Cost-utility of celecoxib use in different treatment strategies for osteoarthritis and rheumatoid arthritis from the Quebec healthcare system perspective

Abstract

Objective: To assess the cost-utility of celecoxib in three treatment strategies for arthritis in Quebec, considering both upper gastrointestinal (GI) and cardiovascular (CV) events.

Methods: A Markov analytic framework was used to model patients with osteoarthritis and rheumatoid arthritis at low/average and high risk of GI and CV toxicity over 5 years with monthly cycles. Treatment strategies were modelled in line with Canadian clinical practice. In first-line treatment, patients started on celecoxib; second-line, patients started on a non-selective non-steroidal anti-inflammatory drug (NSAID) and switched to celecoxib after a first GI event; third-line, patients started on a non-selective NSAID, added a proton pump inhibitor (PPI) after a first GI event, and switched to celecoxib after a second GI event (while maintaining the PPI). Model inputs were determined through comprehensive literature searches (MEDLINE and EMBASE) from 1995 to 2006. Included studies evaluated GI (dyspepsia, uncomplicated and complicated ulcers, death) and CV (myocardial infarction, stroke, death) events. Drug and procedure costs were derived from Canadian published sources (Can$2005).

Results: Total costs per patient for celecoxib first-, second-, and third-line treatment were Can$4,790, $3,390, and $3,466, and total quality-adjusted life-years (QALY) were 3.251, 3.231, and 3.230, respectively. In all risk categories, celecoxib second-line was less costly and as effective as celecoxib third-line, producing savings to the healthcare system. Although celecoxib first-line generated incremental expenditures versus celecoxib second-line, it was also more effective. The resulting cost-utility ratio for the high-risk population was Can$54,696/QALY. Based on this analytical approach, a treatment strategy where celecoxib is used before the combination of a non-selective NSAID plus a PPI possesses cost advantages for the Quebec provincial drug programme. One-way sensitivity analysis (varying GI and CV event rates, utilities, and cost) generally showed second-line treatment with celecoxib as the dominant strategy compared with third-line treatment with celecoxib.

Conclusion: Although effectiveness of second- and third-line celecoxib use is similar, total cost is lower for second-line. These results suggest that the use of celecoxib before the combination of a non-selective NSAID plus a PPI is relatively cost-effective in the treatment of arthritis pain and support the full benefit listing of celecoxib in Quebec's drug programme.

Key words::

• celecoxib
• cost-effectiveness
• cost-utility
• NSAID
• osteoarthritis
• Quebec
• rheumatoid arthritis

Introduction

Osteoarthritis (OA) and rheumatoid arthritis (RA) are two of the most common forms of arthritis, a widespread and potentially disabling condition. Non-selective non-steroidal anti-inflammatory drugs (NSAIDs) and cyclooxygenase 2 (COX-2) selective NSAIDs^1–3 are often used in these conditions as symptomatic treatment to decrease pain and indirectly improve functional capacity. Arthritis management and arthritis-related disability place a significant medical and economic burden on society⁴.

Inhibition of the COX-1 isoenzyme during NSAID use results in depletion of prostaglandins in the gastric mucosa and an increased potential for gastrointestinal (GI) toxicity⁵; more specifically, adverse GI events can range from minor discomfort to more serious GI complications such as intestinal perforations, ulcers and bleeds (PUB)⁶. In Canada, NSAID-induced complications are estimated to cause 3,897 hospitalisations and 365 deaths annually⁷. Overall, NSAID-related GI events place a substantial burden on healthcare budgets.

COX-2 selective NSAIDs such as celecoxib have been associated with a superior upper and lower GI safety and tolerability profile^8,9 and lower healthcare resource utilisation related to GI toxicity^10–12. However, recent reports suggesting an increased risk of cardiovascular (CV) adverse events for both COX-2 selective and non-selective NSAIDs^13–16 have led to confusion over the most appropriate pain management in patients with OA and RA. In the updated evidence-based recommendations of the Canadian Consensus Group on the use of non-selective and COX-2 selective NSAIDs in the treatment of OA and RA³, COX-2 selective NSAIDs are still considered the agents of choice in patients with risk factors for PUB, depending on the CV risks of the patient.

Most economic analyses of the cost-effectiveness of COX-2 selective NSAIDs in the treatment of OA and RA have been carried out from the GI perspective^12,17. COSMO (Cost-Utility Outcomes Simulation Model for OA and RA patients), a Markov decision-analytic model, was constructed to assess the cost-effectiveness of celecoxib in different treatment strategies, incorporating both upper GI toxicity and CV events¹⁸. This model can assist decision makers, clinicians and third-party payers in adjudicating the economic benefit of one treatment strategy compared with another.

The objective of this study was to utilise COSMO in assessing treatment strategies for OA/RA involving celecoxib as first-, second- and third-line treatment, from the perspective of the Quebec provincial drug programme.

Methods

Model design

A Markov decision-analytic model was developed to quantify the cost-utility of celecoxib as first-, second- and third-line treatment options from the perspective of a third-party payer (the Quebec provincial drug programme). This analysis was carried out over a 5-year period, with monthly cycles, to reflect the chronic nature of both OA and RA (Figure 1).

Populations

Two patient subpopulations were simulated: (1) arthritis patients aged <65 years, and (2) those aged ≥65 years. Patients in the younger subpopulation were considered to be at low-to-average risk of developing an adverse event (GI or CV) while on NSAID therapy, and those in the elder subpopulation to be at high risk of an adverse event. The 65 years cut-off is supported by the Canadian Consensus Group on the use of NSAIDs³. Furthermore, several publications have shown age to be a major risk factor for the development of upper GI^19–23 and CV²⁴ adverse events during NSAID treatment. The younger subpopulation encompassed 60% of all patients²⁵.

Treatment strategies

Three treatment strategies were defined based on the occurrence of GI or CV events. A GI event was defined as an ulcer (complicated or uncomplicated) or GI discomfort (dyspepsia only). A CV event was defined as myocardial infarction (MI), stroke and CV death. In the first-line treatment strategy, patients initiated treatment with celecoxib. A proton pump inhibitor (PPI) was added to their treatment if patients experienced a first GI event; after a second GI event, patients remained on the combination of celecoxib plus a PPI (unless the GI event was a second ulcer resulting in discontinuation of treatment).

In the second-line treatment strategy, patients initiated treatment with a non-selective NSAID (ibuprofen, diclofenac or naproxen). Following a first GI event (other than a complicated ulcer), the non-selective NSAID was replaced by celecoxib; following a second GI event, a PPI was added to their treatment (unless the GI event was a second ulcer, resulting in discontinuation of treatment). If the first GI event was a complicated ulcer, the non-selective NSAID was replaced directly by celecoxib plus a PPI. This second-line strategy is considered the study reference scenario as it more closely depicts the current utilisation of celecoxib in Quebec²⁶.

In the third-line treatment strategy, patients initiated treatment on a non-selective NSAID. Following a first GI event a PPI was added to their treatment; following a second GI event, treatment was replaced by celecoxib plus a PPI (unless the GI event was a second ulcer resulting in discontinuation of treatment). This third-line strategy was considered as the main comparator because it depicted the potential utilisation of celecoxib if access were to be restricted by the province of Quebec's drug programme.

If a patient experienced a CV event or a second ulcer in any of the three strategies, treatment was discontinued; the patient was considered as having ‘uncontrolled arthritis’ and was not allowed to resume therapy for the remainder for the 5-year model time frame. The treatment strategies were validated by an independent clinical expert to ensure maximum clinical relevance.

Model inputs

Model inputs included efficacy; upper GI (dyspepsia, uncomplicated and complicated ulcers, death), CV (MI, stroke, other CV event, CV death), and overall death event rates; utility values; and costs (medication and clinical events).

Efficacy

The model assumed equivalent effectiveness of celecoxib and non-selective NSAIDs for the treatment of arthritis pain. This was based on the results of published studies that used standard validated methods to assess the efficacy of celecoxib compared with non-selective NSAIDs in patients with OA and RA^8,27–30. It was assumed that no patient discontinued celecoxib or non-selective NSAID treatment because of loss of efficacy. Discontinuation was a consequence of adverse events.

Literature search

GI and CV event rates for celecoxib and non-selective NSAIDs were determined through a comprehensive MEDLINE and EMBASE search from 1995 to 2005/2006, for relevant comparative studies (meta-analyses, clinical trials and cohort studies), including abstracts. Definitions of GI events were based on the Celecoxib Long-term Arthritis Safety Study (CLASS) study, a 6-month randomised controlled trial in OA and RA patients to compare the upper GI toxicity associated with celecoxib and non-selective NSAIDs (ibuprofen and diclofenac)⁵. Definitions of CV events were the same as used in the Anti-platelet Trialists’ Collaboration (ATPC) study³¹. For GI events, the search criteria were comparative studies (celecoxib vs. any NSAID) reporting GI events (dyspepsia, uncomplicated and complicated ulcers and death) in an OA/RA population. Only studies presenting data in time-dependent GI events (clinical events per patient year) were considered. For CV events, the search criteria were all studies in any patient population using celecoxib or NSAIDs reporting the following clinical events: MI, angina, stroke, CV disease, congestive heart failure, coronary artery disease, acute coronary syndrome, or cerebrovascular disease. However, only studies evaluating MI, stroke and CV death outcomes, and those for which the rate of CV events (events per 100 patient years) could be determined, were considered as inputs in the model.

Event rates and related inputs

The refined literature search for incident GI events found only one study (the CLASS trial⁵), out of 15 initially identified, that provided all GI event rates as time-dependent events. For CV events, only one study (a meta-analysis of the CV safety of celecoxib involving 41 randomised controlled trials with 44,330 patients³²) out of 19 initially identified met APTC end points and provided outcomes in an event per time horizon format. Table 1 shows the event rate inputs for GI and CV events.

Some additional GI parameters were also included in the model: the percentage of patients with a complicated ulcer who required hospitalisation was set at 62.7%, and of these 8.5% required surgery³⁴. The distribution of complicated ulcers was set at 91% for bleeding ulcer, 9% for obstruction and 0% for perforation⁵, with an overall complicated ulcer death rate of 4.9%³⁵.

Event rate assumptions

CV and GI events were assumed to be mutually exclusive and, unless a patient experienced a GI or CV event, long-term event rates remained constant over the 5-year period^13,36,37. Discontinuation due to non-GI- or CV-related adverse events was not considered. Nor did the model consider the specific safety profile of patients taking low-dose aspirin. Based on the findings of a meta-analysis in which no significant difference in risk was found^32,33, the same CV event rates were used for celecoxib and non-selective NSAIDs in the base-case analysis.

GI event rates in base-case analysis for non-selective NSAIDs were based on those for ibuprofen (800 mg three times per day) and diclofenac (75 mg twice a day)⁵, while rates for celecoxib related to a supratherapeutic dose (400 mg twice a day)⁵, twice the recommended dose for RA and four times the dose for OA. Treatment with a non-selective NSAID plus PPI was assumed to be associated with rates of dyspepsia and uncomplicated ulcer equal to those for celecoxib (expert opinion). For a complicated ulcer, the event rate was 25% higher for the treatment with non-selective NSAID plus PPI than for celecoxib³⁸. The incidence of a complicated ulcer in patients taking celecoxib plus a PPI was assumed to be 0%³⁹; this assumption was also applied to uncomplicated ulcers. PPI use in patients taking celecoxib was also assumed to confer a 50% reduction in the risk of dyspepsia (expert opinion). In the model, deaths caused by a clinical GI event occurred following a complicated ulcer only. The older subpopulation (patients aged ≥65 years) was assumed to have a 2.75 times (midpoint risk) increased risk of clinical GI events (complicated/uncomplicated ulcer)⁴⁰. Following a GI event (complicated/uncomplicated ulcer), those at both low/average and high GI risk were assumed to have a 3.25 times increased risk of subsequent upper GI events⁴⁰. For patients experiencing a CV event, it was assumed they would discontinue treatment after a first CV event. However, these patients still continued transitioning through the model, accumulating quality-adjusted life years (QALYs), until the end of the model time horizon. As a consequence, the risk of a subsequent CV event was applied to these patients. To account for the fact that the risk of a subsequent CV event is higher, we took the background risk of a previous non-fatal MI, stroke, or other CV/cerebrovascular event (observed in the Canadian general population) and applied a factor of 1.07, 1.08 and 1.13, respectively⁴¹.

Mortality

Two types of mortality rates were considered in this model. The average death rate of 0.005869/year based on the Statistics Canada mortality rate in 2003⁴¹ was used as the background mortality rate for the average Canadian population. The model then applied an increased death rate due to each of the CV events.

Utility values

Utility values were obtained from published literature (MEDLINE and EMBASE) and were assigned to outcomes to assess QALYs. Search terms included ‘cost-utility analysis’, ‘quality of life’, ‘arthritis’, ‘gastrointestinal disorders’ and ‘cardiovascular disorders’. Table 2 shows the utility values used in the model for arthritis (controlled and uncontrolled), GI events (GI discomfort, complicated and uncomplicated ulcer), and CV events (non-fatal MI, non-fatal stroke and non-fatal other CV events). Utilities were applied for 30 days in the month in which the event occurred. Individuals were assumed to return to their controlled arthritis state 1 month after a GI event, and 2 months after a non-fatal MI or non-fatal other CV event (with no residual utility deficit due to the CV event). After a stroke, individuals remained in their stroke state indefinitely; if they went on to have a GI event, they maintained the stroke utility but did not have further utility decrements related to the GI event.

Costs

The direct costs included in this analysis (medication, hospitalisations and adverse events), presented in 2005 Canadian dollars, are shown in Table 3. For non-selective NSAID inputs, drug acquisition costs specific to generic naproxen (Quebec's lowest cost NSAID) were used. Costs for MI, stroke and other CV events were included for the acute phase only; potential costs after 1 month following the event were not included. Cost of surgery, post-surgery and medical treatment in the ‘uncontrolled arthritis’ health state was assumed to be zero. Costs and benefits were discounted at 3% annually.

Sensitivity analysis

Multiple one-way sensitivity analyses were performed, varying individual parameters from the base case. For the different treatment strategies, the percentage of patients in the high-risk category, the incidence of GI and CV events, medication costs and the utility values for clinical events (except for loss of efficacy and controlled arthritis) were all varied by ±20% from base case.

Results

Tables 4A & B show the summary results of the base case analysis for the total OA/RA population, as well as for younger and older (≥65 years) patients. Based on the Quebec population data, 40% of patients are assumed to be aged ≥65 years; age is a major risk factor for developing an adverse GI or CV event while on NSAID therapy.

Clinical outcomes

Over 5 years, for the total population studied, first-line celecoxib was associated with fewer GI events per patient than either second- or third-line treatment (1.287, 1.641, 1.642, respectively) (Table 4A). Conversely, in the first-line model, celecoxib was associated with a greater number of CV events than either second- or third-line treatment (0.087, 0.083, 0.083, respectively). These CV results are a consequence of the model structure rather than individual drug properties. Based on the literature, the analysis used the same CV event rates for both celecoxib and non-selective NSAIDs. For the total population, the total QALYs (per person) associated with first-, second- and third-line celecoxib treatment were 3.251, 3.231, 3.230, respectively.

In the subgroup analysis, results similar to the base case were observed for both the younger (<65 years old) and older (≥65 years old) patient populations (Table 4B).

Economic outcomes

Over 5 years, for the total population studied, first-line celecoxib treatment was associated with higher total costs per patient than second-line or third-line celecoxib treatment (Can$4,790, $3,390, $3,466, respectively) (Table 4A). The costs for managing both fatal and non-fatal CV events were also higher for first-line ($1,051) celecoxib treatment compared with second-line and third-line treatment ($1,006 and $1,005, respectively); however, the costs of managing fatal and non-fatal GI events were higher for patients receiving third-line celecoxib treatment ($264) than either first- or second-line treatment ($141 and $255). Finally, the total cost of medication per patient was greater for first-line celecoxib treatment, than second- or third-line treatment ($3,598, $2,129, $2,197, respectively).

Again, in the subgroup analysis, results similar to the base case were observed for the younger and older populations (Table 4B).

Incremental analysis

Of the three treatment strategies, first-line celecoxib use was associated with the highest QALYs but also the greatest total cost per patient. The resulting cost-utility ratio was Can$54,696/QALY for the older or high-risk population (Table 4B). Effectiveness (QALYs) of second- and third-line use of celecoxib was similar. However, in all risk categories, the total cost of second-line treatment was lower than that of third-line treatment; thus, from a pharmacoeconomic perspective, second-line dominated third-line (same effectiveness, lower cost).

Sensitivity analysis

The results of multiple one-way sensitivity analyses to test the uncertainty surrounding model inputs showed second-line treatment dominating third-line treatment in almost every case (Table 5). The only scenarios where second-line treatment was not dominating third-line treatment were where: (1) celecoxib GI discomfort rate was increased by 20% and, (2) celecoxib drug cost was increased by 20%. The sensitivity analyses also showed that the incremental cost-utility ratios of celecoxib first-line versus celecoxib second-line treatment remained stable over most of the parameters tested in the model. Overall, the following parameters were the most sensitive to change: celecoxib GI discomfort rate, utilities and celecoxib drug cost. Attributing a better safety profile to non-selective NSAIDs did not modify the overall results.

Discussion

This analysis assessed the cost-effectiveness of celecoxib used in first-, second-, or third-line treatment for patients with OA/RA. Using a Markov decision-analytic model, which considers both GI toxicity and CV events, first-line treatment with celecoxib was associated with the highest QALYs but also, significantly, the greatest total cost. In those at low/average (younger population) or high risk (older population) of a GI or CV event, the results showed similar effectiveness of second- and third-line celecoxib treatment in terms of QALYs, but a lower total cost for second-line compared to third-line treatment. These results indicate that second-line treatment dominates third-line treatment.

As expected in this analysis, the younger subpopulation, which was assumed to be at low/average risk of an adverse event, had higher QALYs (3.249 for both second- and third-line treatment) and lower costs (Can$3,304 and $3,380, second- and third-line, respectively) than the older (high risk) subpopulation (total QALYs 3.203 and 3.202, second- and third-line, respectively; total costs Can$3,519 and $3,597, second- and third-line, respectively). This is explained by the fact that younger patients experience fewer GI and CV events over time and consequently accumulate more QALYs and incur less cost.

In the older subpopulation, the gain in QALYs (0.025 additional QALYs) associated with the use of celecoxib in first-line as opposed to second- or third-line treatment is greater than the same gain in the low/average GI/CV risk subpopulation (0.018 additional QALYs). This shows that using celecoxib earlier in therapy is more important in patients aged ≥65 years who are at high risk of developing a GI event. Moreover, the use of celecoxib as first-line rather than second-line therapy is more cost effective in older patients than in younger patients (Can$54,696/QALY vs. $80,077/QALY, respectively).

In terms of CV events, first-line celecoxib was also associated with more CV events than either second- or third-line treatment. In the study model, its was assumed that the CV event rate for non-selective NSAIDs was equivalent to that of celecoxib. In addition, celecoxib was assumed to have a lower GI event death rate, compared with non-selective NSAIDs. Therefore, the decrease in the number of individuals who die from a GI event while taking celecoxib as a first-line treatment strategy translates into an increased number of patients transitioning through the model to subsequent years and getting exposed to the standard rate of CV events; thus explaining why there is an increase in the number of CV-related events for patients on the first-line treatment strategy.

Direct comparisons with other studies are limited by the different model structures employed; however, the study results for the younger (low/average GI and CV risk) and older (high GI and CV risk) subpopulations are consistent with previous findings that patients at high risk of developing GI events are most likely to gain greater benefit from celecoxib therapy⁴⁹, and that celecoxib has greater cost-effectiveness in this group of patients^17,34,50. Analyses utilising the Celecoxib Outcomes Measurement Evaluation Tool (COMET) model have been applied to Canada and results indicate that celecoxib use in patients aged ≥65 years would not impose an excessive incremental impact on a Canadian provincial healthcare budget¹⁷. Another Canadian cost-utility model, conducted from the perspective of a third-party payer, found celecoxib to be cost-effective versus non-selective NSAIDs in OA/RA patients who had previously experienced upper GI events. Analyses by age indicated that celecoxib was cost-effective in those aged >76 years, although cost-effectiveness ratios were not provided³⁴.

The current model incorporates several features that reflect the real-life clinical setting. Unlike most previous analyses of the cost-effectiveness of celecoxib and non-selective NSAIDs, this model emulates clinical practice by including treatment switches following adverse events. This gives a much more realistic representation of treatment patterns and, rather than simply evaluating one treatment compared with another, enables comparison of the cost-effectiveness of treatment strategies involving three or more different medications. However, it is recognised that the three treatment strategies evaluated in this economic evaluation do not reflect all the possible pathways observed in clinical practice. For example, in the real-world setting, a clinician may choose to put a patient who has experienced a second uncomplicated ulcer on a combination of celecoxib plus a PPI; however, in the model it was assumed that these patients would discontinue anti-inflammatory treatment.

Furthermore, for the second- and third-line treatment strategies, the model assumes the older subpopulation would initiate treatment on a non-selective NSAID alone whereas current clinical practice guidelines recommend the addition of gastroprotection. However, a study performed in Quebec showed that the co-prescription of gastroprotection with non-selective NSAIDs in patients at moderate-to-high risk of GI events remained suboptimal with only 35% of non-selective NSAID prescriptions being co-prescribed with any gastroprotection⁵¹. In addition, it is worth noting that for the older subpopulation, the addition of a PPI would not affect the outcome of the incremental analysis for the second- and the third-line treatment strategies as the addition of a PPI would be applied to both treatment strategies. Nevertheless, from a modelling standpoint, it was impossible to consider all possible treatment pathways and it is felt that those selected were appropriate.

Another strength of this model is the consideration of CV events in addition to upper GI events. Reports have indicated an increased risk of CV events compared with placebo for both COX-2 selective and non-selective NSAIDs^13–16, and in the study model the same CV event rates were used for celecoxib and non-selective NSAIDs based on the results of a meta-analysis^32,33. However, as current data on the relative CV risks of celecoxib and non-selective NSAIDs are from trials that were not primarily designed or powered to assess CV event rates, there is still some uncertainty surrounding CV event input rates. Given this uncertainty, a sensitivity analyses was performed using lower CV rates for the non-selective NSAIDs (to potentially reflect those of naproxen). The outcomes of the sensitivity analyses demonstrate that the results are robust to variation in CV event rates.

As with all modelling exercises, this study has its limitations and has employed a number of assumptions. For example, GI event rates for celecoxib were based on a supratherapeutic dose (800 mg/day), corresponding to twice the recommended dose for RA and four times the dose for OA, and thus may underestimate the GI benefit associated with celecoxib use. The model also does not take into consideration the particular safety profile of patients taking low-dose aspirin for CV prophylaxis, and may, therefore, under- and overestimate the true number of GI events and CV events, respectively. It was also assumed that there was 100% adherence to PPI therapy, although several studies have demonstrated that PPI co-therapy is underutilised and associated with low compliance^52–56. The study model may therefore underestimate the real GI advantage of celecoxib compared with non-selective NSAIDs plus PPI. In addition, by assuming for the medication cost that naproxen, the lowest cost NSAID, would be representative of the whole class of non-selective NSAIDs, the third-line treatment strategy was favoured over the first- and second-line treatment strategies. It was also assumed that patients discontinued after failing the three treatment strategies have ‘uncontrolled arthritis’ and would not resume therapy in order to limit treatment choices to the NSAID class of drugs. Such patients, however, have other treatment options which are beyond the scope of this review. Finally, in the absence of precise drug exposure data, no consideration is given to episodic use of drug treatments. This constant-use assumption leads to increased adverse events and greater costs than would be expected in the real world, but as episodic use introduces no differential between non-selective NSAIDs and celecoxib the qualitative comparisons remain valid.

Conclusion

Utilisation studies, including one performed in Quebec using the RAMQ database (Régie de l'assurance maladie du Québec), indicate that celecoxib is commonly being used in OA and RA patients at high risk of GI complications^26,57, in line with evidence-based clinical practice guidelines³. As such, it appears that celecoxib is currently considered in the treatment algorithm for arthritic pain as a first-line agent in patients with high GI risk or as a second-line agent to be used in patients who have previously experienced a GI event while using a non-selective NSAID. The current study provides the Quebec provincial drug programme with evidence to support the cost-effectiveness of celecoxib in these treatment strategies.

Figure 1. Decision tree – COSMO model.

Table 1. Event rate inputs for GI and CV events for celecoxib and non-selective NSAIDs.

Event (source)	Event rate per patient year	Celecoxib event rate per month (%)*	Non-selective NSAID event rate per month (%)*
GI discomfort (dyspepsia only)(Silverstein et al, 2000^5)	C: 575/1,441N: 640/1,384	3.325	3.853
Uncomplicated ulcer(Silverstein et al, 2000^5)	C: 19/1,441N: 29/1,384	0.110	0.175
Complicated ulcer(Silverstein et al, 2000^5)	C: 11/1,441N: 20/1,384	0.064	0.120
Non-fatal MI^†(Simon et al, 2005^32 andWhite et al, 2007^33)	C: 35/5,651.2N: 19/4,386.4	0.036	0.036
Non-fatal stroke^†(Simon et al, 2005^32 andWhite et al, 2007^33)	C: 7/5,651.2N: 16/4,386.4	0.030	0.030
Non-fatal other CV events^†(Simon et al, 2005 ^32 andWhite et al, 2007^33)	C: 39/5,651.2N: 38/4,386.4	0.072	0.072
CV death^‡(Simon et al, 2005^32 andWhite et al, 2007^33)	C: 15/5,651.2N: 19/4,386.4	0.036	0.036

*Event rates per patient year were extracted from included studies and converted to event rate per month using the formula r_(t)= 1 – e^–r/t, where the monthly rate = 1 – e^{(–annual rate/12)}, based on the model structure.

^†The same CV monthly event rates were used for celecoxib and non-selective NSAIDs.

^‡CV death is a composite of MI plus stroke and other CV death.

N, non-selective NSAID; C, celecoxib.

Table 2. Utility values used in the model.

Variable	Utility value	Source
Controlled arthritisUncontrolled arthritis	0.7800.688	Mittmann et al, 1999^42Maetzel et al, 2003^34
GI events* GI discomfort Uncomplicated ulcer Complicated ulcer	0.7700.7500.440	Kristiansen et al, 2002^43Kristiansen et al, 2002^43Kristiansen et al, 2002^43
CV events Non-fatal MI^† Non-fatal stroke^‡ Non-fatal other CV events^†	0.5000.4500.500	Tengs et al, 2003^44

*Utility applied in the month the event occurred.

^†Utility applied for the 2 months (including the month the event occurred). These utility values were scale-adjusted from the utility value of non-fatal stroke.

^‡Utility applied in all months post-event.

Table 3. Summary of costs (Can$2005).

Parameter	Cost (Can$)*	Source
Medication		RAMQ^45
Celecoxib 100 mg bid/   200 mg bid	1.66^†
Naproxen 500 mg bid	0.42
Omeprazole 20 mg qd	1.38^‡
GI events		Maetzel et al, 2003^34
GI discomfort	35
Uncomplicated ulcer	514
Complicated ulcer   Hospitalisation/surgery   Hospitalisation/medication   Outpatient	8,5613,633548
CV events
Non-fatal MI	14,803	Maetzel et al, 2003^34
Non-fatal stroke	16,986	Lamy A et al, 2003^46
Non-fatal other event	4,543^§	Casciano R et al, 2004^47
Fatal events   MI   Stroke   Other CV events   GI event	14,80316,98614,8038,561	AssumptionAssumptionAssumptionAssumption

*Any non-2005 costs inflated using the consumer price index⁴⁸.

^†Based on an average daily dose of celecoxib of 230 mg/day.

^‡Average price of omeprazole in Quebec, which includes a 13% market share for Losec (AstraZeneca Canada Inc., Mississaugua, ON).

^§Average of cardiac arrest and congestive heart failure events.

bid, twice daily; qd, once daily.

Table 4A. Total population: summary results per person over 5 years.

	Celecoxib 1st line	Celecoxib 2nd line	Celecoxib 3rd line
Economic outcomes
Total costs (Can$2005)	4,790	3,390	3,466
Costs to manage GI events*	141	255	264
Costs to manage CV events*	1,051	1,006	1,005
Total medication costs^†	3,598	2,129	2,197
Clinical outcomes
Total QALYs	3.251	3.231	3.230
No. GI-related events*	1.287	1.641	1.642
No. CV-related events*	0.087	0.083	0.083
Incremental analysis
Cost per QALY gained (2nd vs. 3rd line)	2nd line dominates 3rd line
Cost per QALY gained (1st vs. 2nd line)	67,882

*Non-fatal and fatal events.

^†Includes cost of celecoxib, non-selective NSAIDs and PPIs.

Table 4B. Younger and older subpopulations: summary results per person over 5 years.

	Celecoxib 1st line	Celecoxib 2nd line	Celecoxib 3rd line
	Younger* (low/average GI risk) subpopulation
Economic outcomes
Total costs (Can$2005)	4,734	3,304	3,380
Costs to manage GI events^†	105	195	203
Costs to manage CV events^†	1,034	993	993
Total medication costs^‡	3,595	2,115	2,183
Clinical outcomes
Total QALYs	3.267	3.249	3.249
No. GI-related events^†	1.277	1.623	1.624
No. CV-related events^†	0.087	0.084	0.084
Incremental analysis
Cost per QALY gained (2nd vs. 3rd line)	2nd line dominates 3rd line
Cost per QALY gained (1st vs. 2nd line)	80,077
	Older^§ (high GI risk) subpopulation
Economic outcomes
Total costs (Can$2005)	4,874	3,519	3,597
Costs to manage GI-events^†	196	344	354
Costs to manage CV-events^†	1,076	1,025	1,024
Total medication costs^‡	3,602	2,150	2,219
Clinical outcomes
Total QALYs	3.227	3.203	3.202
No. GI-related events^†	1.303	1.668	1.670
No. CV-related events^†	0.086	0.082	0.082
Incremental analysis
Cost per QALY gained (2nd vs. 3rd line)	2nd line dominates 3rd line
Cost per QALY gained (1st vs. 2nd line)	54,696

*Arthritis patients aged <65 years.

^†Non-fatal and fatal events.

^§Arthritis patients aged ≥65 years.

^‡Includes cost of celecoxib, non-selective NSAIDs and PPIs.

Table 5. Results of sensitivity analyses: incremental cost/QALY gained (Can$ 2005).

Cost/QALY gained	Values		General population		Low-risk population		High-risk population
	Base case	Sensitivity analysis	2nd vs. 3rd line	1st vs. 2nd line	2nd vs. 3rd line	1st vs. 2nd line	2nd vs. 3rd line	1st vs. 2nd line
Base case			2nd dom*	67,882	2nd dom	80,077	2nd dom	54,696
Non-fatal MI
nsNSAID	0.036%	0.029%0.032%	2nd dom2nd dom	74,14870,895	2nd dom2nd dom	87,69583,735	2nd dom2nd dom	59,56657,040
Non-fatal stroke
nsNSAID	0.030%	0.024%0.027%	2nd dom2nd dom	70,57469,188	2nd dom2nd dom	83,04381,518	2nd dom2nd dom	57,01155,817
Non-fatal other CV events
nsNSAID	0.072%	0.058%0.065%	2nd dom2nd dom	89,68177,306	2nd dom2nd dom	103,88890,475	2nd dom2nd dom	73,73262,833
Death: CV
nsNSAID	0.036%	0.029%0.032%	2nd dom2nd dom	118,62786,419	2nd dom2nd dom	162,662107,430	2nd dom2nd dom	83,01865,981
GI discomfort
nsNSAID	3.853%	3.082%4.624%	2nd dom2nd dom	80,08559,284	2nd dom2nd dom	100,80666,729	2nd dom2nd dom	60,10750,525
Celecoxib	3.325%	2.660%3.990%	2nd dom32,949	70,47067,223	2nd dom30,504	82,61780,157	2nd dom37,314	57,31753,439
Uncomplicated ulcer
nsNSAID	0.175%	0.140%0.210%	2nd dom2nd dom	69,27466,550	2nd dom2nd dom	81,44678,751	2nd dom2nd dom	56,08853,376
Celecoxib	0.110%	0.089%0.132%	2nd dom2nd dom	68,19167,582	2nd dom2nd dom	80,49379,673	2nd dom2nd dom	54,83554,559
Complicated ulcer
nsNSAID	0.120%	0.096%0.144%	2nd dom2nd dom	79,72058,938	2nd dom2nd dom	89,64672,244	2nd dom2nd dom	67,99445,463
Celecoxib	0.064%	0.051%0.076%	2nd dom2nd dom	64,79071,211	2nd dom2nd dom	78,39081,827	2nd dom2nd dom	50,72059,175
nsNSAID + PPI	0.080%	0.064%0.095%	2nd dom2nd dom	67,88267,882	2nd dom2nd dom	80,07780,077	2nd dom2nd dom	54,69654,696
Reduction in probability of GI event due to adding a PPI
GI discomfort
Celecoxib	50%	40%60%	2nd dom2nd dom	69,36167,913	2nd dom2nd dom	82,18480,157	2nd dom2nd dom	55,61954,682
Uncomplicated ulcer
Celecoxib	100%	80%100%	2nd dom2nd dom	68,18267,882	2nd dom2nd dom	80,34680,077	2nd dom2nd dom	54,98554,696
Complicated ulcer
Celecoxib	100%	80%100%	2nd dom2nd dom	69,81267,882	2nd dom2nd dom	82,43280,077	2nd dom2nd dom	56,19254,696
Patients with higher risk of GI	40%	32%48%	2nd dom2nd dom	70,05265,826	2nd dom2nd dom	80,07780,077	2nd dom2nd dom	54,69654,696
Utility values	(All utility values except loss of efficacy and controlled arthritis)	–20% base case+20% base case	2nd dom2nd dom	30,9612nd dom	2nd dom2nd dom	33,9432nd dom	2nd dom2nd dom	25,7822nd dom
Drug costs
nsNSAID	$0.42	$0.336$0.504	2nd dom2nd dom	70,07965,686	2nd dom2nd dom	82,71077,443	2nd dom2nd dom	56,42152,972
Celecoxib	$1.66	$1.328$1.992	2nd dom106,822	57,79677,969	2nd dom118,718	68,12392,030	2nd dom93,228	46,62862,764
PPI	$1.38	$1.104$1.656	2nd dom2nd dom	61,39174,374	2nd dom2nd dom	72,66087,493	2nd dom2nd dom	49,20560,188

*Same effectiveness but is at a lower cost. dom, dominates.

Acknowledgements

Declaration of interest: This study was sponsored by Pfizer Inc. M. Lu of Dymaxium Inc. provided technical assistance in running the analyses in the model. Editorial support was provided by Leigh Prevost, BSc, of Parexel and was funded by Pfizer Inc. L.B. has disclosed that he has received an honorarium from Pfizer Canada Inc. for the validation of the model structure. N.R., N.M. and J.H. have disclosed that they were paid consultants to Pfizer in connection with the development of this manuscript. At the time of the analysis, N.R. and N.M. were employees of the HOPE Research Centre, Toronto, Canada; J.H. was a full-time employee of Dymaxium Inc. and J.P.R. and G.Z. were full-time employees of Pfizer. All authors were involved in the writing, review and approval of this paper.

Data in this paper were previously presented as a poster at the 29th Annual Meeting of the Society for Medical Decision Making, October 20–24, 2007, Pittsburgh, PA, USA.

Notes