• acetaminophen;
  • cost-effectiveness;
  • knee osteoarthritis;
  • NSAIDs;
  • selective COX-2 inhibitors


  1. Top of page
  3. Introduction
  4. Subjects and Methods
  5. Results
  6. Discussion
  7. References

Objective: The objective of this study was to conduct an economic evaluation of rofecoxib and celecoxib compared with high-dose acetaminophen or ibuprofen with and without misoprostol for patients with symptomatic knee osteoarthritis (OA).

Methods: A decision analysis model was designed over 6 months using two measures of effectiveness: 1) number of upper gastrointestinal (GI) adverse events averted; and 2) number of patients who achieved perceptible pain relief. Separate analyses were conducted for all patients and for those who did not respond to acetaminophen. Outcome probabilities were obtained from a comprehensive review of randomized controlled trials and observational studies. Costs were derived from actual resource utilization of OA patients.

Results: In terms of averting GI events, acetaminophen dominates the other options for an average risk patient population. For patients who did not respond to acetaminophen, rofecoxib had the lowest incremental cost-effectiveness ratio (ICER) per GI event avoided ($32,000) relative to ibuprofen. In terms of pain control, ibuprofen had an ICER of $610.77 per additional patient achieving minimal perceptible clinical improvement (MPCI) relative to acetaminophen, while rofecoxib had an ICER of $12,000 relative to ibuprofen. For patients who did not respond to acetaminophen and who are at high risk of developing an adverse GI event, rofecoxib dominates ibuprofen as the preferred alternative for both measures of effectiveness. One-way, two-way, and probabilistic sensitivity analyses established that these results were generally robust.

Conclusions: Our results suggest that for average-risk knee OA patients, acetaminophen dominates the other therapies in terms of cost per GI event averted. In terms of pain relief, cost-effectiveness acceptability curves indicate that if one values pain relief below $275 per patient achieving MPCI, acetaminophen is the therapy most likely to be optimal; between $275 and $14,150, ibuprofen is most likely to be optimal; and above $14,150, rofecoxib is most likely to be optimal.


  1. Top of page
  3. Introduction
  4. Subjects and Methods
  5. Results
  6. Discussion
  7. References

It has been estimated that approximately 21 mil-lion Americans have symptomatic osteoarthritis (OA) [1]. Nonsteroidal anti-inflammatory drugs (NSAIDs) have been the most widely used drugs to treat OA. Approximately 70 to 75 million NSAID prescriptions are written annually in the United States alone, at a cost of approximately $2 billion [2,3]. Over the past two decades, evidence documenting NSAID toxicity has been overwhelming. Notably, the majority of societal expenses attributable to NSAIDs are related to the treatment of adverse effects, particularly adverse gastrointestinal (GI) events [4]. In fact, acetaminophen is currently recommended as the first-line therapy for OA of the knee and hip according to the current treatment guidelines for the management of OA published by the American College of Rheumatology (ACR) [5] and others [6], partly due to the toxicity of NSAIDs.

NSAIDs inhibit cyclooxygenase (COX)-1 and COX-2 to varying degrees. The therapeutic effect of NSAIDs is derived from COX-2 inhibition, while many of the toxic effects result from COX-1 inhibition. The newer NSAIDs (e.g., celecoxib, rofecoxib) selectively inhibit COX-2 and have been shown to have effective anti-inflammatory and analgesic properties with fewer GI toxic effects [7,8]. However, these drugs are considerably more expensive than both traditional NSAIDs [9] and acetaminophen. Thus, there is a need to examine their cost-effectiveness (CE).

The objective of this study is to assess the CE of celecoxib and rofecoxib to treat symptomatic knee OA relative to the more conventional NSAIDs (ibuprofen) used with or without misoprostol prophylaxis and relative to acetaminophen. Methods of clinical decision analysis and economic evaluation were used, focusing largely on the direct costs of managing GI adverse events associated with the use of each drug. An institutional/payer perspective to the analysis was adopted.

Subjects and Methods

  1. Top of page
  3. Introduction
  4. Subjects and Methods
  5. Results
  6. Discussion
  7. References

Decision Analysis Model

We developed a decision analysis model (Fig. 1) using DATA Professional, version 4.0 (TREEAGE Software, Williamstown, MA, USA) to compare the costs and outcomes of five treatment options for knee OA: acetaminophen (1 g QID), ibuprofen (800 mg TID), ibuprofen (800 mg TID) plus misoprostol (200 μg TID), celecoxib (200 mg QD), and rofecoxib (25 mg QD). The target population for the study were patients, typically over 50 years of age, who had radiographically identified knee OA, for whom the above therapeutic drugs are indicated for pain relief. Our model was based, in part, on a previously published decision model [10], which reflected the actual management of patients with clinically suspicious or confirmed upper GI complications, enrolled in a large-scale multicenter trial.


Figure 1. Decision tree for OA treatment with pain reduction and adverse GI events averted as the effectiveness measures.

Download figure to PowerPoint

Two different effectiveness outcomes were examined: effectiveness of each drug in avoiding upper GI toxicity and in reducing pain. The first measure of effectiveness is defined as the number of confirmed symptomatic upper GI adverse events averted. The second measure of effectiveness is defined as the number of patients who achieved the minimum perceptible clinical improvement (MPCI) from a patient's perspective according to the Western Ontario and McMaster Universities Osteoarthritis Index (WOMAC) pain subscale [11].

Two separate analyses were conducted. The first focused on all OA patients where acetaminophen, ibuprofen, and selective COX-2 inhibitors were included in the analysis. The second was limited to a subgroup of OA patients for whom acetaminophen failed to provide adequate pain relief (acetaminophen nonresponders). Acetaminophen was not included in the analysis of nonresponders. The same outcomes were employed for both analyses.

The decision tree shows the sequence of events considered in our analyses, which may occur within 6 months of the initiation of each drug (Fig. 1). Patients using the drug may or may not experience an adverse event in the form of a clinically confirmed upper GI ulcer, perforation, obstruction, or bleed (PUB). The term “confirmed” was used to indicate that a symptomatic ulcer was ascertained by endoscopy or upper GI tract radiography to be a confirmed PUB. Some of these PUBs present as complications, namely perforation, bleeding, and/or obstruction, necessitating either surgery or inpatient medical management. The remaining or uncomplicated PUBs are symptomatic ulcers and could require outpatient management with or without endoscopy. We assume similar rates of resource use to treat confirmed PUBs associated with each drug.

Patients who do not experience confirmed PUBs may still be evaluated because of clinical suspicion of a PUB or because they present with other GI events such as dyspepsia, diarrhea, or reflux. Suspected PUBs could require inpatient or outpatient procedures with or without endoscopy. The other GI events (e.g., dyspepsia) are typically treated on an outpatient basis where a gastroprotective agent (GPA) is generally prescribed. We assume different rates of inpatient and outpatient procedures and GPA use to treat suspected PUBs and other GI events associated with each drug. Clinical outcomes at the end of 6 months included the number of confirmed PUBs experienced as well as pain relief experienced with use of each drug.

We considered only short-term, nondiscounted direct medical costs, which include the costs of the drugs and the costs associated with monitoring and treating adverse events associated with each drug. The latter costs include costs associated with hospitalization (including surgery), outpatient procedures and consultations, and coprescriptions of GPAs. Incremental CE ratios were calculated for the nondominated options as the additional cost per patient achieving a unit of effectiveness compared with the next less costly, nondominated option.

Data and Assumptions

The base-case estimates for each variable and the ranges used for sensitivity analyses are shown in Table 1.

Table 1.  Base-case probability estimates and their ranges for sensitivity analyses over a period of 6 months
VariableBaseline estimate (range)References
  • *

    Ranges for sensitivity analyses defined as 25% above and below baseline estimate.

  • Mayo Clinic cost utilization data (Cost data).

Pain efficacy: probability of achieving defined minimum perceptible clinical improvement in pain as measured on the WOMAC pain subscale
 Acetaminophen.75 (.68–.8)33–36
 Ibuprofen.83 (.8–.87)12,44
 Celecoxib.79 (.73–.85)17,35,43
 Rofecoxib.86 (.82–.90)12,35,44
GI toxicity
 Probability of confirmed PUB
  Acetaminophen.005 (.005–.01)7,13,14,18–20,23–26,28–32
  Ibuprofen.0205 (.009–.047)7,8,13–16,18–26,41
  Celecoxib.0097 (.0065–.021)7,13,16
  Rofecoxib.0092 (.0065–.021)8,12,14,15
  Effectiveness of misoprostol vs. placebo in preventing PUB.4 (.4–.75)27
 Probability of complications from confirmed PUB
  Acetaminophen.002 (.002–.004)7,13,14,18–20,23–26,28–32
  Ibuprofen.0075 (.0055–.02)7,8,13–16,18–26,41
  Celecoxib.0037 (.002–.0076)7,13,16
  Rofecoxib.0020 (.002–.0076)8,12,14,15
  Effectiveness of misoprostol vs. placebo in preventing complications.16 (0–.4)27
 Probability of (unconfirmed) suspected PUBs*
  Acetaminophen.02 (.015–.025)Assumed
  Ibuprofen.022 (.0165–.0275)27
  Ibuprofen with misoprostol.0176 (.0132–.022)27
  Celecoxib.02 (.015–.025)7
  Rofecoxib.02 (.015–.025)Assumed
 Probability of other GI events (e.g., dyspepsia)*
  Acetaminophen.23 (.115–.23)33–36,39,40
  Ibuprofen.23 (.115–.23)7,8,27,40
  Ibuprofen with misoprostol.31 (.155–.31)27
  Celecoxib.25 (.125–.25)7,35,37
  Rofecoxib.16 (.08–.16)8,35,38
Utilization of health-care resources
 Probability of hospitalization for a complicated PUB.63 (.56–.67)10
 Probability of surgery for complicated PUB.36 (.29–.39)10
 Probability of endoscopy for uncomplicated PUB (symptomatic ulcer).27 (.15–.35)10
 Probability of inpatient medical management of suspected PUB*
  Acetaminophen.0175 (.013–.022)40
  Ibuprofen.023 (.017–.028)10
  Ibuprofen with misoprostol.023 (.017–.028)10
  Celecoxib.012 (.009–.015)Assumed
  Rofecoxib.012 (.009–.015)41
 Probability of outpatient management of suspected PUB with endoscopy*
  Acetaminophen.155 (.12–.19)40
  Ibuprofen.158 (.12–.20)41
  Ibuprofen with misoprostol.08 (.06–.1)10
  Celecoxib.124 (.09–.15)Assumed
  Rofecoxib.124 (.09–.15)41
 Probability of outpatient management of other GI distress with GPAs*
  Acetaminophen.267 (.20–.33)40
  Ibuprofen.322 (.24–.40)41
  Ibuprofen with misoprostol.12 (.09–.15)10
  Celecoxib.255 (.19–.32)Assumed
  Rofecoxib.255 (.19–.32)41
Direct costs
 Medication costs ($) (cost of drug and dispensing fee      AWP price for  prescribed dose per patient for 6 months     )
  Acetaminophen30 (22.5–42)9
  Ibuprofen50 (35–86)9
  Misoprostol480 (320–640)9
  Celecoxib435 (326–544)9
  Rofecoxib435 (326–544)9
 GPAs for treatment of ulcers355.8 (177.9–533.7)9
 GPAs for treatment of other GI events32.4 (21.6–43.2)9
GI toxicity monitoring costs ($)
 Inpatient surgical management for confirmed PUB16,670 (14,235–19,105)(Cost data)
 Inpatient medical management for confirmed PUB12,980 (10,837–15,122)(Cost data)
 Outpatient management of confirmed PUB with endoscopy1587 (1,435–1740)(Cost data)
 Outpatient management of confirmed PUB without endoscopy1047 (950–1145)(Cost data)
 Outpatient management of suspected PUB with endoscopy1267 (1,150–1385)(Cost data)
 Outpatient management of suspected PUB without endoscopy709 (685–733)(Cost data)
 Inpatient management of suspected PUB11,365 (10,485–12,245)(Cost data)
 Outpatient physician consult47 (35–59)(Cost data)

Adverse GI events. The first measure of effectiveness was defined as the number of confirmed symptomatic PUBs averted. PUBs include symptomatic upper GI ulcers and ulcer complications (bleeding, perforation, obstruction). Four groups of adverse GI events were considered to determine the total cost of treating adverse GI events with the use of each drug: all confirmed PUBs, complicated confirmed PUBs, suspected but not confirmed PUBs, and other less serious adverse GI events, each of which have different rates of morbidity and health-care resource utilization. Baseline probability estimates (including confidence intervals CIs]) of adverse GI events for each branch of the tree were derived primarily from randomized controlled trials, population-based cohort studies, several published reports, and some unpublished reports [12,13]. We begin with the more recent evidence for the COX-2 inhibitor drugs versus NSAIDs.

PUB probability estimates for celecoxib were obtained from the recently published large safety trial CLASS study [7]. Food and Drug Administration reviews were used to extract 6 months of overall cumulative incidence data [13]. Estimates used for sensitivity analyses considered data from the nonaspirin group as well as the overall PUB rates from the CLASS study. PUB probability estimates for rofecoxib were obtained from a pooled analysis of Phase II/IIIA randomized controlled clinical trials in OA [14]. Six months of cumulative incidence data as presented in the New Drug Application (NDA) documents were also used [15]. Sensitivity analysis estimates were derived from the recent large safety trial in rheumatoid arthritis VIGOR study [8] and the CLASS study, assuming that the risk associated with rofecoxib is equal to celecoxib. PUB probability estimates for ibuprofen were also derived from 6 months of cumulative incidence data obtained from the CLASS study [7,13]. CIs for sensitivity analysis reflected the range of NSAID data available from the pooled analyses of the rofecoxib and celecoxib randomized controlled clinical trials [12,14,16,17]; the VIGOR study in rheumatoid arthritis [8], assuming that the risk of ulcer complications with ibuprofen may be comparable to naproxen; and observational cohort studies [2,18–26], which considered a variety of outcomes mainly reflective of ulcer complications. Estimates of the protective effect of misoprostol in the ibuprofen plus misoprostol arm were derived from the MUCOSA study [27]. We assumed the same protective effect of misoprostol on suspected PUBs as on confirmed PUBs.

PUB probability estimates for acetaminophen were calculated using ibuprofen data from the CLASS study [7,13], assuming that there was no risk of PUBs associated with acetaminophen exposure compared to a relative risk of up to 4 with ibuprofen. Recent data suggest that higher doses of acetaminophen also may be associated with a higher risk of PUBs [28,29]. The higher range for sensitivity analyses assumed a relative risk of 2 to accommodate findings of recent observational studies with high-dose acetaminophen [28–32].

Other GI events, which are not classified as confirmed PUBs (e.g., dyspepsia, abdominal pain, and diarrhea and referred to as GI distress in the decision trees), included all adverse events classified under the system organ class GI system disorders. Data from the clinical trials, which included acetaminophen [33–36] and product monograms [37,38], were used to derive estimates for acetaminophen, ibuprofen, celecoxib, and rofecoxib. For the ibuprofen plus misoprostol arm, we assumed a 35% increase in other GI events due to misoprostol-related GI events (e.g., diarrhea) [27]. The frequency of other GI events with ibuprofen in the MUCOSA study was 50% lower than the CLASS study. These data constituted the lower range for ibuprofen in sensitivity analyses. The acetaminophen, celecoxib, and rofecoxib estimates were also scaled down by a similar amount to generate lower ranges for sensitivity analysis. The higher range for the sensitivity analysis considered values reported in the acetaminophen trials [33–36,39,40].

Probability estimates of resource use (endoscopies, upper GI radiographic series, hospitalization, and surgery) for both complicated and uncomplicated PUBs were derived from the MUCOSA study [10,27]. We assumed that resource use for confirmed PUBs, which included surgical management, hospitalization (without surgery), and outpatient management (with and without endoscopy), would be the same for all drugs. Outpatient management also included the use of GPAs to treat ulcers. The baseline estimates refer to the estimates for resource utilization (pooled event rates) for all patients in the moderate and high-risk subgroups of the MUCOSA study, while the upper and lower sensitivity range estimates reflect event rates of all patients in the misoprostol and placebo arms, respectively.

Estimates of resource use for suspected PUBs and other GI events such as dyspepsia and abdominal pain were derived from the VIGOR [8] and CLASS [7] studies and a large-scale observational study on resource utilization of OA patients on NSAIDs compared to acetaminophen [40]. Resource use for suspected PUBs included hospitalization and outpatient management (with and without GI procedures) and the use of GPAs to manage these events. Resource use and medical management for these less serious adverse GI events were demonstrated to be different for different drugs [41]. We conservatively assumed the same rates of resource utilization for celecoxib as rofecoxib, derived from the VIGOR study. Resource use estimates for ibuprofen were also obtained from the VIGOR study. Differences in resource use between NSAIDs and acetaminophen were estimated based on the results of a recent observational study [40].

Pain efficacy. The second measure of effectiveness was defined as the number of patients who achieved the MPCI with the WOMAC pain subscale [11]. The MPCI represents the difference or change on the pain subscale associated with the smallest change in health states detectable by the patient. A previous study determined that MPCI is equivalent to a change of 9.7 units on a 0 to 100 VAS WOMAC pain scale [11].

The probability of a patient achieving MPCI was estimated assuming a normal distribution of mean change scores from baseline and by calculating the probability of mean change score being less than the MPCI, standardized by the mean and standard deviation. In other words,

  • image(1)

where Φ is the standard normal cumulative density function, μ is the mean change in scores from baseline, and σ is the standard deviation of mean in change scores. This method is similar to the those used in other CE studies [42] to assess the comparative efficacy of therapeutic options in a rheumatoid arthritis population using the ACR 20 improvement criteria.

Efficacy data were extracted from NDA reviews conducted by the FDA's Arthritis Advisory Committee for celecoxib and rofecoxib [12,17] publications of some of these development studies included in the NDA [43,44] and other recently published studies that included acetaminophen [35,36]. These data included the mean change on all efficacy scores from baseline and confidence intervals or standard deviations of these differences. Data were abstracted for all of the efficacy measures used in each study. After an extensive review of completeness of data and comparability of studies, we chose to use the WOMAC pain subscale.

Data from Geba et al. [35] were used for celecoxib and rofecoxib baseline estimates. The range for sensitivity analysis considered values calculated from Bensen et al. [43] for celecoxib and Day et al. [44] for rofecoxib. We also used the Day et al. study to derive estimates for ibuprofen. We assumed that the use of misoprostol had no impact on pain efficacy of ibuprofen. Data from Pincus et al. [36] and Geba et al. [35] were used to elicit the baseline estimate and range for sensitivity analysis of acetaminophen in terms of pain relief (Table 1).

Cost estimates. We used 5 years of institutional billing data for patients living in a single county who gave authorization for their data to be used for research purposes and who were treated at our institution with a diagnosis of OA to determine the cost of hospitalization and outpatient management of confirmed and suspected PUBs. To account for inflationary changes in the overall cost of providing care over the time period 1995 to 2000, we inflation-adjusted costs incurred each year to year 2000 levels using the All Urban Consumers U.S. City Average Consumer Price Index for Medical Care [45]. Because only outcomes within 6 months were considered, no further discounting was necessary. The robust 5% winsorized average episode costs were used for each group. The winsorized mean is a robust estimator of the central tendency of the distribution accounting for the presence of skewness by replacing a proportion of extreme observations with the value of the nearest available observation [46]. A range of two standard errors (mean) on both sides of the estimates was used as a range for sensitivity analysis.

The cost of outpatient management of confirmed and suspected PUBs included the cost of physician visits but did not include the cost of drugs. We assumed that an average patient would be treated with 40 mg of omeprazole per day for 60 days. For the ibuprofen plus misoprostol arm, we assumed that the treatment with misoprostol would be discontinued while the patient is on omeprazole.

For the cost of other GI adverse events, we assumed that patients seek a single consultation with a general practice physician. As a proxy for this cost, we used year 2000 Medicare Fee Schedule average reimbursement of $47 for CPT-499213 [47], which is an office or other outpatient visit for the evaluation and management of an established patient with problems of mild-to-moderate GI symptom severity, requiring approximately 15 minutes.

Medication costs were average wholesale prices (AWP) from the 2000 Red Book [9]. Costs of acetaminophen and ibuprofen were the lowest AWP for generic brands. The baseline cost for misoprostol was calculated at 600 μg per day. To determine the cost of treatment with GPAs over a 6-month period, we assumed that a patient experiencing other GI events (e.g., dyspepsia, diarrhea) would purchase the cheapest over-the-counter GPAs for an average consumption for 90 days: 400 mg cimetidine QD, 20 mg famotidine QD, or 150 mg ranitidine QD.

Base-case CE analyses. We conducted an analysis of incremental cost, incremental effectiveness, and incremental CE of each option relative to the next most costly option, using both measures of effectiveness. These analyses were performed for all patients and separately for acetaminophen nonresponders. Effectiveness values were assumed to be the same for both analyses.

Sensitivity analyses. One-way sensitivity analyses were performed on every variable in the model to examine the robustness of the results in terms of plausible variations in variable values in the model. Two-way and threshold sensitivity analyses were also performed on the variables which were identified as critical to the decision choice in one-way sensitivity analyses. In addition, we conducted a probabilistic sensitivity analysis using the Monte Carlo method to create 10,000 simulated trials of the patient cohort to assess the impact of simultaneous variations in the distribution of important variables around their point estimates. Using an acceptability curve approach [48], we also assessed how the proportion of the 10,000 simulated trials favoring one treatment arm relative to others varies as the monetary value placed on a unit of effectiveness changes.


  1. Top of page
  3. Introduction
  4. Subjects and Methods
  5. Results
  6. Discussion
  7. References

Base-Case CE Analyses

Using the baseline estimates in Table 1, costs, effectiveness, and incremental CE ratios for both measures of effectiveness are shown in Table 2. For a cohort of 1000 patients, acetaminophen generated the lowest total costs ($63,000), while ibuprofen plus misoprostol generated the highest total costs ($556,000). Table 2 includes the base-case CE results for all patients and for acetaminophen nonresponders with effectiveness defined as adverse events averted and percentage of patient achieving MPCI on the WOMAC pain scale, respectively.

Table 2.  Base-case total costs, effectiveness, and incremental CE of different therapeutic strategies in a hypothetical cohort of 1000 OA patients followed for 6 months for effectiveness defined as adverse events averted and pain efficacy
StrategyCost ($)Effectiveness=adverse events avertedEffectiveness=patients achieving MPCI response
EffectivenessIncremental CE ratioEffectivenessIncremental CE ratio
All patients*Acetaminophen nonrespondersAll patientsAcetaminophen nonresponders
  • *

    The difference in cost divided by the difference in effectiveness for each therapeutic strategy compared with the next best nondominated strategy. Dominance occurs when one strategy is always less costly and more effective than the others.

  • Note: Results are shown separately for all patients and for acetaminophen nonresponders.

Acetaminophen 63,000994.9  750  
Ibuprofen with prophylaxis556,000987.7DominatedDominated830DominatedDominated

For all patients, when effectiveness was defined as adverse events averted, the acetaminophen option was both the most effective (994.9 adverse events averted in a cohort of 1000) and the least costly ($63,000) of the five options. The other four options both cost more and were less effective; thus, these were “dominated” by acetaminophen. When we replicated the analysis with effectiveness defined as the number of patients achieving MPCI on the WOMAC pain scale, acetaminophen was again the least costly option. Ibuprofen, the next nondominated option increased total costs by $49,000 (i.e., $112,000-$63,000), but also increased the probability of achieving MPCI response by 8% (or 80 additional patients, i.e., 830–750, in a cohort of 1000), resulting in an incremental CE ratio of $610.77 per additional MPCI response relative to acetaminophen. Rofecoxib the other nondominated option further increased total costs and effectiveness by $359,000 and 3%, respectively, yielding an incremental CE ratio of approximately $12,000 over ibuprofen, the next cheaper option. The celecoxib and ibuprofen plus misoprostol options were both more costly and less effective than the other options and were therefore dominated. Another set of analyses focused on acetaminophen nonresponders and excluded acetaminophen as an option in the decision tree. The base-case results indicated that for both definitions of effectiveness, ibuprofen was the least costly ($112,000) and rofecoxib was the most effective (990.8 adverse events averted and 860 patients achieving MPCI on the WOMAC pain scale). Relative to ibuprofen, rofecoxib (the only nondominated option) had an incremental CE ratio of approximately $32,000 for every additional adverse event averted and $12,000 for every additional patient achieving MPCI on the WOMAC pain scale.

Sensitivity Analyses

One-way sensitivity analysis results remained robust for both models (with and without acetaminophen) across the specified range of all variables in the models. Threshold analysis on the model with effectiveness defined as GI events averted indicated that ibuprofen is the least costly option when the cost of acetaminophen increases beyond $80, compared to baseline generic brand price of $30. Similarly, when the probability of a confirmed PUB from acetaminophen is > .04 (compared to baseline estimate of .005) or when the probability of a suspected PUB from acetaminophen was over .07 (compared to a baseline estimate of .02), ibuprofen is the least costly option. When effectiveness was defined as the number of patients who receive perceptible pain relief, the results were robust across all variables except level of pain relief from acetaminophen. Threshold analysis indicated that when acetaminophen provided perceptible pain relief to less than 48% of patients, or when the probability of a confirmed PUB from acetaminophen was over .03 or a suspected PUB from acetaminophen was over .06, ibuprofen is the least costly option. These probabilities of confirmed or suspected PUBs from acetaminophen are compatible with a relative risk of 6.0 times above the general population [28,29]. Although it has been generally accepted that acetaminophen is not associated with an increased risk of adverse GI events, observational studies report relative risks as high as 3.6 at high dosages [28,29].

Threshold analysis of the models relating to acetaminophen nonresponders indicates that when effectiveness is defined as GI events averted, rofecoxib is less costly than ibuprofen to treat patients with a high risk of developing GI ulcers (around 14-fold higher than the average NSAID user population). This threshold value is higher when effectiveness is defined in terms of pain relief; that is, rofecoxib is less costly than ibuprofen when patients have 20-fold higher risk than the average NSAID user population.

Analysis of the model was repeated with higher estimates for both the probability of a confirmed PUB for acetaminophen (.009) and the probability of PUB complications from acetaminophen (.0035) to examine whether or not the higher estimates from recent observational studies were likely to impact our baseline results. When efficacy was defined as GI events averted, we found that acetaminophen continues to be the least costly and most effective, dominating the other options. When efficacy was defined in terms of pain relief, once again, acetaminophen was least costly; ibuprofen had an incremental CE ratio of $534 per additional MPCI response relative to acetaminophen.

Probabilistic sensitivity analysis [49] was conducted using the Monte Carlo method to repeat the simulation 10,000 times with a different set of randomly generated parameters drawing each parameter from a triangular prior distribution for each simulation. The resulting distribution of effectiveness and cost estimates (summarized in Table 3) captures the global effect of uncertainty about all model parameters.

Table 3.  Results of probabilistic sensitivity analysis using 10,000 Monte Carlo simulations
StrategyMeanSDMinimumMaximum% Optimal at willingness to pay*
Acetaminophen respondersAcetaminophen nonresponders
  • *

    Refers the percentage of the 10,000 Monte Carlo trials in which the indicated strategy had the highest net health benefit relative to all other strategies, when a one unit improvement in effectiveness is valued at $0, $300, $5,000, $15,000, or $25,000.

  • NA, not included in the analysis.

Effectiveness=adverse events averted
 Ibuprofen974.529.57935.1994.19  10080.050.9
 Ibuprofen with prophylaxis986.945.39958.09997.65     
 Rofecoxib987.624.02971.1995.65   10.325.7
 Celecoxib987.784.04971.79996.12   8.923.0
Effectiveness=patients achieving MPCI response
 Ibuprofen with prophylaxis833.4714.47800.23869.88     
 Rofecoxib860.0716.36820.99899.48  1.252.370
 Celecoxib789.7324.22730.84849.35   0.10.1
Costs ($, in thousands)
 Ibuprofen with prophylaxis556.5813.82519.09623.40     

The Monte Carlo simulation results of mean CE ratios, their distributions, and the percentage of each strategy is most likely to be the optimal choice of the decision maker's willingness to pay (WTP) for any incremental improvement in outcome (Table 3). In this analysis, we defined the dollar amounts that decision makers would be willing to pay for each GI event averted or for each person achieving an MPCI response. In probabilistic sensitivity analysis, a therapy is most likely to be optimal or preferred when it has the highest net health benefit of all therapies in the greatest proportion of the 10,000 simulated trials. As decision makers’ value of improvement in outcomes changes, so does the net health benefit of each treatment relative to the others. Therefore, in the probabilistic sensitivity analysis, the determination of the “most-likely to be optimal” therapy depends on the value of averting adverse events or pain improvement.

Table 3 indicates that when effectiveness was defined as the number of GI events averted, acetaminophen had the highest average net health benefit in 100% of the Monte Carlo simulations and therefore was most likely to be optimal for all patients, regardless of WTP. For acetaminophen nonresponders, ibuprofen was most likely to be optimal (100 and 80% optimal) when WTP for averting a GI event was $5,000 and $15,000, respectively, and borderline (50.9% optimal) when WTP for averting a GI event was $25,000. Ibuprofen was surpassed by rofecoxib when WTP exceeded $35,000. For effectiveness defined in terms of pain relief, when WTP per patient achieving MPCI was $300, ibuprofen was optimal in approximately 56.2% of the Monte Carlo simulations. For acetaminophen nonresponders, when WTP per patient achieving MPCI increased to $15,000, rofecoxib was optimal in 52.3% of the simulations, and at $25,000, rofecoxib was optimal in 70% of the simulations.

A CE acceptability curve generated from probabilistic sensitivity data, showing the thresholds of decision-makers’ value at which different therapies become most likely preferred (or optimal), is presented in Fig. 2. Acceptability curves simultaneously assess the impact of uncertainty about parameter values and uncertainty about decision-makers’ WTP for averting GI adverse events or pain improvement [50]. This analysis (Fig. 2) is shown only for effectiveness defined as improvement in pain, because acetaminophen was optimal for all values of WTP when effectiveness was defined as GI adverse event averted.


Figure 2. CE acceptability curves showing the values at which each of the therapeutic options becomes the optimal choice when effectiveness is defined as pain improvement.

Download figure to PowerPoint

The acceptability curves in Fig. 2 illustrate that acetaminophen was most likely to be optimal only when the value of an additional patient achieving MPCI was less than $275. Ibuprofen was most likely to be optimal until WTP exceeded $14,150, after which rofecoxib became the most likely to be optimal. It should be noted that the acceptability curve thresholds for determining the therapy most likely to be optimal ($275 and $14,150) differ from the base-case incremental CE ratios ($610 and $12,000). This difference occurs because the triangular distributions chosen for the random parameters in the probabilistic sensitivity analysis were asymmetric, so that the average parameter value in the probabilistic sensitivity analysis did not necessarily equal the base-case parameter value.


  1. Top of page
  3. Introduction
  4. Subjects and Methods
  5. Results
  6. Discussion
  7. References

Our objective was to illustrate the trade-offs in the costs and effectiveness of different drug options to treat a cohort of patients with OA over 6 months, specifically the incremental costs versus benefits of choosing the more recent therapeutic options over the current standard for treatment (NSAIDs with and without misoprostol and acetaminophen). Both adverse events averted and pain efficacy were used as measures of effectiveness, first on a model including all patients and then on a subset of patients for whom acetaminophen failed to provide adequate pain relief and was therefore excluded from the analysis.

Our analyses indicated that acetaminophen was the least toxic and the cheapest option and therefore the most cost-effective in averting major adverse events. It was also the most cost-effective in providing pain relief for OA patients. Acetaminophen remained the most cost-effective alternative even when we assumed an increased risk (i.e., twice the baseline estimate) of developing adverse GI events at high doses, as suggested by a recent observational study [28]. A switch to rofecoxib, the most effective drug at reducing pain, would come at considerable additional expense for patients who would otherwise respond to acetaminophen. These findings are consistent with the ACR recommendations [5].

Our analyses indicated that for patients who did not respond to acetaminophen, ibuprofen was the most cost effective in averting adverse GI events as well as in relieving pain, but these results were sensitive to a patient's relative risk of developing a GI adverse event. Our findings suggested that in high-risk patients (more that 14-fold higher than the average NSAID user population), rofecoxib was more cost-effective than ibuprofen with or without misoprostol. Such patient groups include the elderly with multiple risk factors such as multiple concomitant medications and history of GI problems including ulcers or bleeds. Taken together, these two findings underscore the need to carefully rule out acetaminophen and ibuprofen as treatment options before considering more costly alternatives. The latter agents should only be considered for patients at very high risk for ulcer perforations or bleeds (Fig. 3).


Figure 3. Decision grid based on CE findings.

Download figure to PowerPoint

The generic prices of acetaminophen and ibuprofen give these drugs a considerable advantage over the newer drugs. Our findings also indicated that the newer drugs, rofecoxib, celecoxib, and misoprostol, which cost approximately seven times that of the generic version of acetaminophen and ibuprofen, generated much less expenditure in treatment of the GI adverse events than those associated with ibuprofen. These drugs are only indicated for patients who are at very high risk of these adverse events, balancing the additional expense of drug costs with costs of treating adverse events.

The strength of our study lies in the use of a previously published decision model [10], which reflected the actual management of patients with clinically suspicious or confirmed upper GI complications, enrolled in a large-scale multicenter trial. Another strength of our study was the high quality of clinical and epidemiologic sources of probability estimates. The probability estimates used to inform the model were based on a comprehensive review of recent randomized controlled trials and observational studies. Pooled estimates were also used where available making results more widely generalizable to a broader range of patients, drug utilization patterns, and practice settings. Further, we used estimates of actual resource utilization rather than charges or expert opinions to assess costs. For a very small part of our tree, where data was unavailable, resource use was imputed from the published literature by making a few modeling assumptions. As far as possible, our sensitivity analyses incorporated the range of plausible estimates obtained from the literature.

The findings from our analyses both complement and extend the conclusions of previous studies examining the CE of these drugs. We compared our methodology and findings with those from the six published CE studies [51–56] addressing similar research questions. Two of these analyses were conducted in the United States [51,52], one in Canada [53], one in Switzerland [54], one in Sweden [55], and one in Norway [56]. The models used were largely the same as ours. The major differences between the models were the comparator drugs used in the models. Only Holzer et al. [51] examined acetaminophen, but they did not consider the COX-2 inhibitors.

Our findings are similar to those of Holzer et al. [51], which showed that acetaminophen was the least costly and most cost-effective alternative compared to NSAIDs with and without the use of GPAs. Our findings are also similar to Motheral et al. [52], who reported rofecoxib to be the most cost-effective relative to NSAIDs when acetaminophen was not included in the analysis. However, the time horizon of their model was 3 months. We also could not assess whether the drug costs in their model included the cost of GPAs to prevent ulcers, similar to the role of misoprostol in our model. Our findings are also similar to those of Zabinski et al. [53], where NSAIDs were found to be more cost-effective than celecoxib, which in turn was more cost-effective than the NSAIDs plus GPAs. In contrast, the findings of Chancellor et al. [54], Haglund and Svarvar [55], and Svarvar and Aly [56] indicated that celecoxib was the most cost-effective. The divergent findings of these latter studies could be attributable to differences in efficacy and toxicity estimates and possibly differences in the national health systems in which these studies were conducted.

Generic acetaminophen is most cost-effective, in terms of GI toxicity, to treat OA patients irrespective of the risk of GI ulcers, supporting the current guidelines for OA drug therapy. Ibuprofen is the least costly for patients who do not respond to acetaminophen; typically those patients with more severe pain. Rofecoxib is cost-effective for patients who both are at a higher risk of GI events and do not respond to acetaminophen (Fig. 3).

Our analysis is unique in two ways. First, we attempt to examine two outcomes, one of which to our knowledge has not previously been examined in economic analysis for OA, that is, pain relief. This strategy enables us to compare the incremental CE ratios for both measures of effectiveness. Second, our model examines a spectrum of drugs, including the ones recommended by ACR—that is, acetaminophen as a first line of treatment and the newer COX-2 inhibitors for patients at high risk of perforations, ulcers or bleeds. This enabled us to analyze the threshold risk level at which it becomes more cost-effective to use the newer drugs.

Our primary analysis focused on an OA patient cohort where the ACR recommends acetaminophen as the first-line treatment. Except for Holzer et al. [51], none of the other CE studies included acetaminophen as the drug of choice. In addition, all models assumed equal pain relief across arms. Ours does not because there is good evidence supporting differences in pain efficacy and patient preferences between acetaminophen and the other drugs used in the model [36,57,58]. In our study, we use pain relief as an outcome measure, like studies in rheumatoid arthritis. We also analyzed the drugs indicated for higher pain relief separately because of the different scale in the costs of these drugs. We were thus able to assess the CE for different patient populations (all patients vs. acetaminophen nonresponders; Fig. 3).

These results, however, must be interpreted in light of certain limitations:

  • First, we used a static, deterministic model to analyze the series of events over 6 months. A Markov model could have been useful in extending the findings of this study to a longer period, while at the same time taking into account patient compliance with drug therapy, switches between drugs [59], and long-term adverse effects. However, our probability estimates were derived from the existing literature and, therefore, limited to the duration of studies. Unfortunately, no long-term data are available to inform a Markov model.
  • Second, we limited our analysis to direct medical costs over a 6-month period. Again, because of the absence of data, a more comprehensive cost analysis was not feasible.
  • Third, the use of mostly clinical trial data may be a source of bias in estimating effectiveness as well as the true incidence of drug-related adverse events in the community. Patients enrolled in such trials may differ in important ways from population-based cohorts of patients with OA, thus introducing a potential referral bias. Thus, probability and cost estimates derived from trials may be biased. However, large trials (such as VIGOR and CLASS) more closely reflect the real-life setting than smaller clinical trials. In addition, we used data from observational studies to provide estimates for the ranges for sensitivity analyses.
  • Fourth, there is a need for a preference-based utility measure for OA that combines toxicity and effectiveness in pain relief outcomes in a single scale. This is an area for potential future research.
  • Fifth, we did not include in our model other potentially important adverse events, focusing instead on GI adverse events. Although recent data on rofecoxib and celecoxib suggest the importance of hepatic, renal, and cardiovascular adverse events, it is unclear to what degree the renal and hepatic laboratory abnormalities would manifest clinically and generate resource utilization. Moreover, the incidence of severe hepatic and renal disease is very low [60] (<1/1000 person-years) making these events relatively less influential from a CE point of view compared with GI adverse events. Regarding common and less severe forms of hepatic and renal adverse events (e.g., lab abnormalities), it is almost impossible to obtain exactly comparable data (inclusion of the same adverse events) across drugs. Cardiovascular adverse events would have been important to include from a CE point of view because the prevalence is high among the elderly. Evidence and quantitative findings for thrombotic cardiovascular events have been documented in the context of VIGOR and CLASS studies but quantitative data for similar events are scanty for NSAIDs and acetaminophen [61]. Inclusion of these adverse events is an important area for future research.
  • Finally, recent evidence strengthens the value of Helicobacter pylori eradication in reducing NSAID related GI toxicity [62,63], and eradication strategies need to be incorporated into future CE analyses.

In summary, our results suggest that for OA patients, acetaminophen dominates the other therapies in terms of cost per GI event averted, supporting the current guidelines for OA drug therapy. In terms of CE acceptability for average-risk patients who did not respond to acetaminophen, rofecoxib is preferred to ibuprofen only if the value of avoiding a GI event exceeds $30,000. In terms of pain control, if the value of pain relief is below $275 per patient achieving MPCI, acetaminophen is preferred; between $275 and $14,150, ibuprofen is preferred; and above $14,150, rofecoxib is preferred.

The authors thank Drs. Andreas Maetzel, Patrick Kamath, and Michele Doran who provided comments on earlier versions of this document and Dr. Claire Bombardier for her helpful guidance and consultation.

This study was funded in part by McNeil Consumer Healthcare.


  1. Top of page
  3. Introduction
  4. Subjects and Methods
  5. Results
  6. Discussion
  7. References
  • 1
    Lawrence, RC, Helmick, CG, Arnett, FC, Et, Al. Estimates of the prevalence of arthritis and selected musculoskeletal disorders in the United States. Arthritis Rheum 1998; 41: 77899.
  • 2
    Gabriel, SE, Fehring, RA. Trends in the utilization of non-steroidal anti-inflammatory drugs in the United States, 1986–1990. J Clin Epidemiol 1992; 45: 10414.
  • 3
    Frost & Sullivan Market Intelligence. Pain management markets (August 1993). Report no. 894–52. Mountain View (CA): Frost & Sullivan, 1995.
  • 4
    Rothschild, BM. Cost savings and prescribing protocols for nonsteroidal anti-inflammatory drugs. JAMA 1996; 276: 5256.
  • 5
    Anonymous. Recommendations for the medical management of osteoarthritis of the hip and knee: 2000 update. American College of Rheumatology Subcommittee on Osteoarthritis Guidelines. Arthritis Rheum 2000; 43: 190515.
  • 6
    Eccles, M, Freemantle, N, Mason, J. North of England evidence based guideline development project: summary guideline for non-steroidal anti-inflammatory drugs versus basic analgesia in treating the pain of degenerative arthritis. The North of England Non-steroidal Anti-inflammatory Drug Guideline Development Group. BMJ 1998; 317: 52630.
  • 7
    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. JAMA 2000; 284: 124755.
  • 8
    Bombardier, C, Laine, L, Reicin, A, Et, Al. Comparison of upper gastrointestinal toxicity of rofecoxib and naproxen in patients with rheumatoid arthritis. N Engl J Med 2000; 343: 15208.
  • 9
    Medical Economics, Company. 2000 drug topics red book. Montvale: The Company, 2000.
  • 10
    Maetzel, A, Ferraz, MB, Bombardier, C. The cost-effectiveness of misoprostol in preventing serious gastrointestinal events associated with the use of nonsteroidal antiinflammatory drugs. Arthritis Rheum 1998; 41: 1625.
  • 11
    Ehrich, EW, Davies, GM, Watson, DJ, Et, Al. Minimal perceptible clinical improvement with the Western Ontario and McMaster Universities osteoarthritis index questionnaire and global assessments in patients with osteoarthritis. J Rheumatol 2000; 27: 263541.
  • 12
    Anonymous. Vioxx (Rofecoxib) tablets review. Washington: Food and Drug Administration, Center for Drug Evaluation and Research, Division of Anti-Inflammatory, Analgesic, and Opthalmic Drug Products, 1999.
  • 13
    Witter, J. Celebrex capsules (Celecoxib). NDA 20–998/S-009. Washington: Medical Officer Review, Food and Drug Administration, Center for Drug Evaluation and Research, Arthritis Advisory Committee, 2001. p. 1–99.
  • 14
    Langman, M, Jensen, DM, Watson, DJ, Et, Al. Adverse upper gastrointestinal effects of rofecoxib compared with NSAIDs. JAMA 1999; 282: 1929 33.
  • 15
    Anonymous. VIOXX tablets: VIOXX oral suspension (Rofecoxib). VIOXX Gastrointestinal Outcomes Research Study (VIGOR) briefing information. NDA 21–052. Washington: Food Drug Administration, Center for Drug Evaluation Research, Arthritis Advisory Committee, 2001. p. 1–112.
  • 16
    Goldstein, JL, Silverstein, FE, Agrawal, NM, Et, Al. Reduced risk of upper gastrointestinal ulcer complications with celecoxib, a novel COX-2 inhibitor. Am J Gastroenterol 2000; 95: 168190.
    Direct Link:
  • 17
    Anonymous. Celebrex (celecoxib) capsules review: osteoarthritis statistical review. Washington: Food and Drug Administration, Center for Drug Evaluation and Research, Division of Anti-inflammatory, Analgesic, and Opthalmic Drug Products, 1998.
  • 18
    Hernandez-Diaz, S, Rodriguez, LA. Association between nonsteroidal anti-inflammatory drugs and upper gastrointestinal tract bleeding/perforation: an overview of epidemiologic studies published in the 1990s. Arch Intern Med 2000; 160: 20939.
  • 19
    MacDonald, TM, Morant, SV, Robinson, GC, Et, Al. Association of upper gastrointestinal toxicity of non-steroidal anti-inflammatory drugs with continued exposure: cohort study. BMJ 1997; 315: 13337.
  • 20
    Hallas, J, Lauritsen, J, Villadsen, HD, Gram, LF. Nonsteroidal anti-inflammatory drugs and upper gastrointestinal bleeding, identifying high-risk groups by excess risk estimates. Scand J Gastroenterol 1995; 30: 43844.
  • 21
    Lanza, FL, Codispoti, JR, Nelson, EB. An endoscopic comparison of gastroduodenal injury with over-the-counter doses of ketoprofen and acetaminophen. Am J Gastroenterol 1998; 93: 10514.
    Direct Link:
  • 22
    Smalley, WE, Ray, WA, Daugherty, JR, Griffin, MR. Nonsteroidal anti-inflammatory drugs and the incidence of hospitalizations for peptic ulcer disease in elderly persons. Am J Epidemiol 1995; 141: 53945.
  • 23
    Andrade, SE, Martinez, C, Walker, AM. Comparative safety evaluation of non-narcotic analgesics. J Clin Epidemiol 1998; 51: 135765.
  • 24
    Garcia, RLAG, Walker, AM, Gutthann, SP. Nonsteroidal antiinflammatory drugs and gastrointestinal hospitalizations in Saskatchewan: a cohort study. Epidemiology 1992; 3: 33742.
  • 25
    Lanza, LL, Walker, AM, Bortnichak, EA, Dreyer, NA. Peptic ulcer and gastrointestinal hemorrhage associated with nonsteroidal anti-inflammatory drug use in patients younger than 65 years. Arch Intern Med 1995; 155: 13717.
  • 26
    Henry, D, Lim, LL, Garcia Rodriguez, LA, Et, Al. Variability in risk of gastrointestinal complications with individual non-steroidal anti-inflammatory drugs: results of a collaborative meta-analysis. BMJ 1996; 312: 15636.
  • 27
    Silverstein, FE, Graham, DY, Senior, JR, Et, Al. Misoprostol reduces serious gastrointestinal complications in patients with rheumatoid arthritis receiving nonsteroidal anti-inflammatory drugs: a randomized, double-blind, placebo-controlled trial. Ann Intern Med 1995; 123: 2419.
  • 28
    Garcia Rodriguez, L, Hernandez-Diaz, S. The relative risk of upper gastrointestinal complications among users of acetaminophen and nonsteroidal anti-inflammatory drugs. Epidemiology 2001; 12: 5706.
  • 29
    Rahme, E, Pettitt, D, LeLorier, J. Determinants and sequelae associated with utilization of acetaminophen versus traditional nonsteroidal antiinflammatory drugs in an elderly population. Arthritis Rheum 2002; 46: 304654.
  • 30
    Laporte, JR, Carne, X, Vidal, X, Et, Al. Upper gastrointestinal bleeding in relation to previous use of analgesics and non-steroidal anti-inflammatory drugs. Catalan Countries Study on Upper Gastrointestinal Bleeding. Lancet 1991; 337: 859.
  • 31
    Matzke, GR. Nonrenal toxicities of acetaminophen, aspirin, and nonsteroidal anti-inflammatory agents. Am J Kidney Dis 1996; 28(Suppl 1): S6370.
  • 32
    Savage, RL, Moller, PW, Ballantyne, CL, Wells, JE. Variation in the risk of peptic ulcer complications with nonsteroidal antiinflammatory drug therapy. Arthritis Rheum 1993; 36: 8490.
  • 33
    Williams, HJ, Ward, JR, Egger, MJ, Et, Al. Comparison of naproxen and acetaminophen in a two-year study of treatment of osteoarthritis of the knee. Arthritis Rheum 1993; 36: 1196206.
  • 34
    Bradley, JD, Brandt, KD, Katz, BP, Et, Al. Comparison of an antiinflammatory dose of ibuprofen, an analgesic dose of ibuprofen, and acetaminophen in the treatment of patients with osteoarthritis of the knee. N Engl J Med 1991; 325: 8791.
  • 35
    Geba, GP, Weaver, AL, Polis, AB, Et, Al. Efficacy of rofecoxib, celecoxib, and acetaminophen in osteoarthritis of the knee—a randomized trial. JAMA 2001; 287: 6471.
  • 36
    Pincus, T, Koch, GG, Sokka, T, Et, Al. A randomized, double-blind, crossover clinical trial of diclofenac plus misoprostol versus acetaminophen in patients with osteoarthritis of the hip or knee. Arthritis Rheum 2001; 44: 158798.
  • 37
    Pharmacia, Corporation. US prescribing information for Celebrex [Internet]. Peapack (NJ): The Corporation, 2000 [revised 2000 Apr 24]. Available from:
  • 38
    Merck & Co. US prescribing information for Vioxx [Internet]. Witehouse Station: The Company, 2000.
  • 39
    Moore, N, Van Ganse, E, JM, LP, Et, Al. The PAIN study: paracetamol, aspirin, and ibuprofen new tolerability study—a large-scale randomized clinical trial comparing the tolerability of aspirin, ibuprofen and paracetamol for short-term analgesia. Clin Drug Invest 1999; 18: 8998.
  • 40
    Rahme, E, Joseph, L, Kong, SX, Et, Al. Gastrointestinal health care resource use and costs associated with nonsteroidal antiinflammatory drugs versus acetaminophen: retrospective cohort study of an elderly population. Arthritis Rheum 2000; 43: 91724.
  • 41
    Bombardier, C, Laine, L, Reicin, A, Et, Al. Fewer gastrointestinal protective agents, procedures, and hospitalizations with rofecoxib vs. naproxen in the VIGOR (Vioxx Outcomes Research) Study [Abstract]. Arthritis Rheum 2000; 43(Suppl 9): S225.
  • 42
    Choi, HK, Seeger, JD, Kuntz, KM. A cost-effectiveness analysis of treatment options for patients with methotrexate-resistant rheumatoid arthritis. Arthritis Rheum 2000; 43: 231627.
  • 43
    Bensen, WG, Fiechtner, JJ, McMillen, JI, Et, Al. Treatment of osteoarthritis with celecoxib, a cyclooxygenase-2 inhibitor: a randomized controlled trial. Mayo Clinic Proc 1999; 74: 1095105.
  • 44
    Day, R, Morrison, B, Luza, A, Et, Al. A randomized trial of the efficacy and tolerability of the COX-2 inhibitor rofecoxib vs ibuprofen in patients with osteoarthritis. Rofecoxib/Ibuprofen Comparator Study Group. Arch Intern Med 2000; 160: 17817.
  • 45
    US Department of Labor Bureau of Labor Statistics. All urban consumer price index for medical care. Washington: Bureau, 2001.
  • 46
    Humber, PJ. Robust estimation of a location parameter. Ann Math Stat 1964; 35: 73101.
  • 47
    Practice Management Information, Corporation. ADP/context medical fees in the United States. Los Angeles: The Corporation, 2000.
  • 48
    Van Hout, BA, Al, MJ, Gordon, GS, Rutten, FF. Costs, effects and C/E-ratios alongside a clinical trial. Health Econ 1994; 3: 30919.
  • 49
    Briggs, A. Handling uncertainty in cost-effectiveness models. Pharmacoeconomics 2000; 17: 479500.
  • 50
    Stinnett, AA, Paltiel, AD. Estimating CE ratios under second-order uncertainty: the mean ratio versus the ratio of means. Med Decis Making 1997; 17: 4839.
  • 51
    Holzer, SS, Cuerdon, T. Development of an economic model comparing acetaminophen to NSAIDs in the treatment of mild-to-moderate osteoarthritis. Am J Managcare 1996; II(Suppl S15): 26.
  • 52
    Motheral, BR, Bataoel, JR, Armstrong, EP. A strategy for evaluating the novel COX-2 inhibitors versus NSAIDs for arthritis. Hospital Form 1999; 34: 85563.
  • 53
    Zabinski, RA, Burke, TA, Johnson, J, Et, Al. An economic model for determining the costs and consequences of using various treatment alternatives for the management of arthritis in Canada. Pharmacoeconomics 2001; 19(Suppl 1): 4958.
  • 54
    Chancellor, JV, Hunsche, E, De Cruz, E, Sarasin, FP. Economic evaluation of celecoxib, a new cyclo-oxygenase 2 specific inhibitor, in Switzerland. Pharmacoeconomics 2001; 19(Suppl 1): 5975.
  • 55
    Haglund, U, Svarvar, P. The Swedish ACCES model: predicting the health economic impact of celecoxib in patients with osteoarthritis or rheumatoid arthritis. Rheumatology (Oxford) 2000; 39(Suppl 2): 516.
  • 56
    Svarvar, P, Aly, A. Use of the ACCES model to predict the health economic impact of celecoxib in patients with osteoarthritis or rheumatoid arthritis in Norway. Rheumatology (Oxford) 2000; 39(Suppl 2): 4350.
  • 57
    Pincus, T, Swearingen, C, Cummins, P, Callahan, LF. Preference for nonsteroidal antiinflammatory drugs versus acetaminophen and concomitant use of both types of drugs in patients with osteoarthritis. J Rheumatol 2000; 27: 10207.
  • 58
    Felson, DT. The verdict favors nonsteroidal antiinflammatory drugs for treatment of osteoarthritis and a plea for more evidence on other treatments. Arthritis Rheum 2001; 44: 147780.
  • 59
    Langman, M, Kahler, KH, Kong, SX, Et, Al. Drug switching patterns among patients taking non-streroidal anti-inflammatory drugs. a retrospective cohort study of a general practitioners database in the United Kingdom. Pharmacoepidemiol Drug Safety 2001; 10: 51724.
  • 60
    Walker, AM. Quantitative studies of the risk of serious hepatic injury in persons using nonsteroidal antiinflammatory drugs. Arthritis Rheum 1997; 40: 2018.
  • 61
    Ray, WA, Stein, CM, Hall, K, Et, Al. Non-steroidal anti-inflammatory drugs and risk of serious coronary heart disease: an observational cohort study. Lancet 2002; 359: 11823.
  • 62
    Huang, JQ, Sridhar, S, Hunt, RH. Role of Helicobacter pylori infection and non-steroidal anti-inflammatory drugs in peptic-ulcer disease: a meta-analysis. Lancet 2002; 359: 1422.
  • 63
    Chan, FKL, To, KF, Wu, JCY, Et, Al. Eradication of Helicobacter pylori and risk of peptic ulcers in patients starting long-term treatment with non-steroidal anti-inflammatory drugs: a randomised trial. Lancet 2002; 359: 913.