IJE Advance Access published online on February 8, 2008
International Journal of Epidemiology, doi:10.1093/ije/dym296
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
What is the harm–benefit ratio of Cox-2 inhibitors?
1 General Practice Research Database, Medicines and Healthcare products Regulatory Agency, London, UK.
2 Utrecht Institute for Pharmaceutical Sciences, Utrecht University, Utrecht, The Netherlands.
3 Department of Epidemiology and Population Health, London School of Hygiene and Tropical Medicine, London, UK.
4 Department of Medical Epidemiology and Biostatistics, Karolinska Institute, Stockholm, Sweden.
*Corresponding author. General Practice Research Database, Medicines and Healthcare products Regulatory Agency, 1 Nine Elms Lane, London SW8 5NQ, UK. E-mail: Tjeerd.vanstaa{at}GPRD.com
 |
Abstract |
|---|
 Top Abstract IntroductionMethodsResultsDiscussionAcknowledgementsReferences |
|---|
Background Selective cyclooxygenase-2 (Cox-2) inhibitors, developed to reduce the risk of NSAID-related gastrointestinal (GI) complications, have been associated with an increased risk of cardiovascular events. Our objective was to determine the balance of potential harm and benefit related to Cox-2 inhibitors’ exposure.
Methods The study population included patients aged 40+ years who received a prescription for Cox-2 inhibitors and were included in the General Practice Research Database. The incidence of upper GI events, myocardial infarction (MI) and stroke was estimated in this cohort. It was assumed that patients had experienced the upper GI and cardiovascular effects, as observed in clinical trials [relative rate (RR) of 0.49 for upper GI and 1.86 for MI]. Simulation methodology was used to estimate attributable risks, i.e. the difference between exposed and unexposed event probabilities.
Results The study population included 155 439 Cox-2 users. The number of upper GI events prevented by Cox-2 inhibitors was 179, while the number of excess MI cases was 83 per 10 000 patients treated for 4 years. A strong association was found between extent of GI benefit and cardiovascular harm. There was a large difference in the frequency of benefit over harm in only 6% of the patients (difference of 1% or more); 23% of the patients had more harm than benefit, including those with a history of ischaemic heart disease.
Conclusions The benefit of Cox-2 inhibitors in reducing the frequency of upper GI events may be offset by their cardiovascular harm, particularly in patients with risk factors for cardiovascular disease.
Keywords NSAIDs, Cox-2 inhibitors, risk, benefit, Myocardial infarction, Upper gastrointestinal event
Accepted 19 December 2007
|
Introduction |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionAcknowledgementsReferences |
|---|
Cyclooxygenase-2 (Cox-2) inhibitors are a selective type of non-steroidal anti-inflammatory drugs (NSAIDs). They were developed to minimize the upper gastrointestinal (GI) side-effects of conventional NSAIDs. Two large randomized clinical trials (RCTs) reported a halving of the risk of upper GI events in patients using selective Cox-2 inhibitors compared with conventional NSAIDs.1,2 Celecoxib was subsequently indicated in the UK for pain and inflammation in osteoarthritis or rheumatoid arthritis and rofecoxib for pain and inflammation in osteoarthritis. Extrapolating from the number of patients in the UK General Practice Research Database (GPRD), over 1 million patients were prescribed selective Cox-2 inhibitors in the UK since their launch in 2000. In the US in 2002, 34.7 million visits were made to community or hospital-based outpatients practices in which therapy with selective Cox-2 inhibitors was either initiated or continued.3
A possible signal of cardiovascular toxicity of selective Cox-2 inhibitors was observed in the first large trial with rofecoxib. Users of 50 mg rofecoxib daily had a 5-fold increased risk of myocardial infarction (MI) compared with naproxen users.1 This finding was, however, attributed to a possible protective effect of naproxen on the risk of MI, as naproxen may inhibit the production of thromboxane and inhibit platelet aggregation.4 Several epidemiological studies later reported similar signals of cardiovascular toxicity, especially at higher daily dosages of selective Cox-2 inhibitors.5 But epidemiological studies are considered by some to be prone to bias and not ‘robust’ enough for decision making and risk assessment.6 The signal of cardiovascular toxicity was finally confirmed in two recent RCTs, which were initiated to expand the indication of use of selective Cox-2 inhibitors to the prevention of colorectal adenomas.7,8
It has not yet been tested whether their benefit on reducing the probability of upper GI events is offset by any cardiovascular toxicity of selective Cox-2 inhibitors. In this study, we estimated the attributable risks for beneficial and adverse outcomes with selective Cox-2 inhibitors. Attributable risks are the probability of the occurrence of a particular event over a specific time period as a result of exposure. Attributable risks, rather than relative rates (RRs), are of key importance in the assessment of the harm and benefit of drug therapies. An adverse event with a large RR that occurs only rarely may be less important than an event with a small RR occurring frequently. We used simulation methodology to estimate the net public health impact of selective Cox-2 inhibitors using data from a large cohort of users in actual clinical practice.
|
Methods |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionAcknowledgementsReferences |
|---|
Study population
The study population consisted of men and women aged 40 years or older who received a first prescription for selective Cox-2 inhibitors and who were included in GPRD. The GPRD comprises the computerized medical records of general practitioners (GPs). In the UK, GPs are responsible for primary health care and specialist referrals. Patients are semi-permanently affiliated to a practice, which centralizes the medical information from the GPs, specialist referrals and hospitalizations. The data recorded in the GPRD include demographic information, including death, prescription details, clinical events, preventive care provided, laboratory results, specialist referrals, hospital admissions and their major outcomes.9
Overall design of decision model
We estimated in the study population the probabilities during current exposure to selective Cox-2 inhibitors of MI or unstable angina, stroke and upper GI events (including gastroduodenal ulcers and complications such as upper GI haemorrhage). These probabilities were estimated specific for age, gender and clinical risk factors. Using various assumptions for the effects of selective Cox-2 inhibitors (i.e. the RRs), we then estimated the probabilities of these outcomes if the patients would have been unexposed to selective Cox-2 inhibitors. If the effect of a drug (i.e. the RR) is known, the underlying (unexposed) event probability can be estimated by dividing the event probability as observed in the exposed GPRD patients through the RR. The attributable risk is the difference between the exposed and unexposed probability. As an example, if the observed event probability during exposure was 3% and if the RR of drug effect was 0.6 (i.e. 40% reduction of event probability due to drug exposure), the probability during non-exposure would have been 5%. The decision model estimated the number of GI cases prevented (benefit) and the number of excess cases of MI or stroke (harm) under different scenarios of beneficial and adverse effects of selective Cox-2 inhibitors. In other words, the decision model evaluated what would have been the harm–benefit ratio of selective Cox-2 inhibitors in a real-life study population under different scenarios of drug effect.
Scenarios of effects of selective Cox-2 inhibitors
Our decision model compared the potential harm and benefit of selective Cox-2 inhibitors to that of conventional NSAIDs (non-selective Cox-2 inhibitors). Four different scenarios of adverse and beneficial effects of selective Cox-2 inhibitors were evaluated in this study:
- RCT GI benefit—RCT cardiovascular harm. This analysis assumed that the effects of selective Cox-2 inhibitors had been similar in the study population to those found in RCTs. Information on the beneficial effects of selective Cox-2 inhibitors in reducing the risk of upper GI events was obtained from a meta-analysis of 11 RCTs that reported an RR of 0.49 on symptomatic ulcers in users of selective Cox-2 compared with conventional NSAIDs.10 The adverse effects on cardiovascular events were obtained from a meta-analysis of 121 RCTs that reported a RR of 1.86 for MI and RR of 1.02 for stroke with selective Cox-2 inhibitors compared with placebo.11
- RCT GI benefit—GPRD cardiovascular harm. In this scenario, the RR for upper GI events was based on the RCT meta-analysis10 and the RRs for cardiovascular harm were based on an analysis in GPRD comparing users of selective Cox-2 inhibitors to non-users who were matched by age, sex and propensity score for MI or stroke (this propensity score was estimated using the risk factors as outlined previously). In GPRD, the RR was 1.43 for MI and 1.13 for stroke. Although bias due to residual confounding may be likely in these observational RR estimates, the objective of the decision model was to estimate the possible impact of a signal.
- RCT GI benefit—midpoint cardiovascular harm. In this scenario, the RR for upper GI events was based on the RCT meta-analysis10 and the RRs for cardiovascular harm were based on the midpoint of the RRs in GPRD (comparing users of selective Cox-2 inhibitors to propensity matched controls) and no effect (i.e. RR of 1.22 for MI and 1.07 for stroke). This analysis assumed that only half of the adverse cardiovascular effects as observed in GPRD were related to the exposure to selective Cox-2 inhibitors.
- Midpoint GI benefit—GPRD cardiovascular harm. This scenario evaluated the impact of a smaller upper GI benefit of selective Cox-2 inhibitors. In this analysis, we used the midpoint of the RRs of upper GI events in the RCT meta-analysis10 and in GPRD comparing users of selective Cox-2 inhibitors with conventional NSAIDs (i.e. RR = 0.69). The RRs for cardiovascular harm were based on the GPRD analysis (RR of 1.43 for MI and 1.13 for stroke).
Event probabilities during exposure to selective Cox-2 inhibitors
The cumulative probabilities of each of the outcomes and of mortality during exposure to selective Cox-2 inhibitors were estimated in the study population using the survivor function in Cox proportional regression.12 For each set of patient characteristics, the Cox model allows calculation of an individual's probability of an outcome. The follow-up period was the time from the first prescription up to 3 months after the last selective Cox-2 prescription or the time of censoring, whichever date came first. The probabilities of these outcomes were individualized, i.e. estimated specific for age, gender, clinical risk factors and duration of therapy with selective Cox-2 inhibitors. The following clinical risk factors were considered in the analysis: smoking history and body mass index (where available), socioeconomic class and region of the practice location and number of visits to the GP in the 6–12 months prior. Risk factors specific for MI and stroke included history of diabetes, hypertension, systemic inflammation (rheumatoid arthritis or systemic lupus erythematosus), atrial fibrillation or heart valve disorders, ischemic heart disease, cerebrovascular disease, renal failure, bilateral oophorectomy and prescribing in the 6 months before of diuretics, statins, oral glucocorticoids, aspirin, anticoagulants and cardiac glycosides; for upper GI events, additional risk factors included history of diabetes, hypertension, systemic inflammation (rheumatoid arthritis or systemic lupus erythematosus), upper GI events or abdominal pain, rheumatoid arthritis or osteoarthritis and prescribing in the 6 months before of peptic ulcer healing drugs or antacids, oral glucocorticoids, aspirin, paracetamol, anticoagulants, statins, antidepressants and nitrates or calcium blockers. For risk factors with missing data, indicator variables for missing values were included in the regression models (smoking information was not recorded for 37.6% and BMI for 26.8% of the patients). There were only small differences in the parameter estimates between models excluding patients with missing alcohol or BMI information and models including indicators variables for missing values. In order to simplify the regression models and improve their fit to the data, forward selection of risk factors was used to identify the predictors for the various outcomes using a statistical significance level of 0.05. Almost all risk factors were selected and there were no major differences between the parameters in the model with all risk factors and that based on forward selection. In order to improve the fit of the models, we also investigated possible statistical interactions between risk factors and age and sex, which were included using a statistical significance level of 0.01 or lower. Various methods were used to test the fit of the final Cox regression models. The proportional hazards assumption was evaluated by visual examination of the Schoenfeld residuals. We also compared the observed 5-year probability of each of the outcomes (based on the Kaplan-Meier estimate) to the probability predicted by the Cox models. The individual probabilities for death were also estimated using similar methods and using the risk factors selected in the analyses for MI, stroke or upper GI.
Simulation model
Simulation methodology was used to compare the outcomes with selective Cox-2 inhibitors to those with conventional NSAIDs. Out of the study population, 5000 patients were randomly sampled. For each patient, the event probabilities during exposure to selective Cox-2 inhibitors and to conventional NSAIDs were estimated. The event probabilities during exposure to selective Cox-2 inhibitors were based on the estimates of the Cox regression models specific for the patient's characteristics. The event probabilities for conventional NSAIDs were based on the Cox regression estimates divided by the RR of drug effect. Over the course of the model (up to a maximum of 4 years of therapy), individual probabilities were adjusted, at each 1-month period, for increasing age. Using the individual event probabilities, each simulation estimated the number of cases with upper GI events, stroke or MI with each exposure. The differences in the number of cases with selective Cox-2 inhibitors and conventional NSAIDs were then estimated.
The random variability in the differences in the number of cases was determined as follows. The event probabilities were randomly selected from a normal distribution based on the mean and standard deviation of the parameter in the Cox regression models. The RR of drug effect was randomly selected from a normal distribution based on the RR and its 95% CI. Two-hundred different cohorts of 5000 patients were analysed and non-parametric bootstrapping techniques were then used to estimate the 95% confidence intervals (CI). The 95% CI was based on the 2.5 and 97.5% percentile of the distribution of the bootstrapping results.13
Weighting of harm and benefit
To establish an overall estimate of the harm–benefit ratio of selective Cox-2 inhibitors, each of the different outcomes was weighted by 1-year post-event mortality. Life expectancy has previously been used to standardize outcomes of different medical interventions.14 Patients with MI had the lowest 1-year survival and, therefore, MI was considered the reference event. Patients with an upper GI event had a 30% (RR = 0.70) lower 1-year probability of mortality. For this reason, 1.4 upper GI events were considered equivalent to 1 MI event (i.e. 1/RR).
|
Results |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionAcknowledgementsReferences |
|---|
Table 1 shows the characteristics of the selective Cox-2 users. The study population included 155 439 users. The mean age was 64.9 years and 64.5% were women. Only a minority of patients prescribed selective Cox-2 inhibitors had a history of osteoarthritis or rheumatoid arthritis.
|
As shown in Table 2, the number of upper GI events prevented by selective Cox-2 inhibitors was 179 cases per 10 000 patients treated for 4 years (using the scenario for drug effect of RCT GI benefit—RCT cardiovascular harm). The number needed to prevent one upper GI event over 4 years of selective Cox-2 therapy was 56. On the other hand, there would be an excess of 125 MI cases and the number needed to cause one MI was 80. Weighting events by mortality, the upper GI benefit of selective Cox-2 inhibitors did not substantially outweigh cardiovascular harm with this scenario of drug effect. When taking into account the real-life duration of use, both the number of GI events prevented by selective Cox-2 inhibitors and the excess number of cases of MI/stroke was low. In GPRD, only 16% of the patients who started selective Cox-2 inhibitors received treatment for >1 year.
|
Stratifying the cohort by baseline probability for upper GI events, benefits were largest in those with the highest baseline probability for upper GI events (Table 3). But this subgroup also experienced more MIs and strokes. Restricting selective Cox-2 inhibitors to patients with a low baseline MI probability would reduce the excess numbers of MIs and strokes. But the number of upper GI events prevented by selective Cox-2 inhibitors was also smaller.
|
Figure 1 shows the extent of heterogeneity in potential harm and benefit over 4 years with selective Cox-2 inhibitors using the four scenarios of drug effect. With the scenario for drug effect of RCT GI benefit—RCT cardiovascular harm, 23% of the patients would experience more harm than benefit (weighted by mortality). In only 6% of the patients, the difference between benefit and potential harm exceeded 1% in this scenario of drug effect.
|
Table 4 shows the predictors for benefit over harm of selective Cox-2 inhibitors. Patients with a history of ischaemic heart disease had potentially more harm than benefit compared with those without ischaemic heart disease, irrespective of the scenario of drug effect.
|
Figure 2 shows a two-way sensitivity analysis of the effects of changing the probabilities of upper GI events, MI and stroke. With the scenario for drug effect of RCT GI benefit—RCT cardiovascular harm, the number of upper GI events prevented was 257 per 10 000 over 4 years of treatment and the excess number of MI cases was 63 in case of an increase of 50% in the probability of upper GI events and a decrease of 50% of the MI/stroke probabilities. The weighted benefit minus harm would be 115 with these changes.
|
|
Discussion |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionAcknowledgementsReferences |
|---|
This study found that the benefit of Cox-2 inhibitors in reducing the frequency of upper GI events may be offset by their cardiovascular harm, particularly in patients with risk factors for cardiovascular disease even with RCT efficacy on upper GI events (compared with conventional NSAIDs). If the beneficial effects of Cox-2 inhibitors would be lower in actual clinical practice than observed in RCTs, the potential harm of Cox-2 inhibitors would outweigh their upper GI benefit.
This study used four different scenarios for the adverse and beneficial effects of selective Cox-2 inhibitors, including some derived from RCTs and some from observational comparisons. It is likely that the estimates of drug effects, which were based on observational comparisons may have been biased due to residual confounding. Users of selective Cox-2 inhibitors may have had higher probabilities of cardiovascular events due to underlying disease severity compared with controls. Statistical adjustment may reduce these baseline differences, but is unlikely to completely resolve this confounding. With respect to upper GI effects, one scenario was based on the midpoint between RCT efficacy and that observed in GPRD comparing selective Cox-2 inhibitors with conventional NSAIDs. Newly introduced drugs may be preferentially prescribed to patients with more complicated histories. It has also been reported that selective Cox-2 inhibitors were prescribed to patients at increased risk of upper GI adverse drug events.15 On the other hand, the drug effects as found in the RCTs may have high internal validity (due to randomization) but these estimates may not be fully generalizable to GPRD patients in actual clinical practice. The patients in the RCTs were restricted to patients with rheumatoid arthritis or osteoarthritis using selective Cox-2 inhibitors daily at high dosages for prolonged periods of time.1,2 In contrast, GPRD patients often used these drugs intermittently at lower dosages for short periods of time and most did not have rheumatoid arthritis or osteoarthritis. There is only limited RCT evidence for GI protective effects of selective Cox-2 inhibitors at lower dosages and with intermittent use. Given these difficulties in establishing drug effects in actual clinical practice, we used a range of estimates for the effects of selective Cox-2 inhibitors.
It is often argued that observational data are not ‘robust’ enough for regulatory decision-making and risk assessment and that observational data should only be used if causality is firmly established and that one should wait until evidence of highest quality appears.6 Unfortunately, RCTs are rarely conducted to specifically study safety concerns. As an example, the APPROVe study that evaluated the effects of rofecoxib in the prevention of colorectal adenomas was not initiated to address earlier signals of cardiovascular toxicity of rofecoxib, but to extend the market and indication for use.7 Furthermore, the introduction of medical (and other) technologies should be guided by the ‘precautionary principle’, which states that lack of full scientific certainty shall not be used as a reason for postponing action where there is a threat of serious damage and potentially major impact.16 The approach used in this study proposes to use the precautionary principle in the assessment of harm–benefit ratio of drug therapies. In this approach, the potential magnitude of harm is first established. For a signal with a potential magnitude of harm that may outweigh benefit, causality can then be determined in further studies that specifically address these signals. A ‘wait and see’ approach to signals with a potentially large harm seems less preferable, as absence of perfect (RCT) evidence is not evidence of absence of harm.
An advantage of the harm–benefit model used in this study is that the estimates for attributable risks can be individualized and estimated for specific patient groups in actual clinical practice. A recent analysis by Choi et al. evaluated the trade-off between upper GI benefit and MI harm of rofecoxib.17 The probabilities for upper GI and MI used in this decision analysis were based on the averages in the VIGOR RCT.1 But the average patient in actual clinical practice may be different from the average patient enrolled in a RCT. As an example, the incidence rate in GPRD of upper GI events in current users of selective Cox-2 inhibitors was about three times lower than that reported in the large RCTs with rofecoxib and celecoxib,1,2 while the rates of MI were similar. Also, the use of averages in decision modelling does not capture any variability in the harm–benefit profile across a population. The harm and benefit may not be evenly distributed across a population. Drug harm may occur more frequently in particular patient groups. We found that patients with risk factors for cardiovascular disease had a considerably more negative harm–benefit profile compared with patients without these risk factors. This suggests that safety studies in patients at high risk of cardiovascular disease could be useful before recommending long-term use of selective Cox-2 inhibitors in these patients.
One of the limitations of this study was that we did not differentiate between different types of selective Cox-2 inhibitors. The reason for this was that there was no difference in the RR of MI between rofecoxib and celecoxib in GPRD. This is consistent with a recent meta-analysis of RCTs.11 However, there are other data that indicate different cardiovascular effects of celecoxib and rofecoxib, including a study by Solomon et al.18 Another potential limitation is that we assumed immediate cardiovascular toxicity of selective Cox-2 inhibitors. The APPROVe study found a statistically different risk in MI only after 18 months of rofecoxib therapy, although this study was not powered to statistically detect earlier effects, with wide confidence intervals indicating that observed results were consistent with proportional hazards over time.7 Another RCT with rofecoxib reported an adverse effect on MI within 6 months1 and a meta-analysis found no difference between short- and long-term studies.19 In GPRD, the RR of MI did not vary with duration of selective Cox-2 therapy. Another limitation of this study was that we did not consider any adverse effects of conventional NSAIDs on the risk of stroke or MI compared with no treatment. There is some epidemiological evidence suggesting that the risks of MI may also be increased with conventional NSAIDs.11 Our analysis was restricted to evaluating the harm–benefit ratio of selective Cox-2 inhibitors relative to conventional NSAIDs. An evaluation of the harm–benefit ratio of selective Cox-2 inhibitors relative to no treatment would require information on the level of pain and quality of life in patients using NSAIDs, which is not available in GPRD.
In conclusion, the benefits of selective Cox-2 inhibitors in reducing the frequency of upper GI events may be offset by their cardiovascular harm, particularly in patients with risk factors for cardiovascular disease. Patient selection may be important in ensuring that the potential cardiovascular harm of selective Cox-2 inhibitors does not outweigh their upper GI benefit.
|
Acknowledgements |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionAcknowledgementsReferences |
|---|
The views expressed in this paper are those of the authors and do not reflect the official policy or position of the Medicines and Healthcare products Regulatory Agency, UK.
Conflict of interest: GPRD receives funding from, among others, several pharmaceutical companies. No other authors have declared any conflicts of interest.
|
References |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionAcknowledgementsReferences |
|---|
1 Bombardier C, Laine L, Reicin A, et al. Comparison of upper gastrointestinal toxicity of rofecoxib and naproxen in patients with rheumatoid arthritis. VIGOR Study Group. N Engl J Med (2000) 343:1520–28.
2 Silverstein FE, Faich G, Goldstein JL, et al. Gastrointestinal toxicity with celecoxib vs nonsteroidal anti-inflammatory drugs for osteoarthritis and rheumatoid arthritis: the CLASS study: a randomized controlled trial. Celecoxib Long-term Arthritis Safety Study. J Am Med Assoc (2000) 284:1247–55.
3 Dai C, Stafford RS, Alexander C. National trends in cyclooxygenase-2 inhibitor use since market release. Arch Intern Med (2005) 165:171–77.
4 Van Hecken A, Schwartz JI, Depre M, et al. Comparative inhibitory activity of rofecoxib, meloxicam, diclofenac, ibuprofen, and naproxen on COX-2 versus COX-1 in healthy volunteers. J Clin Pharmacol (2000) 40:1109–20.[Abstract]
5 Arellano F. Observational studies and the withdrawal of rofecoxib. Pharmacoepidemiol Drug Saf (2006) 15:203–5.[CrossRef][Web of Science]
6 Watson DJ, Santanello NC. Observational studies and the withdrawal of rofecoxib. Pharmacoepidemiol Drug Saf (2006) 15:199–201.[CrossRef][Web of Science][Medline]
7 Bresalier RS, Sandler RS, Quan H, et al. Cardiovascular events associated with rofecoxib in a colorectal adenoma chemoprevention trial. N Engl J Med (2005) 352:1092–102.
8 Solomon SD, McMurray JJ, Pfeffer MA, et al. Cardiovascular risk associated with celecoxib in a clinical trial for colorectal adenoma prevention. N Engl J Med (2005) 352:1071–80.
9 Walley T, Mantgani A. The UK General Practice Research Database. Lancet (1997) 350:1097–99.[CrossRef][Web of Science][Medline]
10 Hooper L, Brown TJ, Elliott R, Payne K, Roberts C, Symmons D. The effectiveness of five strategies for the prevention of gastrointestinal toxicity induced by non-steroidal anti-inflammatory drugs: systematic review. Br Med J (2004) 329:948–57.
11 Kearney PM, Baigent C, Godwin J, Halls H, Emberson JR, Patrono C. Do selective cyclo-oxygenase-2 inhibitors and traditional non-steroidal anti-inflammatory drugs increase the risk of atherothrombosis? Meta-analysis of randomised trials. Br Med J (2006) 332:1302–8.
12 van Staa TP, Geusens P, Pols HA, de Laet C, Leufkens HG, Cooper C. A simple score for estimating the long-term risk of fracture in patients using oral glucocorticoids. Q J Med (2005) 98:191–98.[Web of Science]
13 Glick HA, Briggs AH, Polsky D. Quantifying stochastic uncertainty and presenting results of cost-effectiveness analyses. Expert Rev Pharmacoeconomics Outcomes Res (2001) 1:25–36.[CrossRef]
14 Wright JC, Weinstein MC. Gains in life expectancy from medical interventions – standardizing data on outcomes. N Engl J Med (1998) 339:380–86.
15 MacDonald TM, Morant SV, Goldstein JL, Burke TA, Pettitt D. Channelling bias and the incidence of gastrointestinal haemorrhage in users of meloxicam, coxibs, and older, non-specific non-steroidal anti-inflammatory drugs. Gut (2003) 52:1265–70.
16 United Nations Environment Programme: Rio Declaration on Environment and Development. Available at: http://www.unep.org/Documents.multilingual/Default.asp?DocumentID=78&ArticleID=1163 [downloaded September 13, 2006].
17 Choi HK, Seeger JD, Kuntz KM. Effects of rofecoxib and naproxen on life expectancy among patients with rheumatoid arthritis: a decision analysis. Am J Med (2004) 116:621–29.[CrossRef][Web of Science][Medline]
18 Solomon DH, Schneeweiss S, Glynn RJ, et al. Relationship between selective cyclooxygenase-2 inhibitors and acute myocardial infarction in older adults. Circulation (2004) 109:2068–73.
19 Juni P, Nartey L, Reichenbach S, Sterchi R, Dieppe PA, Egger M. Risk of cardiovascular events and rofecoxib: cumulative meta-analysis. Lancet (2004) 364:2021–29.[CrossRef][Web of Science][Medline]
CiteULike 
Connotea 
Del.icio.us What's this?
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||