Experimental studies have shown that administration of nonsteroidal anti-inflammatory drugs (NSAIDs) to susceptible individuals can lead to the development of congestive heart failure (CHF). There have been few epidemiological investigations of the importance of this adverse effect.
To estimate the relative risk of first admission to a hospital with CHF in recent users of NSAIDs, compared with nonusers, and to determine whether the estimated relative risk was increased in those with a history of heart disease and the extent to which the level of risk varied with the dose and half-life of the drugs consumed.
We conducted a matched case-control study of the relationship between recent use of NSAIDs and hospitalization with CHF. Cases (n = 365) were patients admitted to hospitals with a primary diagnosis of CHF. Controls (n = 658) were patients without CHF who were admitted to the same hospitals as case patients. Structured interviews were used to obtain information on several study factors, including recent use of aspirin and other NSAIDs.
Use of NSAIDs (other than low-dose aspirin) in the previous week was associated with a doubling of the odds of a hospital admission with CHF (adjusted odds ratio, 2.1; 95% confidence interval, 1.2-3.3). Use of NSAIDs by patients with a history of heart disease was associated with an odds ratio of 10.5 (95% confidence interval, 2.5-44.9) for first admission with heart failure, compared with 1.6 (95% confidence interval, 0.7-3.7) in those without such a history. The odds of a first admission to a hospital with CHF was positively related to the dose of NSAID consumed in the previous week, and was increased to a greater extent with long half-life than with short half-life drugs. Assuming these relationships are causal, NSAIDs were responsible for approximately 19% of hospital admissions with CHF.
The burden of illness resulting from NSAID-related CHF may exceed that resulting from gastrointestinal tract damage. NSAIDs should be used with caution in patients with a history of cardiovascular disease.
THE BENEFICIAL and harmful effects of nonsteroidal anti-inflammatory drugs (NSAIDs) are related primarily to their inhibition of prostaglandin synthesis.1 Although prostaglandins have both vasodilator and vasoconstrictor actions, the overall effects of NSAIDs are to raise systemic vascular resistance and to reduce renal perfusion in susceptible individuals.2,3 In some individuals with impaired ventricular function, these mechanisms can exacerbate their tendency to develop congestive heart failure (CHF).3 There have been experimental studies demonstrating the deleterious hemodynamic effects of NSAIDs in persons with preexisting CHF.3 However, there have been few pharmacoepidemiological studies designed to estimate the risk of CHF in users of NSAIDs in the population.4 NSAIDs are widely used by elderly patients.1 Congestive heart failure is a common reason for admission to the hospital, and is associated with poor survival following discharge from the hospital.5-7 A modest increase in the risk of this condition in users of NSAIDs (compared with nonusers) could translate into a substantial increase in morbidity and mortality, both for the individual user and for the community.
We describe herein the results of a case-control study designed to investigate these issues. Our main aim was to estimate the relative risk of first admission to a hospital with CHF in recent users of NSAIDs, compared with nonusers. We were also interested in whether the estimated relative risk was increased in those with a history of heart disease and in the extent to which the level of risk varied with the dose and half-life of the drugs consumed. From a consideration of these factors we thought it possible to provide specific advice to improve the safety of these treatments.
We carried out a matched case-control study between 1993 and 1995. The study was based in 2 public hospitals in the Hunter Region of New South Wales (NSW). These hospitals provide most of the acute medical care for a population of about 450,000. Cases and controls were recruited from a list of postal codes that represented the catchment areas for the participating hospitals. The study was approved by the ethics committees of the University of Newcastle, Newcastle, NSW, and the Hunter Area Health Service. All participants gave their informed consent.
Most cases of CHF admitted to the study hospitals were cared for by general internal medicine specialists who made the diagnosis on the basis of clinical and radiologic criteria and did not routinely perform investigations of ventricular function. Consequently, we used a "pragmatic" case definition. Eligible cases were consecutive patients admitted to either study hospital where the resident medical officer who admitted the case and the attending physician were in agreement that the primary reason for admission was CHF. Patients admitted to hospitals for other reasons who were found incidentally to have CHF were excluded. The study nurses ensured that all met the Framingham criteria for definite cases of CHF.8 Case patients who did not meet these criteria were excluded.
We aimed to match 2 controls to each case. Controls were patients of the same sex and age (±5 years) as the cases, who were admitted as emergencies to the same hospital as the cases and who had no clinical or radiologic evidence of CHF. Because CHF is a chronic disorder, we excluded potential controls if they had a history of hospital admission with CHF. We also excluded subjects who had primary admission diagnoses that were known to be complications of, or indications for, NSAID use. Patients with upper gastrointestinal tract bleeding or ulceration, large bowel bleeding, or a primary admission diagnosis of a rheumatic disease were excluded. Patients who were admitted for other reasons, but who also had rheumatic diseases or a history of conditions that are known complications of NSAID therapy, were not excluded; this rule also applied to the cases.
After termination of recruitment an audit was performed of controls who had been admitted to the study because a decision about enrollment had sometimes to be made before a final diagnosis was available. Our concern was that we might have unwittingly enrolled controls who proved later to have been admitted with CHF. The audit involved a review of all clinical data from the hospital admission, including the results of radiologic investigations. The final decision regarding exclusion of doubtful controls was made by the chief investigator (D.H.), who was not informed about the drug consumption history of the individual concerned. As a result of this audit, 22 controls were withdrawn from the study.
The index day for cases and controls was the day of hospital admission. Information on all drugs that had been used in the week and month prior to the index day, recent and past history of heart disease, hypertension, diabetes, demographic details, alcohol consumption, and smoking history was obtained by structured interview. All interviews were conducted within 1 week of admission to hospital, and more than 80% were conducted within 4 days. The trained interviewers were aware of whether a subject was a case or control, but adhered strictly to the wording of the questionnaire. Each question could be repeated once for clarification. Interviewees were unaware of the study hypotheses. We were particularly interested in the consumption of analgesic and anti-inflammatory medications. This information was obtained by a series of open-ended questions inquiring about any drugs used for painful complaints. These were followed by more direct questions and presentation of flash cards listing, in large type, the generic and trade names of all relevant products on the Australian market. Information on prescribed NSAIDs was confirmed by contact with the patient's family practitioner. In addition to the information obtained by interview, we abstracted the following details from medical records and hospital computerized records: chest x-ray reports, hematologic and biochemical data, clinical notes regarding previous admissions with CHF, medical history, and all other active diagnoses that were thought to be present at the time of hospital admission. A history of heart disease was recorded if patients responded positively to specific questions about preexisting heart problems or if such problems were recorded in their medical records.
We were aware of the difficulties of performing a case-control study of a condition that leads to recurrent admissions. In many patients, CHF is a prevalent disease and because this would likely lead to modification of NSAID therapy prior to entry to the study, our primary interest was the analysis of case patients who experienced their first admission for CHF.
The anticipated prevalence of NSAID use in the controls was about 20%.9 The mean daily dose of NSAIDs in the week prior to admission was standardized by using the manufacturers' recommended minimum daily maintenance doses for rheumatoid arthritis as 1 dose unit.9,10 The dose calculation excluded regular low-dose aspirin (daily dose of 100 to 300 mg), as this was likely to have been taken as prophylaxis against cardiovascular events, making a causal association with CHF unlikely.
The odds ratio (OR) was regarded as a valid estimate of the relative risk. We aimed to recruit a total of 400 cases and 800 controls to detect an OR of 1.5 with use of NSAIDs, with a power of 80% and type I error of .05.11 Nonuse of NSAIDs and aspirin was the reference for calculation of exposure ORs. The relationship between the OR for CHF and the plasma half-lives of the drugs was examined using published values from a standard reference source.10 We used multiple conditional logistic regression analysis to adjust for potential confounders (age in years; history of heart disease, renal disease, respiratory disease, and peripheral arterial disease; and use of β-blockers, calcium channel antagonists, antidiabetic drugs, and oral glucocorticosteroids).12 Multiple conditional logistic regression was also used to assess relationships between drug doses and plasma half-life values. The hypothesis that prior heart disease might increase the adverse effects of NSAID therapy was stated a priori. Effect modification was assessed by including product terms (use of NSAIDs in the previous week × history of heart disease) in the regression models, and P values (2 sided) were calculated by performing log likelihood ratio tests.
Estimation of disease burden
To calculate the excess risk for the individual user of NSAIDs, we referred the ORs for cases experiencing their first episode of CHF to hospital admission data for NSW in 1993-1994.13 From the number of admissions to NSW hospitals with a principal diagnosis of CHF we estimated numbers of patients, using data on the frequency and number of readmissions experienced by a cohort of patients with CHF from Westmead Hospital in Sydney, NSW.14 Etiologic fractions were calculated from the adjusted ORs, and the prevalence of NSAID use among case patients was determined using the method of Bruzzi et al.15 The etiologic fraction is an estimate of the proportion of cases in the community that can be attributed to use of the drugs, assuming the associations are causal.
Three hundred sixty-five case patients with CHF and 658 controls were recruited. It was possible to find only one matched control for 50 of the case patients with CHF. One hundred forty-nine case patients were hospitalized for the first time with CHF.
Characteristics of all cases and controls
Most case patients had moderate or severe CHF; 92% had New York Heart Association functional grades III and IV. Almost all case patients had experienced dyspnea and 58% complained of peripheral edema. Seventy-two percent of case patients had elevated jugular venous pressure, and almost all had lung crackles on examination. Eighty-six percent of case patients had cardiomegaly noted on their chest x-ray reports. The clinical features of first-time cases were very similar, although a slightly lower proportion (78%) had radiographic evidence of cardiomegaly.
Table 1 and Table 2 summarize the baseline characteristics of all cardiac failure cases and controls and first-time cases and controls, respectively. Despite matching age to within 5 years, the case patients were older than the controls. Compared with controls, a greater proportion of all cases were ex-smokers, while fewer were current smokers. Cases were more likely than controls to have a history of heart disease and respiratory disease. Ninety percent of individuals with a history of heart disease had experienced a myocardial infarction or angina pectoris. The prevalence of other diseases in the 2 groups was similar.
Drug use by all cases and controls
As expected, the use of drugs for cardiovascular disorders differed between the groups. The greatest differences were seen with loop diuretics (furosemide and bumetanide), which were used more often by cases than controls. In contrast, thiazide diuretics had been used by similar numbers of cases and controls. Major differences were seen in prior use of angiotensin-converting enzyme (ACE) inhibitors, which were taken by twice as many cases as controls. Overall, calcium channel blocking drugs were used more frequently by cases than controls, but the difference was confined to diltiazem, which has a mixed vascular and cardiac action. Cases had a lower use of β-blocking agents than controls. In the case of antidiabetic drugs, the reverse was true.
USE OF NSAIDs BY ALL CASES AND CONTROLS
Consumption of nonaspirin NSAIDs in the previous week was higher among cases (17.3%) than controls (12.0%) (Table 3). The crude OR for developing CHF was 1.5 (95% confidence interval [CI], 1.0-2.2). Adjustment was made for the following potential confounders: a history of heart disease, renal disease, respiratory disease, and peripheral arterial disease; the use of β-blockers, calcium channel antagonists, antidiabetic drugs, and oral glucocorticosteroids; and age in years. These adjustments increased the ORs for CHF with use of nonaspirin NSAIDs in the previous week to 2.1 (95% CI, 1.2-3.3).
Characteristics of patients with a first admission with chf
One hundred forty-nine case patients were hospitalized for the first time with CHF. The distribution of most baseline characteristics in the cases with cardiac failure for the first time and their matched controls (n = 272) was similar to the whole group (Table 2). As expected, a history of heart disease in first-time cases was lower than that for the whole group of cases, and they had a lower consumption of ACE inhibitors and high-ceiling diuretics.
USE OF NSAIDs BY PATIENTS WITH A FIRST ADMISSION WITH CHF
The results for the first-time cases are summarized in Table 3. The ORs calculated on the basis of nonaspirin NSAIDs or any NSAIDs (including aspirin other than low dose) used in the previous week and the previous month were slightly higher than those seen with the entire group of cases. The ORs increased modestly after adjustments for potential confounders. Use of NSAIDs in the previous week was associated with an adjusted OR for first admission with heart failure of 2.8 (95% CI, 1.5-5.1). The test for interaction with a history of heart disease was statistically significant (P = .01). Among first-time cases with a history of heart disease, use of nonaspirin NSAIDs in the week prior to admission was associated with an OR of 10.5 (95% CI, 2.5-44.9). The corresponding OR in those without a history of heart disease was 1.6 (95% CI, 0.7-3.7). The importance of the interaction between NSAID use and a history of heart disease is illustrated in Figure 1. Compared with subjects who had no history of heart disease and were nonusers of NSAIDs, the OR for the first episode of heart failure in subjects who had a history of heart disease and had used nonaspirin NSAIDs in the previous week was 26.3 (95% CI, 5.8-119.1).
Importance of dose and drug half-life
A dose-response relationship was apparent in first-time cases. However, the dose-response relationship varied according to the history of heart disease, and the test for an interaction with this variable was significant (P = .002). The estimate of the adjusted OR per unit increase in daily dose in those without a preexisting history of heart disease was nonsignificant: 1.2 (95% CI, 0.7-2.0). In contrast, the estimate in those with a preexisting history of heart disease was 7.5 per unit increase in daily dose (95% CI, 2.0-27.6).
In patients with their first episode of CHF there was a statistically significant relationship between the OR for developing CHF and published values for the plasma half-life of the NSAID used in the week prior to hospitalization. However, this relationship was only apparent in subjects with a history of heart disease (Table 4). The test for an interaction between the half-life relationship and a history of heart disease was significant (P = .004). In those with a history of heart disease, the test for log linear trend was statistically significant (P = .03).
Calculation of etiologic fractions
Among patients who were experiencing their first episode of heart failure the prevalence of use of any NSAID in the week prior to admission among case patients was 29.5%. Assuming a nonconfounded association, the etiologic fraction due to consumption of NSAIDs in this population was calculated to be 19%.
According to data from the New South Wales Department of Health in 1993-1994 there were 11,336 admissions of patients aged 65 years and older with a primary discharge diagnosis of CHF.14 Using the data from the Westmead Hospital cohort we calculated that approximately 18% of patients admitted to hospitals with CHF were readmitted with CHF within 6 months, in some cases more than once.14 From the numbers of readmissions we calculated that every 100 patients will experience 133 admissions for CHF during a defined 12-month period. This rate was referred to the NSW statistics to derive an estimate of 8523 patients, aged 65 years and older, admitted with CHF during 1993-1994. From the NSW population statistics for that year we calculated a hospitalization rate for this age group of 1162 per 100,000 per year. Allowing for differences in case definition, this is consistent with estimates we have published for the Hunter Region of NSW.7 The yearly rate for first admission with CHF was 465 per 100,000. Using the data on prevalence of use of NSAIDs and the adjusted ORs, we estimate that the excess risk of first admission with CHF for users of NSAIDs is approximately 600 per 100,000 users per year.
In this study we found that recent use of NSAIDs by elderly patients doubles the odds of being admitted to hospital with an episode of CHF. The estimated relative risk was higher for first admission with heart failure, and the risk of this outcome was increased substantially by NSAID use in those with a history of heart disease (OR, 10.5; 95% CI, 2.5-44.9). In these patients, both high-dose and long plasma drug half-life seemed to increase the risk of precipitation of CHF. Assuming the association between use of NSAIDs and CHF is unconfounded, the disease burden attributable to these drugs may be large—approaching the levels of morbidity and mortality that we have previously documented for serious upper gastrointestinal complications of NSAID use in NSW.9,16
This study has limitations that need to be considered when interpreting the results. Because it may be chronic, CHF is not an ideal condition to study using case-control methodology. It is difficult to specify the index day with confidence, and previous episodes of CHF may have led to interventions, including cessation of NSAID treatment. Previous mild episodes of CHF may be overlooked in both cases and controls.
We looked only at recent use of NSAIDs, and distinguishing between a true etiologic role for the drugs and an exacerbating effect in a predisposed individual is difficult. In practice, we think the latter is likely. There is no evidence that NSAIDs cause morphological changes in the heart through direct myocardial damage, nor are they thought to have a cumulative effect. Instead, the drugs are capable of inducing short-term hemodynamic changes through inhibition of prostaglandin synthesis.3 These considerations may have some general significance for the disease burden created by the drugs. It can be argued that many patients with exacerbations of CHF because of NSAID therapy would have gone on to develop this problem at a later stage without the drugs. In other words, the effect of NSAIDs has been to accelerate the problem.
An additional consideration is the tendency of NSAIDs to antagonize the actions of diuretics and ACE inhibitors.17 We did not specifically analyze for an interaction between NSAIDs and diuretics or ACE inhibitors in this study because the drugs are used a treatment for the condition that defines the cases, and they cannot be studied in both cases and controls.
Achieving a highly specific case definition was difficult in this study, and it is possible that we misclassified cases. We relied on a clinical case definition as this was easily operationalized and has been shown to have similar prognostic value to tests of ventricular function.18 Often the day of onset of CHF could not be specified with confidence, and there is a possibility that medical intervention had reduced NSAID use after the onset of CHF, but prior to recruitment, thereby diluting any apparent association.
Our study can also be criticized for choosing a control group from hospital inpatients. We recognize that this is not an ideal group, because the prevalence of use of NSAIDs by hospital patients may not be representative of the source population from which the cases were drawn. To minimize bias we excluded control patients whose primary reason for hospital admission was a condition that is known to be directly associated with the use of NSAIDs.19 We had previously recruited hospital controls in a case-control study of the relationship between consumption of NSAIDs and development of upper gastrointestinal tract bleeding and perforation, and found similar use of NSAIDs by these hospital controls, a simultaneously recruited community control group, and by a community survey sample.9,20 In the present work we used the similar selection criteria for choosing hospital controls.
An additional concern was that we would recruit controls who had unrecognized CHF at the time of enrollment, and we performed an audit of controls to minimize this source of bias. Cases were defined at entry by the presence of the relevant clinical features, and the Framingham diagnostic criteria were applied by the research nurses.7 In contrast, the controls were defined based on the absence of features of heart failure and often the necessary information was incomplete at the time of recruitment. Despite these efforts it is likely that some controls who remained in the study will have been potential cases and some cases may not have had CHF. This represents a potential bias to the null. In addition, we excluded controls if they had a previous hospitalization with CHF. This definition was probably fairly specific for heart failure, but insensitive in that we may have missed prior episodes of mild heart failure that were treated in the community and did not result in hospitalization. A similar exclusion rule was not applied to case patients. Consequently, our controls were most appropriate in the analysis of first-time heart failure cases. In the analysis of all subjects, the retention of cases (but not controls) with a history of hospitalization with CHF may have introduced an additional bias to the null, as some of the cases may have been advised to avoid treatment with NSAIDs. Consequently, we believe that, for a number of reasons, our estimates of risk may be conservative.
This study builds on experimental work that showed that the introduction of NSAID therapy in subjects with established or incipient heart failure produced an increase in systemic vascular resistance and a decline in the cardiac output.3 The present study is concerned with how often, and under what circumstances, these adverse effects occur, and with the size of the disease burden they create in a community with a high intake of NSAIDs.20,21 The existence of plausible mechanisms, the findings of dose-response relationships, and the nature of the effect modification point to a real effect of NSAIDs. In patients who had been admitted with CHF for the first time the dose-response relationship was steep in those who had a history of cardiac disease, but not in those without this history. These findings are consistent with an important effect of NSAIDs in patients with a damaged, but not previously decompensated, left ventricle.
NSAIDs had been used by nearly 30% of first-time cases and, assuming an unconfounded relationship, we calculate that approximately 19% of first episodes of CHF in our community are attributable to the use of these drugs. We made no direct measurement of the rate of hospitalization with CHF, and therefore we cannot estimate directly the attributable risks for an individual user of NSAIDs, or for the population. Accordingly, we used hospitalization data for NSW and from published cohort studies to derive these values.7,13,14 The estimated excess risk of first admission for CHF in NSAID users of about 500 per 100 000 per year is close to what we have previously calculated for major gastrointestinal tract bleeding and ulcer perforation in NSW.9,16 Although the epidemiological relationship between the use of these drugs and the heart failure is not nearly so well established as for gastrointestinal adverse effects, these data indicate that the adverse effects on the cardiovascular system potentially pose an equal threat to the public health.
The findings of this study are consistent with the recent report by Heerdink et al.4 Using a prospective observational design, these investigators found an approximate doubling of the risk of hospitalization with CHF when patients receiving diuretics were given NSAIDs. The risk was greatest in those who had existing CHF. The present study was not confined to users of diuretics, and indicates that there is an increased risk to subjects who have a previous history of heart disease, even if they have not previously had heart failure. The findings of these 2 studies have important clinical implications. Congestive heart failure is a common cause of morbidity and mortality in the community. Even a small relative risk will translate into a significant disease burden. Once a patient has been hospitalized with CHF for the first time, treatment is unlikely to restore the patient's function to the previous level. Thus, there is an imperative to avoid this development. The risk with NSAIDs seems to be concentrated in patients with a history of cardiac disease. If this is correct, it has significance for the ways in which NSAIDs are used by such patients. NSAIDs are used mainly for symptom relief, often seem to be used for minor ailments, and may be continued unnecessarily as long-term therapy. Arguably, guidelines should discourage the use of NSAIDs in individuals with a damaged but compensated left ventricle. We recommend that these drugs should be used with caution in such individuals, in the lowest possible dose, and that drugs with a long plasma half-life should be avoided. It is possible that drugs that are selective inhibitors of the inducible cyclo-oxygenase 2 will have a lower rate of adverse effects on the kidney and cardiovascular system, but this remains to be established in well-designed pharmacoepidemiological studies.22
Accepted for publication July 1, 1999.
This study was supported by grants from the National Health and Medical Research Council of Australia and Pfizer Australia Pty Ltd.
We are indebted to the general practitioners in the Hunter Region of New South Wales for their support of this project. We also wish to recognize the excellent work of Beverley Barry and Duncan Cruikshank, who worked as research assistants on this project.
Corresponding author: David Henry, MBChB, Discipline of Clinical Pharmacology, Clinical Sciences Building, Mater Misericordiae Hospital, Waratah New South Wales 2900, Australia (e-mail: firstname.lastname@example.org).
RO Non-steroidal anti-inflammatory drugs: differences and similarities. N Engl J Med.
1991;3241716- 1725Google ScholarCrossref
JS Renal syndromes associated with nonsteroidal anti-inflammatory drugs. N Engl J Med.
1984;310563- 572Google ScholarCrossref
et al. Prostaglandins in severe congestive heart failure. N Engl J Med.
1984;310347- 352Google ScholarCrossref
A NSAIDs associated with increased risk of congestive heart failure in elderly patients taking diuretics. Arch Intern Med.
1998;1581108- 1112Google ScholarCrossref
Studies of Left Ventricular Dysfunction (SOLVD) Investigators, Effect of enalapril on survival in patients with reduced left ventricular ejection fractions and congestive heart failure. N Engl J Med.
1991;325293- 302Google ScholarCrossref
CONSENSUS Trial Study Group, Effects of enalapril on mortality in severe congestive heart failure. N Engl J Med.
1987;3161429- 1435Google ScholarCrossref
et al. Management and outcomes of congestive heart failure: a prospective study of hospitalised patients. Med J Aust.
1998;168115- 118Google Scholar
WB The natural history of congestive heart failure: The Framingham Study. N Engl J Med.
1971;2851441- 1446Google ScholarCrossref
C Variability in the risk of major gastrointestinal complications from non-aspirin non-steroidal anti-inflammatory drugs. Gastroenterology.
1993;1051078- 1088Google Scholar
Not Available, Australian Prescription Products Guide. 24th ed. Hawthorn, Victoria Australian Pharmaceutical Publishing Co Ltd1995;
JJ Case Control Studies: Design, Conduct, Analysis. New York, NY Oxford University Press1982;
D Modelling Binary Data. London, England Chapman & Hall1991;
Not Available, NSW Inpatient Statistics Collection: Users Manual. Sydney, Australia New South Wales Dept of Health1992;
et al. Burdens and outcomes of hospitalisation for congestive heart failure. Med J Aust.
1997;16767- 70Google Scholar
et al. Estimating the population attributable risk for multiple risk factors using case-control data. Am J Epidemiol.
1985;122904- 914Google Scholar
J Non-steroidal anti-inflammatory drugs and peptic ulcer hospitalization rates in New South Wales. Gastroenterology.
1993;1041083- 1091Google Scholar
W Counteraction of the vasodilator effects of enalapril by aspirin in severe heart failure. J Am Coll Cardiol.
1992;201549- 1555Google ScholarCrossref
JP Prognosis in ischemic heart disease. Arch Intern Med.
1992;1522433- 2437Google ScholarCrossref
NE Statistical Methods in Cancer Research: Volume 1—The Analysis of Case-Control Studies. Lyon, France International Agency for Research on Cancer1980;25
R Drug use in Australia: a community prevalence study. Med J Aust.
1992;156759- 764Google Scholar
et al. Pattern of use of non-steroidal anti-inflammatory drugs in Australia 1990-1994. Med J Aust.
1996;164589- 592Google Scholar
JRI Selectivity of non-steroidal anti-inflammatory drugs as inhibitors of constitutive and inducible cyclo-oxygenase. Proc Natl Acad Sci U S A.
1993;9011693- 11697Google ScholarCrossref