This Article Abstract Submit a response to this article Services E-mail this article to a friend Alert me to new issues of the journal

D A T A W A T C H
C O X - 2  U S E
W E B  E X C L U S I V E   

18 February 2004  
  

The Impact Of Drug Coverage 
On COX-2 Inhibitor Use In Medicare

Elders with the most generous drug coverage in 2000,
not those who met certain clinical criteria, were most likely to use COX-2 inhibitors.

By Jalpa A. Doshi, Nicole Brandt, and Bruce Stuart

ABSTRACT:

Passage of the Medicare drug benefit legislation has renewed attention to the relationship between drug coverage and medication use. This study assesses the impact of drug coverage on COX-2 inhibitor use among elderly people with osteoarthritis, taking into account risk for adverse gastrointestinal events. COX-2 use among aged beneficiaries with the most generous coverage was twice that of those with no third-party coverage. COX-2 use also increased with increasing gastrointestinal risk. However, this risk differential in COX-2 use disappears among those with the most generous coverage. Potential overuse of costly medications should be addressed as the Medicare drug benefit is being phased in.

With the passage of the Medicare drug benefit legislation in November 2003, policymakers have begun to evaluate how the new law will affect drug use among Medicare beneficiaries.¹ Virtually all experts agree that expanding drug coverage to beneficiaries who lack it will lead these beneficiaries to use more medications.² The presumption is that more use is a good thing. Several recent studies have shown that elderly people with chronic disease are less likely to receive costly but essential medications if they are uninsured.³ However, the law may also increase access to expensive therapies that either are not needed or have less costly substitutes. Although the line between necessary and unnecessary care is generally difficult to draw precisely, the opportunities for unnecessary use of prescription medications are greater than for most other kinds of health services. The widespread overuse of antibiotics in treating respiratory infections is a case in point.⁴ Yet, surprisingly, we know little about the extent of overuse of drug treatments for the kinds of chronic conditions most commonly affecting elderly Medicare beneficiaries.

Osteoarthritis (OA) affects more than twenty-one million Americans; its prevalence is particularly high among older adults.⁵ Nonsteroidal anti-inflammatory drugs (NSAIDs) are commonly used for pain relief; however, these agents have a potential for adverse gastrointestinal (GI) effects such as bleeding and ulcers, particularly in the presence of risk factors such as older age, history of peptic ulcer disease, and concomitant use of corticosteroids and anticoagulants.⁶ Cyclooxygenase-2 (COX-2) selective inhibitors are newer drugs launched in late 1998 (celecoxib) and mid-1999 (rofecoxib) that have a GI safety advantage over traditional NSAIDs.⁷

Although COX-2s are potentially safer, they are more expensive than other NSAIDs, many of which are available as generic over-the-counter (OTC) formulations. Furthermore, they offer no added benefit in pain relief over traditional NSAIDs.⁸ If drugs are prescribed strictly in terms of relative cost and benefit, one would expect to see high rates of COX-2 use among arthritis sufferers with sizable risk of GI complications and low rates among those with lesser risk.⁹ However, drug coverage distorts this equation. By making expensive medications cheaper to the patient (sometimes cheaper than OTC substitutes), insurance increases the likelihood that people at low risk of adverse reactions will get the expensive product. The more generous the insurance, the greater the potential distortion.

We designed this study to examine the influence of drug coverage on use of COX-2 inhibitors among elderly Medicare beneficiaries with OA and varying levels of GI risk. We address three questions: (1) Are elderly OA patients who are at greater risk of adverse GI effects more likely than those at lower risk to be prescribed a COX-2? (2) Are OA patients who have more generous drug coverage more likely than those with limited or no drug coverage to be prescribed a COX-2? (3) Does drug coverage influence differentiation in prescribing COX-2s by level of GI risk?

Study Data And Methods

Data source. Data for the study were taken from the 2000 Medicare Current Beneficiary Survey (MCBS), a nationally representative survey of the Medicare population conducted under the auspices of the Centers for Medicare and Medicaid Services (CMS). The use of 2000 data is particularly suitable for this study since COX-2 inhibitors were well established in the market by then, while concerns related to their adverse cardiovascular effects (which may have affected physicians’ prescribing patterns) were not published until 2001.¹⁰

The MCBS collects extensive information on sociodemographics, health and functional status, health insurance, health care use, and costs and is linked to Medicare claims. Prescription data are based on self-reports. To assure accurate recall, respondents are asked to keep medication logs, save insurance receipts, and show the interviewers all of their medication containers during the thrice-yearly interviews. The MCBS does not capture use of OTC medications.

Study sample. Our study sample comprised community-dwelling beneficiaries age sixty-five and older who self-reported OA in the MCBS health survey and were enrolled in fee-for-service (FFS) Medicare for all of 2000. We used self-reported OA rather than diagnosed arthritis in the Medicare claims because common chronic conditions are frequently undercounted in claims files. Beneficiaries in long-term care facilities were excluded because drug use for this population is not included in the MCBS files. We excluded beneficiaries in Medicare health maintenance organizations (HMOs) since Medicare claims (necessary to identify adverse GI events) are not available for this group. The final sample consisted of 4,601 beneficiaries (weighted N = 14.3 million).

Measures. Our dependent variable was any COX-2 inhibitor prescription in 2000. Independent variables included age, sex, race, income, risk factors for GI events, and generosity of drug coverage. We used the Veterans Health Administration’s (VHA’s) Gastrointestinal Risk Assessment Tool to quantify each patient’s risk for NSAID-induced GI events.¹¹ This validated scoring tool, developed by Gurkirpal Singh and colleagues at Stanford University, was adopted by the VHA Pharmacy Benefits Management Strategic Healthcare Group and the Medical Advisory Panel as part of the VHA approval criteria for the nonformulary use of COX-2 inhibitors in veteran patients.¹² It comprises six questions with points assigned based on the patient’s response. These include age (12 points for ages 61–65, up to 18 points for older than age 85), self-reported health status (0–4 points), self-reported rheumatoid arthritis (0 or 2 points), months of exposure to corticosteroid medications (0–5 points), GI bleeding or ulcers (0 or 8 points), and GI symptoms with NSAID use in the absence of GI events (0 or 2 points). Patients with a risk score below 11 are considered at low risk of GI problems (given the age range of our sample, none of our subjects was at “low risk”), and nonselective NSAIDs are the recommended therapy. A score of 12–15 is considered “moderate risk,” but nonselective NSAIDs are still recommended. Patients with a score of 16–20 are defined as being at “significant risk,” and a standard NSAID should be considered if used for less than thirty days or intermittently. Patients with more than thirty days’ use should be considered for a COX-2 only if they are intolerant to etodolac or salsalate or if therapy fails. Finally, beneficiaries with a risk score of 21 or greater are at “substantial risk,” and a COX-2 can be prescribed. In general, the VHA recommends that all patients with OA undergo a therapeutic trial of acetaminophen (4,000 mg daily) or salsalate before taking a COX-2.

We could map the first three questions in the scoring tool to information in the MCBS. The remainder required use of assumptions or proxy measures. Since days’ supply is not reported in the MCBS, we treated each corticosteroid mention as a thirty-day supply. We identified GI events from Medicare claims (ICD-9 codes 531–534, 578) occurring at any time during 2000. The MCBS does not query respondents about GI symptoms (such as heartburn, stomachache, nausea, or vomiting), and these are not well captured in claims data. As a proxy measure, we used evidence of use of gastroprotective agents. We also captured use of anticoagulants or antiplatelets, or both, and the number of OA-related physician visits as additional risk factors.
We defined the generosity of prescription drug coverage as the share of beneficiaries’ annual drug spending paid for by insurance. Other studies have used type of drug coverage (for example, employer-sponsored, self-purchased) when assessing access to essential medicines, but there is such heterogeneity within types that we chose a more direct measure.¹³ We grouped beneficiaries into four categories: 0 percent, 1–50 percent, 51–75 percent, and 76–100 percent. The bottom category represents primarily beneficiaries with no drug coverage but also includes a small group with such limited drug coverage that none of their drug spending was paid for by the third party. On the other end of the spectrum, beneficiaries in the most generous category had at least three-quarters of their annual drug bill paid for by insurance.

Statistical analysis. We constructed cross-tabulations showing the bivariate relationships between COX-2 inhibitor use, GI risk scores, and generosity of drug coverage. We used logistic regressions to obtain adjusted-odds ratios for COX-2 use conditional on the various study variables. We estimated separate regressions for each insurance generosity subsample to examine GI risk differentiation in COX-2 use. All analyses were conducted in STATA 7.0 using the survey estimators to account for the complex design of the MCBS.

Study Results

Exhibit 1 shows characteristics of the 4,601 people in the sample, weighted to represent the 14.3 million elderly Medicare beneficiaries with OA. The sample had a mean age of 76.4 years, was mostly female, and was predominantly white. While 30 percent had no third-party drug payments, about 36 percent fell in the most generous drug coverage group. In terms of GI risk scores, 18 percent had scores in the “moderate risk” range (12–15), while 22 percent had scores in the “substantial risk” range (21 or higher).

Exhibit 2 reports prevalence of COX-2 inhibitor use by GI risk scores as well as by individual risk factors. While the overall prevalence of COX-2 use was about 20 percent in our sample, we found a positive correlation between GI risk score and COX-2 use. Beneficiaries with GI risk scores of 21 or greater (26 percent) had 50 percent higher COX-2 use than those with scores of 12–15 (17 percent). Individual GI risk factors significantly associated with higher use of COX-2s (p < .05) were fair-to-poor health, rheumatoid arthritis, use of corticosteroids and gastroprotective agents, and number of OA visits during the year. While COX-2 use was very similar for Medicare beneficiaries ages 65–79 (20–21 percent), it dropped to 18 percent for beneficiaries age eighty and older.

Exhibit 3 depicts the relationship between COX-2 use and the generosity of drug coverage. As generosity increased, COX-2 use rose for all risk groups. The relationship is strong and statistically significant (p < .05) for all groups. However, the relative increase in COX-2 use was much higher for beneficiaries in the “moderate risk” group than for those in the “substantial risk” group. As a result, the high level of risk differentiation evident in the group with no third-party drug payments essentially disappears among those with the most generous drug coverage. This same pattern holds for each individual risk factor displayed in Exhibit 4. More generous drug coverage is associated with higher COX-2 use among those at greatest risk, however measured, but the increase is always greater for those at lesser risk. It is interesting to note that the greatest percentage difference (more than 500 percent) in COX-2 use is between beneficiaries with no OA-related physician visits and no third-party payments (9.2 percent) and those with five or more OA visits and the most generous drug coverage (55.9 percent).

The logistic regression results presented in Exhibit 5 are consistent with the bivariate relationships: Increasing risk of GI problems is associated with greater odds of being prescribed a COX-2, but generous drug coverage has a much larger impact. Controlling for all other factors, beneficiaries having a GI risk score of 21 or more had about 60 percent higher odds [odds ratio 1.63; 95 percent confidence interval (1.20–2.20)] of receiving a COX-2 compared with those having a GI risk score of 12–15. All else being equal, those with the most generous insurance coverage had more than double the odds of getting the expensive COX-2s [OR 2.22; 95 percent CI (1.80–2.75)] compared with those having no third-party payments.¹⁴ Separate logistic regression estimates for the subgroup of beneficiaries with no third-party payments (not shown) confirmed a high level of GI risk differentiation in COX-2 use [OR 2.34; 95 percent CI (1.13–4.86) for GI risk scores of 16–20 and OR 3.55; 95 percent CI (1.72–7.33) for GI risk scores above 21, compared with the reference group having scores of 12–15]. This risk differentiation was not evident in the subgroup analysis of those with the most generous drug coverage [OR 1.08; 95 percent CI (0.71–1.65) for GI risk scores of 16–20 and OR 1.23; 95 percent CI (0.75–2.02) for GI risk scores above 21, compared with the reference group having scores of 12–15].

Discussion

Our study examined variations in the use of an expensive class of pain-relief drugs as a function of GI risk factors and the generosity of drug coverage among aged Medicare beneficiaries with OA. As expected, we found that COX-2 use increases with GI risk. On average, only a quarter of those at “substantial risk” of a gastrointestinal problem received a COX-2 in 2000, which suggests possible underuse of these medications. We also found that COX-2 use rises with increasing generosity of drug coverage. The average likelihood of using a COX-2 rose from 13 percent among the least well insured to 26 percent among those with the most generous coverage. Generosity of drug coverage substantially mediates the odds of being prescribed a COX-2 as a function of GI risk. Aged Medicare beneficiaries with the most generous drug coverage and only moderate risk of a GI problem were actually more likely to get a COX-2 (25 percent) than beneficiaries with no or limited coverage but at substantial risk (20 percent). Among those with the best coverage, GI risk had no independent effect on who got the COX-2s. In other words, irrespective of the GI risk, people with the most generous prescription plans in 2000 were more likely to use COX-2 inhibitors.

In sum, while drug coverage is clearly associated with greater use of expensive COX-2 inhibitors, most of the increase in use is among those least in need. This conclusion must be interpreted in light of several important study limitations. First, our study sample excluded aged Medicare beneficiaries in Medicare HMOs, which have a strong incentive to manage the use of costly drugs. Second, our data set did not capture use of OTC NSAID medications, which made it impossible to estimate the true level of NSAID use. Hence, we cannot comment on the extent to which beneficiaries across the insurance categories, particularly those at high GI risk and with no COX-2 use, were exposed to risk for NSAID-induced GI complications. Similarly, we were unable to assess possible substitution of OTC acetaminophen for COX-2 use. It seems reasonable to suspect that elderly beneficiaries with no drug coverage would be more likely than those with generous coverage to select acetaminophen. This could help explain the low level of COX-2 use by people with no coverage, but it cannot explain the lack of risk differentiation in COX-2 use for people with generous coverage. In other words, while the lack of data on OTC medications means that our findings cannot be used to establish the true degree of underuse of COX-2s, this omission does not affect our conclusion on potential overuse of COX-2s by the well-insured at low GI risk.

Two measurement issues also deserve mention. First, we measured generosity of drug coverage using payment data rather than information on benefit design (which is unavailable in the MCBS). It is possible that we classified people with the same type of coverage into different generosity categories depending on their annual drug use. Although this reduces the precision of our estimates, it is not expected to impart any systematic bias to our results. The issue is behavior based on exposure to price signals, so the fact that there might be differing exposure to price signals stemming from differing levels of consumption is just as relevant for behavior based on those signals as is a difference based on the administrative categorization of an insurance policy. A second potential source of error arises in the measurement of GI risk factors. We may have misclassified some beneficiaries because of the need to use proxy measures or assumptions on duration of corticosteroid use, GI events, and GI symptoms. However, there is no reason to believe that any such misclassification would be systematically related to generosity of drug coverage and lead to biased results. Finally, our study used the VHA scoring tool to define beneficiaries in different GI risk levels. Although it has been implemented nationally in the VHA system, it is not a universal tool or guideline for COX-2 prescribing, and a cautious physician might have prescribed COX-2s by virtue of the patient’s older age alone. However, our assessment of COX-2 use by each of the individual GI risk factors (age, adverse GI events, corticosteroid use, and anticoagulant use) also indicated similar results, so that the high level of risk differentiation evident in the group with no third-party drug payments greatly diminished among people with the most generous drug coverage.

The findings from this study are timely, as Congress recently enacted legislation to add drug coverage to the Medicare benefit package. Prior research on the relationship between drug coverage and drug use by the elderly has focused on underuse of essential medications. That research has shown that providing all Medicare beneficiaries with a meaningful drug benefit will increase access to these medications. Our study suggests that policymakers should also be concerned with potential overuse of drug therapy by Medicare beneficiaries once the benefit is implemented. Given our findings, the provisions outlined for prescription drug plan sponsors to have cost and utilization management and medication therapy management in Section 1860 D-4(c) of the law take on added importance. Ideally, the drug benefit should assure that those at greatest need receive appropriate therapy. At the same time, the application of drug utilization management tools may be required to ensure that costly medications are not excessively used by those for whom less costly substitutes are suitable. Only then will Medicare be able to provide an affordable, sustainable, and cost-effective drug benefit.

The authors thank Jennifer Hardesty for her help with drug nomenclature in the MCBS and Becky Briesacher for her comments on an earlier draft. This work was presented in part at the AcademyHealth Annual Meeting, Nashville, Tennessee, June 2003. Research support was provided by the Commonwealth Fund. The views presented here are those of the authors and not necessarily those of the Commonwealth Fund, its directors, officers, or staff.

NOTES

1. Medicare Prescription Drug, Improvement, and Modernization Act of 2003, P.L. 108-173 (8 December 2003).
2. J.A. Poisal and L.A. Murray, “Growing Differences between Medicare Beneficiaries with and without Drug Coverage,” Health Affairs (Mar/Apr 2001): 74–85; B. Stuart and J. Grana, “Ability to Pay and the Decision to Medicate,” Medical Care 36, no. 2 (1998): 202–211; and F. Gianfrancesco, A. Baines, and D. Richards, “Utilization Effects of Prescription Drug Benefits in an Aging Population,” Health Care Financing Review 15, no. 3 (1994): 113–126.
3. J. Blustein, “Drug Coverage and Drug Purchases by Medicare Beneficiaries with Hypertension,” Health Affairs (Mar/Apr 2000): 219–230; A.S. Adams, S.B. Soumerai, and D. Ross-Degnan, “Use of Antihypertensive Drugs by Medicare Enrollees: Does Type of Coverage Matter?” Health Affairs (Jan/Feb 2001): 276–286; and A.D. Federman et al., “Supplemental Insurance and Use of Effective Cardiovascular Drugs among Elderly Medicare Beneficiaries with Coronary Heart Disease,” Journal of the American Medical Association 286, no. 4 (2001): 1732–1739.
4. A.C. Nyquist et al., “Antibiotic Prescribing for Children with Colds, Upper Respiratory Tract Infections, and Bronchitis,” Journal of the American Medical Association 279, no. 11 (1998): 875–877; and M.A. Steinman, C.S. Landefeld, and R. Gonzales, “Predictors of Broad-Spectrum Antibiotic Prescribing for Acute Respiratory Tract Infections in Adult Primary Care,” Journal of the American Medical Association 289, no. 6 (2003): 719–725.
5. R.C. Lawrence et al., “Estimates of the Prevalence of Arthritis and Selected Musculoskeletal Disorders in the United States,” Arthritis and Rheumatism 41, no. 5 (1998): 778–799; and S.L. Hughes and D. Dunlop, “The Prevalence and Impact of Arthritis in Older Persons,” Arthritis Care Research 8, no. 4 (1995): 257–264.
6. G. Singh and D.R. Ramey, “NSAID-Induced Gastrointestinal Complications: The ARAMIS Perspective—1997,” Journal of Rheumatology 25, Supp. 51 (1998): 8–16; and G. Singh and G. Triadafilopoulus, “Epidemiology of NSAID-Induced GI Complications,” Journal of Rheumatology 26, Supp. 56 (1999): 18–24.
7. F.E. Silverstein et al., “Gastrointestinal Toxicity with Celecoxib versus Nonsteroidal Anti-Inflammatory Drugs for Osteoarthritis and Rheumatoid Arthritis, The CLASS Study: A Randomized Controlled Trial,” Journal of the American Medical Association 284, no. 10 (2000): 1247–1255; and C. Bombardier et al., “Comparison of Upper Gastrointestinal Toxicity of Rofecoxib and Naproxen in Patients with Rheumatoid Arthritis: VIGOR Study Group,” New England Journal of Medicine 343, no. 21 (2000): 1520–1528.
8. Ibid.
9. B.M. Spiegel et al., “The Cost-Effectiveness of Cyclooxygenase-2 Selective Inhibitors in the Management of Chronic Arthritis,” Annals of Internal Medicine 138, no. 10 (2003): 795–806; and A. Maetzel, M. Krahn, and G. Naglie, “The Cost Effectiveness of Rofecoxib and Celecoxib in Patients with Osteoarthritis or Rheumatoid Arthritis,” Arthritis and Rheumatism 49, no. 3 (2003): 283–292.
10. D. Mukherjee, S.E. Nissen, and E.J. Topol, “Risk Of Cardiovascular Events Associated with Selective COX-2 Inhibitors,” Journal of the American Medical Association 286, no. 8 (2001): 954–959; and S.L. Targum, U.S. Food and Drug Administration, “Memorandum: Review of Cardiovascular Safety Database,” 1 February 2001, www.fda.gov/ohrms/dockets/ac/01/briefing/3677b2_06_cardio.pdf (14 January 2004).
11. Pharmacy Benefits Management Strategic Healthcare Group and the Medical Advisory Panel, Veterans Health Adminstration, Department of Veterans Affairs, “Use of Cyclooxygenase (COX) 2 Inhibitors Celecoxib (Celebrex) or Rofecoxib (Vioxx) in Veterans,” April 2001, www.vapbm.org/criteria/coxcriteria.pdf (3 February 2004); G. Singh et al., “GI Score: A Simple Self-Assessment Instrument to Quantify the Risk of Serious NSAID-Related GI Complications in RA and OA” (abstract), Arthritis and Rheumatism 41, Supp. (1998): S75; and M.M. Wolfe, D.R. Lichtenstein, and G. Singh, “Gastrointestinal Toxicity of Nonsteroidal Antiinflammatory Drugs,” New England Journal of Medicine 340, no. 24 (1999): 1888–1899.
12. Singh et al., “GI Score.”
13. Adams et al., “Use of Antihypertensive Drugs by Medicare Enrollees”; and Federman et al., “Supplemental Insurance and Use of Effective Cardiovascular Drugs.”
14. Reestimation of the logistic regression with the individual GI risk factors (see Exhibit 4) instead of the GI risk score categories resulted in similar adjusted-odds ratios for the generosity-of-coverage categories.

Jalpa Doshi (jdoshi{at}mail.med.upenn.edu) is a health services research scientist in the Division of General Internal Medicine, University of Pennsylvania, in Philadelphia. Nicole Brandt is an assistant professor, Department of Pharmacy Practice and Science, University of Maryland School of Pharmacy, in Baltimore. Bruce Stuart is a professor in the Department of Pharmaceutical Health Services Research there.

DOI: 10.1377/hlthaff.W4.94
©2004 Project HOPE–The People-to-People Health Foundation, Inc.