137
369

 

THS

l
3160‘s
#30077?

 

LIBRARY
Michigan State
University

 

 

 

This is to certify that the

l
thesis entitled ‘

Selective Cyclooxygenase-Z (COX-2) Inhibition
and Risk of Acute Myocardial Infarction or
Stroke in a Population of Rheumatoid Arthritis and
Osteoarthritis Patients From an Administrative

Medical grégénfiegaogaset

William M. Spalding

has been accepted towards fulfillment
of the requirements for

Master' 5 degree in Epidemiology

Mae? 0
Major professor

Date May 28, 2003

MS U i: an Affirmative Action/Equal Opportunity Institution

 

 

PLACE IN RETURN BOX to remove this checkout from your record.
To AVOID FINES return on or before date due.
MAY BE RECALLED with earlier due date if requested.

 

DATE DUE DATE DUE DATE DUE

 

45%

 

M v

 

 

 

 

 

 

 

 

 

 

 

 

 

6/01 cJCIRC/DateDuepSS—p. 1 5

SELECTIVE CYCLOOXYGENASE-Z (COX-2) INHIBITION AND RISK OF
ACUTE MYOCARDIAL INFARCTION OR STROKE IN A POPULATION OF
RHEUMATOID ARTHRITIS AND OSTEOARTHRITIS PATIENTS FROM AN

ADMINISTRATIVE MEDICAL CLAIMS DATASET

By

William Michael Spalding

A THESIS
Submitted to
Michigan State University
In partial fulfillment of the requirements
For the degree of
MASTER OF SCIENCE
Department of Epidemiology

2003

U .IIIIEDI’II § I’ ‘3} I? ,I I

ABSTRACT
SELECTIVE CYCLOOXYGENASE-2 (COX-2) INHIBITION AND RISK OF
ACUTE MYOCARDIAL INFARCTION OR STROKE IN A POPULATION OF
RHEUMATOID ARTHRITIS AND OSTEOARTHRITIS PATIENTS FROM AN
ADMINISTRATIVE MEDICAL CLAIMS DATASET
By
William M. Spalding

It has been hypothesized that treatment with selective COX-2 inhibitors
(coxibs) creates a prothrombotic state by decreasing the synthesis of the
prostaglandin prostacyclin, which has vasodilatory and platelet anti-aggregation
effects, leading to adverse cardiovascular (CV) events.

This thesis was based on an analysis of a population based, retrospective
cohort study of NSAID use in an adult RA or OA population generated from
administrative medical and pharmacy claims data from a large private insurer.
We compared incidence rates for CV events in subjects treated with coxibs,
NSAle, combination users (coxibs and NSAIDs), and non-users of NSAIDs
utilizing multivariate Cox proportional hazards models.

Relative to non-users of any NSAID, users of coxibs were found to be at
greater risk for a CV event than were subjects treated with nonselective NSAle,
with the effect most pronounced in treated hypertensive subjects (HR=2.30, 95%
CI 2.24-2.37 and HR=1.69, 95% CI 1.29-2.20, respectively). Among coxib treated
subjects, those treated with rofecoxib were two times more likely to experience a

CV event than were celecoxib treated subjects (HR=2.77, 95% CI 1.36-5.60 and

HR=1.44, 95% CI 1.09-1.79, respectively).

To Patti, Katelyn, and Alexander, for your continuous encouragement and

suppon.

ACKNOWLEDGMENTS

I would like to acknowledge the following people for their significant contributions
to this work. First and foremost, I’d like to acknowledge my graduate advisor Dr.
Mat Reeves, who made a positive impact on this work through his always
insightful suggestions and critiques throughout all phases of the project, and for
the hours spent reviewing and editing of this manuscript. Sandy Suh and Steve
Zlotnick are acknowledged for their assistance with the literature search, and
clinical interpretation of results. I’d like to thank Dr. Charles Barr for his
continuous support and medical guidance. Lastly, I’d like to thank Lora McAdams

for her assistance in the negotiation of all the logistical hurdles in completing this

thesis.

TABLE OF CONTENTS

LIST OF TABLES
INTRODUCTION

METHODS
Study Design
Cohort Definition
Drug Exposure Definition
Outcome Definition
Confounder Definition
Statistical Analysis

RESULTS
Primary Analysis
Non-selective NSAID subgroup
Selective COX-2 Subgroup
DISCUSSION
TABLES

REFERENCES

vi

10
10
11
11
12
12
13

15
17
20
21
23
33

41

Table 1.

Table 2.

Table 3.

Table 4.

Table 5.

Table 6.

Table 7.

Table 8.

Table 9.

LIST OF TABLES

Description of COX-2 randomized control trials included in
Mukherjee meta-analysis

Characteristics of aspirin primary prevention trials included in
Mukherjee meta-analysis

Covariates assessed in the analysis as potential confounders
of the relationship between treatment and cardiovascular
events. Univariate Cox models.

Covariate groupings assessed as potential confounders of the
treatment relationship to cardiovascular events. Univariate Cox
Models

Characteristics of NSAID treatment cohorts

Primary Analysis: Relationship between NSAID use and CV
events using the main effects Cox Proportional Hazards model
— Adjusted Hazard Ratios + 95% Cl

Primary Analysis: Relationship between NSAID use and CV
events using the Cox Proportional Hazards interaction model —
Adjusted Hazard Ratios + 95% Cl

Secondary Analysis: Relationship between naproxen and non-
naproxen NSAID use and CV events relative to non-users in
the Cox Proportional Hazards model — Adjusted Hazard Ratios
+ 95% Cl

Secondary Analysis: Secondary analysis: Relationship
between Celecoxib and Rofecoxib and CV

Events using the Cox Proportional Hazards Model - Adjusted
Hazard Ratios + 95% Cl

vi

33

34

35

36

37

38

39

40

41

INTRODUCTION

Nonsteroidal anti-inflammatory drugs (NSAIDs) have been used ever
since ancient Egyptians learned that salicylate-containing plants were efficacious
in the treatment of rheumatism. [1,2] The first commercially synthesized and
marketed NSAID, aspirin, was introduced in 1899. Since then, aspirin and other
compounds in the NSAID class of products have become one of the most
commonly prescribed pharmacological agents worldwide, with more than 100
million prescriptions filled annually in the United States, United Kingdom, and
Canada alone. [3] NSAIDs are efficacious in the treatment of inflammation and
pain, and are commonly prescribed for treatment of rheumatoid arthritis (RA),
and osteoarthritis (OA), chronic inflammatory conditions of the joints.

Although the efficacy of NSAIDs for the treatment of OA and RA is well
documented [4-9.44], they can be associated with adverse gastrointestinal (GI)
side effects. [10-13] It has been hypothesized that GI related side effects are
attributable to the inhibition of the cyclooxygenase (COX) —1 enzyme, the
enzyme that catalyzes prostaglandins that protect the gastric mucosa. [14] It is
estimated that 10%-20% of NSAID treated patients experience dyspepsia, and
that between 50,000 and 100,000 hospitalizations annually are attributable to
NSAID induced ulceration of the GI tract. Approximately 16,000 deaths annually
can be attributed to NSAID related complications. [14,15]

In addition to its role in the protection of the gastric mucosa, the
expression of the COX enzyme plays numerous housekeeping functions

responsible for maintenance of physiologic homeostasis in other body systems,

including the kidney where disturbances in renal function can lead to the
development or exacerbation of peripheral edema and hypertension. COX
expression also plays a role in the physiologic response to tissue injury and
inflammation, and is involved in the mediation of the pain signal originating in the
nociceptors of damaged tissue. [2,16,17] There are two known isoforms of the
COX enzyme. The COX-1 isoforrn is ubiquitous and constitutively expressed in
all normal tissue cells, and is the isofonn primarily involved in physiologic
homeostasis. The COX-2 isofonn is inducible, and its expression is primarily
promoted as a result of pathophysiologic processes. When tissue becomes
damaged and inflamed, the COX-2 isoforrn is expressed in response to the
release of inflammatory mediators such as cytokines, mitogens and growth
factors, and catalyzes the synthesis of prostaglandins that regulate their function.
[2,16,17]

The prostaglandins thromboxane-Az (TxA2) and prostacyclin (PGI2) are
involved in the regulation of platelet-vessel wall interactions. TXA2 elicits platelet
activation, vasoconstriction, and smooth muscle proliferation. PGI2 opposes
TXA2 and elicits vasodilatory and anti-aggregatory effects [18]. COX-1 and COX-
2 are involved in the synthesis of TXA2 and PGI2, respectively. Because of the
antagonistic relationship of TXA2 and PGI2, it has been theorized that selective
inhibition of the COX-2 enzyme might produce a net prothrombotic state by
creating an imbalance between TXA2 and PGI2 synthesis [19], resulting in an

overall increase in serious thromboembolic cardiovascular events, including

myocardial infarction and stroke. This increased risk may be particularly evident
in patients with an already elevated CVD risk.

Recently, several publications have questioned the cardiovascular safety
of the COX-2 inhibitor class of NSAIDs. [18-23] The primary focus of these
publications has been the cardiovascular safety results of the Vioxx
Gastrointestinal Outcomes Research Study (VIGOR), a 52-week randomized
control trial (RCT) designed to assess the GI safety of the selective COX-2
inhibitor rofecoxib compared to the non-selective NSAID naproxen in 8 076
patients diagnosed with rheumatoid arthritis (Table 1). The VIGOR trial revealed
that patients treated with high-dose rofecoxib experienced a significantly higher
incidence rate of serious thrombotic cardiovascular events compared to those
treated with naproxen (1.1% vs. 0.5%, respectively). [24] Rofecoxib treated
patients also experienced a significantly higher incidence of myocardial
infarctions (0.4% vs. 0.1%, respectively), although the mortality rates due to
adverse cardiovascular events were identical in both treatment groups (0.2%).
[24] During a CV safety review presented to the FDA, the trial’s sponsor
hypothesized that the NSAID comparator naproxen has cardioprotective
properties similar to aspirin, and that the observed differences in serious
thrombotic events of the VIGOR trial were due to these aspirin-like
cardioprotective properties. [25]

In contrast to the results observed in VIGOR, the Celecoxib Long Term
Arthritis Safety Study trial (CLASS) did not demonstrate a similar increase in CV

risk. CLASS was a 52-week, randomized, double-blind study of the GI safety of

celecoxib compared to the NSAID comparators ibuprofen and diclofenac in a
population of 8 059 patients with a diagnosis of 0A or RA. [28] In the CLASS
trial, a serious CV event was defined as an event that was considered life-
threatening, required hospitalization, or was otherwise medically significant.
None of the CV safety comparisons was statistically significant. F iffy-two events
(1.3%) were observed in celecoxib treated patients, and 49 events (1.2%) in the
combined NSAID arm (RR=1.1, 95% CI 0.7-1.6). Nineteen (0.5%) total AMls
were observed in the celecoxib cohort compared to 13 (0.3%) in the NSAID
cohort (RR=1.46, 95% CI 0.72-2.95, p=0.29). Seven (0.2%) of the celecoxib
AMls were fatal compared to 2 (0.05%) AMI deaths in the NSAID arm (RR=3.49,
95% CI 0.73-16.81, p=0.10). NSAID treated patients experienced a statistically
significant greater number of cerebrovascular events compared to those treated
with celecoxib, with 12 (0.3%) events versus 4 (0.1%) respectively (RR=3.0, 95%
CI 0.97-9.31, p=0.045). [26,27]

A meta-analysis (published by Mukherjee et al [19]) reviewed the
cardiovascular safety results of 4 randomized clinical trials that included VIGOR,
CLASS, and two smaller rofecoxib placebo controlled RCTs (studies 085 and
090). Study 085 was as 6-week randomized, double-blind, placebo-controlled
efficacy and safety study of rofecoxib compared to the NSAID nabumetone in 1
042 patients diagnosed with OA of the knee. Study 090 replicated the design of
study 085, and enrolled 978 patients. Studies 085 and 090 were not published
however, data were made available through the FDA website. In the rofecoxib

085 study, 3 CV events were observed: 1 (0.2%) event in the rofecoxib arm

versus 2 (0.4%) for those treated with nabumetone, and no events observed in
the placebo-treated patients. Comparisons between rofecoxib and nabumetone
were not statistically significant although the estimate was highly imprecise
(RR=0.48, 95% CI 0.04-5.31, p=0.54). In the rofecoxib 090 study, 6 (1.5%)
serious CV events were observed among rofecoxib treated patients, 2 (0.5%)
among those treated with nabumetone, and 1 (0.5%) in the placebo arm.
Comparisons of serious CV incidence rates between rofecoxib and nabumetone
were not statistically significant although again the estimate was highly imprecise
(RR=3.01, 95% Cl 061-1485, p=0.15).

Mukherjee’s primary analysis was a comparison of the CV safety results of
VIGOR and CLASS relative to the results observed from the combined placebo
arms from 4 randomized, long-term (> 4 years) trials that explored the
cardioprotective benefit of aspirin (Table 2). From this analysis, Mukherjee
reported that the annualized AMI incidence rate of the placebo arms from the
combined aspirin trials was lower than that observed in the rofecoxib treated
patients in VIGOR (0.52% vs. 0.74%, =0.04) and celecoxib treated patients in
CLASS (0.52% vs. 0.80%, p=0.02). Comparisons of the NSAID treated patients
to the placebo group were not reported. From these findings, Mukherjee
concluded that long-term exposure to selective COX-2 inhibitors increases risk
for a thrombotic cardiovascular event.

Two retrospective cohort studies utilizing administrative medical claims
data have been recently published [29,30] that wholly or partially oppose the

conclusions of Mukherjee. These studies are important in that their findings

support the CV safety of the COX-2 agents as used in large representative
populations.

Ray et al evaluated risk of acute myocardial infarction (AMI) or death from
coronary heart disease (CHD) in patients enrolled in the Tennessee Medicare
(TennCare) health plan between January 1, 1999 and June 30, 2001. [30]
Patients included in the analysis were between the ages of 50 and 84, must have
been eligible for TennCare benefits for 365 days continuously, were not in a
nursing home, and had no history of non-cardiovascular life-threatening illness.
Over 378,000 subjects satisfied the inclusion criteria. Subjects were classified
into one of 6 NSAID treatment groups: a) non-user, b) ibuprofen, c) naproxen, d)
celecoxib, e) rofecoxib s 25 mg, and f) rofecoxib > 25 mg. The mean age of all
subjects was 61.5 years, however users of celecoxib and rofecoxib s 25 mg
tended to be slightly older than the other comparator groups. The population was
predominantly female, with users of celecoxib and rofecoxib more likely to be
female than non-users. The primary endpoint for the Ray for this analysis was
hospital admission for AMI, or CHD related death as coded on death certificates.
Ray found no increased risk for AMI or CHD death in users of celecoxib
(RR=0.96, 95% CI 0.76-1.21, p=0.19) or low dose rofecoxib (<= 25 mg)
(RR=1.03, 95% CI = 0.78-1.35) compared to a non-treated cohort. However,
patients exposed to 50 mg doses of rofecoxib experienced a 70% increase in
risk. Although these results were not statistically significant, (RR=1], 95% CI

0.98-2.95, p=0.056) the point estimate was reasonably precise, and the data

suggests that a threshold effect might exist. A similar analysis of the dose-
response for celecoxib was not reported.

Mamdani et al evaluated the CV risk of selective COX-2 inhibitors utilizing
administrative health care data from the period of April 1, 1998 through March
31, 2001 for residents of Ontario, Canada. [30] The analysis included only
subjects 66 years or older. All elderly residents of Ontario, Canada have
universal coverage for prescription drugs, hospital, and primary care, and claims
data for more than 1.44 million subjects over the age of 66 was available for
analysis. Celecoxib and rofecoxib both became available through the drug
benefit program in April, 2000, but were limited to patients that either failed
traditional NSAID therapy, had demonstrated an intolerance for NSAIDs, or had a
history of upper GI hemorrhage or ulcer. Celecoxib was only approved for use in
osteoarthritis and rheumatoid arthritis, and rofecoxib was approved for
osteoarthritis only. No similar restrictions were place for NSAIDS. Approximately
167 000 subjects were included in the analysis, of which 100 000 were randomly
selected non-users of any NSAID agent. Users of NSAID agents were assigned
to one of 4 treatments based on prescription history: a) celecoxib, b) rofecoxib),
c) naproxen, and d) non-naproxen nonselective NSAIDs. The mean age of the
study population was 75.7 years, and was similar across cohorts. The study
population was predominantly female, and users of celecoxib and rofecoxib were
more likely to be female than either non-users or users of NSAIDs (70%, 71%,
56%, and 62% respectively). The primary endpoint of the analysis was admission

to hospital for AMI. Relative to non-treated controls, Mamdani found no evidence

that users of celecoxib or rofecoxib were at increased risk for an AMI (RR=0.9,
95% CI 0.7-1.2 and RR=1.0, 95% CI 0.8-1.4, respectively). As a secondary
analysis, Mamdani also tested the hypothesis that naproxen is cardioprotective.
Results of this analysis demonstrated that users of naproxen had a similar
incidence rate for AMI as non-treated controls (RR=1.0, 95% CI 0.6-1.7).

Given the variability of findings related to the safety of selective COX-2
inhibitors, no definitive conclusions can be drawn regarding the risk of adverse
CV events in users of these agents. However, because of their potential clinical
benefit in patients at risk for an upper GI hemorrhage or ulcer, we chose to
initiate this study to further add to the evidence regarding exposure to COX-2
inhibitors and CV risk so that the tradeoffs between CV and GI risk can be better
understood. This paper reports on the findings of a retrospective cohort analysis
exploring the relationship between exposure to selective COX-2 inhibitors and
the risk for an acute myocardial infarction or stroke requiring hospitalization in a
population of more than 45 000 adult osteoathritis and rheumatoid arthritis
patients from a database obtained from a large Northeastern United States
insurer. The primary objective of this study was to compare the risk of
cardiovascular events between 4 treatment groups: a) user of selective COX-2
inhibitors, b) users of nonselective NSAIDs, c) users of COX-2 inhibitors and
nonselective NSAIDs in combination, and d) non-users in a population of arthritis
patients. Secondary objectives were to evaluate the cardioprotective properties

of naproxen relative to other non-naproxen NSAIDs, and to compare the risk of

adverse cardiovascular events between celecoxib and rofecoxib, the two

predominant forms of selective COX-2 inhibitors.

METHODS
Design

This study was designed as a population based retrospective cohort study
using data from over 3 million subjects aged 18 years or older from the
northeastern United States, who were enrolled in a private medical insurance
plan managed by a large regional insurer. The period covered by this analysis
was January 1, 1999 through June 30, 2001, and roughly coincides with the
product launches of the selective COX-2 agents Celebrex (celecoxib) and Vioxx
(rofecoxib), and with the time period in which these products were made
available on the formulary of the insurer.

These data are comprised of over 95 million medical and 32 million
pharmacy claim records. Medical claims data consisted of both outpatient visits
and hospital inpatient stays. Outpatient claims include visits to a primary care
physician, urgent care facilities, and hospital emergency department not resulting
in an admission. Enrollment, medical and pharmacy claim records were linked
using an encrypted patient identifier. Patient identifiers were encrypted by the
data vendor to maintain patient confidentiality. Subjects are primarily employed
individuals and their dependents. The mean age of all enrolled individuals is 37.5
years, with a median age of approximately 40 years. Fifty-two percent of the

enrolled population was female.

Cohort Definition

All patients with a previous diagnosis of rheumatoid arthritis or
osteoarthritis that were at least 18 years of age were eligible for inclusion in this
study. Arthritis status was determined by review of all available outpatient claims
records, and patients with at least one claim with a primary or secondary
International Classification of Diseases, Ninth Revision ( ICD-9-CM) diagnosis
code in the range of 714 — 715 were identified as eligible for inclusion. Although
nonsteroidal anti-inflammatory drugs are used in other inflammatory conditions,
this population was selected because the COX-2 class of drugs is primarily
indicated for use in an OA and RA population, and because this population would
likely require regular and long term exposure to NSAID treatment for relief of
inflammation and pain. [41, 42]

For all subjects, continuous enrollment in a medical insurance plan with
drug coverage was required. A break in coverage of less than 45 days was
considered as having had continuous coverage. Subjects that were continuously
enrolled but discontinued drug coverage were only eligible during the period in

which drug coverage was available.

Drug Exposure Definition
For the primary analysis, subjects were assigned to one of 4 treatment
cohorts; COX-2 only, NSAID only, Combination of COX-2 and NSAID, and non-

users. A non-user was deemed as being untreated if no pharmacy claims were

10

observed for a COX-2 inhibitor or any non-selective NSAID. The COX-2 cohort
consisted of subjects dispensed celecoxib, rofecoxib, or both. The NSAID cohort
included subjects dispensed naproxen, ibuprofen, diclofenac, or any combination
of those agents. Subjects that filed overlapping claims for both NSAID and COX-
2 were assigned to the combination cohort unless one therapy exceeded 90% of
the exposure, in which case they were assigned to that exposure cohort. In order
to eliminate sporadic users, treated subjects were required to be persistent in
their use of therapy. Persistency was defined as having been dispensed at least
2 consecutive prescriptions for the treatment to which the subject was assigned.
If a break occurred between the calculated end date of the first prescription and
the dispense date of the subsequent prescription, a 30 day grace period was
applied. Subjects that changed therapy were assigned to the cohort in which they
were last observed if they met the criteria for persistency described above, and
the observation period began as of the date of the switch. Subjects that failed to
meet the persistency criteria following the switch in therapy were dropped from
the analysis. For the treated cohorts, the duration of exposure was defined as the
period of time in which subjects were first dispensed a prescription for the study
drug to which they were assigned, ending when the last prescription would have
been exhausted based on intended duration of therapy. The dispense date of the
first prescription was established as the index date of observation. Data on the
number of days of supply was used as an indicator of the intended duration of
therapy, and was added to the dispense date to derive the expected ending date

of each prescription. If the next prescription was dispensed prior to the ending

11

date of the previous prescription, the excess supply was carried over to the next
period and added to that prescription’s days supply estimate. Subjects that failed
to refill their prescription by the end of the 30-day grace period were deemed to
have discontinued therapy, and their observation period ended. For the
untreated subjects, the index date was established as the earlier of January 1,
1999, or the date at which drug coverage became available if drug coverage
began after January 1, 1999. Observation ended when one of the following
events occurred; a) admission to hospital with a CV event, b) discontinued
therapy with the study drug to which they were assigned, or reached the end of

the observation period of June 30, 2001.

Outcome Definition

Subjects were classified as having a major CV event if they were admitted
to the hospital with an acute myocardial infarction or stroke. Events were
identified by reviewing medical claims for hospital admission that contained a
primary lCD-9-CM code in the range of 410-414 (AMI) or 430-436 (Stroke) during
the observation period. Subjects that may have had more than one event during
the study period were only included once in the analysis, and the date of the
event was based on the first event occurrence. The date of the event was

assigned as the date of admission to the hospital.

Confounder Definition

12

Prior CVD or known CV risk factors were regarded as potential
confounder of the primary exposure-outcome relationship. Potential confounders
were identified using lCD-9-CM codes listed as primary or secondary diagnoses
in any outpatient medical claim occurring prior to the CV event. Diagnosis codes
were defined as in Table 3. The classification of previous stroke and previous
AMI were treated as exceptions. To avoid misclassifying a current outcome
disease process as a previous event, outpatient claims with a diagnosis code for
AMI or stroke that were observed within 90 days of an admission for one of these
two conditions were identified as being part of the disease process that led to the

hospital admission.

Statistical Analysis

Time-to-event analysis methods [31] were used to estimate the hazard
ratios for a CV event for the four treatment cohorts. Relative risks were
calculated using Cox proportional hazards models, with the untreated controls as
the referent group. Since all risk estimates are relative to the untreated controls,
linear contrasts were used to make statistical comparisons between each of the
active treatment groups. The proportional hazards assumption was assessed for
each univariate model, and for the final main effects models by plotting the
scaled Schoenfeld residuals for each covariate included in the model.

The relationships between potential confounders and the risk of a CV
event were assessed using univariate proportional hazards models. All

covariates found to be significant predictors of a CV event at the p s 0.05 level

13

were assessed in the multivariate modeling process (Table 3). With the exception
of age, which was allowed to remain as a continuous variable, all confounders
were coded as dichotomous design variables.

Because of the large number of statistically significant covariates that
were identified as potential confounders during the univariate analysis phase, we
elected to simplify the model by combining covariates into clinically similar
groupings as shown in Table 4. These grouped variables were assessed using
univariate analysis, and treated in the same manner as the other covariates.

Creation of design variables for hypertension, hypertension treatment,
hyperlipidema, and hyperlipidemia treatment required a 3-Ievel design since
inclusion of the treatment variable is dependent upon the recognition of the
disease. Hence, the logical relationship between disease and treatment was
modeled as no disease, disease with no treatment, and disease with treatment
(Table 4.).

Proportional hazards models were fit using backward selection methods.
Covariates that were not statistically significant at the p < 0.05 level in the model
were removed one at a time, and the model refit with the remaining covariates.
This process was continued until all remaining variables were statistically
significant. Age and gender were allowed to remain in the models regardless of
their contribution to the final model. Models were also fit using forward selection
methods and compared to the final backward selection models. Forward

selection resulted in the same models as backward selection.

14

Each confounder remaining in the main effects model was assessed for
treatment interactions by stratifying the analysis on the confounding variable. If
marked changes in the estimate of treatment effects were observed after
stratification on the confounder, an interaction term between the confounder and
treatment was created and tested in the model. Interaction terms that were
significant at the p s 0.05 level were retained in the final model after testing
validity of the proportional hazards assumption. All analyses were performed
using SAS for Windows, version 8.2 (SAS Institute Inc., Cary NC) and Stata/SE

version 8.0 (Stata Corp, College Station TX).

RESULTS

Of the approximately 3 million adult subjects for which medical and
pharmacy data were available for analysis, 40 666 subjects satisfied the study
inclusion criteria for continuous enrollment and a diagnosis of OA (88%) or RA
(12%). The mean age of the study cohort was 54.25 years, and was 61% female.
The most common cardiovascular risk factors were hypertension (40.2%), and
hyperlipidemia (34.2%). Seven percent of the cohort were identified as having
had a previous AMI, and approximately 2% had a previous stroke. Twenty-five
percent of the cohort were treated for hypertension, and 19% treated for
hyperlipidemia at some time during the study period.

From these subjects, we identified 32 087 (79%) as being treated with
nonsteroidal anti—inflammatory agents, and 8 579 non-treated controls for whom

no NSAID pharmacy claims wee observed during the 30-month study period.

15

Among the treated cohort, we identified 15 950 users of non-selective NSAIDs, 9
608 users of selective COX-2 inhibitors, and 6 529 individuals whose prescription
pattern indicated combination therapy (Table 5).

Descriptively, significant differences were observed in baseline
characteristics between cohorts. Relative to the total study population, users of
selective COX-2 inhibitors were more likely to be older (mean age 57.6 years vs.
54.2 years), and female (67% vs. 61%). Users of selective COX-2 inhibitors only
were consistently more likely to have comorbidities at baseline that predispose
subjects to risk for CV events, although some observed differences were minor
(Table 5). Compared to the study population, the COX-2 only group were more
likely to be hypertensive (46.2% vs. 40.2%), and to have been medically treated
with anti-hypertensive agents (31.1% vs. 25.2%). Other CV risk factor differences
observed were presence of hyperlipidemia (37.7% vs. 28.4%), previous serious
thrombotic CV event, including AMI (8.8% vs. 6.9%) and ischemic stroke (1.8%
vs. 1.3%). Subjects in the NSAID arm tended to be slightly younger than average
(mean age 52.0 years vs. 54.2), and less likely to be female (57.7% vs. 60.7%)
relative to the study cohort. In contrast to what was observed in the COX-2 only
group, NSAID only treated subjects were consistently less likely to have major
CV risk comorbidities at baseline, even though some of the differences were
small. Compared to the study cohort, these subjects were less likely to be
hypertensive (34.7% vs. 40.2%), medically treated for hypertension (21.7% vs.
25.2%), and had fewer thrombotic CV events, including AMI (4.6% vs. 6.9%) and

ischemic stroke (0.8% vs. 1.3%). In direct comparisons between the COX-2 only

16

group and the NSAID only group, the COX-2 only group was more likely to have
other severe coronary heart disease comorbidities, including congestive heart
failure (4.1% vs. 1.6%), and cardiomyopathy (1.1% vs. 0.6%). In general, the
characteristics of the combination therapy group tended to be similar to those of
the COX-2 only group. No differences were observed in the arthritis status
between groups. The average length of follow-up was 739 days for the untreated
controls, 518 days for combination therapy, 277 days for COX-2 only and 243

days for NSAID only (Table 5).

Primary Analysis

For the entire study cohort, 44 516 person-years of follow-up and 1 111
CV events were observed (Table 6). The COX-2 only cohort had the highest
unadjusted crude event rate, with 34.6 events per 1000 person-years, followed
by combination therapy (27.2 events/1000 person-years), untreated controls
(25.7 events/1000 person-years) and NSAID only (15.1 events/1000 person-
years). Relative to untreated controls, subjects treated with selective COX-2
inhibitors, both in the COX-2 only and combination therapy groups, were more
likely to have experienced an event after adjusting for age, gender, and presence
of CV risk factors (Table 6). The greatest risk was apparent in subjects taking
combination therapy, with a 55% increase in the hazard ratio relative to untreated
controls, and was statistically significant (HR=1.55, 95% Cl=1.31-1.83). The
observed risk increase in the COX-2 was similar to that observed in the

combination group (HR=1.51, 95% CI 1.28-1.78). The effects of treatment on the

17

NSAID only group tended to be neutral, with a hazard ratio of approximately 1.0
relative to untreated controls (HR=1.04, 95% CI 0.86-1.26).

All covariates that remained in the main effects model had statistically
significant association with the risk of a CV event. The largest effect was
observed in subjects that had a coronary artery bypass graft surgery during the
study (HR=3.6, 95% CI 1.76-7.29), although this estimate was imprecise. Large
and highly precise point estimates were calculated for history of coronary heart
disease, including previous AMI (HR=2.59, 95% CI 2.22-3.01), previous stroke
(HR=1.77, 95% CI 1.47-2.13), and other miscellaneous heart disease (HR=1.59,
95% CI 1.36-1.84). Treated hypertensive subjects had a 55% increase in CV risk
compared to non-hypertensives (HR=1.55, 95% CI 1.34-1.79) however, risk in
the non-treated hypertensives was equivalent to that of non-hypertensives
(HR=0.98, 95% CI 0.82-1.18).

A statistically significant interaction was identified between treatment
group and hypertension status (p=0.01), and the analysis expanded to
accommodate exploration of this interaction (Table 7). Inclusion of this term into
the model modified the referent group to be non-users of NSAIDS that are also
non-hypertensive.

Among non-users of NSAIDs, there was a 36% increase in the risk for a
CV event in the treated hypertension group compared to the referent no
hypertension group (HR=1.36, 95% CI 1.09-1.70). There was no observed
change in risk in non-users of NSAIDs that were hypertensive but not treated

(HR=0.97, 95% CI 0.75-1.27) (Table 7). Among users of COX-2 only, a similar

18

slight increase in risk was observed in both the non-hypertensives (HR=1.33,
95% CI 0.99-1.80, p=0.052) and hypertensive without treatment subgroups
(HR=1.27, 95% CI 0.86-1.87, p=0.22) relative to the non-hypertension non-user
referent group. Neither estimate was statistically significant although the risk
estimate for the COX-2 non-hypertensives was very close to being significant. A
greater than two-fold increase in risk was observed among COX-2 only users
that were also being treated for hypertension relative to the non-hypertensive
non-user referent group, and was highly statistically significant (HR=2.30, 95% Cl
224-237, p<0.0001).

Non-hypertensive users of non-selective NSAIDs had a slightly
decreased risk for a CV event compared to the non-hypertensive non-user
referent group, and was borderline statistically significant (HR=0.71, 95% CI
0.50-1.00, p=0.052). The risk estimate in the non-treated hypertensives was
nearly identical to that observed in the COX-2 only group (HR=1.26, 95% CI
0.86-1.86, p=0.99). There was a statistically significant 69% increase in the risk
of a CV event among the NSAID treated subjects that were also being treated for
hypertension although the increase in risk was not as pronounced as that
observed in the COX-2 only group (HR=1.69, 95% CI 1.29-2.20, p=0.02).

The only statistically significant risk estimate among non-hypertensive
subjects occurred in the combination therapy group, where a 67% increase in
risk was observed (HR=1.67, 95% CI 1.28-2.19, p=0.0002). Despite the
difference observed in the non-hypertension group compared to the other

treatments, the combination therapy group tended to be more similar to the other

19

treatments in the non-treated hypertensives subgroup (HR=1.09, 95% Cl 0.72-
1.65, p=0.608). Among the treated hypertension subgroup, the risk estimates of
the combination therapy group were similar to that observed in the COX-2 only
group, and was highly statistically significant (HR=2.23, 95% CI 1.76-2.82,

p<0.0001).

Non-selective NSAID Subgroup: Naproxen vs. Non-naproxen vs. Non-
treated

Subjects included in this subgroup analysis were selected from the NSAID
cohort included in the primary analysis. Subjects were categorized into either a
naproxen group that had prescriptions for only naproxen during the observation
period, or a non-naproxen group that consisted of subjects dispensed only
diclofenac or ibuprofen. The risk of a CV event was then compared to the non-
user control population used in the primary analysis.

The average length of follow-up for both the naproxen and non-naproxen
treatment groups was 168 days (Table 8). Demographic and medical history
characteristics were also similar between treatments (data not reported).
Comparisons between treatment groups were made adjusting for age, gender,
and pre-existing CV risk factors (previous AMI, previous stroke, diabetes, prior
non-CV heart disease, and hypertension status). There were no observed
statistically significant differences in CV risk compared to non-users of NSAIDs
for either of the naproxen or non-naproxen cohorts (HR=0.92, 95% CI 0.61-1.64

and HR=1.02, 95% CI 0.69-1.50 respectiveIY). nor was there any differences

20

observed between the two NSAID treatment groups when tested using linear
contrasts (p=0.71) (Table 8).

Second-order interactions between the treatments and included covariates
were assessed, with a moderately significant interaction detected between
treatment and hypertension status (p<0.05). Proportional hazards models were fit
with the treatment and hypertension status interaction included, and impact on
the conclusions above assessed. In all subgroups, no differences were observed
between either of the treatment groups and controls, or between the two
treatments that would change the conclusions drawn from the data presented

above and so, these data were excluded from this analysis.

COX-2 Subgroup: Celecoxib vs. Rofecoxib

Eligible subjects for this subgroup analysis were drawn from the selective
COX-2 inhibitor cohort included in the primary analysis, and for which no other
NSAID prescriptions were observed. Subjects were categorized into either a
celecoxib group or a rofecoxib group. Subjects that switched between celecoxib
and rofecoxib therapy were excluded from the analysis.

The average length of follow-up for the celecoxib treated subjects was 245
days compared to an average follow-up of 174 days for subjects treated with
rofecoxib, (Table 9) and demographic and medical history characteristics were
similar between treatments (data not reported). Hazard ratio estimates were
calculated using proportional hazards models with age, gender, previous AMI,

previous stroke, diabetes, and previous non-CV heart disease as covariates.

21

Rofecoxib treated subjects were used as the referent group. In a direct
comparison between all celecoxib and rofecoxib treated subjects, a 27% lower
risk for a CV event in the celecoxib group was observed (HR=0.73, 95% Cl 0.52-
1.02), which was borderline significant.

The second-order interaction between treatment and hypertension status
was again assessed and found to be statistically significant (p=0.015), and to
have clinically meaningful differences. Consequently, this term was included into
the final model, and the results presented according to the 3 treatment
subgroups; a) non-hypertensives, b) hypertensive but not treated, and c)
hypertensive with treatment. The referent group in this analysis was rofecoxib
treated subjects that were not hypertensive.

In non-hypertensive subjects, a slightly higher risk for a CV event was
observed in celecoxib treated subjects relative to rofecoxib treated subjects, but
this finding was not statistically significant (HR=1.44, 95% CI 0.710294,
p=0.312). Estimates of relative risk were similar for both celecoxib and rofecoxib
treated subjects that were hypertensive but not treated medically, and were not
statistically significant (HR=1.09, 95% CI 0.78-1.53 and HR=1.01, 95% Cl 0.34-
2.98, respectively). Among treated hypertensive subjects, a 44% increase in risk
for a CV event was observed in the celecoxib treated subjects relative to non-
hypertensive rofecoxib treated subjects, and was statistically significant
(HR=1.44, 95% CI = 1.12-1.85). Among users of rofecoxib, subjects that were
treated hypertensives were 2.77 times more likely to have a CV event than those

subjects that were not hypertensive (HR=2.77, 95% Cl 136-56). A direct

22

comparison of risk between celecoxib and rofecoxib in treated hypertensives
revealed a highly statistically significant difference in risk between treatments in

this high-risk subgroup (p=0.002).

DISCUSSION

This study had several important findings that could impact how selective
COX-2 inhibitors and non-selective NSAIDS are used in clinical practice. First, it
appears that patients already at high risk for a CV event are channeled towards
treatment with a COX-2 inhibitor, and away from non-selective NSAIDs. Zhao
has previously described this channeling effect. [40] This finding is particularly
relevant with regards to patients that are hypertensive or have experienced a CV
event for which treatment with angiotensin converting enzyme inhibitors (ACEI),
Beta-Blockers, or loop diuretics are indicated. The mechanism of action of these
antihypertensive agents are dependent upon the production of vascular
prostaglandins and therefore, their efficacy may be blunted when used in
combination with NSAIDs that inhibit vascular prostaglandin synthesis. [32, 43]
Since the disruption of the mechanism of action of these antihypertensive agents
by NSAIDs is well known, and since these drug interaction effects had not been
previously reported in the COX-2 inhibitors, it is reasonable to assume that
physicians treating patients with antihypertensives would have preferentially
selected COX-2 inhibitors over traditional NSAIDs when therapy with
nonsteroidal anti-inflammatory agents was necessitated, which likely explains the

observed differences in CV risk factors at baseline between NSAID treated and

23

COX-2 treated subjects. Recently however, two randomized double-blind control
trials comparing the treatment effects of celecoxib and rofecoxib on systolic and
diastolic blood pressure have demonstrated that a potential drug interaction
resulting in blood pressure destabilization may occur in some subjects treated
with a selective COX-2 inhibitor and anti-hypertensive agents, particularly ACE
inhibitors and Beta-blockers, and that the effect was most pronounced in subjects
in the rofecoxib arm. [31,32] The most recent package insert (Pl) filed with the
FDA requires a warning of a potential interaction between rofecoxib and ACEI
that is not similarly required in the PI for celecoxib. [41,42]

The main effects model in our analysis revealed a statistically significant
51% increase in risk for a CV event in COX-2 users versus non-treated control
subjects, even after adjustment for baseline risk factors. A similar statistically
significant increase of 55% was observed in the combination treatment cohort.
Conversely, the NSAID only group had a risk ratio that was essential 1.0 versus
non-users of NSAIDs. These findings support the hypothesis of increased CV
risk associated with use of selective COX-2 inhibitors proposed by Mukherjee.
The finding that the combination treated group had a similar risk ratio as the
COX-2 only group was particularly interesting. It was expected that the incidence
of CV events in this group would fall somewhere in-between that of the NSAID
only and COX-2 only groups. That the combination treated group had a similar
incidence rate as that observed in the COX-2 only group suggests that any

exposure to COX-2 inhibitors elevates risk, even if not used exclusively.

24

Our analysis based on the interaction model revealed that the association
between COX-2 use and CV events is primarily driven by the increase in risk
observed in treated hypertensive subjects. Relative to non-treated controls, a
33% and 27% increase in the incidence of CV events in users of COX-2
inhibitors was observed in non-hypertensives and hypertensives not being
treated respectively however, these differences were not statistically significant.
A similar increase of 26% was observed among the non-treated hypertensives
using nonselective NSAIDs. Given these similar findings, it appears that
hypertensives users of any NSAID-type agent may have some increased risk of a
CV event. Given that a similar increase in risk was not observed in non-users of
NSAIDs, it is likely that this effect is real, even though the results were not
significant statistically. A greater than 2-fold increase in CV events was observed
in the treated hypertensives using any COX-2 inhibitors, and a 69% increase in
risk was observed in NSAID users. Using the 36% increase in CV events
observed in the non-users as a reference, users of any NSAID appear to be at
increased risk for a CV event if they are also being treated for hypertension,
although the magnitude of effect appears to be largest in the COX-2 and combo
using groups. However, some of the differences observed between the COX-2
group and the NSAID group are likely attributable to the baseline risk differences
and so, the differences observed may have been partially confounded, and are
likely to be smaller than what was reported here.

To further explore the risk of CV events in users of COX-2 inhibitors, we

elected to conduct a subgroup analysis in subjects that were treated exclusively

25

with either rofecoxib or celecoxib to determine whether this appeared to be a
class effect, or whether the observed risk was related to a specific compound. In
the subgroup analysis, no statistically significant differences were observed
between these two compounds in non-hypertensive and hypertensive patients
not on anti-hypertensive treatment, although point estimates of risk tended to
favor rofecoxib. In the treated hypertension subgroup however, a nearly 3-fold
increase in the incidence of CV events was observed in the rofecoxib group
relative to non-hypertensive users of rofecoxib, and was highly statistically
significant. A statistically significant 44% increase in CV events was also
observed in users of celecoxib however, assuming that as much as 36% of this
increase may be attributable to being hypertensive and on treatment, the
increased risk associated with celecoxib appears to be marginal. Conversely, the
magnitude of effect observed in the rofecoxib users indicates that an interaction
between rofecoxib and anti-hypertensive treatment may be occurring that either;
a) negates the treatment effects of the anti-hypertensives, or b) interacts in a
fashion that promotes a CV event. This finding is consistent with two double-blind
randomized control studies by Whelton that demonstrated a destabilization of
blood pressure in rofecoxib treated subjects that were also treated with ACEI and
B-blockers that was not observed in similarly treated celecoxib treated subjects.
[32,33] The results of this study and those of the randomized control trials
support the hypothesis that increased risk for CV events among selective COX-2
users in populations already at high-risk are compound related, and are not a

class effect.

26

To assess the claim by Merck and Company that differences observed in
AMI incidence in the VIGOR trial was due to the cardioprotective properties of
naproxen, we chose to conduct a subgroup analysis comparing CV event rates in
naproxen only users, non-naproxen NSAID users, and non-users of any NSAID
agent. Since no significant interactions were discovered between treatments and
hypertension status, no stratification of the analysis was done by hypertension
status, resulting in three relevant comparisons; a) naproxen versus non-users of
NSAIDs, b) non-naproxen NSAID users versus non-users, and c) naproxen
versus non-naproxen NSAID users.

Using crude rates as measure, there would appear to be some
cardioprotective benefit associated with use of naproxen versus controls
(25.7/1000 person-years and 13.9/1000 person-years respectively). However,
the non-naproxen users had similar results, with a crude rate of 15.9/1000
person-years. These differences did not hold when the model was adjusted for
baseline risk factors. After adjusting for age, gender and CV-related
comorbidities, the incidence rate observed in both naproxen and non-naproxen
NSAID users were comparable to that observed in non-users (p=0.68 and p=0.71
respectively). Based on these findings, it is not likely that the results observed in
the VIGOR trial can be explained by cardioprotective benefit of the naproxen
comparator.

The analysis by Mukherjee (JAMA, 2001) linking COX-2 inhibitors to CV
thromboembolic risk was the pivotal study that initiated the debate regarding the

CV safety of these agents. The primary analysis of the Mukherjee study is a

27

comparison of myocardial infarction incidence rates for the selective COX-2
treated subjects in VIGOR and CLASS versus the placebo arm of a meta-
analysis of aspirin primary prevention trials. In this analysis, Mukherjee reports
that the COX-2 agents rofecoxib and celecoxib had annualized incidence rates
for myocardial infarction that exceeded the rate for the placebo treated arm in the
meta-analysis (0.74% and 0.80% versus 0.52% respectively), and that these
results were statistically significant for both comparisons (p < 0.05).

Several authors have published critical reviews of the Mukherjee meta-
analysis [34-38], and these publications have raised several valid issues pointing
to methodological flaws committed by Mukherjee that are primarily the result of
invalid assumptions. First, in making comparisons across studies, an assumption
is made that the populations studied are homogenous, classification of events
are similar, and that methods of outcome ascertainment would reasonably
identify the same number of events. These assumptions did not hold in the
Mukherjee meta-analysis. The 4 combined aspirin primary prevention
randomized control trials (RCT) were heterogeneous with respect to the base
populations from which participants were drawn (Table 2). For instance, 3 of the
trials enrolled only male subjects while the fourth was 47% female. One trial
enrolled exclusively subjects diagnosed with systemic hypertension whereas in
the 3 remaining trials, prevalence of hypertension was between 9% and 26%.
Similar differences were observed in the prevalence of smokers between the 4
trials. With regards to outcome, Strand identified a 4-fold difference between the

lowest and highest observed crude incidence rates for cardiovascular events

28

within the placebo arms of the 4 trials, and recommended use of a mean MI
incidence rate reported per 100 patient-years rather than the weighted mean
incidence rate reported by Mukherjee, which would have changed the placebo MI
incidence rate estimate from 0.52% to 0.77%, comparable rates to that observed
in CLASS and VIGOR. With regards to baseline characteristics for comparisons
of the 4 aspirin RCTs to CLASS and VIGOR, the 4 aspirin RCTs were
predominantly male compared to a primarily female population in CLASS and
VIGOR (31% and 20% respectively), and tended to be younger. Significant
variations in CV risk factors were also observed between the 4 aspirin RCTs,
CLASS, and VIGOR.

Because of the findings of Mukherjee, Ray and Mamdani explored the CV
risk of COX-2 inhibitors as experienced in clinical practice in population based
studies. The Mamdani study did not report any evidence to support the
Mukherjee hypothesis however, this study included only subjects aged 66 years
or older, thus the study population was about 10 years older than the average
age of patients likely to begin NSAID therapy, and therefore may not be
representative of the typical NSAID using population. Given the older age of the
Mamdani population, it is also very likely that a higher incidence of CV events will
be observed regardless of treatment status, potentially biasing results towards
the null hypothesis of no difference reported by Mamdani. The Ray analysis
selected a population that was more generalizable to the typical NSAID using
population with regards to age however, allowing subjects who switched

therapies to be observed within multiple treatment groups complicated the

29

analysis. Although switching of therapy is a common phenomenon in
retrospective analyses of NSAID usage, and it is appropriate to model therapy
switchers, it is unclear from a biological or mechanistic perspective when the
risks from the new treatment begin to override the effects associated with the
previous treatment. In the Ray design, the lag and induction periods are not well
defined, and it appears that exposure risks for each treatment begin and end at
the time of the switch, potentially resulting in misattribution of events to
treatments, although this risk is likely to be small. The Ray analysis also did not
appear to exclude sporadic users of NSAID agents, and these subjects may bias
the results toward the null hypothesis of no effect since it is unlikely to observe
an event in short-term exposure periods.

This study had several limitations that deserve mention. This study was
conducted utilizing administrative data primarily used for reimbursement.
Consequently, use of lCD-9-CM coding for the assignment of risk factors and
study endpoints may have led to misclassification of some patients. However, it
is likely that misclassification is random and similar across cohorts, resulting in
any potential bias towards the null hypothesis. These data also do not provide
information on other significant potential confounding effects, such as smoking
status, use of cardioprotective low-dose aspirin in high-risk subjects, and use of
over-the-counter (OTC) NSAID products. The lack of data regarding use of OTC
NSAIDs is particularly important as this could have led to a misclassification of
exposure, particularly in subjects identified as non-users of NSAIDs, resulting in

biased estimates of CV risk in this cohort towards the null hypothesis. Assuming

3O

that need for regular treatment with NSAIDs would have resulted in a
prescription, we believe that the potential bias introduced into this analysis is
small as OTC NSAID use would likely be sporadic. With regards to exposure, an
assumption was made that study drugs were used regularly over time, and not
used sporadically, and that all drug was consumed. It is possible, that exposure
time for each drug was overestimated. Lastly, although we attempted to control
for confounding through the inclusion of CV related covariates, there were
substantial differences in characteristics observed between cohorts at baseline
that likely resulted from channeling of subjects with CV risk into the COX-2
treatment group. The channeling of at-risk subjects for CV events towards one of
the treatment arms leads to confounding by indication, biasing results such that a
greater difference in effects may be observed than would otherwise be expected
had patients in both treatment arms had a similar risk profile. It is very likely that
our final models did not entirely adjust for confounding by indication and that
likewise, our risk estimate differences were biased as a result.

In summary, this study addressed some of the weaknesses identified
previously in the Ray and Mamdani studies by including all adults in an OA and
RA population who were persistent users of NSAID agents, and by including a
novel combined treatment group associated with dual use of selective COX-2
inhibitors and non-selective NSAIDS. Our analysis showed that arthritic subjects
requiring NSAID therapy that are at high risk for CV events tend to be channeled
towards treatment with a selective COX-2 inhibitor and away from non-selective

NSAIDs. There is evidence of increased risk for a CV event in users of COX-2

31

inhibitors relative to non-selective NSAIDs however, the risks appear to be
predominantly associated with use of rofecoxib in treated hypertensive subjects,
a finding that was consistent with at least one double-blind randomized control
trial. Lastly, this study failed to demonstrate that naproxen has any
cardioprotective benefits that are similar to those observed with low-dose aspirin,
and that the differences in AMI rates observed in the VIGOR trial are related to

exposure to high-dose rofecoxib.

32

Table 1. Description of COX-2 Randomized Control Trials included In Mukherjee Meta-

 

 

analysis
VIGOR CLASS Study 085 Study 090
N (ITT) 8056 7968 1042 978
COX-2 Rofecoxib Celecoxib Rofecoxib Rofecoxib
NSAID comparator Naproxen Ibuprofen Nabumetone Nabumetone
chlofenac

Mean duration of 9 months 9 months 6 weeks 6 weeks
follow-up

OA/RA % 0/100 72/28 100/0 100/0
Male % 20 31 32 30
Mean Age 58.2 60.1 63.1 62.7
Smoker % 20 16 NR NR
Hypertension % 29 39 42 41
Diabetes % 8 9 NR NR
CV aspirin use % 0 22 11.9 12.2

 

 

 

 

 

 

33

 

Table 2. Characteristics of aspirin primary prevention trials included in Mukherjee meta-

 

 

analysis
US Physician’s UK Doctor’s Thrombosis Hypertension
Health Study Study Prevention Trial Optimal
Treatment Trial

N 2,207 5,139 2,540 18,790
Mean duration of 5.0 years 6.0 years 6.8 years 3.8 years
follow-up

Male % 100 100 100 53
Mean Age 75 47 NR NR
Hypertension % 9 10 26 100
Smoker % 11 13 41 16
Diabetes % 2 2 NR 8

 

 

 

 

 

 

NR=Not Reported

34

 

Table 3. Covariates assessed In the analysis as potential confounders of the relationship
between treatment and cardiovascular events. Univariate Cox models.

 

 

 

Covariate“ Hazard Ratio1 95% CI Codlng:

Procedures

Previous CABG 18.56 9.26 - 37.21 33500-33506
33510-33516
33533-33545

Previous PTCA 8.87 3.32 — 23.67 35450-35460

Diagnosis

Previous cerebrovascular disease 8.12 6.66 - 9.89 436-437

Previous hemorrhagic stroke 6.92 4.01 — 11.96 430-432

Congestive heart failure 6.58 5.66 - 7.65 428

Previous AMI 6.47 5.70 - 7.34 410-414

Atherosclerosis 5.12 4.19 — 6.26 440

Previous lschemic Stroke 4.71 3.70 - 6.00 433-434

Previous transient ischemic attack 4.56 3.41 — 6.10 435

Ill-defined conditions of the heart 4.28 3.63 - 5.05 429

Pulmonary heart disease 3.95 2.87 - 5.44 415-416

Cardiomyopathy 3.24 2.31 - 4.56 425

Hyperlipidemia 0.80 0.70 - 0.91 272

Obesity 0.71 0.53 - 0.95 278

Diabetes 2.33 2.04 - 2.66 250

Family history of CVD 1.91 1.08 — 3.38 V17

Hypertension 1.84 1.63 - 2.07 401-405

Drug Treatment3

Treatment with antihyperlipidemics 2.80 2.49 - 3.16

Treatment with calcium channel 2.69 2.26 —- 3.20

blockers

Treatment with diuretics 2.66 2.34 - 3.01

Treatment with ACE II receptor blocker 2.50 2.05 - 3.05

Treatment with Beta Blockers 2.38 2.06 - 2.74

Treatment with ACEI 2.26 1.92 - 2.66

Demographic

Gender (female as referent) 1.68 1.49 - 1.89

Age (5 year increments) 1.41 1.38 - 1.44

 

 

 

 

 

Hazard Ratio for CV event given presence of risk factor versus no risk factor
2 Diagnosis coded using ICD-9-CM, procedures coded using CPT-4
3 Drugs identified by generic compound and brand
* All covariates were significant at p<0.05 level

35

Table 4. Covariate groupings assessed as potential confounders of the relationship
between treatment and cardiovascular events. Univariate Cox models.

 

 

Grouped Covariate Included covariates Hazard 95% Cl
Ratio
Hypertension Treatment2 ACEI 2.15 1.82-2.54

ACE ll Receptor Blocker
Calcium Channel Blocker
Beta Blocker

Diuretic
Prior Miscellaneous Ill-defined conditions of the heart 5.04 4.42-5.75
Coronary Heart Disease Congestive heart failure

Pulmonary heart disease

Prior Stroke Prior ischemic stroke 5.75 4.84-6.83
Prior hemorrhagic stroke

Prior transient ischemic attack
Prior cerebrovascular disease

 

 

 

 

Hypertension Status3 No hypertension 1.0 Reference
Hypertension with no treatment 1.61 1.35-1.93
Hypertension with treatment 3.47 3.03-3.97
Hyperlipidemia Status4 No hyperlipidemia 1.0 Reference
Hyperlipidemia without treatment 0.45 0.37-0.56
Hyperlipidemia with treatment 1.19 1.02-1.38

 

1 Hazard Ratio for CV event given presence of risk factor versus no risk factor

2 Group variable assessed with hypertension included in the model.

3 The 2 level hypertension treatment and hypertension design variables were combined into a
single design variable containing 3 levels: No hypertension, hypertension with no treatment,
hypertension with treatment.

4 The 2 level hyperlipidemia treatment and hyperlipidemia design variables were combined into a
single design variable containing 3 levels: No hyperlipidemia, hyperlipidemia but no treatment,
hyperlipidemia with treatment.

36

 

Table 5. Characteristics of NSAID Treatment Groups

 

 

 

 

 

 

 

 

 

 

 

 

Study Cohorts
Non- COX-2 NSAID Combo Totals
users

No. of subjects (% Female) 8579 9608 15950 6529 40666
Age

Median 54 57 52 54 54

Mean 54.5 57.6 52.0 54.5 54.2

SD 14.1 12.3 12.4 11.4 12.8

Range 18—101 18-99 18—103 18—97 18-103
Female % 56.1 66.9 57.7 65.2 60.7
Rheumatoid Arthritis % 11.3 11.5 13.1 12.0 11.9
Osteoarthritis % 88.7 88.5 86.9 88.0 88.1
CV risk factors %
Diabetes 12.7 14.0 10.3 14.1 12.3
Hypertension 38.6 46.2 34.7 47.1 40.2
Atherosclerosis 2.1 2.2 1.1 1.9 1.7
Pulmonary heart disease 0.9 1.0 0.4 1.0 0.75
Ill-defined conditions of heart 3.5 4.4 2.1 4.2 3.27
Cardiomyopathy 1.0 1.1 0.6 1.0 0.8
Congestive heart failure 3.5 4.1 1.6 3.7 2.9
Obesity 4.1 5.1 4.0 6.3 4.7
Hyperlipidemia 34.6 37.7 29.3 40.4 28.4
Previous AMI 11.1 8.8 4.6 4.2 6.9
Previous hemorrhagic stroke 0.3 0.2 0.1 0.1 0.2
Previous TIA 1.8 1.3 0.4 0.4 0.9
Previous ischemic stroke 2.4 1.8 0.8 0.6 1.3
Previous cerebrovascular disease 2.5 1.7 0.7 0.5 1.3
Previous PTCA 0.03 0.05 0.03 0.03 0.04
Previous CABG 0.07 0.04 0.0 0.05 0.03
Family history of CVD 0.6 0.5 0.7 0.8 0.6
Treatment of CVD %
Any antihypertensive 21.0 31.1 21.7 30.9 25.2
ACEI 5.9 7.9 5.8 8.2 6.7
ACE II receptor blocker 3.1 5.4 2.9 5.4 3.9
Beta Blockers 9.1 12.0 9.1 11.7 10.2
Calcium Channel Blockers 4.2 7.0 3.7 6.1 5.0
Diuretics 11.6 18.5 11.8 19.1 14.5
Any antihyperlipidemia 16.1 23.0 15.8 23.4 18.8
Grouped Factors
Prior non-CV CHD 6.7 8.1 3.6 7.3 5.9
Prior Stroke 5.1 3.6 1.5 1.2 2.7
Treated Hypertension 21.0 31.1 21.7 30.9 25.2
Treated Hyperlipidemia 11.5 15.4 10.9 16.9 13.1

 

37

 

Table 6. Primary Analysis: Relationship between NSAID use and CV events using the main

effects Cox Proportional Hazards model — Adjusted Hazard Ratios + 95% CI

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Study Cohorts
Non-users COX-2 NSAID Combo

Number of observed events 446 252 161 252
Days of follow-up, mean 1: SD 739 :I: 247 277 1 247 243 :t 272 518 t 270
Total follow-up, person years 17 353 7 278 10 628 9 257
Crude event rate/1000 person 25.7 34.6 15.1 27.2
years
Main Effects Model
Hazard Ratios (95% Cl)* 1.0 1.51 1.04 1.55

(reference) (1.28-1.78) (0.86-1.26) (1.31-1.83)
Model Details

Variable Hazard 95% CI p-value
Ratio

COX-2 treatment 1.508 1 .28-1 .78 <.0001
NSAID treatment 1.041 0.86-1.26 0.6846
Combination treatment 1.548 1.31 -1 .83 <.0001
Age 1.050 1.04-1.06 <.0001
Gender 1.707 1.51-1.93 <.0001
Coronary Artery Bypass Graft 3.581 1.76—7.29 0.0004
Prior AMI 2.585 2.22-3.01 <.0001
Family history of heart disease 2.088 1.18-3.70 0.0115
Atherosclerosis 1.233 1.00-1.53 0.0554
Diabetes 1.321 1.15-1.52 <.0001
Prior coronary heart disease 1.586 1.36-1.84 <.0001
Prior stroke 1.769 1 .47-2. 13 <.0001
Hypertension treatment no1 0.983 0.82-1.18 0.8542
Hypertension treatment yes1 1.550 1 34-1.79 <.0001
Hyperlipidemia treatment no2 0.512 0.41-0.63 <.0001
Hyperlipidemia treatment yes2 0.805 0.69-0.94 0.0061

 

Referent group is no hypertension
2 Referent group is no hyperiipidemia

* Adjusted for age, gender, CV risk factors

38

 

Table 7. Primary Analysis: Relationship between NSAID use and CV events using the Cox
Proportional Hazards interaction model - Adjusted Hazard Ratios + 95% Cl

 

 

 

 

 

 

 

 

 

 

 

 

Study Cohorts
Non-users COX-2 NSAID Combo
Interaction Model
Hazard Ratios (95% CI)*
No Hypertension 1.0 1.33 0.71 1.67
(reference) (099-180) (0.50-1.00) (1 .28-2.19)
Hypertension with no treatment 0.97 1.27 1.26 1.09
(0.75-1.27) (0.86-1.87) (0.86-1.86) (0.72-1.65)
Hypertension with treatment 1.36 2.30 1.69 2.23
(1 09-170) (224-237) (1 .29—2.20) (1 76-282)
Linear Contrasts“ COX-2 vs. NSAID vs. COX-2 vs. Combo vs.
Control Control NSAID Control
No Hypertension 0.061 0.052 0.002 0.0002
Hypertension with no treatment 0.212 0.220 0.992 0.608
Hypertension with treatment <0.0001 0.1 13 0.02 <0.0001
Model Details
Variable Hazard 95% Cl p-value
Ratio
COX-2 treatment 1.333 0.99—1.80 0.0608
NSAID treatment 0.710 0.50-1.00 0.0519
Combination treatment 1.674 1.28-2.19 0.0002
Age 1.050 1.04-1.05 <.0001
Gender 1.699 1.51-1.92 <.0001
Coronary artery bypass graft 3.648 1.79-7.43 0.0004
Prior AMI 2.608 224-303 <.0001
Family history of heart disease 2.052 1.16-3.63 0.0137
Atherosclerosis 1.233 1.00-1 .53 0.0556
Diabetes 1.319 1.15-1.51 <.0001
Prior coronary heart disease 1.587 1.37-1.84 <.0001
Prior stroke 1.775 1.47-2. 14 <.0001
Hypertension treatment no1 0.973 0.75-1.27 0.0402
Hypertension treatment yes1 1.364 1 09-170 7.6468
Hyperlipidemia treatment no2 0.507 0 41-0.63 <.0001
Hyperlipidemia treatment yes2 0.802 069-094 0.0053
COX-2*hypertension treatment no3 0.976 0.59-1.62 0.9238
COX-2*hypertension treatment 1 .266 0.88—1.82 0.2024
yes
NSAID*hypertension treatment no“ 1.828 1.07-3.12 0.0266
NSAID*hypertension treatment 1.742 1.14-2.66 0.0104
yes
Coambo*hypertension treatment 0.670 0.40-1.1 1 0.1215
no
Combo*hypertension treatment 0.975 0.69-1.38 0.8848
yes

 

Three level design variable combining status of hypertension and treatment for hypertension

 

2 Three level design variable combining status of hyperlipidemia and treatment for hyperlipidemia
3 Interaction term between treatment and hypertension status

39

Table 8. Secondary Analysis: Relationship between naproxen and non-naproxen NSAID
use and CV events relative to non-users using the Cox Proportional Hazards model -

Adjusted Hazard Ratios + 95% Cl

 

 

 

Study Cohort
Non-users Naproxen Non-naproxen
NSAID

Number of observed events 446 26 31
Days of follow-up, mean 1 SD 739 t 247 168 t 233 168 :I: 236
Total follow-up, person years 17 353 1 870 1 954
Crude event rate/1000 person 25.7 13.9 15.9
years
Relative Risk Ratios (95% CI)* 1.0 0.92 1.02

(Reference) (0.61 -1 .64) (069-150)

 

Linear Contrasts‘

Naproxen vs.
Non-users

Non-naproxen
vs. Non-users

Naproxen vs.
Non-naproxen

 

 

 

p-value

0.683

 

0.937

 

0.707

 

" Adjusted for age, gender and prior CV comorbidities

40

 

Table 9. Secondary analysis: Relationship between Celecoxib and Rofecoxib and CV
Events using the Cox Proportional Hazards Model - Adjusted Hazard Ratios + 95% CI

 

Celecoxib

Rofecoxib

 

Number of observed events
Days of follow-up, mean :t SD
Total follow-up, person years
Crude event rate/1000 person
years

102

245 :t 242
2896

35.2

57

174 i 173
1378
41.4

 

 

Main Effects Mode?

 

Hazard Ratios (95% Cl)

0.73 (052-102)

1.0 (reference)

 

Interactions Model2

 

No hypertension
Hypertension with no treatment
Hypertension with treatment

1.44 (0.71-2.94)
1.09 (0.76-1.49)
1.44 Q09-1.79)

1.0 (reference)
1.01 (0.34-2.98)
2.77 (1 .36-5.60)

 

 

 

 

 

 

Linear Contrasts“ p-value
No Hypertension 0.312
Hypertension without treatment 0.899
Hypertension with treatment 0.002

 

Adjusted for age, gender, previous AMI, previous stroke, diabetes, previous non-CV
heart disease
2 Adjusted for age, gender, previous AMI, previous stroke, diabetes, previous non-CV
heart disease, hypertension status, treatment*hypertension status

41

References

1.

10.

11.

12.

13.

14.

15.

16.

17.

Vane JR, Botting RM, Anti-inflammatory drugs and their mechanism of action, lnflamm. Res.
47, Supplement 2 1998: 878-887

Steinmeyer J, Pharmacological basis for the therapy of pain and inflammation with
nonsteroidal anti-inflammatory drugs, Arth Res. 2000;2z379—85

Hungin APS, Kean WF, Nonsteroidal Anti-Inflammatory Drugs: Overused or Underused in
Osteoarthritis?, Am J Med. 2001;110(1A):8$-11S

Hochberg MC, Dougados M., Pharmacological therapy of osteoarthritis, Best Pract Res Clin
Rheumatol 2001 Oct;15(4):583-93

Pasero G, Marcolongo R, Semi U, Parnham MJ, Ferrer F., A multi-centre, double-blind
comparative study of the efficacy and safety of aceclofenac and diclofenac in the treatment of
rheumatoid arthritis, Curr Med Res Opin 1995;13(6):305-15

Beaver WT., Review of the analgesic efficacy of ibuprofen. Int J Clin Pract Suppl 2003
Apr;(135):13-7

Hochberg M0,, New directions in symptomatic therapy for patients with osteoarthritis and
rheumatoid arthritis. Semin Arthritis Rheum 2002 Dec;32(3 Suppl 1):4-14

Schiff MH., A comparison of Naprelan and Naprosyn in the treatment of osteoarthritis of the
knee, Am J Orthop 1996 Sep;25(9 Suppl):14-20

Baumgartner H, Hohmeister R, Blumenberg-Novoselac M, An observer-blind crossover study
to compare the efficacies of flurbiprofen, indomethacin and naproxen given orally and rectally
in the relief of night pain and morning stiffness due to rheumatoid arthritis, J Int Med Res
1988 May-Jun;16(3):189-96

Steen KS, Lems WF, Aersten J et al., Incidence of clinically manifest ulcers and their
complications in patients with rheumatoid arthritis, Ann Rheum Dis. 2001;60:443-7

Simon LS, Are the biologic and clinical effects of the COX-2 specific inhibitors an advance
compared with the effects of traditional NSAIDs?, Curr Opin Rheumatol. 2000;12:163-70

Graumlich JF, Preventing gastrointestinal complications of NSAIDs. Risk factors, recent
advances, and latest strategies. Postgrad Med. 2001;109:117-20

Griffin MR, Scheiman JM, Prospects for changing the burden of nonsteroidal anti-
inflammatory drug toxicity, Am J Med. 2001;110:33S-37S

Fitzgerald GA, Cardiovascular Pharmacology of Nonselective Nonsteroidal Anti-Inflammatory
Drugs and Coxibs: Clinical Considerations, Am J Cardiol 2002;89(suppl):260—320

Oviedo JA, Wolfe MM, Clinical Potential of CycIo-Oxygenase-2 Inhibitors, BioDrugs
2001;15(9): 563-72

Cashman, JN, The Mechanisms of Action of NSAIDs in Analgesia, Drugs 1996; 52 Suppl. 5:
13-23

Goodwin JS, Mechanism of Action of Nonsteroidal Anti-Inflammatory Agents, Amer J Med
1984 Jul 13;77(1A):57-64

42

18.

19.

20.

21.

22.

23.

24.

25.

26.

27.

28.

29.

30.

31.
32.

33.

Catella-Lawson C, Crofford L, Cyclooxygenase Inhibition and Thrombogenicity, Am J Med
2001;110(3A):28$-32S

Mukherjee D, Nissan 8, Topol E, Risk of Cardiovascular Events Associated with Selective
COX-2 Inhibitors, JAMA 2001;286:954-959

Mukherjee D, Selective cyclooxygenase-2 (COX-2) inhibitors and potential risk of
cardiovascular events, Biochem Pharmacol 2002;63(5):817-21

Bing RJ, Lomnicka M, Why do cyclooxygenase-2 inhibitors cause cardiovascular events?, J
Am Coll Cardiol 2002;39(3):521-2

Califf RM, The Coxib Story: Some Lessons and More Questions, Am J Cardiol 2002;89:971-
2

Epstein M, Cardiovascular and Renal Effects of COX-2 Specific Inhibitors: Recent Insights
and Evolving Clinical Implications, Am J Ther 2001;8281-3

Bombardier C, Laine L, Reicin A, Shapiro D et al, Comparison of Upper Gastrointestinal
Toxicity of Rofecoxib and Naproxen in Patients With Rheumatoid Arthritis, NEJM
2000;343(21):1520-1528

Food and Drug Administration. Cardiovascular Safety Review. Rockville, MD: Food and Drug
Administraion;2001. Available at

http://www.fda.qov/ohrms/dockets/ac/01lbriefinq/3677b2 06 ca_rdio.pdf. Accessibility verified
May 18, 2003

White WB, Cyclooxygenase-2 inhibitors and cardiovascular thrombotic events, J Am Coll
Cardiol 2002;40(3):577-8

White WB, Faich G, Whelton A, Maurath C, Ridge NJ, Verburg KM, Geis GS, Lefkowith JB,
Comparison of thromboembolic events in patients treated with celecoxib, a cyclooxygenase-2
specific inhibitor, versus ibuprofen or diclofenac, Am J Cardiol 2002;89(4):425-30

Silverstein FE, Faich G, Goldstein JL, Simon LS, Pincus T, Whelton A, Lakuch R, Eisen G et
al, Gastrointestinal toxicity with celecoxib vs nonsteroidal anti-inflammatory drugs for
osteoarthritis an rheumatoid arthritis: the CLASS study: A randomized controlled trial.
Celecoxib Long-term Arthritis Safety Study

Ray WA, Stein CM, Daugherty JR, Hall K, Arbogast PG, Griffin MR, COX-2 selective non-
steroidal anti-inflammatory drugs and risk of serious coronary heart disese., The Lancet
2002;360:1071-1073

Mamdani M, Rochon P, Juurlink DN, Anderson GM, Kopp A, Naglie G, Austin PC, Laupacis
A, Effect of Selective Cyclooxygenase 2 Inhibitors and Naproxen on Short-term Risk of Acute
Myocardial Infarction in the Elderly, Arch Intern Med 2003;163:481-486

Cox DR, Oakes D, Analysis of Survival Data, 1984 Chapman Hall, London, UK.

Whelton A, Fort JG, Puma JA, Normandin D, Bello AE, Verburg KM, Cyclooxygenase-2-
Specific Inhibitors and Cardiorenal Function: A Randomized, Controlled Trial of Celecoxib
and Rofecoxib in Older Hypertensive Osteoarthritis Patients, Am J of Therap. 2001; 8:85-95
Whelton A, White WB, Bello AE, Puma JA, Fort JG et al., Effects of Celecoxib and Rofecoxib

on Blood Pressure and Edema in Patients 2 65 Years of Age With Systemic Hypertension
and Osteoarthritis, Am J Cardiol 2002;90:959-63

43

34.

35.
36.
37.

38.

39.

40.

41.

42.

43.

Strand V, Hochberg MC, The risk of cardiovascular thrombotic events with selective
cyclooxygenase-2 inhibitors, Arthritis Rheum 2002;47(4):349-55

Konstam MA, Cardiovascular events and COX-2 Inhibitors, JAMA 2001;286(22):2807-2812
Grant KD, Cardiovascular events and COX-2 Inhibitors, JAMA 2001;286(22):2807-2812
Singh G, Cardiovascular events and COX-2 Inhibitors, JAMA 2001;286(22):2807-2812

White WB, Cardiovascular events and COX-2 Inhibitors, JAMA 2001;286(22):2807-2812

Wolfe F, Flowers N, Burke TA, Arguelles LM, Pettitt D.lncrease in lifetime adverse drug
reactions, service utilization, and disease severity among patients who will start COX-2
specific inhibitors: quantitative assessment of channeling bias and confounding by indication
in 6689 patients with rheumatoid arthritis and osteoarthritis. J Rheumatol 2002
May;29(5):1015-22

Zhao, SZ, Burke TA, Whelton A, von Allmen H, Henderson SC, Comparison of the Baseline
Cardiovascular Risk Profile Among Hypertensive Patients Prescribed COX-2-Specific
Inhibitors or Nonspecific NSAIDs: Data From Real-Life Practice, Am J Managed Care Oct
2002;8:S392-S400

Food and Drug Administration, Vioxx Package Insert, Rockville, MD: Food and Drug
Administraion;2001. Available at http://www.fda.qov/cder/foi/labeI/2002/21042$7lbl.pdf.
Accessibility verified May 18, 2003

Food and Drug Administration. Celebrex Package Insert, Rockville, MD: Food and Drug
Administraion;2001. Available at http://www.ida.qov/cder/foi/laLbel/2001/20998s10lbl.pdf.
Accessibility verified May 18, 2003

Whelton A, Nephrotoxicity of Nonsteroidal Anti-inflammatory Drugs: Physiologic Foundations
and Clinical Implactions, Am J Med 1999;106(58):13$-24S

. Steinmeyer J, Pharmacological basis for the therapy of pain and inflammation of nonsteroidal

anti-inflammatory drugs, Arth Res 2000, 22379-85

44