Nonsteroidal anti-inflammatory drugs (NSAIDs) are

Transcription

1 GASTROENTEROLOGY 2003;125: Gastrointestinal Health Care Resource Utilization With Chronic Use of COX-2 Specific Inhibitors Versus Traditional NSAIDs LOREN LAINE,* JENIFER WOGEN, and HOLLY YU *University of Southern California School of Medicine, Los Angeles, California; and The Institute for Effectiveness Research, LLC, Bridgewater, New Jersey Background & Aims: Cyclooxygenase 2 (COX-2) specific inhibitors (coxibs) decrease gastrointestinal (GI) events in controlled trials, but results in clinical practice are unknown. We assessed GI-related resource use and costs in patients switching from chronic nonsteroidal anti-inflammatory drug (NSAID) therapy to chronic coxib therapy and in patients starting chronic NSAID therapy vs. chronic coxib therapy in a U.S. administrative claims database of >8 million lives. Methods: Switchers (n 2246) were assessed in the 12-month periods before and after switching from chronic NSAID therapy to coxib therapy. New NSAID (n 25,989) and new coxib (n 2125) groups were assessed for the 12-month periods before and after the initial prescription. Proportions of patients with GI resource use (odds ratio [OR] adjusted for relevant covariates) and costs were compared. Results: The adjusted OR for any GI resource use (coxib vs. NSAID period) among switchers was 0.86 ( ). The decrease was due to less GI cotherapy (OR 0.82 [ ]). Costs were not significantly lower after switching to coxibs (mean difference, $19; 95% CI: $139, $55), although after adding NSAID/coxib costs, the total cost in the coxib period was significantly higher (mean increase, $377; $271, $488). Adjusted OR for GI resource use for new-coxib vs. new-nsaid was 1.04 ( ), but GI costs were significantly lower in new-nsaid patients. Conclusions: Patients switching from chronic NSAID therapy to chronic coxib therapy had a slight decrease in the proportion using GI-related resources but not in GI costs. When NSAID/coxib drug costs were included, costs were significantly less with NSAIDs than with coxibs. The potential GI-related cost savings suggested in coxib clinical trials may not be fully realized in real-world settings. Nonsteroidal anti-inflammatory drugs (NSAIDs) are among the most widely used drugs in the world. Over $6.5 billion is spent annually on NSAIDs in the United States, and over 111 million prescriptions are filled. 1 Use of traditional nonselective NSAIDs significantly increases the risk of ulcers, gastrointestinal (GI) clinical events such as bleeding, and GI symptoms such as dyspepsia. 1 Furthermore, the cost of managing NSAID-associated GI side effects is substantial. 1 5 Cyclooxygenase 2 (COX-2) specific inhibitors (coxibs) have been demonstrated to decrease the development of ulcers, GI clinical events such as bleeding, and GI health care resource utilization as compared with traditional nonselective NSAIDs in double-blind trials. 1,6 10 For this reason, coxibs are frequently prescribed in lieu of traditional NSAIDs, accounting for at least one third of United States NSAID prescriptions and 60% of NSAID prescription costs. 1 However, whether the use of coxibs in place of traditional NSAIDs decreases GI resource utilization in routine clinical practice is unknown. Health care reform and escalating health care costs make the evaluation of resource utilization increasingly important. Although knowledge about clinical endpoints is important, the process of care and the costs associated with the use of NSAIDs and coxibs is also relevant to physicians, patients, and payors. Primary end points such as ulcers or GI bleeding may be assessed well in clinical trials, but health care resource utilization generally cannot be properly evaluated in controlled clinical trials because many of the determinants of resource utilization are mandated by the protocol. We therefore used an integrated administrative claims database of more than 8 million lives to assess GI-related resource use in patients switching from chronic NSAID therapy to chronic coxib therapy and also compared resource use in patients starting chronic NSAID therapy and patients beginning chronic coxib therapy. Materials and Methods This longitudinal, retrospective cohort study used medical and pharmacy claims data from a geographically diverse cohort of 8 million people in the United States residing Abbreviations used in this paper: COX-2, cyclooxygenase 2; coxibs, cyclooxygenase 2 specific inhibitors; GI, gastrointestinal; PPI, proton pump inhibitors by the American Gastroenterological Association /03/$30.00 doi: /s (03)

2 390 LAINE ET AL. GASTROENTEROLOGY Vol. 125, No. 2 in over 20 states. The database used is composed of longitudinal, member-linked data on medical services provided through commercial HMO, PPO, and other indemnity products that are marketed to employers and other commercial groups; 72% of the database s members belong to HMO or PPO plans. Patients were eligible if they filled a prescription for either a traditional NSAID or a coxib between January 1, 1999, and June 30, 2000 ( index prescription ), were continuously benefit eligible 365 days before and after the index prescription, had 2 refills of the index agent or another drug in the same class in addition to the index fill, and had no fills of the other index agent in the year following the index prescription. Patients were divided into switchers ( 3 prescriptions for traditional NSAIDs during the 12-month period immediately preceding an index coxib prescription) and newto-therapy NSAID or coxib users (no NSAIDs or coxibs during the 12 months before the index prescription). Administrative claims data 1 year before and after the index prescription were extracted for any inpatient hospitalization, emergency department (ED) visit, or outpatient visit with an ICD-9 or procedure code associated with an upper or lower GI diagnosis (list of codes used available upon request). Utilization patterns for resource use with only upper GI codes were similar to those including upper and lower GI codes, so we combined upper and lower GI-related resource utilization for our outcome measurements. Pharmacy claims for proton pump inhibitors (PPI), H2-receptor antagonists, sucralfate, and misoprostol were also extracted for all subjects. Patient chronic disease burden was assessed using a modified Chronic Disease Score (CDS) 11 based on that of Clark et al. 12 The CDS utilized the presence of drug markers filled during the 12 months before the index prescription to identify the presence of 27 chronic diseases, including hypertension, cardiac disease, depression, and hyperlipidemia. These were weighted and summed to derive an overall composite CDS for each patient that represents the patient s total chronic disease burden. Patients were subsequently classified as mild (0 3), moderate (4 11), or severe ( 11) to enable comparison of baseline risk between study groups. Differences between groups were examined for relevant demographic and clinical characteristics, as well as GI-related resource utilization during pre- and postindex periods. For statistical comparisons in both new-to-therapy groups, levels of significance were defined a priori and were calculated based on independent group t tests for continuous variables and 2 tests for categorical variables. In the switcher group, paired comparison t tests were used to compare post- vs. prior-period differences. Statistical comparisons were tested at a significance level of Prior to statistical modeling, univariate and bivariate descriptive analyses as well as assessment of the clinical relevance of independent variables were used to determine the appropriateness of variables to be included in the modeling process. Comparisons of the proportions of patients with GI-related resource use were performed, assessing inpatient hospitalization, outpatient and emergency department visits, prescriptions, and any GI resource use. In the switcher cohort, repeated measures logistic regression was used to estimate odds ratios (ORs) with 95% confidence intervals (CIs) for any GI-related use for the postindex coxib period vs. the prior traditional NSAID period, controlling for relevant independent variables. For the new-to-therapy groups, multiple logistic regression analysis was used to calculate ORs with 95% CIs for binomial outcome variables for chronic coxib use vs. chronic NSAID use during the postindex period, while controlling for relevant covariates, including GI-related resource use prior to NSAID or coxib therapy. GI-related medical resource costs were summed using the amount paid to the service provider for encounters. Drug ingredient cost was used to calculate medication costs. Total costs, consisting of GI-related costs in addition to the cost of NSAID or coxib prescriptions, were also determined. Because of potential inaccuracies in billing diagnoses, we first examined the GI resource cost using 2 different approaches for inpatient care: including hospitalizations with any GI diagnosis or only those with a primary GI diagnosis. Because the results indicated similar patterns, we conservatively report GI cost including inpatient costs for only those visits with a primary GI diagnosis. GI-related outpatient medical resource costs were analyzed using primary, secondary, or tertiary ICD-9 codes and the associated cost component of the visit. The average GI-related costs and total costs were estimated utilizing both the mean and median of the data in the original U.S. dollar scale. The cost differences for the switchers during the coxib period as compared with the NSAID period, as well as between new coxib and new NSAID patients, were compared by taking the difference of the group means. The uncertainty surrounding the cost difference was addressed by using the bootstrapping technique 13 (1000 repetitions, sampling with replacement) to calculate the corresponding 95% bootstrap CIs. The computation was done using the R Boot Package ( Because different population characteristics could account for higher levels of utilization, we further examined the relationship between new-to-therapy patients and GI-related costs or combined GI and NASID/coxib costs by using linear regression models with a log transformation of the dependent variables (GI-related costs or combined GI and NSAID/coxib costs) after taking differences in demographics, medical morbidity, and index drug usage into account. The covariates in the models included age, gender, prior GI-related resource use ($), preexisting diagnosis of pain or arthritis, steroid use, and number of index drug prescriptions. Partial financial support for data extraction and statistical analysis was provided by AstraZeneca (Wilmington, DE). No AstraZeneca employee had involvement in the analysis or drafting of this report. Results Characteristics of the Study Cohorts A total of 30,360 subjects were included in the study cohorts: 2246 switchers, 25,989 new-to-

3 August 2003 RESOURCE USE WITH COXIBS VS. NSAIDS 391 Table 1. Baseline Characteristics of the Study Groups Switchers (NSAID3Coxib) (n 2246) NSAID (n 25,989) New-to-therapy Coxib (n 2125) Age (yr) a Female (%) a 64.3 Severe chronic disease score (%) a 24.0 Osteoarthritis (%) a 32.0 Rheumatoid arthritis (%) a 7.6 Any GI resource use prior to index prescription (%) a 51 a P vs. switchers, new-to-therapy coxib. therapy NSAID, and 2125 new-to-therapy coxib. Selected characteristics of the study groups are summarized in Table 1. The mean chronic disease score at baseline was lower for new NSAID patients ( ) than new coxib patients ( [P 0.001]) or switched coxib patients ( [P ]). Patients with higher chronic disease scores were more likely to have a GI-related claim during the prior period: e.g., GI claims in 16% of mild, 53% of moderate, and 77% of severe new coxib patients; and 10% of mild, 34% of moderate, and 62% of severe new NSAID patients. The times of initial GI visits or prescriptions were split evenly across the 12-month period before the index prescription and were comparable in the 3 cohorts: 26% 30% of first GI resource use in the year prior to the index prescription occurred in the 3 months before the index prescription. In addition, no relation was seen between GI-related resource use and proximity of index prescription to the date of the launch of coxibs in the United States. GI Resource Utilization and Costs for Switchers (NSAIDs to Coxibs) A slightly lower proportion of switchers used GI resources during the chronic coxib period (54.9%) as compared with the chronic NSAID period (59.1%) (Table 2). The adjusted ORs (coxib vs. NSAID periods) were 0.86 ( ) for any GI resource utilization and 0.82 ( ) for a GI prescription. Adjusted ORs for GI-related oupatient/ed visits and hospitalizations were not significant (Table 2). PPI therapy was used by 603 (26.8%) of the switchers during the preswitch NSAID period, and 424 (70.3%) of these 603 patients remained on PPIs after switching to coxibs. No PPI therapy was used by 1643 (73.2%) of the switchers during the NSAID period, and 176 (10.7%) of these 1643 started PPIs during the coxib period. Thus, PPI therapy was used by 603 (26.8%) of patients before switching and 600 (26.7%) after switching. The sum total of GI-related dollars spent for all patients in the switcher group combined was 2.9% lower in the coxib period than the NSAID period. Analysis of the 1562 (69.5%) switchers with any GI-related resource use during the NSAID or coxib period showed a slight decrease in GI costs during the coxib period (mean, $650; median, $211) vs. the NSAID period (mean, $669; median, $242). The mean cost difference in GIrelated cost between the coxib period and the NSAID period of $19 was not statistically significant using the 95% bootstrap confidence interval (95% CI: $139, $55). When the cost of the NSAID or coxib was added to the GI-related costs, the mean combined cost was $1290 (median, $929) during the coxib period vs. $913 Table 2. Crude Proportions of Patients With GI Healthcare Resource Utilization and Adjusted Odds Ratios in Patients Switching From Chronic NSAID to Chronic Coxib Therapy GI resource use NSAID (n 2246) Coxib (n 2246) Adjusted OR (95% CI) Coxib vs. NSAID Any GI resource a 59.1% 54.9% 0.86 ( ) GI prescription 48.4% 44.7% 0.82 ( ) GI outpatient/ed visit 34.9% 32.6% 0.94 ( ) GI hospitalization 3.3% 2.8% 0.98 ( ) a OR adjusted for age; sex; chronic disease score; number of GI prescriptions in the prior period; number of GI-related outpatient, inpatient, or ER visits; prior diagnosis of joint pain, back pain, or skeletomuscular pain; and number of NSAID/coxib prescriptions during coxib and NSAID periods.

4 392 LAINE ET AL. GASTROENTEROLOGY Vol. 125, No. 2 Table 3. Crude Proportions of Patients With GI Healthcare Resource Utilization and Adjusted Odds Ratios in Patients Initiating Chronic NSAID vs. Chronic Coxib Therapy New-to-Therapy GI resource use NSAID (n 25,989) Coxib (n 2125) Adjusted OR (95% CI) Coxib vs. NSAID Any GI resource a 32.6% 51.0% 1.04 ( ) GI prescription 21.9% 41.5% 1.13 ( ) GI outpatient/ed visit 19.8% 27.3% 0.98 ( ) GI hospitalization 1.6% 2.5%.90 ( ) a Adjusted for patient age; gender; prior GI diagnoses; number of prior GI prescriptions; number of GI-related outpatient, inpatient, or ED visits; chronic disease score; timing of initiation of NSAID/coxib therapy; and number of NSAID/coxib prescriptions. (median, $539) during the NSAID period, a significant increase of $377 (95% CI: $271, $488) after the switch from NSAIDs to coxibs. GI Resource Utilization and Costs for New-to-Therapy Patients The crude proportions and ORs for GI-related resource utilization during the 12-month study period for new-to-therapy coxib vs. NSAID patients are shown in Table 3. Compared with patients beginning NSAID therapy, new coxib patients had higher crude proportions of any GI prescriptions, any outpatient or ED GI-related visits, any GI hospitalizations, and any GI resource utilization. However, adjusted ORs did not show evidence of a significant difference in any of these utilization categories. GI-related medical resource cost differences in newto-therapy patients were evaluated only for patients with GI-related costs after the index prescription (n 9372). The mean GI consumption for new-to-therapy coxib patients was $835 (median, $377) vs. $616 (median, $175) for new NSAID patients. The mean cost difference (new coxib-new NSAID) was $219 (95% CI: $93, $351). When the drug cost of NSAID or coxib therapy was included, the mean combined cost was $1370 (median, $962) for new coxib vs. $714 (median, $285) for new NSAID patients. The mean cost difference was $656 (95% CI: $533, $774). The new coxib patients had significantly higher GI-related costs and combined costs (coxib therapy GI-related costs) than the new NSAID group. The positive associations between new coxib patients and higher GI-related costs and combined costs were also examined in the regression analyses after adjusting for covariates; this adjustment did not alter the conclusion that new coxib patients consumed more GIrelated dollars and had higher combined costs than new NSAID patients (P 0.001). Costs Related to Lower GI Diagnoses The proportion of overall GI costs because of lower GI diagnoses was also determined. The mean costs for lower GI diagnoses were 13.4% ($90/$669) of the mean total GI costs for the NSAID period in switchers, 20.8% ($135/$650) for the coxib period in switchers, 29.8% ($184/$617) for the new-to-therapy NSAID patients, and 16.5% ($138/$835) for new-to-therapy coxib patients. The mean difference in lower GI costs for the switchers of $45 (95% CI: $20, $123) was not significant. The mean difference in lower GI costs of $46 (95% CI: $104, $7) between new coxib patients and new NSAID patients also was not significant. Discussion NSAID-related GI side effects are responsible for a large economic burden on the health care system. North American studies indicate that, for each dollar spent on NSAIDs, $0.66 to $1.25 is spent on GI side effects, 2,4 and that nearly one third of total medical costs in arthritis patients may be related to GI adverse events. 3 Furthermore, prior analyses may have underestimated the full spectrum of NSAID-related GI health care resource utilization because they focus on upper GI processes or do not define the specific GI diagnoses included. However, NSAIDs also may increase the risk of lower GI complications. 14 Lower GI diagnoses accounted for 32% of all GI hospitalizations in osteoarthritis patients and 13% in rheumatoid arthritis patients in the ARAMIS data bank, 15 whereas lower GI events accounted for 40% of all serious GI events in a recent outcomes trial of NSAID use in rheumatoid arthritis patients. 16 Coxibs were developed to provide antiinflammatory efficacy comparable with nonselective NSAIDs with decreased GI toxicity. Double-blind GI outcome trials demonstrated a reduction in upper and lower GI clinical events 8,16 and in GI health care resource utilization. 9,10 Nevertheless, results from trials with detailed protocols that govern or influence the care of enrolled patients may not reflect events in standard clinical practice. Therefore, we used claims from a large integrated database to determine the real world effects of the introduction of coxibs on GI health care resource utilization. In addition,

5 August 2003 RESOURCE USE WITH COXIBS VS. NSAIDS 393 we specifically included lower GI diagnoses in our study to more accurately reflect all NSAID-related GI effects. Our results show that coxibs are prescribed more frequently than traditional NSAIDs to patients who are older, have more severe concurrent illness, or have a diagnosis of osteoarthritis or rheumatoid arthritis. As might be expected, a greater proportion of coxib users also had preexisting GI disease or symptoms. We performed 2 separate comparisons to assess the effect of coxibs on GI health care resource utilization. First, we examined chronic NSAID users who switched to chronic coxib therapy, using the initial NSAID period as a control for the coxib period. Second, we compared concurrent groups of patients who started either chronic NSAID or coxib therapy without use of either agent in the preceding 12 months, adjusting for imbalances in clinical factors between the new NSAID and new coxib groups. In both cases, we examined the proportion of patients with utilization of GI health care resources (hospitalization, outpatient, or emergency visit or prescription for antisecretory drugs, misoprostol, or sucralfate) and the reimbursements paid to providers for these resources. Patients switching from chronic NSAID therapy to chronic coxib therapy had a small decrease in the use of any GI resource during the 12-month coxib period (adjusted OR: 0.86 [ ]) because of a small decrease in the proportion using medical cotherapy (adjusted OR: 0.82 [ ]). Among patients using PPIs with chronic NSAID therapy, 70.3% remained on PPIs after switching to coxibs, whereas 10.7% of those not using PPIs with NSAIDs started PPIs after switching to coxibs. No significant decrease in GI-related health care costs was seen with the switch to coxibs. When the cost of the patients antiinflammatory medication, NSAID or coxib, was added to the GI costs, however, the combined costs during the coxib period were markedly higher, with a mean increase of $377. When patients initiating chronic NSAID or coxib therapy at concurrent times were compared, the proportion with GI resource utilization was similar for all categories, but the new coxib group had higher costs than the new NSAID group after adjusting for important covariates. The strengths of a claims database study are that it allows examination of a large number of patients and determination of actual practice patterns. An inherent limitation of retrospective database studies is that the populations are not randomized to treatment, so selection bias resulting from prescriber treatment decisions may occur. We have attempted to minimize this by use of a pre vs. post design in patients switching from NSAIDs to coxibs so that each patient is essentially his or her own control and through statistical adjustment in the new-to-therapy cohort. Nevertheless, it is difficult to ascertain the clinical reason for initiating or switching to a coxib with a retrospective database, still leaving open the potential of systematic bias. Furthermore, disease severity may increase over time, increasing the need for GI medications, visits, and costs in the period after the switch. In this case, GI resource utilization and costs could arguably have been even higher had the patients remained on an NSAID rather than switching to a coxib. Several potential explanations for the failure of coxibs to decrease GI health care resource utilization in this evaluation of standard clinical practice may be postulated, including variations in clinical practice, patient characteristics such as baseline comorbidity, patient therapy preferences, and the fact that health care resource utilization may not parallel clinical outcomes. For example, physicians might replace an NSAID with a coxib but continue the practice pattern (e.g., PPIs) they use for patients taking traditional NSAIDs. In addition, many patients taking coxibs appropriately receive medical cotherapy, e.g., patients at very high risk for ulcer complications and patients with other indications for GI medications, such as reflux or dyspepsia. GI symptoms associated with coxibs also may lead to doctor visits and diagnostic evaluations, further blunting any decrease in costs associated with coxib therapy. Potential explanations for the lack of change in hospitalizations with coxibs include discontinuation of GI cotherapy in some patients and closer monitoring of higher-risk patients leading to earlier diagnosis and treatment. Furthermore, the absolute risk reduction in GI complications in randomized trials is 0.3% 1.3%, 8,16,17 a difference that may be difficult to demonstrate even in large studies. In addition, coxibs conceivably may not have provided significant protection in this real-world group of patients, e.g., if a large proportion of patients was taking concurrent aspirin or over-the-counter NSAIDs, which cannot be measured in a claims database study. Over-thecounter NSAIDs could conceivably be more common in patients with other comorbidities and act as an unmeasured confounding factor. Finally, a greater proportion of coxib patients had rheumatoid arthritis and osteoarthritis. These patients are more likely to take long-term therapy, and rheumatoid arthritis patients may have a higher risk for GI events and are more likely to take other medications such as prednisone. The timing of initial GI encounters also could bias our results against coxibs. Coxibs would be expected to provide little or no decrease in GI resource use if a large

6 394 LAINE ET AL. GASTROENTEROLOGY Vol. 125, No. 2 proportion of patients developed their first GI problem just prior to beginning a coxib. However, the times of initial GI encounters in our patients were split evenly across the 12-month period prior to the index prescription and were comparable in all cohorts. Potential explanations also need to be considered for the fact that GI-related costs were higher in patients starting coxibs for the first time as compared with those starting NSAIDs for the first time. Unlike the switchers analysis, in which a paired comparison in the same patients was performed, the comparison of the new-totherapy groups was between 2 separate, nonrandomized cohorts. In cohort studies, differences between the study groups other than the exposure being assessed may be present, and these differences may bias the results. Although we attempted to correct for these differences with statistical methods to adjust for clinically relevant variables, the possibility that unmeasured factors may confound results always exists. For example, if patients who are started on coxibs are sicker than those started on NSAIDs, but this is not completely accounted for by the variables included in the regression model, then the increase in cost might be due to the degree of sickness rather than the choice of antiinflammatory agent. The fact that new-to-coxib patients had higher GI costs without a higher proportion using GI resources supports this theory: Sicker patients may have more complex and expensive diagnostic testing, hospitalizations, or drug use. Alternatively, doctors who choose the more expensive coxib as initial therapy may also use other more expensive practice patterns, e.g., more expensive and/or longer courses of medical cotherapies, more expensive diagnostic tests, longer hospitalizations. Prior studies suggest that medications and hospitalizations are the major determinants of GI-related costs in NSAID users, 3,4 whereas others have reported that medications alone account for most of the GI-related costs in NSAID users. 5 We found that medications represented the largest proportion of GI-related costs in NSAID users, approximately 3-fold higher than the cost of hospitalization, even in the relatively high-risk patients who were switched to coxibs. Prior studies also have suggested that outpatient care accounted for a very small part of overall GI costs in NSAID users. 4,5 However, we found that costs for outpatient and emergency visits were more than double those for inpatient care among switchers and approximately 50% higher for new-to-therapy patients, perhaps reflecting the recent greater emphasis on outpatient management of disease. Finally, although lower GI events also are increased with NSAID use, 14 prior analyses have not assessed health care resource costs related to lower GI disorders in patients taking NSAIDs. We found that approximately 15% 30% of overall GI costs in NSAID users were due to lower GI diagnoses. Thus, lower GI clinical effects of NSAIDs need to be specifically sought when assessing the GI costs of NSAID use. In conclusion, coxibs were more commonly prescribed to patients who are older, have more severe concurrent medical illnesses, and have preexisting GI diseases or symptoms. In addition, GI medications make up the largest proportion of GI-related costs in patients taking NSAIDs. Patients switching from traditional NSAID therapy to coxib therapy had a small decrease in the proportion using any GI resources because of a decrease in use of GI medications. The costs of GI health care were not decreased significantly with coxib therapy. When costs of the antiinflammatory medication (traditional NSAID or coxib) were included, the costs during the coxib period were considerably higher. Among patients without prior NSAID or coxib use, after adjustment for confounding factors, the proportion with GI health care resource use was comparable in those receiving chronic coxib and traditional NSAID therapy, but the costs were higher in the coxib group. Thus, although GI side effects are decreased with coxibs as compared with traditional NSAIDs in randomized controlled trials, this benefit appears to have not yet been translated into an important decrease in GI health care costs in real world clinical practice. References 1. Laine L. Approaches to NSAID use in the high-risk patient. Gastroenterology 2001;120: Rahme E, Joseph L, Kong SX, Watson KJ, LeLorier J. 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: Bloom BS. Direct medical costs of disease and gastrointestinal side effects during treatment for arthritis. Am J Med 1986; 84(suppl 2A): Smalley WE, Griffin MR, Fought RL, Ray WA. Excess costs from gastrointestinal disease associated with nonsteroidal anti-inflammatory drugs. J Gen Intern Med 1996;11: Lanes SF, Lanza LL, Radensky PW, Yood RA, Meenan RF, Walker AM, Dreyer NA. Resource utilization and cost of care for rheumatoid arthritis in a managed care setting. The importance of drug and surgery costs. Arthritis Rheum 1997;40: Laine L, Harper S, Simon T, Bath R, Johanson J, Schwartz H, Stern S, Quan H, Bolognese J. A randomized trial comparing the effect of rofecoxib, a COX-2 specific inhibitor, to ibuprofen on the gastroduodenal mucosa of osteoarthritis patients. Gastroenterology 1999;117: Simon LS, Weaver AL, Graham DY, Kivitz AJ, Lipsky PE, Hubbard RC, Isakson PC, Verburg KM, Yu SS, Zhao WW, Geis GS. Antiinflammatory and upper gastrointestinal effects of celecoxib in rheumatoid arthritis. A randomized controlled trial. JAMA 1999; 282:

7 August 2003 RESOURCE USE WITH COXIBS VS. NSAIDS Bombardier C, Laine L, Reicin A, Shapiro D, Burgos-Vargas R, Davis B, Day R, Bosi Ferraz M, Hawkey CJ, Hochberg MC, Kvien TK, Schnitzer TJ. Comparison of upper gastrointestinal toxicity of rofecoxib and naproxen in patients with rheumatoid arthritis. N Engl J Med 2000;343: Laine L, Bombardier C, Reicin A, Hawkey C, Watson DJ, Ramey DR, Brett C. Gastrointestinal (GI) co-therapy, procedures, and hospitalizations in a GI outcomes study of rofecoxib vs naproxen in rheumatoid arthritis. Am J Gastroenterol 2000;95:2633 (abstr). 10. Goldstein JL, Eisen G, Stenson W, Agrawal N, Burke T, Pettitt D, Wilson AM, Fort J. Celecoxib significantly reduces gastrointestinal-related healthcare resource utilization compared to NSAID: SUCCESS-1 in ostearthritis (OA) trial (abstr). Am J Gastroenterol 2001;96(suppl):S Zhao Z, Boscarino JA, Deverka PA, Epstein RS, Wang Chin J. Predicting drug spending: update and evaluation of a revised chronic disease score. Presented at ISPOR Fifth International Meeting, May Clark DO, Von Korff M, Saunders K, Baluch WM, Simon G. A chronic disease score with empirically derived weights. Med Care 1995;33: Efron B, Tibshirani RJ. An introduction to the bootstrap, New York: Chapman and Hall, Bjarnason I, Hayllar H, Macpherson AJ, Russell AS. Side effects of nonsteroidal anti-inflammatory drugs on the small and large intestine in humans. Gastroenterology 1993;104: Singh G, Rosen Ramey D. NSAID induced gastrointestinal complications: the ARAMIS perspective Arthritis, Rheumatism, and Aging Medical Information System. J Rheumatol 1998; 25(suppl 51): Laine L, Connors LG, Reicin A, Hawkey CJ, Burgos-Vargas R, Schnitzer TJ, Yu Q, Bombardier C. Serious lower GI clinical events with non-selective NSAID or coxib use. Gastroenterology 2003; 124: FDA Arthritis Advisory Committee meeting. February 7, Available at: Accessed June 16, Received January 9, Accepted May 1, Address correspondence to: Loren Laine, M.D., GI Division, Department of Medicine, U.S.C. School of Medicine, 2025 Zonal Avenue, Los Angeles, California llaine@usc.edu; fax: (323) Supported in part (for data extraction and statistical analysis) by AstraZeneca (Wilmington, Delaware). Dr. Laine received research support from Astra Zeneca, TAP, Merck, and Johnson & Johnson; and is a consultant to Astra Zeneca, Merck, Pfizer, and TAP. The Institute for Effectiveness Research, LLC, is a subsidiary of Medco Health Solutions, Inc., a Merck Company.