Assessing The Comparative Effect of Tocilizumab

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
by
FENGLONG XIE
PAUL MUNTNER, COMMITTEE CHAIR

JEFFREY R. CURTIS
ELIZABETH DELZELL
NENGJUN YI
HUIFENG YUN
A DISSERTATION
Submitted to the graduate faculty of The University of Alabama at Birmingham,

in partial fulfillment of the requirements for the degree of
Doctor of Philosophy
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
Tocilizumab is a humanized monoclonal anti-body against the interleukin-6
receptor and is effective in the treatment of rheumatoid arthritis (RA) [1]. Multiple
studies have observed unfavorable changes in the lipid profile of tocilizumab-
treated RA patients. Few studies have compared the cardiovascular disease
(CVD) risk associated with tocilizumab to other biologic disease-modifying
antirheumatic drugs (bDMARDS). Due to limitations in existing studies, the real-
world association of tocilizumab with CVD risk remains uncertain.
We conducted a retrospective cohort study using 2006-2015 Medicare and
MarketScan claims data to assess the comparative effect of tocilizumab on CVD
risk. Medicare claims data provide a unique opportunity to estimate disease
burden and conduct comparative effectiveness studies with much larger sample
sizes than cohorts, which require primary data collection. However, Medicare
claims data lack information on cause of death. To address this limitation, we
developed claims-based algorithms for identifying fatal CVD in Medicare claims
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
fatal CVD events with high sensitivity and specificity.
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
with increased or decreased CVD risk compared to abatacept or rituximab. We
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
CVD risk associated with infliximab.
We also conducted a retrospective cohort study using Medicare claims data to
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
methotrexate compared to tocilizumab only was 0.66 (95%CI: 0.40-1.09).
KEYWORDS: tocilizumab, biologic DMARDS, CVD risk, Medicare claims,
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
me while working to finish my Ph.D. Specifically, I want to thank my mentor, Dr.
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.
Additionally, I want to thank the other members of my graduate committee, Dr.
Nengjun Yi and Dr. Huifeng Yun, for their excellent suggestions and instructive
criticism concerning my research project.
Furthermore, I want to express thanks to my colleagues, Dr. Lang Chen, Shuo
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
always with me.
vi
TABLE OF CONTENTS
ABSTRACT ................................................................................................................................. iii
DEDICATION............................................................................................................................... v
ACKNOWLEDGEMENTS ...................................................................................................... vi
LIST OF TABLES ..................................................................................................................... ix
LIST OF FIGURES .................................................................................................................. xii
INTRODUCTION ....................................................................................................................... 1
Rheumatoid arthritis ..................................................................................................... 1
Rheumatoid arthritis and the risk for cardiovascular disease ......................... 1
Treatment for rheumatoid arthritis ........................................................................... 2
Treatment for rheumatoid arthritis and the risk for cardiovascular disease 3
Tocilizumab and the risk for cardiovascular disease ......................................... 3
Clinical trial and observational study ...................................................................... 5
Significance .................................................................................................................... 5
DEVELOPMENT OF ALGORITHMS FOR IDENTIFYING FATAL

CARDIOVASCULAR DISEASE IN MEDICARE CLAIMS ............................................. 7
TOCILIZUMAB AND THE RISK FOR CARDIOVASCULAR DISEASE: A DIRECT

COMPARISON AMONG BIOLOGIC DISEASE-MODIFYING ANTIRHEUMATIC
DRUGS FOR RHEUMATOID ARTHRITIS PATIENTS IN REAL WORLD
SETTING .................................................................................................................................... 52
METHOTREXATE USE AND THE RISK FOR CARDIOVASCULAR DISEASE

AMONG RHEUMATOID ARTHRITIS PATIENTS INITIATING BIOLOGIC
DISEASE-MODIFYING ANTIRHEMATIC DRUGS ....................................................... 95
SUMMARY .............................................................................................................................. 124
vii
GENERAL LIST OF REFERENCES ............................................................................... 127
APPENDIX A: IRB approval ................................................... ………………………………………..136
viii
LIST OF TABLES
Table Page

CARDIOVASCULAR DISEASE IN MEDICARE CLAIMS
1 Definitions of fatal events in the REGARDS study…………………………………………31
2 Distribution of pre-specified candidate predictors for participants with and

without an adjudicated fatal CVD, CHD and stroke……………………………………....32
3 Performance of the direct and indirect approaches for discriminating fatal

cardiovascular disease and coronary heart disease…………………..…………………34
4 Performance of algorithms for discriminating cause of death under

different cut-points of predicted probability compared with using hospital
discharge diagnosis codes……………………………………………………………………………………35
S1 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………………………………………………………………….………………………………………………………41
S2 Beta-coefficients from logistic regression models for predictor variables for

the outcomes of fatal cardiovascular disease, coronary heart disease,
stroke, myocardial infarction and coronary heart disease death…………………42
S3 Net reclassification index for the claims-based algorithm developed in the

current study compared to hospital discharge diagnosis codes for fatal
cardiovascular disease…………………………………………………………………..……………………..44

coronary heart disease………………………………………………………………………………………….45

stroke..............................................................................................................................46
ix
S6 Example of the calculation of the predicted probability for having a fatal
stroke……………………………………………………………..………………………………….…………………...…47

coronary heart disease event…………………………………………………………………..............48

cardiovascular disease event………………………………………………………….………..………...50

SETTING
1a Distribution of baseline characteristics by biologic DMARDS for Medicare
…………………………………………………………………………………………………………………..………………..79
1b Distribution of characteristics by biologic DMARDS for MarketScan….……80
2 Incidence rates for composite cardiovascular event and its components

(Acute myocardial infarction, stroke or fatal CVD) by biologic DMARDS by
data source……………………………………………………………………………………………………………….81
3 Incidence rates for composite cardiovascular event by History of other

cardiovascular disease‡ and biologic DMARDS by data source…………………82
S1 Cardiovascular disease related diagnosis, procedure and Current

Procedure Terminology code used for defining outcome, covariate or for
exclusion……………………………………………………………………………………………………………….....87
S2 Incidence rates for composite cardiovascular event† by biologic DMARDS

for biologic DMARDS exposed by data source……………………………………………….88
S3 Incidence rates for composite cardiovascular event by definition for fatal

cardiovascular disease and biologic DMARDS by data source………………….89
S4 Distribution of baseline RA disease activity as assessed by the multi-

biomarker disease activity (MBDA) prior to initiation of various biologic
DMARDs ........................................................................................................................90
x
DISEASE-MODIFYING ANTIRHEMATIC DRUGS
1 Distribution of characteristics by methotrexate use on initiation date........117
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
S3 Distribution of baseline RA disease activity as assessed by the multi-

biomarker disease activity (MBDA), by methotrexate use on initiation
date .............................................................................................................................. ..123
xi
LIST OF FIGURES
Figure Page

1 Direct and indirect approaches for developing the claims-based

algorithms……………………………………………………………………………………………………………….…37
2 Flow chart for cohort selection………………………………………………….…………..…………….38
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

SETTING
1 Flow chart for cohort selection………………………………….…………………………………..…….83
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
S3 Forest plot for adjusted hazard ratio by definition of fatal cardiovascular

disease and data source……………………………………………………....93
xii
1 Flow chart for cohort selection…………………………………………………………….…….......120
xiii
1
INTRODUCTION
Rheumatoid arthritis (RA)
RA is a chronic inflammatory polyarthritis affecting about 0.5-2% of the
population [2-4]. Without effective treatment, RA often leads to joint deformity
and loss of function, disability, poor quality of life, and shortened life expectancy
[5-11]. Prevalence of RA in women is twice that in men. The pathogenesis of RA
is believed to be both genetic and environmental. More than 100 genes, such as
HLA-DRB1, PTPN22, TNFAIP3 [12], are found to be associated with disease
initiation and progression. Cigarette smoking is a major environmental factor that
is associated with higher rheumatoid factor titer and later disease severity [13].
Rheumatoid arthritis and the risk for cardiovascular disease (CVD)
CVD-related deaths account for 50% of excess premature mortality in RA
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
excessive risks for CVD[15]. The inflammatory milieu is characterized by
elevated production of a variety of cytokines and inflammatory markers including
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
RA and have been shown to promote atherosclerosis [16-18].

2
Treatment for rheumatoid arthritis
Although the FDA has approved more than 50 drugs to treat RA, RA is not
curable. Generally, there are two categories of treatment: symptomatic therapy
that aims to reduce patients’ pain, and disease-modifying antirheumatic drugs
(DMARDS) that target disease progression management.
DMARDS fall into three categories based on whether they target specific steps in
the inflammatory process, and whether they are biologic or small molecule.
Traditional DMARDS, including methotrexate, hydroxychloroquine, sulfasalazine,
leflunomide, cyclophosphamide, and azathioprine, suppress the whole immune
system without a specific target and slow disease activity by modifying the
course of the disease. In contrast, biologic DMARDS are a group of antibodies
targeting specific cytokine, cytokine receptor, T cell, or B cell. Abatacept
interferes with T cell costimulation [19]; tumor necrosis factor inhibitors (TNFi),
including adalimumab, certolizumab, etanercept, golimumab and infliximab,
control inflammation by binding to TNF [20-24]; tocilizumab reduces inflammation
by binding to interleukin 6 (IL-6) receptors; and rituximab controls the immune
response by binding to and depleting CD20 B cells [25]. RA patients who do not
have an adequate response to traditional DMARDS (mainly methotrexate),
typically add (combination therapy) or switch to biologic DMARDS
(monotherapy). While some therapies appear to have appreciable efficacy in the
absence of concomitant methotrexate, other therapies are clearly more effective
when combined with methotrexate [26].

3
Tofacitinib is another sub-category of DMARDS and is known as the “Janus
kinase inhibitor”. Tofacitinib controls inflammation by blocking the Janus kinase
pathways. Unlike biologic DMARDS, tofacitinib is small molecule and can be
taken orally.
Treatment for rheumatoid arthritis and the risk for cardiovascular disease
Several studies and meta-analyses have demonstrated that methotrexate is
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
concluded mechanistically that this effect is mediated primarily by controlling
systemic inflammation. However, in a setting of other therapies that also reduce
systemic inflammation (e.g. biologics), the incremental benefit of concomitant
methotrexate on CVD risk is not clear.
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].
Tocilizumab and the risk for cardiovascular disease
Tocilizumab (TCZ) is a humanized monoclonal antibody against the IL-6
receptor[38]; it is effective in the treatment for RA patients with or without
methotrexate [1]. Multiple studies have observed seemingly unfavorable changes
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
CVD in RA patients. A recently completed randomized clinical trial of 3,080

4
patients (ENTRACTE, ClinicalTrials.gov: NCT01331837) showed that there is no
increased risk (hazard ratio of 1.05) of CVD associated with tocilizumab
compared to etanercept among RA patients with risk factors for CVD [42]. Two
observational studies likewise found that there are no increased risks of
myocardial infarction (MI), stroke, or other causes of mortality associated with
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
other therapies (ENTRACTE, ClinicalTrials.gov: NCT01331837); the two
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
tocilizumab-exposed groups (83 major adverse cardiovascular events in the trial,
and 32 and 22 composite CVD events in the two observational studies,
respectively). Moreover, failure to include fatal CVD as an outcome may under-
estimate the incidence of true CVD-related events. This may be even more
problematic in studies involving RA patients since RA patients experience a
higher proportion of sudden coronary heart disease (CHD) death [45]. Due to
sample size limitations, one of the observational studies compared tocilizumab to
a pooled TNFi exposure group without evaluating individual TNFi therapies,
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
potential confounding of RA severity. Previous research has shown that RA
disease activity is associated with higher CVD risk [46].
Clinical trial and observational study
While a clinical trial is the best way to assess the causality of exposure and
outcome, it is expensive and sometimes impossible for rare events.
Administrative claims data make it possible to conduct observational studies
economically and timely. 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 [47]. In 2015, 55 million US adults received health insurance
through the Medicare program [48]. Medicare claims data provide the opportunity
to estimate disease burden, conduct comparative effectiveness research, and
safety studies with much larger sample sizes than cohort studies that require
primary data collection. A limitation of Medicare claims data is the lack of
information on cause of death. While assessing the association of exposures with
non-fatal outcomes and all-cause mortality can provide useful information, cause
of death is important to address many study questions.
Significance
This dissertation has both methodological significance and clinical 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
among Medicare beneficiaries. Adding CVD-related fatal events, identified with
these algorithms, to non-fatal CVD events identified through claims data should
provide a more accurate event rate and less biased exposure-outcome
associations.
Through specific aim 2, we confirmed results from a large randomized trial in a
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.
Through specific aim 3, we concluded that concomitant methotrexate use is
associated with an overall 27% reduction in CVD risk. The strength of association
varies among concomitant biologic DMARDS. This finding has two clinical
implications: 1) Although it may not increase effectiveness, adding methotrexate
to tocilizumab initiators may help to reduce the CVD risk; 2) Because RA patients
initiating biologic DMARDS usually have inadequate response to methotrexate,
and biologic DMARDS like tocilizumab are effective with and without
methotrexate, suggests that methotrexate may have a mechanism for reducing
CVD risk in addition to controlling inflammation, implying that methotrexate may
reduce CVD risk among non-RA patients with high CVD risk.
7

by
FENGLONG XIE, ISANDRO D. COLANTONIO, JEFFREY R. CURTIS,

MEREDITH L. KILGORE, EMILY B. LEVITAN, KERI L. MONDA,
MONIKA M. SAFFORD, BEN TAYLOR, MARK WOODWARD,
PAUL MUNTNER
Pharmacoepidemiology and Drug Safety
Copyright
2018
By
Wiley
Used by permission
Format adapted [and errata corrected] for dissertation

8
Development of algorithms for identifying fatal cardiovascular disease in

Medicare claims
Running title: algorithms for fatal CVD
Fenglong Xie1,2, Lisandro D. Colantonio1, Jeffrey R. Curtis1,2, Meredith L.

Kilgore3, Emily B. Levitan1, Keri L. Monda4, Monika M. Safford5, Ben Taylor4,
Mark Woodward6,7,8, Paul Muntner1
1 Department of Epidemiology, School of Public Health, University of Alabama at

Birmingham, Birmingham, AL, USA
2 Department of Medicine, Division of Clinical Immunology and Rheumatology,
School of Medicine, University of Alabama at Birmingham, Birmingham, AL, USA
3 Department of Health Care Organization and Policy, School of Public Health,
University of Alabama at Birmingham, Birmingham, AL, USA
4 The Center for Observational Research, Amgen Inc. USA.
5 Department of Medicine, Division of General Internal Medicine, Weill Cornell
Medicine, New York, NY, USA
6 The George Institute for Global Health, University of Oxford, UK
7 The George Institute for Global Health, University of New South Wales,
Sydney, Australia
8 Department of Epidemiology, Johns Hopkins University, Baltimore, MD, USA
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
Keywords: algorithm, fatal cardiovascular disease, Medicare claims
Word count
Abstract: 249
Manuscript: 2,960
Tables: 4
Figures: 4
9
Abstract
Background Cause of death is often not available in administrative claims data.
Objective To develop claims-based algorithms to identify deaths due to fatal
cardiovascular disease (CVD; i.e., fatal coronary heart disease [CHD] or stroke),
CHD, and stroke.
Methods Reasons for Geographic and Racial Differences in Stroke (REGARDS)
study data were linked with Medicare claims to develop the algorithms. Events
adjudicated by REGARDS study investigators were used as the gold standard.
Stepwise selection was used to choose predictors from Medicare data for
inclusion in the algorithms. C-index, sensitivity, specificity, positive predictive
value (PPV) and negative predictive value (NPV) were used to assess algorithm
performance. Net reclassification index (NRI) was used to compare the
algorithms to an approach of classifying all deaths within 28 days following
hospitalization for myocardial infarction and stroke to be fatal CVD.
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
closest observed-to-predicted number of fatal CVD events, the sensitivity was
0.64, specificity 0.90, PPV 0.65 and NPV 0.90. The algorithms resulted in
positive NRIs compared with using deaths within 28 days following
hospitalization for myocardial infarction and stroke. Claims-based algorithms for
discriminating fatal CHD and fatal stroke performed similarly to fatal CVD.
10
Conclusion The claims-based algorithms developed to discriminate fatal CVD
events from other causes of death performed better than the method of using
hospital discharge diagnosis codes.

11
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
2015, 55 million US adults received health insurance through the Medicare
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
limitation of Medicare claims data is the lack of information on cause of death.
While assessing the association of exposures with non-fatal outcomes and all-
cause mortality can provide useful information, cause of death is relevant for
addressing many study questions.
Cardiovascular disease (CVD), including coronary heart disease (CHD) and
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
discharge diagnoses and discharge disposition (i.e., deceased or alive), an
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
Medicare beneficiaries can result in the inability to accurately estimate CVD
event rates, as many events are fatal [9], and in biased exposure-outcome
associations due to censoring some CVD-related deaths as non-events. The goal
of the current analysis was to develop claims-based algorithms to discriminate

12
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
(REGARDS) study, a US population-based cohort that enrolled 30,239 non-
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
University of Alabama at Birmingham and all participants provided written
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. We required Medicare fee-for-service coverage for 182 days prior to

13
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
for Medicare due to disability or end-stage renal disease. Finally, we excluded
REGARDS study participants with a hospital discharge diagnosis code for
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
REGARDS study investigators. These participants were excluded as it could not
be determined whether or not their deaths were fatal CVD events.
CVD, CHD and stroke outcomes in REGARDS
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
adjudicated by two physicians independently [9]. Table 1 provides the definitions
of fatal CVD, fatal CHD and fatal stroke events in the REGARDS study.
Hospital discharge diagnosis code method
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
28 days prior to a participant’s death date (hospital discharge diagnosis method).

14
Using this approach, we defined fatal CHD as a death preceded by an inpatient
claim with a hospital discharge diagnosis code for MI (Supplemental Table 1).
Fatal hemorrhagic or ischemic stroke was defined by a death preceded by an
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
definition of fatal MI provided by the “Case definitions for acute CHD in
epidemiology and clinical research studies” [12].
Statistical methods
Candidate predictors
All potential predictors considered in the development of algorithms were defined
using Medicare data. Candidate predictors, pre-specified by the authors, included
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
candidate predictors using variables defined by the Agency for Healthcare
Research and Quality Clinical Classifications Software (AHRQ CCS) [13]. We
calculated summary statistics for the pre-specified potential predictors among
deceased REGARDS study participants included in the analysis.

15
Approaches for developing algorithms for composite outcomes
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
outcomes: a direct approach and two indirect approaches (Figure 1).
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-
specified candidate predictors were included in a single logistic regression model
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
using this approach.

16
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
approach. Participants with a predicted probability higher than a pre-specified
cut-point for fatal MI, CHD death or fatal stroke were defined as having a high
predicted probability for fatal CVD. Participants with a predicted probability
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-
specified cut-points were selected is provided in the Comparison of the Direct
and Indirect Approaches section below.
Indirect approach 2
The second indirect approach modeled the outcomes of fatal MI, CHD death and
fatal stroke, separately, as described in the indirect approach 1 above. The
predicted probabilities for each outcome from the three separate models were
used as independent variables in a logistic regression model with fatal CVD as
the outcome. In a separate analysis, the predicted probabilities of fatal MI and
CHD death were used to model fatal CHD as the outcome.
Comparison of the Direct and Indirect approaches
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
Participants with a predicted probability above this cut-point were categorized as
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
composite outcomes, we calculated sensitivity, specificity, positive predictive
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.
Evaluation of final algorithms
We assessed the performance of the final algorithms using sensitivity, specificity,
NPV and PPV and plotted the receiver operating characteristic curve. We
assessed the improvement in discrimination for the final algorithms compared
with hospital discharge diagnosis codes using the net reclassification index (NRI)
[16, 17]. Choosing the cut-point of predicted probability that provides the closest
number of predicted and observed events to define a high probability of an
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
SAS 9.4 (SAS Institute Inc., Cary, NC).

18
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
in current analyses (Figure 2). Among participants included in current analyses,
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.
Comparison of the direct and indirect approaches for algorithm
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
diagnosis codes for MI in any position within 28 days of death date
(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
to model fatal CVD and fatal CHD.
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
defining high probability using cut-points resulting in the closest percentage of
predicted and observed events, maximum specificity with sensitivity ≥0.80 or the
maximum sum of sensitivity and specificity. When cut-points of predicted
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
CHD, the algorithms resulted in a positive NRI compared to hospital- discharge
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
(Table 4 and Supplemental Table 5). When choosing a cut-point to achieve

20
maximum NPV while maintaining a PPV ≥0.80, the algorithm for fatal stroke
resulted in a positive, non-significant NRI.
Discussion
We developed claims-based algorithms for discriminating fatal CVD, fatal CHD,
and fatal stroke, separately, versus other causes of death. Each algorithm
showed good discrimination. The algorithms had a positive NRI compared with
using hospital discharge diagnoses. We provided cut-points for identifying a high
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
agreement on cause of death between death certificates and clinician
adjudicators is moderate (kappa statistic=0.54) [19, 20]. Studies assessing CVD
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
study. The claims-based algorithms that we developed had higher sensitivity
compared with this approach, while maintaining high specificity, and improving
discrimination as indicated by a positive NRI. To minimize the misclassification of
CVD events, we recommend using the algorithms developed herein rather than
hospital discharge diagnoses.
Those interested in using the algorithms that we developed may have different
objectives. Therefore, we evaluated thresholds with high calibration, sensitivity,
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
as sensitivity analyses to assess the robustness of findings.
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
investigators may be interested in calculating CVD events rates in claims data
using the outcomes from the Pooled Cohort risk equations, this has not been
22
possible previously. A high-predicted probability for a fatal CVD event based on
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.
To avoid data inconsistencies, we excluded 130 REGARDS participants with
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
participants, there was no death date recorded in the Medicare beneficiary
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
have had an effect on our algorithm.
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
REGARDS study were adjudicated by two trained investigators with an
adjudication committee resolving any disagreement. The REGARDS study
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
be evaluated in other race/ethnic groups. Also, we restricted the analysis 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
needs to be tested. These algorithms should also be validated in external
datasets.
In conclusion, we developed claims-based algorithms that can be used to identify
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
probability of being CVD-related on these algorithms to non-fatal CVD events
identified through claims data should provide a more accurate event rate and
less biased exposure-outcome associations.

24
Acknowledgments
The work for this manuscript was funded by an industry /academic collaboration
between Amgen Inc. and University of Alabama at Birmingham. REGARDS
project is supported by a cooperative agreement (U01 NS041588) from the
National Institute of Neurological Disorders and Stroke, National Institutes of
Health, Department of Health and Human Services. Additional funding was
provided by grants from the Agency for Healthcare Research and Quality (R01-
HS-8517) and the National Institutes of Health (R01HL080477 and
K24HL111154-K24).
Conflict of interest
F. Xie and L. D. Colantonio report no disclosure. E. B. Levitan, M. Kilgore, M. M.
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
from Novartis. P. Muntner receives honoraria from Amgen. K. L. Monda and B.
Taylor are employees of Amgen.

25
References
1. Administration on Aging: Administration for Community Living. Peofile of
older Americans: 2013, Washington, DC: US Department of Health and
Human Services; 2014.
http://www.aoa.gov/Aging_Statistics/Profile/2013/index.aspx. (Accessed at
April 25, 2017).
2. Nonell L: Total Number of Medicare Beneficiaries. Timeframe: 2015 2017.
http://kff.org/medicare/state-indicator/total-medicare-beneficiaries/?
currentTimeframe=0&sortModel= %7B%22colId%22:%22Location%22,
%22sort%22:%22asc%22%7D. (Accessed at April 25, 2017).
3. Heron M: National Vital Statistics Report. 2015, 64(10).
http://www.cdc.gov/ nchs/data_access/Vitalstatsonline.htm. (Accessed at
April 25, 2017).
4. Serban MC, Colantonio LD, Manthripragada AD, Monda KL, Bittner VA,
Banach M, Chen L, Huang L, Dent R, Kent ST et al: Statin Intolerance and
Risk of Coronary Heart Events and All-Cause Mortality Following
Myocardial Infarction. Journal of the American College of Cardiology 2017,
69(11):1386-1395.
5. Zhang J, Xie F, Yun H, Chen L, Muntner P, Levitan EB, Safford MM, Kent
ST, Osterman MT, Lewis JD et al: Comparative effects of biologics on
cardiovascular risk among older patients with rheumatoid arthritis. Annals
of the Rheumatic Diseases 2016, 75(10): 1813-18.
6. Spatz ES, Beckman AL, Wang Y, Desai NR, Krumholz HM: Geographic
Variation in Trends and Disparities in Acute Myocardial Infarction

26
Hospitalization and Mortality by Income Levels, 1999-2013. JAMA
Cardiology 2016, 1(3):255-265.
7. Henry J. Kaiser Family Foundation: Distribution of Medicare Beneficiaries
by Eligibility Category. State Health Facts 2013.
8. Roth DL, Skarupski KA, Crews DC, Howard VJ, Locher JL: Distinct age
and self-rated health crossover mortality effects for African Americans:
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
JH, Shikany JM, Prineas RJ, Samdarshi T, Bittner VA et al: Association of
race and sex with risk of incident acute coronary heart disease events.
JAMA : the Journal of the American Medical Association 2012,
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
differences in stroke study: objectives and design. Neuroepidemiology
2005, 25(3):135-143.
11. Xie F, Colantonio LD, Curtis JR, Safford MM, Levitan EB, Howard G,
Muntner P: Linkage of a Population-Based Cohort With Primary Data
Collection to Medicare Claims: The Reasons for Geographic and Racial
Differences in Stroke Study. American Journal of Epidemiology 2016,
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
acute coronary heart disease in epidemiology and clinical research
studies: a statement from the AHA Council on Epidemiology and
Prevention; AHA Statistics Committee; World Heart Federation Council on
Epidemiology and Prevention; the European Society of Cardiology
Working Group on Epidemiology and Prevention; Centers for Disease
Control and Prevention; and the National Heart, Lung, and Blood Institute.
Circulation 2003, 108(20):2543-2549.
13. Healthcare Cost and Utilization Project.Agency for Healthcare Research
and Quality. Rockville M: Clinical Classifications Software (CCS) for ICD-
9-CM. https://www.hcup-us.ahrq.gov/toolssoftware/ccs/ccs.jsp. (Accessed
at January 2017).
14. Stoltzfus JC: Logistic regression: a brief primer. Academic Emergency
Medicine : Official Journal of the Society for Academic Emergency
Medicine 2011, 18(10):1099-1104.
15. Efron B, Halloran E, Holmes S: Bootstrap confidence levels for
phylogenetic trees. Proceedings of the National Academy of Sciences of
the United States of America 1996, 93(23):13429-13434.
16. Maarten LJG MV, Jacqueline WCM, Michael PJ , and Ewout SW: Net
Reclassification Improvement: Computation, Interpretation,and
Controversies. Annals of Internal Medicine 2014, 160:10.

28
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
the ROC curve to reclassification and beyond. Statistics in Medicine 2008,
27(2):157-172; Discussion 207-112.
18. Pearte CA, Furberg CD, O'Meara ES, Psaty BM, Kuller L, Powe NR,
Manolio T: Characteristics and baseline clinical predictors of future fatal
versus nonfatal coronary heart disease events in older adults: the
Cardiovascular Health Study. Circulation 2006, 113(18):2177-2185.
19. Halanych JH, Shuaib F, Parmar G, Tanikella R, Howard VJ, Roth DL,
Prineas RJ, Safford MM: Agreement on cause of death between proxies,
death certificates, and clinician adjudicators in the Reasons for
Geographic and Racial Differences in Stroke (REGARDS) study.
American journal of epidemiology 2011, 173(11):1319-1326.
20. Olubowale OT, Safford MM, Brown TM, Durant RW, Howard VJ, Gamboa
C, Glasser SP, Rhodes JD, Levitan EB: Comparison of Expert Adjudicated
Coronary Heart Disease and Cardiovascular Disease Mortality With the
National Death Index: Results From the REasons for Geographic And
Racial Differences in Stroke (REGARDS) Study. Journal of the American
Heart Association 2017, 6(5).
21. Brown TM, Deng L, Becker DJ, Bittner V, Levitan EB, Rosenson RS,
Safford MM, Muntner P: Trends in mortality and recurrent coronary heart
disease events after an acute myocardial infarction among Medicare
beneficiaries, 2001-2009. American heart journal 2015, 170(2):249-255.

29
22. Graham DJ, Ouellet-Hellstrom R, MaCurdy TE, Ali F, Sholley C, Worrall C,
Kelman JA: Risk of acute myocardial infarction, stroke, heart failure, and
death in elderly Medicare patients treated with rosiglitazone or
pioglitazone. JAMA : the journal of the American Medical Association
2010, 304(4):411-418.
23. Soliman EZ, Prineas RJ, Case LD, Russell G, Rosamond W, Rea T,
Sotoodehnia N, Post WS, Siscovick D, Psaty BM et al:
Electrocardiographic and clinical predictors separating atherosclerotic
sudden cardiac death from incident coronary heart disease. Heart 2011,
97(19):1597-1601.
24. Matthew D. Ritchey FL, Hilary K. Wall, Claudia A. Steiner, Cathleen
Gillespie, Mary G. George, Janet S. Wright: Million Hearts: Description of
the National Surveillance and Modeling Methodology Used to Monitor the
Number of Cardiovascular Events Prevented During 2012–2016. Journal
of the American Heart Association 2017, 6(5).
25. Goff DC, Jr., Lloyd-Jones DM, Bennett G, Coady S, D'Agostino RB, Sr.,
Gibbons R, Greenland P, Lackland DT, Levy D, O'Donnell CJ et al: 2013
ACC/AHA guideline on the assessment of cardiovascular risk: a report of
the American College of Cardiology/American Heart Association Task
Force on Practice Guidelines. Journal of the American College of
Cardiology 2014, 63(25 Pt B):2935-2959.
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
cholesterol to reduce atherosclerotic cardiovascular disease risk in adults:
synopsis of the 2013 American College of Cardiology/American Heart
Association cholesterol guideline. Annals of internal medicine 2014,
160(5):339-343.
27. Kenneth D. Kochanek MA, Sherry L. Murphy, B. S., Jiaquan Xu, M.D.,
Betzaida Tejada-Vera, M. S., Division of Vita Statistics: Deaths: Final Data
for 2014. National Vital Statistics Reports 2016, 65(4):122.
28. WHO: Stroke--1989. Recommendations on stroke prevention, diagnosis,
and therapy. Report of the WHO Task Force on Stroke and other
Cerebrovascular Disorders. Stroke; a journal of cerebral circulation 1989,
20(10):1407-1431.
31
Table 1: Definitions of fatal events in the REGARDS study.
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
Table 2: Distribution of pre-specified candidate predictors for participants with

and without an adjudicated fatal CVD, CHD, and stroke.
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
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
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
Stroke 13.1% 12.3% 15.8% 13.5% 11.3% 12.1% 31.7%
Physician visit diagnosis code, beyond 28 days of death
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
Malignancy 31.9% 37.8% 11.7% 36.4% 11.3% 33.0% 12.7%

Other heart disease 57.5% 56.5% 60.7% 56.6% 61.6% 57.5% 57.0%
Atrial fibrillation 34.8% 34.0% 37.5% 34.5% 36.3% 34.4% 43.0%
Sudden death or ventricular 14.6% 12.6% 21.6% 12.6% 23.8% 14.7% 13.4%
arrhythmias
Other and unspecified 1.3% 1.0% 2.3% 1.5% 0.4% 0.9% 8.5%
intracranial hemorrhage
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
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%
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)
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)
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)
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)
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)
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)
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
Figure 1: Direct and indirect approaches for developing the claims-based
algorithms
CHD: Coronary heart disease; CVD: Cardiovascular disease; MI: myocardial infarction
38
Figure 2: Flow chart for cohort selection

REGARDS
participants
N=30,239
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
Excluded participants with an adjudicated death date in N=130

REGARDS that differs from the death date in Medicare Excluded
N=4,197
Excluded participants younger than 65.5 years of age on N=300

death date using date from Medicare Excluded
N=3,897
Excluded participants without 182 consecutive days of N=1,142

fee-for-service coverage prior to death Excluded
N=2,755
Excluded participants with hospital discharge diagnosis code N=32

for stroke within 28 days of death but no medical record was Excluded
obtained for REGARDS adjudication
N=2,723
Excluded participants with hospital discharge diagnosis code N=38 Excluded

for MI within 28 days but no medical record was
obtained for REGARDS adjudication
Final study
population
N=2,685
39
Figure 3: Receiver operating characteristic curve for fatal CVD, fatal CHD, and
fatal stroke
CHD: Coronary heart disease; CVD: Cardiovascular disease

40
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
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
Supplemental Table 2: Beta-coefficients from logistic regression models for

predictor variables for the outcomes of fatal cardiovascular disease, coronary
heart disease, stroke, myocardial infarction, and coronary heart disease death.
Predictors Fatal Fatal Fatal Fatal MI CHD
CVD† CHD† stroke death
Intercept 0.5519 -1.2950 -4.6061 -5.1933 -1.3376
Age at death in years -0.0229 - - - -
Black race - - - -0.3827 -
Number oncologist visits within 6 months of death -0.0525 - - - -
Number vascular visit within 6 months of death -0.1738 -0.2663 - - -0.3337
Number cardiologist visit within 6 months of death - - - 0.0193 -
Hospital discharge diagnosis code (any position) within 28 days of death
Myocardial infarction 3.3520 3.4409 - 3.3121 2.4135
Stroke 3.0341 - 3.8227 - -
Outpatient physician visit diagnosis code within 28 days of death
Myocardial infarction 1.5437 1.4851 - 2.1178 -
Stroke 0.5438 - 1.8607 - -
Hospital discharge diagnosis code (any position), any time prior to death
Chronic obstructive pulmonary disease - -0.2779 - - -0.3839
End stage renal disease and renal failure - 0.4687 - 0.7346
Malignancy -0.5460 -0.8352 - - -0.6103
Other and unspecified intracranial hemorrhage 1.0886 - 2.4584 - -
Other acute and subacute forms of ischemic heart - 0.3747 - - -
disease
Hospital discharge, emergency room visit or nursing home visit diagnosis code, any time before death
Sudden death 1.0606 0.9951 - - 0.7573
Hospital discharge status, any time before death
Discharged to skill nursing facility -0.4797 -0.5936 - -0.8439 -0.4434
Discharged to hospice -0.6684 -0.7276 - 0.5703 -1.5154
Died in hospital - - - 1.3590 -1.2441
28 days or more for last hospitalization - - - -1.4271 -
ICD-9-CM procedure code or CPT code, any time prior to death
Coronary artery bypass grafting or percutaneous 0.3335 - - - -
coronary intervention
Liver, gall and pancreas disease -0.3156 - - - -
Other symptoms involving abdomen and pelvis - - - - -0.4041
Sudden death or ventricular arrhythmias 0.5941 - - -
Other forms of chronic ischemic heart disease - - - - 0.4051
AHRQ classification based on diagnosis code from Hospital discharge or physician visit
100 Acute myocardial infarction - 0.4710 - - 0.6590
101 Coronary atherosclerosis and other heart 0.4916 - - - -
disease
107 Cardiac arrest and ventricular fibrillation - 0.7685 - - 1.0274
108 Congestive heart failure; non-hypertensive - - - 1.1515 -
109 Acute cerebrovascular disease - - 1.8620 - -
113 Late effects of cerebrovascular disease -0.8000 -0.6793
120 Hemorrhoids -0.4077 - - - -
132 Lung disease due to external agents - -1.2412 1.4820 - -
137 Disease of mouth; excluding dental - - - 0.8780 -
144 Regional enteritis and ulcerative colitis - - - 0.9789 -
147 Anal and rectal conditions - - - - -0.8248
148 Peritonitis and intestinal abscess - - - - 0.9585
158 Chronic kidney disease - - -0.8042 - -
162 Other disease of bladder and urethra - - - -0.6951 -
19 Cancer of bronchus; lung -0.9380 -0.7994 - - -0.9709
198 Other inflammatory condition of skin - - - -0.5745 -
2 Septicemia (except in labor) - - - 0.4467 -
203 Osteoarthritis - - - 0.7322 -
208 Acquired foot deformities - - - -0.9189 -
211 Other connective tissue diseases - - -1.1380 - -
43
Supplemental Table 2: Continuation.

Predictors Fatal Fatal Fatal Fatal - CHD
CVD† CHD† stroke MI death
AHRQ classification based on diagnosis code from hospital discharge or physician visit
237 Complication of device; implant or graft - - - - 0.3969
242 Poisoning by other medications and drugs - - 0.7573 - -
254 Rehabilitation care; fitting of prostheses; and 0.4232 0.5099 - - -
adjustment of devices
256 Medical examination/evaluation - 0.3389 - - -
2620 E Code: Unspecified - - -2.7530 - -
3 Bacterial infection; unspecified site - - 0.6041 - -
42 Secondary malignancy -1.3967 -1.3657 -2.3296 -1.6132 -1.1103
48 Thyroid disorders - - -0.6801 - -
49 Diabetes mellitus without complication 0.2637 - - - -
50 Diabetes mellitus with complication 0.3625 - - - 0.3537
52 Nutritional deficiencies - - - - -0.4222
55 Fluid and electrolyte disorders -0.4160 -0.4614 - - -
63 Disease of white blood cells -0.3827 - - -0.5436
64 Other hematologic conditions -0.6405 - - - -
651 Anxiety disorders -0.6807 -0.6371 - -0.7614 -0.3774
653 Delirium, dementia, and amnestic and other -0.3318 -0.4459 - - -0.4516
cognitive disorders
657 Mood disorders - - 0.6994 - -
9 Sexually transmitted infections (not HIV or - - - 1.0945 -
hepatitis)
90 Inflammation; infection of eyes (except that - - - 0.6237 -
caused by tuberculosis or sexually transmitted
disease)
96 heart valve disorders 0.3822 0.3500 - - 0.3973
97 Peri-, endo-, and myocarditis, cardiomyopathy - 0.4155 - - 0.3204
(except that caused by tuberculosis or sexually
transmitted disease)
-: Blank
AHRQ: Agency for healthcare research and quality; CHD: Coronary heart disease; CPT: Current procedure terminology;
CVD: Cardiovascular disease; ICD-9-CM: International classification of diseases, 9th version, clinical modification.
†
Direct approach.
For indirect approach 2 for CHD: intercept=-3.0188; coefficient of predicted probability for MI=5.8050; coefficient of
predicted probability for CHD death=6.0630.
For indirect approach 2 for CVD: intercept=-2.8527; coefficient of predicted probability for stroke= 5.0264; coefficient of
predicted probability for stroke=5.2219; coefficient of predicted probability for CHD death=5.7589.
Supplemental Table 3: Net reclassification index for the claims-based algorithm developed in the current study compared
to hospital discharge diagnosis codes for fatal cardiovascular disease.
REGARDS Model – Not fatal CVD REGARDS Model – Fatal CVD
(same prevalence) (same prevalence) NRI (95% CI)
Hospital model – Hospital model – Hospital model – Hospital model –
Not Fatal CVD Fatal CVD Not Fatal CVD Fatal CVD
Adjudicated fatal CVD Overall 0.285 (0.244, 0.327)
Yes 203 (33.4%) 4 (0.7%) 232 (38.2%) 169 (27.8%) Event 0.375 (0.335, 0.415)
No 1,859 (89.5%) 3 (0.1%) 190 (9.2%) 25 (1.2%) Non-Event -0.090 (-0.104, -0.078)
(Sensitivity ≥0.80) (Sensitivity ≥0.80)
Yes 111 (18.3) 1 (0.2%) 324 (53.3%) 172 (28.3%) Event 0.531 (0.490, 0.571)
No 1,540 (74.2%) 0 (0.0%) 509 (24.5%) 28 (1.4) Non-Event -0.245 (-0.264, -0.227)
(PPV ≥0.80) (PPV ≥0.80)
Yes 319 (52.5%) 7 (1.2%) 116 (19.1%) 166 (27.3%) Event 0.179 (0.147, 0.214)
No 2,000 (96.3) 10 (0.5%) 49 (2.4%) 18 (0.9%) Non-Event -0.019 (-0.027, -0.014)
(Maximum Sum of sensitivity and (Maximum sum of sensitivity and
specificity) specificity)
Yes 131 (21.6%) 1 (0.2%) 304 (50.0%) 172 (28.3) Event 0.498 (0.457, 0.539)
No 1,627 (78.3%) 1 (0.1%) 422 (20.3%) 27 (1.3%) Non-Event -0.203 (-0.221, -0.186)
CVD: Cardiovascular 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.
44
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)
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)
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)
(PPV ≥0.80) (PPV ≥0.80)
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)
(Maximum Sum of sensitivity and (Maximum Sum of sensitivity and
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.
45
to hospital discharge diagnosis codes for fatal stroke.
REGARDS Model – Not fatal stroke REGARDS Model – Fatal stroke
(same prevalence) (same prevalence)
NRI (95% CI)
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)
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)
(PPV ≥0.80) (PPV ≥0.80)
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)
(Maximum Sum of sensitivity and (Maximum Sum of sensitivity and
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.
46
47
Supplemental Table 6: Example of the calculation of the predicted probability for

having a fatal stroke
Predictors* Column A Column B Column C
(Column A x Column B)
Intercept 1 -4.6061 -4.6061
Hospital discharge diagnosis code for stroke within 28 days 1 3.8227 3.8227
of death
Physician visit diagnosis code for stroke within 28 days of 1 1.8607 1.8607
death
Hospital discharge diagnosis code for Other and 0 2.4584 0
unspecified intracranial hemorrhage any time prior to death
†
AHRQ CCS category
109 Acute cerebrovascular disease 1 1.8620 1.8620
132 Lung disease due to external agents 0 1.4820 0
158 Chronic kidney disease 0 -0.8042 0
211 Other connective tissue diseases 1 -1.1380 -1.1380
242 Poisoning by other medications and drugs 0 0.7573 0
2620 E Code: Unspecified 0 -2.7530 0
3 Bacterial infection; unspecified site 0 0.6041 0
42 Secondary malignancy 0 -2.3296 0
48 Thyroid disorders 0 -0.6801 0
657 Mood disorders 0 0.6994 0
Final score (column sum)‡ 1.8013
§
Final calculation of the probability of having a fatal stroke :
Probability of having fatal stroke=exp(1.8013)/(1+exp(1.8013))=0.8583
*The example provided is for a patient with one or more:
• Inpatient claim(s) with a discharge diagnosis code for stroke within 28 days of death date,
• Outpatient claims with a diagnosis code(s) for stroke within 28 days of death date,
• Inpatient or outpatient claim(s) with a diagnosis code(s) mapped to AHRQ CCS 109 (Acute
cerebrovascular disease) any time prior to death date,
• Outpatient claims with a diagnosis code(s) mapped to AHRQ CCS 211 (Other connective tissue
diseases) any time prior to death date.
Column A: All values in Column A for predictors included in algorithm for fatal stroke are dichotomous, with 1 meaning
that the beneficiary had the claim, and 0 that the beneficiary did not have the claim. The value for the intercept in Column
A is 1 for all people.
Column B: Contains the coefficients for each predictor and the intercept term from the fatal stroke algorithm.
Column C: Contains the product of Columns A and B.
†
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 final score is calculated as the sum of all values in Column C.
§
The final probability for having a fatal stroke is calculated as probability = exp(final score)/(1+exp(final score)).
48

having a fatal coronary heart disease event.
Predictors Column A Column B Column C Column D Column E
Intercept 1 -5.1933 -5.1933 -1.3376 -1.3376
Black race 0 -0.3827 0 - -
Number of vascular visits within 6 months of the death date 0 - - -0.3337 0
Number of cardiologist visits within 6 months of the death date 3 0.0193 0.0579 - -
Hospital discharge diagnosis code for myocardial infarction 0 3.3121 0 2.4135 0
within 28 days of death date
Physician visit diagnosis code for Myocardial infarction within 0 2.1178 0 - -
28 days of death date
Hospital discharge diagnosis code, any time before death
date
End stage renal disease 0 0.7346 0 - -
Chronic obstructive pulmonary disease 0 - - -0.3839 0
Malignancy 0 - - -0.6103 0
Sudden death 0 - - 0.7573 0
Discharged to skill nursing facility, any time prior to death date 0 -0.8439 0 -0.4434 0
Discharged to hospice, any time prior to death date 0 0.5703 0 -1.5154 0
Died in hospital 0 1.3590 0 -1.2441 0
28 days or more for last hospitalization 0 -1.4271 0 - -
Physician visit diagnosis code, any time before death date
Other symptoms involving abdomen and pelvis 0 - - -0.4041 0
Other forms of chronic ischemic heart disease 1 - - 0.4051 0.4051
AHRQ CCS category†
100 Acute myocardial infarction 0 - - 0.6590 0
107 Cardiac arrest and ventricular fibrillation 1 - - 1.0274 1.0274
108 Congestive heart failure; non-hypertensive 0 1.1515 0 - -
113 Late effects of cerebrovascular disease 0 - - -0.6793 0
137 Disease of mouth; excluding dental 0 0.8780 0 - -
144 Regional enteritis and ulcerative colitis 0 0.9789 0 - -
147 Anal and rectal conditions 0 - - -0.8248 0
148 Peritonitis and intestinal abscess 0 - - 0.9585 0
162 Other disease of bladder and urethra 0 -0.6951 0 - -
19 Cancer of bronchus; lung 0 - - -0.9709 0
198 Other inflammatory condition of skin 0 -0.5745 0 - -
2 Septicemia (except in labor) 0 0.4467 0 - -
203 Osteoarthritis 1 0.7322 0.7322 - -
208 Acquired foot deformities 0 -0.9189 0 - -
237 Complication of device; implant or graft 0 - - 0.3969 0
42 Secondary malignancy 0 -1.6132 0 -1.1103 0
50 Diabetes mellitus with complication 0 - - 0.3537 0
52 Nutritional deficiencies 0 - - -0.4222 0
63 Disease of white blood cells 0 - - -0.5436 0
651 Anxiety disorders 0 -0.7614 0 -0.3774 0
653 Delirium, dementia, and amnestic and other cognitive 0 - - -0.4516 0
disorders
9 Sexually transmitted infections (not HIV or hepatitis) 0 1.0945 0 - -
90 Inflammation; infection of eyes 1 0.6237 0.6237 - -
96 heart valve disorders 0 - - 0.3973 0
97 Peri-, endo-, and myocarditis, cardiomyopathy 0 - - 0.3204 0
Intermediate scores (column sum) -3.7795 0.0949
Probability for each intermediate outcome‡ 0.0223 0.5237
Final score§: -3.0188+5.8050x0.0223+6.0630x0.5237=0.2859
Probability for having fatal CHD=exp(0.2859)/(1+exp(0.2859))=0.5710
The example provided is for a patient with:
• Three cardiologist visits within 6 months of their date of death.
• One or more outpatient claim(s) with a diagnosis code(s) for other forms of chronic ischemic heart disease any time prior to
their date of death.
• One or more inpatient or outpatient claim(s) with a diagnosis code(s) mapped to AHRQ CCS 203 (Osteoarthritis) any time
prior to their date of death.
• One or more inpatient or outpatient claim(s) with a diagnosis code(s) mapped to AHRQ CCS 90 (Inflammation; infection of
eyes) any time prior to their date of death.
Column A: All values (except number of vascular visits and number of cardiologist visits within 6 months of the death date) in Column A are
for variables 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. 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.
Column C: Contains the product of Columns A and B.
49
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

having a fatal cardiovascular disease event.
Predictors Column Column Column Column Column Column Column
A B C D E F G
Intercept 1 -5.1933 -5.1933 -1.3376 -1.3376 -4.6061 -4.6061
Black race 0 -0.3827 0 - - - -
Number vascular visit within 6 months of death 0 -0.3337 0 - -
Number cardiologist visit within 6 months of death 1 0.0193 0.0193 - - - -
Hospital discharge diagnosis code for Myocardial 0 3.3121 0 2.4135 0 - -
infarction within 28 days of death
Hospital discharge diagnosis code for stroke 1 - - - - 3.8227 3.8227
within 28 days of death
Physician visit diagnosis code for Myocardial 0 2.1178 0 - - - -
infarction within 28 days of death
Physician visit diagnosis code for Stroke within 28 1 - - - - 1.8607 1.8607
days of death
Hospital discharge diagnosis code, any time prior
to death date
Chronic obstructive pulmonary disease 0 - - -0.3839 0 - -
End stage renal disease and renal failure 0 0.7346 0 - - - -
Malignancy 1 - - -0.6103 -0.6103 - -
Other and unspecified intracranial hemorrhage 0 - - - - 2.4584 0
Sudden death† 0 - - 0.7573 0 - -
Discharged to skill nursing facility, any time prior 1 -0.8439 -0.8439 -0.4434 -0.4434 - -
to death date
Discharged to hospice, any time prior to death 0 0.5703 0 -1.5154 0 - -
date
Died in hospital 1 1.3590 1.3590 -1.2441 -1.2441 - -
28 days or more stay for last hospitalization 0 -1.4271 0 - - - -
††
Other symptoms involving abdomen and pelvis 0 - - -0.4041 0 - -
††
Other forms of chronic ischemic heart disease 1 - - 0.4051 0.4051 - -
AHRQ CCS category†††
100 Acute myocardial infarction 1 - - 0.6590 0.6590 - -
107 Cardiac arrest and ventricular fibrillation 0 - - 1.0274 0 - -
108 Congestive heart failure; non-hypertensive 1 1.1515 1.1515 - - - -
109 Acute cerebrovascular disease 1 - - - - 1.8620 1.8620
113 Late effects of cerebrovascular disease 0 - - -0.6793 0
132 Lung disease due to external agents 0 - - - - 1.4820 0
137 Disease of mouth ; excluding dental 0 0.8780 0 - - - -
144 Regional enteritis and ulcerative colitis 0 0.9789 0 - - - -
147 Anal and rectal conditions 0 - - -0.8248 0 - -
148 Peritonitis and intestinal abscess 0 - - 0.9585 0 - -
158 Chronic kidney disease 1 - - - - -0.8042 -0.8042
162 Other disease of bladder and urethra 0 -0.6951 0 - - - -
19 Cancer of bronchus; lung 0 - - -0.9709 0 - -
198 Other inflammatory condition of skin 0 -0.5745 - - 0 - -
2 Septicemia (except in labor) 0 0.4467 - - 0 - -
203 Osteoarthritis 1 0.7322 0.7322 - - - -
208 Acquired foot deformities 1 -0.9189 -0.9189 - - - -
211 Other connective tissue diseases 1 - - - - -1.1380 -1.1380
237 Complication of device; implant or graft 0 - - 0.3969 0 - -
242 Poisoning by other medications and drugs 0 - - - - 0.7573 0
2620 E Code: Unspecified 0 - - - - -2.7530 0
3 Bacterial infection; unspecified site 0 - - - - 0.6041 0
42 Secondary malignancy 0 -1.6132 0 -1.1103 0 -2.3296 0
48 Thyroid disorders 1 - - - - -0.6801 -0.6801
50 Diabetes mellitus with complication 1 - - 0.3537 0.3537 - -
52 Nutritional deficiencies 0 - - -0.4222 0 - -
63 Disease of white blood cells 0 - - -0.5436 0 - -
65 Anxiety disorders 0 -0.7614 0 -0.3774 0 - -
653 Delirium, dementia, and amnestic and other 0 - - -0.4516 0 - -
cognitive disorders
657 Mood disorders 0 - - - - 0.6994 0
9 Sexually transmitted infections (not HIV or 0 1.0945 0 - - - -
hepatitis)
90 Inflammation; infection of eyes 0 0.6237 0 - - - -
96 heart valve disorders 1 - - 0.3973 0.3973 - -
97 Peri-; endo-; and myocarditis; 0 - - 0.3204 0 - -
cardiomyopathy
Intermediate scores (sum of column value -5.2558 -1.8203 0.3170
above)
‡
Probability for each intermediate outcome 0.0052 0.1394 0.5786
Final score§:-2.8527+5.2219x0.0052+5.7589x0.1394+5.0264x0.5786=0.8855
Probability for having fatal CVD=exp(0.8855)/(1+exp(0.8855))=0.7080
51
Supplemental Table 8: Continuation.

The example provided is for a patient with:
• One cardiologist visits within 6 months of their date of death.
• One or more inpatient claim(s) with discharge diagnosis code(s) for stroke within 28 days of death date.
• One or more outpatient claim(s) with diagnosis code(s) for stroke within 28 days of death date.
• One or more inpatient claim(s) with diagnosis code(s) for malignancy any time prior to death date.
• Discharged to skilled nursing facility any time prior to death date.
• Died in hospital.
• One or more outpatient claim(s) with diagnosis code(s) for other forms of chronic ischemic heart disease.
• One or more inpatient or outpatient claim(s) with a diagnosis code(s) mapped to AHRQ CCS 100 (Acute myocardial
infarction) any time prior to death date.
• One or more inpatient or outpatient claim(s) with a diagnosis code(s) mapped to AHRQ CCS 108 (Congestive heart
failure; non-hypertensive) any time prior to death date.
• One or more inpatient or outpatient claim(s) with a diagnosis code(s) mapped to AHRQ CCS 109 (Acute
cerebrovascular disease) any time prior to death date.
• One or more inpatient or outpatient claim(s) with a diagnosis code(s) mapped to AHRQ CCS 158 (Chronic kidney
disease) any time prior to death date.
• One or more inpatient or outpatient claim(s) with a diagnosis code(s) mapped to AHRQ CCS 203 (Osteoarthritis) any
time prior to death date.
• One or more inpatient or outpatient claim(s) with a diagnosis code(s) mapped to AHRQ CCS 208 (Acquired foot
deformities) any time prior to death date.
• One or more inpatient or outpatient claim(s) with a diagnosis code(s) mapped to AHRQ CCS 211 (Other connective
tissue diseases) any time prior to death date.
• One or more inpatient or outpatient claim(s) with a diagnosis code(s) mapped to AHRQ CCS 48 (Thyroid disorders) any
time prior to death date.
• One or more inpatient or outpatient claim(s) with a diagnosis code(s) mapped to AHRQ CCS 50 (Diabetes mellitus with
complication) any time prior to death date.
• One or more inpatient or outpatient claim(s) with a diagnosis code(s) mapped to AHRQ CCS 96 (heart valve disorders)
any time prior to death date.
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.
††
†††
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 REAL WORLD SETTING
FENGLONG XIE, EMILY B. LEVITAN, HUIFENG YUN, PAUL MUNTNER,

JEFFREY R. CURTIS
Submitted to Arthritis care & research
Format adapted for dissertation

53
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
Fenglong Xie1, 2, Huifeng Yun1, 2, Emily B. Levitan2, Paul Muntner2, Jeffrey R.

Curtis1, 2
1 Division of Clinical Immunology and Rheumatology, 2 Department of
Epidemiology, University of Alabama at Birmingham, Birmingham, AL, USA
Keywords: biologic DMARDS, tocilizumab, myocardial infarction, stroke, fatal
cardiovascular disease
Word count:
Abstract: 250
Main context: 3,734
Tables: 3
Figures: 3
54
Background: Multiple studies have observed seemingly unfavorable changes in
lipid profiles associated with IL6 receptor antagonists (IL-6R) and some other
rheumatoid arthritis (RA) therapies [1-3]. The real-world cardiovascular disease
(CVD) risk associated with the first approved anti IL-6R medication for RA,
tocilizumab, remains uncertain.
Methods: A cohort study using 2006-2015 Medicare and MarketScan claims
data was conducted, studying RA patients who initiated biologic disease-
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
method. A variety of subgroup analyses were conducted to ensure consistency.
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.
There were no statistically significant differences in CVD risk between
tocilizumab and any other biologic using MarketScan data. Results were robust
in numerous subgroup analyses.

55
Conclusion: Consistent with findings of a recently completed safety trial in RA,
tocilizumab was associated with a comparable CVD risk compared to etanercept,
as well as a number of other RA biologics, in two large data sources.
Significance and Innovation:
• Despite seemingly unfavorable changes in lipid profiles associated with IL-
6R therapies for RA, the net effect on CVD risk is unclear
• This observational analysis evaluated the comparative CVD risk of
tocilizumab referent to a variety of other biologic therapies for RA in two
large real-world datasets. Consistent with the results of a large clinical
trial, there was no increased risk for patients treated with TCZ compared
to most other RA biologics including etanercept and abatacept.
• The robustness of these findings was confirmed in a variety of key
subgroup analyses that extended the generalizability of the main results
and included patients at low CVD risk

56
Rheumatoid arthritis (RA) is an autoimmune disease and affects about 0.5-1.0%
of the adult population [4, 5]. If RA is uncontrolled, poor quality of life and
shortened life expectancy often ensue [6-11]. Cardiovascular disease (CVD)-
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
patients[13]. The inflammatory milieu is characterized by elevated production of a
variety of cytokines and inflammatory markers including interleukin-1 and (IL-6),
C-reactive protein (CRP) and tumor necrosis factor (TNF). These cytokines and
inflammatory markers play an important pro-inflammatory role in RA patients and
also have been shown to promote atherosclerosis [14-16].
Biologic disease-modifying antirheumatic drugs (bDMARDS) are effective
treatments for RA patients with inadequate response to conventional DMARDS
[17-23] such as methotrexate (MTX). bDMARDS reduce the inflammatory
response through varying mechanisms of action: abatacept (ABA) interferes with
T cell costimulation [23]; TNF inhibitors (TNFi) including adalimumab (ADA),
certolizumab (CER), etanercept (ETA), golimumab (GOL) and infliximab (INF)
control inflammation by binding to TNF[17-19, 21, 22]; and rituximab (RIT)
controls the immune response by binding to and depleting CD20 B cells[20].
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].
Tocilizumab (TCZ) is a humanized monoclonal antibody against the IL-6 receptor
[25]; it is effective in the treatment for RA patients [26]. Multiple studies have
57
observed seemingly unfavorable changes in lipid profile in RA patients treated
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
recently completed randomized clinical trial of 3,080 patients (ENTRACTE,
ClinicalTrials.gov: NCT01331837) compared the CVD risk associated with TCZ
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
observational studies likewise found no increased risk of myocardial infarction
(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
CVD events in the two observational studies, respectively). Failure to include
fatal CVD as an outcome may under-estimate the incidence of true CVD-related
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
compared TCZ to a pooled TNFi exposure group without evaluating individual
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
disease activity is associated with higher CVD risk.
The objective of this study was to assess the CVD risk associated with TCZ
compared to individual TNFi therapies, as well as to other biologics used for RA
(e.g. RIT, ABA). As part of the investigation, we evaluated the potential for
confounding using externally linked laboratory data measuring RA disease
activity. We also provided additional evidence in clinically important subgroups by
stratifying patients with respect to key CVD risk factors to identify both higher and
lower CVD risk patients.
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
claims data for RA patients. Medicare is a U.S., government sponsored health
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
that aggregate information from a variety of employer and health-plans.
Cohort inclusion: To be included in the cohort, Medicare beneficiaries and
MarketScan participants must have 1) initiated at least one of the bDMARDS for
the treatment of RA (all of which are generally prescribed only by

59
rheumatologists) on January 1, 2010 or thereafter (when TCZ became available
in the US); 2) were diagnosed with having RA, defined by two International
Classification of Disease, Ninth Revision, Clinical Modification (ICD-9-CM)
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
coverage immediately preceding initiation of bDMARDS; For Medicare
beneficiaries, observability was defined by enrollment in Medicare Part A
(hospital), Part B (outpatient) and Part D (pharmacy), and not enrolled in Part C
(Medicare advantage, as claims for beneficiaries enrolled in part C are not
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-
9-CM diagnosis code for other autoimmune/inflammatory diseases including
inflammatory bowel disease, psoriatic arthritis, psoriasis, or ankylosing
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
current procedural terminology (CPT) code for percutaneous coronary
intervention (PCI) and coronary artery bypass grafting (CABG) (Supplemental
Table 1); 3) had history of malignancy (ignoring non-melanoma skin cancer), HIV
or organ transplantation, also using all available prior data.
Follow-up: Follow-up started at the initiation of each bDMARDS and ended at
earliest of 1) end of exposure of the specific agent plus a 90 day extension (see
60
below for details) 2) switch to other bDMARDS or tofacitinib; 3) outcome 4) death
date (inclusive of fatal CVD outcome); 5) loss of coverage; 6) end of study
(September 30, 2015, when the U.S. transitioned to ICD-10 coding).
Main exposure: Exposure to bDMARDS was determined based on prescription
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
infusion drugs were identified using Healthcare Common Procedure Coding
System (HCPCS) codes in Part B. Infusion drugs (ABA, CER and TCZ) in
Medicare before a specific J code were available were identified using a
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].
Other medication exposures and covariates including conventional synthetic
DMARDS, prescription non-steroidal anti-inflammatory drug, statins and other
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
most recent prescription in the baseline period[33]. Average daily dose of
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
prior to follow up start. CVD risk factors (diabetes, hyperlipidemia, hypertension,
obesity, chronic kidney disease, and smoking), other co-morbidities and health
behaviors and health care utilization were assessed in baseline, chosen based
on subject matter expertise. A history of CVD-related conditions (hxCVD)
including other acute, subacute forms of ischemic heart disease, angina pectoris,
and other forms of chronic ischemic heart disease, other and unspecified
intracranial hemorrhage, and transient cerebral ischemia (Supplemental Table 1)
was assessed using all available data before initiation.
Outcome assessment: The primary outcome was composite of incident MI,
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
[PPV] ≥94%)[34]. Incident hemorrhagic or ischemic stroke was defined as: at
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
claims based algorithm with PPV ≥ 0.80[30],
Statistical methods
Mean and standard deviation were calculated for continuous variables and
proportions were calculated for categorical variables. Incidence rates (IRs) were
calculated by dividing number of events by person years exposed; 95%
confidence intervals (CIs) were calculated with Poisson regression. Cox
regressions were used to calculate unadjusted and multivariable adjusted hazard

62
ratios (aHR) and 95% CIs. Demographic character ( age, sex), co-morbidities
(hxCVD, 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), healthcare utilization (any hospitalization, number of physician visit),
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
age (e.g. disability) were also adjusted for Medicare data.
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
claims-based algorithm for fatal CVD resulted in higher sensitivity (≥0.80), or
resulted in equal sensitivity and PPV, wherein the predicted proportion of events
equal observed proportion of events. Using external adjustment methods,
potential confounding for RA disease activity was examined using the Multi-
Biomarker Disease Activity (MBDA) test results, according to the method
described by Schneeweiss et.al. [38]. The prevalence of moderate or high RA
disease activity (P C1 and P C0 ) by medication exposure group was assessed in a
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
the University of Alabama at Birmingham approved the study.
RESULTS
We identified 354,486 (Medicare 206,275 + MarketScan 148,211) RA patients
and 463,446 (Medicare 271,832 + MarketScan 191,614) initiations of bDMARDS
(Figure 1). After applying the inclusion and exclusion criteria, the final cohort
64
contained 88,463 (Medicare 46,648/ MarketScan 41,815) RA patients and
117,493 (Medicare 61,715/ MarketScan 55,778) initiations, ranging from a low of
8,469 episodes for RIT to 77,048 episodes for TNFi. The mean (SD) age was
64.7 (12.1) in Medicare, 52.2 (12.3) in MarketScan, 83.9% of patients in
Medicare and 80.5% in MarketScan were female, 68.6% were non-Hispanic
White in Medicare.
Table 1a and 1b show the distribution of select demographic and clinical
characteristics stratified by bDMARDS initiated. TCZ users were similar to ABA
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
baseline and to be naïve to bDMARDS. The mean follow-up time was
approximately 1 year for all exposure groups.
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
medications) was 15.0 (95%CI: 13.9-16.3). The corresponding aHRs referent to
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
CVD accounted 6% to 35% of composite CVD events, depending on the
sensitivity of the chosen algorithm for fatal CVD.
Subgroup analysis showed that when compared to RA patients without hxCVD
(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
RA patients with hxCVD (Figure 3, first panel) or without hxCVD (Figure 3,
second panel) were numerically similar to aHRs for whole cohort respectively,
but with wider 95%CI.
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
primary algorithm. As expected based on the smaller number of events, the
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
(95%CI: 1.9-3.8) to 3.7 (95%CI: 2.2-6.1) for stroke.
The adjusted HRs from MarketScan data were similar to corresponding HRs from
Medicare data, but with much wider 95%CI.
Assessment of confounding by RA disease activity as assessed by the multi-
biomarker disease activity
Results from the MBDA lab tests linked to Medicare data are shown in
supplemental table 4. The highest proportion of patients in high disease activity

67
categories were found amongst RTX and TCZ users, and TNFi users were more
likely to be in lower disease activity categories. The computed true HR for
composite CVD comparing ETA to TCZ was minimally different (HR=1.11) after
adjusting for RA disease activity as measured using the MBDA (categorized as
moderate or high disease activity, referent to low/remission).
DISCUSSION
In this retrospective cohort study of Medicare- and commercially-insured RA
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
significantly higher risk for CVD.
Several randomized clinical trials have observed increase in LDL cholesterol
levels in TCZ treated RA patients, higher than the increase in LDL cholesterol
level seen with other conventional DMARDS or bDMARDS treated RA patients
[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
completed randomized controlled trial (ENTRACTE, ClinicalTrials.gov:
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
group, providing additional evidence to the finding of no increased risk.
A randomized trial among 3,080 RA patients was recently conducted comparing
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
factors than we studied. They reported a non-significant, HR of 1.05 (TCZ vs.
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
stroke associated with TCZ as compared to ETA, as was numerically suggested
in the randomized trial, although based only on 42 total stroke events.
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
were included. We used a validated algorithm to identify infusion drugs (ABA,
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
separately, as we did. We also included both bDMARDS naïve and experienced
RA sub groups; we found the number of previous bDMARDS was not
significantly associated with increased or decreased risk for CVD. Recognizing
the potential for residual confounding due to unmeasured factors such as RA
disease activity, we used laboratory measure of RA disease activity, the MBDA,
to assess the distribution of disease activity. TCZ users had a higher proportion
of individuals in higher disease activity at baseline compared to TNFi users. The
computed HR for composite CVD comparing ETA to TCZ was no different from
the adjusted HR without external MBDA adjustment. This addresses concerns
that RA disease activity, which has been previously shown to be associated with
CVD risk [40], might confound the associations that we observed.

70
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
exposure. This is not a problem when studying intravenous therapies, however,
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
adjudicated, although we relied on claims-based algorithms that have in excess
of 80-90% in their accuracy, and there is little reason to expect that performance
of these would be differential by specific biologic exposure.
In conclusion, our observational study confirmed results from a large randomized
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
to slightly increased CVD risk associated with infliximab.

71
References
1. Kawashiri SY, Kawakami A, Yamasaki S, Imazato T, Iwamoto N, Fujikawa
K, Aramaki T, Tamai M, Nakamura H, Ida H et al: Effects of the anti-
interleukin-6 receptor antibody, tocilizumab, on serum lipid levels in
patients with rheumatoid arthritis. Rheumatology international 2011,
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
changes and decreased hepatic LDL receptor expression by tocilizumab
in rheumatoid arthritis. Atherosclerosis 2013, 229(1):174-181.
3. Gabay C, McInnes IB, Kavanaugh A, Tuckwell K, Klearman M, Pulley J,
Sattar N: Comparison of lipid and lipid-associated cardiovascular risk
marker changes after treatment with tocilizumab or adalimumab in
patients with rheumatoid arthritis. Annals of the rheumatic diseases 2016,
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
prevalence of arthritis and other rheumatic conditions in the United States.
Part I. Arthritis and rheumatism 2008, 58(1):15-25.
5. Gabriel SE: The epidemiology of rheumatoid arthritis. Rheumatic diseases
clinics of North America 2001, 27(2):269-281.

72
6. Lillegraven S, Kvien TK: Measuring disability and quality of life in
established rheumatoid arthritis. Best practice & research Clinical
rheumatology 2007, 21(5):827-840.
7. Myllykangas-Luosujarvi R, Aho K, Kautiainen H, Isomaki H:
Cardiovascular mortality in women with rheumatoid arthritis. The Journal
of rheumatology 1995, 22(6):1065-1067.
8. Solomon DH, Karlson EW, Rimm EB, Cannuscio CC, Mandl LA, Manson
JE, Stampfer MJ, Curhan GC: Cardiovascular morbidity and mortality in
women diagnosed with rheumatoid arthritis. Circulation 2003,
107(9):1303-1307.
9. Fischer LM, Schlienger RG, Matter C, Jick H, Meier CR: Effect of
rheumatoid arthritis or systemic lupus erythematosus on the risk of first-
time acute myocardial infarction. The American journal of cardiology 2004,
93(2):198-200.
10. Wong JB, Ramey DR, Singh G: Long-term morbidity, mortality, and
economics of rheumatoid arthritis. Arthritis and rheumatism 2001,
44(12):2746-2749.
11. Mok CC, Kwok CL, Ho LY, Chan PT, Yip SF: Life expectancy,
standardized mortality ratios, and causes of death in six rheumatic
diseases in Hong Kong, China. Arthritis and rheumatism 2011,
63(5):1182-1189.
12. Avina-Zubieta JA, Choi HK, Sadatsafavi M, Etminan M, Esdaile JM,
Lacaille D: Risk of cardiovascular mortality in patients with rheumatoid

73
arthritis: a meta-analysis of observational studies. Arthritis and
rheumatism 2008, 59(12):1690-1697.
13. del Rincon ID, Williams K, Stern MP, Freeman GL, Escalante A: High
incidence of cardiovascular events in a rheumatoid arthritis cohort not
explained by traditional cardiac risk factors. Arthritis and rheumatism
2001, 44(12):2737-2745.
14. Ridker PM, Cushman M, Stampfer MJ, Tracy RP, Hennekens CH:
Inflammation, aspirin, and the risk of cardiovascular disease in apparently
healthy men. The New England journal of medicine 1997, 336(14):973-
979.
15. Ridker PM, Rifai N, Stampfer MJ, Hennekens CH: Plasma concentration
of interleukin-6 and the risk of future myocardial infarction among
apparently healthy men. Circulation 2000, 101(15):1767-1772.
16. Kofler S, Nickel T, Weis M: Role of cytokines in cardiovascular diseases: a
focus on endothelial responses to inflammation. Clinical science 2005,
108(3):205-213.
17. Moreland LW, Baumgartner SW, Schiff MH, Tindall EA, Fleischmann RM,
Weaver AL, Ettlinger RE, Cohen S, Koopman WJ, Mohler K et al:
Treatment of rheumatoid arthritis with a recombinant human tumor
necrosis factor receptor (p75)-Fc fusion protein. The New England journal
of medicine 1997, 337(3):141-147.
18. Maini R, St Clair EW, Breedveld F, Furst D, Kalden J, Weisman M,
Smolen J, Emery P, Harriman G, Feldmann M et al: Infliximab (chimeric

74
anti-tumour necrosis factor alpha monoclonal antibody) versus placebo in
rheumatoid arthritis patients receiving concomitant methotrexate: a
randomised phase III trial. ATTRACT Study Group. Lancet 1999,
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
human anti-tumor necrosis factor alpha monoclonal antibody, for the
treatment of rheumatoid arthritis in patients taking concomitant
methotrexate: the ARMADA trial. Arthritis and rheumatism 2003, 48(1):35-
45.
20. Emery P, Fleischmann R, Filipowicz-Sosnowska A, Schechtman J,
Szczepanski L, Kavanaugh A, Racewicz AJ, van Vollenhoven RF, Li NF,
Agarwal S et al: The efficacy and safety of rituximab in patients with active
rheumatoid arthritis despite methotrexate treatment: results of a phase IIB
randomized, double-blind, placebo-controlled, dose-ranging trial. Arthritis
and rheumatism 2006, 54(5):1390-1400.
21. Smolen J, Landewe RB, Mease P, Brzezicki J, Mason D, Luijtens K, van
Vollenhoven RF, Kavanaugh A, Schiff M, Burmester GR et al: Efficacy and
safety of certolizumab pegol plus methotrexate in active rheumatoid
arthritis: the RAPID 2 study. A randomised controlled trial. Annals of the
rheumatic diseases 2009, 68(6):797-804.
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
active rheumatoid arthritis after treatment with tumour necrosis factor
alpha inhibitors (GO-AFTER study): a multicentre, randomised, double-
blind, placebo-controlled, phase III trial. Lancet 2009, 374(9685):210-221.
23. Kremer JM, Genant HK, Moreland LW, Russell AS, Emery P, Abud-
Mendoza C, Szechinski J, Li T, Ge Z, Becker JC et al: Effects of abatacept
in patients with methotrexate-resistant active rheumatoid arthritis: a
randomized trial. Annals of internal medicine 2006, 144(12):865-876.
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
cardiovascular disease in patients with rheumatoid arthritis: a systematic
literature review. Rheumatology 2011, 50(3):518-531.
25. Sheppard M, Laskou F, Stapleton PP, Hadavi S, Dasgupta B: Tocilizumab
(Actemra). Human vaccines & immunotherapeutics 2017:0.
26. Maini RN, Taylor PC, Szechinski J, Pavelka K, Broll J, Balint G, Emery P,
Raemen F, Petersen J, Smolen J et al: Double-blind randomized
controlled clinical trial of the interleukin-6 receptor antagonist, tocilizumab,
in European patients with rheumatoid arthritis who had an incomplete
response to methotrexate. Arthritis and rheumatism 2006, 54(9):2817-
2829.
27. al JTGe: Comparative Cardiovascular Safety of Tocilizumab Vs
Etanercept in Rheumatoid Arthritis: Results of a Randomized, Parallel-
Group, Multicenter, Noninferiority, Phase 4 Clinical Trial. ACR abstract
2016, late-Breaking Abstract Poster Session(3L).

76
28. Kim SC, Solomon DH, Rogers JR, Gale S, Klearman M, Sarsour K,
Schneeweiss S: Cardiovascular Safety of Tocilizumab Versus Tumor
Necrosis Factor Inhibitors in Patients With Rheumatoid Arthritis: A Multi-
Database Cohort Study. Arthritis & rheumatology 2017, 69(6):1154-1164.
of the rheumatic diseases 2016.
30. Fenglong Xie LDC, Jeffrey R. Curtis, Meredith L. Kilgore, Emily B. Levitan,
Keri L. Monda, Monika M. Safford, Ben Taylor, Mark Woodward, Paul
Muntner.: Development of algorithms for identifying fatal cardiovascular
disease in Medicare claims. Pharmacoepidemiology & Drug safety 2017,
submitted.
31. Maradit-Kremers H, Crowson CS, Nicola PJ, Ballman KV, Roger VL,
Jacobsen SJ, Gabriel SE: Increased unrecognized coronary heart disease
and sudden deaths in rheumatoid arthritis: a population-based cohort
study. Arthritis and rheumatism 2005, 52(2):402-411.
32. Curtis JR, Xie F, Chen R, Chen L, Kilgore ML, Lewis JD, Yun H, Zhang J,
Wright NC, Delzell E: Identifying newly approved medications in Medicare
claims data: a case study using tocilizumab. Pharmacoepidemiology and
drug safety 2013, 22(11):1214-1221.

77
33. Audrey L Fan JNF, R Van Harrison, Elizabeth A Jackson, Marie A
Marcelino.: Screening and Management of Lipids. UMHS Lipid Therapy
Guideline 2014.
34. Kiyota Y, Schneeweiss S, Glynn RJ, Cannuscio CC, Avorn J, Solomon
DH: Accuracy of Medicare claims-based diagnosis of acute myocardial
infarction: estimating positive predictive value on the basis of review of
hospital records. American heart journal 2004, 148(1):99-104.
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-
based stroke algorithms in contemporary Medicare data: reasons for
geographic and racial differences in stroke (REGARDS) study linked with
medicare claims. Circulation Cardiovascular quality and outcomes 2014,
7(4):611-619.
36. Lin D, Wei, L. J., and Ying, Z. : Checking the Cox Model with Cumulative
Sums of Martingale-Based Residuals. Biometrika 1993, 80:16.
37. Lee EW, Wei, L. J. Amato, D.: Cox-Type Regression Analysis for Large
Numbers of Small Groups of Correlated Failure Time Observations.
Netherlands: Kluwer Academic 1992:11.
38. Schneeweiss S: Sensitivity analysis and external adjustment for
unmeasured confounders in epidemiologic database studies of
therapeutics. Pharmacoepidemiology and drug safety 2006, 15(5):291-
303.
78
39. Curtis JR, Xie F, Chen L, Saag KG, Yun H, Muntner P: Biomarker-related
risk for myocardial infarction and serious infections in patients with
rheumatoid arthritis: a population-based study. Annals of the rheumatic
diseases 2017.
40. Dixon WG, Watson KD, Lunt M, Hyrich KL, British Society for
Rheumatology Biologics Register Control Centre C, Silman AJ, Symmons
DP, British Society for Rheumatology Biologics R: Reduction in the
incidence of myocardial infarction in patients with rheumatoid arthritis who
respond to anti-tumor necrosis factor alpha therapy: results from the
British Society for Rheumatology Biologics Register. Arthritis and
rheumatism 2007, 56(9):2905-2912.

79
Table 1a: Distribution of baseline characteristics by biologic DMARDS for

Medicare
TNFi
TCZ All ADA ETA INF ABA RIT
N=7,369 N=36,895 N=8,571 N=8,082 N=9,561 N=11,979 N=5,472
Age in years, mean (STD) 64.7 64.2 61.2 61.8 67.2 65.9 65.3
(12.3) (12.1) (12.6) (12.4) (10.6) (11.7) (11.7)
Female, % 85.0 83.1 8.31 82.7 81.9 85.9 82.8
Race, %
White 70.5 67.3 60.0 61.7 74.5 70.1 71.5
Black 8.0 10.6 13.8 12.6 8.4 8.4 8.3
Other 21.5 22.1 26.2 25.6 17.1 21.5 20.2
Follow up time in days, mean (STD) 402 399 (397) 371 392 514 402 (385) 428
(385) (386) (386) (462) (365)
Co morbidity, %
†
History of cardiovascular disease 23.0 20.7 19.1 20.5 20.5 23.1 23.4
Other acute, subacute forms of 2.2 1.7 1.7 1.9 1.5 2.1 1.9
ischemic heart disease
Angina pectoris 5.0 4.4 4.4 4.6 3.8 4.9 4.5
Other forms of chronic ischemic 17.4 16.2 14.7 15.9 16.1 17.9 18.5
heart disease
Other and unspecified 0.2 0.1 0.1 0.2 0.1 0.2 0.3
Transient cerebral ischemia 3.4 2.7 2.6 2.6 2.5 3.2 3.0
Heart failure 5.5 4.7 4.7 5.1 4.4 6.2 7.0
Atrial fibrillation 5.7 4.8 3.5 4.0 5.4 6.1 6.6
Abdominal aortic aneurism 0.6 0.6 0.5 0.7 0.7 0.6 0.7
Peripheral arterial disease 1.9 2.1 1.8 2.1 2.3 2.2 2.2
Diabetes 21.8 23.3 24.7 25.0 22.2 22.9 22.7
Hyperlipidemia 45.2 44.0 40.5 42.0 45.5 45.9 44.3
Hypertension 60.4 61.1 59.5 59.5 62.2 61.9 62.7
Obesity 7.1 6.2 6.5 6.5 5.1 6.3 6.4
Chronic kidney disease 9.8 9.5 8.9 9.2 9.4 10.3 12.5
Chronic obstructive pulmonary 27.9 26.7 28.2 28.5 24.7 26.7 29.6
disease
Fibromyalgia 21.9 20.4 21.1 21.5 17.8 21.6 19.4
Hospitalized infection 8.6 7.9 7.7 9.0 6.9 10.3 13.6
Medication use, %
Methotrexate 53.1 63.7 63.2 62.4 71.5 57.1 52.4
Nonsteroidal anti-inflammatory 41.4 47.5 50.7 50.1 45.6 42.7 39.3
drug
‡
Prior biologic DMARDS
0 9.3 50.8 50.7 62.3 65.1 27.7 23.4
1 35.0 31.8 38.7 27.9 24.7 45.0 34.7
2 33.0 10.8 7.4 6.9 7.3 20.2 25.5
3 and more 22.7 6.6 3.2 2.9 2.9 7.1 16.4
Statin
None 65.3 64.4 66.3 65.4 63.3 64.0 67.3
Low potency 5.2 5.3 4.6 5.1 5.7 5.8 5.0
Moderate potency 24.0 24.7 23.5 23.9 25.6 24.6 22.6
High potency 5.5 5.6 5.6 5.6 5.4 5.6 5.1
Other lipid lowering drug 9.4 9.2 8.8 8.8 9.2 9.9 .5
Steroid
None 32.8 36.6 37.1 35.7 35.1 35.0 28.7
0< <=7.5 50.2 48.7 48.7 48.7 49.5 49.9 48.4
>7.5 17.0 14.7 14.2 15.7 15.4 15.1 22.9
Health behavior and healthcare utility
Smoking 14.1 15.4 17.4 18.6 13.2 13.5 16.7
Prostate specific antigen (male 41.5 43.0 35.1 38.4 49.1 45.3 39.9
only)
PAP smear (female only) 13.4 13.0 13.4 13.3 12.4 12.8 12.4
Mammography (female only) 42.0 40.1 36.4 37.4 44.5 42.9 40.9
Any hospitalization 21.0 18.7 18.2 19.8 17.9 21.8 26.2
Number physician visit, mean 18.2 16.9 16.3 16.8 16.8 18.2 18.9
(STD) (10.2) (10.0) (10.0) (10.3) (9.4) (10.2) (10.8)
State buy in program (i.e. low 31.9 40.8 57.9 54.0 24.7 33.3 31.4
income)
Eligibility for Medicare for reason 53.2 52.6 65.8 62.8 38.2 48.3 51.8
other than age (e.g. disability)
ABA: Abatacept; ADA: Adalimumab; DMARDS: Disease-modifying anti-rheumatic drug. ETA: Etanercept; INF: Infliximab; RIT: Rituximab;
TCZ: Tocilizumab; TNFi: Tumor necrosis factor inhibitor (including, ADA, ETA, INF, certolizumab, and golimumab).
†
Excluded: Acute myocardial infarction, old myocardial infarction, and stroke.
‡
Assessed using all available data.
80
Table 1b: Distribution of characteristics by biologic DMARDS for MarketScan

TNFi
TCZ All ADA ETA INF ABA RIT
N=4,523 N=40,153 N=14,336 N=13,874 N=4,610 N=8,105 N=2,997
Age in years, mean (STD) 52.9 51.7 50.9 51.2 54.4 53.6 54.1
(12.2) (12.3) (11.9) (12.1) (13.1) (12.4) (12.1)
Female, % 83.9 79.4 78.9 78.5 80.1 83.1 82.1
Follow up time in days, mean 329 399 (397) 331 (330) 358 (353) 408 348 392
(STD) (323) (391) (336) (337)
Co morbidity, %
History of cardiovascular 8.1 7.2 6.8 6.6 8.6 9.1 10.3
†
disease
Other acute, subacute forms of 1.1 0.9 0.9 0.8 0.8 1.2 1.2
ischemic heart disease
Angina pectoris 2.0 1.6 1.5 1.3 1.9 1.8 2.4
Other forms of chronic ischemic 5.1 4.7 4.3 4.3 6.0 6.0 7.1
heart disease
Other and unspecified 0.0 0.1 0.1 0.1 0.0 0.1 0.0
Transient cerebral ischemia 1.4 0.9 0.9 0.9 1.2 1.3 1.6
Heart failure 1.4 0.9 0.7 0.9 1.3 1.6 2.3
Atrial fibrillation 1.5 1.4 1.2 1.2 1.9 2.2 2.6
Abdominal aortic aneurism 0.2 0.2 0.1 0.2 0.3 0.1 0.3
Peripheral arterial disease 0.4 0.4 0.3 0.4 0.5 0.6 0.6
Diabetes 12.6 10.9 10.4 10.8 12.7 12.3 15.0
Hyperlipidemia 23.1 20.6 19.5 20.4 23.1 22.6 23.1
Hypertension 33.8 30.8 30.0 29.7 34.8 33.7 35.4
Obesity 3.3 2.6 2.5 2.7 2.4 2.5 3.0
Chronic kidney disease 3.4 3.1 2.8 3.0 3.8 3.9 6.1
Chronic obstructive pulmonary 14.5 12.4 11.8 12.7 14.1 14.0 17.0
disease
Fibromyalgia 15.8 12.9 12.4 13.0 13.6 13.4 12.6
Hospitalized infection 4.3 3.4 2.9 3.6 4.4 5.3 8.2
Medication use, %
Methotrexate 55.2 68.6 30.2 29.3 29.3 57.3 54.4
Nonsteroidal anti-inflammatory 48.3 54.4 55.2 56.3 51.9 48.6 44.3
drug
‡
Prior biologic DMARDS
0 7.3 57.9 60.8 73.8 52.4 20.3 22.0
1 34.9 30.0 33.6 21.4 28.2 40.9 32.4
2 32.0 8.1 4.1 3.4 13.3 29.3 25.6
3 and more 25.8 4.0 1.4 1.3 6.1 9.5 20.0
Statin
None 76.4 79.2 80.2 79.5 77.3 77.8 76.7
Low potency 3.0 2.8 2.7 2.9 3.1 3.1 3.2
Moderate potency 16.5 14.6 13.7 14.4 15.5 15.2 16.0
High potency 4.1 3.4 3.3 3.3 4.2 3.9 3.4
Other lipid lowering drug 6.2 6.2 5.6 5.9 7.5 7.0 7.3
Steroid
None 35.6 38.9 39.4 37.3 39.3 37.9 32.6
0< <=7.5 46.0 47.3 47.0 48.4 45.2 47.9 45.3
>7.5 18.4 13.8 13.6 14.3 15.5 14.2 22.1
Health behavior and healthcare
utility
Smoking 4.4 4.7 4.8 5.0 4.2 4.1 5.0
Prostate specific antigen (male 29.5 29.4 28.6 31.0 29.9 29.1 26.1
only)
PAP smear (female only) 31.6 34.1 34.6 35.0 29.4 31.3 30.0
Mammography (female only) 33.9 36.5 36.8 36.6 35.0 35.8 34.9
Any hospitalization 12.5 10.2 9.4 10.4 11.8 13.1 17.4
Number physician visit, mean 14.6 12.7 12.2 12.6 14.2 14.0 14.1
(STD) (8.2) (7.5) (7.1) (7.4) (8.0) (8.1) (12.1)
ABA: Abatacept; ADA: Adalimumab; DMARDS: Disease-modifying anti-rheumatic drug. ETA: Etanercept; INF: Infliximab; RIT: Rituximab;
TCZ: Tocilizumab; TNFi: Tumor necrosis factor inhibitor (including, ADA, ETA, INF, certolizumab, and golimumab).
†
Excluded: Acute myocardial infarction, old myocardial infarction, and stroke.
‡
Assessed using all available data.
81
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)
Tocilizumab 56 8,073 6.9 (5.3, 9.0) 8 4,068 2.0 (1.0, 3.9)

Pooled TNFi 308 40,113 7.7 (6.9, 8.6) 109 38,173 2.9 (2.4, 3.4)
Myocardial Adalimumab 62 8,660 7.2 (5.6, 9.2) 34 12,956 2.6 (1.9, 3.7)
infarction Etanercept 55 8,643 6.4 (4.9, 8.3) 35 13,584 2.6 (1.8, 3.6)
Infliximab 124 13,369 9.3 (7.8, 11.1) 20 5,139 3.9 (2.5, 6.0)
Abatacept 107 14,613 7.3 (6.1, 8.8) 39 7,702 5.1 (3.7, 6.9)
Rituximab 52 8,643 8.2 (6.2, 10.7) 23 3,200 7.2 (4.8, 10.8)
Tocilizumab 50 8,083 6.2 (4.7, 8.2) 15 4,071 3.7 (2.2, 6.1)

Pooled TNFi 275 40,173 6.8 (6.1, 7.7) 118 3,8168 3.1 (2.6, 3.7)
Adalimumab 52 8,683 6.0 (4.6, 7.9) 35 12,955 2.7 (1.9, 3.8)
Stroke Etanercept 46 8,652 5.3 (4.0, 7.1) 41 13,572 3.0 (2.2, 4.1)
Infliximab 102 13,396 7.6 (6.3, 9.2) 22 13,572 3.0 (2.2, 4.1)
Abatacept 94 14,644 6.4 (5.2, 7.9) 28 7,711 3.6 (2.5, 5.3)
Rituximab 47 6,374 7.4 (5.5, 9.8) 11 3,209 3.4 (1.9, 6.2)
Tocilizumab 15 8,112 1.8 (1.1, 3.1) 1 4,076 0.2 (0.0, 1.7)

Fatal Pooled TNFi 113 40,317 2.8 (2.3, 3.4) 14 38,240 0.4 (0.2, 0.6)
cardiovascul Adalimumab 30 8,710 3.4 (2.4, 4.9) 1 12,975 0.1 (0.0, 0.5)
ar disease Etanercept 16 8,673 1.8 (1.1, 3.0) 3 13,604 0.2 (0.1, 0.7)
(Primary Infliximab 37 13,458 2.7 (2.0, 3.8) 7 5,156 1.4 (0.6, 2.8)
definition) Abatacept 32 14,696 2.2 (1.5, 3.1) 5 7,729 0.6 (0.3, 1.6)
Rituximab 23 6,406 3.6 (2.4, 5.4) 2 3,219 0.6 (0.2, 2.5)
CVD: Cardiovascular disease; DMARDS: Disease-modifying anti-rheumatic drug.
†
Fatal CVD: identified by published algorithm, cut-points=0.27 for predicted proportion closest to observed proportion.
Table 3: Incidence rates for composite cardiovascular† event by History of other cardiovascular disease‡ and biologic
DMARDS by data source.
Data source Medicare MarketScan

Exposure Number Person Incidence rate Number of Person Incidence rate
of event years (95%CI) event years (95%CI)
Tocilizumab 71 6,377 11.1 (8.8, 14.0) 18 3,745 4.8 (3.0, 7.6)
Without Pooled TNFi 415 32.323 12.8 (11.7, 14.1) 171 35,458 4.8 (4.2, 5.6)
history of Adalimumab 79 7.161 11.0 (8.8, 13.8) 52 12,107 4.3 (3.3, 5.6)
other Etanercept 76 6.988 10.9 (8.7, 13.6) 60 12,741 4.7 (3.7, 6.1)
cardiovascular Infliximab 155 10.737 14.4 (12.3, 16.9) 32 4,647 6.9 (4.9, 9.7)
‡
disease Abatacept 131 11,421 11.5 (9.7, 13.6) 46 7,025. 6.5 (4.9, 8.7)
Rituximab 70 4,895 14.3 (11.3, 18.1) 25 2,870 8.7 (5.9, 12.9)
Tocilizumab 33 1,668 19.8 (14.1, 27.8) 3 319 9.4 (3.0, 29.1)
With Pooled TNFi 185 7,650 24.2 (20.9, 27.9) 51 2,644 19.3 (14.7, 25.4)
history of Adalimumab 42 1,473 28.5 (21.1, 38.6) 15 830 18.1 (10.9, 30.0)
other Etanercept 26 1,635 15.9 (10.8, 23.4) 15 812 18.5 (11.1, 30.6)
cardiovascular Infliximab 75 2,571 29.2 (23.3, 36.6) 10 482 20.8 (11.2, 38.6)
‡
disease Abatacept 68 3,140 21.7 (17.1, 27.5) 21 661 31.8 (20.7, 48.7)
Rituximab 35 1,442 24.3 (17.4, 33.8) 10 320 31.2 (16.8, 58.1)
DMARDS: Disease-modifying anti-rheumatic drug.
†
Including incident myocardial infarction, incident stroke and fatal cardiovascular disease. Fatal cardiovascular disease was identified by published
algorithm, cut-points=0.27 for predicted proportion closest to observed proportion.
‡
History of other cardiovascular disease: Other acute and subacute forms of ischemic heart disease (International classification of diseases, ninth
revision (ICD-9):411), angina pectoris (ICD-9: 413), other forms of chronic ischemic heart disease (ICD-9: 414), other and unspecified intracranial
hemorrhage (ICD-9: 432) or transient cerebral ischemia (ICD-9:435) prior to initiation.
82
83
Figure 1: Flow chart for cohort selection. Initiation of target Biologic

Medicare / MarketScan
N subject=206,275/148,211
N episode=271,832/191,614
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
Excluded due to no paired diagnosis code N subject=9,264/15,831

for rheumatoid arthritis before initiation
N episode=11,340/20,748
N subject=76,048/52,585
N episode=101,740/70,766
Excluded due to diagnosis code for other N subject=110,680/5,499

Auto immune/inflammatory disease in N episode=15,197/7,873
baseline
N subject=65,368/47,086
N episode=86,543/62,893
Excluded due to diagnosis code for HIV N subject=9,169/2,890

Malignancy and organ transplantation N episode=11,894/3,790
N subject=56,199/44,196
N episode=74,649/59,103
Excluded due to diagnosis code for MI, N subject=9,551/2,381

old MI or stroke before initiation N episode=12,934/3,325
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
other cardiovascular disease and data source.
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
86
87
Supplemental table 1: Cardiovascular disease related diagnosis, procedure and

Current Procedure Terminology code used for defining outcome, covariate or for
exclusion.
Code type
Outcome Myocardial infarction 410.x*1 ICD-9-CM diagnosis
Outcome Stroke 430.xx, 431.xx, 433.xx, 434.x1, ICD-9-CM diagnosis
436.xx
Exclusion Myocardial infarction 410.xx ICD-9-CM diagnosis
Exclusion Old myocardial infarction 412.xx ICD-9-CM diagnosis
Exclusion Stroke 430.xx, 431.xx, 433.xx, 434.xx, ICD-9-CM diagnosis
436.xx
Percutaneous coronary 00.66, 36.01-36.09 ICD-9-CM procedure
intervention 92980-92996 Current Procedure
Terminology
Coronary artery bypass grafting 36.10-36.19 ICD-9-CM procedure
33510-33536 Current Procedure
Terminology
Covariate Other acute and subacute forms 411.xx ICD-9-CM diagnosis
of ischemic heart disease
Covariate Angina pectoris 413.xx ICD-9-CM diagnosis
Covariate Other forms of chronic ischemic 414.xx ICD-9-CM diagnosis
heart disease
Covariate Other and unspecified intracranial 432.xx ICD-9-CM diagnosis
hemorrhage
Covariate Transient cerebral ischemia 435.xx ICD-9-CM diagnosis
*X represent any digit or missing.
88
Supplemental table 2: Incidence rates for composite cardiovascular event† by

biologic DMARDS for biologic DMARDS exposed by data source.
Data source Medicare MarketScan
Exposure Number Person Incidence rate Number of Person Incidence rate
of event years (95%CI) event years (95%CI)
Tocilizumab 89 7,244 12.3 (10.0, 15.1) 21 3755 5.6 (3.6, 8.6)
Pooled TNFi 287 17,860 16.1 (14.3, 18.0) 87 14557 6.0 (4.8, 7.4)
Adalimumab 61 4.125 14.8 (11.5, 19.0) 26 4754 5.5 (3.7, 8.0)
Etanercept 45 3.076 14.6 (10.9, 19.6) 16 3026 5.3 (3.2, 8.6)
Infliximab 74 4.096 18.1 (14.4, 22.7) 18 2159 8.3 (5.3, 13.2)
Abatacept 140 10,298 13.6 (11.5. 16.0) 45 6004 7.5 (5.6, 10.0)
Rituximab 74 5,504 14.8 (11.5, 19.0) 27 2554 10.6 (7.2, 15.4)
†
Including incident myocardial infarction, incident stroke and fatal cardiovascular disease. Fatal cardiovascular disease
was identified by published algorithm, cut-points=0.27 for predicted proportion closest to observed proportion.
Supplemental table 3: Incidence rates for composite cardiovascular event by definition for fatal cardiovascular disease
and biologic DMARDS by data source.
Medicare MarketScan
Sensitive definition‡ Equal sensitivity and PPV§ Sensitive definition‡ Equal sensitivity and PPV§
Exposure Person Number Incidence rate Number Incidence rate Person Number Incidence rate Number of Incidence rate
years of event (95%CI) of event (95%CI) years of event (95%CI) event (95%CI)
Tocilizumab 8,045 131 16.3 (13.7, 19.3) 112 13.9 (11.6, 16.8) 4,064 24 5.9 (4.0, 8.8) 23 5.7 (3.8, 8.5)
Pooled TNFi 39,973 767 19.2 (17.9, 20.6) 654 16.4 (15.2, 17.7) 38,102 244 6.4 (5.6, 7.3) 228 6.0 (5.3, 6.8)
Adalimumab 8,634 157 18.2 (15.6, 21.3) 132 15.3 (12.9, 18.1) 12,938 74 5.7 (4.6, 7.2) 71 5.5 (4.3, 6.9)
Etanercept 8,623 146 16.9 (14.4, 19.9) 115 13.3 (11.1, 16.0) 13,553 79 5.8 (4.7, 7.3) 75 5.5 (4.4, 6.9)
Infliximab 13,308 289 21.7 (19.4, 24.4) 246 18.5 (16.3, 20.9) 5,128 47 9.2 (6.9, 12.2) 42 8.2 (6.1, 11.1)
Abatacept 14,561 263 18.1 (16.0, 20.4) 220 15.1 (13.2, 17.2) 7,686 71 9.2 (7.3, 11.7) 69 9.0 (7.1, 11.4)
Rituximab 6,337 161 25.4 (21.8, 29.6) 123 19.4 (16.3, 20.9) 3,190 40 12.5 (9.2, 17.1) 36 11.3 (8.1, 15.6)
PPV: positive predictive value.
Composite cardiovascular events Include incident myocardial infarction, incident stroke and fatal cardiovascular disease, fatal cardiovascular disease was identified by published
algorithm
†
Cut-points=0.54, for PPV≥0.80.
‡
Cut-points=0.15 for sensitivity≥0.80.
§
Cut-points=0.27 for Equal sensitivity and PPV, predicted proportion of event equals observed proportion of event.
89
90
Supplemental table 4: Distribution of baseline RA disease activity as assessed by

the multi-biomarker disease activity (MBDA) prior to initiation of various biologic
DMARDS
MBDA score Abatacept Rituximab Tocilizumab Adalimumab Etanercept Infliximab Pooled
category N=933 N=281 N=678 N=407 N=310 N=466 TNFi†
N=2,264
Low (1-29), % 7.18 2.85 5.75 6.88 7.42 5.79 6.67
Medium or
high (30-100), % 92.82 97.15 94.25 93.12 92.58 94.21 93.33
†
Includes five TNFi therapies (adalimumab, certolizumab, etanercept, golimumab, and infliximab).
Note: RR=ARR/( (Pc1*(RRcd-1)+1)/ (Pc0*(RRcd-1)+1) );RR="True" or fully adjusted exposure relative risk;
ARR=Apparent exposure relative risk (1.10); RRcd=Association between confounder and disease outcome (1.5);
Pc1=Prevalence of confounder in the exposed (0.9258 );Pc0=Prevalence of confounder in the unexposed (0.9425), the
calculated MBDA adjusted HR for comparing etanercept to tocilizumab was: 1.11.
91
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:
92
Supplemental figure 2: Forest plot for adjusted hazard ratio for primary outcome
for biologic DMARDS experienced by data source.
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:
93
Supplemental figure 3: Forest plot for adjusted hazard ratio by definition of fatal
cardiovascular disease and data source.
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

FENGLONG XIE, EMILY B. LEVITAN, HUIFENG YUN, LANG CHEN,

PAUL MUNTNER, JEFFREY R. CURTIS
In preparation for Arthritis care & research
Format adapted for dissertation

96
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
1 Division of Clinical Immunology and Rheumatology, 2 Department of
Epidemiology, University of Alabama at Birmingham, Birmingham, AL, USA
Keywords: methotrexate, biologic initiators, myocardial infarction, stroke, fatal
cardiovascular disease
Word count:
Abstract: 250
Main context: 3,139
Tables: 4
Figures: 1
97
Background: Methotrexate has been associated with reduced risk for
cardiovascular disease (CVD) among rheumatoid arthritis (RA) patients never
exposed to biologic disease-modifying antirheumatic drugs (bDMARDS). Effect of
methotrexate on CVD risk among RA patients initiating bDMARDS remains
unknown.
Methods: A retrospective cohort study was conducted to assess the effect of
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
Poisson regression. Association between methotrexate and risk of CVD were
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
varying methotrexate respectively. The overall multivariable adjusted HR was
0.73 (0.65-0.82) for time varying methotrexate. Interaction between
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
sensitivity and subgroup analyses.
Conclusions: Our observational study suggests an overall 27% reduction of

98
CVD risk associated with time varying methotrexate use. The association
strengths vary among underlying bDMARDS.

99
Rheumatoid arthritis (RA) is a chronic inflammatory polyarthritis affecting about
0.5-2% of the population [1-3]. Without effective treatment, RA often leads to joint
destruction, deformity, loss of function, disability, poor quality of life and
shortened life expectancy [4-10]. Cardiovascular disease (CVD) related deaths
account for more than 50% of the excess mortality in RA patients [11].
Inflammatory features of the disease are characterized by elevated production of
cytokines and inflammatory markers (interleukin 1: IL-1, IL-6, C-reactive protein:
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
promote atherosclerosis [12-14].
Methotrexate is the first line disease-modifying antirhematic drug (DMARDS)
recommended for the treatment for RA [15, 16]. RA patient without adequate
response to methotrexate typically add or switch to biologic DMARDS
(bDMARDS). While some bDMARDS appear to have appreciable efficacy for the
treatment for RA even in the absence of concomitant methotrexate use, other
bDMARDS are clearly more effective when administered with concomitant
methotrexate.
A number of studies and meta-analyses have demonstrated that methotrexate is
associated with reduced risk for CVD [17-26]. Most studies have concluded
mechanistically that this effect is mediated primarily by controlling systemic
inflammation. However, in the setting of other therapies that can also reduce
systemic inflammation (e.g. bDMARDS), the incremental benefit of concomitant
methotrexate on CVD risk is not clear.

100
The objective of this study was to assess the effect of methotrexate on CVD risk
when administered concomitantly with bDMARDS.
METHODS
Population and data source: This study utilized Medicare claims data for 100%
RA patients from January 1, 2006 to September 30, 2015. Medicare is a
government sponsored health plan provided for US residents 65 or older.
Medicare also covers disabled individuals (e.g. due to RA) or with end-stage of
renal disease who are younger than 65 years of age.
Study design: A retrospective cohort study was conducted among RA patients
who initiated a bDMARDS between January 1, 2007 and September 30, 2015.
RA was identified using two International Classification of Disease, Ninth
Revision, Clinical Modification (ICD-9-CM) codes for RA assigned by a physician,
separated by 7 days and occurring within one year. Injection bDMARDS were
identified from Medicare Part D using national drug codes (NDC) and parental
bDMARDS were identified using Healthcare Common Procedure Coding System
(HCPCS) codes in Part B. Infusion drugs (abatacept, certolizumab and
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
PPV 97%)[27]. To be considered as a new user for each bDMARD, RA patients
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
referred as baseline period for assessing co-morbidities. RA patients initiating a
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
malignancy other than non-melanoma skin cancer, HIV, or organ transplantation
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
ICD-9-CM diagnosis codes for other autoimmune/ inflammatory disease
(inflammatory bowel disease, psoriasis, psoriatic arthritis, and ankylosing
spondylitis) in baseline were excluded.
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);
2) switch to other bDMARDS or tofacitinib; 3) outcome; 4) death date (inclusive
for fatal CVD); 5) loss of Medicare coverage 6) end of study (September 30,
2015).
Main exposure: Time varying methotrexate was defined as current exposure
(days of supply without extension). In a sensitivity analysis, a 90 days extension
was added to days of supply of methotrexate.

102
Demographic characteristics, co-morbidities, and health behaviors: Age
was updated at the time of each bDMARD initiation. Co-morbidities and
healthcare utilization were defined using ICD-9-CM diagnosis code in the
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
healthcare utilization were defined dichotomously. Glucocorticoids dose was
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
and defined dichotomously, similarly to methotrexate exposures.
Outcome assessment: The primary outcome was a composite of incident MI,
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
the primary outcome.

103
Statistical methods
Descriptive statistics were calculated for patients with and without methotrexate
use. Mean and standard deviation were calculated for continuous variables and
proportions were calculated for categorical variables. Standard mean differences
(SMD) were calculated for both continuous and categorical variables, SMD
greater than 0.1 was considered as indication for non-balance. Incidence rates
(IRs) were calculated by dividing number of events by person years exposed;
95% confidence intervals (CIs) were estimated with Poisson regression.
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
disease, diabetes, hyperlipidemia, hypertension, obesity, chronic kidney disease,
chronic obstructive pulmonary disease, fibromyalgia, any hospitalized infection,
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
the multivariable adjusted model. Robust sandwich covariance matrix estimate
was used to account for the clustered nature of the data since each patient could
initiate multiple bDMARDS over time. No proportional hazard assumption was
assessed since methotrexate was defined as time varying.
Interaction between time varying methotrexate and back ground bDMARDS was
tested.
104
One variable with 16 levels (e.g. abatacept with methotrexate, abatacept without
methotrexate, adalimumab with methotrexate, …) was created and included in a
multivariable adjusted COX model, the effect of methotrexate on the risk for CVD
for each bDMARDS was calculated with contrast statement in SAS.
A subgroup analyses were performed for patients ever previously exposed to
methotrexate in order to assure that the not current methotrexate exposure group
was at some point deemed well enough to have received methotrexate. This
subgroup would expect to therefore exclude individuals with strong (but
potentially unmeasured) confounding related to factors for which clinicians would
not give patients methotrexate (e.g. alcoholism) that might not be measured well
in health plan data.
A sensitivity analysis was performed for the time-varying methotrexate analysis
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
methotrexate was experiencing outcome in some cases.
Using external adjustment methods, potential confounding for RA disease activity
was examined using the Multi-Biomarker Disease Activity (MBDA) test results
[32], according to the method described by Schneeweiss et.al [33]. The
prevalence of moderate or high RA disease activity (P C1 and P C0 ) by medication
exposure group (i.e. methotrexate vs. no methotrexate) was assessed in a
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
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 the University of Alabama at Birmingham approved
the study.
RESULTS
We identified 475,651 initiations of bDMARDS from 334,092 patients. After
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
(12.3) years, 84.0% were female, 68.2% were non-Hispanic white.
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
infliximab, were previously exposed to methotrexate in the baseline, with folic
acid use, and be bDMARD naïve. Patients with methotrexate use on bDMARDS
initiation date had less co-morbidity.
Main analysis, all bDMARDS
We identified 537 composite CVD events over 44,312-person years with time
varying methotrexate exposure, resulting in an IR of 12.1 (95%CI: 11.1-13.2)
events per 1,000-person years (Table 2). We identified 1,324 composite CVD
106
events over 74,127-person years without time varying methotrexate exposure,
resulting in an IR of 17.9 (95%CI: 16.9-18.8). The unadjusted HR and the multi-
variable adjusted HR for time varying methotrexate exposure was 0.68 (95%CI:
0.61-0.75) and 0.73 (95%CI: 0.65-0.82) respectively.
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-
0.86) for stroke and 0.74 (95%CI: 0.66-0.83) for MI or stroke.
Interaction between methotrexate and background bDMARDS
P-value was 0.0030 for test for interaction between time varying methotrexate
and background bDMARDS.
By bDMARDS, 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)
for infliximab, 0.66 (95%CI: 0.40-1.09) for tocilizumab.
Contrasts for each component of CVD were similar to that for composite CVD
(Table 3).
Subgroup analysis limited to RA patients who used methotrexate prior to initiation
of bDMARDS resulted in similar IR and HR to that from analysis for all RA
patients (Supplementary table 1). Sensitivity analysis for time varying
methotrexate with 90 days extension resulted in similar IR and HR results to that
from analysis for time varying methotrexate with no extension (Supplementary
table 2).
External adjustment for the estimate for hazard ratio
Results from the MBDA lab tests linked to Medicare data are shown in
supplementary table 3. The proportions of RA patients with moderate or high
MBDA score were similar between bDMARDS initiators with and without
methotrexate. The computed true HR for composite CVD time varying
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
estimate of effect of MBDA on MI [32].
DISCUSSION
In this retrospective cohort study of Medicare-insured RA patients who initiated
bDMARDS for the treatment for RA, we observed an approximately 27%

108
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.
Several studies have demonstrated that methotrexate associated with reduced
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
these studies were conducted among RA patients never exposed to bDMARDS.
All these estimates are highly compatible with our aHRs associated with
methotrexate which were approximately 0.73 (95%CI: 0.65-0.82).
Our study has significant clinical implications. While combination with
methotrexate may not meaningfully increase the efficacy of some bDMARDS
(e.g. tocilizumab), combination therapy may further reduce the risk for CVD. An
ongoing randomized clinical trial (CIRT, ClinicalTrials.gov: NCT01594333) plans
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
to detect a 25 percent relative risk reduction. This hypothesized effect size of
methotrexate in this clinical trial is quite close to what we observed in RA patients
treated with bDMARDS. Our study suggests that when tolerability is not a
problem, methotrexate should be recommended to all RA patients treated with
bDMARDS to further reduce the risk for CVD.
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
potential for residual confounding due to unmeasured factors such as RA
disease activity, we used laboratory measure of RA disease activity, the MBDA,
to assess the distribution of disease activity. There was no difference in 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
associations that we observed.
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
exposure. Finally, CVD events were not adjudicated individually, although we
relied on claims-based algorithms that have in excess of 80-90% in their
accuracy, and there is little reason to expect that performance of these would be
differential by specific biologic exposure or methotrexate use.
In conclusion, our observational study suggests an overall 27% reduction of CVD
risk associated with time varying methotrexate use. The strengths of association
vary among background bDMARDS.

111
References
3. Scott DL, Shipley M, Dawson A, Edwards S, Symmons DP, Woolf AD:
The clinical management of rheumatoid arthritis and osteoarthritis:
strategies for improving clinical effectiveness. British journal of
rheumatology 1998, 37(5):546-554.
rheumatology 2007, 21(5):827-840.
5. Riise T, Jacobsen BK, Gran JT, Haga HJ, Arnesen E: Total mortality is
increased in rheumatoid arthritis. A 17-year prospective study. Clinical
rheumatology 2001, 20(2):123-127.
of rheumatology 1995, 22(6):1065-1067.
107(9):1303-1307.
112
93(2):198-200.
44(12):2746-2749.
63(5):1182-1189.
rheumatism 2008, 59(12):1690-1697.
979.

113
108(3):205-213.
15. Singh JA, Saag KG, Bridges SL, Jr., Akl EA, Bannuru RR, Sullivan MC,
Vaysbrot E, McNaughton C, Osani M, Shmerling RH et al: 2015 American
College of Rheumatology Guideline for the Treatment of Rheumatoid
Arthritis. Arthritis & rheumatology 2016, 68(1):1-26.
16. Smolen JS, Landewe R, Bijlsma J, Burmester G, Chatzidionysiou K,
Dougados M, Nam J, Ramiro S, Voshaar M, van Vollenhoven R et al:
EULAR recommendations for the management of rheumatoid arthritis with
synthetic and biological disease-modifying antirheumatic drugs: 2016
update. Annals of the rheumatic diseases 2017, 76(6):960-977.
17. Choi HK, Hernan MA, Seeger JD, Robins JM, Wolfe F: Methotrexate and
mortality in patients with rheumatoid arthritis: a prospective study. Lancet
2002, 359(9313):1173-1177.
18. Prodanovich S, Ma F, Taylor JR, Pezon C, Fasihi T, Kirsner RS:
Methotrexate reduces incidence of vascular diseases in veterans with
psoriasis or rheumatoid arthritis. Journal of the American Academy of
Dermatology 2005, 52(2):262-267.
19. van Halm VP, Nurmohamed MT, Twisk JW, Dijkmans BA, Voskuyl AE:
Disease-modifying antirheumatic drugs are associated with a reduced risk
for cardiovascular disease in patients with rheumatoid arthritis: a case
control study. Arthritis research & therapy 2006, 8(5):R151.

114
20. Suissa S, Bernatsky S, Hudson M: Antirheumatic drug use and the risk of
acute myocardial infarction. Arthritis and rheumatism 2006, 55(4):531-536.
21. Naranjo A, Sokka T, Descalzo MA, Calvo-Alen J, Horslev-Petersen K,
Luukkainen RK, Combe B, Burmester GR, Devlin J, Ferraccioli G et al:
Cardiovascular disease in patients with rheumatoid arthritis: results from
the QUEST-RA study. Arthritis research & therapy 2008, 10(2):R30.
22. Radovits BJ, Popa-Diaconu DA, Popa C, Eijsbouts A, Laan RF, van Riel
PL, Fransen J: Disease activity as a risk factor for myocardial infarction in
rheumatoid arthritis. Annals of the rheumatic diseases 2009, 68(8):1271-
1276.
23. Westlake SL, Colebatch AN, Baird J, Kiely P, Quinn M, Choy E, Ostor AJ,
Edwards CJ: The effect of methotrexate on cardiovascular disease in
patients with rheumatoid arthritis: a systematic literature review.
Rheumatology 2010, 49(2):295-307.
24. Micha R, Imamura F, Wyler von Ballmoos M, Solomon DH, Hernan MA,
Ridker PM, Mozaffarian D: Systematic review and meta-analysis of
methotrexate use and risk of cardiovascular disease. The American
journal of cardiology 2011, 108(9):1362-1370.
25. Almalag HM, Mangoni AA, Crilly MA: Methotrexate and risk of
cardiovascular disease. The American journal of cardiology 2012,
109(9):1383-1384.
26. De Vecchis R, Baldi C, Palmisani L: Protective effects of methotrexate
against ischemic cardiovascular disorders in patients treated for

115
rheumatoid arthritis or psoriasis: novel therapeutic insights coming from a
meta-analysis of the literature data. Anadolu kardiyoloji dergisi : AKD = the
Anatolian journal of cardiology 2015.
27. Curtis JR, Xie F, Chen R, Chen L, Kilgore ML, Lewis JD, Yun H, Zhang J,
Wright NC, Delzell E: Identifying newly approved medications in Medicare
claims data: a case study using tocilizumab. Pharmacoepidemiology and
drug safety 2013, 22(11):1214-1221.
28. Audrey L Fan JNF, R Van Harrison, Elizabeth A Jackson, Marie A
Marcelino.: Screening and Management of Lipids. UMHS Lipid Therapy
Guideline 2014.
29. Kiyota Y, Schneeweiss S, Glynn RJ, Cannuscio CC, Avorn J, Solomon
DH: Accuracy of Medicare claims-based diagnosis of acute myocardial
infarction: estimating positive predictive value on the basis of review of
hospital records. American heart journal 2004, 148(1):99-104.
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-
based stroke algorithms in contemporary Medicare data: reasons for
geographic and racial differences in stroke (REGARDS) study linked with
medicare claims. Circulation Cardiovascular quality and outcomes 2014,
7(4):611-619.
31. Fenglong Xie LDC, Jeffrey R. Curtis, Meredith L. Kilgore, Emily B. Levitan,
Keri L. Monda, Monika M. Safford, Ben Taylor, Mark Woodward, Paul
Muntner.: Development of algorithms for identifying fatal cardiovascular

116
disease in Medicare claims. Pharmacoepidemiology & Drug safety 2017,
submitted.
diseases 2017.
33. Schneeweiss S: Sensitivity analysis and external adjustment for
unmeasured confounders in epidemiologic database studies of
therapeutics. Pharmacoepidemiology and drug safety 2006, 15(5):291-
303.
34. Dixon WG, Watson KD, Lunt M, Hyrich KL, British Society for
Rheumatology Biologics Register Control Centre C, Silman AJ, Symmons
DP, British Society for Rheumatology Biologics R: Reduction in the
incidence of myocardial infarction in patients with rheumatoid arthritis who
respond to anti-tumor necrosis factor alpha therapy: results from the
British Society for Rheumatology Biologics Register. Arthritis and
rheumatism 2007, 56(9):2905-2912.

117
Table 1: Distribution of characteristics by methotrexate use on initiation date.

With MTX† Without MTX† SMD§
N=34,200 N=54,055
Age in years, mean (STD) 65.2 (11.8) 64.1 (12.6) 0.09
Female, % 82.9 84.7 0.05
Race, % 0.04
White 69.4 67.4
Black 9.8 9.8
Other 20.8 22.7
Follow up time in days, mean (STD) 547 (589) 454 (503) 0.17
Biologic initiated, % 0.26
Abatacept 18.3 21.8
Adalimumab 16.3 14.8
Certolizumab 6.6 7.4
Etanercept 15.4 14.7
Golimumab 5.5 5.7
Infliximab 22.9 15.6
Rituximab 8.1 10.7
Tocilizumab 6.8 9.3
Comorbidity, %
†
History of cardiovascular disease 19.8 21.9 0.05
Heart failure 4.7 6.3 0.07
Atrial fibrillation 4.8 5.4 0.03
Abdominal aortic aneurism 0.5 0.6 0.01
Peripheral arterial disease 2.0 2.2 0.02
Diabetes 21.8 23.7 0.05
Hyperlipidemia 43.5 42.9 0.01
Hypertension 60.3 61.1 0.02
Obesity 4.4 5.0 0.03
Chronic kidney disease 7.0 10.6 0.13
Chronic obstructive pulmonary disease 24.0 28.6 0.10
Fibromyalgia 17.8 20.8 0.08
Hospitalized infection 7.4 10.8 0.12
Medication use, %
Baseline Methotrexate 100.0 36.9
Folic acid use‡ 2.4 0.9 0.14
Nonsteroidal anti-inflammatory drug 48.3 45.3 0.06
Prior biologic DMARDS 0.38
0 49.4 39.3
1 32.8 36.7
2 12.4 15.7
3 and more 5.4 8.3
Statin 0.15
None 64.8 67.1
Low potency 5.2 4.6
Moderate potency 24.9 23.2
High potency 5.1 5.1
Other lipid lowering drug 9.7 10.9 0.04
Prednisone equivalent, mg/day 0.06
None 34.8 35.6
<=7.5 49.7 47.9
>7.5 15.4 16.5
Health behavior and healthcare utility
Smoking 12.2 14.2 0.06
Prostate specific antigen (male only) 44.5 42.0 0.05
PAP smear (female only) 14.0 13.6 0.01
Mammography (female only) 42.3 38.9 0.07
Any hospitalization 19.2 23.4 0.10
Number physician visit, mean (STD) 16.6 (9.6) 18.0 (10.5) 0.14
State buy in program (proxy for low income) 37.1 40.1 0.06
Reason other than age eligible for Medicare 48.8 54.9 0.12
DMARDS: Disease-modifying anti-rheumatic drug; MTX: methotrexate; SMD: Standardized mean difference.
†
MTX was defined as MTX exposure at biologic DMARDS initiation date.
‡
Folic acid was defined as folic acid exposure at biologic DMARDS initiation date.
§
SMD >0.1 was considered as potentially important (i.e., a large difference in the distributions between groups).
Table 2: Incidence rate and hazard ratio of composite CVD and its component by associated with methotrexate use.
‡ ‡
With time varying MTX Without time varying MTX
§
Event Person IR Event Person IR Unadjusted HR Adjusted HR
years (95% CI) years (95% CI) (95% CI) (95% CI)
Myocardial infarction, 12.1 17.9 0.68 0.73
stroke or fatal CVD 537 44312 (11.1, 13.2) 1324 74127 (16.9, 18.8) (0.61, 0.75) (0.65, 0.82)
6.5 9.3 0.70 0.73
Myocardial infarction 289 44497 (5.8, 7.3) 694 74526 (8.6, 10.0) (0.61, 0.80) (0.63, 0.86)
5.4 8.0 0.67 0.72
Stroke 239 44555 (4.7, 6.1) 599 74732 (7.4, 8.7) (0.58, 0.78) (0.61, 0.86)
Myocardial infarction 11.6 16.8 0.69 0.74
or stroke 513 44312 (10.6, 12.6) 1242 74127 (15.8, 17.7) (0.62, 0.76) (0.66, 0.83)
CVD: Cardiovascular disease; CI: Confidence interval; MTX: Methotrexate; HR: Hazard ratio; IR: Incidence rate.
†
Concomitant MTX was defined as MTX prescription within 120 days after initiation of biologic disease-modifying anti-rheumatic drugs (DMARDS).
‡
Time varying MTX was defined as days of supply with no extension.
§
Adjusted for age, sex, race, biologic DMARDS initiated, 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.
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)
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)
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
Figure 1: Flow chart for cohort selection.

Initiation of target Biologic
N subject=334,092/242,254
N episode=475,651/334,428
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
Excluded due to no paired diagnosis code N subject=13,801/22,425

for rheumatoid arthritis before initiation
N episode=17,761/31,126
N subject=100,153/70,933
N episode=140,498/100,426
Excluded due to diagnosis code for other N subject=13,157/7,142

Auto immune/inflammatory disease in N episode=19,971/11,014
baseline
N subject=86,996/63,791
N episode=120,527/89,412
Excluded due to diagnosis code for HIV N subject=11,238/3,708

Malignancy and organ transplantation N episode=15,367/5,615
N subject=75,758/60,083
N episode=105,160/84,247
Excluded due to diagnosis code for MI, N subject=11,540/2,875

old MI or stroke before initiation
N episode=16,905/4,414
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
§ §
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)
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.
§
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
† † ‡
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.
†
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
Supplementary table 3: Distribution of baseline RA disease activity as assessed

by the multi-biomarker disease activity (MBDA), by methotrexate use on initiation
date.
MBDA score category With concomitant methotrexate Without concomitant methotrexate
N=2,354 N=2,702
Low (1-29), % 6.04 6.67
Medium, or high (30-100), % 93.96 93.33
RA: Rheumatoid arthritis.
Note: RR=ARR/( (Pc1*(RRcd-1)+1)/ (Pc0*(RRcd-1)+1) );RR="True" or fully adjusted exposure relative risk;
ARR=Apparent exposure relative risk (0.77); RRcd=Association between confounder and disease outcome (1.5);
Pc1=Prevalence of confounder in the exposed (0.9396 );Pc0=Prevalence of confounder in the unexposed (0.9333), the
calculated MBDA adjusted HR for comparing etanercept to tocilizumab was: 0.77.
124
SUMMARY
Cardiovascular disease (CVD) is the leading cause of death in United States.
Rheumatoid arthritis (RA) patients are twice as likely to experience CVD
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;
likewise, treatment having the potential to decrease CVD risk is worth
investigating.
Tocilizumab is a humanized monoclonal antibody against IL-6 receptor and is
approved by the FDA to treat RA. Unfavorable lipid profile changes in RA
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
(ENTRACTE, ClinicalTrials.gov: NCT01331837) compared tocilizumab only to
etanercept but no other bDMARDS. This limitation cannot be overcome by
another clinical trial due to excessive costs and the improbability of recruiting
enough RA patients to compare tocilizumab to all other bDMARDS. 2) The two
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
an outcome may under-estimate the incidence of true CVD-related events. This
could be more problematic in studies involving RA patients since RA patients
experience a higher proportion of sudden coronary heart disease (CHD) death

125
than do non-RA patients. One of these observational studies compared
tocilizumab to a pooled tumor necrosis factor inhibitor (TNFi) exposure group
without evaluating individual TNFi therapies, which could also be problematic,
since each TNFi may have drug specific risks for CVD. Disease activity is
associated with CVD risk. Confounding by RA disease activity was not
addressed in the observational studies.
Tocilizumab is effective for the treatment of RA along or in combination with
methotrexate. Rheumatologists tend not to prescribe methotrexate to RA patients
initiating tocilizumab. Prior studies have shown that methotrexate is associated
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
among RA patients when concomitantly used with bDMARDS.
In this dissertation, we developed claims-based algorithms for identifying fatal
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
discrimination when compared to methods using discharge diagnosis code within
28 days prior to death. The approaches can be adapted by other investigators to
develop claims-based algorithms for other fatal events and fatal events in other
claims data.
With the claims-based algorithms to identify fatal CVD, we conducted a
retrospective cohort study to assess the comparative effect of tocilizumab on the
risk of CVD. In a more economical way, we confirmed results from a large

126
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
associated with a reduction of CVD risk when compared to a pooled TNFi
exposure.
We also conducted a retrospective cohort study to assess the effect of
methotrexate on the risk of CVD and concluded that concomitant methotrexate
use is associated with an overall 27% reduction of CVD risk. The strength of
association varies among concomitant biologic DMARDS. This finding suggests
that adding methotrexate to tocilizumab initiators may reduce the CVD risk
compared to tocilizumab monotherapy. Given that RA patients initiating
tocilizumab together with methotrexate showed little benefit in effectiveness,
methotrexate may have an additional mechanism other than controlling
inflammation. Thus, methotrexate may reduce CVD risk among non-RA patients
with increased CVD risk.
My study also addressed another limitation associated claims data: lack of
disease severity. We conducted external adjustment with laboratory data with
disease severity linked to Medicare claims through multiple non-unique identifiers
for a sub group of the study cohort.

127
GENERAL REFERENCES
3. Scott DL, Shipley M, Dawson A, Edwards S, Symmons DP, Woolf AD:
The clinical management of rheumatoid arthritis and osteoarthritis:
strategies for improving clinical effectiveness. British journal of
rheumatology 1998, 37(5):546-554.
rheumatology 2007, 21(5):827-840.
5. Riise T, Jacobsen BK, Gran JT, Haga HJ, Arnesen E: Total mortality is
increased in rheumatoid arthritis. A 17-year prospective study. Clinical
rheumatology 2001, 20(2):123-127.
of rheumatology 1995, 22(6):1065-1067.

128
107(9):1303-1307.
93(2):198-200.
44(12):2746-2749.
63(5):1182-1189.
11. Okada Y, Wu D, Trynka G, Raj T, Terao C, Ikari K, Kochi Y, Ohmura K,
Suzuki A, Yoshida S et al: Genetics of rheumatoid arthritis contributes to
biology and drug discovery. Nature 2014, 506(7488):376-381.
12. Mikuls TR, Sayles H, Yu F, Levan T, Gould KA, Thiele GM, Conn DL,
Jonas BL, Callahan LF, Smith E et al: Associations of cigarette smoking
with rheumatoid arthritis in African Americans. Arthritis and rheumatism
2010, 62(12):3560-3568.
129
rheumatism 2008, 59(12):1690-1697.
14. del Rincon ID, Williams K, Stern MP, Freeman GL, Escalante A: High
incidence of cardiovascular events in a rheumatoid arthritis cohort not
explained by traditional cardiac risk factors. Arthritis and rheumatism
2001, 44(12):2737-2745.
979.
108(3):205-213.
18. Kremer JM, Genant HK, Moreland LW, Russell AS, Emery P, Abud-
Mendoza C, Szechinski J, Li T, Ge Z, Becker JC et al: Effects of abatacept
in patients with methotrexate-resistant active rheumatoid arthritis: a
randomized trial. Annals of internal medicine 2006, 144(12):865-876.

130
19. Moreland LW, Baumgartner SW, Schiff MH, Tindall EA, Fleischmann RM,
Weaver AL, Ettlinger RE, Cohen S, Koopman WJ, Mohler K et al:
Treatment of rheumatoid arthritis with a recombinant human tumor
necrosis factor receptor (p75)-Fc fusion protein. The New England journal
of medicine 1997, 337(3):141-147.
20. Maini R, St Clair EW, Breedveld F, Furst D, Kalden J, Weisman M,
Smolen J, Emery P, Harriman G, Feldmann M et al: Infliximab (chimeric
anti-tumour necrosis factor alpha monoclonal antibody) versus placebo in
rheumatoid arthritis patients receiving concomitant methotrexate: a
randomised phase III trial. ATTRACT Study Group. Lancet 1999,
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
human anti-tumor necrosis factor alpha monoclonal antibody, for the
treatment of rheumatoid arthritis in patients taking concomitant
methotrexate: the ARMADA trial. Arthritis and rheumatism 2003, 48(1):35-
45.
22. Smolen J, Landewe RB, Mease P, Brzezicki J, Mason D, Luijtens K, van
Vollenhoven RF, Kavanaugh A, Schiff M, Burmester GR et al: Efficacy and
safety of certolizumab pegol plus methotrexate in active rheumatoid
arthritis: the RAPID 2 study. A randomised controlled trial. Annals of the
rheumatic diseases 2009, 68(6):797-804.

131
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
active rheumatoid arthritis after treatment with tumour necrosis factor
alpha inhibitors (GO-AFTER study): a multicentre, randomised, double-
blind, placebo-controlled, phase III trial. Lancet 2009, 374(9685):210-221.
24. Emery P, Fleischmann R, Filipowicz-Sosnowska A, Schechtman J,
Szczepanski L, Kavanaugh A, Racewicz AJ, van Vollenhoven RF, Li NF,
Agarwal S et al: The efficacy and safety of rituximab in patients with active
rheumatoid arthritis despite methotrexate treatment: results of a phase IIB
randomized, double-blind, placebo-controlled, dose-ranging trial. Arthritis
and rheumatism 2006, 54(5):1390-1400.
25. Antoni C, Kalden JR: Combination therapy of the chimeric monoclonal
anti-tumor necrosis factor alpha antibody (infliximab) with methotrexate in
patients with rheumatoid arthritis. Clinical and experimental rheumatology
1999, 17(6 Suppl 18):S73-77.
26. Choi HK, Hernan MA, Seeger JD, Robins JM, Wolfe F: Methotrexate and
mortality in patients with rheumatoid arthritis: a prospective study. Lancet
2002, 359(9313):1173-1177.
27. Prodanovich S, Ma F, Taylor JR, Pezon C, Fasihi T, Kirsner RS:
Methotrexate reduces incidence of vascular diseases in veterans with
psoriasis or rheumatoid arthritis. Journal of the American Academy of
Dermatology 2005, 52(2):262-267.

132
28. van Halm VP, Nurmohamed MT, Twisk JW, Dijkmans BA, Voskuyl AE:
Disease-modifying antirheumatic drugs are associated with a reduced risk
for cardiovascular disease in patients with rheumatoid arthritis: a case
control study. Arthritis research & therapy 2006, 8(5):R151.
29. Suissa S, Bernatsky S, Hudson M: Antirheumatic drug use and the risk of
acute myocardial infarction. Arthritis and rheumatism 2006, 55(4):531-536.
30. Naranjo A, Sokka T, Descalzo MA, Calvo-Alen J, Horslev-Petersen K,
Luukkainen RK, Combe B, Burmester GR, Devlin J, Ferraccioli G et al:
Cardiovascular disease in patients with rheumatoid arthritis: results from
the QUEST-RA study. Arthritis research & therapy 2008, 10(2):R30.
31. Radovits BJ, Popa-Diaconu DA, Popa C, Eijsbouts A, Laan RF, van Riel
PL, Fransen J: Disease activity as a risk factor for myocardial infarction in
rheumatoid arthritis. Annals of the rheumatic diseases 2009, 68(8):1271-
1276.
32. Westlake SL, Colebatch AN, Baird J, Kiely P, Quinn M, Choy E, Ostor AJ,
Edwards CJ: The effect of methotrexate on cardiovascular disease in
patients with rheumatoid arthritis: a systematic literature review.
Rheumatology 2010, 49(2):295-307.
33. Micha R, Imamura F, Wyler von Ballmoos M, Solomon DH, Hernan MA,
Ridker PM, Mozaffarian D: Systematic review and meta-analysis of
methotrexate use and risk of cardiovascular disease. The American
journal of cardiology 2011, 108(9):1362-1370.

133
34. Almalag HM, Mangoni AA, Crilly MA: Methotrexate and risk of
cardiovascular disease. The American journal of cardiology 2012,
109(9):1383-1384.
35. De Vecchis R, Baldi C, Palmisani L: Protective effects of methotrexate
against ischemic cardiovascular disorders in patients treated for
rheumatoid arthritis or psoriasis: novel therapeutic insights coming from a
meta-analysis of the literature data. Anadolu kardiyoloji dergisi : AKD = the
Anatolian journal of cardiology 2015.
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
cardiovascular disease in patients with rheumatoid arthritis: a systematic
literature review. Rheumatology 2011, 50(3):518-531.
37. Sheppard M, Laskou F, Stapleton PP, Hadavi S, Dasgupta B: Tocilizumab
(Actemra). Human vaccines & immunotherapeutics 2017:0.
38. Maini RN, Taylor PC, Szechinski J, Pavelka K, Broll J, Balint G, Emery P,
Raemen F, Petersen J, Smolen J et al: Double-blind randomized
controlled clinical trial of the interleukin-6 receptor antagonist, tocilizumab,
in European patients with rheumatoid arthritis who had an incomplete
response to methotrexate. Arthritis and rheumatism 2006, 54(9):2817-
2829.
39. Kawashiri SY, Kawakami A, Yamasaki S, Imazato T, Iwamoto N, Fujikawa
K, Aramaki T, Tamai M, Nakamura H, Ida H et al: Effects of the anti-
interleukin-6 receptor antibody, tocilizumab, on serum lipid levels in

134
patients with rheumatoid arthritis. Rheumatology international 2011,
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
changes and decreased hepatic LDL receptor expression by tocilizumab
in rheumatoid arthritis. Atherosclerosis 2013, 229(1):174-181.
41. Gabay C, McInnes IB, Kavanaugh A, Tuckwell K, Klearman M, Pulley J,
Sattar N: Comparison of lipid and lipid-associated cardiovascular risk
marker changes after treatment with tocilizumab or adalimumab in
patients with rheumatoid arthritis. Annals of the rheumatic diseases 2016,
75(10):1806-1812.
42. al JTGe: Comparative Cardiovascular Safety of Tocilizumab Vs
Etanercept in Rheumatoid Arthritis: Results of a Randomized, Parallel-
Group, Multicenter, Noninferiority, Phase 4 Clinical Trial. ACR abstract
2016, late-Breaking Abstract Poster Session(3L).
43. Kim SC, Solomon DH, Rogers JR, Gale S, Klearman M, Sarsour K,
Schneeweiss S: Cardiovascular Safety of Tocilizumab Versus Tumor
Necrosis Factor Inhibitors in Patients With Rheumatoid Arthritis: A Multi-
Database Cohort Study. Arthritis & rheumatology 2017, 69(6):1154-1164.
of the rheumatic diseases 2016.

135
45. Maradit-Kremers H, Crowson CS, Nicola PJ, Ballman KV, Roger VL,
Jacobsen SJ, Gabriel SE: Increased unrecognized coronary heart disease
and sudden deaths in rheumatoid arthritis: a population-based cohort
study. Arthritis and rheumatism 2005, 52(2):402-411.
diseases 2017.
47. Administration on Aging: Administration for Community Living. Peofile of
older Americans: 2013, Washington, DC: US Department of Health and
Human Services; 2014.
48. Nonell L: Total Number of Medicare Beneficiaries. Timeframe: 2015 2017.
49. Heron M: National Vital Statistics Report. 2015, 64(10).
50. Marks JL, Edwards CJ: Protective effect of methotrexate in patients with
rheumatoid arthritis and cardiovascular comorbidity. Therapeutic
advances in musculoskeletal disease 2012, 4(3):149-157.
51. Hjeltnes G, Hollan I, Forre O, Wiik A, Lyberg T, Mikkelsen K, Agewall S:
Relations of serum COMP to cardiovascular risk factors and endothelial
function in patients with rheumatoid arthritis treated with methotrexate and
TNF-alpha inhibitors. The Journal of rheumatology 2012, 39(7):1341-
1347.
136
APPENDIX A
IRB APPROVAL
137

Assessing The Comparative Effect of Tocilizumab

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Assessing The Comparative Effect of Tocilizumab

Uploaded by

Copyright:

Available Formats

ASSESSING THE COMPARATIVE EFFECT OF TOCILIZUMAB ON RISK OF

CARDIOVASCULAR DISEASE AMONG RHEUMATOID ARTHRITIS

PAUL MUNTNER, COMMITTEE CHAIR

Submitted to the graduate faculty of The University of Alabama at Birmingham,

Tocilizumab is a humanized monoclonal anti-body against the interleukin-6

studies have observed unfavorable changes in the lipid profile of tocilizumab-

treated RA patients. Few studies have compared the cardiovascular disease

(CVD) risk associated with tocilizumab to other biologic disease-modifying

antirheumatic drugs (bDMARDS). Due to limitations in existing studies, the real-

world association of tocilizumab with CVD risk remains uncertain.

We conducted a retrospective cohort study using 2006-2015 Medicare and

MarketScan claims data to assess the comparative effect of tocilizumab on CVD

risk. Medicare claims data provide a unique opportunity to estimate disease

claims data lack information on cause of death. To address this limitation, we

developed claims-based algorithms for identifying fatal CVD in Medicare claims

fatal CVD events with high sensitivity and specificity.

with increased or decreased CVD risk compared to abatacept or rituximab. We

CVD risk associated with infliximab.

We also conducted a retrospective cohort study using Medicare claims data to

methotrexate compared to tocilizumab only was 0.66 (95%CI: 0.40-1.09).

KEYWORDS: tocilizumab, biologic DMARDS, CVD risk, Medicare claims,

me while working to finish my Ph.D. Specifically, I want to thank my mentor, Dr.

Additionally, I want to thank the other members of my graduate committee, Dr.

criticism concerning my research project.

Furthermore, I want to express thanks to my colleagues, Dr. Lang Chen, Shuo

always with me.

ABSTRACT ................................................................................................................................. iii

LIST OF TABLES ..................................................................................................................... ix

LIST OF FIGURES .................................................................................................................. xii

Rheumatoid arthritis ..................................................................................................... 1

Rheumatoid arthritis and the risk for cardiovascular disease ......................... 1

Treatment for rheumatoid arthritis ........................................................................... 2

Tocilizumab and the risk for cardiovascular disease ......................................... 3

Clinical trial and observational study ...................................................................... 5

DEVELOPMENT OF ALGORITHMS FOR IDENTIFYING FATAL

TOCILIZUMAB AND THE RISK FOR CARDIOVASCULAR DISEASE: A DIRECT

METHOTREXATE USE AND THE RISK FOR CARDIOVASCULAR DISEASE

SUMMARY .............................................................................................................................. 124

APPENDIX A: IRB approval ................................................... ………………………………………..136

DEVELOPMENT OF ALGORITHMS FOR IDENTIFYING FATAL

1 Definitions of fatal events in the REGARDS study…………………………………………31

2 Distribution of pre-specified candidate predictors for participants with and

3 Performance of the direct and indirect approaches for discriminating fatal

4 Performance of algorithms for discriminating cause of death under

S1 Diagnosis, procedure and current procedure terminology codes used as

S2 Beta-coefficients from logistic regression models for predictor variables for

S3 Net reclassification index for the claims-based algorithm developed in the

S4 Net reclassification index for the claims-based algorithm developed in the

S5 Net reclassification index for the claims-based algorithm developed in the

S7 Example of the calculation of the predicted probability for having a fatal

S8 Example of the calculation of the predicted probability for having a fatal

TOCILIZUMAB AND THE RISK FOR CARDIOVASCULAR DISEASE: A DIRECT

1a Distribution of baseline characteristics by biologic DMARDS for Medicare

1b Distribution of characteristics by biologic DMARDS for MarketScan….……80

2 Incidence rates for composite cardiovascular event and its components

3 Incidence rates for composite cardiovascular event by History of other

S1 Cardiovascular disease related diagnosis, procedure and Current

S2 Incidence rates for composite cardiovascular event† by biologic DMARDS

S3 Incidence rates for composite cardiovascular event by definition for fatal

S4 Distribution of baseline RA disease activity as assessed by the multi-

1 Distribution of characteristics by methotrexate use on initiation date........117

S3 Distribution of baseline RA disease activity as assessed by the multi-

DEVELOPMENT OF ALGORITHMS FOR IDENTIFYING FATAL