias
Other and unspecified 3.3% 2.7% 5.1% 3.7% 1.5% 2.5% 16.9%
intracranial hemorrhage
Transient cerebral ischemia 21.0% 20.9% 21.2% 21.8% 17.1% 20.3% 33.8%
Viral or opportunistic infection 32.4% 33.1% 30.1% 33.1% 29.4% 32.5% 31.7%
End stage renal disease 20.9% 20.7% 21.9% 20.0% 25.1% 21.5% 10.6%
Chronic obstructive pulmonary 59.0% 59.5% 57.6% 59.2% 58.3% 59.2% 56.3%
disease
Organ transplant 0.9% 0.9% 1.2% 0.9% 1.3% 0.9% 0.7%
Gastrointestinal perforation 1.5% 1.7% 0.7% 1.6% 0.8% 1.5% 0.0%
Interstitial lung disease 7.9% 8.3% 6.7% 8.1% 7.1% 8.1% 5.6%
Hernia, Intestinal obstruction, 39.3% 40.3% 35.9% 40.3% 34.7% 39.2% 41.6%
Crohn’s disease
Liver, gall and pancreas 21.5% 22.9% 16.5% 22.5% 16.5% 21.8% 16.2%
disease
General symptoms 88.5% 89.2% 86.2% 89.5% 83.9% 88.1% 95.1%
Symptom involving respiratory 85.0% 85.3% 83.9% 85.1% 84.6% 85.2% 81.7%
and other chest symptoms
Other symptoms involving 57.8% 60.1% 50.2% 59.7% 49.5% 58.2% 52.1%
abdomen and pelvis
Hospital discharge (any position) or physician visit diagnosis code, any time prior to death
Diabetes 53.5% 51.3% 61.2% 51.8% 61.4% 53.1% 62.0%
Obesity 18.6% 18.6% 18.6% 18.3% 20.3% 18.9% 13.4%
Hyperlipidemia 79.0% 78.3% 81.4% 78.5% 81.4% 78.9% 81.7%
Hypertension 93.1% 92.6% 94.6% 92.9% 93.7% 92.8% 97.9%
Smoking 37.5% 39.0% 32.4% 38.5% 32.8% 37.9% 29.6%
ICD-9-CM procedure code or CPT code, any time prior to death
CABG and PCI 14.7% 14.7% 23.0% 14.6% 25.5% 16.7% 14.8%
Any time prior to death or otherwise specified
Died in hospital 32.3% 30.1% 39.5% 31.4% 36.1% 31.0% 52.1%
Discharge to skilled nursing 42.6% 45.2% 33.7% 45.3% 30.3% 42.4% 46.5%
facility
Discharged to hospice 18.4% 20.4% 11.5% 20.4% 9.4% 18.3% 19.7%
28 days or more for last 3.7% 4.0% 2.6% 4.0% 2.3% 3.7% 4.2%
hospitalization
Number of cardiologist visit 4.9 (8.0) 4.6 (7.8) 6.1 (8.5) 4.6 (7.7) 6.4 (9.0) 4.9 (8.1) 4.7 (5.9)
within 6 months prior to death
Number of oncologist visit 2.5 (7.7) 3.1 (8.6) 0.5 (2.1) 3.0 (8.4) 0.6 (2.3) 2.7 (7.9) 0.2 (0.9)
within 6 months prior to death
Number of neurologist visit 1.1 (3.6) 1.0 (3.5) 1.5 (3.9) 1.1 (3.7) 1.0 (3.0) 1.0 (3.4) 3.3 (5.8)
within 6 months prior to death
Numbers are mean (standard deviation) or percentage.
CHD: Coronary heart disease; CVD: Cardiovascular disease. ICD-9-CM: International classification of diseases, 9th version, clinical
modification.CPT: Current procedure terminology code. STD: Standard deviation.
Table 3: Performance of the direct and indirect approaches for discriminating fatal cardiovascular disease and coronary
heart disease.
Outcome Number C-index Cut- Sensitivity Specificity PPV NPV
predictors (95% CI) points† (95% CI) (95% CI) (95% CI) (95% CI)
Fatal CVD
Direct approach‡ 27 0.86 (0.84, 0.88) 0.32 0.61 (0.57, 0.65) 0.88 (0.87, 0.89) 0.60 (0.56, 0.63) 0.88 (0.87, 0.90)
Indirect approach 1§ NA NA NA 0.62 (0.58, 0.65) 0.89 (0.88, 0.90) 0.62 (0.58, 0.66) 0.89 (0.87, 0.90)
Indirect approach 2¶ # 0.87 (0.85, 0.89) 0.27 0.64 (0.61, 0.68) 0.90 (0.88, 0.91) 0.65 (0.61, 0.69) 0.90 (0.88, 0.91)
Fatal CHD
Direct approach† 26 0.86 (0.84, 0.88) 0.30 0.58 (0.54, 0.63) 0.91 (0.89, 0.92) 0.57 (0.53, 0.62) 0.91 (0.90, 0.92)
Indirect approach 1§ NA NA NA 0.56 (0.51, 0.60) 0.91 (0.90, 0.92) 0.56 (0.52, 0.61) 0.90 (0.89, 0.92)
Indirect approach 2¶ # 0.86 (0.84, 0.88) 0.24 0.58 (0.54, 0.62) 0.91 (0.90, 0.92) 0.59 (0.54, 0.63) 0.91 (0.90, 0.92)
Components used in the indirect approaches
Fatal MI 20 0.91 (0.89, 0.94) 0.22 0.59 (0.52, 0.66) 0.97 (0.96, 0.98) 0.59 (0.52, 0.66) 0.97 (0.96, 0.98)
CHD death 25 0.87 (0.85, 0.89) 0.29 0.52 (0.46, 0.57) 0.92 (0.91, 0.93) 0.51 (0.46, 0.56) 0.93 (0.91, 0.94)
Fatal stroke 13 0.95 (0.92, 0.97) 0.27 0.70 (0.63, 0.78) 0.98 (0.98, 0.99) 0.72 (0.64, 0.79) 0.98 (0.98, 0.99)
CHD: Coronary heart disease; CI: confidence interval; CVD: Cardiovascular disease; MI: myocardial infarction; NA: not applicable, fatal event here was identified based on the
identification of each component; NPV: Negative predictive value; PPV: Positive predictive value.
†:
Cut-points: Cut-points were chosen to provide the closest observed-to-predicted events. Participants with a predicted probability equal to or greater than the cut-points were
considered to have a positive test result.
‡:
Direct approach: Model probability of event (e.g., fatal cardiovascular disease) on candidate predictors.
§:
First indirect approach: conducted separate logistic regression models for fatal MI, CHD death and fatal stroke, using stepwise selection for each outcome, as described above. For
each of these outcomes, we identified the cut-point of predicted probability result in a closest observed-to-predicted event and had a predicted probability above the cut-point (first
indirect approach). Participants with a predicted probability higher than the specific cut-point for any outcome were considered to have a high probability of fatal CVD (i.e., a positive
test result).
¶:
Second indirect approach: used the predicted probabilities for fatal MI, CHD death and fatal stroke from the three separate models described above as the independent variables in a
logistic regression model with fatal CVD as the outcome (second indirect approach).
#: Predicted probability for fatal CVD was derived using a model with three components (predicted probability of fatal MI, fatal stroke, and CHD death). Predicted probability of each
component was determined in separate models (see Components used in the indirect approaches in this table and Table 3). Predicted probability for the 2 components (fatal MI, and
CHD death).
34
Table 4: Performance of algorithms for discriminating cause of death under different cut-points of predicted probability
compared with using hospital discharge diagnosis codes.
Outcome Approach for defining high Cut- Predicted Sensitivity Specificity PPV NPV Sum of NRI
(Observed probability of the outcome points† Proportion% (95% CI) (95% CI) (95% CI) (95% CI) sensitivity and (95% CI)
proportion) specificity
Fatal CVD Discharge diagnosis NA 7.5 0.28 0.99 0.86 0.82 1.27 Reference
(22.7%) codes‡ (0.25, 0.32) (0.98, 0.99) (0.81, 0.91) (0.81, 0.84)
Closest observed-to- 0.27 22.6 0.64 0.90 0.65 0.90 1.56 0.285
predicted (0.61, 0.68) (0.88, 0.91) (0.61, 69) (0.89, 0.91) (0.244, 0.327)
≥0.80 sensitivity 0.15 38.5 0.82 0.74 0.48 0.93 1.56 0.286
(0.78, 0.85) (0.72, 0.76) (0.45, 0.51) (0.92, 0.94) (0.242, 0.330)
≥0.80 PPV 0.54 13.0 0.46 0.97 0.81 0.86 1.43 0.160
(0.42, 0.50) (0.96, 0.98) (0.77, 0.85) (0.85, 0.87) (0.130, 0.194)
Maximize the sum of 0.17 34.5 0.78 0.78 0.51 0.93 1.57 0.296
sensitivity and specificity (0.75, 0.82) (0.77, 0.80) (0.48, 0.55) (0.91, 0.94) (0.251, 0.340)
Fatal CHD Discharge diagnosis NA 3.5 0.18 1.00 0.91 0.85 1.17 Reference
(17.8%) codes§ (0.14, 0.21) (0.99, 1.00) (0.86, 0.97) (0.83, 0.86)
Closest observed-to- 0.24 17.7 0.58 0.91 0.59 0.91 1.49 0.317
predicted (0.54, 0.62) (0.90, 0.92) (0.54, 0.63) (0.90, 0.92) (0.272, 0.364)
≥0.80 sensitivity 0.12 35.3 0.81 0.75 0.41 0.95 1.56 0.385
(0.78, 0.85) (0.73, 0.77) (0.38, 0.44) (0.94, 0.96) (0.337, 0.432)
≥0.80 PPV 0.57 8.2 0.36 0.98 0.80 0.88 1.35 0.174
(0.32, 0.41) (0.97, 0.99) (0.75, 0.85) (0.86, 0.89) (0.140, 0.212)
Maximize the sum of 0.13 32.7 0.79 0.77 0.43 0.95 1.56 0.388
sensitivity and specificity (0.75, 0.83) (0.76, 0.79) (0.40, 0.46) (0.94, 0.96) (0.339, 0.435)
Fatal stroke Discharge diagnosis NA 4.1 0.60 0.99 0.77 0.98 1.59 Reference
(5.3%) codes¶ (0.52, 0.68) (0.99, 0.99) (0.69, 0.84) (0.97, 0.98)
Closest observed-to- 0.27 5.2 0.70 0.98 0.72 0.98 1.69 0.114
predicted (0.63, 0.78) (0.98, 0.99) (0.64, 0.79) (0.98, 0.99) (0.103, 0.182)
≥0.80 sensitivity 0.13 7.5 0.80 0.97 0.57 0.99 1.77 0.180
(0.74, 0.87) (0.96, 0.97) (0.50, 0.64) (0.98, 0.99) (0.119, 0.258)
≥0.80 PPV 0.44 4.1 0.63 0.99 0.81 0.98 1.62 0.030
(0.55, 0.71) (0.99, 1.00) (0.74, 0.88) (0.97, 0.98) (-0.031, 0.090)
Maximize the sum of 0.06 12.6 0.87 0.92 0.36 0.99 1.78 0.194
sensitivity and specificity (0.81, 0.92) (0.91, 0.93) (0.31, 0.42) (0.99, 1.00) (0.124, 0.276)
CHD: Coronary heart disease; CI: Confidence interval; CVD: Cardiovascular disease; NA: Not applicable, dichotomous variable; NPV: Negative predictive value; NRI: Net
reclassification improvement; PPV: Positive predictive value.
†:
Cut-point: Participants with a predicted probability above this cut-point are identified as having a fatal event by algorithms.
‡:
Discharge diagnosis code for cardiovascular disease included International Classification of Diseases, 9th version, clinical modification (ICD-9-CM) 410.xx, 430.xx, 431.xx, 433.x1,
35
434.x1, 436.xx.
§:
Discharge diagnosis code for coronary heart disease included ICD-9-CM 410.xx.
¶:
Discharge diagnosis codes for stroke included ICD-9-CM 430.xx, 431.xx, 433.x1, 434.x1, 436.xx
35
37
algorithms
CHD: Coronary heart disease; CVD: Cardiovascular disease; MI: myocardial infarction
38
N=9,836
Excluded participants for not being linked to Medicare claims Excluded
N=20,403
N=16,076
Excluded participants who were still alive on December 31, 2013 Excluded
N=4,327
N=4,197
N=3,897
N=2,755
Figure 3: Receiver operating characteristic curve for fatal CVD, fatal CHD, and
fatal stroke
Figure 4: Steps to define fatal stroke, fatal CHD and fatal CVD using the claims-
based algorithms
AHRQ CCS category: Agency for Healthcare Research and Quality Clinical Classifications
Software; CHD: Coronary heart disease; CVD: Cardiovascular disease
41
Supplemental Material
Supplemental Table 1: Diagnosis, procedure, and current procedure terminology
codes used as pre-specified candidate predictors for developing algorithms to
discriminate fatal cardiovascular disease, coronary heart disease, and stroke.
†
Candidate predictors Code Code type
Myocardial infarction 410.xx ICD-9-CM diagnosis
Coronary heart disease 411.xx, 412.xx, 413.xx, 414.xx ICD-9-CM diagnosis
Stroke 430.xx, 431.xx, 433.x1, 434.x1, 436.xx ICD-9-CM diagnosis
Abdominal aortic aneurism 441.3x, 441.4x, 441.5x, 441.6x, 441.7x,441.9x ICD-9-CM diagnosis
38.44, 39.25, 39.77 ICD-9-CM procedure
34800, 34802-34805. 34808, 34812, 34813, Current Procedure
34820, 34825, 34826, 34830-34834 Terminology
Peripheral arterial disease 440.20, 440.21, 440.22, 440.23, 440.24, ICD-9-CM diagnosis
440.31, 444.81, 444.2x
37205, 75962 Current Procedure
Terminology
Atrial fibrillation 427.31 ICD-9-CM diagnosis
Malignancy 140.xx-172.99, 174.xx-208.99 ICD-9-CM diagnosis
Other heart disease 390.xx-398.xx, 402.xx, 404.xx, 415.xx, 416.xx, ICD-9-CM diagnosis
421.xx-426.xx, 427.0x, 427.2x, 427.3x, 427.6x,
427.8x, 427.9x, excluded 427.31
Sudden death or ventricular 427.1x, 427.4x, 427.41, 427.42,427.5, ICD-9-CM diagnosis
arrhythmias 798, 798.1, 798.2
Other and unspecified 432.xx ICD-9-CM diagnosis
intracranial hemorrhage
Transient cerebral ischemia 435.xx ICD-9-CM diagnosis
End stage renal disease 585.6x, 586.xx ICD-9-CM diagnosis
Chronic obstructive 490.xx- 496.xx, 500.xx-505.xx, 506.4x ICD-9-CM diagnosis
pulmonary disease
Organ transplant V42.0x, V42.6x, V42.7x, V42.8x ICD-9-CM diagnosis
Gastrointestinal perforation 530.4x, 531.1x, 531.2x, 531.5x, 531.6x, ICD-9-CM diagnosis
532.1x, 532.2x, 532.5x, 532.6x, 533.1x,
533.2x, 533.5x, 533.6x, 534.1x, 534.2x,
534.6x, 569.83
Interstitial lung disease 515.xx, 516.8x, 516.9x, 516.30, 516.31-516.37 ICD-9-CM diagnosis
714.81, 710.9x
Hernia, Intestinal 550.xx- 553.xx, 555.xx-558.xx, 560.xx ICD-9-CM diagnosis
obstruction, Crohn’s
Liver, gall and pancreas 570.xx-577.xx ICD-9-CM diagnosis
disease
General symptoms 780.xx ICD-9-CM diagnosis
Symptom involving 786.xx ICD-9-CM diagnosis
respiratory and other chest
symptoms
Other symptoms involving 789.xx ICD-9-CM diagnosis
abdomen and pelvis
††
Sudden death 427.5x, 798, 798.1x, 798.2x ICD-9-CM diagnosis
Diabetes 250.xx, 357.2x, 362.0x, 366.41 ICD-9-CM diagnosis
Hyperlipidemia 272.0x, 272.1x, 272.2x, 272.4x ICD-9-CM diagnosis
Hypertension 401.0x, 401.1x, 401.9x ICD-9-CM diagnosis
Obesity V85.2x, V85.3x, V85.4x, 278.00, 278.01, ICD-9-CM diagnosis
278.02, 793.91
Smoking 305.1x, V15.82 ICD-9-CM diagnosis
th
ICD-9-CM: International classification of diseases, 9 version, clinical modification.
†
: Any ICD-9-CM diagnosis code or current procedure terminology code for defining a predictor.
††:
Hospital discharge or emergency room or nursing home visit.
42
44
Supplemental Table 4: Net reclassification index for the claims-based algorithm developed in the current study compared
to hospital discharge diagnosis codes for fatal coronary heart disease.
REGARDS Model – Not fatal CHD REGARDS Model – Fatal CHD
(same prevalence) (same prevalence)
NRI (95% CI)
Hospital model – Hospital model – Hospital model – Hospital model –
Not Fatal CHD Fatal CHD Not Fatal CHD Fatal CHD
Adjudicated fatal CHD Overall 0.317 (0.272, 0.364)
Yes 201(42.0%) 0 (0.0%) 193 (40.3%) 85 (17.8%) Event 0.403 (0.358, 0.449)
No 2006 (90.9 %) 3 (0.1 %) 192 (8.7%) 5 (0.2%) Non-Event -0.087 (-0.098, -0.074)
REGARDS Model – Not fatal CHD REGARDS Model – Fatal CHD
(Sensitivity ≥0.80) (Sensitivity ≥0.80)
Hospital model – Hospital model – Hospital model – Hospital model –
Not Fatal CHD Fatal CHD Not Fatal CHD Fatal CHD
Adjudicated fatal CHD Overall 0.385 (0.337, 0.432)
Yes 90 (18.8%) 0 (0.0%) 304 (63.5%) 85 (17.8%) Event 0.635 (0.589, 0.678)
No 1648 (74.7%) 0 (0.0%) 550 (24.9%) 8 (0.4%) Non-Event -0.249 (-0.268, -0.231)
REGARDS Model – Not fatal CHD REGARDS Model – Fatal CHD
(PPV ≥0.80) (PPV ≥0.80)
Hospital model – Hospital model – Hospital model – Hospital model –
Not Fatal CHD Fatal CHD Not Fatal CHD Fatal CHD
Adjudicated fatal CHD Overall 0.174 (0.140, 0.212)
Yes 302 (63.1%) 1 (0.2%) 92 (19.2%) 84 (17.5%) Event 0.190 (0.154, 0.229)
No 2159 (97.9%) 3 (0.1%) 39 (1.8%) 5 (0.2%) Non-Event -0.016 (-0.023, -0.011)
REGARDS Model – Not fatal CHD REGARDS Model – Fatal CHD
(Maximum Sum of sensitivity and (Maximum Sum of sensitivity and
specificity) specificity)
Hospital model – Hospital model – Hospital model – Hospital model –
Not Fatal CHD Fatal CHD Not Fatal CHD Fatal CHD
Adjudicated fatal CHD Overall 0.388 (0.339, 0.435)
Yes 101 (21.1%) 0 (0.0%) 293 (61.2%) 85 (17.8%) Event 0.612 (0.565, 0.655)
No 1705 (77.3%) 0 (0.0%) 493 (22.4%) 8 (0.4%) Non-Event -0.223 (-0.242, -0.206)
CHD: Coronary heart disease; NRI: Net reclassification index.
Overall NRI calculated as event NRI minus non-event NRI.
Event NRI calculated as the percentage of participants with a higher probability of an event minus the percentage with a lower probability of an event among
participants who had a fatal CVD event.
Non-event NRI calculated as the percentage of participants with a lower probability of an event minus the percentage with a higher probability of an event among
participants who did not have a fatal CVD event.
45
Supplemental Table 5: Net reclassification index for the claims-based algorithm developed in the current study compared
to hospital discharge diagnosis codes for fatal stroke.
REGARDS Model – Not fatal stroke REGARDS Model – Fatal stroke
(same prevalence) (same prevalence)
NRI (95% CI)
Hospital model – Hospital model – Hospital model – Hospital model –
Not Fatal stroke Fatal stroke Not Fatal stroke Fatal stroke
Adjudicated fatal stroke Overall 0.114 (0.103, 0.182)
Yes 39 (27.5%) 1 (0.7%) 18 (12.7%) 84 (59.2%) Event 0.120 (0.059, 0.189)
No 2,499 (98.3%) 3 (0.1%) 18 (0.7%) 23 (0.9%) Non-Event -0.006 (-0.010, -0.002)
REGARDS Model – Not fatal stroke REGARDS Model – Fatal stroke
(Sensitivity ≥0.80) (Sensitivity ≥0.80)
Hospital model – Hospital model – Hospital model – Hospital model –
Not Fatal stroke Fatal stroke Not Fatal stroke Fatal stroke
Adjudicated fatal stroke Overall 0.180 (0.119, 0.258)
Yes 27 (19.0%) 1 (0.7%) 30 (21.1%) 84 (59.2%) Event 0.204 (0.132, 0.283)
No 2,455 (96.5%) 2 (0.1%) 62 (2.4%) 24 (0.9%) Non-Event -0.024 (-0.031, -0.018)
REGARDS Model – Not fatal stroke REGARDS Model – Fatal stroke
(PPV ≥0.80) (PPV ≥0.80)
Hospital model – Hospital model – Hospital model – Hospital model –
Not Fatal stroke Fatal stroke Not Fatal stroke Fatal stroke
Adjudicated fatal stroke Overall 0.030 (-0.036, 0.090)
Yes 48 (33.8%) 5 (3.5%) 9 (6.3%) 80 (56.3%) Event 0.028 (-0.031, 0.089)
No 2,514 (98.9%) 8 (0.3%) 3 (0.1%) 18 (0.7%) Non-Event 0.002 (-0.001, 0.005)
REGARDS Model – Not fatal stroke REGARDS Model – Fatal stroke
(Maximum Sum of sensitivity and (Maximum Sum of sensitivity and
specificity) specificity)
Hospital model – Hospital model – Hospital model – Hospital model –
Not Fatal stroke Fatal stroke Not Fatal stroke Fatal stroke
Adjudicated fatal stroke Overall 0.194 (0.124, 0.276)
Yes 18 (12.7%) 1 (0.7%) 39 (27.5%) 84 (59.9%) Event 0.268 (0.188, 0.350)
No 2,328 (91.2%) 1 (0.0%) 189 (7.4%) 25 (1.0%) Non-Event -0.074 (-0.085, -0.064)
NRI: Net reclassification index.
Overall NRI calculated as event NRI minus non-event NRI.
Event NRI calculated as the percentage of participants with a higher probability of an event minus the percentage with a lower probability of an event among
participants who had a fatal CVD event.
Non-event NRI calculated as the percentage of participants with a lower probability of an event minus the percentage with a higher probability of an event among
participants who did not have a fatal CVD event.
46
47
Column D: Contains the coefficients for each predictor and the intercept from the coronary heart disease death algorithm.
Column E: Contains the product of Columns A and D.
†
AHRQ CCS category: Agency for Healthcare Research and Quality Clinical Classifications Software code based on ICD-9-CM diagnosis
code from hospital discharge or physician visit any time prior to death date. A list of these codes are available at https://www.hcup-
us.ahrq.gov/toolssoftware/ccs/ccs.jsp
‡
The probability for having each intermediate outcome (i.e., fatal myocardial infarction and coronary heart disease death, separately) is
calculated as probability = exp(intermediate score)/(1+exp(intermediate score)).
§
The final score is calculated as -3.0188+5.8050*predicted probability of having fatal myocardial infarction +6.0630*predicted probability of
having fatal coronary heart disease, where -3.0188 is the intercept, and 5.8050 and 6.0630 are the coefficients for fatal myocardial
infarction and coronary heart disease death, respectively, calculated from a logistic regression model with the gold standard being
REGARDS adjudicated fatal events. The probability of having fatal CHD was calculated as probability = exp(final score)/(1+exp(final
score)).
50
Column A: All values (except number of vascular visits and number of cardiologist visits within 6 months of the death date) in Column A for
predictors included in algorithm for fatal coronary heart disease are dichotomous, with 1 meaning that the beneficiary had the claim (or is
black), and 0 that the beneficiary did not have the claim (or is not black). For number of vascular visits and number of cardiologist visits
within 6 months of the death date, Column A is an integer for the number of visits. The value for the intercept in Column A is 1 for all
people.
Column B: Contains the coefficients for each predictor and the intercept from the fatal myocardial infarction algorithm. The value for the
intercept in column B must be 1 for all the beneficiaries as this is a constant.
Column C: Contains specific values for each beneficiary used to calculate the Intermediate scores. Values in this column are calculated by
multiplying values for each predictor and the intercept in Column A by Column B.
Column D: Contains the coefficients for each predictor and the intercept from the coronary heart disease death algorithm. The value for the
intercept in column D must be 1 for all the beneficiaries as this is a constant.
Column E: Contains specific values for each beneficiary used to calculate the Intermediate scores. Values in this column are calculated by
multiplying values for each predictor and the intercept in Column A by Column D.
Column F: Contains the coefficients for each predictor and the intercept from the fatal stroke algorithm. The value for the intercept in
column F must be 1 for all the beneficiaries as this is a constant.
Column G: Contains specific values for each beneficiary used to calculate the Intermediate scores. Values in this column are calculated by
multiplying values for each predictor and the intercept in Column A by Column F.
†
Hospital discharge, emergency room or nursing home visit diagnosis, any time prior to death date.
††
Physician visit diagnosis code, any time prior to death
†††
Agency for Healthcare Research and Quality Clinical Classifications Software code based on ICD-9-CM diagnosis code from hospital
discharge or physician visit prior to death date. A list of these codes are available at https://www.hcup-
us.ahrq.gov/toolssoftware/ccs/ccs.jsp.
‡
The probability for having each intermediate outcome (i.e., fatal myocardial infarction, coronary heart disease death or fatal stroke) is
calculated as probability = exp(score)/(1+exp(score)).
§
The final score is calculated as -2.8527+5.2219*predicted probability of having fatal myocardial infarction +5.7589*predicted probability of
having fatal coronary heart disease + 5.0264*predicted probability of having fatal stroke, where -2.8527 is the intercept, and 5.2219, 5.7589
and 5.0264 are the coefficients for fatal myocardial infarction, coronary heart disease death and fatal stroke, respectively, calculated using
a logistic regression model using REGARDS study procedures as the gold standard. The probability of having fatal cardiovascular disease
was calculated as probability = exp(final score)/(1+exp(final score)).
52
Tocilizumab and the risk for cardiovascular disease: a direct comparison among
biologic disease-modifying antirheumatic drugs for rheumatoid arthritis patients in
a real world setting
cardiovascular disease
Word count:
Abstract: 250
Tables: 3
Figures: 3
54
lipid profiles associated with IL6 receptor antagonists (IL-6R) and some other
(CVD) risk associated with the first approved anti IL-6R medication for RA,
modifying antirheumatic drugs after January 1, 2010 and with 365 days medical
and pharmacy coverage before initiation. The primary outcome was a composite
of myocardial infarction (MI), stroke and fatal CVD, assessed using a validated
Results: Total of 88,463 RA patients were included. The crude incidence rate
(IR) per 1000 patient-years for composite CVD among Medicare patients ranged
from 11.8 (95%CI: 9.7-14.4) for etanercept to 17.3 (95%CI: 15.2-19.7) for
rituximab users. The crude incidence rate for pooled TNFi users was 15.0 (13.9-
16.3). Compared to tocilizumab, the adjusted hazard ratios were 1.01 (0. 79-
1.28) for abatacept, 1.16 (0.89-1.53) for rituximab, 1.10 (0.80-1.51) for
etanercept, 1.33 (0.99-1.80) for adalimumab, and 1.61 (1.22-2.12) for infliximab.
tocilizumab and any other biologic using MarketScan data. Results were robust
trial, there was no increased risk for patients treated with TCZ compared
of the adult population [4, 5]. If RA is uncontrolled, poor quality of life and
related death accounts for a 50% excess premature mortality in RA patients [12].
Traditional risk factors do not fully explain the excessive risks for CVD in RA
C-reactive protein (CRP) and tumor necrosis factor (TNF). These cytokines and
Many studies have suggested that treatment with TNFi reduces the risk of CVD
in RA patients, but results from such investigations are not always consistent
[24].
[25]; it is effective in the treatment for RA patients [26]. Multiple studies have
57
with TCZ [1-3]; these results raise the concern for the possibility that TCZ may
increase the risk of CVD events. Few studies have compared the risk for CVD
associated with TCZ to other bDMARDS for the treatment for RA patient. A
to ETA among high-risk RA patients with one of several risk factors for CVD and
found no increased CVD risk associated with TCZ compared to ETA [27]. Two
(MI), stroke, or all-cause mortality associated with TCZ compared to TNFi or ABA
[28, 29]. All these studies had important limitations: the clinical trial only
compared TCZ to ETA and not to other therapies; the two observational studies
did not include fatal CVD as an outcome and all three studies were relatively
limited in their sample size and number of events in the TCZ-exposed groups (83
major adverse cardiovascular events total in the trial, and 32 and 22 composite
events [30]. This may be even more problematic in studies involving RA patients
since they experience a higher proportion of sudden CHD death than non-RA
subjects [31]. Due to sample size limitations, one of the observational studies
TNFi therapies, and each of the medications in this class may have somewhat
different drug-specific risk for CVD. Finally, the observational studies did not
58
assess the impact of disease activity among biologic users to address the
potential for confounding by RA disease activity. Prior work has shown that RA
The objective of this study was to assess the CVD risk associated with TCZ
(e.g. RIT, ABA). As part of the investigation, we evaluated the potential for
stratifying patients with respect to key CVD risk factors to identify both higher and
METHODS
Data source and study design: We conducted a retrospective cohort study using
January 1, 2006 through September 30, 2015 in both Medicare and MarketScan
plan provided for US residents 65 or older, and Medicare also covers disabled
patients and those with end stage of renal disease who are younger than age 65
years of age. RA is one of the reasons for qualifying for disability. MarketScan
data contains claims for commercially insured and relatively younger RA patients
MarketScan participants must have 1) initiated at least one of the bDMARDS for
in the US); 2) were diagnosed with having RA, defined by two International
diagnosis codes for RA, separated by 7 days and within one year before
initiation; 3) had at least 365 days observable with both medical and pharmacy
(hospital), Part B (outpatient) and Part D (pharmacy), and not enrolled in Part C
complete). This 365 days baseline period was used for assessing co-morbidities
and ensuring the initiation was the first use of each specific medication.
Cohort exclusion: RA patients were excluded from the cohort if they 1) had ICD-
spondylitis in the baseline to ensure that biologic treatment was for RA; 2) using
all available data preceding the start of follow-up (not just the 12 month baseline),
had any ICD-9-CM diagnosis code for past MI, stroke, ICD-9 procedure code or
Table 1); 3) had history of malignancy (ignoring non-melanoma skin cancer), HIV
earliest of 1) end of exposure of the specific agent plus a 90 day extension (see
60
date and days of supply for injection drugs and recommended dosing intervals
for infusion drug (30 days for intravenous ABA and TCZ, 56 days for INF and 180
days for RIT) with a 90 days extension to both injection and infusion drugs.
Injection drugs were identified using national drug codes (NDC) in Part D, and
System (HCPCS) codes in Part B. Infusion drugs (ABA, CER and TCZ) in
published and validated algorithm based on units, unit price, diagnosis code and
infusion/injection code associated with use of the non-specific J code 3490 and
3590 (sensitivity 94%, specificity 100% and positive predictive value (PPV) 97%)
[32].
lipid lowering drug, were identified using NDC codes in Part D in the baseline
period; statins were classified as low, moderate and high potency based on the
glucocorticoids was calculated for the most recent 6-month interval in baseline
and was classified as none, low dose (<7.5 mg/day) or higher dose (≥7.5
mg/day). Number of past bDMARDS use was assessed using all available data
61
obesity, chronic kidney disease, and smoking), other co-morbidities and health
behaviors and health care utilization were assessed in baseline, chosen based
including other acute, subacute forms of ischemic heart disease, angina pectoris,
and other forms of chronic ischemic heart disease, other and unspecified
incident stroke and fatal CVD. Incident MI was defined as: at least one ICD-9-CM
diagnosis code (Supplemental Table 1) for MI from hospital discharge and with at
least one night stay in hospital unless the patient died (positive predictive value
least one ICD-9-CM diagnosis code (Supplemental Table 1) for stroke from
hospital discharge (PPV ≥90%) [35]. Fatal CVD was identified by a validated
Statistical methods
Mean and standard deviation were calculated for continuous variables and
proportions were calculated for categorical variables. Incidence rates (IRs) were
ratios (aHR) and 95% CIs. Demographic character ( age, sex), co-morbidities
drug utilization (methotrexate, NSAIDS, statin potency, other lipid lowering drug
use in baseline, number of biologic used prior initiation, oral steroid dose in six
months before initiation), smoking were included in the adjusted model. Race,
State Buy In (a proxy for low income), reason for Medicare eligibility other than
The proportional hazard assumption were assessed using the Lin, Wei and Ying
method [36]. Robust sandwich covariance matrix estimation was used to account
for the clustered nature of the data since one patient could have contributed to
multiple bDMARDS exposure groups if they initiated multiple biologics over the
study period.
Subgroup analyses for lower risk and higher risk patients on the basis of a
hxCVD were conducted. Another subgroup analysis was performed that required
prior exposure to at least one bDMARDS, given that most TCZ users were
previously exposed to other bDMARDS. Finally, the three most commonly used
TNFi (ADA, ETA and INF) were compared individually to TCZ in separate
models[37].
63
We performed several sensitivity analyses for the fatal CVD outcome. By varying
the threshold for defining a patients as having high probability for fatal CVD, the
resulted in equal sensitivity and PPV, wherein the predicted proportion of events
potential confounding for RA disease activity was examined using the Multi-
subgroup of patients who have MBDA test results (within six months before
medication initiation); the effect size of MBDA (RR CD ) on CVD risk was obtained
from published data [39] that found that the risk of MI for patients in moderate or
high RA disease activity was 1.5-fold elevated compared to people who had low
RA disease activity.
All analyses were conducted using SAS 9.4 (SAS Institute Inc., Cary, NC). Use
of the Medicare data was governed by a Data Use Agreement (DUA) with the
Centre for Medicare and Medicaid Services. The institutional Review Board of
RESULTS
(Figure 1). After applying the inclusion and exclusion criteria, the final cohort
64
8,469 episodes for RIT to 77,048 episodes for TNFi. The mean (SD) age was
White in Medicare.
and RIT users except that TCZ users were less likely to be naïve to bDMARDS.
Compared to TNFi users, TCZ users were more likely to be white, with hxCVD,
heart failure and atrial fibrillation, hospitalized and had more physician visits in
baseline; TCZ users were less likely to be diabetic, use methotrexate in the
Medicare data
The crude IRs for the composite CVD outcome ranged from a low of 11.8
(95%CI: 9.7-14.4) for ETA to a high of 17.3 (95%CI: 15.2-19.7) for INF per 1,000-
person years (PYs) (Table 2, left, top panel). The crude IR for TCZ was 12.9
(10.7, 15.7). The IR for a pooled TNFi exposure category (for all 5 TNFi
TCZ (Figure 2, top panel) were 1.10 (95%CI: 0.80-1.51) for ETA, 1.01 (95%CI: 0.
79-1.28) for ABA, 1.16 (95%CI: 0.89-1.53) for RIT and 1.61 (95%CI: 1.22-2.12)
65
for INF. Pooling all 5 TNFi therapies, the aHR was 1.27 (95%CI: 1.02-1.59)
referent to TCZ.
Trends were similar looking at individual events that comprised the composite
CVD outcome. ETA had the lowest and INF had the highest IRs for both MI and
stroke. (Table 2, left). ETA and TCZ had the lowest and RIT had the highest IRs
for fatal CVD. The HRs for MI and stroke were shown in supplemental figure 1.
Among RA patients in Medicare cohort who died, 44% died in hospital. Fatal
(Table 3, left, top panel), RA patients with hxCVD had a much higher risk for
composite CVD (Table 3, left, bottom panel). The aHRs for each bDMARDS for
second panel) were numerically similar to aHRs for whole cohort respectively,
Results for sub group analysis restricted to RA patients who had prior experience
with bDMARDS (44.4% of overall cohort) are shown in supplemental table 2 (left
panel). TCZ users had the lowest (12.3 events per 1000 person- years) and INF
users the highest (18.1 events per 1000 person- years) crude IR for composite
CVD; Crude IRs for ABA, ETA, ADA, RIT and pooled TNFi were between the IRs
for TCZ and INF. The corresponding aHR were shown in supplemental figure 2
(top panel).
66
Results from the sensitivity analysis for using more sensitive and for equal
sensitivity and PPV algorithm to identify fatal CVD were shown in supplemental
table 3 (left panel) and supplemental figure 3 (first and second panel). The
increases in crude IRs when using more sensitive algorithm and using equal
sensitivity and PPV algorithm were similar among the bDMARDS (data not
shown). The aHR for each bDMARDS when using more sensitive or equal
sensitivity and PPV algorithm were numerically similar to corresponding aHRs for
aHRs for analysis with more sensitive algorithm had a narrower 95%CI.
MarketScan data
The crude IRs from MarketScan data were about 50% of the corresponding IRs
from Medicare data, as Market Scan patients were more than 10 years younger
than Medicare patients. The IRs ranged from 5.2 (95%CI: 4.1-6.6) to 11.0
(95%CI: 7.92-15.3) for composite CVD 1,000 person years (PYs) (Table 2, right,
top panel); 2.0 (95CI: 1.0-3.9) to 7.2 (95%CI: 4.8-10.8) for MI; and from 2.7
The adjusted HRs from MarketScan data were similar to corresponding HRs from
Results from the MBDA lab tests linked to Medicare data are shown in
categories were found amongst RTX and TCZ users, and TNFi users were more
composite CVD comparing ETA to TCZ was minimally different (HR=1.11) after
DISCUSSION
patients who initiated one or more bDMARDS for the treatment for RA, we
observed comparable risks for composite CVD associated with TCZ versus most
other biologic therapies including ETA, ABA and a variety of other therapies,
irrespective of patients’ history of CVD risk factors. Among the three most used
TNFi, ETA was associated with lowest risk and INF associated with highest risk
for CVD. Compared to TCZ, INF was the only RA biologic associated with a
levels in TCZ treated RA patients, higher than the increase in LDL cholesterol
[1-3]. These observations raise the concern that TCZ may associated with
increased risk for CVD. However, these findings do not appear to translate into a
higher risk for CVD events based on our results and those from a recently
NCT01331837).
68
In a study using three data sources (including Medicare, accounting for 54% of
all events), Kim et al. compared TCZ to TNFi for the risk for MI, stroke and all-
cause mortality [28] and showed that TCZ was associated with numerically but
not significant lower risk for all outcome studied except heart failure . However,
the data was based only 32 events in their composite outcome for TCZ-exposed
patients. In our study, 125 CVD events were observed in the TCZ exposure
the CVD risk associated with TCZ to ETA among adult RA patient with increased
risk for CVD. The MACE outcome included incident MI, stroke, and CVD death,
similar to the definition for our primary outcome. They found 83 MACE outcomes
in 4,900 person years in TCZ, yielding an IR of 16.9 per 1000 person years [27].
The IR is between our IR for patient without a history of CVD (11.9) and patient
with a history of CVD (21.6). This was to be expected given that their definition
for increased risk for CVD was more inclusive but included weaker CVD risk
ETA). The corresponding HR for TCZ vs. ETA in our study was 0.91 (1/1.10),
well within their 95%CI: (0.77-1.43). We did not observe an increased risk for
Our study has several strengths. We included fatal CVD as a part of a CVD
composite outcome. Fatal CVD was identified through a validated claims based
algorithm [30], and not including fatal CVD events may underestimate the IR for
69
CVD and may result in a biased estimate. In our study, more than half of fatal
events would have been missed in the Medicare data if only inpatient deaths
CER and TCZ) represented with non-specific J code 3490 and 3590 before a
specific J code was available, failure to identify these infusion drugs will reduce
the sample size and more importantly will misclassify some non-new user as new
users and also misclassify the date of drug initiation. Our large sample size made
it possible to assess CVD risk in subgroups with and without a history of CVD; to
estimate IR and HRs with high precision; and to compare TCZ to each individual
TNFi. Given some differences in CVD risk between TNFi therapies, we would
suggest that it is undesirable to pool all TNFi together if they could be compared
to assess the distribution of disease activity. TCZ users had a higher proportion
computed HR for composite CVD comparing ETA to TCZ was no different from
that RA disease activity, which has been previously shown to be associated with
Our study has some limitations common to other studies using administrative
data. Claims data only has prescription filling information, and patients filling a
prescription not necessary take the medication, and thus we may misclassify
which represents most of the TCZ, ABA and INF exposure in this data. Also, the
number of outcomes in MarketScan was relatively small, and thus the estimates
in that data source may not be stable. Finally, CVD events were not individually
of 80-90% in their accuracy, and there is little reason to expect that performance
trial in an economic way that provides high precision: TCZ was not associated
with increased or reduced CVD risk compared to ETA. However, unlike the
clinical trial which enrolled only higher risk patients, we extended this finding to
“low CVD risk” RA patients. We further showed that TCZ was associated with
reduced CVD risk when compared to a pooled TNFi exposure, mostly attributable
References
31(4):451-456.
2. Strang AC, Bisoendial RJ, Kootte RS, Schulte DM, Dallinga-Thie GM,
Levels JH, Kok M, Vos K, Tas SW, Tietge UJ et al: Pro-atherogenic lipid
75(10):1806-1812.
4. Helmick CG, Felson DT, Lawrence RC, Gabriel S, Hirsch R, Kwoh CK,
Liang MH, Kremers HM, Mayes MD, Merkel PA et al: Estimates of the
8. Solomon DH, Karlson EW, Rimm EB, Cannuscio CC, Mandl LA, Manson
107(9):1303-1307.
93(2):198-200.
10. Wong JB, Ramey DR, Singh G: Long-term morbidity, mortality, and
44(12):2746-2749.
11. Mok CC, Kwok CL, Ho LY, Chan PT, Yip SF: Life expectancy,
63(5):1182-1189.
13. del Rincon ID, Williams K, Stern MP, Freeman GL, Escalante A: High
2001, 44(12):2737-2745.
14. Ridker PM, Cushman M, Stampfer MJ, Tracy RP, Hennekens CH:
979.
15. Ridker PM, Rifai N, Stampfer MJ, Hennekens CH: Plasma concentration
108(3):205-213.
17. Moreland LW, Baumgartner SW, Schiff MH, Tindall EA, Fleischmann RM,
necrosis factor receptor (p75)-Fc fusion protein. The New England journal
354(9194):1932-1939.
19. Weinblatt ME, Keystone EC, Furst DE, Moreland LW, Weisman MH,
Birbara CA, Teoh LA, Fischkoff SA, Chartash EK: Adalimumab, a fully
45.
Agarwal S et al: The efficacy and safety of rituximab in patients with active
22. Smolen JS, Kay J, Doyle MK, Landewe R, Matteson EL, Wollenhaupt J,
Gaylis N, Murphy FT, Neal JS, Zhou Y et al: Golimumab in patients with
75
23. Kremer JM, Genant HK, Moreland LW, Russell AS, Emery P, Abud-
24. Westlake SL, Colebatch AN, Baird J, Curzen N, Kiely P, Quinn M, Choy E,
Ostor AJ, Edwards CJ: Tumour necrosis factor antagonists and the risk of
26. Maini RN, Taylor PC, Szechinski J, Pavelka K, Broll J, Balint G, Emery P,
2829.
28. Kim SC, Solomon DH, Rogers JR, Gale S, Klearman M, Sarsour K,
29. Zhang J, Xie F, Yun H, Chen L, Muntner P, Levitan EB, Safford MM, Kent
30. Fenglong Xie LDC, Jeffrey R. Curtis, Meredith L. Kilgore, Emily B. Levitan,
submitted.
31. Maradit-Kremers H, Crowson CS, Nicola PJ, Ballman KV, Roger VL,
32. Curtis JR, Xie F, Chen R, Chen L, Kilgore ML, Lewis JD, Yun H, Zhang J,
Guideline 2014.
35. Kumamaru H, Judd SE, Curtis JR, Ramachandran R, Hardy NC, Rhodes
JD, Safford MM, Kissela BM, Howard G, Jalbert JJ et al: Validity of claims-
7(4):611-619.
36. Lin D, Wei, L. J., and Ying, Z. : Checking the Cox Model with Cumulative
37. Lee EW, Wei, L. J. Amato, D.: Cox-Type Regression Analysis for Large
303.
78
39. Curtis JR, Xie F, Chen L, Saag KG, Yun H, Muntner P: Biomarker-related
diseases 2017.
40. Dixon WG, Watson KD, Lunt M, Hyrich KL, British Society for
Table 2: Incidence rates for composite cardiovascular event and its components
(Acute myocardial infarction, stroke or fatal CVD†) by biologic DMARDS by data
source.
Medicare MarketScan
Number Person Incidence rate Number Person Incidence rate
Outcome Exposure of event years (95%CI) of event years (95%CI)
Tocilizumab 104 8,045 12.9 (10.7, 15.7) 21 4,064 5.2 (3.4, 7.9)
Myocardial Pooled TNFi 600 39,973 15.0 (13.9, 16.3) 222 38,102 5.8 (5.1, 6.6)
infarction, Adalimumab 121 8,634 14.0 (11.7, 16.8) 67 12,938 5.2 (4.1, 6.6)
stroke, fatal Etanercept 102 8,623 11.8 (9.7, 14.4) 75 13,553 5.5 (4.4, 6.9)
cardiovascul Infliximab 230 13,308 17.3 (15.2, 19.7) 42 5,128 8.2 (6.1, 11.1)
ar disease Abatacept 199 14,561 13.7 (11.9, 15.7) 67 7,686 8.7 (6.9, 11.1)
Rituximab 105 6,337 16.6 (13.7, 20.1) 35 3,190 11.0 (7.9, 15.3)
82
83
N subject=126,563/77,834
Excluded due to without 365 days observable
N episode=168,697/97,528
N subject=85,314/70,377
N episode=113,082/94,086
N subject=2/1,961
Excluded due to age younger than 18
N episode=2/2,572
N subject=85,312/68,416
N episode=113,080/91,514
N subject=76,048/52,585
N episode=101,740/70,766
N subject=65,368/47,086
N episode=86,543/62,893
N subject=56,199/44,196
N episode=74,649/59,103
N subject=46,648/41,815
N episode=61,715/55,778
84
Figure 2: Forest plot for adjusted hazard ratio for primary outcome by data
source.
CI: confidence interval; CVD: Cardiovascular disease; HR: hazard ratio; IR: incidence rate;
Pooled TNFi: Pooled tumor necrosis factor inhibitor including adalimumab, certolizumab,
etanercept, golimumab and infliximab; Primary outcome including myocardial infarction, stroke
and fatal CVD identified with algorithm with cut-points for positive predictive value ≥0.80.HR:
Adjusted for History of cardiovascular disease(excluded acute myocardial infarction, old
myocardial infarction, stroke), age, sex, Heart failure, Atrial fibrillation, Abdominal aortic aneurism,
Peripheral arterial disease, Diabetes, Hyperlipidemia, Hypertension, Obesity, Chronic kidney
disease, Chronic obstructive pulmonary disease, Fibromyalgia, any hospitalized infection, any
hospitalization, number of physician visit, Methotrexate, NSAIDS, Statin potency, other lipid
lowering drug use in baseline, number of biologic used prior initiation, Oral steroid dose in six
month before initiation, Smoking. Race, State Buy In (poor), Reason other than age for eligible for
Medicare were also adjusted for Medicare data.
85
Figure 3: Forest plot for adjusted hazard ratio for primary outcome by history of
CI: confidence interval; CVD: Cardiovascular disease; HR: hazard ratio; IR: incidence rate;
Pooled TNFi: tumor necrosis factor inhibitor. Pooled tumor necrosis factor inhibitor including
adalimumab, certolizumab, etanercept, golimumab and infliximab; Primary outcome including
myocardial infarction, stroke and fatal CVD identified with algorithm with cut-points for positive
predictive value ≥0.80.HR: Adjusted for age, sex, Heart failure, Atrial fibrillation, Abdominal aortic
aneurism, Peripheral arterial disease, Diabetes, Hyperlipidemia, Hypertension, Obesity, Chronic
kidney disease, Chronic obstructive pulmonary disease, Fibromyalgia, any hospitalized infection,
any hospitalization, number of physician visit, Methotrexate, NSAIDS, Statin potency, other lipid
lowering drug use in baseline, number of biologic used prior initiation, Oral steroid dose in six
86
month before initiation, Smoking. Race, State Buy In (poor), Reason other than age for eligible for
Medicare were also adjusted for Medicare data.
87
89
90
Supplemental figure 1: Forest plot for adjusted hazard ratio for myocardial
infarction and stroke by data source.
CI: confidence interval; HR: hazard ratio; IR: incidence rate; Pooled TNFi: Pooled tumor necrosis
factor inhibitor including adalimumab, certolizumab, etanercept, golimumab and infliximab.HR:
Adjusted for History of cardiovascular disease(excluded acute myocardial infarction, old
myocardial infarction, stroke), age, sex, Heart failure, Atrial fibrillation, Abdominal aortic aneurism,
Peripheral arterial disease, Diabetes, Hyperlipidemia, Hypertension, Obesity, Chronic kidney
disease, Chronic obstructive pulmonary disease, Fibromyalgia, any hospitalized infection, any
hospitalization, number of physician visit, Methotrexate, NSAIDS, Statin potency, other lipid
lowering drug use in baseline, number of biologic used prior initiation, Oral steroid dose in six
month before initiation, Smoking. Race, State Buy In (poor), Reason other than age for eligible for
Medicare were also adjusted for Medicare data.
92
Supplemental figure 2: Forest plot for adjusted hazard ratio for primary outcome
for biologic DMARDS experienced by data source.
CI: confidence interval; CVD: Cardiovascular disease; HR: hazard ratio; IR: incidence rate;
Pooled TNFi: Pooled tumor necrosis factor inhibitor including adalimumab, certolizumab,
etanercept, golimumab and infliximab; Primary outcome including myocardial infarction, stroke
and fatal CVD identified with algorithm with cut-points for positive predictive value ≥0.80. HR:
Adjusted for History of cardiovascular disease(excluded acute myocardial infarction, old
myocardial infarction, stroke), age, sex, Heart failure, Atrial fibrillation, Abdominal aortic aneurism,
Peripheral arterial disease, Diabetes, Hyperlipidemia, Hypertension, Obesity, Chronic kidney
disease, Chronic obstructive pulmonary disease, Fibromyalgia, any hospitalized infection, any
hospitalization, number of physician visit, Methotrexate, NSAIDS, Statin potency, other lipid
lowering drug use in baseline, number of biologic used prior initiation, Oral steroid dose in six
month before initiation, Smoking. Race, State Buy In (poor), Reason other than age for eligible for
Medicare were also adjusted for Medicare data.
93
Supplemental figure 3: Forest plot for adjusted hazard ratio by definition of fatal
cardiovascular disease and data source.
CI: confidence interval; CVD: Cardiovascular disease; HR: hazard ratio; IR: incidence rate;
Pooled TNFi: tumor necrosis factor inhibitor. Pooled tumor necrosis factor inhibitor including
adalimumab, certolizumab, etanercept, golimumab and infliximab; PPV: positive predictive value;
Sensitive definition including myocardial infarction, stroke and fatal CVD identified with algorithm
with cut-points for sensitivity ≥0.80; Equal sensitivity and PPV including myocardial infarction,
stroke and fatal CVD identified with algorithm with cut-points for sensitivity equal PPV (Predicted
proportion of events equal observed proportion of events.HR: Adjusted for History of
cardiovascular disease(excluded acute myocardial infarction, old myocardial infarction, stroke),
age, sex, Heart failure, Atrial fibrillation, Abdominal aortic aneurism, Peripheral arterial disease,
Diabetes, Hyperlipidemia, Hypertension, Obesity, Chronic kidney disease, Chronic obstructive
pulmonary disease, Fibromyalgia, any hospitalized infection, any hospitalization, number of
physician visit, Methotrexate, NSAIDS, Statin potency, other lipid lowering drug use in baseline,
number of biologic used prior initiation, Oral steroid dose in six month before initiation, Smoking.
94
Race, State Buy In (poor), Reason other than age for eligible for Medicare were also adjusted for
Medicare data.
95
Methotrexate use and the risk for cardiovascular disease among rheumatoid
arthritis patients initiating biologic disease-modifying antirhematic drugs
Fenglong Xie1, 2, Lang Chen1, Huifeng Yun1, 2, Emily B. Levitan2, Paul Muntner2,
Jeffrey R. Curtis1, 2
cardiovascular disease
Word count:
Abstract: 250
Tables: 4
Figures: 1
97
unknown.
methotrexate on CVD risk using 2006-2015 Medicare claims data for RA patients
initiating bDMARDS. Methotrexate use was defined as time varying. The primary
outcome was composite of incident myocardial infarction, stroke, and fatal CVD.
Incidence rates (IR) and 95% confidence intervals (CI) were calculated using
assessed using Cox regression for overall, and for each underlying bDMARDS.
Results: A total of 88,255 DMARDS initiations were included in this study. The
average age was 64.6 (12.3) years, 84.0% were female, 68.2% were non-
Hispanic white. The crude IRs for CVD were 12.1(95%CI: 11.1-13.2) and 17.9
(16.9-18.8) events per 1,000-person years for RA patients with and without time
methotrexate and bDMARDS was significant. The contrast HRs for time varying
methotrexate (Yes vs. No) ranged from 0.58 (0.35, 0.96) for certolizumab
initiators to 0.91 (0.69, 1.20) for etanercept initiators. Results were robust in
CVD risk associated with time varying methotrexate use. The association
0.5-2% of the population [1-3]. Without effective treatment, RA often leads to joint
account for more than 50% of the excess mortality in RA patients [11].
CRP and tumor necrosis factor: TNF). These play an important role in the
increased risk for CVD in RA patients, and they also have been shown to
recommended for the treatment for RA [15, 16]. RA patient without adequate
(bDMARDS). While some bDMARDS appear to have appreciable efficacy for the
methotrexate.
associated with reduced risk for CVD [17-26]. Most studies have concluded
inflammation. However, in the setting of other therapies that can also reduce
The objective of this study was to assess the effect of methotrexate on CVD risk
METHODS
Population and data source: This study utilized Medicare claims data for 100%
Medicare also covers disabled individuals (e.g. due to RA) or with end-stage of
who initiated a bDMARDS between January 1, 2007 and September 30, 2015.
separated by 7 days and occurring within one year. Injection bDMARDS were
identified from Medicare Part D using national drug codes (NDC) and parental
tocilizumab) in the first year after regulatory approval and before a permanent
HCPCS code was available were identified using a published algorithm based on
units, unit price, diagnosis code and infusion/injection codes associated with the
nonspecific HCPCS codes 3490 and 3590 (sensitivity 94%, specificity 100% and
had to have no previous claim for that specific therapy using all available prior
data and had to have a minimum of 365 days of prior data. These 365 days were
101
bDMARD who did not meet this criterion would have been prevalent users and
while excluded from analysis of that specific bDMARD, might still have qualified
for initiation of other bDMARDS. Patients with ICD-9-CM diagnosis code for
before initiation date (using all available data) were also excluded. RA patients
with diagnosis code for myocardial infarction (MI: ICD-9-CM 410.xx), old MI (ICD-
9-CM 412.xx) and stroke (ICD-9-CM 430.xx, 431.xx, 433.xx, 434.xx and 436.xx)
before initiation date (using all available data) were excluded so as to allow for
the study of incident CVD events. For all other covariates, 365 days prior to each
biologic initiation date served as baseline for assessing co-morbidities and other
medication exposures. To make sure the treatment was for RA, patients with
Follow up: Follow up start at the initiation of the bDMARD and end at earliest of
1) end of exposure of the specific agent (days of supply plus 90 days extension);
for fatal CVD); 5) loss of Medicare coverage 6) end of study (September 30,
2015).
baseline period with exception of the history of CVD (hxCVD) other than MI or
stroke, which was identified using all available data. All co-morbidities and
calculated as a 183 days average before initiation and classified as none, low
dose (<7.5 mg/day) or higher dose (≥7.5 mg/day). Statin use was identified using
NDC codes and classified as none, low, medium and high potency [28] based on
the last prescription in baseline. Folic acid use was identified using NDC codes
incident hemorrhagic or ischemic stroke and fatal CVD. Incident MI was defined
as: at least one ICD-9-CM diagnosis code for MI (410.x1) from hospital discharge
and with at least one night stay in hospital unless the patient died
(PPV≥90%)[29]. Incident stroke was defined as: at least one ICD-9-CM diagnosis
code for stroke (430.xx, 431.xx, 433.x1, 434.x1 and 436.xx) from hospital
discharge (PPV ≥90%) [30].Fatal CVD was identified by a claims based algorithm
with PPV ≥80% [31]. The secondary outcomes were individual components of
Statistical methods
Descriptive statistics were calculated for patients with and without methotrexate
use. Mean and standard deviation were calculated for continuous variables and
(SMD) were calculated for both continuous and categorical variables, SMD
greater than 0.1 was considered as indication for non-balance. Incidence rates
Unadjusted and multivariable adjusted hazard ratios (aHR) and 95%CI were
estimated using Cox regression (using initiation date as time origin). Age, sex,
race, heart failure, atrial fibrillation, abdominal aortic aneurism, peripheral arterial
any hospitalization, number of physician visit, folic acid, NSAIDS, statin potency,
other lipid lowering drug use in baseline, number of biologic used prior initiation,
oral steroid dose in six month before initiation, smoking, poor (enrolled in state
buy-In program), reason other than age for eligible for Medicare were included in
was used to account for the clustered nature of the data since each patient could
Interaction between time varying methotrexate and back ground bDMARDS was
tested.
104
One variable with 16 levels (e.g. abatacept with methotrexate, abatacept without
multivariable adjusted COX model, the effect of methotrexate on the risk for CVD
methotrexate in order to assure that the not current methotrexate exposure group
was at some point deemed well enough to have received methotrexate. This
not give patients methotrexate (e.g. alcoholism) that might not be measured well
that added a 90-day extension to days of supply of methotrexate, for the effect of
methotrexate may last after quilting methotrexate or the reason for quilting
was examined using the Multi-Biomarker Disease Activity (MBDA) test results
subgroup of patients who had their data linked to MBDA test results (within six
months before bDMARDS initiation); the effect size of MBDA (RR CD ) on CVD risk
105
was obtained from published data [32]that found that the risk of MI for patients in
who had low RA disease activity. All analyses were conducted using SAS 9.4
(SAS Institute Inc., Cary, NC). Use of the Medicare data was governed by a Data
Use Agreement (DUA) with the Centre for Medicare and Medicaid Services. The
the study.
RESULTS
applying the inclusion and exclusion criteria, 88,255 initiations (64,218 patients)
were included in this study (Figure 1). The average age at initiation was 64.6
Patients with methotrexate use (38.8%) on bDMARDS initiation date were similar
to patients without methotrexate use in age, sex, and race (Table 1). Patients
with methotrexate use on bDMARDS initiation date were more likely to initiate
acid use, and be bDMARD naïve. Patients with methotrexate use on bDMARDS
We identified 537 composite CVD events over 44,312-person years with time
events per 1,000-person years (Table 2). We identified 1,324 composite CVD
106
variable adjusted HR for time varying methotrexate exposure was 0.68 (95%CI:
Multi-variable adjusted HRs for each component of CVD were similar to that for
composite CVD (Table 2). The multi-variable adjusted HR for time varying
methotrexate exposure was 0.73 (95%CI: 0.63-0.86) for MI; 0.72 (95%CI: 0.61-
P-value was 0.0030 for test for interaction between time varying methotrexate
The IRs for composite CVD for bDMARDS with time varying methotrexate
exposure ranged from low of 8.6 (95%CI: 5.5-13.5) for tocilizumab initiators to
high of 13.6 (95%CI: 10.2-18.3) for rituximab initiators (Table 4, top panel, left).
The IRs for composite CVD for bDMARDS without time varying methotrexate
exposure ranged from low 14.6 (95% CI: 11.8-18.0) for tocilizumab initiators to
high 22.4 (95%CI: 20.1-25.0) for infliximab initiators (Table 4, top panel, middle).
Contrasts for comparing with and without time varying methotrexate exposure
ranged from low of 0.58 (95%CI: 0.35-0.96) for certolizumab initiators to high of
0.91 (95%CI: 0.68-1.20) for etanercept initiators. Contrast for comparing with and
without time varying methotrexate exposure was 0.73 (95%CI: 0.58-0.92) for
107
abatacept, 0.58 (95%CI: 0.35-0.96) for certolizumab and 0.63 (95%CI: 0.52-0.75)
Contrasts for each component of CVD were similar to that for composite CVD
(Table 3).
table 2).
Results from the MBDA lab tests linked to Medicare data are shown in
MBDA score were similar between bDMARDS initiators with and without
methotrexate was 0.73 when 1.5 was used as HR associated with unmeasured
confounder (Moderate or high MBDA score); the number of 1.5 was the point
DISCUSSION
reduction in the risk for CVD associated with time varying methotrexate use. The
magnitude in the reduction of the risk for CVD varied among bDMARDS
initiators. Methotrexate use associated with significant lower risk for CVD among
abatacept and infliximab initiators and numerically lower risk for CVD among
other bDMARDS.
risk for CVD among bDMARDS naïve RA patients [17-20]. In a prospective study
including 1,240 RA patients attended the Wichita Arthritis Center between Jan 1,
1981 and Dec 31, 1999, Choi et al. found the HR associated with methotrexate
use for CVD death was 0.3 (95%CI: 0.2-0.7) [17]. In a study including veterans
with RA, Prodanowich et al. found the risk ratio for CVD for methotrexate was
0.83 (95%CI: 0.71-0.96) [18]. Suissa et al. conducted a cohort study using the
PharMetric Patient-centric Outcome database and found the risk ratio for
myocardial infarction for methotrexate use was 0.81 (95%CI: 0.60-1.08)[20]. All
All these estimates are highly compatible with our aHRs associated with
(e.g. tocilizumab), combination therapy may further reduce the risk for CVD. An
to recruit 7,000 stable coronary artery disease patients with type 2 diabetes or
metabolic syndrome to assess the effect of low-dose methotrexate on the risk for
109
heart attack, stroke or death. The trial was designed to provide 90 percent power
treated with bDMARDS. Our study suggests that when tolerability is not a
Our study has several strengths. We included fatal CVD as a part of a CVD
composite outcome. Fatal CVD was identified through a validated claims based
algorithm[31], and not including fatal CVD events may underestimate the IR for
CVD and may result in a biased estimate. We used a validated algorithm [27] to
identify infusion drug (ABA, CER and TCZ) represented with non-specific
HCPCS codes 3490 and 3590 before a permanent HCPCS code were available,
failed to identify these infusion drugs will reduce sample size and more
importantly will misclassify some non-new user as new user and also misclassify
their initiation date. We used Medicare claims data for 100% RA patients, and
our large sample size made it possible to assess effect of methotrexate on CVD
risk by bDMARDS; to estimate IR and HRs with high precision. Recognizing the
distribution of MBDA score between bDMARDS with and without time varying
methotrexate. This addresses concern that RA disease activity, which has been
110
previously shown to be associated with CVD risk [34], might confound the
Our study has some limitations common to other studies using administrative
data. Claims data only has prescription filling information, and patients filling a
prescription not necessary take the medication, and thus we may misclassify
accuracy, and there is little reason to expect that performance of these would be
risk associated with time varying methotrexate use. The strengths of association
References
1. Helmick CG, Felson DT, Lawrence RC, Gabriel S, Hirsch R, Kwoh CK,
Liang MH, Kremers HM, Mayes MD, Merkel PA et al: Estimates of the
5. Riise T, Jacobsen BK, Gran JT, Haga HJ, Arnesen E: Total mortality is
7. Solomon DH, Karlson EW, Rimm EB, Cannuscio CC, Mandl LA, Manson
107(9):1303-1307.
112
93(2):198-200.
44(12):2746-2749.
10. Mok CC, Kwok CL, Ho LY, Chan PT, Yip SF: Life expectancy,
63(5):1182-1189.
12. Ridker PM, Cushman M, Stampfer MJ, Tracy RP, Hennekens CH:
979.
13. Ridker PM, Rifai N, Stampfer MJ, Hennekens CH: Plasma concentration
108(3):205-213.
15. Singh JA, Saag KG, Bridges SL, Jr., Akl EA, Bannuru RR, Sullivan MC,
17. Choi HK, Hernan MA, Seeger JD, Robins JM, Wolfe F: Methotrexate and
2002, 359(9313):1173-1177.
19. van Halm VP, Nurmohamed MT, Twisk JW, Dijkmans BA, Voskuyl AE:
20. Suissa S, Bernatsky S, Hudson M: Antirheumatic drug use and the risk of
22. Radovits BJ, Popa-Diaconu DA, Popa C, Eijsbouts A, Laan RF, van Riel
1276.
23. Westlake SL, Colebatch AN, Baird J, Kiely P, Quinn M, Choy E, Ostor AJ,
24. Micha R, Imamura F, Wyler von Ballmoos M, Solomon DH, Hernan MA,
25. Almalag HM, Mangoni AA, Crilly MA: Methotrexate and risk of
109(9):1383-1384.
27. Curtis JR, Xie F, Chen R, Chen L, Kilgore ML, Lewis JD, Yun H, Zhang J,
Guideline 2014.
30. Kumamaru H, Judd SE, Curtis JR, Ramachandran R, Hardy NC, Rhodes
JD, Safford MM, Kissela BM, Howard G, Jalbert JJ et al: Validity of claims-
7(4):611-619.
31. Fenglong Xie LDC, Jeffrey R. Curtis, Meredith L. Kilgore, Emily B. Levitan,
submitted.
32. Curtis JR, Xie F, Chen L, Saag KG, Yun H, Muntner P: Biomarker-related
diseases 2017.
303.
34. Dixon WG, Watson KD, Lunt M, Hyrich KL, British Society for
118
Table 3: Incidence rates and hazard ratios for composite CVD and its component associated with time varying
methotrexate, by biologic DMARDS.
With time varying methotrexate† Without time varying methotrexate†
‡ ‡ §
Biologic Event Person Incidence rate Adjusted HR Event Person Incidence rate Adjusted HR Contrast
DMARDS years (95% CI) (95% CI) years (95% CI) (95% CI)
Myocardial infarction or stroke Abatacept 102 8909 11.4 (9.4, 13.9) 0.82 (0.61, 1.10) 317 18237 17.4 (15.6, 19.4) 1.12 (0.88, 1.43) 0.73 (0.58, 0.92)
or fatal CVD Adalimumab 81 6130 13.2 (10.6, 16.4) 1.14 (0.83, 1.56) 161 9852 16.3 (14.0, 19.1) 1.28 (0.98, 1.67) 0.89 (0.68, 1.17)
Certolizumab 20 2152 9.3 (6.0, 14.4) 0.70 (0.43, 1.15) 70 3819 18.3 (14.5, 23.2) 1.21 (0.88, 1.67) 0.58 (0.35, 0.96)
Etanercept 74 6042 12.2 (9.8, 15.4) 1.00 (0.72, 1.37) 163 11189 14.6 (12.5, 17.0) 1.10 (0.54, 1.43) 0.91 (0.68, 1.20)
Golimumab 16 1503 10.6 (6.5, 17.4) 0.88 (0.51, 1.50) 54 2646 20.4 (15.6, 26.6) 1.52 (1.07, 2.14) 0.58 (0.33, 1.01)
Infliximab 180 14060 12.8 (11.1, 14.8) 0.93 (0.71, 1.22) 334 14891 22.4 (20.1, 25.0) 1.49 (1.17, 1.90) 0.63 (0.52, 0.75)
Rituximab 45 3302 13.6 (10.2, 18.3) 1.03 (0.71, 1.49) 140 7660 18.3 (15.5, 21.6) 1.15 (0.88, 1.50) 0.90 (0.64, 1.26)
Tocilizumab 19 2214 8.6 (5.5, 13.5) 0.66 (0.40, 1.09) 85 5834 14.6 (11.8, 18.0) reference 0.66 (0.40, 1.09)
Myocardial infarction Abatacept 52 8,944 5.8 (4.4, 7.6) 0.78 (0.52, 1.18) 166 18,327 9.1 (7.8, 10.6) 1.11 (0.80, 1.55) 0.70 (0.51, 0.97)
Adalimumab 43 6,147 7.0 (5.2, 9.4) 1.15 (0.75, 1.78) 79 9,900 8.0 (6.4, 10.0) 1.19 (0.82, 1.73) 0.97 (0.66, 1.41)
Certolizumab 13 2,157 6.0 (3.5, 10.4) 0.86 (0.46, 1.61) 33 3,833 8.6 (6.1, 12.1) 1.08 (0.69, 1.69) 0.80 (0.41, 1.53)
Etanercept 40 6,056 6.6 (4.9, 9.0) 1.02 (0.66, 1.58) 85 11,238 7.6 (6.1, 9.4) 1.08 (0.75, 1.55) 0.94 (0.64, 1.39)
Golimumab 9 1,506 6.0 (3.1, 11.5) 0.92 (0.45, 1.90) 19 2,659 7.1 (4.6, 11.2) 1.00 (0.59, 1.72) 0.92 (0.42, 2.04)
Infliximab 94 14,153 6.6 (5.4, 8.1) 0.90 (0.62, 1.31) 198 15,010 13.2 (11.5, 15.2) 1.66 (1.20, 2.30) 0.55 (0.42, 0.70)
Rituximab 27 3,314 8.2 (5.6, 11.9) 1.15 (0.71, 1.87) 69 7,704 9.0 (7.1, 11.3) 1.05 (0.72, 1.53) 1.09 (0.69, 1.72)
Tocilizumab 11 2,221 5.0 (2.7, 9.0) 0.72 (0.37, 1.40) 45 5,854 7.7 (5.7, 10.3) reference 0.72 (0.37, 1.40)
Stroke Abatacept 45 8,960 5.0 (3.8, 6.7) 0.73 (0.47, 1.14) 155 18,413 8.4 (7.2, 9.9) 1.13 (0.80, 1.60) 0.65 (0.46, 0.91)
Adalimumab 31 6,155 5.0 (3.5, 7.2) 0.90 (0.55, 1.46) 71 9,919 7.2 (5.7, 9.0) 1.18 (0.80, 1.75) 0.76 (0.50, 1.17)
Certolizumab 8 2,163 3.7 (1.9, 7.4) 0.58 (0.27, 1.26) 31 3,833 8.1 (5.7, 11.5) 1.14 (0.71, 1.82) 0.51 (0.23, 1.13)
Etanercept 31 6,065 5.1 (3.6, 7.3) 0.86 (0.53, 1.40) 68 11,262 6.0 (4.8, 7.7) 0.96 (0.65, 1.43) 0.90 (0.58, 1.38)
Golimumab 7 1,505 4.7 (2.2, 9.8) 0.82 (0.37, 1.86) 33 2,658 12.4 (8.8, 17.5) 2.00 (1.26, 3.18) 0.41 (0.18, 0.94)
Infliximab 89 14,164 6.3 (5.1, 7.7) 0.94 (0.64, 1.39) 136 15,069 9.0 (7.6, 10.7) 1.24 (0.87, 1.78) 0.76 (0.58, 1.00)
Rituximab 18 3,321 5.4 (3.4, 8.6) 0.86 (0.49, 1.51) 65 7,715 8.4 (6.6, 10.7) 1.13 (0.76, 1.69) 0.76 (0.44, 1.29)
Tocilizumab 10 2,221 4.5 (2.4, 8.4) 0.73 (0.36, 1.46) 40 5,864 6.8 (5.0, 9.3) reference 0.73 (0.36, 1.46)
Myocardial infarction or stroke Abatacept 95 8,909 10.7 (8.7, 13.0) 0.79 (0.58, 1.07) 304 18,237 16.7 (14.9, 18.7) 1.12 (0.88, 1.44) 0.70 (0.55, 0.89)
Adalimumab 74 6,130 12.1 (9.6, 15.2) 1.09 (0.79, 1.51) 148 9,852 15.0 (12.8, 17.7) 1.23 (0.94, 1.62) 0.88 (0.66, 1.17)
Certolizumab 20 2,152 9.3 (6.0, 14.4) 0.73 (0.44, 1.20) 62 3,819 16.2 (12.7, 20.8) 1.13 (0.81, 1.57) 0.65 (0.39, 1.08)
Etanercept 70 6,042 11.6 (9.2, 14.6) 0.98 (0.71, 1.37) 145 11,189 13.0 (11.0, 15.3) 1.02 (0.78, 1.35) 0.96 (0.72, 1.28)
Golimumab 15 1,503 10.0 (6.0, 16.6) 0.86 (0.49, 1.51) 52 2,646 19.7 (15.0, 25.8) 1.54 (1.08, 2.18) 0.56 (0.31, 1.00)
Infliximab 175 14,060 12.5 (10.7, 14.4) 0.94 (0.71, 1.24) 323 14,891 21.7 (19.5, 24.2) 1.50 (1.17, 1.92) 0.63 (0.52, 0.76)
Rituximab 45 3,302 13.6 (10.2, 18.3) 1.07 (0.74, 1.55) 127 7,660 16.6 (13.9, 19.7) 1.09 (0.83, 1.44) 0.98 (0.69, 1.39)
Tocilizumab 19 2,214 8.6 (5.5, 13.5) 0.69 (0.42, 1.14) 81 5,834 13.9 (11.2, 17.3) reference 0.69 (0.42, 1.14)
CVD: Cardiovascular disease; DMARDS: Disease-modifying antirhematic drugs; HR: Hazard ratio; IR: Incidence rate.
†
Time varying methotrexate was defined as methotrexate day of supply with no extension.
‡
Comparison to infliximab without MTX, adjusted for age, sex, race, baseline MTX, heart failure, atrial fibrillation, abdominal aortic aneurism, peripheral arterial disease, diabetes, hyperlipidemia, hypertension,
obesity, chronic kidney disease, chronic obstructive pulmonary disease, fibromyalgia, any hospitalized infection, any hospitalization, number of physician visit, time varying folic acid, nonsteroidal anti-
inflammatory drug, statin potency, other lipid lowering drug use in baseline, number of biologic used prior initiation, oral steroid dose in six month before initiation, smoking, State Buy-In (poor), reasons other
than age for eligible for Medicare.
§
In one model for each outcome, compares risk for time varying MTX versus nonuse within each bDMARD exposure row.
119
120
N subject=220,135/146,317
Excluded due to without 365 days observable N episode=317,389/199,364
N subject=113,957/95,937
N episode=158,262/135,064
N subject=3/2,579
Excluded due to age younger than 18 N episode=3/3,512
N subject=113,954/93,358
N episode=158,259/131,552
N subject=100,153/70,933
N episode=140,498/100,426
N subject=86,996/63,791
N episode=120,527/89,412
N subject=75,758/60,083
N episode=105,160/84,247
N subject=64,218/57,208
N episode=88,255/79,833
Supplementary table 1: Incidence rate and hazard ratio for CVD associated with methotrexate use, by biologic DMARDS
for patients exposed to methotrexate before initiation of biologic DMARDS.
‡ ‡
With time varying methotrexate Without time varying methotrexate
§ §
Biologic Event Person Incidence rate Adjusted HR Event Person Incidence rate Adjusted HR Contrast
DMARDS years (95% CI) (95% CI) years (95% CI) (95% CI)
Abatacept 99 8,689 11.4 (9.4, 13.9) 0.80 (0.57, 1.13) 186 11,309 16.4 (14.2, 19.0) 1.12 (0.83, 1.53) 0.71 (0.56, 0.91)
Adalimumab 78 5,939 13.1 (10.5, 16.4) 1.12 (0.78, 1.61) 92 6,065 15.2 (12.4, 18.6) 1.27 (0.90, 1.79) 0.88 (0.65, 1.20)
Myocardial infarction or Certolizumab 18 2,100 8.6 (5.4, 13.6) 0.64 (0.37, 1.10) 48 2,399 20.0 (15.1, 26.6) 1.38 (0.93, 2.03) 0.47 (0.27, 0.81)
stroke or fatal CVD Etanercept 68 5,793 11.7 (9.3, 14.9) 0.95 (0.66, 1.37) 98 6,939 14.1 (11.6, 17.2) 1.11 (0.79, 1.56) 0.85 (0.62, 1.16)
Golimumab 16 1,458 11.0 (6.7, 17.9) 0.90 (0.51, 1.58) 30 1,587 18.9 (13.2, 27.0) 1.44 (0.92, 2.26) 0.62 (0.34, 1.14)
Infliximab 175 13,660 12.8 (11.0, 14.9) 0.93 (0.68, 1.28) 217 10,464 20.7 (18.2, 23.7) 1.47 (1.08, 1.99) 0.64 (0.52, 0.78)
Rituximab 45 3,178 14.2 (10.6, 19.0) 1.05 (0.70, 1.57) 66 4,497 14.7 (11.5, 18.7) 0.99 (0.69, 1.43) 1.06 (0.72, 1.55)
Tocilizumab 19 2,164 8.8 (5.6, 13.8) 0.66 (0.39, 1.13) 54 3,873 13.9 (10.7, 18.2) reference 0.66 (0.39, 1.13)
CVD: Cardiovascular disease; CI: Confidence interval; MTX: Methotrexate; HR: Hazard ratio.
†
Concomitant methotrexate was defined as methotrexate prescription within 120 days after initiation of biologic disease-modifying anti-rheumatic drugs (DMARDS).
‡
Time varying methotrexate was defined as days of supply with no extension.
§
Comparison to infliximab without MTX, adjusted for age, sex, race, baseline MTX, heart failure, atrial fibrillation, abdominal aortic aneurism, peripheral arterial disease, diabetes, hyperlipidemia, hypertension,
obesity, chronic kidney disease, chronic obstructive pulmonary disease, fibromyalgia, any hospitalized infection, any hospitalization, number of physician visit, folic acid, nonsteroidal anti-inflammatory drug,
statin potency, other lipid lowering drug use in baseline, number of biologic used prior initiation, oral steroid dose in six month before initiation, smoking, State Buy-In (poor), reasons other than age for eligible
for Medicare.
¶
In one model, compares risk for MTX versus non use within each bDMARD exposure row.
121
Supplementary table 2: Incidence rate and hazard ratio for CVD associated with time varying methotrexate use with 90
days extension, by biologic DMARDS.
With time varying methotrexate Without time varying methotrexate
† † ‡
Biologic Event Person Incidence rate Adjusted HR Event Person Incidence rate Adjusted HR Contrast
DMARDS years (95% CI) (95% CI) years (95% CI) (95% CI)
Myocardial infarction or stroke Abatacept 152 12232 12.4 (10.6, 14.6) 0.91 (0.68, 1.22) 267 14914 17.9 (15.9, 20.2) 1.15 (0.88, 1.49) 0.79 (0.64, 0.99)
or fatal CVD Adalimumab 117 8490 13.8 (11.5, 16.5) 1.22 (0.90, 1.67) 125 7492 16.7 (14.0, 19.9) 1.27 (0.95, 1.70) 0.96 (0.74, 1.26)
Certolizumab 30 2887 10.4 (7.3, 14.9) 0.80 (0.52, 1.24) 60 3084 19.5 (15.1, 25.1) 1.27 (0.90, 1.79) 0.63 (0.40, 0.99)
Etanercept 102 8390 12.2 (10.0, 14.8) 1.01 (0.74, 1.39) 135 8841 15.3 (12.9, 18.1) 1.14 (0.85, 1.52) 0.89 (0.68, 1.16)
Golimumab 22 2042 10.8 (7.1, 16.4) 0.91 (0.56, 1.49) 48 2107 22.8 (17.2, 30.2) 1.66 (1.15, 2.39) 0.55 (0.33, 0.92)
Infliximab 273 18649 14.6 (13.0, 16.5) 1.09 (0.83, 1.44) 241 10301 23.4 (20.6, 26.5) 1.52 (1.16, 1.98) 0.72 (0.60, 0.87)
Rituximab 64 4692 13.6 (10.7, 17.4) 1.06 (0.75, 1.50) 121 6269 19.3 (16.2, 23.1) 1.18 (0.88, 1.58) 0.90 (0.66, 1.23)
Tocilizumab 32 3170 10.1 (7.1, 14.3) 0.80 (0.52, 1.23) 72 4878 14.8 (11.7, 18.6) reference 0.80 (0.52, 1.23)
CVD: Cardiovascular disease; CI: Confidence interval; HR: Hazard ratio.
†
Comparison to infliximab without MTX, adjusted for age, sex, race, baseline MTX, heart failure, atrial fibrillation, abdominal aortic aneurism, peripheral arterial disease, diabetes, hyperlipidemia, hypertension,
obesity, chronic kidney disease, chronic obstructive pulmonary disease, fibromyalgia, any hospitalized infection, any hospitalization, number of physician visit, time varying folic acid, nonsteroidal anti-
inflammatory drug, statin potency, other lipid lowering drug use in baseline, number of biologic used prior initiation, oral steroid dose in six month before initiation, smoking, State Buy-In (poor), reasons other
than age for eligible for Medicare.
‡
In one model, compares risk for MTX versus nonuse within each bDMARD exposure row.
122
123
SUMMARY
compared to non-RA patients. Any treatment or practice that has the potential to
increase the risk for CVD among RA patients will definitely raise safety concerns;
investigating.
patients treated with tocilizumab have been reported, raising important safety
issues. Even so, some important limitations were observed in the few studies that
evaluated the association between tocilizumab and CVD risk. 1) The clinical trial
another clinical trial due to excessive costs and the improbability of recruiting
observational studies failed to include fatal CVD as an outcome due to the lack of
the cause of death in administrative claims data. Failure to include fatal CVD as
since each TNFi may have drug specific risks for CVD. Disease activity is
with reduced risk for CVD among RA patients naïve to bDMARDS. No studies
have been conducted to assess the effect of methotrexate on the risk of CVD
CVD, fatal CHD, and fatal stroke events in Medicare claims data. The algorithms
can identify fatal CVD events with high sensitivity and specificity and have better
develop claims-based algorithms for other fatal events and fatal events in other
claims data.
randomized trial that tocilizumab was not associated with increased or decreased
CVD risk compared to etanercept. The hazard ratio for tocilizumab compared to
etanercept was 0.91 (95%CI: 0.66, 1.25). We also extended our findings from
high-risk patients with at least one CVD risk factor to “low CVD risk” RA patients.
Our results showed that tocilizumab was not associated with CVD risk compared
to abatacept and rituximab. Further, our results indicated that tocilizumab was
exposure.
use is associated with an overall 27% reduction of CVD risk. The strength of
that adding methotrexate to tocilizumab initiators may reduce the CVD risk
inflammation. Thus, methotrexate may reduce CVD risk among non-RA patients
GENERAL REFERENCES
1. Helmick CG, Felson DT, Lawrence RC, Gabriel S, Hirsch R, Kwoh CK,
Liang MH, Kremers HM, Mayes MD, Merkel PA et al: Estimates of the
5. Riise T, Jacobsen BK, Gran JT, Haga HJ, Arnesen E: Total mortality is
7. Solomon DH, Karlson EW, Rimm EB, Cannuscio CC, Mandl LA, Manson
107(9):1303-1307.
93(2):198-200.
44(12):2746-2749.
10. Mok CC, Kwok CL, Ho LY, Chan PT, Yip SF: Life expectancy,
63(5):1182-1189.
12. Mikuls TR, Sayles H, Yu F, Levan T, Gould KA, Thiele GM, Conn DL,
2010, 62(12):3560-3568.
129
14. del Rincon ID, Williams K, Stern MP, Freeman GL, Escalante A: High
2001, 44(12):2737-2745.
15. Ridker PM, Cushman M, Stampfer MJ, Tracy RP, Hennekens CH:
979.
16. Ridker PM, Rifai N, Stampfer MJ, Hennekens CH: Plasma concentration
108(3):205-213.
18. Kremer JM, Genant HK, Moreland LW, Russell AS, Emery P, Abud-
19. Moreland LW, Baumgartner SW, Schiff MH, Tindall EA, Fleischmann RM,
necrosis factor receptor (p75)-Fc fusion protein. The New England journal
354(9194):1932-1939.
21. Weinblatt ME, Keystone EC, Furst DE, Moreland LW, Weisman MH,
Birbara CA, Teoh LA, Fischkoff SA, Chartash EK: Adalimumab, a fully
45.
23. Smolen JS, Kay J, Doyle MK, Landewe R, Matteson EL, Wollenhaupt J,
Gaylis N, Murphy FT, Neal JS, Zhou Y et al: Golimumab in patients with
Agarwal S et al: The efficacy and safety of rituximab in patients with active
26. Choi HK, Hernan MA, Seeger JD, Robins JM, Wolfe F: Methotrexate and
2002, 359(9313):1173-1177.
28. van Halm VP, Nurmohamed MT, Twisk JW, Dijkmans BA, Voskuyl AE:
29. Suissa S, Bernatsky S, Hudson M: Antirheumatic drug use and the risk of
31. Radovits BJ, Popa-Diaconu DA, Popa C, Eijsbouts A, Laan RF, van Riel
1276.
32. Westlake SL, Colebatch AN, Baird J, Kiely P, Quinn M, Choy E, Ostor AJ,
33. Micha R, Imamura F, Wyler von Ballmoos M, Solomon DH, Hernan MA,
34. Almalag HM, Mangoni AA, Crilly MA: Methotrexate and risk of
109(9):1383-1384.
36. Westlake SL, Colebatch AN, Baird J, Curzen N, Kiely P, Quinn M, Choy E,
Ostor AJ, Edwards CJ: Tumour necrosis factor antagonists and the risk of
38. Maini RN, Taylor PC, Szechinski J, Pavelka K, Broll J, Balint G, Emery P,
2829.
31(4):451-456.
40. Strang AC, Bisoendial RJ, Kootte RS, Schulte DM, Dallinga-Thie GM,
Levels JH, Kok M, Vos K, Tas SW, Tietge UJ et al: Pro-atherogenic lipid
75(10):1806-1812.
43. Kim SC, Solomon DH, Rogers JR, Gale S, Klearman M, Sarsour K,
44. Zhang J, Xie F, Yun H, Chen L, Muntner P, Levitan EB, Safford MM, Kent
45. Maradit-Kremers H, Crowson CS, Nicola PJ, Ballman KV, Roger VL,
46. Curtis JR, Xie F, Chen L, Saag KG, Yun H, Muntner P: Biomarker-related
diseases 2017.
50. Marks JL, Edwards CJ: Protective effect of methotrexate in patients with
1347.
136
APPENDIX A
IRB APPROVAL
137