Impact of NSAIDS on Mortality and the Effect of Preexisting Coronary Artery Disease in US Veterans

Methods 

To examine the risk of exposure to NSAIDs in a cohort at risk for long-term NSAID use, we conducted this nested case-control study in patients with osteoarthritis.

The study population was US veterans using the Veterans Health Administration health care system. As of April 2000, there were more than 26.5 million veterans.20 Of the 26.5 million veterans, approximately 4.4 million were enrolled and receiving health care services through the Veterans Health Administration.20

Patients with a diagnosis for osteoarthritis (International Classification of Diseases, Ninth Revision, 715) between October 1, 2000, and September 30, 2001, were identified. Patients were followed for events from October 1, 2001, to September 30, 2004. Patients were required to have received at least two medication dispensings from a Veterans Health Administration pharmacy to ensure use of pharmacy services. To attribute risk to specific NSAIDs, patients receiving more than one NSAID during follow-up were excluded. To ensure that events identified during follow-up were new events and not diagnoses from previous events listed as a comorbidity at an encounter, patients were excluded if they had any diagnosis for a myocardial infarction (ICD-9 410, 411.0, 412) or stroke (ICD-9 431, 433, 434) before October 1, 2001.

Patients meeting all criteria were divided into 2 cohorts according to the presence of preexisting CAD. We included patients in the CAD cohort if they had a diagnosis for angina (413) or ischemic heart disease (414.0) between October 1, 1999, and September 30, 2001. All other patients were included in the non-CAD group. After identification, the groups were further restricted by additional exclusion criteria. Because prescriptions filled at Veterans Health Administration pharmacies are frequently 90-day supplies, we excluded those who filled an NSAID prescription within the first 90 days of follow-up to avoid misclassifying exposure status and to identify an inception cohort to limit the impact of survivorship bias in patients previously using NSAIDs. Thus, to avoid immortal time bias, those with an event in the first 90 days of follow-up also were excluded. Finally, those with more than 90 inpatient days before their index date and those with cancer were excluded.

Cases were identified on the basis of occurrence of a cardiovascular or cerebrovascular event or death. Patients were identified as having a cardiovascular event if there was a diagnosis of acute myocardial infarction (410), old myocardial infarction (412) (to capture out-of-system events that occurred during follow-up), postmyocardial infarction syndrome (411), acute or subacute ischemic heart disease (411.1, 411.8), or an inpatient event with a primary diagnosis of angina (413). The angina diagnosis was only included in the non-CAD group as a cardiovascular event (0.55% of cases had angina as qualifying event). Patients were identified as having a cerebrovascular event with codes 433, 434, or 435. Validation studies for cardiovascular and cerebrovascular events in these data have shown that diagnosis codes are valid compared with medical records.21 Both inpatient and outpatient records were used to identify events, because acute events may not have occurred at a Veterans Health Administration facility. Rather, they may occur at the nearest health care facility, and thus, outpatient diagnoses were also used to identify events. Deaths were identified in the beneficiary database, which captures between 90% and 95% of veterans who die.22, 23 Deaths identified represent all-cause mortality rather than cause-specific mortality. For the composite end point, the first occurring event was identified. In separate analyses of mortality, all deaths during follow-up were included as cases.

An incidence density sampling approach was used, and a maximum of 20 controls were randomly selected for each case from eligible patients at risk at the time of the case event.24, 25 Controls were individually matched on age (±2 years) and sex, and given the same index date as cases. There was an average of 11.3 controls per case, with 24 cases having fewer than 5 controls (1 case 3 controls, 23 cases 4 controls).

NSAID exposure was determined from pharmacy-dispensing records. Patients with at least one dispensing during follow-up were defined as exposed. Specific drugs examined were celecoxib, rofecoxib, naproxen, ibuprofen, diclofenac, etodolac, and indomethacin. These represented the most frequently used NSAIDs during follow-up. All other nonselective NSAIDS (excluding aspirin) were combined into an “other NSAID” category.

Covariates were identified using inpatient and outpatient data from October 1, 2000, to the index date. Pharmacy-dispensing data were used to identify medications. Inpatient and outpatient diagnoses were used to determine comorbidities. The annual number of hospitalizations and outpatient physician visits were counted.

Because of concerns for possible bias introduced by case-mix differences between populations, we used an event risk score similar to other analyses investigating cardiovascular events as an outcome.3, 9, 18 To create the risk scores, all patients not exposed to NSAIDs during follow-up were identified. A logistic regression model was used to create an event risk score separately in each cohort. The final model for the risk score included age, sex, anticoagulant, antiplatelet, aspirin, digoxin, beta-blocker, calcium channel blocker, nitrates, antiarrhythmics, statins, diuretics, angiotensin-converting enzyme inhibitors, angiotensin receptor blockers, hypoglycemics, angina, chronic heart failure, peripheral vascular disease, obesity, or tobacco or alcohol dependence. The predicted values for each individual were divided into deciles, and 10 risk score categories were created. For individuals included in the case-control study, an event risk score category was assigned based on covariates up to their index date.

Conditional logistic regression was used to examine the association between NSAID exposure and events. Three comparisons were made within cohorts: no exposure versus any NSAID; no exposure versus COX-2 and non–COX-2; and no exposure versus individual medications. Comparisons were made for the composite end point, cardiovascular and cerebrovascular events combined, and each event individually. Odds ratios (ORs) and 95% confidence intervals were calculated. Adjusted models included event risk score and comorbidities not in the risk score.

The Hines Veterans Affairs Hospital institutional review board approved the research.

Results 

We identified 565,451 patients with an osteoarthritis diagnosis, of whom 490,507 met initial eligibility criteria. A total of 371,897 patients (75.8%) were included in the non-CAD group, and 188,610 patients (24.2%) were included in the CAD cohort. Figure 1 shows the number of cases after each restriction. There were 16,869 cases in the non-CAD cohort and 11,912 cases in the CAD cohort. In the non-CAD cohort there were 3022 (17.9%) cardiovascular cases, 4787 (28.4%) cerebrovascular cases, and 9060 (53.7%) deaths. For the all-cause mortality analysis there were 9636 deaths during follow-up. In the CAD cohort there were 3213 (27.0%) cardiovascular cases, 2862 (24.0%) cerebrovascular cases, and 5837 (49.0%) deaths. A total of 6452 deaths occurred during follow-up in this group.

View Large Image Download to PowerPoint

Figure 1. Study inclusion flow chart for CAD and non-CAD cohorts and number of cases after further inclusion criteria were applied. MI = myocardial infarction; VHA = Veterans Health Administration; CAD = coronary artery disease; NSAID = nonsteroidal anti-inflammatory drug; Rx = prescription; Dx = diagnosis.

The average age was 69.8 years (standard deviation = 11.9) in the non-CAD group and 71.8 years (standard deviation = 10.0) in the CAD group. A slightly larger proportion of cases and controls were exposed to prescription NSAIDs in the non-CAD group (18.4% of cases, 20.1% of controls; P < .001) than in the CAD group (15.8% of cases, 16.0% of controls; P = .433) (Table 1).

Table 1. Baseline Characteristics of Cases and Controls by Presence of Preexisting Coronary Artery Disease

		Non-CAD Cohort			CAD Cohort
		Cases	Controls	P Value	Cases	Controls	P Value
	N	16,869	190,296		11,912	135,379
	Total person-years of follow-up	25,617	283,166		17,120	191,977
	Male, %	96.8	96.6		98.3	98.2
	Ages, %
	<60 y	22.0	21.5		14.0	12.8
	60-74 y	34.2	33.8		38.4	38.0
	≥75 y	43.8	44.8		47.5	49.2
	NSAID exposure, %
	Any NSAID	18.4	20.1	<.001	15.8	16.0	.433
	Any COX-2	1.6	1.6	.701	1.9	1.8	.714
	Celecoxib	0.7	0.8	.388	0.8	0.8	.697
	Rofecoxib	0.9	0.8	.171	1.1	1.1	.884
	Naproxen	5.1	5.7	.001	4.0	4.4	.068
	Ibuprofen	6.5	7.4	<.001	5.4	5.4	.786
	Diclofenac	1.0	1.0	.281	0.8	0.9	.717
	Etodolac	1.2	1.2	.932	0.9	1.0	.279
	Indomethacin	1.0	1.0	.735	1.0	0.8	.095
	Other	2.0	2.2	.050	1.8	1.8	.783
	Non-NSAID medication exposure, %
	HIV medication	0.2	0.1	<.001	0.1	0.1	.646
	Anticoagulant	9.5	5.9	<.001	18.6	13.5	<.001
	Antiplatelet	8.8	3.6	<.001	13.	9.3	<.001
	Aspirin	30.3	22.3	<.001	49.7	43.0	<.001
	Digoxin	9.1	5.4	<.001	21.9	15.1	<.001
	Beta-blocker	29.4	22.3	<.001	57.9	55.1	<.001
	Alpha-blocker	22.7	22.2	.121	23.5	24.1	.126
	Calcium channel blocker	29.0	24.8	<.001	35.9	34.5	.002
	Antihypertensive (other)	4.5	2.6	<.001	5.9	3.6	<.001
	Nitrate	13.6	6.9	<.001	54.0	42.7	<.001
	Antiarrhythmic	2.1	1.3	<.001	6.1	4.3	<.001
	Statin	31.9	31.6	.405	56.7	60.5	<.001
	Non-statin lipid lowering	5.2	4.7	.005	8.8	7.8	<.001
	Thiazide diuretic	17.9	17.9	.953	16.6	16.2	.191
	Loop diuretic	22.9	11.7	<.001	43.6	29.6	<.001
	K+ sparing diuretic	9.5	7.7	<.001	11.4	9.0	<.001
	ACE inhibitor	40.3	34.8	<.001	54.4	48.6	<.001
	ARB	4.4	4.0	.021	7.1	6.5	.012
	Insulin	8.3	4.2	<.001	13.6	8.6	<.001
	Hypoglycemic	19.5	15.1	<.001	26.3	22.2	<.001
	Noncardiac comorbidities, %
	COPD	31.9	18.9	<.001	38.9	27.1	<.001
	GERD	5.4	4.5	<.001	6.1	5.0	<.001
	Peptic ulcer disease	7.8	5.9	<.001	9.8	7.8	<.001
	Obesity	18.9	17.9	.001	22.3	21.3	.011
	Liver disorder	4.3	1.4	<.001	2.3	1.2	<.001
	Chronic renal failure	8.0	3.7	<.001	11.7	6.5	<.001

CAD = coronary artery disease; NSAID = nonsteroidal anti-inflammatory drug; COX-2 = cyclooxygenase-2; HIV = human immunodeficiency virus; ACE = angiotensin-converting enzyme; ARB = angiotensin receptor blocker; COPD = chronic obstructive pulmonary disease; GERD = gastroesophageal reflux disease.

Exposure to any NSAID during follow-up was associated with an adjusted OR of 0.91 (95% confidence interval [CI], 0.87-0.95) for the composite end point in the non-CAD group and 0.99 (95% CI, 0.94-1.05) in the CAD group (Table 2). For cardiovascular or cerebrovascular events, the OR was 1.14 (95% CI, 1.08-1.21) in the non-CAD group and 1.18 (95% CI, 1.11-1.27) in the CAD group. Exposure to NSAIDs was associated with a decreased risk of mortality in both the non-CAD (OR = 0.72 [95% CI, 0.68-0.77]) and CAD (OR = 0.79 [95% CI, 0.73-0.86]) groups.

Table 2. Risk of Events Associated with Nonsteroidal Anti-inflammatory Drug Exposure Compared with No Exposure Stratified by Preexisting Coronary Artery Disease

		Any Event		Cardiovascular or Cerebrovascular		Cardiovascular Event		Cerebrovascular Event		Death
		OR⁎	95% CI	OR⁎	95% CI	OR⁎	95% CI	OR⁎	95% CI	OR⁎	95% CI
	Non-CAD
	No exposure	1	–	1	–	1	–	1	–	1	–
	NSAID	0.91	(0.87-0.95)	1.14	(1.08-1.21)	1.14	(1.03-1.25)	1.14	(1.05-1.22)	0.72	(0.68-0.77)
	No exposure	1	–	1	–	1	–	1	–	1	–
	Non–COX-2	0.90	(0.87-0.95)	1.14	(1.08-1.21)	1.15	(1.05-1.27)	1.13	(1.04-1.22)	0.71	(0.67-0.76)
	COX-2	0.94	(0.83-1.07)	1.12	(0.94-1.35)	0.94	(0.68-1.31)	1.22	(0.98-1.52)	0.81	(0.68-0.97)
	CAD
	No exposure	1	–	1	–	1	–	1	–	1	–
	NSAID	0.99	(0.94-1.05)	1.18	(1.11-1.27)	1.23	(1.12-1.35)	1.13	(1.02-1.25)	0.79	(0.73-0.86)
	No exposure	1	–	1	–	1	–	1	–	1	–
	Non–COX-2	0.99	(0.94-1.05)	1.20	(1.12-1.29)	1.27	(1.14-1.40)	1.12	(1.01-1.25)	0.77	(0.71-0.85)
	COX-2	0.97	(0.84-1.12)	1.08	(0.89-1.30)	0.95	(0.71-1.27)	1.21	(0.94-1.57)	0.93	(0.76-1.13)

OR = odds ratio; CI = confidence interval; CAD = coronary artery disease; NSAID = nonsteroidal anti-inflammatory drug; COX-2 = cyclooxygenase-2.

⁎ Adjusted for event risk score and comorbidities.

When exposure was stratified by COX-2 and nonselective NSAIDs, the association between exposure and events was similar for nonselective NSAIDs as with NSAIDs overall. There was an association with increased risk for cardiovascular or cerebrovascular events in both groups and decreased risk for overall mortality (Table 2). For COX-2 as a class, exposure was not associated with the composite end point of any event, nor was there a significant increase in cardiovascular or cerebrovascular event risk.

When the associations with individual medications were examined, only naproxen (OR = 0.91 [95% CI, 0.84-0.98]) and ibuprofen (OR = 0.88 [95% CI, 0.83-0.94]) had a significant association with the composite end point in the non-CAD group, and there were no significant associations with the composite end point in the CAD group (Table 3). The decreased risk observed in the composite measure was largely dependent on the decreased risk associated with all-cause mortality for naproxen and ibuprofen (Figure 2). Of the other medications examined, diclofenac, etodolac, and the other NSAID group were associated with significant protective effects for all-cause mortality in the non-CAD group, whereas only the “other NSAID” group was significant in the CAD group. Although every medication did not show a statistically significant OR, there was an overall trend in the point estimates for all-cause mortality showing a protective effect with each medication.

Table 3. Risk of Events in Noncoronary Artery Disease and Coronary Artery Dsease Cohorts by Nonsteroidal Anti-inflammatory Drug Exposure

		Any Event		Cardiovascular or Cerebrovascular Event		Cardiovascular Event		Cerebrovascular Event
		OR⁎	95% CI	OR⁎	95% CI	OR⁎	95% CI	OR⁎	95% CI
	Non-CAD
	No NSAID	1	–	1	–	1	–	1	–
	Celecoxib	0.87	(0.71-1.05)	0.91	(0.69-1.22)	0.85	(0.52-1.39)	0.97	(0.68-1.37)
	Rofecoxib	1.02	(0.85-1.21)	1.32	(1.04-1.67)	1.03	(0.66-1.59)	1.45	(1.10-1.92)
	Naproxen	0.91	(0.84-0.98)	1.18	(1.07-1.30)	1.21	(1.04-1.40)	1.15	(1.01-1.31)
	Ibuprofen	0.88	(0.83-0.94)	1.11	(1.02-1.22)	1.10	(0.96-1.27)	1.11	(0.99-1.25)
	Diclofenac	0.91	(0.78-1.08)	1.32	(1.08-1.62)	1.42	(1.03-1.94)	1.24	(0.95-1.63)
	Etodolac	1.00	(0.86-1.16)	1.33	(1.10-1.62)	1.22	(0.89-1.67)	1.43	(1.12-1.82)
	Indomethacin	0.95	(0.80-1.12)	1.11	(0.89-1.39)	1.06	(0.74-1.51)	1.13	(0.86-1.50)
	Other	0.89	(0.79-1.00)	0.97	(0.83-1.14)	1.05	(0.81-1.36)	0.91	(0.74-1.13)
	CAD
	No NSAID	1	–	1	–	1	–	1	–
	Celecoxib	1.00	(0.81-1.24)	1.17	(0.88-1.55)	0.92	(0.60-1.43)	1.40	(0.96-2.03)
	Rofecoxib	0.95	(0.79-1.14)	1.01	(0.78-1.31)	0.97	(0.67-1.42)	1.09	(0.77-1.54)
	Naproxen	0.94	(0.86-1.04)	1.09	(0.96-1.24)	1.01	(0.84-1.20)	1.20	(1.01-1.43)
	Ibuprofen	1.02	(0.94-1.12)	1.27	(1.15-1.42)	1.45	(1.26-1.67)	1.08	(0.91-1.27)
	Diclofenac	1.03	(0.85-1.25)	1.18	(0.90-1.55)	1.01	(0.67-1.52)	1.33	(0.92-1.91)
	Etodolac	0.93	(0.76-1.14)	1.12	(0.86-1.45)	1.24	(0.88-1.73)	0.94	(0.63-1.42)
	Indomethacin	1.05	(0.85-1.29)	1.15	(0.88-1.50)	1.17	(0.82-1.67)	1.12	(0.76-1.65)
	Other	1.01	(0.87-1.17)	1.31	(1.10-1.57)	1.54	(1.22-1.95)	1.06	(0.80-1.41)

OR = odds ratio; CI = confidence interval; CAD = coronary artery disease; NSAID = nonsteroidal anti-inflammatory drug.

⁎ OR adjusted for event risk score and comorbidities.

View Large Image Download to PowerPoint

Figure 2. Risk of all-cause mortality in non-CAD and CAD cohorts by NSAID exposure (lines are 95% confidence interval; non-CAD cohort, solid line; CAD cohort, dashed line). NSAID = nonsteroidal anti-inflammatory drug.

With regard to cardiovascular and cerebrovascular events, there was an overall trend toward medications being associated with increased risk of events (Table 3). The increased risk ranged from 1.11 (95% CI, 1.02-1.22) for ibuprofen to 1.33 for etodolac (95% CI, 1.10-1.62) in the non-CAD cohort. In the CAD group, ibuprofen was associated with the largest risk of either a cardiovascular or cerebrovascular event (OR = 1.27 [95% CI, 1.15-1.42]).

Discussion 

We found exposure to NSAIDs among patients with osteoarthritis was associated with increased risk of cardiovascular and cerebrovascular events in both patients with and without preexisting CAD. For the NSAIDs as a class, and in particular for the nonselective NSAIDS, there was a decreased risk for all-cause mortality associated with exposure to the medications.

As in several recent studies,2, 3, 5, 6, 7, 12, 16 an increased risk of cardiovascular and cerebrovascular events associated with rofecoxib was found in our study. However, the results of this study suggest the overall risk of events may be modified by CAD. Our analysis would suggest that, whether through its prothrombotic effects26 or its effects on blood pressure control,27 exposure to rofecoxib puts persons who would be expected to have relatively low risks of cardiovascular or cerebrovascular events at an increased risk for those events. However, in patients with underlying risk already present, exposure to rofecoxib may not lead to increases in the risk of events relative to any underlying risk because of CAD. That is, in the patients with preexisting CAD, rofecoxib exposure was not associated with an increased risk of cardiovascular or cerebrovascular events.

The evidence on the association between nonselective NSAIDs and cardiovascular events is inconsistent. For example, Ray et al18 showed an increased risk for ibuprofen exposure (relative risk [RR] = 1.15) and not naproxen (RR = 0.85). Mamdani et al6 found an increased risk for chronic heart failure events in patients exposed to nonselective NSAIDs (RR = 1.4). Levesque et al5 reported an increased risk of cardiovascular events associated with naproxen use (RR = 1.17), albeit not statistically significant. This last study is in contrast with a meta-analysis that showed naproxen was associated with a 14% reduced risk of myocardial infarction.16 Similar to a recently reported study,15 the results of our study suggest the effect of NSAIDs on cardiovascular and cerebrovascular events may be based on characteristics of the class of medications and not individual medicines, particularly for those patients without preexisting CAD.

A notable finding is that although NSAIDs may increase the risk of cardiovascular or cerebrovascular events, they were associated with a decreased risk in all-cause mortality. This is particularly interesting given that mortality is the outcome least likely to suffer from misclassification in this analysis. Studies of NSAID risks have generally focused on specific outcomes, such as upper gastrointestinal events or cardiovascular events, and have not examined all-cause mortality. It should be noted that for this study, we defined the primary outcome as the first occurring event. Therefore, if patients with nonfatal cardiovascular or cerebrovascular events subsequently died, it would be possible that the “protective” effect observed with all-cause mortality would be reduced or completely removed. However, when we examined all deaths during follow-up, the decreased risk for all-cause mortality associated with NSAID exposure remained.

The possible protective effect of NSAIDs on all-cause mortality opens speculation about possible mechanisms. If we assume that the protective mechanisms are not reducing mortality by cardiovascular or cerebrovascular events, then it would have to do so through other causes. In this elderly population the next most common causes of death (excluding cancer) would be respiratory tract infections and dementia. Is it possible that anti-inflammatory properties of NSAIDs reduce the risk of respiratory-related mortality? Are they helping delay mortality related to Alzheimer’s disease or other dementia? Or does the protective effect simply reflect improved vitality or exercise because of more optimal control of chronic pain associated with osteoarthritis?

There are several limitations that need to be acknowledged. A cause-specific mortality analysis was not possible, because the mortality database used does not contain cause of death. The classification of patients as having preexisting CAD was based on the presence of diagnostic codes and not clinical markers; also the cohort did not include patients with previous events. From this analysis, we do not know whether the risk would be different in patients with different levels of coronary heart disease risk or in patients with previous events. The relatively low use of COX-2 medications in this population is likely attributed to Veterans Health Administration recommendations for NSAID use. COX-2 medications are to be reserved for patients at high risk for upper gastrointestinal events. Therefore, patients in this analysis who are prescribed COX-2 medications may be slightly different with respect to upper gastrointestinal event risks than non–COX-2 NSAID users. However, there is no evidence that upper gastrointestinal event risk would lead to differentially prescribing COX-2 medications for cardiovascular disease risk. In addition, we only include patients receiving a single NSAID product during follow-up (the majority of patients treated with NSAIDs used only one product during follow-up). We limited the analysis to this population to examine the effects of individual medications and is another potential explanation for relatively low COX-2 use. Patients prescribed COX-2 medications may be switched from another NSAID after experiencing gastrointestinal discomfort, and by limiting the analysis to patients using only one NSAID, we exclude those individuals. We were unable to account for over-the-counter use, and those classified as unexposed may have used over-the-counter NSAIDs. For this concern, we believe the low medication co-payments (either $0 or $7 per 30-day supply) would have limited over-the-counter use to only the lowest doses of over-the-counter NSAIDs. Finally, the ability to draw conclusions around specific medications is limited by relatively small sample sizes for some drugs; however, point estimates for all-cause mortality were consistent across all medications.

Respective of these limitations, our study demonstrated an increased risk of nonfatal cardiovascular and cerebrovascular events consistent with previous reports. We found that risks may differ for events based on preexisting CAD, and that there may be some degree of protective effect of NSAIDs on all-cause mortality. This study, when added to the rapidly increasing literature of NSAID exposure, suggests the need for better understanding of mechanisms related to major adverse events.