Role of initial NSAID choice and patient risk factors in the prevention of NSAID gastropathy: A decision analysis

Abstract

Objective

To investigate the role of initial nonsteroidal antiinflammatory drug (NSAID) choice in the prevention of NSAID gastropathy, based on relative clinical and economic effects.

Methods

To mimic clinical practice, a symptom-driven decision analytic model was constructed to compare 2 treatment strategies for long-term users of NSAIDs over a 1-year period: Strategy 1—generic NSAID used initially, and safer, more expensive NSAID reserved for treatment failures due to symptomatic gastropathy; and Strategy 2—safer, more expensive NSAID used in all instances. The only distinction between the strategies was the choice of initial NSAID. NSAIDs differed in gastrointestinal safety profiles and acquisition costs. The use and impact of antisecretory medications were included in the model. Because published data on patients' ulcer risk and relative NSAID safety show considerable variability, sensitivity analyses were used to evaluate the key clinical outcomes and costs.

Result

For patients without risk factors for NSAID ulcers (average risk), the model estimated that the strategy restricting use of the safer NSAID resulted in more symptomatic ulcers (Strategy 1, 2.58; Strategy 2, 0.73) and ulcer-related complications (Strategy 1, 1.18; Strategy 2, 0.23) per 100 patient years. The restricted strategy led to a significantly lower cost per patient treated (Strategy 1, $239; Strategy 2, $831 per year). In the principal analysis, the incremental costs to prevent symptomatic and complicated ulcers were $31,900 and $56,700, respectively. The estimated incremental cost per ulcer avoided was sensitive to the relative protection provided by the safer NSAID and fell dramatically as the patients' ulcer risk was increased above average risk.

Conclusion

Unrestricted use of NSAIDs that reduce the risk of symptomatic ulcers has the potential to produce important clinical benefits at incremental cost. The impressive impact of ulcer risk on the incremental cost per ulcer prevented warrants increased attention to risk factor identification when NSAIDs are prescribed.

INTRODUCTION

Nonsteroidal antiinflammatory drugs (NSAIDs) are associated with adverse gastrointestinal (GI) effects ranging from mild dyspepsia to serious complications such as bleeding peptic ulcer. Although the probability is low that any long-term NSAID user will experience a drug-related complication (1), the fact that 2 million Americans regularly use these agents make NSAID gastropathy an important problem from both clinical and economic perspectives. It is estimated that more than 100,000 hospitalizations and 10,000 to 20,000 deaths each year in the United States can be attributed to NSAID-related complications (2). Although the cost of excess hospitalizations has been estimated at $4 billion annually (3), these expenditures greatly underestimate the societal impact of NSAID gastropathy, in that indirect costs such as lost productivity tend not to be included.

The scope of this problem has led the Food and Drug Administration to include a formal warning in the package labeling regarding the risk of adverse GI events for patients using NSAIDs (4). Despite attempts to educate patients, most regular NSAID users have a lack of awareness of potential side effects of NSAIDs (5). Controversy remains among clinicians on how best to weigh the potential clinical benefits of NSAIDs against the possibility of adverse events associated with their use. Identification of risk factors for the development of NSAID-related complications may aid clinicians in identifying patients at highest risk (1, 6).

Although there is no consensus on how best to minimize NSAID-related adverse events, it is clear that assessments of available treatment options must take into account clinical effects and economic consequences. Although recently published prospective trials have evaluated the impact of celecoxib and rofecoxib in narrowly defined cohorts (7, 8), these reports did not include formal economic analyses. Moreover, these investigations may not truly represent clinical practice because of the patient populations evaluated, the exclusion of antisecretory agents, and differences in aspirin use.

Limitations of published cost-effectiveness analyses evaluating strategies to prevent NSAID-related adverse events include a reliance on endoscopically diagnosed lesions as clinical end points (9–11). Symptoms, not the presence of an endoscopic lesion, drive patient satisfaction and health care expenditures. Failure to fully account for patient symptoms, which commonly occur in the absence of an endoscopic lesion, may lead to a significant underestimation of the economic burden of NSAID gastropathy.

Misoprostol is the only FDA-approved drug indicated to prevent NSAID-related adverse events. Published economic analyses suggest that this agent is cost-effective for patients at increased risk for NSAID gastropathy (12). However, these analyses did not include non-ulcer symptoms in their assessment of treatment costs. Misoprostol has not been demonstrated to reduce NSAID-related dyspepsia and may be associated with its own adverse effects (13). Thus, acid inhibitory drugs are widely used to reduce NSAID-associated adverse effects, and misoprostol represents 5% of the anti-ulcer medications prescribed to long-term NSAID users (14, 15). Although H₂ receptor antagonists may reduce NSAID-associated dyspepsia (16), these agents are not effective in preventing NSAID-associated ulcers and their related complications at traditional doses (17). Potent acid suppression with high-dose H₂ antagonists (18) or proton pump inhibitors (PPIs) (19–21) have been demonstrated to heal and prevent the recurrence of endoscopic ulcers in randomized controlled trials. In order to determine the actual impact of NSAID gastropathy in clinical practice, it is imperative that these widely used agents be included in cost-effectiveness calculations.

An alternative to use of supplementary pharmaceuticals to reduce NSAID toxicity is the use of an equally effective antiinflammatory agent with reduced propensity for GI injury. The differences in the relative safety of currently available NSAIDs may be explained by their pharmacologic properties and impact on the 2 cyclooxygenase isoenzymes, COX-1 and COX-2 (22, 23). In order to constrain health care expenditures, drug formularies often recommend using less expensive but less safe NSAIDs (e.g., generically available compounds), while restricting use of equally effective yet safer regimens for treatment failures. Determining which NSAID users have access to these safer, more expensive agents should depend on their clinical and economic effects.

Because long-term NSAID users may fail initial therapy, experience dyspepsia, or suffer a complication necessitating a change in therapy, the clinical and cost consequences of NSAID therapy depend on subsequent diagnostic and treatment decisions that occur over the entire natural history of disease. Therefore, analyses that directly compare the acquisition cost of specific NSAIDs may not reflect actual clinical practice, because available agents are often used in sequence (generic agents first, then safer agents). Thus, the most cost-effective NSAID regimen depends not just on the differences in complication rates and/or treatment costs at time of use, but also on the likelihood of switching medications, the variation in patients' adverse GI symptoms, and resultant ulcer and non–ulcer-related health care expenditures.

Accordingly, to evaluate the role of relative NSAID safety, patients' ulcer risk, and the impact of patient symptoms on the cost-effectiveness of long-term NSAID therapy, we developed a symptom-driven decision analytic model to evaluate 2 NSAID treatment strategies that differed only in the choice of the initial NSAID. Our intent was to develop a framework that explicitly delineated community practice patterns (in contrast to a clinical trial) by allowing changes in NSAIDs prescribed, use of antisecretory therapy, and performance of diagnostic endoscopy at the discretion of the physician.

METHODS

Management strategies evaluated. Using the best available published data, we used decision analysis to estimate the clinical and economic consequences of 2 management strategies in a cohort of long-term NSAID users without risk factors for gastropathy:

• Strategy 1–generic NSAID used initially; safer, more expensive NSAID reserved for patients experiencing a GI adverse event or for patients intolerant of the generic NSAID.

• Strategy 2–safer, more expensive NSAID used in all patients. The only difference between the 2 strategies was the prescription of the initial NSAID. The sequence of events after initial NSAID therapy is demonstrated in Table 1.

Table 1. Sequence of diagnostic testing and therapeutic changes by strategy

NSAID = nonsteroidal antiinflammatory drug.

Strategy 1

Initial visit–all patients receive generic NSAID

1st symptomatic episode–generic H2 antagonist added to generic NSAID, no diagnostic testing

2nd symptomatic episode–endoscopy

Ulcer present–rapid urease test for Helicobacter pylori, H. pylori eradication therapy when appropriate; proton pump inhibitor therapy added; switch to safer NSAID

Ulcer absent–H2 antagonist therapy; switch to safer NSAID

3rd symptomatic episode–maintenance antisecretory therapy based on previous endoscopy; continue safer NSAID

Strategy 2

Initial visit–all patients receive safer NSAID

1st symptomatic episode–generic H2 antagonist added to safer NSAID

2nd symptomatic episode–endoscopy

Ulcer present–rapid urease test for H. pylori, H. pylori eradication therapy when appropriate; proton pump inhibitor therapy added; switch to safer NSAID

Ulcer absent–H2 antagonist therapy added to safer NSAID

3rd symptomatic episode–maintenance antisecretory therapy based on previous endoscopy; continue safer NSAID

Decision analytic model. A Markov model (Figure 1) was constructed to simulate the natural history and calculate health care expenditures of a cohort of long-term NSAID users without risk factors for NSAID gastropathy over a 1-year period (24). The probability of developing symptoms was modeled for patients with and without NSAID-associated ulcers. Subsequent to the initial NSAID prescription, the effects of all diagnostic and therapeutic interventions on clinical outcomes and resource use were measured. Because symptoms are imperfectly related to the presence of an endoscopic lesion, progression through the model was symptom-driven. When clinically indicated, diagnostic tests were incorporated into the simulation; results were used as a basis for all subsequent management decisions.

Figure 1. Markov model. Patients' symptom and ulcer status were independently modeled. However, diagnostic and therapeutic interventions occurred only when accompanied by patient symptoms.

The requirement that a patient must have symptoms or an adverse event before visiting a physician has important ramifications. First, patients with an NSAID-related endoscopic lesion may be asymptomatic (Figure 1, top right). While asymptomatic, neither the patient nor the physician was aware of the lesion, and therefore no additional services were ordered, and no costs other than those for the medications currently prescribed were incurred. Individuals with ulcers or erosions and asymptomatic patients without a silent lesion were conveyed different probabilities of future adverse events in the model. Patients with an asymptomatic lesion may either heal the lesion, develop a symptomatic adverse event, or progress directly to a complication such as hemorrhage or perforation. This modeling distinction allows individuals to present with an ulcer complication as their initial indication of NSAID gastropathy, a situation not uncommon in clinical practice (Figure 1, bottom) (25).

Second, patients without an objective lesion may have symptoms. NSAIDs cause dyspepsia in the absence of objective lesions (Figure 1, lower left) (16). Clinical practice does not mandate that all patients undergo invasive diagnostic testing at predetermined designated intervals to determine ulcer status, as in clinical trials. In the model, diagnostic testing and treatment decisions were based on the number of symptomatic episodes, prior medication use, and available diagnostic test results.

Third, unless a diagnostic test is performed, symptomatic patients with objective lesions cannot be distinguished from those without. Clinical history and physical examination are poor predictors of the presence or absence of ulcers or erosions in symptomatic NSAID users (26). Thus, when empiric antisecretory treatment is prescribed based on symptoms in the absence of a diagnostic test, patients with and without lesions received the same therapy.

Clinical inputs. The published literature on the risk of endoscopic ulcers and clinically significant GI adverse events in NSAID users demonstrates considerable variability. Recognizing this variation regarding the critical clinical inputs, the principal analysis—the result of a critical synthesis of available data—should be viewed only as a point estimate of relative cost-effectiveness between the strategies. The sensitivity analyses, which allow an examination of how changes in relative safety between NSAIDs compared and how patients' ulcer risk affects the incremental cost-effectiveness of the safer NSAID, can be equally viewed as the primary study results (7, 8, 27). This framework allows analysis of recent clinical trial information to be analyzed in the context of current clinical practice (7, 8).

NSAID-associated symptomatic ulcer rates. Many clinical investigations evaluating NSAID safety used detection of endoscopic lesions, many of which were not clinically significant, as the determining factor for the incidence, healing, and prevention of NSAID-associated adverse events (17–21, 27–35). These trials varied considerably in definition of an ulcer (depth 3–5 mm), the underlying ulcer risk of the study population, and the NSAID under investigation (ulcerogenicity varies among available NSAIDs). In contrast to these endoscopy-based studies, one randomized trial, Misoprostol Ulcer Complications Outcomes Safety Assessment (MUCOSA) used clinically significant events as the primary outcome (36). In a cohort of rheumatoid arthritis patients that included individuals at low and high risk for NSAID-related complications, 1.46% of the patients developed a symptomatic adverse event over a 6-month period.

From these clinical trial data, calculations were made to estimate a relationship of endoscopic lesions to clinically important events. Asymptomatic lesions could either heal, remain asymptomatic, or progress to a symptomatic (and potentially complicated) event. In the case of an uncomplicated symptomatic ulcer, the presence of an ulcer was confirmed only if endoscopy was performed. For those patients in whom the diagnosis of ulcer was objectively ascertained, future management reflected this finding (e.g., use of PPI therapy, Helicobacter pylori testing). If the sequence of interventions did not specify that endoscopy be performed for a specific symptomatic episode (Table 1), empiric treatment was based on the number of symptomatic episodes, prior medication use, and past diagnostic test results.

We assumed that 50% of symptomatic ulcers attributable to the generic NSAID were complicated (e.g., accompanied by hemorrhage), necessitating urgent medical care. This estimate is consistent with recent prospective data demonstrating that 43% of symptomatic ulcers were complicated (7). Because of the reduced antiplatelet effect of the safer NSAID, it was assumed that 33% of symptomatic ulcers were complicated (23). Given the uncertainty for this variable, sensitivity analysis was used to further explore its impact over a broad range. For the principal analysis, we assumed that half of these complicated cases required an emergency medical intervention or surgical procedure. Patients who experienced a complicated ulcer were modeled separately for the remainder of the study year, because the probability of a recurrent complication is markedly increased in this patient subgroup (1).

There was assumed to be no difference in the probability that an endoscopic ulcer would be symptomatic if an ulcer did develop. This translated to an annualized perforation, ulcer, and bleed (PUB) rate of 2.7%/year for the generic NSAID. This input represents a reduction from that seen in the MUCOSA trial (1.46% over 6 months), reflecting the inclusion of high-risk patients in that study population. This rate was also slightly less than that seen in the celecoxib (3.0%) and rofecoxib (4.5%) outcome trials (7, 8). Given this variation in the literature, reflecting both NSAID safety and patients' ulcer risk, these critical inputs were assessed separately and in combination using sensitivity analyses.

Relative safety of NSAIDs. In the principal analysis, we assumed that the safer NSAID was associated with a 3.7-fold reduction in the monthly endoscopic ulcer rate when compared with a generic NSAID. This assumption is based on the published literature comparing nabumetone, rofecoxib, and celecoxib with commonly used NSAIDs, which demonstrated a 3- to 7-fold reduction in endoscopic ulcer rate (37–39). Recent data from the rofecoxib outcome trial has supported this level of reduction in symptomatic ulcers dependent upon patient risk factors. Although the overall risk reduction was between 2- and 3-fold, the reduction in GI events among low-risk patients was 88% (40).

Non-ulcer symptom rates. Symptom rates as a result of non-ulcer causes severe enough to lead to a management change (e.g., change NSAID or add antisecretory therapy) were derived from the placebo arms of randomized, controlled trials evaluating the effects of antisecretory agents on NSAID ulcer healing and prevention (17–21, 27–35). In the model, we assumed that the rate of symptomatic episodes unrelated to ulcer disease were equivalent for the 2 NSAIDs; this assumption was supported by subgroup analyses from the recent coxib outcome trials (7, 8). When the sequence of interventions (Table 1) did not prescribe a diagnostic test to establish the diagnosis of ulcer disease, patients with symptoms from ulcer and non-ulcer causes were treated similarly. After endoscopy was performed, treatment decisions reflected the knowledge that an ulcer was or was not present.

Effectiveness of antisecretory therapy on ulcer healing and symptoms (Table2). Estimates of efficacy of antisecretory medications on ulcer healing, prevention of ulcer recurrence, and symptom reduction were based on trials evaluating effects of H₂ receptor antagonists and proton pump receptor inhibitors on ulcer healing and symptom relief (17–21, 27, 29, 32–35). Antisecretory agents were used in similar fashion in each strategy based on the number of symptomatic events and endoscopy results (if performed). PPIs were prescribed if an ulcer was identified during endoscopy; for patients with symptoms but without endoscopic evidence of ulceration, H₂ receptor antagonists were prescribed. The model allowed the discontinuation of antisecretory agents if the patient became asymptomatic following a course of therapy.

Table 2. Clinical and cost inputs

NSAID = nonsteroidal antiinflammatory drug.
Effectiveness of antisecretory therapy on ulcer healing and symptoms
Ulcer healing rate, H2 antagonist	0.52
Ulcer healing rate, proton pump inhibitor	0.66
Ulcer prevention rate, H2 antagonist	0.20
Ulcer prevention rate, proton pump inhibitor	0.72
Symptom prevention rate, H2 antagonist	0.43
Symptom prevention rate, proton pump inhibitor	0.52
Cost inputs
Generic NSAID (month)	$5
Safer NSAID (month)	$60
Generic H2 antagonist (month)	$7
Proton pump inhibitor (month)	$100
Endoscopy	$1,000
Biopsy and rapid urease test for Helicobacter pylori	$100
Complicated ulcer requiring surgery	$10,000
Primary care physician visit	$40

Cost estimates. The payer perspective was used for the economic analysis, and therefore the model calculated only direct medical costs. Costs used were actual payments made for medical services by a large private insurer in the eastern US responsible for about 3 million persons. Estimates for physician visits, procedures, and hospital-related expenses for ulcer complications are found in Table 2. Given that payments vary by geography and payer, we tested high- and low-cost estimates by sensitivity analysis. Ranges used in the sensitivity analysis were derived from payments made by 3 different third-party payers, including Medicare. Costs of outpatient pharmaceuticals were based on average prices determined from an ongoing survey of retail pharmacies. Average wholesale prices plus a dispensing fee were used to establish the lower ranges of pharmaceutical costs. Indirect costs (e.g., lost productivity) and nonmedical direct costs (e.g., travel, lodging) were not included.

Outcomes measured. The computer simulation recorded ulcer and symptom status of the patients as they moved through the model on a monthly basis over the study year. Except for the first visit to receive the initial NSAID prescription, patients incurred nondrug costs only if symptoms occurred that were severe enough to warrant a physician visit. This “symptom-driven” management approach was adopted to mirror actual medical care delivery and allow a realistic estimate of resource utilization in the 1-year episode of care. As a result, asymptomatic ulcers, although tracked by the simulation, incurred no health care costs. However, resources devoted to symptomatic episodes for reasons other than an ulcer were included. Incremental cost-effectiveness ratios (e.g., cost per complicated ulcer prevented) were calculated when one strategy led to better clinical outcomes at higher total costs.

RESULTS

Clinical outcomes. In the principal analysis, the model estimated that for patients without risk factors for NSAID-associated adverse events, unrestricted access to the safer NSAID (Strategy 2) led to a reduction in ulcers and associated complications. When compared with Strategy 2, the strategy that restricted use of the safer NSAID to patients who failed generic therapy (Strategy 1) resulted in more symptomatic ulcers (Strategy 1, 2.58; Strategy 2, 0.73) and ulcer-related complications (Strategy 1, 1.18; Strategy 2, 0.23) per 100 patient years. These rates are lower than those reported in the recently completed outcome trials for celecoxib and rofecoxib (7, 8), reflecting the impact of protective agents incorporated into the model. Because the non-ulcer symptom rate was assumed to be equivalent for both NSAIDs in the principal analysis, no difference in the number of months with non-ulcer symptoms (Strategy 1, 43.3; Strategy 2, 44.9 per 100 patient years) was observed between strategies.

Economic outcomes.Cost per patient treated (Figure 2).

Figure 2. Patient yearly expenditure by strategy, broken down by individual components. NSAID = nonsteroidal antiinflammatory drug.

The cost per patient treated over the 1-year period for the restricted strategy (Strategy 1) was significantly lower than that of the strategy that permitted unrestricted use of the safer NSAID (Strategy 1, $239; Strategy 2, $831). This difference was driven for the most part by the substantial variation in acquisition cost of the 2 NSAIDs. When the cost per patient treated was divided into components (Figure 2), the substantial differential in NSAID cost (Strategy 1, $78; Strategy 2, $720) was slightly offset by a 6-fold reduction in complication costs (Strategy 1, $60; Strategy 2, $11) attributable to more prevalent use of the safer agent.

Incremental cost-effectiveness. In the principal analysis, Strategy 2 led to improved clinical outcomes at incremental expense when compared with the restricted strategy for patients at average risk of NSAID-related adverse events. For each clinical end point, an incremental cost-effectiveness ratio was calculated to quantify the extra expenditures necessary to prevent a single adverse event. The strategy that permitted unrestricted use of the safer NSAID prevented an additional symptomatic ulcer at an incremental cost of $31,900. Similarly, the incremental cost per complicated ulcer avoided was $56,700.

Sensitivity analysis. Given the variation and/or uncertainty regarding certain estimates used in the principal analysis, sensitivity analyses indicated that all outcomes measured were relatively insensitive to assumptions about effectiveness and costs of antisecretory therapy, the likelihood that an ulcer was complicated, and the cost of non-drug services (e.g., endoscopy, hospitalization for ulcer complications). However, the results estimated by the computer model were sensitive to the relative level of GI protection provided by the safer NSAID and the ulcer risk of the patient population.

Protective effect of safer NSAID (Figure3). Figure 3 demonstrates the impact of varying the protective effect of the safer NSAID on cost per patient treated and incremental cost-effectiveness for patients at average ulcer risk. As the relative protection provided by the safer agent was varied from 1 (no difference in safety) to a 10-fold relative reduction in the ulcer rate, the cost per patient treated for Strategy 2 remained substantially higher than that for Strategy 1 (Figure 3A). Although the clinical advantages of unrestricted use of the safer NSAID were enhanced as its relative safety was increased, the cost per patient treated remained substantially different. Figure 3B illustrates how the incremental cost to prevent an ulcer varied as the relative safety between NSAIDs was modified. The figure allows cost-effectiveness estimates for a wide range of relative safety, including those data from recent clinical trials (relative safety profile of approximately 2) (7, 8). Using this relative safety difference, the model estimated that the cost per symptomatic and complicated ulcer prevented by initial use of the safer NSAID is $48,000 and $72,000, respectively. As the relative safety between the NSAIDs exceeds 3, the incremental cost-effectiveness tends to stabilize.

Figure 3. Sensitivity analysis of protective effect of safer nonsteroidal antiinflammatory drug (NSAID). A, Protective effect of safer NSAID on cost per patient treated, by strategy. B, Protective effect of safer NSAID on incremental cost per ulcer prevented.

Patient ulcer risk (Figure4S). The difference in the cost per patient treated between the 2 strategies decreased slightly as the ulcer risk increased (Figure 4A). In order to capture the clinical benefits afforded to unrestricted use of the safer NSAID (Strategy 2), a sensitivity analysis estimating the incremental cost per symptomatic and complicated ulcer avoided as a function of ulcer risk was performed. As Figure 4B demonstrates, the incremental expenditure to prevent an adverse event decreased markedly as ulcer risk was increased from baseline.

Figure 4. Sensitivity analysis of ulcer risk. A, Effect of ulcer risk on cost per patient treated by strategy. B, Effect of ulcer risk on incremental cost per ulcer prevented.

DISCUSSION

Each year, more than 100 million prescriptions for NSAIDs are filled in the United States. Gastrointestinal toxicity associated with these commonly used agents is responsible for significant morbidity and mortality and constitutes a multibillion dollar financial drain on our health care system. The availability of safer but more expensive pharmaceutical agents that may reduce the likelihood of untoward events has raised the question of whether all long-term NSAID users should have access to them. The lack of clear evidence-based justification for their use, coupled with an effort to constrain pharmaceutical expenditures, has contributed to this quandary regarding the relative toxicity, or “COXicity,” of available NSAIDs.

To accurately assess the clinical and economic tradeoffs between a lower rate of drug-related complications and resultant higher overall expenditures, both the incremental costs and benefits should be carefully measured and compared with available alternatives. On the cost side, it is critical to look beyond direct cost comparisons of drugs under investigation. All the health care resources incurred over the entire episode of care must be accounted for, especially because a proportion of individuals prescribed one agent may eventually end up being prescribed the other. The clinical indications for, and side effects of, long-term antiinflammatory drug therapy often necessitate changing NSAIDs or adding co-therapy for prophylaxis or symptom control. Moreover, a formulary restriction that requires a patient to fail a less effective yet less expensive NSAID before a better, more expensive agent may be used is potentially shortsighted. The savings in drug costs diminish (and may be eliminated) as adverse events occur and/or a greater proportion of the patient population eventually switches to the better agent.

As far as health outcomes are concerned, symptoms severe enough to necessitate a visit to a health professional—whether caused by an ulcer or not—drive health care expenditures and must be accounted for. Thus, to mimic clinical practice, we constructed a symptom-driven simulation to capture clinical outcomes and health care costs associated with chronic NSAID use. Specifically, the cost-effectiveness of a practice to restrict use of a potentially safer, more expensive NSAID was compared with a strategy that allowed its unrestricted use. The decision analytic framework allowed the evaluation of discrete practice patterns within each strategy and explicitly quantified the tradeoffs between adverse events avoided and incremental costs. What emerged from our analysis was the finding that decisions regarding access to safer, more expensive NSAIDs depend on the cost differential between agents, relative safety among available agents, and patients' ulcer risk. Another interesting finding was that measurement of symptomatic events attributable to non-ulcer causes could contribute significantly to the overall costs of NSAID-related care.

The model estimated that for chronic NSAID users at average risk, unrestricted use of safer NSAIDs has the potential to decrease ulcer-related adverse events at an incremental cost that approximates published values for misoprostol (12). Sensitivity analysis revealed that, at current differences in acquisition price, under no circumstances would unrestricted use of the safer agent generate cost savings in this average-risk population. Thus, decision-makers must decide whether unrestricted use of a safer agent is worthwhile, given other demands for scarce health care resources. The simulation estimated that the incremental cost to prevent an NSAID-related ulcer fell dramatically as the patients' risk of NSAID-related adverse events increased and as the relative safety between the NSAIDs was enhanced. In the population of patients at above-average ulcer risk (e.g., those with risk factors such as prior GI hemorrhage, concomitant corticosteroid or anticoagulant therapy), the clinical and economic argument to recommend routine use of a safer NSAID stands on merit (41). Given the variation in published data regarding relative NSAID safety, sensitivity analyses allow assessment of this variable over a wide range, including those relative safety estimates from recently published outcome trials examining the COX-2 inhibitors (7, 8).

Independent of ulcer risk, our estimates may have undervalued unrestricted use of the safer NSAID, because we accounted for only direct medical costs to the payer. The results generated by the model included neither the quality-of-life improvements that can be attributed to decreased morbidity and mortality nor any indirect cost advantages that may accrue secondary to fewer adverse events (e.g., lost productivity). If our analysis were expanded to include these additional clinical and economic benefits associated with preventable complications, a recommendation that safer, more expensive NSAIDs be used without restriction might be extended to long-term NSAID users with moderate or possibly mild risk for GI toxicity.

Prospective controlled trials have demonstrated that use of safer NSAIDs leads to a reduction in clinically important GI adverse events in selected patient cohorts (7, 8). However, the incremental costs to achieve these benefits are closely linked to patients' risk of developing an adverse event. This important association between the relative cost-effectiveness of a safer NSAID and ulcer risk obligates the development and validation of user-friendly instrument(s) that accurately identify NSAID users at increased risk for adverse events.